DEV Community: Asher Hu

Texture Depth 0.05 mm: How Much Draft Before It Stops Sticking?

Asher Hu — Tue, 20 Oct 2026 04:13:52 +0000

Start with the number. Draft angle = base 1–1.5° + (actual texture depth ÷ 0.025 mm) × 1°. A 0.02 mm texture gets 2°. A 0.05 mm texture gets 3°. That is the industry's working safety line, and the rest of this article is about the places where the line is not enough.

Take the grade most of our enclosures land on: VDI 3400 #30, Ra 4 μm. The real peak-to-valley depth sits between 0.016 and 0.024 mm, so the formula returns a little over 2°. On drawings we write 2.5°. A coarse #36 runs 0.04–0.06 mm deep and needs 3.5–4.5° on the side wall. Go under that and the first T0 shot tends to come out with scuffed, whitened walls — and the corrective round costs far more than the single degree you saved.

The texture number is a Ra value, not a depth

VDI 3400 is the German standard for spark-eroded and etched textures. Its grade numbers — 24, 27, 30, 33, 36 — map to Ra (arithmetic mean roughness): 1.6 μm, 2.5 μm, 4 μm, 6.3 μm, 10 μm. Those five figures are public. But Ra is an average, not the depth of the grooves. Draft calculation needs peak-to-valley depth, and the working conversion is 4–6 times Ra.

Get that factor wrong and the arithmetic is off by five. A client once read Ra 4 μm as a depth of 0.004 mm. The formula returned 1°, the number went onto the drawing, the texture shop did not flag it, and it went into the tool. At T0 all four side walls came out scuffed. Adding angle and re-texturing the cavity took three weeks. Texture shops quote Ra because that is the convention of their trade; the moment you feed that number into a draft calculation, you have to convert it.

The VDI 3400 table we keep on the wall

VDI 3400 grade	Ra (published standard)	Practical depth	Minimum draft
#24	1.6 μm	0.006–0.010 mm	1.5–2°
#27	2.5 μm	0.010–0.015 mm	2–2.5°
#30	4 μm	0.016–0.024 mm	2.5–3°
#33	6.3 μm	0.026–0.038 mm	3–3.5°
#36	10 μm	0.040–0.060 mm	3.5–4.5°

The Ra column comes from VDI 3400 itself. The depth column is the 4–6× conversion — it drifts with the process (chemical etch, laser, EDM), so confirm the real depth with the texture shop before the cavity is cut. The draft column already carries about half a degree of margin on top of the formula, for the reasons in the next section.

What the formula does not cover

The formula only handles geometric interference in the direction perpendicular to the grain. Three other things cause the same scuffing, and none of them is in the equation.

Shrinkage grip. As plastic cools it shrinks toward the cavity centre, which clamps the part onto the core. That adds normal pressure between side wall and cavity insert, and friction rises with it. Crystalline resins (PP, PA, POM) shrink more than amorphous ones (ABS, PC): at the same 2°, a PP part feels noticeably heavier to eject. On crystalline enclosures we add 0.5–1° on top of the formula value.

Texture depth is not uniform. In chemical etching, deep pockets and corners exchange etchant poorly, so the texture there comes out shallower or patchy. Laser texturing drifts in energy density where curvature changes sharply. The same side wall therefore has different depths at different points — calculate from the deepest point, not the average.

Ejection balance. Even with enough angle, too few ejector pins or badly placed ones let the part tilt as it leaves the core, and a local side wall becomes a negative angle scraping hard against the texture. We learned this on a walkie-talkie housing: the wall had 3°, and one side still came out with band-shaped drag marks. Two extra flat ejector pins and a rebalanced ejection fixed it. The angle was never the problem.

So when a textured part sticks, the angle is only one of four suspects: angle, shrinkage, depth variation, ejection.

When there is no room for angle: three fallbacks

Appearance parts regularly hit a dead end: wall height is fixed, the envelope is fixed, and every extra degree of draft eats wall thickness or overall size. At that point we work down a priority list.

① Add local stock, 0.3–0.5 mm. Reserve material outside the textured face and trade thickness for angle; after ejection it sits where assembly hides it. The cost is uneven wall thickness and a sink risk. Do not do this on transparent or high-gloss parts.

② Move the parting line so the pull direction follows the grain. A lot of "scuffing" is really pull direction fighting grain direction. Texture is etched evenly over the cavity surface, with grooves running roughly perpendicular to the surface; if the parting line makes one face a wipe direction, no amount of angle will save it. Moving the parting line is more effective than adding angle — the price is a new tool structure.

③ Drop the grade, or remove texture locally. Going from #33 to #27 cuts the depth by more than half, and the draft requirement falls from the mid-3° range to the low 2s. At normal viewing distance the appearance difference is not obvious, while the process window widens a lot. We did exactly this on a small-appliance panel, and the client accepted it after seeing samples.

Steel hardness before texturing, and the order of operations

One item gets skipped more than any other: the texture is only as good as the steel under it. Pre-hardened P20 (around HRC30) etches inconsistently, with depth varying across the surface. For production tools we ask for heat treatment to HRC48 or above before texturing. Low hardness also wears: a few hundred shots and the grain polishes bright, so appearance consistency will not survive the production run. Confirm this with the tool shop at design stage — discovering at T0 that the texture looks washed out is a far longer recovery than changing an angle.

Short-life, low-volume parts can sometimes get away with pre-hardened steel plus laser texturing, but agree that with both shops before anyone cuts steel.

The order of operations. Fix the grade → ask the texture shop for the actual depth, not the Ra → run the formula for the base angle → add margin for shrinkage and parting direction → check ejection balance last. Reverse the order and the drawing bounces back from the texture shop.

FAQ

How hard does tool steel need to be before texturing?

HRC48 or above for chemical texturing. Pre-hardened steel such as P20, which sits around HRC30, etches unevenly and is not something we would use directly on a production tool. Below the hardness threshold the texture also wears away — a few hundred shots and the grain polishes bright, and appearance consistency cannot be held across a production run. Short-life, low-volume parts can sometimes use pre-hardened steel with laser texturing, but agree it with both shops first — steel selection belongs in the design stage, not in a T0 correction.

The wall already has 2° and it still scuffs. What do I check next?

First have the texture shop measure the actual depth (not the Ra) and confirm it is inside the range the formula assumed. Then check ejection balance, then whether the resin shrinks more than expected, then whether the parting direction conflicts with the grain direction. Those four checks find the cause in more than 80% of cases. Our own 2° case turned out to be too few ejector pins — the part tilted on ejection and a local wall briefly went negative and scraped the texture. Two extra flat pins solved it; the angle never changed. If the scuffing concentrates in one corner it is usually ejection or local depth; if the whole face scuffs evenly, the angle itself is short.

Do VDI, SPI and Mold-Tech convert to each other?

There is no official conversion table. VDI 3400 publishes Ra values, SPI grades by processing method (A1 to D3), and Mold-Tech is a vendor's own texture library. When crossing systems, ask the texture shop for both the grade number and the actual depth, not the grade alone. We have seen the same grade come out at noticeably different depths at different shops, so "VDI 30" on a drawing is not enough — add a permitted depth tolerance and have the shop sign off a sample. That signature removes a lot of later argument.

Is there a relationship between texture depth and wall thickness?

Keep texture depth clearly below wall thickness — under about 1/10 of it. At 1.5 mm wall thickness, be careful with #33 and coarser grades. Too deep a texture on too thin a wall produces show-through (internal structure visible as shadows on the outer face) and a local loss of strength, and it shows up worst on light-coloured resins and thin walls. On a remote-control top cover we had 1.2 mm wall with #33, and the rib positions were faintly visible on the cosmetic face; we added local thickness and dropped the grade. Check wall thickness and texture grade together at design stage, not at sampling.

Originally published at https://www.hezidesign.com/gongyi/1228.html

Rams' Ten Principles: Three Rules No Longer Survive 2026

Asher Hu — Mon, 19 Oct 2026 10:23:56 +0000

Start with the verdict. Three of Dieter Rams' ten principles cannot be carried over unchanged into 2026: rule 7 (good design is long-lasting), rule 5 (good design is unobtrusive) and rule 10 (good design is as little design as possible). Two have become more important than they were in Rams' day: rule 4 (good design is understandable) and rule 6 (good design is honest). And all ten describe whether a product is good after it is made — none of them says whether it can be repaired. So there is a rule 11 to add: good design is repairable.

The short version: dimmable hardware, an app, and a cloud promise are three different lifetimes. A small appliance we designed for export to Europe went into a structural review with the client's legal counsel in the room, and the conversation was not about the form — it was about the spare-parts obligation. That is when I re-read the ten principles and noticed the premise underneath all of them: powered, standalone, long-lived, offline, with no firmware updates.

First, the origin: it is not an industry standard

The ten principles were not issued in 1976 as a standard. Rams distilled them in the late 1970s while reflecting on his own work; his stated motivation was a world that had become "an impenetrable confusion of forms, colours and noises", and his question was "is my design good design?" — a designer's self-audit, not a rulebook for an industry.

The authoritative source is the Vitsœ website; Rams worked with the British furniture company from 1959. The objects he had in mind were specific: Braun small appliances (radios, calculators, juicers, record players) and Vitsœ furniture. All of them are powered, standalone, long-lived, offline and free of firmware updates. That premise is what the three failures grow out of.

Failure 1 — rule 7, long-lasting: true physically, false commercially

Rams' own catalogue is full of discontinued models. The ET 66 calculator went out of production in 1987; the T 1000 world receiver in 1963. Both are still chased in the second-hand market at healthy prices. Long-lasting holds on the physical level and fails on the commercial one — a manufacturer retires the model to make you buy the next one. Design all you like; the schedule says end-of-life.

In 2026 the lifetime of a product is not decided by its form. It is decided by firmware support, by whether the battery can be replaced, and by when the cloud service shuts down. The working formula: service life = min(physical life, firmware support life, service availability life). Whichever is shortest wins. Braun products had no firmware and no service, so rule 7 tested the designer alone. Today it tests legal and the CFO's roadmap.

One real case: a smart plug whose structure was designed for eight years, while the client would only commit to three years of cloud service. In year four the app came down; the plug still had power, but scheduling and statistics were gone. From the user's side, the product was broken. The fix is not an aesthetic attitude but three requirements written into the spec: replaceable battery, usable offline, exportable data.

Failure 2 — rule 5, unobtrusive: the firmware puts it back

On screen products, 2026 does the opposite of restraint: a five-second brand animation at boot, a breathing LED in standby, a confirmation sound on every interaction. In the appearance review we did achieve restraint — a clean panel, no ornament. Then the prototype reached the software team and restraint was added back, one line at a time.

This is not the designer's call. Rams had no retention or daily-active metrics to satisfy, and Braun did not earn money from time spent. The test is blunt: if a product's revenue model depends on user time, it cannot be unobtrusive. That is a business-model problem, not a design problem. What the industrial designer can do is treat the boot animation as a form element and fight it in the review — not complain after the software schedule is frozen.

Failure 3 — rule 10, as little design as possible: the wrong thing is being removed

Rams meant less visual noise. His products had eight knobs, but the knobs were organised: one function each, fixed positions, positional memory, reachable and operable without looking. Today "less" usually means swapping physical keys for a display. Visually it is cleaner; in use, the reachable entry points are what got removed, and the interaction path got deeper.

Rams reduced ornament; much of the industry now reduces reachable controls. A physical knob has positional memory and can be operated blind — your hand goes there without your eyes. A menu hierarchy has no positional memory: wake the screen, find the entry, go two levels down, every time. Eight knobs to one screen is less to look at and more to think about.

We built a car-mounted device: the first version was all touch, and in testing the client said they did not dare operate it while driving. The second version brought the volume and one switch back as physical knobs and left settings on the screen. It grew two bumps and someone in the review called it less clean, but the test pass rate went up clearly. So rule 10 needs a qualifier today: reduce interference, not entry points.

Rule by rule

Verdict	Principles	Why
Needs rewriting	5 unobtrusive, 7 long-lasting, 10 as little as possible	Premises changed: revenue model, firmware and service life, ornament and entry points conflated
More important	4 understandable, 6 honest, 9 environmentally friendly	Complexity is up, AI claims and inflated specs, EU compliance is already on the schedule
Still valid	1 innovative, 2 useful, 3 aesthetic, 8 thorough	Bound to technology, the three-layer floor, criteria diverge by category, precision only rises

The three that still hold should not be thrown out with the rest. Rule 4 has a low-tech test that works: take the manual away, put the product in front of someone who was not on the project, and see whether they can complete the main task. We now run that in every internal review; it is more informative than looking at renders.

Rule 6 is violated most often by AI exaggeration — calling template matching "AI" — and by inflated specs, such as quoting peak charging power instead of sustained. Designers tend to think this is not their business, but the images and copy on the product page are part of the design delivery.

The rule that should be added: rule 11, good design is repairable

All ten principles judge whether a product is good after it is made. None asks whether it can be repaired. In the late 1970s that was not a gap — repairing a radio was normal. Today the gap has widened.

I would add: good design is repairable. This is not a moral appeal; it is a compliance cost. The EU is pushing a repairability index and spare-parts availability requirements, and any Chinese manufacturer exporting to Europe has to meet them. On export projects our structural review input list now includes a spare-parts list and a disassembly path as fixed items.

Dimension	Executable standard	Design action
Battery	Non-destructive access, regular Phillips or Torx, no proprietary jig	Independent battery hatch, not buried under the mainboard
Wear parts	Buttons, ports and seals individually replaceable	Separate small parts, not double-shot integrated modules
Screws	No more than two specifications across the product	Unify at structural review, cut the tool count
Disassembly	Public service manual; main modules within 12 steps	Measure and record step count at T0 prototype
Firmware	At least two years' notice before end of support, plus offline usability	Core functions do not depend on the cloud

The last one is the hardest. It sits outside the designer's control, yet it decides whether the first four matter — a repairable product with a dead cloud is still a dead product. We raise it separately in the concept presentation and ask the client's technical lead for an answer on the spot, rather than discovering it after tooling.

How to use the ten principles now

Read them as a list of constraints that have changed, not as a standard to copy. After each one, ask whether its premise still holds for this category in 2026. If it does, follow it; if not, rewrite it as a quantified requirement.

In practice: at concept stage we go through all ten, mark the ones that do not hold for the category, turn those into quantified items in the requirements document — replaceable battery, offline usability, disassembly step count, screw specifications — and only then move into form development. It costs very little time and keeps legal and software schedules from overturning the design later.

FAQ

Were the principles issued in 1976?

No. Rams distilled them in the late 1970s while reflecting on his own work; they were never a one-off 1976 industry standard. The authoritative source is Vitsœ, where he has consulted since 1959. Understanding the original objects — powered Braun appliances and Vitsœ furniture, offline, no firmware — is what lets you judge which rules still apply.

Why is rule 7 the hardest to keep?

Because service life is now the shortest of three clocks: physical life, firmware support life and service availability life. Rams' products had no firmware and no cloud, so rule 7 tested the designer alone. ET 66 (1987) and T 1000 (1963) were discontinued and are still collected, which shows long-lasting holds physically and fails commercially. The countermeasures are battery replaceability, offline usability and data export.

What is wrong with rule 10?

It conflates reducing ornament with reducing entry points. Rams reduced visual noise while keeping eight logically arranged, positionally memorised knobs that can be operated blind. Swapping them for one screen looks like less and costs more in cognition, because a menu has no positional memory. Our car-mounted device had to bring the volume knob back after all-touch testing failed.

Why is repairability a compliance cost?

Because the EU is advancing a repairability index and spare-parts availability rules; without them a product cannot be placed on that market. That makes it a schedule and cost issue. The executable standards — non-destructive battery access, individually replaceable wear parts, at most two screw types, modules within 12 steps, two years' notice on firmware — change the structure, so they belong in the review as early as possible.

Originally published at hezidesign.com

The MIM Break-Even Point: One Division Sum Tells You the Order Volume

Asher Hu — Mon, 19 Oct 2026 01:02:37 +0000

We work the MIM break-even out in one division: break-even quantity = (MIM tooling − CNC programming and fixtures) ÷ (CNC unit cost − MIM unit cost), then multiply by 1.5 to 2 for the order we quote. On the small parts we have taken through it, the landing point sits between a few thousand pieces and twenty or thirty thousand: below that the tooling cannot be spread, above several hundred thousand pieces MIM steps aside for powder pressing and cold heading. Cavity count and shape complexity move the line; quantity is only the result.

The rule of thumb: for parts of a few tens of grams or less, the economic batch usually sits at 5,000 to 30,000 pieces per order, placed by tooling cost divided by the unit-cost gap. Under 3,000 pieces tooling amortisation wins for CNC or stamping; over 500,000 pieces powder pressing and cold heading are cheaper per part.

1. Split the cost structures before comparing unit prices

The commonest quoting mistake is setting MIM's unit price against CNC's directly; they are not comparable. MIM pays up front for the injection mould, the setters and fixtures for debinding and sintering, and the sizing mould — small-part tooling usually runs from several tens of thousands of yuan to the low hundreds of thousands. Feedstock costs more than bar stock of the same alloy, and unit cost carries injection labour, sintering charged per furnace, gate removal and sizing.

CNC's fixed investment is far lower — programming plus fixtures, a few thousand to just over ten thousand yuan — but its unit price hides material utilisation: on small 316L or titanium parts, removal rates of 60 to 70% are common, and the cut-away is pure loss. MIM is near net shape, so utilisation is far better.

An 8g 316L part with MIM tooling at 80,000 yuan and unit cost 3.5 yuan, against CNC fixtures at 3,000 yuan and unit 18 yuan, gives (80,000 − 3,000) ÷ (18 − 3.5) ≈ 5,300 pieces. That is not the number to quote. Multiply by 1.5 to 2 — the T0 trial usually needs two rounds of tool change, yield-ramp scrap is yours, the sintering batch must fill a furnace, the surplus sits in inventory — and the honest MOQ is 8,000 to 10,000, because the supplier runs its own furnace and rework rate.

2. Cavity count is the only lever you can turn yourself

The same mould in 4 cavities and in 16 is a tier up in tooling and a tier down in unit cost; on small parts 32 cavities is not unusual. More parts per shot spread injection labour and machine time thinner, so unit cost falls. Above tens of thousands a year we recommend multi-cavity, since tooling is one-off; a first batch testing the market is better on a low-cavity mould, dearer per part, than dropping the whole tooling bill at once.

Alongside cavity count sits a variable rarely printed on a quotation: furnace loading. Debinding and sintering are charged per furnace, and how many parts fit depends on weight and tray layout. A 5g part fills a tray at low energy per part; past 100g the furnace takes few, the amortisation climbs and the advantage decays.

3. The more complex the shape, the earlier the break-even

Shape decides the result. Angled holes, profiled slots, internal cavities and multi-directional holes mean re-clamping, dedicated fixtures and slow passes on CNC, while on MIM they grow out of the mould at almost no extra cost.

Part shape	CNC unit labour time	MIM break-even scale	Verdict
Simple turned or 2.5D part, a few through holes	Short	Tens of thousands	Cold heading or stamping first; MIM is no advantage
Medium complexity: angled holes, profiled slots, local curved surfaces	Medium	The 10,000-piece class	MIM's main battleground; negotiable around 10,000
High complexity: multi-directional holes, internal cavities, surfaces hard to clamp	Long	A few thousand	Worth it at a few thousand if the design demoulds

The upper limit is the same: above several hundred thousand pieces a simple part goes to conventional powder metallurgy pressing, one cost tier below MIM and limited to axial-demould 2.5D shapes, or to cold heading when axisymmetric. MIM sits in between — shape freedom near precision casting, cost near pressing.

The limits that flip MIM from worthwhile to not, from drawings we have changed: wall thickness jumps crack during debinding because thick regions shrink more, so hollow the thick area or add a transition and re-check strength; thin long through-holes bend the core pin, so keep the length-to-diameter ratio within 3; sintering shrinkage runs in the teens up to 20% with general tolerance at ±0.3%, so a critical fit leaves stock or takes a sizing step costed into unit cost.

4. After the division, do three things first

List part weight, maximum wall thickness, minimum hole diameter and whether there are multi-directional features — those four numbers decide whether MIM can do the part and what it costs. Ask for a quote stating cavity count, not one blended unit price; 4-cavity and 16-cavity quotes on the same mould show the real break-even. Then multiply by 1.5 to 2 before quoting, leaving room for T0 tool changes, yield ramp and inventory. Working the sum out early beats reworking the structure later.

FAQ

Q: How many parts before MIM is cheaper than CNC?

There is no fixed quantity; divide the tooling gap by the unit-cost gap. A complex small part pays off at a few thousand pieces, a simple part often needs tens of thousands, and the result is then multiplied by 1.5 to 2.

Q: How large a part can MIM make? Is there a weight ceiling?

A few tens of grams and under is most economical; past 100g per part the advantage visibly decays, because debinding and sintering are charged per furnace and a heavy part lets the furnace hold few. Heavier parts are possible, but the economics fall away: heavy plus modest volume usually gives a better account on CNC or precision casting.

Q: How many cavities should an MIM mould have?

Steady annual volume in the tens of thousands and above can go multi-cavity, with 8 to 16 cavities common on small parts; a first product still proving its market is safer low-cavity. Tooling is one-off and fades over tens of thousands of parts, while a low-cavity mould is cheap to tool and dear per part. The worst move is a high-cavity mould before volume is settled.

Q: What tolerance can MIM parts hold?

Sintering carries linear shrinkage in the teens up to 20%, with general tolerance held to ±0.3%; a critical fit either leaves machining stock or takes a sizing step whose cost belongs in unit cost. Tighter requirements are met by a MIM blank finished on one or two faces by CNC. Judge at design stage which dimensions really need precision.

Originally published at hezidesign.com

Samsung's First Smart Glasses at 41g: Where the Weight Has to Go

Asher Hu — Sun, 18 Oct 2026 04:24:29 +0000

Start with the conclusion: 41 grams is only the entry ticket. What decides whether a pair of smart glasses can be worn all day is where those grams sit — how much loads the nose bridge, how much sits behind the ear. Two pairs at the same total weight can differ between a full day and twenty minutes; the difference is a weight budget and a hinge that does not eat temple volume.

The confirmed facts: Samsung's first smart glasses have surfaced at roughly 41 grams, on the Snapdragon AR1 Gen1 platform, expected next month. Public information gives no battery capacity, runtime, lens optics or ingress rating. (Source: IT Home.)

What the news actually says

Samsung has never shipped consumer smart glasses before; this is the first. Three things are confirmed: about 41 grams, Snapdragon AR1 Gen1, launch next month.

Weight is a spec nobody reads, until it sits on a face. Glasses rest at exactly two points — the nose bridge and behind the ears — and people are sensitive to a few tens of grams there. Two 50-gram pairs can end very differently: one lasts a full day, the other comes off after twenty minutes.

The design point: distribution and hinge routing decide whether smart glasses can be worn like normal glasses. Total weight is the ticket; where the centre of gravity sits is the real difference.

Where the 41 grams go

Conventional acetate frames sit at 20–30 grams. Everything a smart pair adds is battery, mainboard and camera modules, and none can go just anywhere.

The battery is the heaviest block, normally in the rear half of the temple, using the area behind the ear as counterweight to pull the centre of gravity off the nose. Temple width is limited: move the cell a few millimetres forward and the nose-pad load changes; move it back and the clamp behind the ear feels different. Our first step is a weight budget table — a ceiling and allowed shift range per block — handed to the mechanical team before layout. Adding counterweight afterwards leaves almost no room.

One point that gets missed: temple clamping force and centre of gravity must be tuned together. A tighter temple resists slipping but presses on the side of the head for hours. Test both in the same prototype round; adjusting one alone pulls the other off.

The hinge: routing a flex cable and surviving the cycles

The hinge is harder than on normal glasses. A conventional hinge only opens and closes. A smart hinge also passes an FPC from the front frame into the temple, linking camera, sensors and mainboard. That flex rotates with the hinge and has a hard limit on bend radius — too tight, and it fails once the cycles climb.

So the hinge envelope is dictated backwards by routing space. If ID draws temples at normal-eyewear proportions, structural review has to give the FPC room, which means a thicker temple. The routing scheme should exist before ID is frozen, not be requested during structure.

Cycle life for folding structures is normally validated over tens of thousands of open-close cycles, with the actual figure set by where the product sits in the market. On a blood-pressure monitor project, a battery-door latch was off by a fraction of a millimetre on engagement, so the feel went loose then tight, and we changed it three times before it was right.

At the tightest wrap the flex folds through a radius measured in tenths of a millimetre; 0.5mm of extra clearance there often separates an off-the-shelf FPC from a custom one.

Skin contact: heat and CMF are the same problem

The Snapdragon AR1 Gen1 class of AR platform draws more power when shooting and computing continuously, and the heat path runs through the inside of the temple — the skin at the side of the head. Thermal design here is not only a performance question.

The usual treatment routes heat outward and along the temple's top edge, with a low-conductivity material or an air gap at the contact surface so touch and temperature stay acceptable. Wall thickness and the bracket's thermal path must be calculated together; shell material alone will not solve the heat.

Surface finish on the contact face needs its own evaluation. Matte soft-touch paint feels good but resists sebum and sweat poorly; over months it turns shiny and tacky, which on daily wear is very visible. Bead-blasted PC or a UV topcoat holds up better. Sweat is acidic, so validation should include artificial-sweat immersion and wipe tests, not a lab-fresh sample.

The design sequence: weight distribution and centre of gravity first, hinge routing second, CMF last. Reverse that order and every later step patches the one before.

What designers should take away

Given a weight target, build the budget table first and write the ceiling for every block. Centre of gravity is an acceptance criterion measured alongside total weight, not a feeling to discuss at the end.

On any structure with relative motion, fix the routing scheme before the outer form is frozen. Flex bend radius, hinge rotation range and FPC clearance are three parameters that move together — change one and re-check the other two.

Treat heat and touch on skin-contact surfaces as one problem. Plan the thermal path, then choose the surface process, then validate against real sweat and sebum. Skip a step and mass production sends the bill back.

FAQ

Is 41 grams light for smart glasses?

Yes — for a pair with a camera it is on the lighter side, but how it feels depends on centre of gravity and temple clamping force. 41 grams is split between the nose bridge and behind the ears. At the same weight, a battery set further back loads the ear side and relieves the nose; set forward, the nose pads leave marks after a long day. Gram count alone cannot answer whether a pair is comfortable.

What is the hardest part of smart-glasses structural design?

The packing inside the temple and the hinge routing — those two decide what the outside can even look like. Battery, mainboard and antenna compete for temple volume, and the FPC still needs bend space through the hinge. Stack those constraints and the minimum temple cross-section is reverse-engineered by the structure, leaving ID little freedom. For a normal-eyewear silhouette, bring structure in during concept.

What should smart-glasses industrial design watch out for?

Frame thickness and temple diameter: keep them close enough to everyday eyewear that the product does not stand out. Glasses are worn in social settings, and an overtly tech shape invites questions at the office or across a dinner table. Temple diameter is the most direct visible signal of internal packing; bringing it near regular frame levels changes acceptance noticeably.

Originally published at hezidesign.com

Grip Ergonomics: From an 82mm Hand Width to the Right Diameter

Asher Hu — Sun, 18 Oct 2026 03:51:24 +0000

Start with the answer. Take the 50th-percentile adult male hand width from GB/T 10000, the Chinese national anthropometric standard: 82mm. The equivalent grip diameter starts at hand width × 0.4, which gives 33mm; allow 3mm up and 2mm down and you land at 31–36mm. A female hand at 76mm follows the same arithmetic to 28–33mm. For a product that serves both, take the overlap: 30–33mm. Then three more decisions finish the grip — ellipse ratio, grip-section length and finger-groove pitch.

The four numbers: diameter 0.38–0.45 × hand width, about 31–37mm for adult men; cross-section an ellipse at 1.2–1.35; effective grip length not below hand width, in practice 90–110mm; finger-groove pitch 20–24mm.

1. Diameter is how far the jaw can open

Early on we drew handles by copying competitors: measured 32mm, wrote 32mm, and could not explain why clients found them tiring. The number maps to how wide the web between thumb and index finger can open. For an 82mm hand closing four fingers around a cylinder, the palm wrap is roughly 100–115mm, an equivalent diameter of 32–36mm. Under 30mm the fingertips reach the palm first and the jaw is forced tight; over 38mm the little and ring fingers cannot close, and ten minutes makes the user switch hands. So 33mm is a comfortable start, 36mm still holds, 40mm only suits large male hands.

Hand tools we hold to 32–35mm. Cups go to 65–80mm for volume, but that is a ring hold judged differently: whether thumb and index finger close a circle. If they cannot, the fingers press and the feel drops. One allowance: a surface looks thicker in the hand than on the drawing, so on clay we pull the diameter in 1–2mm before judging.

2. Not a true circle: 1.2–1.35 seats in the palm

A true circle fails on anti-rotation and load, not looks. The palm touches at a point, pressure concentrates and the palm goes numb, and there is no orientation cue. An ellipse at 1.2–1.35 gives the palm an area; a 33mm circle pressed into a 36×28mm oval is 1.29, and the meat of the palm lies on the curve.

Direction can be wrong. The major axis must follow the jaw opening — the axis between thumb and index finger. Across it, users rotate the wrist to find the angle every time. At sketch stage we mark the axis angle to the product front at 0°–15°; past 15° the visual "straight" reads crooked and the grip deforms.

3. Grip length and pitch: the first cuts in styling

In a styling review these two go first, because they add nothing to the look and only guarantee the floor. A long grip looks clumsy on the drawing and grooves make the surface busy; cut the grooves, then the length, and the product ends up handsome with the palm in air.

The floor for grip length is hand width, 82mm. Below it the heel of the palm has nothing to land on and the fingertips hook with extra force, so stability drops a grade. We build 90–110mm: the surplus gives the thumb room and gives the eye one continuous grip surface. Pitch runs 20–24mm, matching the proximal width of an adult's four fingers; the fillet must not go below R6mm, and depth stays 1.5–3mm. Past 3mm the fingers lock into the slots and every change of grip needs re-aiming.

4. Edge radius, texture and paint change the feel in production

Every edge in the grip zone should be R3mm or more; below that, texturing pulls the grain and the edge feels sharp. Texture follows VDI 3400, with grip zones usually at VDI 27–33 — clear grain without abrading the hand. Do not cut deep texture where the radius is under 1mm: the mould has its first problem there at ejection and the part arrives with a burr. Soft-touch paint at 30–60μm visually dulls the radius, so draw it duller at render stage.

5. The trade-off order: agree what gives way

Do all four dimensions properly and the form turns honest — it will not carry sweeping surfaces. We protect three: diameter, ellipse ratio and grip length do not move; grooves are negotiated per styling at a shallow 1.5mm, almost invisible but still a cue in the hand. Which yields first must be an explicit order agreed in the review, or feel is sacrificed by default.

The comparison we put on the table:

Dimension	True circular section	Elliptical section	Ellipse + finger grooves
Palm contact	Point contact	Area contact	Area contact + finger location fixed
Orientation cue	None	Weak	Clear
Mould complexity	Low	Low	Medium, draft needs checking
Suitable products	Decorative handles	Drinkware, small handheld appliances	Tools and hand-held parts that take force

FAQ

How do I estimate grip diameter quickly?

Take 0.38–0.45 × hand width. An adult male at 82mm maps to 31–37mm, a female at 76mm to 28–33mm, and the shared band is 30–33mm; the figures come from GB/T 10000. We start at 0.4× (82mm gives 33mm) and float ±2–3mm: tools held with force go thinner, passive consumer products can go thicker. Hand width is the most direct single variable — better than height or weight.

Why is a true circular section not recommended?

A circle meets the palm at a point, pressure concentrates, the palm goes numb, and there is no cue which way is forward. An ellipse at 1.2–1.35 takes an area and steadies the grip. The major axis must sit on the jaw-opening axis at 0°–15° to the front; laid across, it forces wrist rotation. Circles suit decorative handles only.

What pitch and depth should finger grooves use?

Pitch 20–24mm for four adult fingers; depth 1.5–3mm with the groove floor at R6mm or more. Past 3mm the fingers are fixed in the slots and switching grip means re-aligning, which feels worse than smooth. For a cleaner surface we cut shallow at about 1.5mm — barely visible, still a cue.

Do texture and soft-touch paint change how the grip feels?

Yes. Grip zones use VDI 27–33: clear grain, no abrasion. Soft-touch paint at 30–60μm dulls the radius. Under a 1mm radius, deep texture tears at ejection and leaves a burr on the sharpest edge of the product. Draw a duller radius at render stage rather than carrying the bare model's crispness into production.

Originally published at hezidesign.com

Amazon's Bee Recorder: Wearable Form Factor Hits the Microphone First

Asher Hu — Sat, 17 Oct 2026 15:26:23 +0000

Amazon's Bee line now has two AI recorders: Frontier, a wearable, and Eclipse, a magnetic accessory for the back of a phone, from $74.99. The announcement gave form factor and price and nothing else — no dimensions, no microphone count, no runtime. What a structures person notices is one company answering one recording need with two completely different housings. The first debt comes due at the microphone aperture.

The core point: a wearable recorder is not a shrunken phone with a clip. The microphone must hear a voice while the device sits against a moving body, and the port that lets sound in is the port that lets sweat, lint and fabric noise in. The acoustic duct, the local wall thickness, the seal, the battery volume, the part count — all of it follows from where that port goes.

What the announcement actually told us

Frontier is a standalone device with its own power and way of being worn; Eclipse is a phone appendage held by magnets. From $74.99, both lines sit at consumer pricing, not professional recording gear. No specifications were published, so we can only read the news backwards from form, and what comes back is structural constraint, not specification.

1. On the wearable, the microphone fights the clothing first

The pickup opening cannot sit against skin: skin friction is low-frequency noise that post cannot remove. It cannot face down or inward either — one brush of fabric and you get a wash of rustling. That leaves the upper edge of the side wall, already claimed by buttons, LEDs and charging contacts. Whoever claims the spot first sets the housing.

Behind the opening runs a bent acoustic duct that lets sound through while keeping moisture and lint out. The duct eats internal volume, and a thin wall cannot make an effective bend, so the wall thickens locally — which means uneven wall thickness. We have run a power-bank housing at 2.0mm nominal wall where the T0 trial pulled a 0.3mm sink at the lip, and the dished area landed right beside the aperture. Effective port length changed with it and the sound came out noticeably muffled. On small housings, sink and aperture are the same problem.

Recording is a long task. Weight and volume stay low so people keep wearing it, which pins battery capacity; the mic array and wireless link are the hungry loads, so runtime is usually bought with a lower sample rate or intermittent wake-up.

2. The magnetic version pushes the pickup-distance bill to the algorithm

Eclipse on the back of a phone makes alignment easy. The phone's magnet array is a ready-made datum; the accessory mirrors it with a ring magnet plus positioning magnets, snaps into centre, and needs no extra guide features. The space that saves can go to battery or speaker.

The trouble moves down the chain: the microphone sits far from the mouth, so voice and room arrive together at similar levels. The housing cannot fix that — only separation in software can. What the structure can do is avoid making it worse: keep apertures out of the usual grip zone and off the desk contact face, because a phone lying flat seals that hole.

Magnet burial depth is the other unavoidable trade. A magnet must sit close to the outer surface to hold, and thinner plastic over it pulls harder — but too thin and the mould short-shots around the insert, and a drop can walk the magnet out. On this kind of part we press the magnet in with interference or add a dot of adhesive: one more step for a lower escape rate. At $74.99 that step has to be costed.

3. The $74.99 line frames the structure before a line is drawn

Retail price decides part count and process. At this band, housings are plastic injection moulding with a few decorative parts; CNC aluminium is reserved for higher tiers. A small-part tooling set — front housing, rear housing, buttons, decorative ring — runs into the tens of thousands, and that bill belongs on the table before sign-off.

The structural load on the two products is visibly different:

Item	Wearable Frontier	Magnetic Eclipse
Pickup position	Close to the body, disturbed by clothing friction	Far from the mouth, collects more room noise
Wear duration	Limited by weight and volume	Goes wherever the phone goes
Charging	Needs a dedicated dock or contacts	Not stated whether it can share phone power
Housing part count	High, including wearing parts and contact structures	Low, mainly housing plus magnet array
Sealing focus	Sweat, sebum	Pocket lint, dust
Main failure points	Wearing-part fracture, contact corrosion	Magnet escape, magnet strength decay

Not one cell can be filled with a number, because the announcement published none; any "exact value" beyond the public specs would be us inventing specifications on the client's behalf.

4. What designers can take away

Decide how the person carries it, then draw the outside. Form decisions flow backwards into microphone position, battery volume and sealing class; appearance is the result, not the starting point. Change form after the drawing is fixed and tooling cost and schedule burn together.

Claim the acoustic opening in the first structure revision. Even with the appearance open, lock the aperture and duct, or a later surface change shortens the duct and pushes the port onto a corner radius.

Treat magnet burial depth and wall thickness as one decision. Holding force is capped by the plastic over the magnet, and that thickness is capped by the moulding process. Discuss both on the same drawing; set them apart and you rework.

Questions we get most often

What is the biggest design difference between a wearable recorder and a magnetic phone accessory?

The pickup environment and the structural burden differ completely. The wearable handles friction noise from skin and clothing and reserves structure for wearing parts and separate charging contacts; the magnetic part borrows the phone's magnet array for alignment and uses fewer parts, but the pickup distance is long, room noise has to be recovered in software, and its thickness still affects hand feel and heat.

Where do recording products usually put the microphone port?

On the upper edge of the side wall first, clear of the finger grip zone and the desk contact face. Inside the port sits a bent acoustic duct for moisture and lint, and it needs internal volume — so a thin wall thickens locally, and local thickening invites sink. Evaluate the two together.

What does a $74.99 price point mean for structural design?

The process route is essentially locked to plastic injection moulding with a few decorative parts. Every added step — adhesive dispensing, overmoulding, post-machining — has to be re-costed, and the structure balanced between tooling cost, assembly time and yield. You cannot pile on part count the way a high-tier programme does.

Originally published at hezidesign.com

Foamed Parts Shrink Worse Than Solid Ones: Which Mould Details to Change

Asher Hu — Fri, 16 Oct 2026 23:33:40 +0000

The conclusion first: when a foamed part sinks, nine times out of ten it is not about how much foaming agent you added. It is about three things in the mould that were never adjusted for foaming — mould temperature uniformity, venting, and gate size. On a PP microcellular foamed battery-pack lower housing we ran, two of four cavities showed sink at the rib positions at T0. The customer's first instinct was to cut the foaming agent. We measured nine points across the cavity instead: the gate side and the far end differed by 12°C, and every sink was at the far end. After reworking the water circuit, the same material at the same expansion ratio produced no sink at all.

The core point: solid parts solve shrinkage with packing; foamed parts solve it with temperature and pressure control. A foamed part's apparent shrinkage is smaller than the same material solid — typically 60%-80% of the solid value — but the local sink risk is higher, because the skin is only 0.3-0.8mm thick and there is nothing but bubbles underneath to hold it up.

Small overall shrinkage does not mean no local sink

These two things get discussed as if they were one. Overall dimensional change is the linear shrinkage of the whole part after cooling. Local sink is one small area of surface being pulled down by the shrinkage beneath it. In foamed parts, the two behave in opposite directions.

The reason is structural. A foamed part is a skin-core structure: a dense outer skin over a core of bubbles. The core has low density, less heat per unit volume, and lower crystallinity than solid material, so overall dimensional change really is smaller. But wherever wall thickness steps up and material accumulates, the core bubbles keep collapsing after packing ends, the thin skin cannot resist, and a dent gets pulled into the surface.

Item	Solid injection-moulded	Foamed
Shrinkage value	Material standard, PP commonly 1.5%-2.5%	60%-80% of the solid value, must be confirmed by trial
Surface skin	Dense throughout	0.3-0.8mm, weaker resistance to sink
Cavity temperature spread	±10°C usually acceptable	Tighten to ±5°C
Vent depth	Commonly 0.02-0.03mm	One step deeper, commonly 0.04-0.06mm
Cooling time	Baseline	Bubbles insulate; estimate 1.3-1.5× the solid part

Troubleshooting order: mould temperature → venting → packing

When something goes wrong on the floor, do not start by changing the material. Walk these three steps and you can usually localise it within half an hour.

Step one: measure mould temperature. Take a contact pyrometer to nine points across the cavity — front, back, left, right and centre. If the spread exceeds 5°C, the sink position will follow the cold side. Where the mould is cold, melt pressure decays fast at the flow end, the bubbles freeze before they can distribute evenly, and shrinkage concentrates there.

On water circuit layout, foamed moulds usually keep cooling channels 10-15mm from the cavity surface with 30-40mm spacing, and deep cavities or rib-dense areas need their own circuit. Our battery-pack lower housing had a single series circuit covering the rib area, so the far end was barely cooled; we added a dedicated in-and-out pair before the sink came under control.

Step two: look at venting. Foaming releases gas, so the amount of gas in the cavity is far greater than in solid moulding. Keep the vent at 0.02mm and the gas cannot escape, creating trapped gas at rib troughs and around boss columns — visible as local glossy or dull patches, or a whole section collapsing. Foamed mould vents go to 0.04-0.06mm with 3-5mm width, placed at the end of material flow and at the end of every rib trough. Deep vents flash easily, so parting-plane fit accuracy has to rise with them; the two changes belong together.

Step three: reduce packing pressure. This is the counter-intuitive one. Solid parts rely on high packing to compensate shrinkage, and 60MPa or more is normal for PP. Give a foamed part that much and the bubbles are crushed smaller, local density rises, and the part starts behaving like a solid one — shrinkage goes up and sink follows. Foamed parts usually take half the solid packing pressure or less, and the higher the expansion ratio, the lighter the packing.

On gates: point gate diameter is enlarged 20%-30% versus the same solid part — a small solid part running 1.0-1.2mm often starts from 1.5mm when foamed. Too small a gate means shear heating makes the material start foaming near the gate, so it is already unevenly distributed by the time it reaches the cavity.

What has to change in the drawing and the mould

A foamed part's drawing cannot be drawn from solid-part habits. These are the places we have revised repeatedly.

Rib thickness at 0.6-0.7× the main wall. Solid parts are usually pushed to 0.5-0.6×; foamed parts can relax slightly because the low-density core packs less material at the same thickness. The rib root must carry a radius of at least 0.5mm — it spreads stress and gives material a flow path.

Draft angle increased 0.5-1°. A foamed surface has micropores, so friction against the mould is higher than a solid part; too little draft scuffs the surface or tears the skin. Deep parts we usually take to 1.5° or more.

Keep ejector pins off thin-skin areas. Foamed skin has low strength, so pins on a large flat face will whiten or punch through. Put them on ribs, bosses and flanges where there is structure, and increase pin diameter a little.

Measure trial dimensions 24 hours after ejection. After ejection, gas slowly escapes from the core bubbles and dimensions keep moving; readings taken in the first few hours mean nothing. Shrinkage must be set from dimensions taken after 24 hours or longer, otherwise the mould you cut will produce yet another size after a week in the customer's warehouse.

Root-cause criteria: if the sink is in the same place every shot, it is a structural issue — change wall thickness and radii. If it drifts from shot to shot and comes and goes, it is mould temperature or the water circuit. If the same cavities fail every time in a four-cavity mould, it is runner balance. If the whole batch is uniformly undersized, the shrinkage value is wrong or post-shrinkage was not accounted for.

Questions we get most often

Q: What shrinkage should be used for a foamed part?
A: Usually 60%-80% of the shrinkage for the same material in solid form. PP solid is commonly 1.5%-2.5%, which puts a foamed part roughly in the 1%-1.8% range, and the trial data decides. Higher expansion ratio gives lower shrinkage, but the relationship is not linear — mould temperature, packing and wall thickness all matter. Run at least three expansion ratios from the same mould, measure after 24 hours, and only then fix the shrinkage for cutting steel.

Q: Where does sink usually appear on foamed parts?
A: Wall-thickness steps, rib roots, boss bases, the far end opposite the gate, and trapped-gas dead corners such as rib troughs. What they share is a lot of material underneath and a thin skin. The approach is to remove accumulated material first — bring rib thickness toward 0.6-0.7× the main wall, add radii at rib roots, then check whether the vent path at those positions is open. Only when the structure cannot be changed should you move to mould temperature and packing.

Q: Three steps feel like a lot. Which one first?
A: Mould temperature. It is the fastest to measure and the most common cause. Nine points with a contact pyrometer takes minutes, and if the spread exceeds 5°C you already have your answer in most cases. Venting comes second because it explains the localised glossy or dull patches that temperature does not account for. Packing pressure comes last because changing it affects density and mechanical behaviour, and it is the easiest to overshoot.

Q: We added foaming agent and it still sinks — is the material the problem?
A: Usually not. Measure the cavity temperature distribution first, then check vent depth and the packing setting; rule those three out before going back to formulation and foaming agent ratio. Packing is the most overlooked of the three: give a foamed part high packing and the bubbles are crushed, local density approaches that of a solid part, and the shrinkage behaviour follows — everyone is used to solid parts relying on high packing, so it is easy to go the wrong way here.

Q: How deep should venting be on a foamed mould?
A: Commonly 0.04-0.06mm, one step deeper than the 0.02-0.03mm typical of standard injection moulds, with 3-5mm width, placed at the end of flow and at the ends of rib troughs. Deeper vents flash, so parting-plane fit accuracy must be raised at the same time — the two changes go together. Vents also need periodic cleaning; foamed material leaves residue in the slots, and once they clog the sink comes back.

Originally published by the Hezi Industrial Design team, based on our industrial design and structural engineering practice in Dongguan. Learn more: Industrial design services

Rams' Ten Principles: Where Rule 5 and Rule 9 Collide

Asher Hu — Fri, 16 Oct 2026 14:51:01 +0000

Most versions of Dieter Rams' ten principles you find online talk about Braun and the 606 shelving system. Inside our studio we use them differently: as a checklist before a proposal goes out, where each principle gets one line answering "what exactly am I doing about this one?" The principle we cannot answer is usually the one that comes back as rework later. And of the ten, the two that genuinely fight each other are Rule 5, unobtrusive, and Rule 9, environmentally friendly.

The short answer: Rule 5 pushes toward fewer materials, fewer joints and continuous surfaces so the product does not shout. Rule 9 pushes toward separable materials, modularity and structural zoning so the product can be recycled. They collide head-on on one decision: does the housing become a single closed shell, or something that can be opened? Our practice is to put both rules in the same column of the proposal and reconcile them before anyone starts sketching.

Where Rule 5 and Rule 9 actually collide

Start with the specific point of contact. Rule 5, unobtrusive, translates into appearance decisions: less visible parting line, fewer material types, fewer decorative parts, continuous transitions from one face to the next. Rule 9, environmentally friendly, translates into structure: different materials must be separable, the battery must be removable, fastening must be reversible rather than glued, packaging must be reduced.

Those two sets of actions meet on one question: snap fits or screws.

An all-snap housing looks the cleanest. No exposed screw holes, faces that meet almost seamlessly — Rule 5 scores well. But an all-snap housing is hard to open: separating the plastic parts requires destructive disassembly, and if the shell is PC+ABS with a metal bracket and a battery inside, the recycling end cannot separate the materials. Rule 9 loses.

Go all-screw instead and the environmental side is easy to open, but the screw holes either show or need cover caps, and both leave a visual stop on the surface. Rule 5 loses.

Option	Appearance (Rule 5)	Disassembly (Rule 9)	Fits
All snap fits	Good, no exposed holes	Poor, destructive disassembly	Single material, no battery, whole-unit recycling at end of life
All screws with cover caps	Visible stopping points	Good, repeatedly serviceable	Contains battery, multiple materials, after-sales repair
Bottom snaps + hidden side screws	Good — the visible face stays clean	Medium — needs a tool but opens	Most small and mid-size consumer electronics

On a power bank housing we used the third option: the top face is one continuous surface, the snaps sit on the bottom, four screws are recessed into the four bottom corners, and the parting line is placed at the start of the R corner where it is invisible at normal grip angles.

A useful rule of thumb: give the face the user looks at most to Rule 5, and the face the user never turns over in a year to Rule 9. The two rules do not each have to be satisfied on every surface — they can be divided by surface.

Rule 10 gets misread as "a minimalist style"

The short answer: when Rams says less, he means fewer distracting elements. To check whether a subtraction was done correctly, look at Rule 4, understandable — did you cut that too? Take the manual away and see whether a new user can still operate the main functions.

The classic counter-example is collapsing five physical buttons into a single touch area, which then requires the user to memorise single tap, double tap, three-second hold, eight-second hold and power-cycle restart. The manual grows from two pages to eight. The appearance score goes up; understandability falls much further.

"Less, but better" is the most quoted line and the most misunderstood. It means fewer distractions, not fewer functions and less structure. The test for a subtraction is whether it took Rule 4 down with it.

We have seen a real case: a small appliance cut five physical keys down to one touch area in the name of a minimal interface. The result required five interaction patterns — single tap, double tap, three-second hold, eight-second hold, and a power-cycle reset — and the manual went from two pages to eight. The appearance got cleaner, but a first-time user had to read the manual. Rule 4 lost more than Rule 5 gained.

The test for Rule 4 is crude but accurate: take the manual away and have someone who has never used the product complete each main function once, with no questions allowed. If they cannot, the rule is not met.

Indicator lights are where domestic products fail this most often. A single-colour LED encodes five or six meanings through steady, slow blink, fast blink, double blink and breathing patterns, and the user has no choice but to memorise them. Switching to two colours plus different positions carries the same information without asking anyone to remember a rhythm — an understandability fix at the appearance level, with no circuit change.

One more: before cutting buttons, count the user's genuinely high-frequency operations. If there are more than three, collapsing them into one touch area is digging your own pit.

Rule 6 is a hard constraint: no fake parting lines

The short answer: honest, for a designer, is an actionable list: no fake parting lines, no purely decorative ventilation slots, no plated plastic pretending to be metal, no two-shot moulding imitating a real material joint. What these have in common is that the cost moves to after-sales: the user's expectations, inferred from appearance, are not met, and complaints and returns follow.

Fake ventilation slots we have seen more than once. A row of decorative grooves on a plastic shell that reads as a vent grille but is sealed. After a long session the machine gets hot, the user's first thought is "the vents are blocked", and out comes a needle. Puncture it and water gets in; the after-sales team rules it user damage and the user disagrees. A few cents of decoration buys an argument nobody can settle.

Plated plastic pretending to be metal is the other one. The gloss can get close, but weight and thermal feel give it away the moment it is picked up. Worse is what happens when the plating wears: the product looks three years old after three months, and the user feels cheated. If the product is meant to be plastic, finish it as plastic — it is cheaper too.

On a blood pressure monitor we reworked the battery compartment three times, and one of those rounds was because the flap used a plated part: in the prototype the finish did not sit well with the body, and the plating was at risk of wearing at the hinge. Switching to the same matte plastic as the body actually improved the visual consistency.

Compressing the ten principles into one proposal checklist

We keep the ten as a table and walk it before every proposal. The point is not to answer "yes" but to answer "specifically how, this time" — and "which one am I overdoing".

Principle	Ask yourself at proposal stage	Common way to overshoot
1 Innovative	What exactly changed versus the previous generation?	Changing form for novelty, so users cannot find the familiar operation
2 Useful	Does every feature map to a real function?	Piling on features until the main one is harder to use
3 Aesthetic	What is the basis for the proportions, surfaces and CMF?	Sacrificing grip for looks, or pure decoration
4 Understandable	With the manual gone, can a new user operate it?	Encoding too many meanings into one indicator light
5 Unobtrusive	Which elements can go without losing function?	Reducing until there is no visual focus and it reads as unfinished
6 Honest	Any decorative feature that misleads about performance?	Fake vents, fake parting lines, plated plastic as metal
7 Long-lasting	Will this form still hold in three years? Are wear parts replaceable?	Chasing trendy colours and shapes, obsolete in a year
8 Thorough	Are radii, chamfers, gaps and key feel held to one standard?	Gaps so small the process tolerances cannot hold them
9 Environmentally friendly	Can battery, metal and different plastics be separated?	Sacrificing surface continuity and assembly efficiency for disassembly
10 As little design as possible	Where is the line between removed and kept?	Cutting function as "noise" and making the manual thicker

Rule 8, thorough, also has an easy way to overshoot. On a beauty device we first pushed a part gap to 0.1mm; on the drawing it looked clean. Talking to the mould shop, the combined injection and assembly tolerances could not hold it — at T0 the gap jumped between 0.15 and 0.2mm, which looked worse than a uniform 0.25mm. A gap is not better the smaller it gets; it should be a stable value inside the tolerances you can actually hold.

One last note on using the list: the ten are not meant to get equal effort. We normally identify the two or three this particular product will struggle with, satisfy those properly, and hold the line on the rest. Chasing full marks on all ten makes the programme expensive and slow, and the first three often undermine each other. Rule 10 is more useful as a final check than as a slogan at the start.

Questions we get most often

Q: Which of the ten principles conflict most often?
A: Rule 5, unobtrusive, and Rule 9, environmentally friendly, conflict most frequently; Rule 10, as little design as possible, and Rule 4, understandable, also fight regularly. The first pair concentrates on how the housing is joined: all snap fits look clean but cannot be opened, and a multi-material product cannot be separated at the recycling end; all screws open easily, but counterbores and cover caps break the continuity of the visible surface. The second pair concentrates on subtracting from the operation path — cutting buttons and cutting explanatory text tend to happen together, and once the user has to memorise operating logic, the line has been crossed.

Q: Why is "as little design as possible" not the same as a minimalist style?
A: What Rams means by less is fewer visual distractions and less unnecessary decoration; the target is noise, not function or necessary controls. Minimalism is a stylistic direction. The two are not on the same plane. The test is simple: after the subtraction, look at Rule 4. If the manual got thicker, or a first-time user needs a tutorial, what you removed was functional visibility — the wrong thing to cut.

Q: How do you verify that a product is "understandable"?
A: Put the manual away, find someone who has never used the product, and have them complete each main function once without asking questions, while you only observe. If they cannot, the principle is not met. The test needs to cover high-frequency operations, not just powering on. Indicator-light encoding is a common failure: a single-colour LED expressing five or six states through steady, slow blink, fast blink and double blink forces memorisation. Switching to two colours with different positions usually fixes it without a circuit change.

Q: What exactly does "honest design" forbid?
A: Three main categories: purely decorative fake functional features, material disguise, and surface treatments that mislead about performance. Fake ventilation slots, fake parting lines and plated plastic standing in for metal all belong here. The cost lands mostly in after-sales. Fake vents make users think overheating is a blocked vent, and some will probe it with a needle; puncture it and water gets in, and assigning responsibility becomes an argument. Plated plastic wears and discolours, so the product feels three months old after three months of use, and the gap in expectation shows up directly in complaints.

Q: Do all ten have to score full marks in a proposal?
A: No, and we would not recommend it. In practice we identify the two or three principles this product will struggle with and satisfy those properly, then hold the line on the rest. Chasing full marks on all ten makes a programme expensive and slow. Our process is to lay the ten out as a table with "specifically how, this time" and "which one might be overdone" on each row, and walk it before the proposal. Rule 10 is best kept for the final check, to confirm the earlier subtraction did not cut into functional visibility — as an opening slogan it tends to turn into a plain stylistic instruction.

Originally published by the Hezi Industrial Design team, based on our industrial design project practice in Dongguan. Learn more: Industrial design services

Weld Lines Keep Landing on the Cosmetic Face? Check the Gate Position, Then the Wall Thickness

Asher Hu — Fri, 16 Oct 2026 00:31:14 +0000

When a weld line lands on a cosmetic surface, do two things before touching the mould: measure the meeting angle between the two melt fronts, and look at how far each front has travelled from its gate. If the angle is below 135°, the window for fixing it by adjusting mould temperature and injection speed is very narrow — the direction has to change at gate layout and wall-thickness balance. Our order of work on this class of problem is: move the gate if it can be moved, and only then go back to wall thickness, because changing wall thickness drags the entire cooling circuit and ejector layout with it. The cost is not on one drawing.

The core criterion: the position of a weld line is decided by gate position; how visible it is decided by the meeting angle and the temperature difference between the two fronts. Above 135° with a small temperature difference at the meeting point, the weld line can usually be faded until it is essentially invisible. Below 135°, even cranking mould temperature up will not rescue it — what needs to change is the gate layout.

Step one: work out whether the weld line is a head-on collision or a flow-around

Open the mould and look at where the weld line sits. There are two families of cause, and the fixes are completely different.

Meeting type: two gates feed two fronts that collide partway. The weld line sits on the line joining the two gates, usually a straight or gently curved line.

Flow-around type: a single gate feeds, and the melt flows around a core, an insert pin or a deep hole and then closes up behind it. The weld line appears on the back side of the hole or insert, and its shape follows the obstacle — one hole can produce two lines.

We had a blood-pressure monitor top cover with a single side gate. There was a ring of deep holes around the display window. When T0 came back there was a pale thin line to the right of the window, directly facing two snap-fit holes on the back. That is the classic flow-around case. The first instinct was more mould temperature: from 60°C to 80°C, several dozen shots. The line faded a little but never went away. We then moved each of the two snap holes by 6mm so they no longer sat on the same flow path, and the weld line dispersed into the non-cosmetic face.

Why does the distinction matter? The meeting type is solved by moving gates; the flow-around type is solved by moving holes and inserts. The directions are different. The test is simple: lay the weld line position over the structure on the back of the part. If it lines up with a hole, treat it as flow-around first.

Move the gate 10mm and the weld line follows — but the cost varies

The gate is the source of the weld line. Wherever the melt enters, that is where the front starts, and where the two fronts meet is the weld line. So on parts with cosmetic requirements, gate layout should be fixed at the structure review stage; changing it after T0 means reworking the mould cavity insert.

This table is the comparison we use internally when choosing a gate scheme; in a real project we rule options out in this order.

Dimension	Single gate	Two symmetric gates	Multi-point / valve gate
Weld line count	Possible at flow-around holes	Always 1	Increases with gate count
Position controllability	By avoiding with holes and inserts	Fixed between the two gates	Timing can be staggered to shift the meeting point
Injection pressure demand	Long flow path, high pressure	Medium	Low, but equipment demand is high
Mould rework cost	Change the sprue bush position	Rework runner + cavity	Rebuild the hot-runner system

One tip that saves a lot of effort: if the cosmetic face has a hole or a recess, place the gate on the long side of that hole so the melt fills around the hole before moving to the far end. The back of the hole is then much less likely to form a closed flow-around line. We did exactly this on a handheld device shell: the gate moved from the short side to the long side, and the weld line moved from directly above a key to the bottom near the gate, where a structural rib hides it.

Gate adjustment priority: first see whether the weld line can be moved onto a non-cosmetic face (bottom, side face, an area covered by ribs or printing); if it cannot be moved, consider adding gates to shorten the flow path and push the meeting point out of sight; only then consider hot-runner timing control to stagger the two fronts — that is a step up in both equipment and mould cost.

Wall thickness difference is the quiet push behind it

When the gate cannot be moved, the problem is often wall thickness. In the same part, thick regions fill faster and thin regions fill slower; the arrival time difference at the meeting point grows, and the weld line becomes visible.

We keep the ratio of main wall thickness to local wall thickness within 1.5:1. Above that, the velocity difference between flow fronts becomes visible to the eye, and the weld line goes white or shiny, sometimes with a slight sink.

We took a wrong turn once that is worth remembering. It was a small appliance top cover, ABS, main wall 2.0mm, with six internal ribs. At T0 there was a very obvious weld line right in the middle of the cosmetic face. The first assumption was injection parameters, and we tested two rounds: mould temperature from 70°C to 95°C, injection speed from slow to fast. Nothing. Then we sectioned the cover and measured: the rib root thickness was 1.6mm, almost 0.8 times the main wall. It was like standing six low walls across the flow path — the melt split at each rib and had to re-merge after it.

Reducing rib thickness to 1.2mm faded the weld line by roughly seventy percent, and mould texturing covered the rest. There is a precondition: the rib's job is stiffness, and thinning it loses stiffness, so we made up for it in rib height — 0.5mm taller — and recalculated the section moment of inertia.

There is one more easily missed item: the transition fillet between wall thicknesses. A direct step between two thickness regions makes the melt decelerate or even stall, and the weld line lands right there. We generally use a transition fillet of 0.5mm or more so the melt keeps moving through.

Wall thickness criteria: a main-to-local wall ratio above 1.5:1, or a rib root thickness above 0.7 times the main wall, will tend to leave an obvious weld line where the flow splits. The order of change is to take the rib thickness down to 0.5-0.6 times the main wall first, recover stiffness with rib height, and only then consider mould temperature and injection speed.

Half an hour on site to decide what to change

Trial time is tight. Work through this order and in half an hour you will usually know the direction:

Does the weld line line up with a hole or insert pin on the back? If yes, treat it as flow-around: move the hole, change the insert structure.
Does the weld line sit on the line between two gates? If yes, treat it as meeting type: evaluate moving or adding gates.
Measure the meeting angle between the two fronts. Below 135°, the direction is the gate; above 135° but still visible, look at wall thickness and ribs.
Measure the ratio of main wall thickness to rib root thickness. Above 0.7, change the rib first.
If none of the above, go back to injection parameters and look for the window.

This order runs from cheapest to most expensive. Parameter adjustments cost nothing; moving a gate means reworking the cavity insert; changing wall thickness pulls the cooling circuit along and is the biggest change. Do the cheap checks first — it saves a lot of mould rework money.

Questions we get most often

Q: Can a weld line be eliminated completely?
A: With a single gate and no flow-around obstacle, the part can be made essentially free of visible weld lines. With multiple gates or deep-hole inserts, the weld line can only be faded and relocated, not eliminated. In practice we do two things at once: move the weld line to a non-cosmetic face or hide it behind structural ribs, and cover what remains with mould texturing (depth of 0.02mm or more is our usual recommendation) or the direction of the grain. If the customer demands a perfectly clean cosmetic face, the fix has to come from part splitting and gate scheme redesign — not from downstream processing.

Q: Does raising mould temperature reduce the weld line?
A: It helps, but there is a ceiling. Higher mould temperature narrows the temperature difference between the two fronts, so the weld line fades. But when the meeting angle is below 135°, mould temperature alone will not rescue it. Raising mould temperature also has side effects: longer cooling, longer cycle time, and thin-wall parts that are prone to push-out whitening or distortion. Our approach is to try one step up in the parameter phase (say 70°C to 80°C); if the weld line only fades slightly, the direction is wrong and you should go back to the gate and wall thickness.

Q: How much do stiffening ribs affect the weld line?
A: A great deal. When the rib root thickness exceeds 0.7 times the main wall, the melt is clearly split as it passes the rib and an obvious whitened weld line tends to remain at the re-merge point. The usual practice is to hold rib thickness at 0.5-0.6 times the main wall. Thinning loses some stiffness, which is normally recovered by making the rib taller, and the rib root fillet should not be too small — 0.3mm or more — otherwise it creates stress concentration and sink marks.

Q: When must the gate position be changed?
A: When the meeting angle is below 135°, the weld line sits on the primary cosmetic face, and there is no room to adjust either hole positions or wall thickness — then moving the gate is the only effective direction. In that case, evaluate the cost first: repositioning a side gate usually only requires a new gate insert, which is relatively cheap; repositioning a hot-runner gate means rebuilding the whole system, which is a different order of cost. We normally lay the options and their costs in front of the customer before deciding which side changes.

Originally published by the Hezi Industrial Design team, based on our industrial design project practice in Dongguan. Learn more: Industrial design services

E-Coating or Spray Painting? Deep Cavities, Corners, and the Environmental Bill

Asher Hu — Thu, 15 Oct 2026 00:49:50 +0000

E-coating pushes water-based paint particles onto a metal surface with a DC electric field, so welds, the inside of hemmed edges and internal cavity walls all get a film — and a uniform one. Spray painting carries paint to the surface by atomisation and airflow, which frees you on colour and substrate, but corners accumulate paint and surfaces facing away inside a deep cavity are a shadow zone that has to be touched up by hand. The dividing line is fairly clean: if the part is conductive metal, the volume is large enough and the job is a rust-prevention primer, go with e-coating first. If it is a plastic part, a small batch, or a coloured cosmetic surface, spray or powder fits better. The environmental accounting, though, cannot be settled by saying "e-coating is greener".

The one-line conclusion: e-coating needs no operator for the tricky geometry — the field does the work — but it is limited to conductive substrates, is essentially restricted to black and grey, and demands hanging points, drainage holes and a 160-180°C bake. Spray painting is flexible on material and colour and is easy to touch up locally, but it cannot reach the back of a deep cavity, and thin-wall corners build up paint. Many outdoor metal parts end up using both: an e-coat primer for rust protection and coverage, plus a powder or liquid topcoat for colour and weather resistance.

Dead corners: one is covered by a field, the other by hand

E-coating immerses the whole part in a tank and applies current, so the paint particles migrate to the workpiece. Wherever the field reaches, paint reaches. Welds, the inside of folded edges and cavity walls — exactly the places a spray gun cannot reach — actually end up more reliably covered than the outer face. Outer surfaces typically run 15-25μm; internal cavities are thinner, but the film is continuous and does provide protection.

The weakness of an electric field is Faraday shielding. Inside a cavity, at the bottom of a deep blind hole, or in a gap between two plates sitting very close together, field strength drops close to zero — that is where you get no film at all, or a film too thin to protect anything. So the internal geometry of an e-coated part cannot be designed too freely.

On the spray side the problem is the mirror image. Everything in line of sight can be painted, but surface tension makes paint accumulate at corners, so the rim of a thin-wall part swells and the interference only shows up at assembly. A deep cavity facing away from the gun is a shadow zone that no amount of angling will reach. Our usual approach is to convert deep cavities into through cavities, or to add a dedicated paint access hole so the gun can be inserted to touch up.

One more difference that is easy to overlook is rework. Scratch a sprayed part and you can touch it in with matched paint — there will be a colour difference, but at least it can ship. Scratch through an e-coat and matching the colour locally is very difficult; in most cases the whole part goes back through the line. In volume production the probability is not high, but if the structure uses thin, easily deformed parts, the handling damage rate goes up and so does that cost.

Hanging points and baking temperature decide whether the part can be e-coated at all

The hanging fixture must conduct, and at the contact point between fixture and workpiece there is no film, or a very thin one. That point has to be decided at the drawing stage and placed on a non-cosmetic face, or on a face that gets hidden by another part after assembly. Let the coating shop clamp wherever it likes and you are waiting for a customer complaint when it lands on the visible surface.

The number of hanging points is not arbitrary either. A single point on thin sheet metal, swinging in the tank and then heated in the oven, has a high probability of distortion. Cathodic e-coat baking generally sits in the 160-180°C range for 20-30 minutes; long parts and thin sheet parts basically need two or more hanging points, or a support frame.

That temperature also rules out a lot of designs outright. Plastic parts, structures with rubber seals, and assemblies containing electronic components are not going through an e-coat line as a unit. Spray painting is far more forgiving: two-component polyurethane can cure at low temperature or even dry at room temperature, and plastic parts can be painted.

Drainage and venting hole positions matter as well. The lowest point needs a drain and the highest point needs a vent, otherwise the electrolyte trapped in the cavity and the bubbles both turn into coating defects. These holes, the hanging points and the process holes are essentially one family of features — call them out together on the drawing instead of adding them in three separate rounds.

The environmental bill: VOC is only one line item

The e-coat tank is water-based; for solvent-borne spray paint the bulk of VOC comes out of the spray booth and the flash-off zone. Emission requirements can be checked against GB 37822-2019, the standard for fugitive VOC emissions, while limits on hazardous substances in the coating itself are covered by GB 30981-2020. Specific limits differ by product category, so confirm the local scheme when quoting a project.

But e-coating is not free of environmental burden. The pretreatment stage — degreasing, phosphating or zirconium conversion — produces wastewater, and phosphate baths contain phosphorus and heavy metals, which is why many plants have switched to zirconium conversion to reduce the load. Ultrafiltrate rinse has to be recovered and reused, and the tank's solids content, pH and temperature need periodic testing and adjustment. Those are long-term running costs, and they do not come cheap when spread across unit price.

The direct answer: environmentally, you cannot simply say e-coating beats spraying. E-coating has low VOC emissions because the bath is water-based; the price is pretreatment wastewater and tank maintenance. Powder coating has near-zero VOC and overspray powder can be recovered, but colour changes are expensive and the film is on the thick side. Solvent-borne liquid spraying has the highest VOC, though much of it has moved to waterborne. Which one you pick depends on volume, the number of colours and local emission requirements together.

Powder coating is a third route. There is no solvent, overspray powder can be collected and reused, and VOC is effectively negligible. The drawback is that changing colour means cleaning the gun and the booth; for small batch orders cycling through five or six colours a day, the cleaning time alone eats the margin. Film thickness is commonly 60-80μm — noticeably thicker than e-coat — and corner coverage is better than liquid paint, but deep cavities still get nothing.

The usual combination for outdoor metal parts is an e-coat primer plus a powder or liquid topcoat. The e-coat handles rust prevention and dead corners; the topcoat handles colour and weather resistance. Cathodic e-coat on its own has mediocre weather resistance and will chalk under long-term outdoor exposure, so the cosmetic face still depends on the topcoat. That combination is more robust than either process alone.

The cost arithmetic is straightforward: e-coating is a line — pretreatment, tank, ultrafiltration and oven are all mandatory — the tank volume is large and the cost of opening the line is high. Nobody runs e-coat at the sampling stage; you start with spray or powder and switch once order volume climbs. What structural design should do is leave room for both processes in advance, for example designing one sheet metal set with both hanging points and a paint access hole.

Questions we get most often

Q: What film thickness does e-coating normally give?
A: Cathodic e-coat commonly runs 15-25μm on outer surfaces, thinner inside cavities and at welds, but the film is continuous — that is the core advantage over spraying. For comparison, powder coating commonly runs 60-80μm and liquid spraying 20-40μm. Thicker is not better: an over-thick e-coat affects hemming and assembly clearance, so tolerances on the drawing need to account for it.

Q: When should I choose e-coating instead of spraying?
A: When all three conditions hold — the workpiece is conductive metal, the volume is large enough, and rust prevention plus corner coverage are demanding. Typical cases are outdoor sheet metal cabinets, chassis parts and bracket-type structural parts. Plastic parts, small batches, many colours, or assemblies containing rubber and electronics essentially rule e-coating out; spraying and powder are more flexible in those scenarios and far cheaper to set up.

Q: What should structural design watch for on e-coated parts?
A: Three things must be fixed at the drawing stage: hanging point positions, cavity process holes, and a drain at the lowest point. Hanging points should avoid the cosmetic face; process holes are generally φ8 or larger; a single-hole blind cavity should stay within 50mm deep. Thin sheet metal parts also need more hanging points and support frames, because e-coat baking sits at 160-180°C for twenty-odd minutes and a slender part hung from one point distorts easily. If these parameters are not specified up front, the coating shop follows its own habits and the rework rate goes up.

Originally published by the Hezi Industrial Design team, based on our industrial design project practice in Dongguan. Learn more: Industrial design services

Two Hours of Vibration Testing, Three M3 Screws Came Loose

Asher Hu — Wed, 14 Oct 2026 04:14:14 +0000

Loose screws after transport vibration almost always trace back to the same root cause: lateral vibration. A bolted joint under pure axial tension is still held by the self-locking geometry of the thread. But the moment the two clamped parts slide against each other laterally, the thread pair takes a small circumferential slip, preload decays step by step, and once it drops past a critical value the rest takes minutes. Locking measures have to be designed against lateral displacement — a spring washer on its own does very little.

The one-line conclusion: get the preload right, lock the friction pair, and then add a mechanical stop. Only doing one of the three — typically a spring washer — is roughly equivalent to doing nothing. Engagement length matters as much as the fastener: steel on steel wants 0.8d or more, a steel screw into aluminium wants 1.5d upwards, and into plastic 2d. On a plastic housing, driving a self-tapping screw straight in to hold something heavy is a classic failure point.

First, work out which vibration you are actually running

A lot of teams say "vibration testing" when what they actually run is a sine sweep. A sweep is for finding resonances, not for simulating transport. A truck, a parcel sorting line, a sea container — what the product experiences there is random vibration, a broadband spectrum with energy stacked across a whole band. The two tests answer different questions and cannot substitute for each other.

For transport packaging vibration in China, the usual standard is GB/T 4857.23, the random vibration method. Export products are frequently specified against ISTA 3A by the customer. Both work the same way: you are given a power spectral density curve, an acceleration RMS value (Grms) and a test duration.

Our own practice is a two-step approach. First a low-level sine sweep to find the product's first natural frequency. Road transport energy concentrates in the low-frequency band, and if the first natural frequency sits inside that band, the amplitude gets amplified several times over — and the lateral shear on the screws is amplified with it. Once the resonance is located, we run the random vibration endurance test against the customer's specified spectrum.

The sweep is the step most often skipped. People argue that running random vibration for the full duration is enough. But after a random run you only know whether it came loose, not why. Adding a sweep tells you whether the problem is insufficient structural stiffness causing resonance, or the joint itself — and those two are fixed in completely different ways.

How a screw walks itself loose

There is a mature engineering method for verifying thread pair loosening under lateral vibration: the transverse vibration test rig described in ISO 16130. The results show a very typical pattern. An ordinary nut under lateral vibration will lose more than half its preload within tens to a few hundred cycles. Swap in a nylon insert lock nut and it survives into the tens of thousands of cycles. The difference is not bolt strength — it is whether anything prevents the thread pair from rotating relative to each other.

Several variables control how fast loosening happens, and all of them can be managed at the design stage:

Thread pitch. A fine pitch has a smaller helix angle and better self-locking behaviour, at the cost of slower assembly and a higher risk of cross-threading. It is not common on small and medium parts, but in a genuinely high-vibration environment a fine pitch is worth considering.

Engagement length. Steel into steel is generally held at 0.8d or more. A steel screw into aluminium wants 1.5d upwards by industry rule of thumb, and into plastic, 2d and above. With insufficient engagement the bearing pressure on the thread flanks is too high, the threads crush, and preload is gone on the spot.

Joint stiffness. This is the one most often ignored. Plastics, foam pads and soft rubber gaskets creep: they compress and slowly get thinner, preload follows them down, and it never comes back on its own. Driving a self-tapping screw into a plastic shell to hold something heavy is a high-incidence failure pattern for exactly this reason.

Our three-layer locking order, and one fix that went the wrong way

The first layer is preload. Torque values cannot be decided by gut feel, and there is no universal torque table that covers every case. The same M3 stainless screw, into aluminium versus into steel, dry versus lubricated, with or without a washer, will take noticeably different torque — typically floating in the 0.5 to 0.8 N·m range. The real value has to be confirmed by trial assembly with a torque wrench. Too little torque and you have not locked anything; too much and you strip the aluminium.

The second layer is friction locking. Medium-strength threadlocker, the Loctite 243 family, suits joints that need to come apart again — heat or extra force will break it. High-strength threadlocker is effectively semi-permanent and tends to damage the threads during removal. Nylon insert lock nuts, all-metal prevailing torque nuts and serrated flange nuts all live in this layer too. Application technique matters: put the compound on the middle of the engaged thread, not piled at the root of the bolt and not smeared on the outside of the nut. After tightening, the compound has to spread across the engagement zone. Applied in the wrong place, it does nothing.

The third layer is the one we care about most: do not let the screw carry shear. If two parts in the assembly can slide relative to each other, use locating posts, snap fits or ribs to take the lateral load, and let the screw do nothing but clamp. Get this layer right and the requirements on the first two can be relaxed.

We learned this the expensive way. An early aluminium-housing product used M3 screws with spring washers. After the random vibration run, we opened it up: the washers were still sitting in place, but the torque had already dropped. On the same joints we later added locating posts plus threadlocker and raised the engagement length from under 1d to 1.5d and above — only then did it hold. That direction of fix cost almost nothing: a few extra posts in the mould.

Two assembly details also get overlooked on the production line. The first is tightening sequence: with multiple fasteners, tighten diagonally in stages. Going straight to full torque on one point lets it absorb most of the preload and leaves the rest effectively loose. The second is reworked parts: if a threaded hole still holds cured compound from a previous assembly, the torque reading on the second assembly will be wrong — clean it out before applying new compound.

Finally, acceptance criteria. After the vibration run, do not just check whether anything fell off. Our practice is to open the product and re-measure every joint with a torque wrench, comparing against the initial assembly torque. If the decay exceeds a set proportion, the joint fails even if the part is still hanging there. The product may go through several more legs of transport in its life; passing this run does not mean it will pass the next one.

Questions we get most often

Q: Which standard should transport vibration testing follow?
A: In China the common choice is GB/T 4857.23, the random vibration method for transport packages; for export products customers frequently specify ISTA 3A. Both supply a power spectral density curve, a Grms value and a duration, simulating the broadband random vibration of road transport, parcel sorting and sea freight. We recommend a low-level sine sweep before the random run to locate the first natural frequency. If that frequency lands inside the energy-concentrated band, the structural response is amplified and conclusions drawn from the random run alone can easily point you in the wrong direction. Duration and level are set by the customer or target market and vary considerably between transport modes.

Q: Can a spring washer actually stop transport vibration from loosening a screw?
A: Its effect on lateral vibration is limited. A spring washer mainly maintains clamping through axial spring force, so it can compensate for a little preload loss, but it cannot prevent lateral relative slip of the thread pair — and that slip is the primary cause of loosening under transport vibration. For high-vibration locations, switch to threadlocker, a nylon insert lock nut or an all-metal prevailing torque nut, or better, change the structure so the screw does not carry shear at all. That structural layer is the cheapest one to add.

Q: Where exactly should threadlocker be applied?
A: On the middle of the engaged thread, where it can spread across the engagement zone when the joint is tightened. Piled at the root of the bolt or smeared on the outside of the nut, it never reaches the mating surface and does nothing. Medium-strength compound suits joints that need rework; high-strength is effectively permanent. One more point that gets missed: on a reworked threaded hole, the cured compound left inside must be cleaned out before new compound is applied, otherwise the torque reading is inaccurate and the actual preload comes out low. If a removed bolt has an obvious layer of compound on its threads, we generally replace it rather than reuse it.

Q: How do you judge pass or fail after a vibration run?
A: Not by whether parts fell off. Open the product and re-measure each joint with a torque wrench, comparing against the initial assembly torque. Anything exceeding the set decay proportion fails, even when it looks intact. Also inspect for stripped threads and stress-whitened or cracked plastic bosses. The reason is that the product may see several legs of transport over its lifetime, so a single passing run does not prove long-term reliability. We recommend writing the acceptance criterion into the design requirements at project kick-off and agreeing it with the customer, so that nobody argues about pass or fail after the test has already been run.

Originally published by the Hezi Industrial Design team, based on our industrial design project practice in Dongguan. Learn more: Industrial design services

Curved Glass Forming: How to Choose Between Heat Bending, CNC and Precision Pressing

Asher Hu — Tue, 13 Oct 2026 01:00:16 +0000

Which route curved glass takes is decided by volume before anything else. For a few dozen to a few hundred pieces of sampling, CNC milling the profile is the simplest option: no tooling, lead time counted in days. Once monthly volume reaches the thousand-piece range, the tooling cost of heat bending starts to amortise. And for genuinely large curvature with optical-grade surfaces, precision hot pressing is the only way. There is no fixed cut-off point — the larger the glass, the more complex the curvature and the tighter the tolerance, the higher that point moves. The worst mistake I have seen is quoting a batch order using sampling-scale CNC unit pricing; the customer only discovers that every piece loses money after signing the contract.

The one-line conclusion: CNC cold processing needs no tooling and holds tight tolerances but costs a lot per piece, so it suits sampling and small batches. Heat bending forms against a mould, giving a low unit cost but high tooling cost and long lead times, so it suits volume curved parts. Precision hot pressing (3D moulding) delivers the best surface-form accuracy, at the price of short mould life and expensive equipment, which is why it is reserved for optical and high-end cosmetic parts. The three are not substitutes for one another; plenty of projects run heat bending, then CNC trimming, then chemical strengthening in sequence.

Heat bending: the money goes into the mould first, the unit cost drops later

The principle is plain: heat the glass close to its softening point, let it deform under its own weight or under a mould, then cool it slowly along an annealing curve. Ordinary soda-lime glass starts to soften around 550°C, while high-alumina silicate glass needs to go considerably higher, with a softening point near 800°C depending on the grade. A tunnel-type bending furnace typically has a dozen or more temperature zones, controlling preheating, forming, annealing and cooling separately. Get one segment of that curve wrong and the part carries high internal stress, which later shows up as spontaneous cracking during edge trimming or a drop test.

Three mould materials are common: graphite, heat-resistant stainless steel and ceramic. Graphite machines easily and is cheap but wears quickly and sheds powder, and powder landing on the glass surface becomes pitting. Stainless steel is durable but conducts heat fast, making temperature control difficult. Ceramic sits between the two. Mould cost follows size and curvature complexity — tens of thousands of yuan for small parts, several hundred thousand for large or multi-cavity moulds. Tooling lead time starts at a few weeks and reaches one to two months for complex shapes. That money and that time are the wall between heat bending and small-batch sampling.

What heat bending cannot do is equally clear. Profile tolerance generally lands in the ±0.1 to ±0.3 mm band, and deteriorates as the part grows. The formed edge is a rough blank, so it almost always needs a CNC pass afterwards, which means the drawing has to leave a few tenths of a millimetre of machining allowance for the downstream step. Then there is springback: once the glass leaves the mould the curvature contracts slightly and the actual angle is smaller than the mould, so the mould design needs compensation. The compensation amount is tuned by trial forming, and it is normal for the first attempt to miss.

CNC: no tooling, but every piece burns machine time

CNC cold processing mills flat glass directly. A diamond burr mounted on a high-speed spindle, usually running above twenty or thirty thousand rpm, mills the profile first, then drills holes, then grinds and polishes the edges. The advantage is that no mould is required — changing the drawing means changing the program — and samples arrive in three to five days. Dimensional accuracy can hold within ±0.05 mm, an order of magnitude better than heat bending, which is why many projects rely on CNC parts during the validation stage to check assembly.

The price is machine time. For a piece of glass the size of a phone cover, milling the profile plus drilling plus polishing takes anywhere from a few minutes to over ten. Larger parts and more holes mean slower work. CNC output must be chemically strengthened, meaning the glass is immersed in molten salt for ion exchange at around 400°C for several hours. Before strengthening, the edge must not have chipping — a chip is a stress concentration point, and after strengthening the part cracks along that chip the first time it is dropped. The other ceiling for CNC is shape: it can produce 2D profiles and 2.5D chamfers and arc edges, but not a true three-dimensional large curvature. Using CNC to chase a surface with several millimetres of drop is not economical in cost or yield.

How to split by volume: for sampling and anything under a few hundred pieces, prefer CNC and skip the tooling cost and trial-forming time. From the thousand-piece level upwards, with a curved shape, go to heat bending. Only when surface-form accuracy must reach the micron level and transmitted imaging cannot distort should precision hot pressing be considered. It is entirely normal for a project to span two stages — validate assembly with CNC, switch to heat bending for production — but the structure must leave clearance according to heat-bending tolerances from the start.

The three sets of data design must hand to the factory

The first set is the 3D curved surface data. Handing over a 2D drawing with a few R-angle callouts is not enough: the mould shop will extrapolate the surface itself, and the surface it derives is often not the surface the ID team wanted. We learned this the hard way. On a curved panel, the drawing carried the outline dimensions and a few radii but no 3D surface; the mould came back and the edge sat a few tenths of a millimetre off the mid-frame when assembled, with uneven gap width, and the only fix was to rework the mould. Since then we hand over STEP files for every curved part, together with the mating surfaces of the mid-frame, so the factory extrapolates against the real assembly relationship.

The second set is fit clearance and the tolerance chain. Glass is brittle and cannot be forced to act as a structural member. Between the glass and a plastic mid-frame, leave 0.1 to 0.15 mm per side, with foam or dispensed adhesive in between so an elastic element absorbs the tolerance. Too little clearance and the glass is loaded during assembly, so a corner drop cracks it; too much and the gap is visible and dust gets in. The tolerance chain has three layers: the glass body, the position of printing or coating, and assembly. Curved printing uses pad printing or heat transfer, one grade worse in positional accuracy than flat screen printing, so artwork needs generous margin from the edge — otherwise a slight print shift lands on the curved transition zone, where adhesion is worst.

The third set is the downstream process list. Trimming, strengthening, printing, coating and bonding each change dimensions or affect yield, and the factory needs the list up front to sequence the work. Wrong order causes real trouble: strengthen before drilling and the micro-cracks at the hole edge destroy the strengthened layer; print before strengthening and the ink discolours in the hot molten salt. There is also the minimum bend radius. Do not draw an aggressive small radius on the drawing and expect the factory to rescue it — the smaller the radius, the thinner the outer surface is stretched, and the more obvious the optical distortion and stress concentration. Curved panels with a viewing window deserve particular care: looking through them at the LEDs behind often produces ghosting or rainbow banding, and that class of problem can only be solved by enlarging the radius or making the area flat. It cannot be tuned out afterwards.

The item design most often forgets: annealing and stress information. Curved glass must be annealed along a curve after demoulding. Parts whose internal stress has not been fully released can shatter during subsequent CNC trimming or a drop, and the shatter location is usually not the load point. Asking for a few extra samples for drop and thermal-shock testing during the sampling stage is more useful than editing tolerances on the drawing.

Questions we get most often

Q: At the sampling stage, choose CNC or open a heat-bending mould directly?
A: Prefer CNC for sampling, because there is no tooling, changing the drawing means changing the program, and parts arrive in days, which suits validating assembly and appearance. Unless the curvature drop is so large that CNC simply cannot produce the shape, spending tens of thousands on tooling at the sampling stage carries high risk — a heat-bending trial takes weeks, and changing the curvature in between means reworking the mould, so the time cost far exceeds the extra per-piece price of CNC. The sound approach is two rounds of assembly validation on CNC, then tooling once the 3D surface data is locked.

Q: What dimensional tolerance can heat-bent glass hold?
A: Profile tolerance on conventional heat-bent parts lands between ±0.1 and ±0.3 mm, depending on glass size, curvature complexity and mould condition: the larger the part and the more drastic the curvature change, the harder the tolerance is to tighten. For higher accuracy the outline can be brought within ±0.05 mm by downstream CNC trimming, but the surface-form accuracy of the curvature itself is difficult to improve by trimming — that is decided at the forming stage. Where the glass mates with the mid-frame, do not expect the glass to hold the dimension itself; leave a few tenths of a millimetre and absorb the tolerance with foam or dispensed adhesive.

Q: Why does curved glass show optical distortion?
A: Distortion comes from thickness variation and residual stress in the bent region: the tighter the bend, the thinner the stretched outer surface and the more compressed the inner one, so light refracts in inconsistent directions and the image warps or shows rainbow banding. The tighter the bend and the smaller the radius, the more obvious it is; incomplete annealing leaves internal stress that makes it worse. For products with a viewing window looking through at LEDs, we generally recommend enlarging the bend radius, or making the window area flat and keeping curvature in the non-transparent region. It cannot be tuned out afterwards.

Q: How does precision hot pressing of 3D glass differ from heat bending?
A: Heat bending mainly relies on self-weight or a soft mould, so surface-form accuracy is ordinary. Precision hot pressing uses a pair of high-accuracy upper and lower moulds to press softened glass, transferring the mould surface directly onto the glass, with surface-form accuracy reaching the micron level. The cost is that the moulds use hard alloy or special graphite, are extremely demanding to machine, and have short life under high temperature and pressure, so equipment and tooling investment is far higher than heat bending. Only applications such as optical lenses and aspherical elements justify it; ordinary cosmetic parts rarely take this route for surface-form accuracy.

Originally published by the Hezi Industrial Design team, based on our industrial design project practice in Dongguan. Learn more: Industrial design services