If the engine is the heart of a car, the wiring harness is its nervous system. This metaphor has been used for years, but few truly realize that this "nervous system" is undergoing an unprecedented transformation.
A single wire is becoming completely different. Its materials are changing, its length is changing, and its mission is changing. Today, we start with the industrial evolution of automotive wiring harnesses, then move into harness inspection, visual inspection, and quality control at the "last centimeter" — connectors and terminals.
Automotive Harness · Visual Inspection · Connectors · Terminals · MindVision
- A Hidden Billion-Dollar Track Global automotive wiring harness sales were about USD 71 billion in 2025, expected to reach USD 75.1 billion in 2026, and potentially exceed USD 108.9 billion by 2032, with a CAGR of about 6.4%. Under another estimate, the global automotive harness market is about USD 57.9 billion in 2026 and will grow to USD 92.8 billion by 2034. Definitions differ, but the direction is clear: this is a steadily expanding billion-dollar market.
A fascinating feature of the harness industry is that it lacks the glamour of batteries or chips, yet every vehicle cannot do without it. A traditional ICE vehicle carries about RMB 2,500–4,500 of harness value per unit, while a new energy vehicle averages around RMB 5,000 — nearly double. The reason is simple: EVs add high-voltage connection systems among battery, motor, and electronic control, greatly increasing harness complexity and technical barriers.
Globally, Japanese, American, and European Tier 1 giants still dominate. Yazaki, Sumitomo Electric, and Aptiv together hold about 71% of the global market. In China, however, local players are accelerating. Tianhai Electric, for example, reached 8.45% domestic market share in automotive harnesses in 2024, ranking first in China, with connector share also among the top three domestically.
- Aluminum Replacing Copper: The Two-Century "Copper Era" Begins to Loosen If any change best represents the "underlying revolution" in the harness industry, it is the shift in materials.
For nearly two centuries, copper has been the absolute protagonist of automotive wiring. The reasons are solid: excellent conductivity, high mechanical strength, and mature processing. But copper has an increasingly fatal weakness — cost.
LME data shows copper prices surged to a record high of EUR 13,160 per tonne in the first half of 2026, and remain around EUR 11,663 per tonne. Aluminum spot prices are only about EUR 2,719 per tonne — copper is more than four times as expensive. This huge gap is forcing automakers to reconsider material choices.
Ferrari has introduced aluminum high-voltage harnesses in the 296 hybrid and is extending them to its first EV, Luce. BMW's latest e-drive platform uses substantial aluminum wiring in both high- and low-voltage systems. Stellantis has also begun replacing copper with aluminum in some models.
The benefits are clear. Lear's measurements show weight dropping from 21.2 g/m to 11.7 g/m — a 45% reduction. Although aluminum has lower conductivity and needs about 30% more cross-section for the same current capacity, it still enables about 30% lightweighting and significantly lower material cost. Vehicle-wide aluminum harnesses can cut overall cost by about 20%.
But aluminum replacement is not simply swapping materials. Aluminum conductivity is only about 60% of copper, mechanical strength about one-third, and it suffers from creep and oxidation. Joining aluminum to copper terminals is a major pain point: thermal expansion mismatch can create voids at welded interfaces, and aluminum oxide films increase resistance. Ultrasonic welding is currently viable, but the process window is narrow, and the industry is still working on it.
In the short term, aluminum harnesses will first expand in high-voltage systems. Total harness weight can drop by at least 20%, and vehicle weight by over 10 kg. Full copper replacement will take time, but the direction is clear.
- Architecture Revolution: From 5 Kilometers to 100 Meters If aluminum replacing copper changes the material, the E/E architecture revolution changes the very way harnesses exist.
Under traditional distributed architectures, every function had its own ECU — engine management, braking, windows — leading to harness lengths of over 5 km and thousands of connectors, jokingly called "spaghetti" wiring. Domain-centralized architectures cut this to about 3 km. The latest "central computing + zonal control" architecture is pushing it toward the hundred-meter level.
Tesla is the most representative: Model S about 5 km, Model 3 about 1.5 km, Cybertruck down to 500 m. Leapmotor's "Four-Leaf Clover" architecture compresses the harness to 996 m, reduces ECUs to 22, and triples system response speed. Hefei Yitong Electronics' LNN system goes further, simplifying harnesses by 70% and cutting assembly time by 85%.
The core logic of zonal architecture is "local access" — sensors and actuators connect to nearby zonal controllers by physical location, not by function. Power distribution is also restructured: secondary power distribution moves into zonal controllers, shifting from "harness-driven" to "node-driven."
Meanwhile, 48V systems are further slimming harnesses. Tesla Cybertruck has already adopted 48V. At the same power, current drops to one quarter, and conductor cross-section can shrink by about 75%.
- When Harnesses Begin to "Speak": The Rise of High-Speed Data Harnesses Another change brought by intelligence: harnesses must not only transmit power, but also data.
With ADAS, smart cockpits, and connected vehicles, in-vehicle data demand is growing exponentially. Traditional CAN bus runs at hundreds of Kbps to 1 Mbps, while automotive Ethernet supports 100 Mbps to 10 Gbps — tens to hundreds of times faster. 10 Gbps Ethernet has moved from future technology to a smart-driving necessity.
This imposes new requirements: better shielding, lower signal loss, stricter EMC. Traditional harness design is no longer enough. Companies must upgrade materials, twisting processes, and connector design. Backbone connectors between modules now carry both high-voltage power and high-speed data, pushing connectors toward miniaturization and integration.
- Smart Manufacturing: One Harness Every 4 Seconds Another easily overlooked change: production methods are being redefined.
At Langfang Leoni Wiring Systems, smart equipment runs at high speed, producing one finished harness every 4 seconds on average. The company has 31 Mercedes-Benz lines and 18 Xiaomi lines, with annual capacity of about 720,000 sets.
Wuhan's new Vitesco plant uses fully automatic insertion equipment, modular assembly lines, and automatic ultrasonic welding, supported by overhead crane systems and AGVs, achieving full-process automation from raw material to finished product. Its self-developed visual inspection precision exceeds millimeter level.
In Xiaogan High-Tech Zone, a complete harness supply chain is forming — from copper rod drawing at Hubei Chuangxin Electric, to cable processing at Xiaogan Foose, to full vehicle harness assemblies. The local system covers "cable materials — precision harness processing — connector matching — full vehicle harness assembly." In 2025, the zone's harness output reached about RMB 8 billion, growing 10%.
- Visual Inspection: The "Second Pair of Eyes" for Harness Quality Control Harnesses are becoming thinner, shorter, and more complex, posing unprecedented quality control challenges. Traditional manual visual inspection is increasingly unable to keep up with growing variants and precision requirements.
Visual inspection is the core technological path to meet this challenge. It uses industrial cameras and algorithms to "understand" every detail of a harness, completing judgments in milliseconds that once required long, focused human attention.
6.1 Why Harnesses Need to Be "Seen"
Automotive harness manufacturing involves cutting, stripping, crimping, housing insertion, and assembly. Every step can introduce defects: incomplete crimping, insulation damage, wrong wire sequence or color, reversed connector polarity, cable surface scratches, jacket cracks, etc.
Most defects are visible, but manual inspection reliability is dropping sharply. Long periods of high concentration cause visual fatigue, fluctuating miss rates, and inconsistent results. More importantly, modern lines run faster, and harness variants multiply. With dozens of specifications and color combinations, human memory and visual recognition easily fail on high-speed lines.
At the same time, OEMs are tightening quality tolerance. A poorly connected terminal may only fail after tens of thousands of kilometers, and recall costs far exceed the cost of catching defects on the line. This "quality leverage" is turning visual inspection from optional to mandatory.
6.2 What to Inspect: Full-Dimensional Coverage from Surface to Interior
Automotive harness visual inspection is not a single technology, but a cluster of technologies for different inspection objects.
Terminal crimp inspection is one of the most difficult scenarios. Crimping mechanically presses the terminal and wire to form an electrical connection; quality directly determines signal reliability. Visual inspection must judge multiple metrics in a tiny area: crimp length, terminal end dimensions, crimp width, insulation crimp depth, exposed copper, etc.
A team at Soonchunhyang University proposed a Harris corner detection method, extracting six key measurements to judge crimp quality. It achieved average 97% accuracy across all items, with 99.9% and 99.7% for crimp width and core width. Its value lies in relying only on traditional image processing, without large annotated datasets, making it highly deployable for cost-sensitive small and medium lines.
Insulation jacket crack inspection faces another challenge. The jacket is a cylindrical surface. Under ordinary visible light, specular reflection hides fine cracks, and bending/twisting during assembly makes single-view images insufficient. A 2026 patent proposed a dual-domain method combining polarization reflection suppression and flattened geometric modeling, using multi-channel input and cross-domain attention for stable crack recognition and quantitative measurement under complex lighting. Multi-camera approaches are also used: synchronized multi-view image sequences, CNN and optical flow build a 3D center-axis model to precisely locate defect radial positions.
Wire sequence and color inspection is relatively mature. Harnesses contain multiple colored wires, and wrong sequence is one of the most common assembly errors. Semi-automatic machine vision systems can simultaneously verify wire position, connector polarity, and color sequence, supporting linear and circular configurations. The core is an HSV + RGB joint classifier, achieving high-precision color discrimination under controlled lighting.
6.3 How Deep Learning Changes the Game
If traditional machine vision solves "can see," deep learning solves "can see accurately" and "can understand."
The difficulty is the high diversity of defect forms. "Poor crimping" may appear as too shallow, too deep, skewed, exposed copper, excessive insulation insertion, etc. Each has different visual features. Rule-based methods require engineers to manually design features and thresholds for every defect — huge effort and hard to exhaust.
In recent years, lightweight YOLO-based models have become a research hotspot. TCQI-YOLOv5 is designed for terminal crimp quality inspection, integrating C2f, FasterNet, and efficient multi-scale attention. It achieves 98.3% average precision for hard-to-detect defects such as "insulation crimp too shallow," while meeting real-time industrial speed requirements.
Another study proposed ILN-YOLOv8, adding a P2 detection layer and efficient local attention. It reached 96.2% accuracy on a real crimp connector dataset, with parameters and computation reduced by 66.7% and 34.6%. This lightweight nature is critical for edge deployment — models must run in real time on limited compute, not rely on cloud inference.
For wire recognition, YOLO-Slender solves a more basic problem: how to quickly and accurately identify and segment each wire under changing lighting, stacking, and many wire types. With C3k2-RDR and a lightweight segmentation head, it reduces parameters by 28.53% while maintaining accuracy, suitable for industrial devices with limited compute.
6.4 An Easily Overlooked Advance: Few-Shot Learning
The weakness of deep learning is data. Traditionally, a reliable defect model needs thousands or tens of thousands of annotated images. In harness production, some defects are extremely rare — it is hard to collect enough "bad" samples.
FFAD offers a new path. Using a pre-trained vision-language model (CLIP), fixed-position image capture, and learnable positional prompts, it achieves over 90% detection accuracy with only 16 samples per defect class. This means new models or harness variants can be deployed in one or two days, without weeks of sample collection and training.
The practical significance: harness variants are exploding — one model may have dozens of configurations, and differences across models are even greater. If every change required large sample collection and retraining, visual inspection could never keep up with line changeover. Few-shot learning is opening this bottleneck.
6.5 A Key Understanding: Visual Inspection Is Not Electrical Testing
One boundary must be clarified: visual inspection and electrical testing (continuity, withstand voltage, insulation resistance) are complementary, not substitutes.
Visual inspection finds appearance and assembly defects — wrong sequence, abnormal crimp form, jacket damage, incomplete connector mating. It cannot verify whether the conductor is truly connected to the correct terminal, whether insulation remains intact under voltage stress, or whether contact resistance is within spec.
Conversely, a harness passing all electrical tests may still have visible mechanical or assembly defects that can later become failures.
Therefore, modern harness inspection uses a "visual + electrical" double defense: visual inspection catches appearance and assembly defects upstream, electrical testing verifies functional integrity downstream. On highly automated lines, the two are merging — the same station performs imaging and continuity testing, and integrated data enables cross-validation and fast traceability.
- Connectors and Terminals: The "Last Centimeter" of Harness Quality In the harness manufacturing chain, connectors and terminals are the "nerve endings." Whether a harness works reliably often depends not on the cable itself, but on the crimp quality of dozens or hundreds of terminals and connectors, PIN coplanarity, and spring contact height precision.
These parts are often millimeter or sub-millimeter scale, yet carry core signal and current functions. A poorly connected terminal may fail after tens of thousands of kilometers; a PIN bent by a fraction of a millimeter can interrupt an entire connector module.
This is the most difficult and critical "last centimeter" — terminal and connector inspection.
The challenges are multi-dimensional:
Precision vs. speed. Crimp width is often only a few millimeters; PIN pitch can be 0.4 mm or finer; defect features are micron-scale. Meanwhile, high-speed lines require cycle times in seconds or milliseconds — full-dimensional PIN inspection in 0.1 seconds demands extreme resolution and frame rate.
Diverse defect types. Terminal and connector defects are complex: bent, tilted, missing, broken pins, uneven plating, out-of-tolerance dimensions, corrosion, poor connection stability. Each has different visual features; some are invisible under visible light.
Material and geometry complexity. Terminals are metallic and highly reflective; specular highlights hide real defects. Connectors are compact, pins occlude each other, and a single view often cannot cover the whole inspection area.
Inline engineering challenge. No matter how accurate in the lab, inspection must fit line cycle time and space constraints. Systems must work with crimping and housing insertion machines, achieving stable imaging and fast judgment in tight spaces.
A mature connector visual inspection system must cover the full chain from imaging to judgment.
At the inspection item level, key metrics include spring contact height, PIN pitch and coplanarity, PIN length and diameter verification, plating quality, and connection stability. These form the basis of connector reliability.
At the imaging solution level, camera, lens, and lighting must match object size and precision. For PIN coplanarity requiring high-precision Z measurement, 3D or multi-angle vision is necessary. For PIN bending, sufficient resolution is needed to capture tiny angular deviations.
At the algorithm and judgment level, AI-based adaptive threshold algorithms are becoming mainstream. Compared with fixed thresholds, they better handle batch variation and lighting changes, maintaining detection rate while reducing false alarms.
At the system integration level, inspection must seamlessly connect with line control for real-time output and automatic sorting, ensuring defects do not flow downstream.
- Imaging Selection: Industrial Cameras Are the Starting Point Harness inspection places demanding requirements on industrial cameras, and different processes differ greatly.
Terminal crimping stations need high resolution and global shutter. Crimp areas are typically 2–10 mm, defect features often sub-millimeter, requiring sufficient ground resolution and global exposure to avoid geometric distortion in motion. MindVision's PX series, with global shutter CMOS and up to 10 MP, is a reasonable choice — 4608×2176 resolution with proper magnification presents crimp details clearly enough for stable algorithm judgment.
Harness appearance stations face another problem. A harness may contain a dozen colored wires that overlap, bend, or are partly covered. High color fidelity and dynamic range are needed to distinguish colors and boundaries under uneven lighting. MindVision GigE cameras offer full mono/color configurations and up to 25 MP, leaving room for optical selection in multi-variant, multi-color inspection.
High-speed line inspection balances speed and precision. When harnesses pass at high cycle rates, area-scan frame rate and exposure become bottlenecks. MindVision's MV-XGLC83DM-T4-90 line-scan camera, with 8K resolution and up to 150 kHz line rate, targets large-format, high-speed continuous inspection — keeping images sharp without smear, a prerequisite for defect interception. The PCIe line-scan MV-PEL83DM pushes max line rate to 200 kHz, further compressing the imaging window.
For cross-section analysis and copper-aluminum interface inspection after aluminum substitution, the SWIR camera MV-GEC500I uses an InGaAs detector covering 0.4–1.7 μm, "seeing through" interface features invisible in visible light. For aluminum oxide films and Cu-Al weld voids, multispectral imaging is becoming a direction worth watching.
Choosing a camera is essentially choosing whether it can run stably. There is a saying on harness lines: a solution that works in the lab may not survive the first month on the line.
Camera selection often focuses on pixels and frame rate, but in deployment, long-term stability is often determined by less glamorous factors — interface friendliness, SDK maturity, and compatibility cost with existing line control.
Harness lines have distributed stations; cameras are often inside mechanical structures with limited wiring space and narrow maintenance windows. MindVision GigE cameras support 100 m transmission and PoE, providing real value: one cable for power and data, reducing cables and maintenance complexity. They are fully Vision-compliant, with SDK support for Windows, Linux, ARM Linux, and MAC OS, requiring no extra adaptation for HALCON, VisionPro, and other mainstream platforms.
These details rarely appear prominently in PPTs, but they determine the distance between "accepted" and "still running after three years."
From "can capture" to "captures correctly," harness visual inspection is undergoing an interesting shift. Early needs were "can capture" — replacing eyes for basic sequence and appearance checks. Today's need is "captures correctly" — continuously delivering trustworthy images at smaller defect scales, with more complex materials and faster cycles.
This shift imposes systemic requirements. It is no longer a camera plus lens, but coordination of lighting, optics, camera performance, and mechanical positioning. In crimp inspection, if lighting cannot suppress metal highlights, even a high-resolution camera captures only overexposed areas. In color recognition, if color fidelity has systematic bias, even a fine classifier may misjudge after changing wire suppliers.
MindVision's portfolio includes an interesting approach: the MV-NEH501C autofocus industrial camera supports 35 cm to infinity autofocus and uses a platform design, allowing sensor modules with different resolutions to be customized. This flexibility matters in harnesses — one station may handle different lengths and poses; fixed-focus solutions require mechanical repositioning during changeover, while autofocus reduces that cost.
- QA: More About Automotive Harness Visual Inspection Q1: Can visual inspection fully replace manual inspection? Under clear rules, controlled lighting, and coverable defect types, visual inspection can approach full inspection with far better efficiency and consistency. But it cannot fully replace humans. Complex surfaces, rare defects, and fast multi-variant changeovers still need human review and exception handling. The realistic path is "machine full inspection + human review," not a one-size-fits-all replacement.
Q2: What is the relationship between visual inspection and electrical testing?
Complementary, not substitute. Visual inspection handles appearance and assembly — sequence, color, crimp form, jacket damage, connector mating. Electrical testing handles continuity, withstand voltage, insulation resistance. A harness may look perfect but have an internal broken core, or pass electrically but have abnormal crimp form. Modern lines should combine both for cross-validation.
Q3: How precise can terminal crimp visual inspection be?
Under controlled conditions, precision is considerable. Harris corner detection achieves about 97% average accuracy across six measurements — crimp length, terminal end dimensions, crimp width — with some dimensions above 99%. Improved YOLO models reach 98.3% average precision for hard defects like "insulation crimp too shallow." Actual line precision depends heavily on camera resolution, lighting, reflection control, and mechanical positioning.
Q4: Does deep learning inspection always need many defect samples?
Traditionally yes, but not necessarily now. Few-shot learning using pre-trained vision-language models can exceed 90% accuracy with only about 16 samples per defect class. This matters greatly because harness variants are many, changeovers fast, and some defects rare. It allows new models and configurations to be deployed in one or two days instead of weeks.
Q5: How does visual inspection handle many colors, strong reflections, and stacking?
Usually at three levels. First, lighting and optics — polarization, diffusion, multi-angle fill to suppress highlights. Second, algorithms — HSV + RGB classifiers for color, segmentation for stacking. Third, mechanics — fixtures to present the harness in the same pose. For curved jacket cracks, multi-camera or 3D reconstruction is used.
Q6: Can visual inspection keep up with faster line cycles?
Lightweight models are solving this. ILN-YOLOv8 maintains accuracy while reducing parameters and computation by about 66.7% and 34.6%, enabling real-time edge operation. Lines also use parallel stations, triggered capture, and ROI cropping to reduce per-frame computation. Visual inspection is not "capture and compute forever," but a real-time process embedded in the cycle.
Q7: Can visual inspection find internal broken cores or abnormal crimp resistance?
Not directly. Vision sees external form. Internal broken cores, high contact resistance, and internal insulation defects require continuity, micro-resistance, withstand voltage, or even X-ray/ultrasound. Visual inspection's value is catching appearance and assembly defects upstream, not covering every quality dimension.
Q8: Does a new harness require a new inspection system?
Ideally no — just change the recipe. Camera, lens, lighting, and station stay; switch models, thresholds, and inspection regions. Few-shot learning and pre-trained models are compressing changeover from weeks to days or less. If every harness change required new equipment, visual inspection could never scale in this fast-paced industry.
Q9: What metrics matter when selecting visual inspection equipment?
Don't just look at pixels. Look at frame rate vs. cycle, lighting stability, lens depth of field, algorithm platform openness, PLC/MES integration, false positive/negative rates, changeover convenience, and supplier field service. For crimp quality, check whether it closes the loop with cross-section analyzers, pull testers, and crimp force monitors — not an isolated vision device.
Q10: Will aluminum replacing copper bring new inspection challenges?
Yes. Aluminum oxide films, reflection characteristics, and Cu-Al interface visual features differ from pure copper. Aluminum is softer, and crimp deformation differs. Existing models and lighting cannot simply be copied; samples must be recollected and optics/thresholds adjusted. Aluminum substitution propagates into inspection.
Q11: How to calculate ROI for visual inspection?
ROI comes from three areas: reducing manual inspection and recheck costs, lowering recall/claim risk from escapes, and improving first-pass yield and cycle time. Harnesses are safety-related; a poor terminal may fail after tens of thousands of kilometers, and recall cost far exceeds line interception cost. For high-speed lines, visual inspection is not a cost item but a quality lever.
Q12: Where is visual inspection heading next?
From "inspection" to "control." Today it mostly finds defects; tomorrow inspection data will feed back to crimpers, cutters, and strippers for adaptive process adjustment. Edge AI, 3D vision, multispectral imaging, few-shot learning, and digital twins will further converge. When vision can warn of crimp parameter drift before defects appear, harness quality competition enters the next stage.
Final Thoughts
The automotive harness industry has been the "silent majority" for decades — not sexy, not eye-catching, but indispensable infrastructure for every vehicle. Today it stands at the intersection of multiple transformations: materials from copper to aluminum, architecture from distributed to centralized, function from power to power plus data, and production from labor-intensive to smart manufacturing.
Visual inspection is becoming indispensable quality infrastructure in this transformation. Its core value is not simply "replacing human eyes with machines," but moving quality control from post-hoc sampling to real-time full inspection, and from experience-based judgment to data-driven decisions.
When a line deploys dozens of visual inspection nodes, each generating multiple images and measurements per second, that data forms a real-time portrait of crimping process state. When a crimp parameter begins to drift — perhaps due to a slight tolerance change in a terminal batch — the vision system can warn before defects occur. This is the deepest value of visual inspection: not just a quality gate, but an entry point for process digitalization.
In the next decade of the harness industry, whoever moves visual inspection from "inspection" to "control" will hold the core advantage of quality competition. And the starting point is letting the camera capture a sufficiently informative image at the right time and the right position.
Reference: https://www.mindvision.ltd/
Top comments (0)