Across the industrial and logistics industry there is a natural plateau that many organisation hit with their digital transformation. They deploy IoT sensors across their business to monitor assets, identify changes in environmental factors and monitor wear and tear. This has resulted in huge amounts of data being generated.
However for most organisations this has only improved visibility through dashboards. These will provide alerts but need human interpretation.
The issue is known as ‘Operational Inertia’. Even if we can detect anomalies with our digital solutions extremely quickly, the physical world can take hours/days to resolve.
Technical leaders must look to implement Physical AI solutions that can offer dynamic results through automation.
Closed Loop – The 4 Engine Stack
To make any change in the physical world, we must design an architecture that can take a high volume of telemetry data and turn it into physical actions. This is done using our 4 Engine Stack.
Engine 1 – Identification & Location
The first thing we need to do is identify and locate our assets. We must know where they are located, how they are orientated and how they relate to other infrastructure.
Our Identification Engine will do this for us. Using data from various sources (Ultra Wide Band, Bluetooth Low Energy, Active & Passive RFID & Computer Vision) the engine will normalise this data to give us an intelligent spatial view. This will ensure that all our metrics are known to the nearest meter.
Engine 2 – Multi-Sensor
Once we have identified our assets we can begin collecting telemetry from a variety of sensors. This engine will monitor various mechanical factors such as high frequency vibratory patterns, sound, heat, energy and pressure.
This data can be processed and analysed on an edge basis. This means we can filter out localised noise, and benefit from high bandwidth. However mainly, this will help overcome delays caused by transmitting data to the cloud.
Engine 3 – AI Decision
Now that we have normalised data from the edge, we can pass it to locally hosted machine learning models. These have been trained against accurate behavioural profiles for each type of asset.
Rather than using basic thresholding like we would traditionally, our AI Decision Engine will use multi-variable pattern recognition to give us an accurate prediction of how likely the asset will fail. This will also provide us with an instant risk score so we can understand how this will affect our operation.
Engine 4 – Physical AI Actions
This engine will provide us with a closed loop.
Using high confidence algorithms our Physical AI Action engine will produce accurate PLC logic adjustments, automatically dispatch maintenance and adjust machine duty cycles. In cases of critical failure our system can also implement pre-validated safety mechanisms.
Limits On Physical Action Engines
Implementing any kind of physical action engine can cause issues. If you try to operate large machinery, high speed conveyers or high pressure systems using closed loop software, you may face unknown safety issues or process issues.
To ensure we can run smoothly, our Physical AI systems in enterprise will follow a HITL model. We will set clear limits for how our systems can act. These will be decided based on the risk probability.
Low Risk – Automated. Our systems will automatically carry out low risk tasks. These could include adding condition based inspections to log, updating our spatial tags or ordering new parts from our inventory management tool.
Medium Risk – Semi-Automated. Our system will recommend actions but the user will need to press one button to implement them. For example changing motor drive speeds.
High Risk – HITL. Our system will implement pre-validated safety measures. For example if our temperature or vibrations go above critical levels our machines will automatically shut down. This will be escalated to a human with high priority.
By separating our intelligent systems into high throughput engines, we can move past just having visibility and have zero delay in our operations. If your organisation want to understand more about how we can implement edge based systems and look at some advanced industrial deep tech ventures then you can find research from Aperture Venture Studio.
Top comments (0)