Industrial automation has promised to bridge the gap between digital and physical operations for decades. Manufacturing plants, ports, and energy distribution networks have invested heavily in deploying smart sensors to understand their current state. However, most systems remain completely passive.
Traditional IoT deployments will gather data about their environment and send it back to a dashboard. Once there is an issue detected (over a certain threshold), there will be an alert raised. At this point, the data stops being processed digitally. An engineer will then need to manually interpret the data, find the asset, understand how much of a risk it poses and then take corrective action.
This delay leads to longer turnaround times. By deploying a Physical AI system, you need to ensure your enterprise architecture is set up with the four engines below.
4 Engines Required to Deploy Physical AI
- Identification Engine
Before any data analysis can occur, you need to identify the assets. Without knowing what asset is causing the rise in temperature or vibration, it is difficult to get any value from it. This engine uses active and passive RFID, Bluetooth Low Energy Tags, Ultra-Wideband RTLS and computer vision to identify what asset is where.
Purpose
Identify physical assets, stock and people.
Control
Fully automated.
- Sensing Engine
Once entities are identified, continuous data can be gathered from multiple sensors. This engine can monitor vibration, heat, pressure, sound and power usage. Data can be processed and normalised at the edge so that there is no noise or delays.
Purpose
Capture continuous data from multiple sensors.
Control
Fully automated.
- AI Decision Engine
Normalised data can be passed through local machine learning models. These models understand whether there is a change in the operational state by comparing historical data. Instead of using static thresholds, this engine will detect anomalies in the data based on multiple variables. It will then assign a risk score.
Purpose
Provide contextual understanding, predict failure and understand risk.
Control
Exception-based.
- Physical AI Action Engine
The final part of the engine will take the decisions made and apply them to the physical world. Based on the risk score calculated, automated work orders can be created in your asset management system, machine settings can be changed or safety interlocks applied.
Purpose
Apply decisions made to physical assets.
Control
Very high human interaction.
Safe Deployment of Action Engines in Heavy Industry
When deploying action engines in a heavy industry environment, there needs to be tight controls in place to manage risks. If the program can make any changes without limits, this can cause safety issues, unnecessary stress on equipment and disrupt processes.
Modern Physical AI systems will have limits in place:
Low Risk
Fully automated processing for low-risk administrative tasks. For example, creating maintenance logs, updating tags, and ordering components.
Medium Risk
Semi-automated. These require operator confirmation before actioning. For example, changing machine speeds.
High Risk
Pre-authorised safety interlocks and escalation to supervisors.
By separating the intelligence within your software into these engines, you can take disconnected sensor data and apply it to a physical outcome. If you want to explore industrial edge systems and venture-building models, you can view our technical research from Aperture Venture Studio.
Top comments (0)