Across various industries, organisations have become stuck. Over the last decade we have implemented huge networks of connected devices within enterprises. Millions of data points are being collected on factory floors, logistics centres and energy centres. But despite the data, there is still a delay between issues occurring and action being taken.
Traditional IoT solutions are largely unidirectional. Data is captured from sensors at the edge and pushed to analysis tools in the cloud. Dashboards display charts and alerts but once an alert has been triggered, humans are required to investigate the issue.
Leaders are now looking to use modular Physical AI solutions to overcome friction and realise true value from automation.
Beyond Unidirectional Telemetry – 4 Engine Stack
To bridge the gap between data capture and action, the intelligence that occurs in the physical world needs to be broken down into 4 separate engines.
- Identity Engine
Without knowing what exactly the issue is, an alert is useless. The Identify engine will provide full details on each piece of equipment, stock and people throughout your facility. Using active and passive RFID, Bluetooth Low Energy tags, Ultra Wide Band Real Time Location systems and visual markers will ensure every sensor reading has a validated digital record.
Purpose: Provides instant identification of assets/people.
Standard: Fully automated identification of entities.
- Sensing Engine
Once the item has been identified, the Sensing engine will monitor multiple data points. These could include vibration, sound, temperature, pressure and energy usage. This normalised data stream will be processed at the edge to remove any noise from the environment.
Purpose: Normalise multi-sensor data streams.
Standard: Fully automated processing of data streams at the edge.
- AI Decision Engine
The normalised data streams are analysed using Machine Learning models trained on baseline data. This engine will not use rigid rules but instead will analyse complex differences in patterns across multiple data points. This will create a risk score based on the severity of the issue and will predict what impact it may have.
Purpose: Perform pattern recognition, risk prediction and decision making.
Standard: Exception based. Highlight high risk situations.
- Physical AI Engine
The final engine will take the decisions made and validate them by taking action. Based on the risk thresholds set up, the Physical AI engine will automatically trigger work orders to your asset management platform, change the operation of machines or trigger automated safety measures.
Purpose: Trigger actions to change the physical world.
Standard: High human override.
Responsible Automation
To ensure safety, when using automation engines in complex, capital intensive environments, responsible processes must be in place. If the algorithms had complete control over the physical machinery, there could be risks.
To ensure safety of the machinery and people, responsible boundaries are created. This will help to classify how automated the response should be.
Low Risk – Fully automated. This can be used for non-critical admin processes. For example capturing inspection data, updating stock locations and ordering standard parts.
Medium Risk – Semi automated. This will require human verification before any action is taken. For example changing machine speeds.
High Risk – Pre-approved safety measures with high priority notifications to supervisors.
By breaking the intelligence at the edge into these engines, organisations can move away from passive monitoring and create a resilient loop. If you would like to learn more about modular frameworks and how they can be deployed in deep tech environments, Aperture Venture Studio have technical research available.
Top comments (0)