Over the last decade there has been a huge investment in smart devices for enterprise industrial applications. Factories, fulfilment centres and energy facilities now have thousands of sensors collecting information around temperature changes, vibrations, energy consumption and movement. However many enterprise leaders are finding that they have invested millions creating a system that can detect issues but cannot solve them.
Traditional IoT solutions suffer from an inherent flaw. Information collected from devices can be sent up to a cloud solution and then reported through dashboards. Once an alert is raised, there is no automated action; a human must intervene and review the alert, find out which asset is causing the issue, check what is going on in the log files and send out a technician.
Industrial businesses are using Physical AI solutions to close the loop on these delays and mitigate risk. Physical AI turns the raw data from sensors into deterministic action.
Four Stages to Physical Intelligence
Firstly we need to create a system which can use the sensor data and take actions. This can be done by combining four different stages of data processing.
Stage 1 – Asset Identification
We need to know exactly who/what the issue is before we can determine whether there is a problem or how to solve it. To do this we combine passive and active RFID, Bluetooth Low Energy and Real Time Location System tags. This means we can instantly map the data received to the correct asset.
Stage 2 – Data Normalisation
Once we know we are receiving data from the correct asset we can process the physical data being received. This could include mechanical vibrations, soundwaves, pressure changes or temperature changes. This raw data can be processed and normalised at the edge.
Stage 3 – Smart Analysis
The normalised data can be used to run machine learning models either at the edge or in a hybrid cloud. Rather than triggering alerts based on thresholds, predictive algorithms can be used to analyse the changes in multi-variate patterns. From this we can determine a risk score.
Stage 4 – Safe Automation
Finally the predictive intelligence can be used to trigger actions on the physical devices. If the risk is low, work orders can be automatically created and prioritised within the enterprises asset management tool. If the situation is urgent or high risk, PLC’s can be pre-programmed to automate safe state changes.
Managing Risk
It is important to manage risk when looking to automate processes. We do not want to leave critical machinery with no control.
Physical AI solutions have limits so can be divided into three categories.
Low Risk – Full automation. These would include any standard processes such as updating maintenance logs, changing inventory flags or adding supply re-orders.
Medium Risk – Partial automation. Recommendations can be made by the system such as changing machine maintenance schedules or speeds. This just requires one click from a human.
High Risk – Highly limited. If critical parameters are breached, automation will kick in to put the device into a pre-validated safe state whilst also alerting the humans.
By separating the edge data into layers, we can use the data to drive actions. For more information on how enterprise teams are deploying physical AI solutions, look at the technical research from Aperture Venture Studio.
Top comments (0)