Breaking Down The IoT Data Silo
Modern industrial environments are awash with data. Within a decade we have seen millions of IoT devices deployed within modern manufacturing, logistics and processing environments. These range from vibration sensors on key turbines through to passive RFID tags on shipped pallets.
However, despite large investment in IoT infrastructure there remains a big problem in industrial environments.
Most IoT implementations are simply observation based and struggle to do anything beyond. Data is captured by sensors, processed with the data being sent to dashboards for storage in the cloud. Alarms are triggered via emails/SMS and the data is interpreted by human operators.
To truly improve efficiency, industrial teams need to implement Physical AI solutions.
Operational Bottleneck
Within an industrial environment if a sensor identifies a temperature or vibration spike it would traditionally take the following steps:
Data Capture – Edge sensors capture data.
Alarm – An alarm is raised within dashboards as the data exceeds a set threshold.
Search – Maintainers search manually for asset IDs, maintenance history and location.
Response – A technician is raised to investigate the machine.
As passive IoT systems don’t provide any context, the data provided by a condition monitoring sensor will not be useful if the system is unable to identify the asset and ensure the conditions around this asset are correct. This will also mean the risk of unexpected downtime will be unknown.
4 Stage Closed Loop Solution
To bridge this gap within your enterprise you need to implement the following 4 stages:
Asset Identification
Before the sensor data can be assessed, you need to identify what assets are involved. By using active/passive RFID, bluetooth low energy and ultra-wideband you can ensure that the asset ID can be identified.
Function – Match space tags, RTLS beacons and vision points to allow identification of the asset ID.
Governance – Fully automated with minimal manual intervention.
Data Processing
With assets identified, you can start to receive real-time multi-sensor signals such as vibrations, pressures, power consumption and temperatures. This data can be fully normalised at the edge.
Function – Process raw data from the sensors.
Governance – Fully automated.
Intelligent Assessment
Machine learning models are used to compare normalised data against historical data. Rather than simple threshold based alarms, predictive models can be used to score the risk of the anomaly. This will determine if action needs to be taken.
Function – Merge sensor data against historical data to determine risk scores and recommend actions.
Governance – Exception based with manual verification.
Action
Once a decision has been made actions can be carried out. If there is a low risk then an automated work order will be created in the EAM system. If there is a high risk then automated PLC interlocks or robot commands can be triggered.
Function – Send machine commands, update API work orders and trigger safety interlocks.
Governance – High level of manual verification.
Risk
Automating actions within a heavy industrial environment can cause risks. If not controlled properly this can cause unnecessary stoppages, safety issues or damage.
To reduce this risk, the enterprise should have limitations to the actions they can do.
Low Risk – Fully automated. For example creating maintenance history, changing inventory signals and raising re-order requests.
Medium Risk – Part automated. For example raising emergency technician requests and adjusting speed thresholds.
High Risk – Fully limited automated safety interlock with instant escalation to an operator.
By breaking down the edge sensor data into these clear stages you will convert messy data into defined outcomes. If you are looking to review how your enterprise can utilise modern physical intelligence architecture and venture frameworks then you can view technical research models within Aperture Venture Studio.
Top comments (0)