The goal of the experiment was to develop and validate in operational settings a Digital Twin model of a machine running on an edge device, which can be used for predictive maintenance. The proposed solution is based on the Apache toolchain of the MIDIH Reference Architecture and uses both the Data-in-Motion and Data-at-Rest processing pipelines. The first one is responsible for real time acquisition of data and reasoning with a pre-trained deep neural network model in order to predict potential failures. The second manages storage of annotated data and periodical re-training of the network with the goal of continuous improvement of the predictions accuracy. The developed system has been deployed and tested in close-to-operational settings and achieved the expected technical KPIs.
Figure 1 ISAB – the developed edge device
The APEMAN experiment allowed us to develop an embedded device able to collect measurements directly from a machine on a shopfloor via a wide array of sensors (e.g. current, voltage, frequency, vibration, temperature etc.) and to test them against a digital twin of the machine to predict and identify imminent failures of the machine. The data collected during the normal operation of the system is continuously transmitted to the server side of the system where it is used to periodically retrain the model and improve its accuracy over time.
Such an approach allows using ISAB as a portable device which can be easily moved from one machine to another, retrained and used for diagnostics while continuously increasing its performance. Such an achievement was possible only because of the duality of the MIDIH architecture supporting both the real-time Data-in-Motion processing and offline Data-at-Rest analysis.
Figure 2 Architecture of the system
Figure 3 ISAB attached to different devices
The developed system allows prediction and detection of machine malfunctions before a critical failure occurs, which creates obvious benefits for the manufacturing companies. This has been confirmed by 3 external companies willing to become early adopters and rent out the next iteration for test trials in their facilities. The technology provider – MASTA – sees the system as a breakthrough product, which will support the intended transition from a service-based integrator to a knowledge-based product provider and is willing to continue investing in its further development.