Key developments from MODUL4R: From Swarm Intelligence to DataCrop Platform and System Integration
Our team has made significant progress across several MODUL4R tasks.
In the area of swarm intelligence, machine learning models have been developed to enable collaborative learning across different manufacturing nodes, similar to how swarm systems work in nature. Our MODUL4R consortium partners at Netcompany - Intrasoft (INTRA) have been working with two pilot cases, namely : FFT's PCB assembly process where component placement quality using synthetic datasets is analyzed, and NECO's tap placement robot where UNINOVA's Asset Administration Shell (AAS) data to detect operational status through sound analysis is leveraged. Notably, the LSTM model (a type of neural network) for the NECO case has shown particularly promising results, achieving high accuracy in detecting robot status. To make these AI systems more understandable and user-friendly, features that help explain their decisions to users have been added.
Training the LSTM model with AAS synthetic data for the NECO use case gives a Validation Accuracy of 97% and Testing Accuracy of 96%
The DataCROP platform has also been equipped with a new workflow editor, making it easier to set up and configure data analytics for manufacturing processes. The platform is currently supporting three pilot cases:
FFT's leg cutting process during PCB assembly (using v1.0 swarm learning synthetic dataset),
NECO's tap placement robot status recognition utilizing UNINOVA's AAS data (also using v1.0 swarm learning synthetic dataset),
EMO's predictive maintenance with vibration analysis.
MODUL4R’s Workflow editor module defines in a graphical way the components involved and the sequence of execution of an AI pipeline. The Editor can manage the data sources (inputs/outputs), analytics processors and infrastructure resources (workers). The editor exports a building and deployment plan that is used by the Airflow module.
On the MODUL4R platform side, a robust integration infrastructure that helps connect different parts of the MODUL4R system has been established, making it easier for project partners to work together and deploy solutions. This includes setting up development and deployment tools, security measures, and monitoring capabilities. Several integration workshops with partners to ensure smooth collaboration across the project have been conducted as well.
