DVel´s AI exploration

Important results from the project

Through a close collaboration with ESS, the project have in this project worked towards the implementation of ML algorithms in the integrated control system at ESS. This has been fulfilled by learning EPICS, which is a set of software tools and applications used to develop and implement distributed control systems for research facilities and industry.

Expected long term effects

The project resulted in DVel gaining increased knowledge in machine learning (ML) and a toolbox for applying ML to our customer projects. Our customers have a great interest to start using AI and we can after this project better show them the opportunities that come with ML. Now we have a reference project and results to show. ML will add value to our customers, for example by predicting maintenance needs, required to maximize process up-time. We can, with ML, deliver insights that strengthen our customers competitiveness.

Approach and implementation

The project used the Tensorflow and Keras libraries in Python to build several neural networks and to train the ML models. For example, LSTM, LSTM-FCN and CNN were used. Data from ESS was preprocessed in Python to make it ready for ML training. We simulated a process sequence in EPICS using data from ESS, that were unknown to the model. PYEPICS was used to integrate Python into the EPICS environment. In this way we were able to predict future alarms before they occurred.