Evolvable Artificial Intelligence for Predictive Maintenance

Purpose and goal:

The most accurate AI-based predictive maintenance techniques require training data abound past equipment failures or a saved history of equipment´s health trends. This type of data is rarely available in sufficient quantities because equipment should not fail.

In this project, we are going to design an evolvable AI framework for predictive maintenance. The framework should enable models to be improved systematically highly degree automated by using data from other similar equipment, data from anomalies investigated by domain experts and synthetic data. Expected results and effects:

The project is expected to provide an innovative method for the common situation of having only insufficient training data for the introduction of predictive maintenance techniques. The method then enables these techniques to evolve and improve consistently. As the intention is to implement the concepts based on existing open-source MLOps platforms and to make prototypes available to other Swedish companies, the project is expected to benefit the Swedish manufacturing industry as a whole.

Approach and implementation:

The framework will be developed in an iterative process and will be based on an integration of techniques combining state-of-the-art predictive maintenance techniques, probabilistic modelling, transfer learning and synthetic data. A prototype of the framework will be implemented and evaluated for a use case from the forest and paper industry.