Evolvable Artificial Intelligence for Predictive Maintenance

Predictive maintenance has the potential to reduce the efforts and costs associated with the maintenance of industrial production systems. Through the use of AI technology trained to interpret sensor data from the production process, equipment can be more efficiently utilized since it is used until its predicted end-of-life and therefore costly, premature, replacements can be avoided. This benefits both the economic and ecological sustainability of operating production systems.

The most accurate AI-based techniques for predictive maintenance require labeled training data about, e.g., equipment failure in the past, or recorded histories of equipment health trends. This type of data is scarce in many use cases as machinery is not supposed to fail in the first place.

In this project, we intend to design an evolvable AI framework for predictive maintenance. A core element of this approach are continuously improved models. In a systematic and highly automated fashion, such improvement over time is accomplished by, e.g., exploiting data from similar equipment, data about operation anomalies labelled by experts and synthetic data.

The goal is to develop a platform that can easily be adapted by companies that want to benefit from AI-based predictive maintenance despite a lack of labeled training data. The platform will be demonstrated in the context of the forest-/paper industry.