AI to Predict and Explain Machine Maintenance Needs

The Grand Solutions project P³AI₄ – Predictive and Prescriptive Process Analytics for Industry 4.0 will combine industrial research based on the interdisciplinary field of process mining, which lies at the intersection of AI, data science, and process science. Denmark is a global leader in process mining research, and in this project, researchers from the Department of Computer Science at the University of Copenhagen and DTU Compute will develop algorithms capable of handling real-time data that can be applied across industries working with automated data collection in their systems.

- Our process mining algorithms have won several international awards but have so far primarily been used to describe and improve business processes retrospectively. With P³AI₄, we can demonstrate that the technology can also predict machine failures in real time, preventing costly breakdowns, says Tijs Slaats, Project Lead and Associate Professor at the Department of Computer Science.

Intelligent applications such as predictive and recommendation systems are considered strategic technology trends in the future of AI.

Great Potential

One of the project’s three industrial partners is Metroselskabet, whose automatic train control system logs approximately 45 million daily events. The project will support a general shift toward real-time monitoring and proactive maintenance solutions.

- We see great potential in the project to help us generate new insights and better utilize the enormous amounts of data from our automated systems. We constantly strive to improve operations, and the project will enable the development of new data-driven and explainable methods to enhance operations and maintenance, reducing the number of service disruptions in the Metro, says Jan Schelhaas, project participant from Metroselskabet.

Through this project, Everllence will be able to predict breakdowns and failures in their modern ship engines and offer spare parts when needed.

DCR Solutions contributes by developing predictive and prescriptive process models and user-centered software solutions.

Major Challenges

Currently, the biggest obstacles to fully exploiting available data are the need for critical domain knowledge and human factors. For example, the fact that large events significantly increase the risk of door failures on a given evening in the Metro is difficult to integrate into current predictive AI models, and their predictions cannot be explained to users.

These challenges are not unique to Metroselskabet but represent a global societal challenge: how to effectively utilize the massive amounts of data generated by Industry 4.0 systems to deliver accurate and explainable predictions and prescriptions to end users.