Explainable AI For Predictive Maintenance
The XAIPre project (pronounce “Xyper”) aims at developing Explainable Predictive Maintenance (XPdM) algorithms that do not only provide the engineers with a prediction but in addition, with 1) a risk analysis should the maintenance be delayed 2) the criteria or indicators used to make that analysis. By providing more insight into the state of the machine, the engineers are empowered and given control over their maintenance process.
In practice, this might look like this: an engineer would usually maintain the equipment on an offshore location once a week, depending on vessel schedules. Should a vessel schedule suddenly change, the engineer can use the technology to assess the equipment’s current condition. The sensors in the machinery can provide data about key indicators, such as heat or friction, to the algorithm. The algorithm then provides the engineer with a risk analysis (e.g. maintenance is required before the vessel is scheduled to depart) and the key indicators that influence this analysis (e.g. a component is heating up faster than is ideal).
XAIPre is focused on maintenance in the maritime industry. Current maintenance concepts in the maritime industry are based on a fixed maintenance interval with a significant safety margin to minimize incidents. As a consequence, maintenance is always carried out too early making it one of the most inefficient industrial activities and most critical at the same time.
The research partners will work together with Heerema Marine Contractors. This cooperation will focus on the predictive maintenance of thrusters. After the initial focus in this project on the thrusters, the research can be applied to the whole fleet and other components. This will help Heerema Marine Contractors to improve their maintenance schedules, driving down costs and increasing the sustainability of machines and worker safety.
The project team is advised by a user committee, representing Heerema Marine Contractors and different industries with similar components and objectives in predictive maintenance, including ship design (C-Job Naval Architects), steel production (Tata Steel Europe), aircraft engines (Airfrance/KLM) and automotive (Volkswagen Group) and Honda Research Institute Europe), making sure that the technology developed in the project can be applied across multiple industries.
GSAreport is an application to easily generate reports that describe the global sensitivities of your input parameters as best as possible. You can use the reporting application to inspect which features are important for a given real world function / simulator or model.