Explicação Online de Falhas para Manutenção Preditiva

The present proposal aims to develop an effective system for online predictive maintenance with explanations. To achieve the broader goal, the following partial objectives were established. 1. Enhance methods for failure detection in an online operational context. Develop new methods that can better adapt to changing operational behaviours in real-time, thereby enhancing failure detection accuracy under varying conditions. This will be accomplished by leveraging clustering and novelty detection methods to identify normal operational regimes that can continuously learn and adapt to the operational behaviour of equipment in real-time. This will help distinguish between various 'normal' operational regimes (modalities) and transfer of models between different buses. 2.Predicting failures for different time-horizons. This objective aims to improve the accuracy and the timely prediction of failures in operational contexts. For this, we will explore different multi-target deep-learning methods based on autoencoders and transformers to forecast failures. 3. Develop a new online regression rules algorithm for imbalance regression to explain fault symptoms captured by the failure detection algorithms. 4. Explore online rule-based explanation methods for failures detected in an operational context. Provide clear, understandable and actionable explanations for detected failures, making predictive maintenance systems more transparent and trustworthy to maintenance teams. Express explanations in natural language. 5. Establish an evaluation framework to assess the effectiveness of failure detection methods. Use interval temporal logic to define metrics and evaluation strategies to estimate the performance of PM systems and determine how reliable and effective they are in predicting failures. 6. All the methods developed in the context of the objectives above will be tested on the PM case study. The ultimate goal consists of developing an online and explainable predictive maintenance for the pneumatic system of electric buses. This project aims to contribute to the field of PM not only by enhancing failure detection capabilities but also by making an effort to ensure that such systems can be used effectively and trusted by the maintenance teams responsible for operational decisions.