Machine Learning-Based Predictive Maintenance using Data Aggregation via Regularized Clustering

The paper is devoted to the development of a technique for classifying failures in predictive maintenance based on machine learning on aggregated data. A feature of the proposed approach is the preliminary clustering of data in order to reduce the dimension of the trained model based on kernels. A regularized ‘soft clustering’ procedure is proposed, using a regularizing functional that takes into account the imbalance between the number of cluster elements corresponding to the presence of failures and their absence. A cluster-learning kernel regression model of the failure rate is also proposed. The performance and efficiency of the proposed technique are illustrated by the results of computational experiments carried out using the synthetic AI4I Predictive Maintenance Dataset.