Explainable Machinery Faults Prediction Using Ensemble Tree Classifiers: Bagging or Boosting?

Abstract

Proactive maintenance aims to accurately classify temporal trends as early as possible, detect faulty states, and pinpoint the root cause of faults. Neither late nor early maintenance is desirable, as each will incur additional operating costs. While various data-driven techniques have been used to identify faults, many fail to perform when faced with missing values at run time or lack explainability. The present work introduces a framework to identify and classify anomalous signals in spite of missing values and partially labeled data. This framework offers explainability by identifying key performance indicators for each fault family using SHapley Additive exPlanations (SHAP). A comprehensive study was performed on existing algorithms to determine the best fault classifier, and candidates with accuracy greater than 80% were selected. This paper introduces a new missing value imputation technique based on Partial Least Squares (PLS-MV). It also uses fuzzy C-means (FCM) to detect different healthy unlabeled operations in the PHME21 dataset. Our results show that boosting algorithm is best suited for creating a generalized model that is capable of classifying faulty patterns in multi-label datasets which may include missing values.