Logical analysis of data in predictive failure detection and diagnosis

Abstract

PurposeThis study aims to address the critical issue of machine breakdowns in industrial settings, which jeopardize operation economy, worker safety, productivity and environmental compliance. It explores the efficacy of a predictive maintenance program in mitigating these risks by proactively identifying and minimizing failures, thereby optimizing maintenance activities for higher efficiency.Design/methodology/approachThe article implements Logical Analysis of Data (LAD) as a predictive maintenance approach on an industrial machine maintenance dataset. The aim is to (1) detect failure presence and (2) determine specific failure modes. Data resampling is applied to address asymmetrical class distribution.FindingsLAD demonstrates its interpretability by extracting patterns facilitating the failure diagnosis. Results indicate that, in the first case study, LAD exhibits a high recall value for failure records within a balanced dataset. In the second case study involving smaller-scale datasets, enhancement across all evaluation metrics is observed when data is balanced and remains robust in the presence of imbalance, albeit with nuanced differences in between.Originality/valueThis research highlights the importance of transparency in predictive maintenance programs. The research shows the effectiveness of LAD in detecting failures and identifying specific failure modes from diagnostic sensor data. This maintenance strategy exhibits its distinction by offering explainable failure patterns for maintenance teams. The patterns facilitate the failure cause-effect analysis and serve as the core for failure prediction. Hence, this program has the potential to enhance machine reliability, availability and maintainability in industrial environments.