Diesel engine diagnosis based on entropy of vibration signals and machine learning techniques

Abstract

Abstract Compression‐ignition (CI) engines, aka diesel engines, are responsible for an essential percentage of the world‐polluting emissions. Moreover, bearings installed in industrial machinery constitute the most common failure affecting global energy consumption. Since industries’ energy demand has a growing tendency, efficient maintenance is a must. Maintenance requires a fast and accurate diagnosis, commonly based on an intrusive or expensive sensor to capture monitoring signals, i.e. pressure, emissions, temperature, fuel consumption and rotational speed. Here, a vibration signal‐based approach is introduced to combustion engines and bearings diagnosis. Namely, a multi‐scale permutation entropy (MPE)‐based feature extraction is conducted within a variability‐based relevance analysis (VRA) stage to feed a straightforward classifier, the K‐nearest neighbours (KNN). Accuracy was validated using a signals’ database from a single‐cylinder engine under multiple work conditions. Also, the methodology is compared through classification accuracy of a widely known bearing vibration signal database obtaining an outstanding performance.