Fault detection in axial piston hydraulic pumps: integrating principal component analysis with silhouette-based cluster evaluation

This paper presents an approach integrating principal component and silhouette analysis with clustering algorithms for fault detection in hydraulic systems. The methodology was validated through a study in which vibration and pressure signals were collected under normal and fault conditions. These signals were then processed through filtering and normalization, followed by dimensionality reduction using principal component analysis. The resulting lower-dimensional feature vectors retained the critical characteristics of both normal and faulty conditions and were subsequently fed into a clustering algorithm. The quality of the resulting clusters was evaluated using silhouette analysis, which offers a reliable means of assessing cluster quality and visualising the outcomes of fault classification. The study demonstrates the effectiveness of this method in accurately representing the patterns of normal and malfunctioning hydraulic pump conditions, ultimately leading to successful diagnostic results.