Intelligent diagnosis of rotating machinery using dynamic mode decomposition and sequential backward selection of multi-sensor features

This study introduces an innovative framework for intelligent condition monitoring of bearings and gears, employing Dynamic Mode Decomposition (DMD) on multi-sensor data that includes vibration, electrical, and acoustic measurements. To our knowledge, this is the first work to explore the application of DMD for fault diagnosis in this domain. The methodology begins with a preprocessing stage, where DMD is applied to extract dynamic features that capture the intrinsic behaviors of the monitored components. From the reconstructed signals generated via DMD, a comprehensive set of statistical and time–frequency indicators is derived. To enhance diagnostic performance and minimize feature redundancy, the Sequential Backward Selection (SBS) algorithm is implemented, yielding a compact yet informative feature subset. These refined features are subsequently used as inputs to various intelligent classification models for fault detection and categorization. The proposed approach achieves an impressive diagnostic accuracy of 99.20%, demonstrating strong robustness and generalizability. Validation is carried out using four distinct datasets, two vibration-based, one acoustic, and one electrical, that cover different operational scenarios and sensor modalities. The results substantiate the effectiveness of the proposed framework in providing accurate and reliable health evaluations of the components of rotating machinery.