Quantitative condition assessment method for rotating machinery using fuzzy neural network

Abstract

Health condition assessment of rotating machinery has been a persistent challenge. Traditional condition assessment methods often rely on single features, limiting their application to comprehensively measure the health condition of rotating machinery. This study introduced a quantitative condition assessment method for rotating machinery using fuzzy neural network (FNN). Initially, multi-domain features of signals from rotating machinery are extracted to achieve comprehensive representation of signals in the feature space. To eliminate redundant information of various features, a feature dimensionality reduction method is explored based on variance variation and stacked auto-encoder (SAE). Afterward, a normalized health indicator is constructed by integrating the optimized features through FNN, and it can indicate the current conditions of rotating machinery. Furthermore, an early anomaly alarm strategy based on 3σ criterion is designed for rotating machinery. The abnormal signal will be recognized automatically when it exceeds the predetermined thresholds. Last, the effectiveness of the proposed method is validated on IMS bearing dataset and XJTU-SY bearing dataset. The results show that the proposed method can effectively obtain the quantitative indicators that reflect the operation conditions of rotating machinery and can accurately detect the early abnormal signals.