Fault Diagnosis of Rotating Machinery Based on Local Centroid Mean Local Fisher Discriminant Analysis

Purpose Currently, the vibration signals collected from rotating machinery are high dimensional and massive, how to extract sensitive fault characteristic information to make the fault diagnosis results more accurate and reliable is the main task. Methods In this paper, a novel feature reduction method local centroid mean local fisher discriminant analysis (LCMLFDA) is proposed to obtain more inherent sensitive fault information from the multi-class and high-dimensional mechanical operating data. First, the vibration signals are converted into a multi-domain statistical feature dataset. Second, the constructed feature dataset is trained by the proposed LCMLFDA algorithm, which is a novel manifold learning algorithm. Finally, the improved k-nearest neighbor classifier (IKNNC) is used to diagnose the health condition of bearings and gearboxes. Results The proposed LCMLFDA takes into account the cohesiveness and separation of samples while maintain the local geometric structure information of the samples and reflecting the nearest neighbor relationship between sample and the local centroid mean. Extracting the mixed features of the measured vibration signal from multiple angles. The improved IKNNC is used to replace the Euclidean distance with a nearest neighbor probability distance to select the nearest neighbor points, which achieves a better classification effect. Conclusion The effectiveness and feasibility of the proposed method is demonstrated using two case studies. The experimental results demonstrate that this fault diagnosis method can achieve better performance.