Real-time Fault Diagnosis Approach Based on Lifting Wavelet and Distance Seeking Clustering

Abstract

The routine ensemble re-time fault diagnosis method and its upgrading alternatives were found with some problems as these: Single detection method causes a high rate of failure. Unable to remove the noise in the data. The basic calculation of signal method use FFT algorithm. Impossible to separate frequency property and time propert of industrial signal. The basic clustering method use the traditional K-means clustering, their disadvantage are the initial uncertainty and Dimension Trap, K-means cluster defects lead to the overall computational efficiency. In regard to these problems, a new ensemble re-time fault diagnosis method was discussed, this method base on lifting wavelet(LW_ learning) and distance seeking clustering(DBIFWALAMDSC). Firstly, data is denoised by LW transform in real-time. Then LW_learning to identify signal to identify frequency property and time property of industrial signal. The algorithm of feature learning Indirect Feature Weight to eliminate Dimension Trap, DB_index algorithm to solve the initial uncertainty, Distance Seeking Clustering is adopted to protect more rapid convergence and improve the clustering performance. The experiments show that DBIFWALAMDSC algorithm is effective and feasible. DBIFWALAMDSC algorithm can monitor time-varying process. DBIFWALAMDSC is superior to the traditional single FFT Clustering in diagnosis precision, rate and adaptability.