An enhanced sparse representation-based intelligent recognition method for planet bearing fault diagnosis in wind turbines

Abstract

Fault diagnosis techniques are vital to the condition-based maintenance strategy of wind turbines, which enables the reliable and economical operation and maintenance for wind farms. Due to the complex kinematic mechanism and modulation characteristic, planet bearing is the most challenging component for fault diagnosis in wind turbine drivetrains. To address this challenge for planet bearing fault diagnosis, we propose an enhanced sparse representation-based intelligent recognition (ESRIR) method, which involves two stages of structured dictionary designs and intelligent fault recognition. In the first stage, the structured dictionary designs are achieved with the overlapping segmentation strategy, which exploits the strong periodic self-similarity and shift-invariance property in planet-bearing vibration signals to enhance the representation and discrimination power of ESRIR. In the second stage, the intelligent fault recognition of planet bearings is implemented with the sparsity-based diagnosis strategy utilizing the minimum sparse reconstruction error-based discrimination criterion. Finally, the applicability of ESRIR for planet bearing fault diagnosis has been validated with the wind turbine planetary drivetrain test rig, demonstrating that ESRIR yields the superior recognition accuracy of 100% and 99.9% for diagnosing three and four planet-bearing health states, respectively. Comparative studies show that ESRIR outperforms the deep convolution neural network and four classical sparse representation-based classification methods on the recognition performances and computation costs.