Intelligent fault diagnosis method for rotating machinery via dictionary learning and sparse representation-based classification

Wind power has developed rapidly over the past decade where study on wind turbine fault diagnosis methods are of great significance. The conventional intelligent diagnosis framework has led to impressive results in many studies over the last decade. Despite its popularity, the diagnosis result is affected severely by the feature selection and the performance of the classifiers. To address this issue, a novel method to diagnose wind turbine faults via dictionary learning and sparse representation-based classification (SRC) is proposed in this paper. Dictionary learning algorithm is capable of converting the atoms in the dictionary into the inherent structure of raw signals regardless of any prior knowledge, indicating that it is a self-adaptive feature extraction approach, which avoids the challenge of feature selection in traditional methods. Next, recognition and diagnosis can be solved by the simple SRC without additional classifier, exploiting the sparse nature that the key entries in sparse representation vector are assigned to the corresponding fault category for a test sample. The validity and superiority of the proposed method are validated by the experimental analysis. Moreover, we find that, in terms of robustness under variable conditions and anti-noise ability, the performance of the proposed method always significantly outperforms the traditional diagnosis methods, leading to a promising application prospect.