Order spectrum-assisted sparse learning classification approach for wind turbine drivetrain fault diagnostics under variable operating conditions

As the crucial equipment for renewable energy power generation, wind turbines always work under variable operating conditions, making it difficult to identify potential failures promptly. Drivetrain system is the guarantee for reliable and safe operations of wind turbines, but drivetrain fault diagnostics under variable operating conditions remains a challenging problem for diagnostics and maintenance of wind turbines. This paper proposes a novel order spectrum-assisted sparse learning classification (OS-SLC) approach for wind turbine drivetrain fault diagnostics under variable operating conditions. First, order tracking is incorporated into signal feature pre-extraction for order spectrum dictionary design initialization. Then, dictionary learning is developed to excavate latent features within order spectra, facilitating effective sparse representations of various health status. Subsequently, the order spectral sparse representation classification strategy based on minimum sparse approximation error is developed, which enables robust fault diagnosis without any discriminative classifier models. Experiment results from transfer diagnosis tasks of wind turbine drivetrain datasets under variable operating conditions verify that the proposed OS-SLC approach shows remarkable diagnostic accuracy comparable to mainstream domain adaptation methods, whilst demonstrating the diagnostic robustness against noises and hyper-parameters. By integrating order tracking with sparse representation learning, our proposed approach demonstrates the potential for fault diagnosis under nonstationary conditions.