Multi-Channel Fault Signal De-Noising Method for Wind Turbine Generators Based on Tensor Train Decomposition

With the continuous advancement of sensor and information collection technologies, data types are increasingly diverse, while the amount of data grow exponentially. For mechanical components such as wind turbine generators (WTGs), the use of a single sensor for fault diagnosis has limitations, and making it challenging to fully and simultaneously perform auxiliary fault diagnosis on gears, bearings and other components on each shaft. However, the use of multi-channel sensors can solve this problem well, but in actual engineering applications, there is noise interference when multi-channel sensors perform joint fault diagnosis, hindering the accurate extraction of rotating machinery fault characteristics. This paper proposes a feature extraction method based on tensor train fused signal (TTFS), it overcomes the shortcomings of traditional methods, which can only diagnose single-channel composite faults, providing a new approach for rotating machinery composite fault diagnosis. Additionally, the proposed method innovatively combines “big data" with multi-channel fault signals to construct a high-dimensional tensor structure signal. Initially, our focus is on data collection, during which we analyse data in tensor format across three dimensions: time, frequency, and channel. To emphasis the use of tensor train decomposition (TT-decomposition), we introduce a tensor reconstruction method with adaptive filter truncation. Subsequently, the continuous wavelet transform (CWT) method is employed to establish tensor data representation across these dimensions. Gradually, the proposed adaptive filtering truncation tensor reconstruction method is employed to reconstruct the tensor by combining the aforementioned matrices. Finally, the continuous wavelet inverse transform is applied to the reconstructed tensor in order to obtain time-domain signals from different channels. The efficacy of the proposed method is demonstrated through experimental signals, which illustrate the advantages of the method and the improved convergence speed of the objective function.