Gas turbine rotor fault diagnosis method based on FFT-CNN transformer and cross attention mechanism

Gas turbines, as an efficient energy conversion device, are widely used in power generation, aviation, and industrial fields, providing reliable energy and power solutions. A gas turbine rotor fault diagnosis method based on Fast Fourier Transform (FFT), Convolutional Neural Network (CNN) - Transformer, and Cross Attention mechanism is proposed, which has a complex structure, diverse operating conditions, and a small number of fault samples during actual operation, resulting in low fault diagnosis accuracy. Firstly, FFT is used to transform the one-dimensional fault signal, so that the transformed frequency domain signal and its own time domain signal are respectively processed through CNN convolutional pooling layers to extract global features. Then, Transformer extracts temporal features, uses cross attention mechanism to fuse time domain and frequency domain characteristics, calculates attention weights, and finally outputs fault classification results through fully connected layers and softmax. This article uses two types of publicly available bearing data, achieving 100% and 97.2% accuracy, respectively. Using the bearing fault dataset as the source domain and the gas turbine rotor fault dataset as the target domain, the transfer learning method is used to achieve 98.54% accuracy in classifying gas turbine rotor faults.