基于RepLKNet-BiGRU的起重机齿轮箱故障诊断方法OA

To address the issue that crane gearboxes are prone to failures under complex operating conditions and that traditional diagnostic methods have limited feature extraction capability,a multi-modal fault diagnosis model integrating two-dimensional images and one-dimensional vibration signals is proposed.The model first converts the raw vibration signals into 2D image representations with global temporal dependencies using the Gramian Angular field method,thus preserving the evolutionary features of the time series in the visual space.Subsequently,the model employs RepLKNet and BiGRU-Global Attention structures for deep extraction of features from different modalities,respectively.Among them,RepLKNet is able to fully mine the spatial pattern and local structure information of the image by virtue of its large receptive field convolution structure.BiGRU network captures the before-and-after temporal correlation of the signal through the bi-directionally gated recurrent unit and highlights the important dynamic changes of the key time segments by combining with the global attention mechanism.After the two modal features are fused at the feature layer,the accurate identification of different fault types of gearboxes is achieved through the fully connected layer with Softmax classifier.The experimental results show that the proposed model exhibits excellent classification accuracy and robustness with a 98.47%accuracy under multiple typical gearbox fault conditions,which is significantly better than the single-modal or traditional deep learning models.