Fault Diagnosis Platform For Equipment Based On Deep Learning

With the rapid development of sexual technology,the structure of equipment is sophisticated and complex,and traditional diagnostic methods for industrial equipment failures have encountered many challenges in terms of intelligence and timeliness.Some researchers used deep learning algorithms to process the identification and achieved results,but there are still many failures,such as some failures in the captured equipment,including many network models that cannot accurately identify failures.They all have model design experience in the field of image processing,and high hardware configuration is required for model deployment,which is not suitable for some actual industrial scenarios.Since conventional neural network models cannot accurately identify faults in complex environments or have low recognition accuracy,this paper combines the advantages of traditional signal analysis and deep learning to first preprocess the collected one-dimensional data to generate a two-dimensional matrix,and then The input is trained in the fault identification network constructed based on the SK convolution block to achieve fault classification.The experimental accuracy on the Uo C gear data set is about 5% higher than that of similar networks without SK convolution;on the PU bearing data set,the accuracy rate is as high as 99.5%.Using T-SNE to visualize the output features,it can be found that there is a significant improvement.The degree of aggregation of similar samples.Due to energy consumption requirements and resource limitations of edge devices,some scenarios require a lighter network model.This paper first analyzes the principles of traditional convolution and depthwise separable convolution,and compares the parameters of the two.The separation convolution is applied to a onedimensional convolutional network,and a channel attention module is introduced to build a lightweight fault identification model.Experiments are conducted on the Uo C and PU datasets,and the number of parameters is reduced by 62% compared to the one-dimensional Res Net18 network with almost no reduction in fault recognition accuracy.Combined with the deep learning model introduced above,this paper designs and implements a fault analysis and intelligent diagnosis platform,which is convenient for maintenance staff to locate the cause of the fault in time.The platform can monitor equipment operating parameters in real time,and integrates a variety of algorithm models to automatically complete fault identification tasks for different equipment and practical problems.