Research On Sub-health Recognition Algorithm Of Capsule Network Optimized Hierarchical Convolution

The failure of rolling bearings in large industrial equipment will not only cause economic losses,but more importantly,it will endanger people’s lives.With the advancement of technology and rapid development,it has become common to obtain a large number of vibration signals from sensors.So,analyzing the vibration signal of the sub-health sensor of the rolling bearing of large industrial equipment,predicting the possible failure,can promptly warn the state of the equipment or replace the equipment parts,reduce the hidden dangers.It makes sense both for the industry and the operator of the industrial equipment.Therefore,many experts and scholars pay attention to and conduct in-depth research on sub-health recognition algorithms.After analyzing and comparing the results of scholars’ sub-health research in recent years.Aiming at the problems of time-consuming,laborious and low diagnostic accuracy of traditional algorithms feature extraction.This paper proposes a sub-health recognition algorithm that improved capsule network to optimize the hierarchical convolution.The algorithm is based on Convolutional Neural Network(CNN)and Capsule Network(CAPSNET).First of all,to solve the problem that traditional CNN continuously stacks convolutional layers and pooling layers to improve the accuracy of bearing sub-health recognition,resulting in a large amount of training time.On the one hand,a hierarchical CNN is proposed,and the data is input into three parallel convolutions kernel of different sizes for processing,multi-angle extraction of sub-health information in the signal.On the other hand,there are two noise reduction processing methods for the data,namely wavelet denoising method and wavelet packet denoising method,to better retain the useful sub-health information in the original vibration signal.A multi-input hierarchical CNN sub-health recognition model is established,and the proposed model can perform sub-health recognition based on vibration signals.Then,for the problem that CNN cannot identify the whole according to the part,which affects the final sub-health recognition result,the original CAPSNET has huge update parameters,and the capsule vector features are redundant.The convolution result is input to the improved CAPSNET for sub-health recognition.The improved CAPSNET adopts a pruning mechanism.When the content expressed by the low-level capsules and the high-level capsules is inconsistent,the value of the coupling coefficient will become very small.Trim the coupling coefficient according to the threshold.At the same time,the corresponding weight parameters are trimmed to reduce the CAPSNET calculation cost to improve performance.The improved capsule network sub-health recognition algorithm can output sub-health recognition results faster according to the relationship between the part and the whole.Finally,the sub-health recognition algorithm of the improved capsule network optimized hierarchical convolution proposed in this paper is verified by the measured bearing data set of Western Reserve University.According to the experimental result data and comparative experimental result data.It is found that the improved algorithm can effectively identify rolling bearings health status,verifying the effectiveness of the improved algorithm in this paper.