Fault Diagnosis Algorithm Of Ranging Arm Transmission System Of Shearer Based On Deep Transfer Learning

Ranging arm transmission system is the core power transmission mechanism of shearer,and its reliability is of great significance to efficient coal production.Traditional shearer fault diagnosis methods mainly use regular maintenance or manual analysis of fault information for experience judgment,which is time-consuming and laborious and the fault recognition is subjective.With the acceleration of intelligent mine construction,the amount of data that can be collected by underground shearer is greatly increased,and the massive shearer operation data can provide more adequate information for shearer fault diagnosis.Because of the deep network structure,deep learning can automatically represent the feature distribution inside the data after deep mining of massive shearer operation data and further directly establish the relationship between the learned features and the target output to realize the accurate fault diagnosis of shearer.However,shearer fault diagnosis under actual working conditions is faced with many problems such as no standard data,scarce typical fault state data and difficult to guarantee data quality.Transfer learning,as a brand new learning paradigm,can apply existing knowledge to solve different but related problems,so that the model can draw inferences from one another.It is expected to overcome the challenges faced by deep learning-based intelligent fault diagnosis in practical application of shearer.Therefore,this paper focuses on the deep transfer learning algorithm,proposes the transfer algorithm solution to the fault diagnosis problem of the shearer’s cross-working condition transfer and cross-equipment transfer.The main content includes the following three aspects:(1)Aiming at the problem of transfer fault diagnosis across working conditions,the fault data feature distribution difference between source domain and target domain is small,so a fault diagnosis model based on parameter transfer is proposed.The deep neural network model TCNNG was constructed to automatically extract the feature information from the condition data set in the source domain with abundant fault sample information.Further,the strategy of parameter transfer and fine tuning was adopted to transfer the diagnostic knowledge parameters acquired from the source domain data to the condition data in the target domain,so as to improve the fault recognition rate and generalization performance of the model under different working conditions.(2)For the problem of cross-device transfer fault diagnosis,the feature distribution of fault data in source domain and target domain is quite different,and direct parameter transfer cannot accurately and effectively identify the fault in target domain.A deep subdomain adaptive fault diagnosis model based on feature transfer is proposed.By constructing the local mean maximum difference module and embedding it into the fully connected network layer of the deep learning model,the global data feature distribution difference of the fault data in the source domain and the target domain and the feature distribution difference of the two domain subclasses are measured,and the distribution alignment of the related subdomains is carried out.(3)Inspired by the adversarial learning mechanism of generating adversarial network,a dynamic adversarial transfer fault diagnosis model is proposed.The feature extractor is regarded as the generator in the adversarial network,and the distribution measure function is regarded as the discriminator in the adversarial network.At the same time,an adaptive factor is constructed to dynamically measure the importance of the marginal distribution and conditional distribution in the source domain and target domain data.The adversarial learning mechanism is introduced to train the model,which effectively reduces the difference of data feature distribution between the source domain and target domain.The fault diagnosis ability of the model is improved.The vibration signals of MG500/1170-AWD1 type shearer ranging arm drive system gear fault were collected by the shearer ranging arm loading test bench.The experimental results show that the proposed method has good classification accuracy and generalization ability in the transfer task of shearer fault diagnosis,which expands the existing research methods of shearer fault diagnosis.It provides a new idea and means for the fault diagnosis research of the ranging arm drive system of shearer under the complicated condition of scarce fault data and variable working conditions.The proposed transfer fault diagnosis method can effectively solve the problem of insufficient effective fault data samples of shearer and fault diagnosis under the condition of cross-working conditions and cross-equipment.In the actual shearer fault diagnosis process,the appropriate fault diagnosis method can be selected according to the actual situation.When there is fault data of ranging arm of a single shearer,the parameter transfer method can be used for fault diagnosis of the same shearer,and the feature transfer method can be used for fault diagnosis of different shearer models.When the data is scarce or the structure of different shearer models is very different,and the adversarial transfer method can be used.