Fault Diagnosis Of Locomotive Gearbox Based On Information Fusion Technology

Locomotive fault diagnosis and overhaul are an important part of locomotive maintenance,in the meantime,locomotive gearbox is also the core component of locomotive transmission.The status of gearbox directly determines the locomotive quality.Therefore,accurate diagnosis of the fault of locomotives gearbox has become an urgent need in the railway industry.It is a new trend to improve the accuracy of fault diagnosis of locomotive gearbox by integrating the result of the oil analysis of the gearbox of the locomotive with the vibration analysis.The development and application of artificial intelligence are the growing trend of various industries,so it is the fault diagnosis of railway locomotives.BP neural network has great ability of self-learning,self-adaptive,fault-tolerant and ability to deal with nonlinear problems.And D-S evidence theory has the ability to process fuzzy information and unstable information.These two theories are the classical method of AI and machine learning.Combining neural network and evidence theory has great application prospects in fault diagnosis.In this paper,when oil diagnosis technology and vibration diagnosis technology were used,the BP neural network was used to get the initial diagnosis to obtain the oil diagnosis results and vibration diagnosis results;then the DS evidence theory was used to obtain the results of decision-level fusion,and finally more accurate diagnostic results were got.Based on an in-depth understanding of the structure,material and failure mechanism of the locomotive gearbox,the following research work was mainly done in this paper:Firstly,using atomic spectroscopy and direct-reading ferromagnetic spectrum technology,the types and sizes of abrasive elements in the gearbox oil under different failure modes were extracted.Then,based on these data,a BP neural network model of improved algorithm was established.And the author used it to test data and get the diagnosis results.It can be seen from the results that the oil method cannot accurately determine the gear root crack fault.Secondly,after the vibration data of each failure mode of the gearbox is obtained.The data is subjected to trend elimination items and noise reduction processing.The time domain parameters and frequency domain parameters in the fault data are extracted,and the BP neural network of the improved algorithm is established using the obtained data.In the model,the output of three vibration subnetworks is obtained.The result shows that there is a misjudgment of the normal state of the gearbox and bearing failure,and there is a certain ambiguity in the recognition of the root cracking fault.Thirdly,the D-S evidence theory was used to fuse the vibration data.From the fusion result,although the misdiagnosis and missed diagnosis did not occur,the failure support was low and the uncertainty was high;then the initial fusion result and oil were obtained.The results obtained by the neural network are merged at the decision level;finally,the fusion diagnosis results are obtained.From the final result,the uncertainty of the diagnosis is obviously decreased.The misdiagnosis phenomenon does not occur in the entire fault diagnosis,and the accuracy of the diagnosis is also greatly improved.