Fault Diagnosis Research Of Aviation Gear Based On Weighted Domain Adaptation Network

Aeroengine is the important symbol to measure the country’s core competitiveness.As a key component of the aeroengine transmission system,planetary gears are usually affected by harsh environments during actual work,and gear components are prone to failure,which will affect the normal operation of the engine.Therefore,it is of great significance to carry out convenient and timely fault diagnosis on aviation gears.In order to solve the problem of transfer diagnosis of aviation gear faults under different engineering backgrounds,this paper constructs fault diagnosis methods for aviation gears based on weighted domain adaptation network.The core idea is to increase the model’s attention to the samples that are helpful to the domain adaptation process through the weighting mechanism,which solves the conditional domain adaptation and partial domain adaptation and conditional partial domain adaptation diagnosis problem.The main works are as follows:Aiming at the transfer diagnosis problems of aviation gear faults,where the types of target domain faults are the same as the types of source domain faults,due to the influence of noise and working conditions,the samples with low transferability cause negative transfer effects on the transfer diagnosis model.This paper constructs a weighted conditional domain adaptation fault diagnosis method.This method constructs a combined sample matrix and input them into the model,evaluates the sample transferability according to the entropy criterion and the discriminant results of the domain discriminator.So that it can reduces the negative transfer effect on the model,which caused by the samples with low transferability.Finally,the effectiveness of the algorithm is verified by the experiments on the gear dataset.Aiming at the partial domain fault diagnosis of aviation gear,where the source domain has more types of faults than those in the target domain,the existing partial domain diagnostic models only use category weights to select common class samples in the source domain and do not consider the degree of interference of the samples,a weighted partial domain adaptation fault diagnosis method is constructed,which uses adaptive weight-based adversarial domain loss to select common class samples while considering the degree of interference of the samples,and combined with the feature measure loss item to further improve the domain adaptation effect.Finally,the effectiveness of the proposed method is verified by the gear fault dataset experiment.Aiming at the current partial domain adaptation diagnosis models ignore the category information of fault samples,which makes the domain adaptation process deviate and produce negative transfer effects.This paper constructs extended research on the basis of weighted partial domain adaptation fault diagnosis method,proposes a weighted conditional partial domain adaptation diagnosis method.The proposed method inputs the category prediction vector as conditional information into the model,so that the model can effectively align features of the same category between samples in different domains,realize partial domain adaptation based on conditional information.Finally,the gear fault diagnosis experiment proves that the proposed method can achieve better diagnosis results.