Research On Intelligent Diagnosis Method Of Turnout Based On Machine Learning

By the end of 2022,the total railway mileage in China reached 155,000 kilometers,including 42,000 kilometers of high-speed railways.In 2023,more than 3,000 kilometers of new lines are expected to be put into operation,including 2,500 kilometers of high-speed railway.However,due to the long operating lines of railways and the complexity of the environment,ensuring train safety and operation maintenance becomes one of the important issues in railway construction.Among them,the turnout is an important equipment that affects the safety of railway transportation.Besides,its normal operation is the key to ensure the transportation safety and improve the transportation efficiency.Based on the above situation,railway safety requirements for the real-time monitoring of the working status of turnout equipment and the intelligence of fault diagnosis and location of turnout equipment are becoming increasingly stringent.Based on site requirements and analysis of typical faults,this thesis takes the ZDJ9 turnout switch machine as the research object,and conducts research on the intelligent diagnosis method of turnouts based on machine learning to provide theoretical support for turnout maintenance.The main research works of this thesis are as follows:(1)To realize accurate fault diagnosis of turnouts,a fault diagnosis algorithm based on multi-domain feature extraction and an improved extreme learning machine is proposed.Firstly,the multi-domain features including time-domain features,value-domain features and time-frequency domain features of the switching resistance of the turnouts are extracted.Secondly,to reduce the dimensionality and redundant information of the extracted features,linear discriminant analysis and kernel principal component analysis algorithms are adopted to realize effective feature dimension reduction.Finally,in response to the problem of particle swarm optimization easily falling into local optima,the comprehensive learning particle swarm optimization algorithm is proposed and introduced to optimize the extreme learning machine.The proposed method is compared to some existing methods,indicating its effectiveness and feasibility.(2)To achieve early maintenance of turnouts,a health status recognition method for turnouts based on curve similarity is proposed.By calculating the Pearson correlation coefficient and Euclidean distance between the curve to be evaluated and the reference curve,the trend analysis of the turnout status is carried out.Furthermore,to address the issue of data nonlinearity,a health status recognition method is proposed based on the isolation forest algorithm,which can identify three kinds of turnout status: healthy,sub healthy and faulty,anc can guide the maintenance staff to carry out early maintenance on the turnouts that have been identified as sub healthy status.The final experimental results show that the proposed method has certain advantages and can effectively recognize the health status of turnouts.(3)Combined with demand analysis,the turnout intelligent diagnosis software is developed and designed,which has intelligent fault diagnosis and health status recognition functions,promotes efficient maintenance of turnouts,and has a positive effect on reducing workload and maintenance costs,achieving the purpose of improving maintenance efficiency.