Research On Wheel Tread Wear Prediction And Tread Optimization Based On Hybrid Network Model

Heavy haul transportation has become the main way of railway freight development in various countries because of its advantages of large volume and low cost.However,with the development of heavy haul railway,the problem of wheel rail wear is becoming more and more serious while increasing railway traffic and economic benefits.The wheel rail wear problem will change the dynamic performance of the train wheel rail system,have a greater impact on the safety of the train operation,and even cause immeasurable losses.Therefore,it is very necessary to carry out comprehensive research on wheel wear and guide the maintenance of wheel rail system with accurate and scientific principles and methods.Through the prediction research and analysis of wheel tread wear,we can understand the overall law of wheel tread wear,and then optimize the design of wheel profile,so as to arrange the wheel turning cycle more reasonably,increase the service time of wheel,and reduce the maintenance cost of heavy haul line.Focusing on the problem of wheel tread wear and tread optimization of heavy haul train,this paper systematically combs,analyzes and summarizes the existing research.Aiming at the shortcomings of the existing research,a series of research works are carried out in the aspects of improvement and optimization of extreme learning machine,prediction model of wheel tread wear of heavy haul truck,optimization model of wheel profile,etc.Based on the theory of extreme learning machine,extreme learning machine is combined with wheel tread wear prediction and optimization,and wheel wear and optimization are studied,analyzed and innovated from the perspective of artificial intelligence.The main research work and achievements are as follows:(1)The piecewise chaos algorithm and mutation mechanism are introduced to improve the quantum particle swarm optimization(QPSO)algorithm,and an improved quantum particle swarm optimization(IQPSO)algorithm is proposed.In this model,the initial value of particles is mapped by using the piecewise logistic chaos principle,and then the position of particles is updated according to the mutation mechanism.Compared with the traditional algorithm,the optimization ability of IQPSO algorithm is further verified.On the basis of IQPSO single objective optimization algorithm,an improved multi-objective quantum particle swarm optimization(MOIQPSO)algorithm is proposed by introducing external archive set,and the test comparison and analysis are carried out to verify the superiority of the performance of moiqpso algorithm.(2)A hybrid model of deep belief network improved swish derived extreme learning machine(GDBN-ISDELM)for wheel tread wear prediction is proposed.Firstly,a swish derived extreme learning machine(SDELM)model is proposed.Based on the IQPSO optimization algorithm,the parameters of SDELM are optimized(improved swish derived extreme learning machine,ISDELM),and then combined with the improved deep belief network(GDBN).The model uses GDBN for unsupervised training and feature extraction of wear data,and then takes the extracted features as input data,and uses ISDELM for data learning and training,and finally realizes data prediction.Compared with other algorithms,GDBN-ISDELM algorithm has better stability,generalization ability and higher prediction accuracy.(3)According to the relevant theory of vehicle dynamics,the paper analyzes the processing method of nonlinear components in the process of establishing the C80 heavy load truck dynamic model,and establishes its dynamic model by UM software.Based on the Archard model,the main influencing factors of wheel tread wear are simulated and analyzed.The results show that the wear depth of wheel tread will vary with different speed,axle weight and curve radius.(4)Through the GDBN-ISDELM model proposed in this paper,the wheel tread wear value can be predicted.Taking the speed,axle load and curve radius as the input and the maximum tread wear depth as the output,the mapping relationship between input and output is established by GDBN-ISDELM model to realize the prediction of tread wear value.Compared with other models,the GDBN-ISDELM model has smaller prediction error,higher prediction accuracy and better generalization ability.It can better reflect the influence of different parameters on the wheel tread wear value,and can more accurately predict the wheel tread wear,so it has strong applicability.(5)The optimization of wheel tread of heavy haul train is studied,and a new optimization method of wheel profile is proposed.Firstly,a description method of wheel profile constructed by cubic NURBS curve is established.Secondly,on the basis of MOIQPSO algorithm,the concept of penalty function is introduced,and a multiobjective optimization model of wheel profile with constraints is established.In the optimization process,the new tread coordinates generated by each iteration are mapped to the wear index through GDBN-ISDELM model,so as to realize the prediction of wear power and wheel rail lateral force of different tread.Finally,the optimized wheel profile is reconstructed by cubic NURBS curve.The wear of the optimized tread is smaller,and the service life of the optimized tread can be increased.