Research On Crack Fault Diagnosis Of Traction Gear Of Locomotive Based On Volterra-DMSCDA

Traction gear system is the main component of locomotive transmission system,its "health status" is an important factor to ensure the safety of running,and crack fault is a more common early gear failure.The effective identification of gear crack fault has certain practical significance and engineering application value to avoid the further deterioration of the fault and even leading to serious problems such as tooth breakage.It can also meet the needs of traction gear repair according to the situation.In view of the limitations of the current mainstream gear fault detection methods(that is,only using system response as the research object,and rarely considering the effect of excitation on fault feature extraction),this paper analyzes both the excitation and the response of the system at the same time,and mines the dynamic characteristics of traction gear system as a typical nonlinear system in order to extract the nonlinear features more accurately,which is helpful for the extraction of crack fault features of traction gears and the effective identification of crack faults.In this paper,the third-order discrete time-domain Volterra series model is established for the traction gear system of locomotive,and the fault diagnosis flow of traction gear based on Volterra series theory is expounded.At the same time,the dynamic model of the traction gear system of locomotivewith different degrees of crack is established,and the setting method of crack propagation mode at the root of the gear and the calculation method of meshing stiffness and damping parameters are expounded,which lays a foundation for the extraction and simulation of crack fault characteristics of traction gears.Secondly,aiming at the time domain kernel identification of the Volterra series model of traction gear system,this paper improves the basic dragonfly algorithm,and proposes a Dymatic Multiple Sub-population Collaboration Dragonfly Algorithm,which is abbreviated as DMSCDA.The improvement measures are as follows: the chaotic sequence initialization strategy is used to improve the initial randomness of the population,the dynamic multi-subgroup strategy is used to classify the individuals with different qualities and assign optimization tasks according to the characteristics of different subgroups in order to improve the working efficiency of the whole population,and a more elastic boundary treatment method is introduced to further improve the individual utilization efficiency of the population.Four typical test functions are used to simulate and verify the performance of the DMSCDA.The results show that compared with the standard Particle Swarm Optimization,which is abbreviated as PSO and the Dragonfly Algorithm,which is abbreviated as DA,the convergence accuracy,the global search ability and the stability of the DMSCDA are better than that of PSO and DA.Then,for the purpose of constructing fault feature database,the problem of extracting crack feature of traction gear based on Volterra-DMSCDA is studied in this paper.Taking the torque of traction motor output shaft of SS7 electric locomotive in 80km/h uniform operation as input and using Simulink as platform,this paper simulates the dynamic model of traction gear system with different degrees of crack,and then obtains the vertical vibration displacement response of the driven gear shaft for the output of the system.Then,the input /output data used to identify the time domain kernel of the Volterra series model of the traction gear system of SS7 electric locomotive is obtained.On the basis of this,this paper identifiys the time domain kernel of the model by DMSCDA.Taking the torque of traction motor output shaft of SS7 electric locomotive in 80km/h uniform operation as input and using Simulink as platform,the dynamic model of traction gear system with different degrees of crack is simulated and tested.The vertical vibration displacement response of the driven gear shaft is obtained and used as the output of the system.The input / output used to identify the time domain kernel of the Volterra series model of the traction gear system of SS7 electric locomotive is obtained.On the basis of the data,the time domain kernel of the model is identified by DMSCDA.Based on the statistical comparison of the first,second and third order time domain kernel distribution of the model,it is concluded that for the traction gear system with tooth root crack,the first order time domain kernel is not enough to reflect the inherent nonlinear characteristics of the traction gear system,but the second-order or third-order time-domain kernel can effectively reflect the changes of the internal dynamic characteristics of the system under different states.Therefore,it is more reasonable to construct the fault characteristic database by using the first-order,second-order and third-order time-domain kernel as the basic data at the same time.Finally,this paper presents a fault diagnosis method of traction gear crack based on DMSCDA-RBFNN.Three kinds of Radial Basis Function Neural Network,which is abbreviated as RBFNN,fault classifiers with different structures are constructed,and then their parameters are optimized by DMSCDA.After testing and evaluating the performance of three kinds of RBFNN fault classifiers by using the data from the database of crack characteristics of traction gears,it is concluded that the accuracy of using only the first order time domain kernel as fault feature information is not high,and it is difficult to use them.But by synthesizing the first three-order time domain kernel as fault feature information,satisfactory fault diagnosis results can be well obtained.