Research On Automobile ABS Fault Diagnosis Based On AGA-Elman Neural Network

With the development of social economy and the popularization of automobiles,people pay much attention and more stringent on the safety requirement of automobiles,and safety design is also the most important link in modern automobile design.Automobile’s Anti-lock Braking System(ABS),as a key active safety device to improve braking performance,can prevent tires from locking by adjusting the braking force,ensure the driving safety and reduce the probability of accidents.Therefore,using intelligent technology to provide more accurate fault diagnosis for ABS is of great significance.Neural network is widely used in the field of fault diagnosis by virtue of its self-learning,self-adaptive,and strong performance in nonlinear pattern classification.Thus,this paper chooses Elman neural network,combined with the improved adaptive genetic algorithm,to establish an automobile ABS fault diagnosis model based on AGA-Elman neural network.The main research contents are as follows:Firstly,this paper selects ABS fault characteristic factors and obtains sample data.According to the structure and working principle of ABS,the sensor and regulator of ABS are determined as the fault research object,and the longitudinal speed,four-wheel speed and lateral speed are determined as the fault characteristic factor.Then,based on the Carsim/Simulink platform,an ABS fault simulation model is established and simulated,for obtaining the fault characteristic factor data under different fault states of ABS,and inputs the data into the neural network as the sample data set,for learning and establishing the non-discrimination with the fault states of the ABS sensor and regulator.Secondly,the adaptive genetic algorithm(AGA algorithm)is improved and the AGAElman neural network is designed.In the beginning,the network structure and learning algorithm of Elman are introduced.Then,Elman’s learning algorithm is improved,and its network is optimized by genetic algorithm,for overcoming the shortcomings that Elman’s network is easily trapped in local optima and its traning efficiency is slow.After that,considering that the genetic algorithm has slow convergence and the phenomenon of premature maturity,and the traditional adaptive genetic algorithm has the slow evolution in the early stage,thus,an improved adaptive genetic algorithm has been presented,which can speed up the convergence efficiency and enhance its global optimization ability,and uses the improved AGA algorithm to solve the optimal initial weight threshold of Elman,for enhancing Elman’s training efficiency and network performance.Finally,the ABS fault diagnosis model is established,and simulation and comparison tests are carried out.Above all,based on the AGA-Elman neural network,the ABS fault diagnosis model is established,and the parameters of AGA algorithm and Elman neural network structure are selected.And the simulation result shows that diagnostic accuracy of the AGA-Elman model is high,which can meet the application requirements.Then,the performances are compared before and after the model improvement.Specifically,the AGA-Elman model,the Elman model and the GA-Elman model are used for simulation and comparison tests,and the training performance and diagnostic result among them are analyzed and compared,reflecting the diagnostic accuracy of the AGA-Elman model built in this paper is higher,and its training speed is faster.After that,the performances of different algorithms are compared.Specifically,the comparative experiments of the AGA-Elman model,the SVM model and the AGA-BP model are carried out,and the simulation results are analyzed to further show the superiority of the AGA-Elman model built in this paper...