Research On Fault Diagnosis Of Turbocharged System Based On Improved Genetic Algorithm For Neural Network

Turbocharger technology has irreplaceable function and has been adopted by many locomotive because in saving energy, improving power and meet the environmental protection. Failure factors of turbocharging system has the characteristics such as diversity, complexity, difficult to diagnosis. It has great economic value and social application value to research on fault diagnosis of turbocharging system.. Turbocharging system has the fault characteristics and limitation of various conditions, such as complicated noise, hard established testing system, bad operating environment and hard-removed noise, etc.. These factors result in the low efficiency fault diagnosis on turbocharging system. Therefore this essay carries out profound study and research on feasibility of turbocharging system fault diagnosis based on the improved genetic algorithm optimized neural network.First of all, this essay describes the present state and various methods of turbocharging system fault diagnosis and analyses the existing problems and difficulties further and studies the feasibility of turbocharging system fault diagnosis by adopting the improved genetic algorithm optimized neural network.Secondly, this essay studies the working process and the main fault and the fault diagnosis theory of turbocharging system. Due to the strong nonlinear mapping ability, neural network can well reflect the nonlinear relationship between turbocharging system state and the various influencing factors. Therefore a neural network is established as a fault diagnosis model. But the disadvantages of neural network is that the structure is difficult to determine. In order to optimize the structure of neural network, an improved genetic algorithm based on Hooke-Jeeves(Hooke-Jeeves Genetic Algorithm,referred to as “HJGA”)is proposed in this paper. In the HJGA algorithm, Hooke-Jeeves algorithm with the strong local searching ability is introduced into the genetic algorithm, the genetic algorithm provides the Hooke-Jeeves algorithm good initial values, Hooke-Jeeves algorithm searches the optimal individual by changing the step size. Using HJGA algorithm for function optimization simulation, the results show that the HJGA algorithm combines the global search ability of genetic algorithm and local search ability of Hooke- Jeeves algorithm, the computing speed, precision and stability are better than genetic algorithm(GA), the validity of the HJGA algorithm is verifiedFinally, the improved HJGA algorithm is used to optimize BP neural network, using the chromosome coding with three level hierarchical structure and appropriate fitness function, adopting turbocharging system fault data samples to train the BP neural network. The neural network structure, weights and thresholds are optimized at the same time. Then fault diagnosis model of the turbocharged system is got. The well-trained neural network had been tested by a selected test set. The test results meet the requirements of the turbocharging system fault diagnosis. In comparison of the training results and diagnosis results between HJGA optimized neural network system and GA optimized neural network system, HJGA algorithm optimized the BP neural network is superior. Structure of the neural network is more compact, convergence speed is faster and diagnostic data are closer to the true value.