With the rapid development of the auto industrial, the production and the reserving number of the autos have been growing, which brings lots of problems on the repair work. Therefore, the enforcement on the problem diagnosis becomes an important issue to be solved, which attracts much more attention.The ABS is a key part of a vehicle. Once an ABS fault occurs, it would make lots of problems, such as not to break, drift, side-slide, and even rotating U-turn, which greatly affects the safety. As the neural network has the characteristics of parallel distributed processing, auto-adaptive, association, memory, clustering and fault-tolerance, it perfectly fits in the ABS problem diagnosis when the regulator or the sensor gets a fault. According to the fault trait of the regulator and sensor, a BP neural network is established for diagnosis. However, this traditional BP network has a high requirement on the initial value, easily falls into a local minimum, and has a slow converging rate or even unable to converge. In this paper, the genetic algorithm is adopted, which has the ability to search a global minimum, and weights and valves value of the traditional BP network are optimized, which leads to a faster converging rate and excludes the defect of non-convergence. Thereafter, the optimized BP network is applied into the diagnosis of regulator and sensor of ABS, and it improves the reliability.At last, a simulation run under the Matlab shows the result that the GA optimized BP network surpasses the traditional one not only on the speed but also the error rate.
|