Research Of The Real Axle Main Reducer Gear Fault Diagnosis Based On Artificial Intelligence

With the rapid development of modern automotive technology, people are increasing having high demand on the quality of vehicles. Vehicle noise, vibration and harshness that NVH is a comprehensive index to measure the quality of vehicle manufactures. In the vehicle noise, the main micro-reducer is a key component in the vehicle drive system assembly. Gear, as the important parts of main reducer, its vibration noise is one of the main noise source. Gear vibration noise is an objective reality, but if gear noise is too large, it will bring potential safety hazard on the quality of vehicles, at the time, it pollute environment and influence people's ride comfort. At present, on the fault diagnosis of gear, the way of enterprise to judge the quality of gear is depending on the sound of gears turning, it have a high operational requirements for workers that only experienced masters have the ability to take this task. Therefore, for information carried by the gear noise, use relevant equipment to collect it analyze its characteristics and rules, then use modern computer intelligent methods to diagnose and research it, it's a very significant work.The message impressed by gears noise is messy and nonlinear, to seek the law of it, the use of artificial neural network to take research on it brings out the best in each other. Artificial neural network is a non-procedural, adaptability, the brain's style information processing. In particular the BP neural network, it has simple structure, plasticity and already has in a wide range of applications in fault diagnosis field. However, the advantages and disadvantages of BP network is also very obvious, it has a adaptive self-learning ability, strong fault tolerance, and very suitable for handling non-linear complex problems, but its convergence is slow, especially be vulnerable to fall into local optimal point, which limit its performance. Simulated annealing algorithm is suitable to solve large combinatorial optimization problems, the calculation is simple, universal and robust, suitable for parallel processing, especially in the theory, it has been proved to be a global optimization algorithms which with the probability 1 to converge to global optimal solution. Improving BP neural network with simulated annealing algorithm can overcome the defect of falling into local optimal point easily, and further improve the network performance.This article describes relevant research on the gear fault diagnosis, specifically expatiates the BP neural network and simulated annealing algorithm, their strengths and weaknesses and improving methods are analyzed, and the integration of simulated annealing algorithm and BP neural network is used to process information on the diagnosis of gear noise, and diagnosis results are analyzed. Compared result shows that integration of simulated annealing BP neural network has better performance, and is more accurate on the gear fault diagnosis.