基于粒子群神经网络的气阀机构故障诊断  

Valve Train Fault Diagnosis Based on PSO Neural Network

作　　者：游张平[1] 胡小平[1] 

机构地区：[1]丽水学院机械工程系,浙江丽水323000

出　　处：《测控技术》2011年第12期102-105,共4页Measurement & Control Technology

基　　金：丽水学院重点科研项目(KZ201118);丽水学院引进人才科研启动基金项目(2009001);国家863计划资助项目(2008AA042803)

摘　　要：提出应用粒子群神经网络和小波包能量特征的柴油机气阀机构故障诊断方法。为了克服BP算法的缺陷,将粒子群优化(PSO)算法应用于神经网络的学习算法中;为了避免PSO算法在全局最优值附近搜索变慢,采用了一种从PSO搜索到BP搜索的启发式算法;然后,通过模拟柴油机气阀机构的两种常见的主要故障:气阀漏气和气门间隙异常,采集气缸盖表面的振动信号,应用小波降噪处理后,再使用小波包提取能量特征参数作为该粒子群神经网络的输入特征向量,对气阀机构故障进行分类和识别。对比实验表明:与BP算法对比,该算法具有较快的收敛速度和较高的分类准确率,也证实了该方法的正确性和有效性。A novel fault diagnosis method for diesel valve train based on particle swarm optimization（PSO） neural network（NN） and wavelet packet decomposition is proposed. PSO algorithm is employed as learning al- gorithm of NN, to overcome drawbacks of pure BP algorithm. To avoid the slow search speed around global opti- mum in PSO algorithm, a heuristic way is adopted to give a transition from particle swarm search to gradient descending search. By simulating two kinds of main fault of a valve train, which are the gas leak and abnormal lash, the vibration signals of a cylinder head have been measured. The different frequency bands energy of diesel engine vibration signal after wavelet packet decomposition constitute the input vectors of PSO NN as fea- ture vectors. Diesel valve faults are classified by using the PSO NN. A comparative experiment shows that the method has more fast convergence speed and higher diagnosis accuracy than BP algorithm. That also shows the correctness and validity of this method in diesel valve train fault diagnosis.

关 键 词：柴油机 粒子群优化算法 神经网络 故障诊断 

分 类 号：TK428[动力工程及工程热物理—动力机械及工程]