基于自适应LS-SVM的柴油机进排气系统故障诊断  

The Fault Diagnosis for Diesel Engine Inlet and Exhaust System Based on Adaptive LS-SVM

作　　者：游张平[1,2] 胡小平[2] 张凯[1] 叶晓平[2] 

机构地区：[1]同济大学机械工程学院,上海201804 [2]丽水学院机械工程系,浙江丽水323000

出　　处：《科技导报》2010年第8期77-80,共4页Science & Technology Review

基　　金：国家高技术研究发展计划(863计划)项目(2008AA042803);浙江省自然科学基金项目(Y1080434)

摘　　要：提出了应用自适应最小二乘支持向量机和小波包能量特征的柴油机进排气系统故障诊断方法。对气门间隙异常、气阀漏气等几种常见故障和系统正常运行进行小波包分解,提取频带能量作为支持向量机的输入特征向量;然后,利用自适应优化算法对最小二乘支持向量机进行优化;最后,利用基于优化参数和最小输出编码的最小二乘支持向量机进行故障分类和识别。对比实验表明,与BP神经网络和采用交叉验证的传统最小二乘支持向量机相比,该方法可克服训练时间较长、容易陷入局部最小等问题,具有较快的训练速度和较高的分类准确率,提高了传统最小二乘支持向量机算法的寻优速度,在样本数较小时仍可取得较好的效果,能有效诊断柴油机进排气系统故障。A novel fault diagnosis method is proposed for diesel engine inlet and exhaust system based on least squares support vector machine （LS-SVM）. Normal running and several common faults （gas leak and abnormal valve lash） are decomposed into wavelet packets, the energies of different frequency bands after wavelet packet decomposition serve as the input vectors of LS-SVM, the feature vectors. An adaptive optimizing algorithm is used to optimize LS-SVM. The engine fault is identified using LS-SVM based on optimized parameters by the adaptive optimizing algorithm and the minimum output coding. A comparative experiment shows that the proposed approach can overcome the shortcomings of the slow convergence and of failing easily into a local minimum in the BP algorithm and the long training time in the traditional IS-SVM algorithm by cross-validation, and the method enjoys advantages such as fast training speed and high identification accuracy percentage as compared with the BP neural network and the traditional LS-SVM. It also performs well with small number of samples. The simulation verifies its validity in identifying diesel engine inlet and exhaust system faults.

关 键 词：最小二乘支持向量机 小波包 柴油机 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]