基于优化递归小波与改进型ART2网络的内燃机故障诊断

曲轴角振动信号是内燃机故障诊断的有效依据。针对传统小波变换存在数据窗长、实时性差的局限性,构造出一类用递推公式进行小波变换的小波基。基于该类小波基的小波变换不受分析窗的制约,能够实时地提取暂态特征。同时提出此类小波时频特性的优化方法,对内燃机曲轴角振动信号进行递归小波分解,并提取特征。针对传统ART2网络只利用模式的相位信息而丢失幅度信息和网络性能依赖于样本顺序等不足,在网络权值更新时引入平均滤波和关联函数,以便提高网络稳定性和收敛速度,同时降低其对样本输入顺序的敏感性。最后,将改进型ART2网络应用于内燃机故障诊断。实验结果表明,递归小波能反映内燃机状态信号的特征;改进型ART2网络训练耗时小于传统ART2网络的3%,识别率为100%。

Fault Diagnosis of I.C. Engines Based on Optimal Recursive Wavelets and Improved ART2 Network

The angular vibration of the crankshaft is viewed as an important tool of fault diagnosis of I.C.engines.For overcoming the shortcomings of the longer data window and poor real-time characteristics in conventional wavelet transforms,a general method of recursive mother wavelets is introduced.Recursive wavelet transforms are not restricted by the data window,and can extract the transient information in real-time.Moreover,the steps to optimize its time-frequency performance are proposed.Besides,the angular vibration signals of crankshafts of I.C.engines are decomposed using recursive wavelets and the features are extracted.In view of the fact that the traditional ART2 network loses the amplitude information of input patterns and is sensitive to the sequence of input patterns,an improved ART2 is presented by introducing average filtering and relational functions into weight adjustment.The modified ART2 has higher convergent speed,is more robust,and more insensitive to the pattern sequence.Finally,it is used in fault recognition of I.C.engines.The results show that recursive wavelets are practical in analyzing I.C.engine condition signals.The time consumed by the modified ART2 accounts only for less than 3% of that by the traditional ART2,and the recognition rate of the improved ART2 is up to 100%.