In the present work, the wavelet packet technique based on the vibration signals is proposed under normal and fault conditions of the spark ignition (SI) engine. A novelty fault diagnosis technique is considered through the calculation of the maximum energy to Shannon entropy ratio for twenty-five mother wavelets. An optimization approach is conducted for selecting the wavelets and decomposition level to reduce the noise of the captured signal. Feature extraction based on a discrete wavelet transform and energy spectrum is extracted. Effect of the selection of proper de-noising wavelet on the performance of both supervised and unsupervised artificial neural network (ANN) is evaluated. Experimental results show that Coif2_2, dmey_2, and rbio5.5_2 are valuable wavelets for de-noising signal of the SI engine. It is also found that the maximum energy to Shannon entropy ratio is a fast and powerful method to be used in the selection of wavelet families with the best decomposition level. In addition, it indicated that the wavelet packet transform has great potential in detecting spark plug defects. It can be reported that the de-noising with the wavelet revealed the best results on the performance of the ANN for the classification and clustering of fault or normal states.

在目前的工作中，提出了基于振动信号的小波包技术在火花点火（SI）发动机的正常和故障条件下。通过计算二十五个母子波的最大能量与香农熵之比，考虑了一种新颖的故障诊断技术。进行了一种优化方法来选择小波和分解级别，以减少捕获信号的噪声。基于离散小波变换和能量谱的特征提取。评估了选择适当的降噪小波对有监督和无监督人工神经网络（ANN）的性能的影响。实验结果表明，Coif2_2，dmey_2和rbio5.5_2是用于SI引擎信号降噪的有价值的小波。还发现最大能量与香农熵之比是一种用于选择具有最佳分解水平的小波族的快速而有效的方法。另外，表明小波包变换在检测火花塞缺陷方面具有很大的潜力。可以报​​道的是，用小波进行的去噪显示出在神经网络的故障或正常状态的分类和聚类方面的最佳结果。