Yang Yanhe, Bi Xiaoyang, Lee Alamusi, Ma Teng, Sun Yinghui, Kong Wei, Hu Wei, Hu Ning
School of Mechanical Engineering, Hebei University of Technology, Tianjin, 300401, People's Republic of China.
State Key Laboratory of Reliability and Intelligence Electrical Equipment, Hebei University of Technology, Tianjin, 300401, People's Republic of China.
Sci Rep. 2023 Oct 20;13(1):17980. doi: 10.1038/s41598-023-44755-7.
Ignition advance angle is one of the important factors affecting the performance of the engine, when it occurs abnormally will make the engine power and economy worse, and even cause serious damage to the engine. Therefore, it is very necessary to recognize the abnormal ignition advance angle of the engine. However, the engine system is closed and has a complex structure, which makes traditional diagnostic methods difficult. This paper proposes an intelligent identification method based on acoustic emission (AE) signals, which collects the AE signals from the engine surface and divides their spectra into equal parts, and selects the frequency bands with high contribution to the classification based on the minimum distance method to construct feature maps, which is used as the input to the convolutional neural network (CNN). The extracted frequency band features of this method can better characterize the AE signals, and the constructed feature maps make the fault information more obvious. Experiments show that the accuracy of this method for abnormal ignition advance angle under normal operating conditions of piston aero-engine is 100%, which is better than the traditional methods. In addition, the recognition accuracies under the other two operating conditions are 99.75% and 98.5%, respectively, indicating that the method has a certain universality.
点火提前角是影响发动机性能的重要因素之一，其出现异常时会使发动机动力性和经济性变差，甚至对发动机造成严重损害。因此，识别发动机点火提前角异常十分必要。然而，发动机系统是封闭的且结构复杂，这使得传统诊断方法存在困难。本文提出一种基于声发射（AE）信号的智能识别方法，该方法采集发动机表面的AE信号并将其频谱等分为若干部分，基于最小距离法选取对分类贡献度高的频段构建特征图，将其作为卷积神经网络（CNN）的输入。此方法提取的频段特征能更好地表征AE信号，构建的特征图使故障信息更明显。实验表明，该方法对活塞式航空发动机正常工况下点火提前角异常的识别准确率为100%，优于传统方法。此外，在另外两种工况下的识别准确率分别为99.75%和98.5%，表明该方法具有一定的通用性。
