تشخیص عیوب سرسیلندر موتور احتراق داخلی در انتهای خط تولید به کمک پردازش سیگنال صوتی و شبکه عصبی پیچشی

This research examines an innovative method for detecting cylinder head faults in internal combustion engines. The defects under investigation include camshaft axial clearance, oil suction tube blockage, and empty tappets, which are identified at the end of the production line and during the hot test stage using audio signal processing. These defects are common in the production line, and their detection, especially camshaft axial clearance, is challenging using traditional methods. In this study, a combination of signal processing and neural networks is used for defect detection.
Audio signals were recorded at three different operating speeds (1300, 1700, and 2500 rpm) under idle conditions using a handy audio recorder, which is a cost-effective and practical choice that can be easily used in various industrial environments. The recorded audio data contains significant noise due to production line working conditions, which complicates the fault detection process. For signal processing and feature extraction, spectrograms along with Mel and Gammatone filter banks were used. These methods allow for more precise analysis of audio frequencies and can better identify sound patterns associated with defects. Convolutional neural network was used for data classification, which provides good results for separating data with a small number of samples. The research results show that spectrograms are more efficient in feature extraction compared to other common methods such as continuous wavelet transform. At operating speeds of 1700 and 2500 rpm, all defects are detected with approximately 98% accuracy. This method is particularly effective in detecting defects such as camshaft axial clearance, which are not detectable by traditional methods. Additionally, this method can distinguish between oil suction tube blockage and empty tappets, which are typically identified as a single fault by production line experts.