Diagnosing the clearance fault of the camshaft in the hot test stage at the end of the production line by using of audio signal processing

This research addresses the detection of camshaft axial clearance defects in internal combustion engines during end-of-line hot testing using acoustic signal processing. The study's importance lies in reducing manufacturing costs and enhancing customer satisfaction through improved product quality. This defect is prevalent in the production line and challenging to identify using conventional methods.The research proposes an intelligent solution for defect detection by combining time-frequency domain signal processing techniques with artificial neural networks. The axial clearance defect was simulated at various severity levels, and acoustic signals were recorded using a handy sound recording device at three different engine speeds (1300, 1700, and 2500 RPM) under no-load conditions. The engines undergo a seven-minute test at these different speeds to ensure proper functionality. The choice of a handy recorder device was based on the manufacturer's request for a portable and cost-effective solution.It's noteworthy that the recorded audio data contains noise due to production line conditions, adding complexity to the fault diagnosis process. For signal processing and feature extraction, two methods were employed: Continuous Wavelet Transform (CWT) and Mel Spectrogram.The results demonstrate that the Mel Spectrogram is more effective for feature extraction compared to the Continuous Wavelet Transform. At operating speeds of 1700 RPM and 2500 RPM, all defect levels are detected with an average accuracy of 99% by using Convolutional Neural Network.This study contributes to the field of non-invasive fault detection in automotive manufacturing, offering a reliable and cost-effective method for identifying camshaft axial clearance defects. The high accuracy achieved at specific operating speeds suggests that this approach could be implemented in real-world production environments, potentially leading to significant improvements in quality control processes and overall product reliability.The success of this method highlights the potential of acoustic signal processing and machine learning techniques in solving complex industrial quality control challenges.