Fault Diagnosis of Bolt Loosening Based on LightGBM Recognition of Sound Signal Features
Mengxian Guo, Yong Guo, Yanfeng Peng, Weijie Zhang, Qihui Ling
Abstract
Aiming at the strong vibration condition of the inner curve radial plunger hydraulic motor, the bolt connection of the motor base is easy to loosen, and it is difficult to distinguish the early loosening faults online in time by using the vibration signals. In this article, we propose a bolt loosening fault diagnosis method based on LightGBM to recognize sound signal features, and this method can achieve online monitoring of bolt loosening faults. Through the vibration energy recovery test platform, the sound signals of four different bolt preload forces during the normal operation of the equipment were collected, and the bolt preload force was increased from completely loosened 0–60 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{N}\cdot \text{m}$ </tex-math></inline-formula> , with an increment of 20 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\text{N}\cdot \text{m}$ </tex-math></inline-formula> each time, and the sound signals were denoised using wavelet threshold denoising method. The LightGBM bolt loosening fault diagnosis model is constructed based on the gradient of one-sided sampling and mutually exclusive feature bundle algorithms. By extracting the time- and frequency-domain features of the denoised sound signals, a dataset containing labels of normal and three faulty signals is generated for training and diagnosis. Finally, the diagnostic accuracy of this method is compared and verified. The results show that the LightGBM algorithm after wavelet threshold denoising improves the diagnostic accuracy by 2.17% over the no-denoising LightGBM and by 5.47% and 2.21% over the XGboost algorithm before and after denoising, respectively.