Research on drill bit sound anomaly detection method based on deep learning

Detecting anomalies in mechanical equipment is crucial in the manufacturing industry, which helps to prevent accidents and thus minimize economic losses. However, due to the complexity of drilling sound waveform, short detection time, coupled with the small size of the balanced dataset, the traditional machine learning method is a bit challenging to extract the most valuable features. This study introduces a novel approach for identifying faults in drilling machines based on analyzing the sound of the drill bit. The method combines the improved Global Context Block (GCBlock) with Convolutional Neural Networks (CNN) to obtain the most critical information from the spectrum. Besides, Gated Recurrent Unit (GRU) and attention mechanism is proposed to focus on abnormal phenomena in sound. In particular, in order to increase the quantity and diversity of the dataset, data augmentation methods were used. The experimental data demonstrates that this method excels in accuracy on the small-scale, short-duration drill bit sound dataset, and the overall accuracy of the abnormal and normal classes are 95.08%. Moreover, experiments conducted on a dataset containing a diverse range of classes and prolonged sound signals also yield high accuracy. This method has been proven to exhibit high efficiency and resilience in the face of challenges, demonstrating its generalization ability and reliability.