Operational State Recognition of Continuous Worm Gear Grinding Process Based on Wavelet Packet Energy Ratio and Double-Layer SVM Classifier

In the process of worm gear grinding, chatter is a kind of unsteady vibration that directly affects the stability of worm gear grinding. It will not only leave vibration marks on the surface of the gear, reduce its surface integrity and performance, but also cause certain damage to the machine tool, thus reducing the processing quality and processing efficiency. Therefore, it is of great significance to monitor and suppress chatter in real time. This paper presents a fast, real-time and accurate state recognition method for the key process of worm gear grinding. Initially, the three-coordinate vibration signal in the production process of worm gear grinding unit was collected. Subsequently, Besides time domain and frequency domain analysis, a variety of time-frequency analysis methods were used to analyze and extract the signal features, and the feature scale was compared. Finally, the energy ratio of the three-layer wavelet packet transform is selected to construct the feature dataset and a 2-layer support vector machine is trained to compare with other classification models. The final results show that the processing anomalies and machining process can be quickly and real-time monitored with 98.25% and 98.75% accuracy respectively on the 2-layer SVM with 3-layer wavelet packet energy ratio. The method proposed in this paper can alarm the abnormal state in time in actual machining, and can select the best time to adjust the machining in time, which effectively guarantees the quality and efficiency of gear machining.