Incremental transfer learning for robot drilling state monitoring under multiple working conditions

Robot drilling is widely used in industrial scenarios, and the quality of the hole affects the quality of the finished product. Drilling for different workpieces under varying working conditions is a common phenomenon in industrial scenarios, so a drilling quality monitoring method suitable for multiple working conditions is needed. In this paper, a continuous transfer learning method for drilling quality classification based on vibration signals is proposed to detect whether a hole is vertical or not under multiple operating conditions. Firstly, the factors affecting the vibration signal of robot drilling were analyzed through the vibration model of robot drilling. Then, the domain adaptation method was used to extract the characteristics of vibration signals under the hyperplane with different working conditions, and the correlation of data between different domains was calculated and the correlation coefficient of data between different domains was constructed. Finally, through the continuous learning method of model parameter reservation, the classification model is dynamically inherited by using the correlation between domains, and a deep learning model suitable for multi-domain data is realized. In the laboratory environment, several groups of vibration signals of robot drilling under different working conditions are collected, and the average classification accuracy is more than 90% under the dynamic input of different working conditions data. The results prove the effectiveness of the method.