Machining process condition monitoring based on ordinal pattern analysis and image matching

Stable machining process state is critical to product quality. However, existing method for monitoring the cutting process state cannot manifest the actual machining situation accurately because it only focuses on a single abnormality while ignoring the simultaneous occurrence of different abnormal states. Aiming at the three typical anomalies and multi-factor anomalies commonly observed in machining, a multi-delay ordinal pattern (OP) (MDOP) image feature based on OP analysis and a complex machining state recognition method based on dictionary template matching are proposed. First, an OP method is developed to extract the inherent wave pattern of monitoring signal under specific parameters. Second, the MDOP image features based on multi-parameters are established to optimize the parameter selection process and enrich the state information. With strong anti-noise and easy data processing, the MDOP method can obtain rich processing state information from the perspective of visual knowledge. Third, an image matching method based on multi-template and a machining state recognition method based on maximum template matching degree are proposed. Finally, a machining experiment that included eight states was designed to verify the effectiveness of the method. Result shows that the proposed method can identify different cutting states accurately, and the multi-source signals can improve the accuracy of the model further. Compared with other methods, the MDOP method has evident advantages.

Automated summary

Stable machining process state is critical to product quality. However, existing method for monitoring the cutting process state cannot accurately manifest the actual machining situation. This study proposes a multi-delay ordinal pattern (OP) (MDOP) image feature based on OP analysis and a complex machining state recognition method based on dictionary template matching. The proposed method accurately identifies different cutting states, and the multi-source signals further improve the accuracy of the model.