Classification of Sensor Signal Features for Ti Alloy Turning Process Optimization

In recent years, the utilization of sensor monitoring systems during the machining has been widely accepted for process optimization and control. With the purpose of process optimization, sensor signals are correlated with process conditions so that the effects of process parameter changes can be realized on the basis of the sensor signal features rather than empirically based on skilled operators’ experience. This research work focuses on the utilization of acceleration sensor signals for classification of machining process conditions acceptability. The concerned machining process is turning of Ti alloy, where the acceptable and not-acceptable process conditions are correlated with real time sensor signal features. The classification and the identification of the sensor signal features are carried out by utilization of a SOM (Self-Organizing Map) based approach. SOM neural networks (NN) provide for an unsupervised NN methodology used here to classify the sensor signal features in clearly separate clusters. The SOM NN performance for Ti alloys turning process condition identification was examined by considering single acceleration components. Moreover, SOM maps were utilized as a powerful tool to identify ambiguous data vectors lying in both acceptable and not-acceptable process condition clusters. The principal objective and scope of this work is the reliable and robust classification of acceptable / not- acceptable process conditions during turning of Ti alloys. Nine case studies were considered: in the first three (Case 1, Case 2, and Case 3) case studies single cutting acceleration components Ax, Ay, and Az are utilized separately, where the sensor signals have four features (a1, a2, a3 and a4). Similarly, in the other case studies (case 4 to case 9) the first three cases have eight features and the last three cases consider sixteen features. A comparative study of process condition identification for all nine case studies along with data refinement procedures have been carried out. SOM NN has shown a very high performance to correctly identify the process condition under these operation situations. Moreover, refinement of the ambiguous data has presented a significant improvement in the results, achieving about 90 % of identification success rate.