This study adopted AI to identify the normal and abnormal vibration signals generated during robotic grinding. This study classified four fundamental factors affecting grinding into three levels to obtain a widely used result and designed an L9(34)ort...
This study adopted AI to identify the normal and abnormal vibration signals generated during robotic grinding. This study classified four fundamental factors affecting grinding into three levels to obtain a widely used result and designed an L9(34)orthogonal array for the grinding experiment. During experimentation, part of the grinding wheels was added weight to produce abnormal vibration signals, which an accelerometer would measure. The study transformed the collected vibration signals into recurrence plots and conducted model training with VGG16 CNN architecture. Finally, this study tested a model with 89.6% training accuracy. The results showed the model could identify whether the recurrence plots stand for normal or abnormal vibrations, with an accuracy of 85%. This means it could predict normal and abnormal grinding conditions and help avoid problems caused by abnormal vibrations.