Electrochemical discharge machining (ECDM) is an effective and promising technology for the micro-hole drilling of hard and brittle insulating materials such as glass, quartz and ceramics. Hole exit quality is an important part of through-hole machining quality in ECDM. The hole exit quality is influenced by two different machining states before and after the hole breakout. Therefore, it is necessary to identify hole breakout occurrence to provide a basis for achieving a machining strategy that matches the changing machining state and thus improves the hole exit quality. However, currently, there is no effective method to detect the hole breakout in the ECDM of micro-holes. In this paper, a self-adaptive micro-hole breakout detection method based on CNN- BiLSTM was proposed. The hole breakout detection can be expressed as a binary classification problem according to the analysis of the current signals collected before and after the hole breakout. The sufficiency of the raw current signal information was discussed. A CNN-BiLSTM model was established to detect hole breakout using the current signals as the training dataset. The CNN-BiLSTM model showed better performance compared to CNN, LSTM, BiLSTM and CNN-LSTM by comprehensive considering of accuracy, F1 score, AUC value, training time, predict time, breakout detection results and Dunn test results. Furthermore, the length of current signals used for each single detection and the threshold value used for breakout detection were discussed. An optimal threshold value was calculated, and the delay of detection was within 66 ms. The proposed method was successfully applied to the hole breakout detection in the through-hole drilling experiments and the success rate was 100 %. The hole exit area just after breakout were measured and compared to the hole exit area after hole completion, the average ratios of the former to the latter were 11.69 %, 27.29 %, 38.43 % and 44.41 % for the voltage of 39 V, 41 V, 43 V and 45 V respectively.
Read full abstract