WIRE BREAKAGE PREDICTION OF WEDM USING VIBRATION SIGNALS IN CUTTING OF SILICON CARBIDE

Silicon carbide (SiC) is used in semiconductor electronics devices that operate at high temperatures or high voltages. A common way of cutting SiC is using the diamond multi-wire saw, which can result in high manufacturing cost due to the substantially high material hardness. For this reason, wire electrical discharge machining (WEDM) was developed for slicing SiC wafers owing to its ability to handle conductive materials. Unfortunately, a severe wire breakage issue was discovered in slicing SiC wafers using WEDM, mainly when a high-duty factor was employed. Despite numerous studies discussing the causes of wire breakage during WEDM, little has been conducted about predicting this phenomenon using sensor signals and machine learning (ML) predictive models. This study thus investigated the potential of using extracted features from vibration signals as in-puts to ML models for predicting the wire breakage time during SiC wafer slicing by WEDM. The performance comparison of several ML classification models was discussed in the research. In addition, the strategy of signal segmentation and overlapping was also considered for the preparation of input features. The results demonstrate that the features extracted from vibration signals in combination with the random forest (RF) model are capable of predicting wire breakage few seconds in advance with a prediction accuracy of 95%. This study lays the groundwork for future research to develop a wire break-age prevention system, which might include later automatically adjusting the machining parameters during the slicing process.