A new method for on-line monitoring discharge pulse in WEDM-MS process

This paper presents a new method for on-line monitoring discharge pulse in wire electrical discharge machining-middle speed (WEDM-MS) process based on digital image processing and machine learning. Discharge pulse monitoring is a key aspect of the WEDM-MS control system, as it directly guides the direction and the speed of the machining. In the proposed system, the discharge pulse is captured using a high-bandwidth oscilloscope because the shortest period of discharge is less than 40 ms. The proposed system workflow consists of image reconstruction, pre-processing, feature extraction, and pulse classification. Wavelet moment analysis (WMA), Hu moment analysis (HMA), fractal dimension analysis (FDA), local geometric characteristics (LGC), and global geometric characteristics (GGC) are all applied in the image pre-processing to extract waveform image features and reduce image dimension. These features are then used in a two-stage classification technique that employs support vector machine (SVM) and random forests (RF) for pulse classification and identification. The first stage method, which uses SVM, discriminates between open circuit, short circuit, and mixed status pulses, while the second stage method, which uses RF, divides the mixed status pulse into spark, transition arc, and arc pulse. In test experiments conducted under different electrical parameters and with different materials, the WEDM-MS discharge pulse monitoring systems were successfully employed with an accuracy of 93.13 %. The proposed SVM-RF approach outperforms learning vector quantization (LVQ) neural network, SVM, or RF discharge pulse WEDM discrimination methods.