Laser Cutting Defect Recognition Using Conversion of Processing Light Information into Spectrogram Images – Spectroscopic Measurements in Multiple Work Surface Conditions and Extraction of Spectral Data Features Based on Processing Principle

At urban production sites, laser cutting is an essential technology for high-speed flexible sheet-metal processing. This study aims to detect defective cuts by sensing laser-cutting-induced light emission and elucidate meaningful features for processing-based detection. The proposed method comprises three steps. In the first step, the sensors installed in the laser head acquire the spectra of light generated during processing, and data analysis software converts the spectral data into spectrograms and stacked-graph images. In the second step, image processing software extracts the edges of both images and emphasizes the periodic features in normal laser cutting. In the final step, a one-class support vector machine recognizes defective cuts from the extracted features. Verification tests using multiple normal and abnormal cut data confirmed that the proposed method accurately detected defective cuts.