A novel contactless measurement framework for punch tool wear using cross-correlation imaging and burr-based deep learning estimation

• A novel contactless measurement framework is proposed for punch tool wear monitoring. • Cross-correlation imaging enables robust, sensorless similarity analysis of punch wear. • Burr height is used as an indirect, process-integrated measurement proxy for wear. • A BiLSTM model forecasts burr evolution with low prediction error (3.59%). • Predicted burrs show high correlation (R² = 0.94) with actual wear, enabling online estimation of actual punch wear. This study introduces a novel, contactless measurement framework for on-cycle monitoring and estimation of punch tool wear in SPCC steel blanking processes. Addressing the limitations of direct measurement and traditional inspection methods, the proposed approach integrates cross-correlation imaging with burr-based forecasting through deep learning to enable accurate, sensorless evaluation of tool condition. A robust image-based similarity metric, derived from cross-correlation analysis of punch head images, ensures resilience to lighting variation and alignment noise. Instead of direct wear quantification, burr height, a visual and process-integrated feature, is utilized as a surrogate wear indicator. A BiLSTM deep learning model was trained on 300 blanking cycles using seven time-series features extracted from image and process data. When tested on Cycles 301–330, the model achieved RMSE 1.83 µm and MAPE 3.59%, accurately predicting future burr formation. A strong correlation (R 2 = 0.94) was observed between predicted burr and actual punch wear measured, validating the use of burr evolution as a reliable proxy for wear progression. This framework enables non-intrusive, uncertainty-aware, real-time wear estimation, offering a new paradigm in intelligent measurement systems for manufacturing diagnostics and predictive maintenance.