Physical model-guided machine learning for accelerating laser induced plasma micro-machining process optimization

The demands for high precision and high efficiency in industrial manufacturing have driven the development of advanced laser-based manufacturing methods, especially laser-induced plasma micro-machining (LIPMM). Limited by incomplete physical understanding and costly, time-consuming experiments, finding the optimal process parameters across the entire process space presents a significant challenge. Although machine learning has been attempted for laser material processing, the lack of sufficient data remains a significant obstacle. To address this issue, a physical model-guided machine learning framework was developed, wherein intermediate mechanism parameters were generated based on the physical model, e.g. peak plasma density and plasma duration, and these were added to the original dataset vectors as extra dimensions to participate in and guide the model training process. The established framework accurately predicted the machining performance of LIPMM and, in combination with a genetic algorithm (GA), collaboratively optimized the multidimensional process parameters, resulting in a comprehensive improvement in both machining depth and material removal rate (MRR). The introduction of physical information enriched the data available for training, improving the model's prediction accuracy with small sample sizes and enhancing the superiority and rationality of the designed processes by avoiding suboptimal solutions. Additionally, coupling the intermediate information into the deep learning can further improve the prediction accuracy of LIPMM performance. The proposed physical modelguided machine learning provides a new strategy for significantly reducing the cost of process optimization, and can be applied to other complex laser processing processes to accelerate process development.