Towards Sustainable Precision: Machine Learning for Laser Micromachining Optimization

TL;DR

This work tackles the challenge of sustainable, real-time monitoring of ultra-short pulse laser micromachining across diverse preprocessing techniques by developing an ML-based framework that predicts final surface roughness from laser parameters and multi-sensor signals. The authors leverage TSFEL to extract rich features, optimize a neural architecture with Bayesian methods, and apply SHAP to identify a compact feature subset, enabling lightweight models. Empirical results show sensor data improves generalization across milling, grinding, polishing, and EDM processes, and SHAP-driven reduction yields near-parallel performance with far fewer inputs using models like Extremely Randomized Trees. The approach lays the groundwork for real-time, energy-efficient monitoring in manufacturing, with demonstrated gains over prior approaches and clear pathways to deployment.

Abstract

In the pursuit of sustainable manufacturing, ultra-short pulse laser micromachining stands out as a promising solution while also offering high-precision and qualitative laser processing. However, unlocking the full potential of ultra-short pulse lasers requires an optimized monitoring system capable of early detection of defective workpieces, regardless of the preprocessing technique employed. While advances in machine learning can help predict process quality features, the complexity of monitoring data necessitates reducing both model size and data dimensionality to enable real-time analysis. To address these challenges, this paper introduces a machine learning framework designed to enhance surface quality assessment across diverse preprocessing techniques. To facilitate real-time laser processing monitoring, our solution aims to optimize the computational requirements of the machine learning model. Experimental results show that the proposed model not only outperforms the generalizability achieved by previous works across diverse preprocessing techniques but also significantly reduces the computational requirements for training. Through these advancements, we aim to establish the baseline for a more sustainable manufacturing process.

Figures (5)

• Figure 1: Data structure for the different experiments leveraged in this project. (Left) Set of workpiece experiments with diverse preprocessing techniques; (Center) For each workpiece, 99 sub-experiments are carried out with varying laser parameters; (Right) Resulting sensor data for each available sub-experiment, consisting of 5684 layers in total per experiment.
• Figure 2: Overall ML-based pipeline for optimized surface roughness prediction, (1) leverage data from the laser parameters and online sensors, (2) reduce sensor data input with feature extraction (see Fig.\ref{['fig:tsfel']}), (3) combine available input data to feed our models, (4) train a multi-layer perceptron (MLP) model for roughness prediction, (5) leverage the MLP weights to select most relevant features and, (6) by reducing the dimensionality of features, apply less computationally intensive ML models to predict roughness.
• Figure 3: Data preprocessing of the input sensor data that performs feature extraction using TSFEL, extracting characteristics from different domains.
• Figure 4: Scatter plots corresponding obtained on the evaluation set, representing data point predictions with respect to ground truth.
• Figure 5: (a): Example of the 20 selected Shapley feature values obtained for grinding preprocessing. (b): R2 test results on all preprocessing techniques, for different thresholds of selected features extracted with SHAP shap. Points on the extreme right of the figure refer to reference results obtained without reducing features.