Laser beam welding (LBW) involves complex and rapid interactions between the laser and material, often resulting in defects such as pore formation. Emissions collected during the process offer valuable insight but are difficult to interpret directly for defect detection. In this study, we propose a data-driven framework to interpret electromagnetic emissions in LBW using both supervised and unsupervised learning. Our framework is implemented in the post-process monitoring stage and can be used as a real-time framework. The supervised approach uses labeled data corresponding to predefined defects (in this work, pore formation is an example of a defined defect). Meanwhile, the unsupervised method is used to identify anomalies without using predefined labels. Supervised and unsupervised learning aims to find reference values in the emissions data to determine the values of signals that lead to defects in welding (enabling quantitative monitoring). A total of 81 welding experiments were conducted, recording real-time emission data across 42 spectral channels. From these signals, statistical, temporal, and shape-based features were extracted, and dimensionality was reduced using Principal Component Analysis (PCA). The LSTM model achieved an average mean squared error (MSE) of 0.0029 and mean absolute error (MAE) of 0.0288 on the testing set across five folds. The Isolation Forest achieved 80% accuracy and 85.7% precision in detecting anomalous welds on a subset with validated defect labels. The proposed framework enhances the interpretability of 4D photonic data and enables both post-process analysis and potential real-time monitoring. It provides a scalable, data-driven approach to weld quality assessment for industrial applications.

To advance quality assurance in the welding process, this study presents a deep learning (DL) model that enables the prediction of two critical welds’ key performance characteristics (KPCs): welding depth and average pore volume. In the proposed approach, a wide range of laser welding key input characteristics (KICs) is utilized, including welding beam geometries, welding feed rates, path repetitions for weld beam geometries, and bright light weld ratios for all paths, all of which were obtained from hairpin welding experiments. Two DL networks are employed with multiple hidden dense layers and linear activation functions to investigate the capabilities of deep neural networks in capturing the complex nonlinear relationships between the welding input and output variables (KPCs and KICs). Applying DL networks to the small numerical experimental hairpin welding dataset has shown promising results, achieving mean absolute error values of 0.1079 for predicting welding depth and 0.0641 for average pore volume. This, in turn, promises significant advantages in controlling welding outcomes, moving beyond the current trend of relying only on defect classification in weld monitoring to capture the correlation between the weld parameters and weld geometries.

Nonlinear multiphysics systems, such as the melt-pool temperature field in laser welding, are difficult to predict efficiently: high-fidelity CFD solvers are accurate but computationally expensive, and the dynamics are non-stationary and characterized by sharp thermal gradients. This work introduces and compares two surrogate models for full-field, time-resolved melt-pool temperature prediction from laser power, speed, and time. The first is a black-box deep learning surrogate that treats each spatial location as a pixel trajectory and uses a two-branch architecture with Temporal Convolutional Network (TCN) blocks followed by a BiLSTM to capture history-driven dynamics, in parallel with a Multi-Layer Perceptron (MLP) that only sees the last time step. A hot-zone-weighted mean-squared error loss emphasizes accuracy in the melt pool. The second is a physics-aware surrogate that applies sliding-window Dynamic Mode Decomposition (SW-DMD) to separate slow and fast thermal modes, projects the fields onto a shared POD basis, and uses two small MLPs to map (power, speed, time) to the complex modal coefficients before reconstructing the temperature field. The novelty of this work lies in the deep-learning design, where an extra MLP branch and a hot-zone-weighted loss increase melt-pool fidelity in the first model, and in the second model, which uses a physics-aware modal representation that enables the network to learn directly on the coefficients of the underlying dynamical modes. Both surrogates are trained on eight Flow-3D laser-weld simulations of AA1050, resampled to 50 snapshots per power–speed combination. On a held-out high-power weld, the SW-DMD–MLP surrogate halves the average RMSE in the 700–950 K solidification range compared with the TCN–BiLSTM+MLP model (≈48 K vs ≈98 K), achieves high overlap of the hot region, and is about four orders of magnitude faster to train. These results show that embedding physics-aware modal structure before learning leads to more accurate and resource-efficient surrogates than black-box data-driven models for this nonlinear weld-dynamics problem.

During laser welding, temperature changes rapidly as the keyhole and melt pool develop. Computational fluid dynamics (CFD) can simulate this behavior, but evaluating many laser power and welding speed combinations is computationally expensive. Using eight FLOW-3D simulations of AA1050 laser welding, resampled to 50 time steps, this study compares two surrogate models with different temperature-field representations. The direct deep learning surrogate predicts temperature at individual spatial locations using a temporal convolutional network-bidirectional long short-term memory (TCN-BiLSTM). Their predictions reconstruct the complete field, while a hot-zone-weighted loss emphasizes high-temperature regions. The physics-aware surrogate applies sliding-window dynamic mode decomposition (SW-DMD) to separate temperature evolution into slow and fast components. Separate proper orthogonal decomposition bases represent these components, and two MLPs predict their coefficients from laser power, welding speed, and time. For the training, validation, and test conditions, respectively, the physics-aware surrogate reduced mean RMSE from 155.4 to 39.4 K (74.6%), from 97.6 to 47.8 K (51.0%), and from 343.2 to 106.0 K (69.1%). Spatial agreement was assessed using mean intersection over union (IoU) at 900 K across three common time steps. For validation, IoU increased from 61.0 to 74.3%. However, it decreased from 79.1 to 74.1% for training and from 17.9 to 16.2% for testing. Thus, the physics-aware representation substantially reduced overall temperature error but did not consistently improve the shape and position of the hottest region. These findings distinguish improved field accuracy from reliable prediction of critical high-temperature spatial boundaries.