A multi-task spatiotemporal deep neural network for predicting penetration depth and morphology in laser welding

2026-06-24 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A novel multi-task deep learning model has been developed to accurately predict penetration state, depth, and weld seam morphology in laser penetration welding. This model integrates spatiotemporal features extracted from top weld pool images, captured by a complementary metal-oxide-semiconductor camera, with welding parameters. It employs a deep learning framework combining convolutional neural networks and state space models for efficient processing of spatial-temporal information. The research also proposes a reliable method for constructing the dataset to enhance robustness and generalization. Validation results show the model achieves 99.35% accuracy for penetration state prediction, a 1.79 millimeter error for penetration depth, and 95.65% accuracy for reconstructing the weld cross-section. This system offers new methodologies for in-situ quality control in laser welding.

Key takeaway

For Machine Learning Engineers developing in-situ quality control for laser welding, this model offers a robust framework. You should consider integrating spatiotemporal deep neural networks, specifically combining CNNs and state space models, to predict penetration depth and morphology. This approach achieves high accuracy, like 99.35% for state prediction. It enables more precise real-time adjustments, significantly improving weld quality assurance in your manufacturing processes.

Key insights

A multi-task spatiotemporal deep neural network accurately predicts laser weld quality using image and parameter data for in-situ control.

Principles

• Integrating spatiotemporal features improves weld quality prediction.
• Multi-task learning enhances prediction of related welding outcomes.
• Robust dataset construction is crucial for model generalization.

Method

The model integrates spatiotemporal features from weld pool images via CNNs and state space models with welding parameters to predict penetration state, depth, and morphology. A reliable dataset construction method is also proposed.

In practice

• Use CMOS cameras for real-time weld pool image capture.
• Combine CNNs and state space models for spatiotemporal data.
• Implement multi-task prediction for comprehensive weld quality.

Topics

• Laser Welding
• Penetration Depth Prediction
• Weld Morphology
• Spatiotemporal Neural Networks
• Multi-task Learning
• In-situ Quality Control

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video — A physics-guided deep spatiotemporal learning framework accurately estimates nearshore wave peak periods from video, leveraging multi-stage transfer learning and physical constraints.
• Asymmetric-Loss-Guided Hybrid CNN-BiLSTM-Attention Model for Industrial RUL Prediction with Interpretable Failure Heatmaps — A hybrid CNN-BiLSTM-Attention model with asymmetric loss improves RUL prediction and interpretability for industrial prognostics.
• Graph In-Context Operator Networks for Generalizable Spatiotemporal Prediction — In-context operator learning with diverse operators significantly outperforms classical methods for complex spatiotemporal prediction.
• Spatio-Temporal Models in AI: Teaching AI to Understand Space and Time — Spatio-Temporal AI models learn interconnected changes across space and time, crucial for dynamic real-world predictions.
• Deep learning-based in silico labeling for analyzing morphological features of MSCs to predict immunomodulatory capacity — A dual AI model non-invasively predicts MSC immunomodulatory capacity from morphology, enhancing cell therapy quality control.
• Data-Efficient Indentation Size Effect Correction in Steels Using Machine Learning and Physics-Guided Augmentation — Physics-guided data augmentation enables accurate ISE correction in steels with small experimental datasets.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.