In-Situ Melt Pool Characterization via Thermal Imaging for Defect Detection in Directed Energy Deposition Using Vision Transformers

2024-11-18

arXiv.org

AI Summary

Real-time detection of internal defects—such as porosity—in the melt pool during Directed Energy Deposition (DED) remains challenging due to limited accessibility and scarcity of labeled training data. Method: This paper proposes an unsupervised, online quality control framework integrating thermal imaging with a Vision Transformer architecture. Specifically, we adapt the Masked Autoencoder (MAE) to learn self-supervised spatiotemporal representations from raw melt pool thermal image sequences—eliminating the need for annotated data—and couple it with a lightweight MLP classifier head fine-tuned via transfer learning for defect identification under few-shot conditions. Results: Evaluated on real-world DED processes, the method achieves an overall accuracy of 95.44%–99.17% and an average F1-score exceeding 80%. It significantly reduces reliance on scarce expert-labeled data while ensuring high accuracy, low computational overhead, and scalability for industrial deployment—establishing a novel paradigm for in-situ quality monitoring in additive manufacturing.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Directed Energy Deposition (DED) offers significant potential for manufacturing complex and multi-material parts. However, internal defects such as porosity and cracks can compromise mechanical properties and overall performance. This study focuses on in-situ monitoring and characterization of melt pools associated with porosity, aiming to improve defect detection and quality control in DED-printed parts. Traditional machine learning approaches for defect identification rely on extensive labeled datasets, often scarce and expensive to generate in real-world manufacturing. To address this, our framework employs self-supervised learning on unlabeled melt pool data using a Vision Transformer-based Masked Autoencoder (MAE) to produce highly representative embeddings. These fine-tuned embeddings are leveraged via transfer learning to train classifiers on a limited labeled dataset, enabling the effective identification of melt pool anomalies. We evaluate two classifiers: (1) a Vision Transformer (ViT) classifier utilizing the fine-tuned MAE Encoder's parameters and (2) the fine-tuned MAE Encoder combined with an MLP classifier head. Our framework achieves overall accuracy ranging from 95.44% to 99.17% and an average F1 score exceeding 80%, with the ViT Classifier slightly outperforming the MAE Encoder Classifier. This demonstrates the scalability and cost-effectiveness of our approach for automated quality control in DED, effectively detecting defects with minimal labeled data.

Problem

• Detect internal defects in DED
• Improve melt pool monitoring
• Enable defect identification with minimal labeled data

Innovation

• Self-supervised learning for unlabeled data
• Vision Transformer-based Masked Autoencoder
• Transfer learning with limited labeled datasets