[Paper Review] Scalable AI Framework for Defect Detection in Metal Additive Manufacturing

Duy Nhat Phan, Srirang Jha+5 more|arXiv (Cornell University)|Nov 1, 2024

Additive Manufacturing Materials and Processes4 citations

TL;DR

This paper presents a scalable AI framework using convolutional neural networks (CNNs) to detect defects in metal additive manufacturing via thermal imaging. By leveraging GAN-generated synthetic data and a denoising pipeline, the framework achieves high-accuracy defect detection without human-labeled data, and integrates into a user-friendly CLADMA interface for practical industrial deployment in AM processes.

ABSTRACT

Additive Manufacturing (AM) is transforming the manufacturing sector by enabling efficient production of intricately designed products and small-batch components. However, metal parts produced via AM can include flaws that cause inferior mechanical properties, including reduced fatigue response, yield strength, and fracture toughness. To address this issue, we leverage convolutional neural networks (CNN) to analyze thermal images of printed layers, automatically identifying anomalies that impact these properties. We also investigate various synthetic data generation techniques to address limited and imbalanced AM training data. Our models' defect detection capabilities were assessed using images of Nickel alloy 718 layers produced on a laser powder bed fusion AM machine and synthetic datasets with and without added noise. Our results show significant accuracy improvements with synthetic data, emphasizing the importance of expanding training sets for reliable defect detection. Specifically, Generative Adversarial Networks (GAN)-generated datasets streamlined data preparation by eliminating human intervention while maintaining high performance, thereby enhancing defect detection capabilities. Additionally, our denoising approach effectively improves image quality, ensuring reliable defect detection. Finally, our work integrates these models in the CLoud ADditive MAnufacturing (CLADMA) module, a user-friendly interface, to enhance their accessibility and practicality for AM applications. This integration supports broader adoption and practical implementation of advanced defect detection in AM processes.

Motivation & Objective

• To address the challenge of limited and imbalanced real-world training data for defect detection in metal additive manufacturing (AM).
• To develop a scalable AI framework that enables automated, high-accuracy defect detection using thermal images from laser powder bed fusion processes.
• To evaluate the effectiveness of synthetic data generation techniques, particularly GANs, in improving model generalization and reducing reliance on human-annotated data.
• To integrate the trained models into a user-friendly cloud-based interface (CLADMA) for practical adoption in industrial AM workflows.

Proposed method

• The framework employs convolutional neural networks (CNNs) to analyze thermal images captured during layer-by-layer metal AM processes.
• Synthetic data is generated using Generative Adversarial Networks (GANs) to augment real training data, improving model robustness and generalization.
• A denoising pipeline is applied to enhance thermal image quality, reducing noise that could interfere with defect detection.
• The framework is evaluated on real thermal images of Nickel alloy 718 layers produced via laser powder bed fusion.
• A CLADMA module is developed to integrate the trained models into a cloud-based, user-friendly interface for industrial deployment.
• The approach compares performance across datasets with and without added noise, validating robustness and scalability.

Experimental results

Research questions

• RQ1 Can GAN-generated synthetic data significantly improve defect detection accuracy in metal AM when real data is scarce or imbalanced?
• RQ2 How does image denoising impact the reliability and performance of defect detection models in thermal imaging for AM?
• RQ3 To what extent can a scalable AI framework reduce dependency on human-annotated data in metal AM defect detection?
• RQ4 How effective is the integration of trained models into a cloud-based, user-friendly interface (CLADMA) for real-world AM applications?

Key findings

• The use of GAN-generated synthetic data led to significant improvements in defect detection accuracy, outperforming models trained on real data alone.
• The denoising pipeline effectively enhanced thermal image quality, resulting in more consistent and reliable defect detection across varying noise levels.
• The CLADMA interface successfully enabled practical deployment of the AI models, demonstrating usability and accessibility for industrial users.
• The framework achieved high performance even with limited real data, highlighting the value of synthetic data in overcoming data scarcity in AM defect detection.
• Models trained on synthetic data with noise demonstrated strong generalization, maintaining high accuracy on real test data.
• The integration of GAN-based data generation eliminated the need for human intervention in data preparation, streamlining the training pipeline.