[Paper Review] Reinforcement Learning-based Defect Mitigation for Quality Assurance of Additive Manufacturing

Jihoon Chung, Bo Shen+3 more|arXiv (Cornell University)|Oct 28, 2022

Additive Manufacturing and 3D Printing Technologies4 citations

TL;DR

This paper proposes Continual G-learning, a model-free reinforcement learning framework that leverages both offline literature-based knowledge and online learning during printing to mitigate unforeseen defects in fused filament fabrication (FFF) additive manufacturing. By integrating prior and real-time knowledge, the method achieves effective defect mitigation with minimal sample requirements, demonstrating superior performance in both numerical and real-world FFF case studies.

ABSTRACT

Additive Manufacturing (AM) is a powerful technology that produces complex 3D geometries using various materials in a layer-by-layer fashion. However, quality assurance is the main challenge in AM industry due to the possible time-varying processing conditions during AM process. Notably, new defects may occur during printing, which cannot be mitigated by offline analysis tools that focus on existing defects. This challenge motivates this work to develop online learning-based methods to deal with the new defects during printing. Since AM typically fabricates a small number of customized products, this paper aims to create an online learning-based strategy to mitigate the new defects in AM process while minimizing the number of samples needed. The proposed method is based on model-free Reinforcement Learning (RL). It is called Continual G-learning since it transfers several sources of prior knowledge to reduce the needed training samples in the AM process. Offline knowledge is obtained from literature, while online knowledge is learned during printing. The proposed method develops a new algorithm for learning the optimal defect mitigation strategies proven the best performance when utilizing both knowledge sources. Numerical and real-world case studies in a fused filament fabrication (FFF) platform are performed and demonstrate the effectiveness of the proposed method.

Motivation & Objective

• To address the challenge of real-time defect mitigation in additive manufacturing due to time-varying process conditions.
• To reduce the number of required training samples for learning defect mitigation strategies in low-volume, customized AM production.
• To develop an online learning framework that combines offline knowledge from literature with online knowledge acquired during printing.
• To create a model-free reinforcement learning method that adapts dynamically to new defects during the AM process.
• To validate the effectiveness of the proposed method in both simulated and real-world fused filament fabrication environments.

Proposed method

• The method employs a model-free reinforcement learning approach to learn optimal control policies for defect mitigation in real time.
• It introduces Continual G-learning, a novel algorithm that integrates offline knowledge (from literature) and online knowledge (acquired during printing) to accelerate learning.
• The framework uses a reward function designed to penalize defect occurrence and encourage process stability during printing.
• It applies function approximation to represent the policy and value functions, enabling generalization across similar defect scenarios.
• The algorithm continuously updates its policy using on-policy experience collected during the printing process.
• Knowledge distillation techniques are used to transfer prior knowledge from offline sources into the initial policy network, reducing sample complexity.

Experimental results

Research questions

• RQ1 Can a reinforcement learning framework effectively mitigate new, unforeseen defects during the additive manufacturing process in real time?
• RQ2 How does combining offline literature-based knowledge with online learning during printing improve sample efficiency in defect mitigation?
• RQ3 What is the performance gain of using both knowledge sources compared to using only online or only offline knowledge?
• RQ4 How does the proposed Continual G-learning algorithm maintain robustness and adaptability across diverse defect types in FFF printing?
• RQ5 To what extent can the method reduce the number of required training samples while achieving high defect mitigation accuracy?

Key findings

• The proposed Continual G-learning method achieved the best defect mitigation performance by effectively combining offline and online knowledge sources.
• The integration of prior knowledge from literature significantly reduced the number of required training samples, improving sample efficiency.
• In real-world FFF case studies, the method successfully mitigated previously unseen defects during the printing process with high reliability.
• The model-free RL approach demonstrated strong adaptability to dynamic and time-varying process conditions in additive manufacturing.
• Numerical studies confirmed that the method outperformed baseline approaches that relied solely on online learning or offline knowledge.
• The framework showed robust generalization across different defect types and printing scenarios, validating its practical applicability in industrial settings.