AI Research

Multi-Head Attention-Based Feature Extractor Integr... | AI Research

Key Takeaways

Multi-Head Attention-Based Feature Extractor Integration with Soft Actor-Critic for Porosity Prediction and Process Parameter Optimization in Additive Manufa...
Additive manufacturing process optimization requires precise parameter control to minimize defects such as porosity.
Traditional reinforcement learning (RL) approaches using discrete action spaces suffer from slow convergence and susceptibility to local optima, limiting their effectiveness for high-precision manufacturing tasks.
This study addresses these limitations by employing a continuous action space combined with a novel architecture that integrates a multi-head attention mechanism with the Soft Actor-Critic (SAC) algorithm.
The proposed methodology achieves a convergence value of 322.79 within 14 episodes, outperforming existing approaches while maintaining stability throughout training.

Paper AbstractExpand

Additive manufacturing process optimization requires precise parameter control to minimize defects such as porosity. Traditional reinforcement learning (RL) approaches using discrete action spaces suffer from slow convergence and susceptibility to local optima, limiting their effectiveness for high-precision manufacturing tasks. This study addresses these limitations by employing a continuous action space combined with a novel architecture that integrates a multi-head attention mechanism with the Soft Actor-Critic (SAC) algorithm. The attention-based feature extractor enhances the agent's ability to capture subtle variations in low-dimensional input features, enabling more effective exploration-exploitation balance for navigating value spaces with local minima. We validate our approach on porosity prediction and process parameter optimization in laser powder bed fusion, demonstrating faster convergence and higher final reward values compared to standard RL methods including DQN, PPO, TD3, and vanilla SAC. The proposed methodology achieves a convergence value of 322.79 within 14 episodes, outperforming existing approaches while maintaining stability throughout training.

Multi-Head Attention-Based Feature Extractor Integration with Soft Actor-Critic for Porosity Prediction and Process Parameter Optimization in Additive Manufacturing
This research addresses the challenge of minimizing defects, such as porosity, in metal additive manufacturing (AM). While traditional reinforcement learning (RL) methods have been used to optimize manufacturing parameters, they often struggle with slow convergence or getting stuck in local optima. This paper introduces a new architecture that combines a multi-head attention mechanism with the Soft Actor-Critic (SAC) algorithm to improve how an agent explores manufacturing environments and optimizes process parameters in a continuous action space.

Enhancing Feature Extraction

The core innovation of this study is the integration of a multi-head attention-based feature extractor into the SAC framework. By using self-attention, the model can better capture subtle variations in input data, such as melt pool depth, width, and laser energy density. This allows the agent to understand the structure of the input space more effectively, enabling it to navigate complex value landscapes—which contain small local minima—more efficiently than standard RL approaches.

Balancing Exploration and Exploitation

The researchers selected the SAC algorithm for its ability to maintain a dynamic balance between exploration and exploitation through maximum entropy. While standard SAC can sometimes be slow to converge due to its high exploration rate, the addition of the attention-based feature extractor allows the agent to map inputs into more informative features. This enhancement helps the agent locate the global optimum faster, providing a more stable and efficient learning process compared to traditional methods like DQN, PPO, and TD3.

Performance and Results

The proposed methodology was validated through simulations focused on porosity prediction and process parameter optimization in laser powder bed fusion. The results indicate that the new approach achieves faster convergence and higher final reward values than the other tested RL methods. Specifically, the proposed model reached a convergence value of 322.79 within just 14 episodes. The study demonstrates that this architecture maintains stability throughout the training process, outperforming existing RL benchmarks in both speed and final performance.

Comments (0)

No comments yet

Be the first to share your thoughts!