This paper presents a new approach to predicting temperature fields in metal additive manufacturing (AM) using a physics-informed neural network framework. Traditional physics-based models are often time-consuming and unsuitable for real-time predictions, while machine learning models can be costly due to their reliance on high-quality datasets. The proposed framework uses a Convolutional Long Short-Term Memory (ConvLSTM) architecture and real-time temperature data to predict future 2D temperature fields across diverse geometries and process parameters. The framework was validated in two scenarios, achieving error rates below 3% and 1% respectively. It has been demonstrated that this approach can be applied across a range of scenarios with varying parameters, geometries, and deposition patterns.
Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.02403
