Optimizing Melt Pool Temperature Prediction Using Convolutional Bilstm with Insights from Dragonfly Behavior in Wire-Arc Additive Manufacturing

Abstract

Wire-Arc Additive Manufacturing (WAAM) has received a lot of attention in recent years because of its ability to create large-scale metallic components layer by layer. Monitoring and controlling the melt pool temperature in real-time, which is a significant factor in deciding the quality of the manufactured part, is a significant problem in WAAM. In this research, we introduce a novel approach for predicting melt pool temperature in wire arc additive manufacturing by employing a Dragonfly optimized convolutional Bi-directional Long Short-Term Memory (DragOCBiLSTM), inspired by insights derived from the behavior of dragonflies. The Convolutional layers efficiently extract spatial characteristics from multi-sensor data, while the Bi-directional LSTM (BiLSTM) layers capture temporal correlations within the data. The utilization of these two elements, refined using the algorithm inspired by dragonfly behavior, presents a significant advantage in comparison to existing predictive models. The data are normalized using the Z-score normalization approach. Principal Component Analysis (PCA) is then used to extract the characteristics from the cleaned data. After that, Dragonfly Optimization (DO) is used to find the best feature subsets.The proposed method’s performance is assessed in terms of Mean Absolute Error (MAE) (10.984), Mean Absolute Percentage Error (MAPE) (3.404), and Mean Squared Error (MSE) (11.25)metrics and compared with existing methods. We provide a promising approach for optimizing the prediction of melt pool temperatures in WAAM, with possible implications for other manufacturing processes, by utilizing the distinctive behavioral insights of dragonflies and merging these with innovative deep learning architectures.