[16P-T10-06] Optimizing Additive Manufacturing Processes: A Predictive System Approach
キーワード：Additive manufacturing, Artificial intelligence, Deep neural network, Density, Hyper-parameter
Additive manufacturing demands precise process control of the process to achieve the high quality of product [1] and has traditionally depended on resource-intensive trial-and-error methods [2]. This study introduces a prediction system designed to enhance the performance through the optimized process conditions. A prediction system based on employed deep neural network (DNN) model and random search method [3] was developed to optimize the process conditions of additive manufacturing. DNN model was trained to identify the relationship between the processing conditions including hatching angle, hatching distance, laser power, layer thickness, and scan speed and density. The training dataset was collected under experimental conditions based on the Taguchi design, supplemented with randomized processing conditions as part of a mixed sampling method. The hyper-parameter tuning was carried out to optimize the DNN model, with the best parameters for activation function, batch size, dropout, epoch, layer, learning rate, and node determined through a random search method. The cross-validation method was additionally used to overcome the small amount of data. The trained model is loaded into a dedicated web GUI, offering a specialized platform for its application in practical field scenarios.