Session: 04-08-01: Dynamics and Control of Aerospace Structures

Paper Number: 71348

Start Time: Monday, 06:10 PM

71348 - A Machine Learning Approach for Predicting Melt-Pool Dynamics of Ti-6Al-4V Alloy in the Laser Powder-Bed Fusion Process 

Over the last decade, powder-bed fusion (PBF) additive manufacturing (AM) has become a wildly popular technique for creating metallic or alloy parts, especially for aerospace, automotive, biomedical, dentistry, and electronics applications. In general, AM techniques save time, reduce costs, and require minimal proficiency of an operator.  However, the process can be defective and costly if the processing parameters are not set and optimized properly. Researchers usually perform trial and error methods using experiments and physics-based numerical modeling to determine the correct combination of the processing parameters, which takes substantial effort, time, and money.  This is where machine learning (ML) comes into play, by giving the scope for modeling and predicting desired outputs in a short time utilizing a very large data set of certain input parameters.  When that data set is obtained and provided as an input to the model, the computer can learn and produce its results.  The digital nature of the PBF process allows ML to identify and resolve the issues in manufacturing conveniently. One of the most important PBF processes is the laser PBF (L-PBF) process, where the melt-pool geometry is a significant output, showing the width and depth of penetration of the laser beam and the heat-affected zone within the workpiece. This output depends on several factors, including the material behavior, environment, and laser parameters. Among these factors, finding the optimized combination of the laser parameters, namely the laser power, scanning speed, and spot size, is extremely crucial when performing the L-PBF process effectively.  This paper presents a supervised neural network (NN) ML model to predict the melt-pool geometries of Ti-6Al-4V alloy in the L-PBF process focusing on the normalized values of the key features (i.e., the laser parameters) – laser power, scanning speed, and spot size. Information about the features and the corresponding target melt-pool width and depth are collected from an extensive literature survey. A trained data set is created with the melt-pool evolution images collected from experiments. The dataset is divided into training and testing sets before any feature engineering, visualization, and analysis, to prevent any data leakage. The k-fold cross-validation technique is applied to minimize the error and find the best performance. Results from the NN model include the best set of hyperparameter configurations and visual representation of the mean squared error as a function of the number of epochs. A logistic regression model is also created and compared with the NN model using the same data set to check the accuracy. The verification of the ML model is performed by comparing its results with the experimental and CFD modeling results for the melt-pool geometry at a given combination of the laser parameters. The melt-pool geometry outputs obtained for the NN model are consistent with the experimental and CFD modeling results.

Presenting Author: Jonathan Ciaccio University of New Orleans

Authors:

M. Shafiqur Rahman University of New Orleans
Jonathan Ciaccio University of New Orleans
Uttam K. Chakravarty University of New Orleans