Developing Machine Learning Models to Predict Influence of Heat Treatments on the Tensile Properties of Ti6Al4V Parts Prepared by Selective Laser Melting

Yang, Zhaotong

Etd

Developing Machine Learning Models to Predict Influence of Heat Treatments on the Tensile Properties of Ti6Al4V Parts Prepared by Selective Laser Melting

Public

This study aimed to establish data-driven models to accurately predict the mechanical properties of Ti6Al4V via selective laser melting with various heat treatments. The heat treatment methods including annealing and hot isostatic pressing (HIP) were studied in depth. The tool of artificial neural network (ANN) was used to establish the models. The data were collected from research publications and reports, using Google Scholar, Science Direct, Research Gate, and Springer between 2006-2020. The models used the mechanical properties of as-printed Ti6Al4V samples and heat treatment parameters as inputs and mechanical properties of post heat treatment samples as outputs. The mechanical properties including yield strength (YS), ultimate tensile strength (UTS), and elongation (EL) were considered. HIP is widely applied on as-printed Ti6Al4V fabricated by selective laser melting (SLM) process to improve the tensile properties, relieve residual stress, reduce the anisotropic behavior, close defects and modify the microstructure. In the process, the metal is heated to a certain temperature and exposed to a high pressure (~100MPa) under protective gas atmosphere, followed by air cooling. There were a total of 40 groups of data used in the HIP study. Typical HIP parameters are holding time, holding temperature and pressure. The pressure value was found to be 100 MPa in most collected reports. So, although pressure is included in the model built by this study, it is a constant. A variation in pressure may be needed in future studies to determine the influence of pressure on final properties of SLM Ti6Al4V. Another heat treatment process widely used on as-printed SLM Ti6AL4V parts is annealing. During annealing, the alloy is exposed to a high temperature for a certain time followed by air cooling. Compared with HIP, annealing can’t close the defects but is more cost-efficient. Similarly, the mechanical properties of as-printed Ti6Al4V, annealing parameters including annealing holding time and temperature were included in the study. There were 44 groups of collected data used to build and test the model for annealing process. The established models show high accuracy in predicting YS and UTS for SLM Ti6Al4V after the heat-treated by annealing process and HIP process. In the annealing study, for YS prediction, 88.6% of prediction results were within the 5% error range for the full dataset, the MAE was found to be 31.2MPa, the accuracy was 96.7%. For UTS, these three numbers were 88.6%, 35.5MPa and 96.6%. In the HIP study, it was found that 87.5% of prediction results were within 5% error range for full dataset for YS. For UTS, this number was 100%. However, the ANN model could not directly reveal relationships between inputs and outputs. It is a well-accepted method to use the established model, to generate predictions under different combinations of input conditions, to reveal the influences of input parameters. In this work, graphs of inputs (initial mechanical properties and heat treatment parameters,) vs. outputs (YS, TS and EL) were generated by the established ANN models and utilized to reveal the influence of a variety of input features. It was found that the final YS and UTS was very sensitive to the holding time and temperature for both annealing and HIP processes. Longer holding time and higher holding temperature resulted in more significant grain growth, and finally led to decreasing in both final YS and UTS of heat-treated Ti6AL4V samples. A new Hall-Petch relationship was proposed to build the relationship between grain size and YS for heat-treated SLM Ti6Al4V parts in this work as Equation 1: σ_Y=767MPa+(191MPa˖〖µm〗^(1/2))/√(D_α ) (1) The results show that the final mechanical properties of SLMed Ti6Al4V parts were sensitive to initial mechanical properties for the annealing process. The higher initial YS and UTS of as-printed samples resulted in higher final YS and UTS after annealing. But, the initial tensile properties were found to be a weak influence factor for SLM Ti6Al4V parts after HIP. Unlike the HIP process, the annealing process cannot remove the defects in the as-printed sample, which means the porosity remains after the annealing treatment. Thus, the printing quality may significantly the quality of final parts after annealing. ASTM F136-13 standard (Standard Specification for Wrought Titanium-6Aluminum-4Vanadium Extra Low Interstitial) was used to guide the prediction experiments. The ANN result suggested that an annealing process of the holding time less than 4 hours and the holding temperature lower than 850°C could be recommended for SLM printed Ti6Al4V parts to reach the YS requirement in ASTM F136-13 standard. As for the HIP process, a holding temperature lower than 970°C and a holding time of around 2 hours, was sufficient for SLM printed Ti6Al4V parts to achieve mechanical properties in compliance with the ASTM F136-13 standard. The established models showed less accuracy in predicting elongation after annealing and HIP. It is well known that a small volume fraction of porosity (<<1%) can lead to poor elongation in as-printed samples. HIP closes some of the defects but not all. The differences among printers and printing processes used in studies that were included in this study meant different defects. A large variation in defects would result in a large scattering of the elongation data, and finally led to the low accuracy of the established models in predicting elongation.

Creator