Machine Learning for Smart Post Processing of Additively Manufactured Ti-6Al-4V for Biomedical Applications

Additive manufacturing (AM) is the emerging manufacturing technology that fabricates near-net shape components directly from CAD data files in a layer-by-layer fashion. Titanium and its alloys have been widely used in healthcare applications, such as dental implants and artificial prosthesis. The AM technology is an attractive technology to fabricate the personalized biomedical products using Ti-6Al-4V. Currently, the healthcare applications accounts for the biggest share in metal AM market. To relieve the residual stress, reduce the porosity, modify the microstructure and achieve isotropic mechanical properties, heat treatment is required for as-fabricated AM Ti-6Al-4V biomedical components. A comprehensive understanding of the heat treatment on AM Ti-6Al-4V parts is needed to produce components that meet the property requirements for biomedical applications.

In this study machine learning is employed to develop a data-driven model and identify the correlations between heat treatment parameters of AM Ti-6Al-4V parts and resulting microstructures and mechanical properties.

A database that includes the processing parameters of various heat treatment approaches, including the Hot Isostatic Pressing (HIP), Mill-Anneal (MA) and Anneal (ANN), has been established. This database also contains microstructure and mechanical properties data, as well as the AM processing parameters. The model is developed using Tensorflow to predict the process-microstructure-properties relationship with an accuracy of 85%. The established model indicates that if the yield strength and tensile strength of the as-fabricated AM samples are “good enough”, their variations are not key influencers for the mechanical properties of the heat treated samples. The mechanical properties after the heat treatment strongly depend on the heat treatment parameters, including temperature, time, pressure, and heat treatment method. However, the sample elongation after heat treatment shows a strong dependence on the elongation of the as-fabricated AM samples. An image learning study is also conducted to simulate the microstructure evolution during heat treatment.