Supervised machine learning with finite element residual stress prediction in a laser peened (Ti-6Al-7Nb) titanium alloy for medical applications

The durability and long-term survival of medical implants are major concerns for patients and surgeons. The laser shock peening (LSP) process can enhance the implant’s in-vivo lifespan through the compressive residual stress introduced on the implant’s surface. In the current research, a novel hybrid machine learning (ML) prediction tool was developed to calculate LSP induced residual stresses. Laser energies of 3 J, 5 J, 7 J with three overlapping levels of 33%, 50% and 67% with a constant laser spot diameter were introduced to a Ti-6Al-7Nb titanium hip implant material. A three-dimensional finite element model was developed incorporating an explicit dynamic analysis to capture the dynamic material response during the LSP process. Furthermore, The random forest ML algorithm was adapted so that the laser energy and overlapping were set as input parameters, while the associated residual stresses were set as output parameters. The mean squared error (MSE), root MSE and coefficient of determination for testing data sets concerning residual stress were 112.9%, 10.6% and 97%, respectively. The FEM and ML results both show a good agreement with the experimental data. The new proposed approach allows development of ML models even with a limited experimental data. The accuracy and performance of each model are discussed and the limitations are addressed. The approach carried out in this paper enables subsequent predictions of surface treatment techniques of implants through residual stress fields and can be applied to a wide range of applications.