A surrogate model for estimating rock stress by a hollow inclusion strain cell in a three-layer medium

• Existing analytical solutions for stress calculation using HI strain cells are limited to double-layer media and are inaccurate for three-layer media, leading to significant errors. • A numerical simulation method using FLAC3D was developed to obtain constitutive matrices relating strain measurements to stress tensors in a three-layer medium. • An artificial neural network surrogate model was trained on 764 numerical models, achieving high accuracy (R2=0.999, MSE=1.254) for predicting constitutive matrices. • The surrogate model was validated through numerical simulations, analytical solutions, and laboratory experiments, and successfully applied to correct stress calculations in a coal mine roadway roof.