Integrated PINN and inverse mapping framework for prescriptive design of hybrid fiber-reinforced concrete

Wasim Abbass , Muhammad Usman Zubair , Fahid Aslam

Abstract

This paper presents a physics informed Bayesian design of Hybrid Fiber Reinforced Concrete. The research changes the emphasis from the prediction of strength to the prescription of mix formulation. The framework was supported by a dataset of 938 samples comprising of 602 laboratory specimens and 336 physics augmented samples. A Physics Informed Neural Network that was trained on composite mechanics constraints had R2=0.93 and a mean absolute error of 2.7 MPa at 28 days. A closed form constitutive relation of compressive behavior was also derived by the model. Bayesian calibration showed the empirical coverage of 92.8% of 95% prediction intervals. An inverse mapping which had been regularized with a cost index, then determined combinations of fibers that would give target strengths between 45 and 85 MPa and still be economically viable. 70 experimental tests of mixtures showed that the mean absolute error is 3.8 MPa which corresponds to approximately 6.5% relative error and that the mixtures saved 18 to 25% cost in case of materials compared with conventional reference designs. These results indicate that the physics-informed neural networks could assist in better mixture design and cost-effective solutions to manufacturing sustainable management and conservation of hybrid fiber reinforced concrete.

Key Words

Bayesian uncertainty; cost optimization; hybrid fiber concrete; inverse design; physics informed neural networks; prescriptive formulation

Address

Wasim Abbass — Department of Civil Engineering, University of Engineering and Technology, Lahore, Pakistan
Muhammad Usman Zubair — Government College University (Roll No. 0572-1-24), Lahore, Pakistan
Fahid Aslam — Department of Civil Engineering, College of Engineering in Al-Kharj, Prince Sattam bin Abdulaziz University, Kharj, 11942, Saudi Arabia

PDF Viewer

Preview is limited to the first 3 pages. Sign in to access the full PDF.

← Back to Volume 38, No. 1