Applicability of Machine Learning to Predict the Flexural Stresses in Jointed Plain Concrete Pavement

Abstract

The development of critical flexural stresses causes variations in the design thickness requirements of jointed plain concrete pavement (JPCP), which are influenced by various combinations of design parameters. It is important to study these design parameters, such as maximum temperature difference, strength of the foundation, slab thickness, and vehicular load, to determine maximum flexural stresses. A total of 480 design conditions (2,880 data sets) for single- and tandem-axle loading each were analyzed and considered covering all practical range of design parameters combinations. Equations for determining maximum flexural stresses in terms of design parameters were derived using machine learning (ML) algorithms: multiple linear regression (MLR), support vector regression (SVR), and random forest (RF) with 3-fold cross-validation. These ML algorithms have shown a good correlation of R2=0.918, 0.927, and 0.929 for single-axle load and R2=0.915, 0.905, and 0.934 for tandem-axle load, respectively. These equations were used to find the critical flexural stresses for the thickness design of JPCP. The computed maximum flexural stresses were also compared with different design approaches such as simplified approach, stress charts, and regression equations for the thickness design of the JPCP slab. Sensitivity analysis was performed to assess the effects of adjusting input variable data by 10% while keeping the other input parameters constant. The radius of relative stiffness and stress coefficient had the greatest influence in maximum flexural stresses. This study would be helpful for the prediction of design parameters effect and precise determination of maximum flexural stresses in JPCP.