Predicting the Recovery and Nonrecoverable Compliance Behaviour of Asphalt Binders Using Artificial Neural Networks

Abstract

Additives are widely used to enhance the rheological and performance properties of asphalt binder to satisfy the demands of extreme loading and climatic conditions. Meanwhile, adding to the complexity of asphalt binder behaviour that requires more time, effort, and material resources during laboratory work. The purpose of this research was to use Artificial Neural Networks (ANNs) to predict the recovery (R) and nonrecoverable compliance (Jnr) behaviour of asphalt binder based on mechanical test parameters and rheological properties of asphalt binder. A comprehensive experimental database consisting of the results of the frequency sweep and Multiple Stress Creep Recovery (MSCR) test using a dynamic shear rheometer (DSR) at five test temperatures (46 ∘C, 52 ∘C, 58 ∘C, 64 ∘C, and 70 ∘C). Prediction models for R and Jnr of asphalt binder modified with different contents of fly ash, fly ash-based geopolymer, glass powder/fly ash-based geopolymer, and styrene–butadiene styrene (SBS) were developed. The ANNs model was developed using five input parameters (temperature, frequency, storage modulus, loss modulus, and viscosity) and one hidden layer with five neurons. The results pointed out that the hybrid and 4%SBS binders achieved the highest ability to resist extremely heavy traffic and to recover the deformation with 60.1% and 85.5% at 46 ∘C, respectively, compared with the other modified asphalt binders. Excellent R-values for the total data set of 0.937, 0.997, 0.985, and 0.987 for Jnr3.2 of unaged binder, Jnr3.2 of aged binder, R3.2 of unaged binder, and R3.2 of aged binder, respectively. Therefore, the ANNs model is appropriate tool to predict the R3.2 and Jnr3.2 using unaged or aged binders at different temperatures.