Investigation of the relationship between indirect tensile stiffness modulus and non-destructive tests in road pavements

Creative Commons License

Serin S., Saltan M., Terzi S.

REVISTA DE LA CONSTRUCCION, cilt.25, sa.2, ss.400-417, 2026 (SCI-Expanded, Scopus)

Özet

Hot Mix Asphalts (HMA) may lose their stiffness over time due to traffic loads and environmental effects, leading to structural deformations. The Indirect Tensile Stiffness Modulus (ITSM) test, which is commonly used to determine such properties, requires coring and measurements conducted in laboratory environments using specialized equipment. Therefore, it is both costly and damaging to the pavement structure. This situation necessitates the investigation of faster, more economical, and non-destructive alternative methods. In this study, statistical and artificial intelligence-based relationships between ITSM and various non-destructive testing (NDT) methods (PQI, NDG, LWD, UPV, BP) were examined. The fieldwork was carried out on a 16 km section of the highway between Antalya and Burdur in Turkey. A total of 80 core samples were taken from 20 different locations at one-year intervals, and various non-destructive tests were also performed at the same points. The ITSM results obtained from the core samples were estimated using the non-destructive test data collected from the field. Both classical correlation analyses and Artificial Neural Network (ANN) models for multivariate prediction were employed in the analyses. The findings revealed that certain non-destructive test methods could estimate ITSM values with high accuracy. In particular, data obtained from LWD and UPV devices showed strong linear relationships with ITSM, yielding correlation coefficients above r>0.75 in single-variable correlation analyses. Among the ANN models developed in the study, the most successful model, M6, was structured using all non-destructive test data collectively. This model demonstrated a remarkable performance with an R² value of 91.18% for ITSM prediction. On the other hand, models based on individual test methods were observed to have relatively limited prediction success, generally ranging between 58% and 66%.