Identification of Fiber and Asphalt Mastic Interface Failure Modes Based on the Improved Growth Curve Mixture Model

The interface failure modes between fibers and asphalt mastic exhibit complex and diverse morphologies under the influence of multiple factors, making cluster analysis difficult. To address this issue, this study proposes an improved growth curve mixture model (IGCMM) for the unsupervised clustering of interface failure modes. Through single-fiber pull-out tests on 90 specimens under three temperatures (-10 °C, 25 °C, and 60 °C) and three fiber types (basalt, glass, and polyester), load-displacement curves were obtained. The multivariate power exponential (MPE) distribution was used to characterize the peak and heavy-tailed features of residuals. Due to the high dimensionality caused by the joint analysis of multiple datasets, principal component analysis (PCA) was employed to compress the 200-dimensional curve data into three principal components, with all model parameters estimated in the low-dimensional space. The Expectation-Maximization (EM) algorithm combined with the extended Bayesian Information Criterion (eBIC) determined the optimal eight failure mode clusters. The results exhibit a high posterior assignment confidence: 96.7% of samples had posterior probabilities exceeding 0.999. The eight statistical patterns were successfully mapped to four theoretical failure modes-fiber pull-out, medium-temperature matrix failure, high-temperature matrix failure, and mixed failure-revealing the coupled regulatory mechanism of temperature and fiber type on interface failure. The framework established in this study-"data-driven clustering-physical parameter anchoring-failure mechanism interpretation"-provides new theoretical tools and methodological support for the mesoscale interface failure diagnosis and crack resistance optimization design of fiber-reinforced asphalt pavement materials.