A diagnostic method for evaluating the condition index of cement-stabilised base using T-S fuzzy neural network

This paper presents a novel diagnostic model for evaluating the damage condition of a cement-stabilised base course using a T-S fuzzy neural network (FNN). An evaluation criterion for core damage rating, base core condition index (BCCI), was established based on the distress features observed in 369 core samples that had been collected from asphalt concrete pavements with cement-stabilised base courses. The core samples of cement-stabilised macadam (CSM) were classified into five levels according to the evaluation criterion. Ten parameters were chosen as inputs to establish non-linear mapping relationships in an FNN model. These input parameters include pavement distresses’ characteristics (crack depth, breadth and lumpiness), Pavement Surface Condition Index (PCI), Riding Quality Index (RQI), Pavement Structure Strength Index (PSSI), Cumulative Equivalent Single Axle Loads (ESALs), thickness of asphalt concrete (AC) layers, annual temperature difference and average annual precipitation. Out of the 369 samples, 169 field cores were used in the FNN model for recursive training, and the parameters for the neural network structure were optimised until the errors between the network outputs and the expected outputs were minimised. The established FNN model was then used for the calculations of predicted quantitative results for remaining sections. The calculation results showed the logical reasoning capability of the T-S fuzzy system and the quantitative data processing capability of the neural network (NN). The comparisons between the objectively predicted results with the subjectively evaluated scores of the testing samples showed a prediction accuracy of 88.4%. The objectively predicted results calculated by the FNN model were also compared with the measurements taken using a Ground Penetrating Radar (GPR). The two groups of results showed reasonable similarity, which again indicated the effectiveness of the developed FNN method.