Evaluation and characteristic analysis of compressive strength and resistivity of EG cement conductive mortar based upon hybrid-BP neural network

Abstract

A concise general equation is difficult to reflect the relationship between physical properties of conductive materials and properties of conductive mortar (CM). To handle this issue, a back propagation neural network (BPNN) model with a topology of 4–10-2 was designed, which was further optimized with the Ant colony optimization (ACO), Genetic algorithm (GA) and Particle swarm optimization (PSO) algorithms. The models intended to establish non-linear relationships among the input layer parameters, viz., expanded graphite (EG), carbon fiber (CF), slag powder (SP), and fly ash (FA) contents, and the output layer parameters including the CM’s compressive strength and resistivity. A total of 112 groups of data points were used for model development, including 80 groups of training set data, and two series of 16 data groups for verification and testing. Influence of conductive materials on compressive strength and resistivity of CM was determined with a correlation analysis. Meanwhile, the importance and total scores of different proportions were analyzed by performing a principal component analysis on the dataset. It was found that both the EG and FA content were correlated negatively with the CM’s compressive strength and resistivity. In addition, the CF and SP contents were correlated positively with the CM’s compressive strength but negatively with resistivity. The order of the conductive substances’ importance in CM’s compressive strength and resistivity was EG > CF > SP > FA, and the group B-3 was determined as the best mix ratio. The PSO-BPNN model had the highest accuracy in predicting both the CM’s compressive strength and resistivity, which can provide guidance for mix design of CM.