Damage localization of concrete structures based on improved BP neural network

The deterioration and damage of concrete structures pose a great threat to engineering safety. The traditional acoustic emission localization algorithm is based on the assumption of a uniform medium in concrete, which has significant errors. To accurately locate internal damage in concrete structures, this paper introduces PSO algorithm and GA algorithm to optimize BP neural network, and proposes two concrete structure damage localization models, PSO-BP and GA-BP, combined with MIC algorithm. Two models were used to locate acoustic emission sources on concrete beams. The results indicate that arrival time, channel number, rise time, counts, energy, duration, amplitude, reverberation frequency, center frequency, and peak frequency can be important parameters for locating acoustic emission sources. Compared to BP neural network, PSO-BP and GA-BP neural networks have reduced average absolute percentage errors by 4.34% and 5.93% respectively; The average absolute error decreased by 10.841 and 15.0927, and the root mean square error decreased by 12.6961 and 17.075, indicating stronger generalization ability and more accurate prediction of acoustic emission source positions. In addition, the concept of specific value error was proposed, and from the perspective of specific value error analysis, it can be concluded that both PSO-BP and GA-BP models have better positioning performance than BP models, and the GA-BP model has the best positioning effect.