Abstract: Modulus inversion based on FWD bending data has been widely applied in asphalt pavement condition detection. However, its accuracy is notably inferior to that of direct structural modulus inversion. Python was utilized to parametrically model finite element software and establish a database of bending basins. A strategy for targeted adjustment of structural modulus is proposed, leveraging the high accuracy of positive deflection data of structural modulus. Utilizing the deflection database, a neural network is employed to directly forecast bending values under varying modulus combinations. The structural modulus is then reversed by predicting the root-mean-square error of the bending value. Results indicate that soil foundation modulus error is less than 0.1%. Through iterative adjustment of the initial modulus based on the root-mean-square error(RMSE) of predicted bending values, the modulus error of surface and base can be reduced to less than 1%, achieving a favorable inverse calculation effect. This method presents a novel approach to inverse modulus calculation based on FWD in the domain of asphalt pavement damage detection.