Abstract: Polygonal wheel is a common wear phenomenon in railway vehicles. With the increase in operating mileage, the degree of wheel wear increases significantly, which seriously affects the ride comfort and operation safety of the trains. It is of great significance to develop the polygonal wheel dynamic detection method with the help of monitoring big data. In this study, a polygonal wheel quantitative identification model is established based on the vertical acceleration signals of the axle box. Firstly, the main orders of the polygonal wheels contained in the acceleration signals of the axle box are identified through order analysis, and the acceleration amplitude corresponding to the main order is obtained at the same time. On this basis, the characteristic matrix of polygonal wheel wear amplitude identification is constructed by the acceleration amplitude of the main order and the entropy characteristic of the acceleration signal. Then the genetic mutation particle swarm optimized multiple kernel extreme learning machine (GMPSO-MKELM) identification model is established. Through the mapping between the characteristic matrix and the wear amplitude of the polygonal wheel, the wear amplitude identification is realized. The results of the simulation and field test show that the proposed model can effectively extract the main order of polygonal wheels from axle box acceleration. The identification accuracy of the proposed model is better than that of the comparison models in wear amplitude, and it has high detection efficiency. The proposed model can achieve accurate identification with a root mean square error of 0.0010 (simulation) and 0.0134 (field test). The proposed polygonal wheel identification model can provide a theoretical basis for the detection and intelligent maintenance of railway vehicle wheels.