Objective To study the morphology and fractal characterization of UHMWPE wear particles by simulation experiment on knee joint, analyze the classification characteristics of wear particles and discuss the correlation between fractal dimension and wear state. Methods The knee joint simulator was used to realize the knee joint wear motion. The forged CoCrMo alloy and UHMWPE were selected as artificial joint prosthesis materials. The wear particle extraction was based on ISO 17853. The automatic extraction and fractal identification system of wear particles were used to investigate fractal characteristics of wear particles. Three kinds of models were established to classify and recognize wear particles, by using the meshing method, cluster analysis and genetic simulated annealing algorithm, respectively. Results The fractal characteristics of UHMWPE single wear particles was very obvious. The fractal dimension calculated by radar fractal method decreased, during the transition from the larger size of strip debris to the smaller size of roundness debris, and the fractal dimension D of spherical debris was close to zero. The weighted sum of squared error values of fractal dimension for wear particle population classification was the smallest by the model of genetic simulated annealing algorithm, and the clustering feature was very obvious. When the wear cycles were low, the large fractal dimension of strip, needle and fibrous abrasive debris with larger fractal dimension had the largest proportion, and the main abrasion modes were the ploughing and spalling wears. With the extension of wear period, the proportion of wear particles with large fractal dimension decreased, and the proportion of flaky, blocky and near-spherical wear particles with low fractal dimension increased clearly. The wear mechanism changed to the fatigue and adhesive wear, and the wear state transited to the composite wear period. During the stable wear period, the proportion of all kinds of abrasive grains changed little. Due to the increase in the number of small particles, the fractal dimension decreased in stable wear state. Conclusions Based on the improved radar graph method, the fractal dimension of wear particles with different profiles could be obtained by automatic extraction and fractal identification system of wear particles. The research findings can be used in shape extraction, fractal dimension calculation and parameter statistics, as well as providing a new digital analysis tool for identification and diagnosis for wear particles of artificial prosthesis.