To describe the dynamic evolutionary law and tribological behavior of the tribopair AISI 52100-AISI 1045, rotational experiments were conducted by sliding a disk against a static pin. The multidimensional phase spaces were reconstructed based on the scalar time-series by the time-delay embedding technique, and the multivariate graph-based method was used to visualize the overall picture of the phase space. The evolution of radar plots and the corresponding multivariate graph centrobaric trajectory (MGCT) is consistent with the description of “running-in, steady-state and increasing friction stages,” and can serve as effective indicators for the friction state transitions. Results show that the radar plot can inform quantitative interpretations of friction process identification. Therefore, the multivariate graph-based method is a useful approach to characterize the nonlinear dynamics of tribological behaviors.