Considering the nonlinear friction damping of the model and the external disturbances of the main drive system during the rolling process， a mathematical model of the main drive system of the strip mill is established. For this system， a nonlinear unknown input observer （UIO） is designed and used for fault detection and fault reconstruction of the rolling mill main drive system. In order to enhance the sensitivity of residuals to the fault signal and improve the fault detection accuracy of the observer， an unknown input observer is constructed to decouple external disturbances from the residuals. Meanwhile， the H∞ performance index is used to improve the robustness of the observer to fault reconstruction， and the convergence analysis of the observation error dynamic system is given by Lyapunov function. To improve the observer design process， the problem of solving the gain matrix is transformed into an optimization problem constrained by linear matrix inequality （LMI）. The residual error is compared with the set threshold to realize fault detection and complete fault reconstruction. Through the simulation research on the main drive system of F4 frame of 2 030 mm cold rolling mill， it is verified that the observer can accurately track the system state detect and estimate the faults of the main drive system.