The online monitoring of fixed sheave block in a drilling rig hoisting system suffers from the limited installation space, coupled reeving ratios that make fault localization difficult and speed fluctuations that can hide weak wear features. Correspondingly this study proposes a machine-vision speed extraction based fault diagnosis method. The constant-density stripes are preset on the sheave rim and an area-scan camera captures an image sequence of stripes during rotation, which converts the sheave rotation into a grayscale-variation signal. The cross-correlation analysis and time-frequency analysis are then used to extract the time-frequency ridges of the cross-correlation coefficient, which enables the synchronous extraction of multiple sheaves’ instantaneous speed. In order to mitigate the effect of overall speed fluctuations under the variable-speed operation, a relative deviation matrix of speed ratios is used as the diagnostic feature. This matrix cancels global disturbances and highlights abnormal speed deviations caused by the rope-groove wear, so the worn sheave can be identified and located. Comparative experiments are carried out on a fault-simulation test bench for the drilling rig hoisting system. Results show that under the constant motor speed (up to 2 800 rpm), time-varying motor speed (1 600-2 800-1 600 rpm), and different loads (40 kg and 80 kg), the method achieves an average accuracy of 86.11%±2.90% and an average F1-score of 86.05%±2.90%. These results show the stable performance under different tests. Overall the online method provides a non-contact, non-sensitive to speed changes solution without relying on the model training, which supports the synchronous monitoring of multiple sheaves for wear identification and provides data support for the later condition assessment and maintenance decisions.