The invention discloses a momentum wheel fault detection method based on the relationship between a voltage and angular momentum. The momentum wheel fault detection method comprises the steps that firstly based on telemetry data of the angular momentum of a momentum wheel, segmentation is performed with different variable quantity of the angular momentum, then an average value of input voltage values in a time interval corresponding to each angular momentum variable segment is calculated, the average voltage values and the angular momentum change time in all time periods are formed to calculate inverse proportional coefficients, then a mean value and a standard deviation of the inverse proportional coefficient in each time period are calculated, an abnormal threshold value is calculated, according to the process, the telemetry data requiring fault detection is processed to obtain an inverse proportional coefficient k' of the average control input voltage value on the current segment and the angular momentum change time interval, if k' is greater than the abnormal threshold value, it is indicated that the momentum wheel is faulty, and if k' is less than or equal to the abnormal threshold value, it is indicated that the momentum wheel is not faulty. The method utilizes the statistical characteristics of extracted feature quantity to realize the detection of micro-faults and the early warning of the faults of the momentum wheel.