Existing fuel pump test platform can only complete the simple functional test and exposes its shortcoming in low fault detection rate under multiple failure modes. In the light of this problem, a fuel pump fault diagnosis experimental device and test program are designed and optimized. The device is used to carry out experiments for seven typical fuel pump failures and accurately capture real-time vibration and outlet pressure signals. Fault features are extracted from the collected signals and different fault feature vectors are built according to fault feature. The sensor layout is optimized by comparing the diagnostic effect of different sensor signals. Experiments show that the device is capable of providing a reliable date for fuel pump failure diagnosis. The detection system is improved by reducing its volume when only using a vibration sensor and pressure sensor.