The invention relates to an experimental device and an experimental method for multi-source information fusion for fault diagnosis of a hydraulic device. The device comprises the following elements: a variable plunger pump, a gear pump, two motors, two suction filters, five throttle and stop valves, two electromagnetic relief valves, an electromagnetic proportional relief valve, two pressure meters, an energy accumulator, two sets of combined sensors, a three-position four-way electromagnet directional valve, a hydraulic motor and a cooler. The experimental device can simulate eight typical running states (faults) of the hydraulic device, including low efficiency, oil and liquid pollution, air suction, leakage, oil pump fault, relief valve failure, motor failure and mechanical failure and can perform cluster analysis and fusion diagnosis to characteristics of the eight faults through a self organizing maps neural network.