The invention discloses a fault detection method applied to helicopter drive system vibration signals. The fault detection method comprises a first step of carrying out noise reduction processing for an acquired helicopter drive system vibration signal to obtain a signal x(t); a second step of decomposing the signal x(t) by using an EMD algorithm to obtain an intrinsic mode function component; a third step of solving for an approximate entropy of the intrinsic mode function component to obtain an approximate entropy value of the intrinsic mode function component; a fourth step of subjecting the approximate entropy value of the intrinsic mode function component to scalar quantization processing, and finally obtaining an approximate entropy value of scalar quantization; a fifth step of bringing the approximate entropy value of scalar quantization into a fuzzy membership function and solving for a membership function value thereof; and a sixth step of obtaining a degree of membership of a certain fault by solving for the maximum membership function value, and finally determining a fault detection result. The invention can be used for improving the safety performance of the helicopter drive system.