Abstract: The nuclear detector is a kind of special random signal converter, and it is difficult to diagnose effectively and quickly by traditional manual method when fault occurs. In this paper, an online intelligent fault detection and classification method was proposed by analyzing the characteristic of the different fault output signals of scintillator detector, and the corresponding fault models were established. Using wavelet packet algorithm and support vector machine (SVM) theory to decompose and extract eigenvectors, the fault type can be judged. Taking ST401 scintillator detector as an example, the simulation experiment was carried out. The experimental results show that the digital method based on signal waveform characteristic can diagnose scintillator detector fault quickly and effectively.