In order to predict the failure of Darlington transistor, a method for fault prognostics based on KPCA and Mahalanobis distance is proposed. Through the failure mechanism analysis and accelerated degradation testing of Darlington transistor, the degradation data of collector current and saturation voltage are obtained. The paper utilizes wavelet packet decomposition and KPCA to process the degradation data and filter out interference signals, obtaining the principal component of the degradation. The Mahalanobis distance is used to fuse these components into health index. And the health index could represent the healthy status of Darlington transistor in changes. Finally, two fault predict algorithms are used to predict the HI. And the availability is proved by the forecasting. The results show that RMS between the predicted value and the true value is within 10%.