A pre-failure feature extraction and modes classification method of board-level package subjected to vibration loading is presented for prognostics and health management of electronics using adaptive spectrum kurtosis and kernel probability distance clustering.Firstly strain response data of electronic components is filtered by empirical mode decomposition method based on maximum spectrum kurtosis

and fault symptom vector is developed by computing and reconstructing the envelope spectrum which contains potential fault information.Secondly nonlinear fault symptom data is mapped and clustered in sparse Hilbert space based on Gaussian radical basis kernel probability distance method.Several cluster centers are formed with the characterizations of the board-level package health state and various failure modes.Finally the current state of board-level package is estimated on basis of its envelope spectrum by computing its probability distance

and the forthcoming failure mode is identified before it happen.The experimental analysis demonstrate the method can recognize and predict the upcoming failure mode of board-level package effectively and serve as a new approach to achieve PHM of electronics.