The embedded coils and kernel principal component analysis (KPCA) together were proposed to detect the bearing faults. 8 detection coils were firstly embedded into the front and rear ends of the motor stator, respectively; and they form 6 groups of differential signals by timesharing multiplexing. The differential signals reflect the operation status of the motor; however, the relationship between the differential signals and motor status was nonlinear. Thus, we employ KPCA to deal with the problem of nonlinearity. The algorithm, detection statistics and steps of KPCA were investigated. Finally, an oil pump was used to carry out the experimental studies. The experiment results were presented to prove the validity and correctness of the proposed fault detection method.