Malfunction prediction is a trend of NC machine tool malfunction diagnosis development and the diagnosis accuracy is heavily dependent on the real and online acquisition of malfunction parameters. Gear malfunction is one of main mechanical malfunctions, and it is very meaningful to forecast its malfunction. The shock vibration of gear malfunction is non-stationary, so it should be processed by time frequency algorithms. Kalman filtering and Laplace wavelet are time frequency algorithms. Klaman filtering is self-adapting algorithm, and can filter noise in real-time. Laplace wavelet can obtain malfunction parameters by correlation filtering when correlation parameter k is the maximum. The proposed technique features two appealing advantages, which include self-adapting Kalman filter-based time-frequency algorithm and a Laplace wavelet-based parameters extraction. A set of simulating gear vibration data was used for verification. It provides a quantitative and more efficient means for obtaining the malfunction parameters to malfunction forecasting system.