A statistical time-frequency spectrum-driven method for damage prediction of high-speed automata is proposed， and then the faults caused by transient shocks can be characterized with changes in time-frequency distributions of the response signals. Firstly， transient shock signals are collected from the performance tracking test of high-speed automata. Secondly， instantaneous frequency features of the signals are calculated to reflect the evolution of frequencies with time. Thirdly， canonical variate analysis model is built on those features to extract the maximal correlation information， such that the information ambiguity among the transient shock signals can be largely reduced. Finally， the threshold for healthy states of high-speed automata is determined with the kernel density estimation， for the aim of monitoring the transient impulse response signals. Different defective conditions of a 12.7 mm automaton are monitored with the proposed work， experimental results confirm the efficiency of the suggested work in detecting transient shock damage.