The vibration caused by mechanical manufacturing will lead to unpredictable changes in product quality, which will increase the manufacturing cost. Plenty of research is imposed to establish a normal vibration coefficient or to develop an accurate and efficient production process. Therefore, various health diagnosis methods and feedback mechanisms are obtained, such as dynamic feature Detection, machine tool state Detection, cutting chatter analysis, health state feedback of specific parts in the machine tool, and so on. This study analyzes the lathe and establishes the vibration analysis and health diagnosis method used for NC lathe or traditional machine tools through the vibration signals generated by different clamping states. Before the spindle speed reaches 2000 rpm, 5 experiments at an interval of 250- rpm and three clamping states are executed. Moreover, the vibration signal is obtained using the intelligent prediction and diagnosis performance system, analyzed in the frequency domain, matched with the root mean square result, and checked for the accuracy of the vibration signal data and vibration eigenvalues. Then, digital tools are used to filter the signal according to the filtered outliers and the matching results of feature extraction. Principal component analysis (PCA) and the projection matrix are used to evaluate 37 features and reduce the dimension of the data, to obtain the vibration data distribution map under different rotating speeds of each clamping state. It can identify the distribution range and tightness of the distribution map under each clamping state. In the above experiment, the signal data of each 250-rpm interval and three clamping states were collected and compared. The results show that it is feasible to use the PCA method to determine the change of vibration value in the CNC lathe and establish a health state feedback data set based on the vibration change mechanism.