The real-time monitoring technology of operation state is an important means to ensure the safe and stable operation of modern mechanical equipment. In view of the vibration signals obtained from real-time monitoring， firstly， the one-dimensional vibration signal is transformed into a two-dimensional gray image. On this basis， a new feature extraction method is proposed， and then by calculating the similarity between adjacent periodic images， a quantitative index that can describe the dynamic characteristics of the machine is established. Finally， the Laida criterion is used to determine whether the operating state has changed. Based on three typical engineering applications： speed change detection， （external） load change detection， early bearing fault detection， the proposed method is verified. The experimental results show that the proposed method achieves 100% detection accuracy in all three tested application scenarios， which reveals the effectiveness of the method.