In this paper, we propose an anomaly detection method for equipment abnormalities using data measured by AreaSensing techniques. The data are gathered from the vibration of equipment installed in a wide area or large scales such as conveyor equipment and bridges. It is difficult to know which data of displacement, velocity, and acceleration is appropriate in advance because each frequency component is different. So, we apply the Hidden Markov Model, which estimates the latent state for each decomposition level by continuously frequency-resolving time series data using Wavelet method. The analysis results show that the normal and abnormally states are estimated. However, as the problem of this method, it is not possible to compare and apply the information criteria such as AIC to the Wavelet decomposed data as it is to appropriately decide which data and model parameters should be used. To overcome the defects, we propose a new evaluation function and developed a method to find a model that can stably estimate the normal and abnormal state transition, even for data separated into different frequencies. Besides, when the current measurement data contains no abnormal state, there was a problem of extracting multiple latent states that are normal but different. We focus on the difference between the state transition probabilities of the normal and unknown model. As the experimental result, the effectiveness of the proposed method has been confirmed. By using the method, it is possible to continuously diagnose abnormalities using vibration measurement data measured by AreaSensing techniques.