Propulsion motors used in ships are typically produced in small numbers, making it challenging to secure operational data for fault diagnosis in advance. Collecting data through measurements is a time-consuming and costly process, highlighting the need to obtain data using a physical model. Ensuring the accuracy of the data generated by this physical model is crucial for effective fault diagnosis. Existing physical models of motors often exhibit analysis errors in vibration data because of insufficient consideration of the ductility effect. This study proposes an improved physical model, termed the fully coupled structure-electric model, which enhances the accuracy of the data obtained from physical modeling. A direct comparison between the experimental measurement data and model data confirmed that high accuracy can be achieved for each motor state. To further investigate the fault diagnosis capabilities using data generated from the physical model, a one-dimensional convolutional neural network classifier was trained on the model data. The effectiveness of the proposed fully coupled structure-electric model is demonstrated by its ability to accurately classify actual operational data, thus confirming the validity of the model for operational data acquisition for fault diagnosis..