Note: This dataset is provided as supplementary material for a manuscript currently under peer review. This dataset contains tri-axial vibration signals of a permanent magnet synchronous motor (PMSM) collected under various sensor locations and physically induced disturbance conditions. It is designed to evaluate the cross-location generalization and false alarm suppression capabilities of fault diagnosis models. [Experimental Setup] Motor: Frameless inrunner PMSM (CubeMars RI60 KV120) Operating Condition: 500 RPM, No-load condition Sensor: LSM6DS3TR-C (STMicroelectronics) tri-axial MEMS accelerometer Sampling Rate: 5 kHz Diagnostic Classes (5 states): Normal (NR), Bearing Fault (BF), Magnet Fault (MF), Stator Fault (SF), and Eccentricity Fault (EF). Recording Duration: 10 seconds per recording. Trials: 5 independent recordings were acquired for each diagnostic class under the specified conditions. Temporal Variability: To introduce realistic distribution shifts, data acquisition for each sensor location (Loc S, Loc A, Loc B, and Loc C) was conducted on different days (separated time intervals). This dataset contains permanent magnet synchronous motor (PMSM) vibration signals collected under various sensor locations and physically induced disturbance conditions. The dataset is structured into six subsets: [Dataset Structure] LocS_Train_Clean: Data measured at the reference location (Loc S, 12 o'clock) under clean conditions without any external physical disturbances. LocS_Train_Noise: Data measured at Loc S (12 o'clock) while applying five types of physically induced disturbances (D1-D5: vibration motor contact and hammer strikes). LocS_Val_Clean: An independent, held-out dataset measured under the same clean conditions at Loc S, strictly separated from the training set. LocC_Val_Test_Noise: Data measured after relocating the sensor to a new evaluation location (Loc C, 2 o'clock) with disturbances (D1-D5) applied. This subset was used for validation during model selection and also included in the final evaluation, so it should be interpreted as a validation-exposed location rather than a strictly unseen test location. LocA_Test_Noise: Data measured at an entirely new location (Loc A, 9 o'clock) with disturbances (D1-D5) applied. This subset represents a strictly unseen test location. LocB_Test_Noise: Data measured at another entirely new location (Loc B, 11 o'clock) with disturbances (D1-D5) applied. This subset also represents a strictly unseen test location.