An Integrated Algorithm for Motor Anomaly Detection Based on t-SNE

Abstract

The invention provides a T-SNE-based integrated algorithm for motor abnormality detection, which belongs to the field of motor abnormality detection. A T-SNE-based motor abnormality detection integration algorithm of the present invention aims at the vibration data of rotating equipment, only uses the normal samples of the equipment itself, performs dimensionality reduction on the samples through a method based on T-SNE, and then uses the method of outlier value detection Carry out multi-dimensional integrated anomaly detection, thus establishing a commercial anomaly detection method that can be used stably. Compared with other inventions, the present invention only needs the normal data of the equipment itself to model during training, and does not need to deliberately collect abnormal samples. Moreover, the present invention does not need to extract different key features or set different thresholds for different data, and only needs to collect a small amount of normal data of the equipment for analysis, so as to achieve a high accuracy rate for abnormal detection of all equipment.

Description

technical field [0001] The invention relates to a T-SNE-based motor abnormality detection integrated algorithm, which belongs to the field of motor abnormality detection. Background technique [0002] The predictive maintenance of the motor is an emerging maintenance method. It can predict the future development trend of the motor state according to the historical state of the motor, so that the maintenance personnel have enough time to deal with the upcoming problems and reduce the loss caused by the failure of the motor. . [0003] At present, the state monitoring of the motor can rely on monitoring signals such as current, voltage, magnetic field, vibration, temperature, and sound, and establish rules or models based on the previously collected state signals. One technique is called anomaly detection. [0004] The patent "CN109236587B is an alarm system for detecting abnormal operation of wind turbines", which discloses a fault monitoring method based on sound signals. ...

AI Technical Summary

Problems solved by technology

The disadvantage of the former is that the accuracy rate is strongly correlated with the extracted features and the selected threshold, and the stability of generalization is often not enough.

For the latter, the generalization achieved by the transfer learning cannot meet the commercial needs because the small samples learned are different or the large samples based on it are too different from the industrial data, and the training requires two positive and abnormal samples.

Images