CNN-based fault classification considered fault location of vibration signals

Lee, Jeong Jun; Cheong, Deok Young; Min, Tae Hong; Park, Dong Hee; Choi, Byeong Keun

doi:10.1007/s12206-023-0909-4

Cited by 2 publications

(1 citation statement)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Spectrogram data has been used frequently in the classification phase in recent years and contains detailed information that provides high classification accuracy. In studies used for rotor fault diagnosis [12,15,16,28,29], it is seen that high classification accuracy is achieved through the spectrogram image.…”

Section: Discussionmentioning

confidence: 99%

Rotor fault characterization study by considering normalization analysis, feature extraction, and a multi-class classifier

Barstuğan,

Arabacı

2024

Eng. Res. Express

View full text Add to dashboard Cite

Background:Rotor faults are the most common malfunctions encountered, especially during the manufacturing stage, in asynchronous motors. These faults cause vibration in the motor torque and a decrease in efficiency. In recent years, the detection of rotor faults has been done using motor current. The reflection of rotor faults on motor current depends on slip, and therefore, the effect increases as the current grows. Good results are achieved in fault detection at nominal loads. However, especially when motor manufacturers are considered, testing the motor by loading it requires expensive testing equipment and long-term test procedures. Therefore, the detection of faults in the motor at no load is emphasized. However, since the effect of the fault decreases when the motor is at no load, fault detection becomes difficult. Generally, small-level faults cannot be detected.Objective:This study focuses on fault detection from the motor current at no load. The development current at no load was used to eliminate the negative effects of slip. However, since the slip is not constant, the change in frequency and amplitude values to be used as a feature makes the diagnosis difficult.Method:In this study, the spectrogram was used to evaluate the change during the start-up time. Thus, a standard dataset was determined for comparison. The texture properties of the spectrogram image were extracted using various methods. The extracted features were subjected to normalization analysis and classified using the k-NN algorithm.Results:In the classification phase, a classification accuracy of 98.66% was achieved using the k-NN method, and it was seen that the proposed method could be used for the detection of rotor faults.Conclusions:The study has successfully demonstrated that broken rotor bar faults in asynchronous motors can be diagnosed using the motor start-up data.

show abstract