| Title |
A technique for frequency converter-fed asynchronous motor vibration monitoring and fault classification, applying continuous wavelet transform and convolutional neural networks / |
| Authors |
Zimnickas, Tomas ; Vanagas, Jonas ; Dambrauskas, Karolis ; Kalvaitis, Artūras |
| DOI |
10.3390/en13143690 |
| Full Text |
|
| Is Part of |
Energies.. Basel : MDPI. 2020, vol. 13, iss. 14, art. no. 3690, p. 1-21.. ISSN 1996-1073 |
| Keywords [eng] |
convolutional neural networks ; deep networks ; continuous wavelet transform ; asynchronous motor ; vibration signals ; classification ; bearings ; short circuit ; frequency converter |
| Abstract [eng] |
In this article, a type of diagnostic tool for an asynchronous motor powered from a frequency converter is proposed. An all-purpose, effective, and simple method for asynchronous motor monitoring is used. This method includes a single vibration measuring device fixed on the motor’s housing to detect faults such as worn-out or broken bearings, shaft misalignment, defective motor support, lost phase to the stator, and short circuit in one of the phase windings in the stator. The gathered vibration data are then standardized and continuous wavelet transform (CWT) is applied for feature extraction. Using morl wavelets, the algorithm is applied to all the datasets in the research and resulting scalograms are then fed to a complex deep convolutional neural network (CNN). Training and testing are done using separate datasets. The resulting model could successfully classify all the defects at an excellent rate and even separate mechanical faults from electrical ones. The best performing model achieved 97.53% accuracy. |
| Published |
Basel : MDPI |
| Type |
Journal article |
| Language |
English |
| Publication date |
2020 |
| CC license |
|
