Have a personal or library account? Click to login
Low-Parameter Critic-Based Multivariate WGAN Model for Clogging Detection in Drives Cover

Low-Parameter Critic-Based Multivariate WGAN Model for Clogging Detection in Drives

Power Electronics and Drives
Volume 10 (2025): Issue 1 (January 2025)
By: Artur Dawid Surówka,  Marcin Kosiba,  Teemu Mikkelä,  Asko Kavala and  Marcin Firla  
Open Access
|May 2025

Figures & Tables

Figure 1.

(a) Layout of the experimental bench; (b) graphical sketch of the setup; (c) localisation of the drive’s parts that were the focus of the study: heatsink, capacitor bank and fan with the airflow indicated and (d) localisation of PT100 temperature sensors inside the drive’s assembly: (1) heatsink temperature (outlet temperature), (2, 3, 4)—IGBT base (in the heatsink under IGBT) and (5)—heatsink temperature (inlet, fan side). DUT, device under test; IGBT, insulated gate bipolar transistor.
(a) Layout of the experimental bench; (b) graphical sketch of the setup; (c) localisation of the drive’s parts that were the focus of the study: heatsink, capacitor bank and fan with the airflow indicated and (d) localisation of PT100 temperature sensors inside the drive’s assembly: (1) heatsink temperature (outlet temperature), (2, 3, 4)—IGBT base (in the heatsink under IGBT) and (5)—heatsink temperature (inlet, fan side). DUT, device under test; IGBT, insulated gate bipolar transistor.

Figure 2.

Ambient-subtracted and scaled: (a) training (upon augmentation by 1D interpolation) and testing data involving (b) inlet, (c) outlet, (d) heatsink and (e) fan clogging scenarios.
Ambient-subtracted and scaled: (a) training (upon augmentation by 1D interpolation) and testing data involving (b) inlet, (c) outlet, (d) heatsink and (e) fan clogging scenarios.

Figure 3.

Model architecture and training scheme (the generator was consistent across all tested model architectures, and the critic, varied); the critic presented in this scheme has one filter only. OBAS, out of bounds anomaly scores.
Model architecture and training scheme (the generator was consistent across all tested model architectures, and the critic, varied); the critic presented in this scheme has one filter only. OBAS, out of bounds anomaly scores.

Figure 4.

(a) Training curve and (b) tSNE plot for the generated vs. real data for the M1 WGAN model.
(a) Training curve and (b) tSNE plot for the generated vs. real data for the M1 WGAN model.

Figure 5.

Values of critic-based OBAS for the VFD system affected by (a) heatsink; (b) outlet; (c) heatsink clogging and (d) fan blockage. The time periods during which the system operates at ambient temperatures of 25 C, 35 C and 45 C are shaded in green. Anomalous operations due to clogging are highlighted with magenta boxes. The letter ‘A’ indicates ambient temperature, while ‘C’ denotes the clogging/blockage level (50% vs. 75%). OBAS, out of bounds anomaly scores; VFD, variable frequency drive.
Values of critic-based OBAS for the VFD system affected by (a) heatsink; (b) outlet; (c) heatsink clogging and (d) fan blockage. The time periods during which the system operates at ambient temperatures of 25 C, 35 C and 45 C are shaded in green. Anomalous operations due to clogging are highlighted with magenta boxes. The letter ‘A’ indicates ambient temperature, while ‘C’ denotes the clogging/blockage level (50% vs. 75%). OBAS, out of bounds anomaly scores; VFD, variable frequency drive.

Figure 6.

Analysis of the minimal clogging/blockage level detectable in the analysed VFD under different ambient conditions: (a) 25 C, (b) 35 C, and (c) 45 C. The letter ‘C’ indicates the clogging/blockage level for which the data are presented. OBAS, out of bounds anomaly scores; VFD, variable frequency drive.
Analysis of the minimal clogging/blockage level detectable in the analysed VFD under different ambient conditions: (a) 25 C, (b) 35 C, and (c) 45 C. The letter ‘C’ indicates the clogging/blockage level for which the data are presented. OBAS, out of bounds anomaly scores; VFD, variable frequency drive.

Comparison of complexities of the critic models_

ArchitectureNo. all parametersNo. of trainable parametersSize in Kb (tflite)
M126325.97
M41731976.46
M85375857.79

Outcomes of the performance assessment of the M1 model_

TypePhaseACCTPRFNRPPVF1
InletStart0.780.790.210.860.83
Cooling0.780.840.160.830.84
Active0.570.360.640.910.51
OutletStart0.650.550.450.810.65
Cooling0.670.670.320.800.73
Active0.440.140.850.820.24
HeatsinkStart0.900.980.020.880.93
Cooling0.881.000.000.840.91
Active0.530.950.690.910.46
FanStart0.871.000.000.790.86
Cooling0.841.000.000.760.86
Active0.660.320.681.000.48
DOI: https://doi.org/10.2478/pead-2025-0008 | Journal eISSN: 2543-4292 | Journal ISSN: 2451-0262
Journal RSS Feed
Language: English
Page range: 125 - 139
Submitted on: Feb 26, 2025
|
Accepted on: Apr 17, 2025
|
Published on: May 17, 2025
Published by: Wroclaw University of Science and Technology
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year
Keywords:
variable frequency drives,
generative adversarial networks,
condition monitoring,
machine learning,
anomaly detection
Related subjects:
Computer sciences,
Artificial intelligence,
Engineering,
Electrical engineering,
Electronics

© 2025 Artur Dawid Surówka, Marcin Kosiba, Teemu Mikkelä, Asko Kavala, Marcin Firla, published by Wroclaw University of Science and Technology
This work is licensed under the Creative Commons Attribution 4.0 License.

Previous articleVolume 10 (2025): Issue 1 (January 2025)Next article