Semi-supervised denoising autoencoder with multiple consistency regularization for process fault classification
Process fault classification constitutes a critical component for ensuring efficient and stable industrial operations. If the collected process data is severely affected by noise or contains a lot of information unrelated to the fault, it will affect the performance of the classification model. To a...
Uloženo v:
| Vydáno v: | Engineering Research Express Ročník 7; číslo 3; s. 35402 - 35420 |
|---|---|
| Hlavní autoři: | , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
IOP Publishing
30.09.2025
|
| Témata: | |
| ISSN: | 2631-8695, 2631-8695 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | Process fault classification constitutes a critical component for ensuring efficient and stable industrial operations. If the collected process data is severely affected by noise or contains a lot of information unrelated to the fault, it will affect the performance of the classification model. To address these issues, this paper proposes a semi-supervised denoising autoencoder method with multi-consistency regularization (MCR-SSDAE). Based on the supervised autoencoder, this paper proposes to add unlabeled inputs and incorporate three intensities of interference into both labeled and unlabeled inputs to overcome the influence of noise on process data and improve the robustness of the model. Labeled data are used for pre-training the model to reduce the impact of irrelevant information. The pseudo-labels of unlabeled data are predicted by using the pre-trained model, and their validity is judged by setting a threshold. The training data is expanded and used for model adjustment to solve the problem of insufficient labeled data and achieve semi-supervised training of the model. In the adjustment process, the pseudo-label consistency and feature consistency under triple interference are proposed to construct a multiple consistency regularization loss function, which can effectively use the information of unlabeled data to improve the prediction ability of the model. The effectiveness of the proposed method is verified in the Tennessee-Eastman (TE) process and the three-phase flow process. The experimental results show that this method can achieve good results with a small amount of labeled data. |
|---|---|
| Bibliografie: | ERX-109057.R1 |
| ISSN: | 2631-8695 2631-8695 |
| DOI: | 10.1088/2631-8695/addd5f |