A multi-scale low rank convolutional autoencoder for process monitoring of nonlinear uncertain systems

In modern industrial process monitoring, due to equipment performance degradation and equipment working environment, process variable measurement can lead to uncertainty in measurement data. Traditional process monitoring methods based on uncertain data typically assume that variables have the same...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Process safety and environmental protection Jg. 188; S. 53 - 63
Hauptverfasser: Yin, Jiawei, Yan, Xuefeng
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.08.2024
Schlagworte:
1D convolutional neural network
Low rank
Multi scale
Process monitoring
Uncertainty process
Process monitoring
Uncertainty process
1D convolutional neural network
Low rank
Multi scale
ISSN:0957-5820, 1744-3598
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In modern industrial process monitoring, due to equipment performance degradation and equipment working environment, process variable measurement can lead to uncertainty in measurement data. Traditional process monitoring methods based on uncertain data typically assume that variables have the same level of uncertainty. However, factors such as the lifespan of different devices and different working environments result in varying levels of uncertainty in variables. To monitor such processes, a multi scale low-rank convolutional autoencoder (MLRCAE) for process monitoring based on uncertain measurement data is proposed. First, to extract robust multi scale features from uncertain input, a multi scale convolution (MSC) module is designed to reduce the impact of different levels of uncertainty on the model. Second, a low-rank constraint (LRC) loss function is used to prevent models from overfitting uncertain data by punishing the rank of hidden layer robust features. In conclusion, we apply this method to numerical simulation, specifically within the Tennessee Eastman process, and wastewater treatment plants to confirm the model’s efficacy and compare it with other advanced methods. The results show that MLRCAE not only reduces the impact of uncertain data, but also maintains stable performance of the model under different levels of uncertainty.
ISSN:0957-5820
1744-3598
DOI:10.1016/j.psep.2024.05.070