Real-time fault detection and diagnosis using sparse principal component analysis

Gajjar, Shriram; Külahçı, Murat; Palazoğlu, Ahmet

doi:10.1016/j.jprocont.2017.03.005

Cited by 127 publications

(54 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• The FDR in the residual space always showed a higher value than that in the principal component space, which matches the results reported in the literature where different techniques were utilized [45,46].…”

Section: Neural Network Hyperparameterssupporting

confidence: 89%

Statistical Process Monitoring of the Tennessee Eastman Process Using Parallel Autoassociative Neural Networks and a Large Dataset

Heo

Lee

2019

Processes

View full text Add to dashboard Cite

In this article, the statistical process monitoring problem of the Tennessee Eastman process is considered using deep learning techniques. This work is motivated by three limitations of the existing works for such problem. First, although deep learning has been used for process monitoring extensively, in the majority of the existing works, the neural networks were trained in a supervised manner assuming that the normal/fault labels were available. However, this is not always the case in real applications. Thus, in this work, autoassociative neural networks are used, which are trained in an unsupervised fashion. Another limitation is that the typical dataset used for the monitoring of the Tennessee Eastman process is comprised of just a small number of data samples, which can be highly limiting for deep learning. The dataset used in this work is 500-times larger than the typically-used dataset and is large enough for deep learning. Lastly, an alternative neural network architecture, which is called parallel autoassociative neural networks, is proposed to decouple the training of different principal components. The proposed architecture is expected to address the co-adaptation issue of the fully-connected autoassociative neural networks. An extensive case study is designed and performed to evaluate the effects of the following neural network settings: neural network size, type of regularization, training objective function, and training epoch. The results are compared with those obtained using linear principal component analysis, and the advantages and limitations of the parallel autoassociative neural networks are illustrated.

show abstract

Section: Neural Network Hyperparameterssupporting

confidence: 89%

Statistical Process Monitoring of the Tennessee Eastman Process Using Parallel Autoassociative Neural Networks and a Large Dataset

Heo

Lee

2019

Processes

View full text Add to dashboard Cite

show abstract

“…This paper proposes a principal component analysis and dynamic active safe semi-supervised support vector machines (PCA-DAS4VM) based fault identification method. Firstly, PCA [18,19] determines the key process variables with a sum weight of more than 80%. The selection of the weight threshold is based on experience with no specific standard in practice.…”

Section: Methodsmentioning

confidence: 99%

“…The process includes 22 continuous measured variables (Table 2) and 20 preset fault modes (Table 3). 3E Feed CMV (14) Separator underflow CMV 4A and C Feed CMV (15) Stripper level CMV (5) Recycle flow CMV (16) Stripper pressure CMV (6) Reactor feed CMV (17) Stripper underflow CMV 7Reactor pressure CMV (18) Stripper temperature CMV (8) Reactor level CMV (19) Stripper steam flow 5Recycle flow CMV (16) Stripper pressure CMV (6) Reactor feed CMV (17) Stripper underflow CMV 7Reactor pressure CMV (18) Stripper temperature CMV (8) Reactor level CMV (19) Stripper steam flow CMV (9) Reactor temperature CMV (20) Compressor work CMV (10) Purge flow CMV (21) Reactor cooling water outlet temperature CMV (11) Separator temperature CMV (22) Condenser cooling water outlet temperature Figure 4 shows that the total variance contribution rate of the first 12 principal components has reached 83.15% (more than 80%), so the first 12 principal components can reflect information about all variables. Figure 5 shows the eigenvalues of the first 12 principal components as well.…”

Section: Process Descriptionmentioning

confidence: 99%

A Dynamic Active Safe Semi-Supervised Learning Framework for Fault Identification in Labeled Expensive Chemical Processes

Jia

Tian

et al. 2020

Processes

View full text Add to dashboard Cite

A novel active semi-supervised learning framework using unlabeled data is proposed for fault identification in labeled expensive chemical processes. A principal component analysis (PCA) feature selection strategy is first given to calculate the weight of the variables. Secondly, the identification model is trained based on the obtained key process variables. Thirdly, the pseudo label confidence of identification model is dynamically optimized with an historical, current, and future pseudo label confidence mean. To increase the upper limit of the identification model that is self-learning with high entropy process data, active learning is used to identify process data and diagnosis fault causes by ontology. Finally, a PCA-dynamic active safe semi-supervised support vector machine (PCA-DAS4VM) for fault identification in labeled expensive chemical processes is built. The application in the Tennessee Eastman (TE) process shows that this hybrid technology is able to: (i) eliminate chemical process noise and redundant process variables simultaneously, (ii) combine historical pseudo label confidence with future pseudo label confidence to improve the identification accuracy of abnormal working conditions, (iii) efficiently select and diagnose high entropy unlabeled process data, and (iv) fully utilize unlabeled data to enhance the identification performance.

show abstract

“…It was noted by Box that the SPE statistic is approximately distributed as a scaled χ 2 ‐distribution with h degrees of freedom, denoted as g χ h 2 . The parameters are given by

θ_{i} = \sum_{j = A + 1}^{p} λ_{j}^{i}, for i = 1, 2; g = \frac{θ_{2}}{θ_{1}}; h = \frac{{θ_{1}}^{2}}{θ_{2}} .

Then, the control limit of SPE statistic would be taken as the (1 − α 2 ) quantile of the g χ h 2 distribution, where α 2 is the expected FAR for the SPE statistic.…”

Section: Common Spc Chartsmentioning

confidence: 99%

A model‐free approach to reduce the effect of autocorrelation on statistical process control charts

Zhang

et al. 2018

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

The presence of autocorrelation in time series data significantly affects the performance of statistical process control charts. For process data collected in real time, consecutive observations result in serial dependence that violates the assumption of sample independence. This paper focuses on the improvement of statistical process control (SPC) charts of individual observations when they are designed to monitor stationary processes in the presence of data autocorrelation. Three widely used SPC tools, Shewhart chart, Hotelling's T 2 chart, and PCA-based SPC charts, are chosen to demonstrate the proposed approach.There has been a modified version of SPC chart shown to be effective in accommodating the autocorrelation behavior, relying on a model-based estimation of chart parameters. Process model estimation, however, is a challenging task, especially for complex systems. In this paper, a model-free monitoring approach is proposed for SPC charts of individual observations, aimed at reducing the effect of autocorrelation on chart performance. It relies on a skipping strategy that creates sample bins with little or no autocorrelation. A single skipping chart and a combined skipping chart (CSC) are established, and a parameter sensitivity study is performed through simulations. In the case studies, the advantages of the CSC are noted as being a model-free approach and having a performance consistent with that of the benchmark modified chart.

show abstract

Real-time fault detection and diagnosis using sparse principal component analysis

Cited by 127 publications

References 34 publications

Statistical Process Monitoring of the Tennessee Eastman Process Using Parallel Autoassociative Neural Networks and a Large Dataset

Statistical Process Monitoring of the Tennessee Eastman Process Using Parallel Autoassociative Neural Networks and a Large Dataset

A Dynamic Active Safe Semi-Supervised Learning Framework for Fault Identification in Labeled Expensive Chemical Processes

A model‐free approach to reduce the effect of autocorrelation on statistical process control charts

Contact Info

Product

Resources

About