Distributed Statistical Process Monitoring Based on Multiblock Canonical Correlation Analysis

Wan, Xinchun; Tong, Chudong; Luo, Lei

doi:10.1021/acs.iecr.9b04971

Cited by 11 publications

(4 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The interactions among units are complex, and the process characteristics are diverse. 18,19,21 Therefore, to overcome the complex relationships between units and variables in large-scale industrial processes and extract local features related to faults, block modeling monitoring is an effective strategy. The methods of multiblock PCA or multiblock PLS were proposed for plant-wide process monitoring.…”

Section: Introductionmentioning

confidence: 99%

“…The number of measured variables is large. The interactions among units are complex, and the process characteristics are diverse. ,, Therefore, to overcome the complex relationships between units and variables in large-scale industrial processes and extract local features related to faults, block modeling monitoring is an effective strategy. The methods of multiblock PCA or multiblock PLS were proposed for plant-wide process monitoring. − Jiang et al proposed a method based on deep learning and blocking strategies to consider the nonlinearity of the process and achieve large-scale fault detection.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Multiblock Adaptive Convolution Kernel Neural Network for Fault Diagnosis in a Large-Scale Industrial Process

Guo

Shi

Tan

et al. 2022

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

The feature extraction method plays a vital role in the fault diagnosis of large-scale processes. In this study, a novel feature extraction method named multiblock adaptive convolution kernel neural network (MBCKN) for fault diagnosis in large-scale processes is proposed. First, the large-scale process is decomposed into several key blocks. Then, several different convolutional kernels are designed in the same layer, and the attention mechanism is used to adaptively select the appropriate convolutional kernel for feature extraction based on the input in each block. Finally, the adaptive fusion method is designed to fuse the features of these blocks for fault diagnosis of the whole process. The proposed method can adaptively select suitable convolutional kernels according to different fault types, and the fusion method considers the interblock relationship under different faults, which brings improvement in diagnostic accuracy for faults with relatively similar features. Through the simulation experiment of the Tennessee Eastman (TE) chemical process, the results show that the proposed method has better performance than the Long Short-Term Memory Neural Network, the single-layer Convolution Neural Network (CNN), and the multi-layer CNN method.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Multiblock Adaptive Convolution Kernel Neural Network for Fault Diagnosis in a Large-Scale Industrial Process

Guo

Shi

Tan

et al. 2022

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

show abstract

“…Upgrading equipment and improving technology can effectively enhance process controllability. At the same time, massive data collected from the process cannot be ignored in the chemical industry, steel, semiconductor and other fields (Archibald et al, 2020;Wan et al, 2020a;Zhang et al, 2019bZhang et al, , 2020a. In addition, scholars also proposed key performance indicators to evaluate the impact of detected faults on system behavior which is important for industrial process analysis (Jiang et al, 2019b(Jiang et al, , 2020.…”

Section: Introductionmentioning

confidence: 99%

A multimode process monitoring strategy via improved variational inference Gaussian mixture model based on locality preserving projections

Guo

Liu

Shi

et al. 2021

Transactions of the Institute of Measurement and Control

View full text Add to dashboard Cite

For multimode process monitoring, accurate mode information is difficult to be obtained, and each mode is monitored separately, which increases the complexity of the system. This paper proposes a multimode process monitoring strategy via improved variational inference Gaussian mixture model based on locality preserving projections (IVIGMM-LPP). First, the raw data are projected to the feature space where samples still maintain the original neighbor structure. Second, a new discriminant condition is introduced to reduce the influence of the initial category parameter on the iteration results in the VIGMM model. Then, the data are updated utilizing modal information, so that the scales of different modes are adjusted to the same level. Next, the deviation vector is introduced to eliminate the multi-center structure of data. Finally, the statistic is built to monitor the process. IVIGMM-LPP establishes one model for monitoring the premise of knowing the mode information, which reduces the complexity of the monitoring process and improves the fault detection rate. The experimental results of a numerical case and the Tennessee Eastman (TE) process verify the effectiveness of IVIGMM-LPP.

show abstract

“…Some of data-driven FDD methods have the potential to weaken the amplification and masking effects, including the methods based on dimension reduction techniques, − sparse models, , and multiblock models. − The use of dimension reduction techniques, e.g., PCA, divides the original data space into two smaller subspaces, i.e., feature and residual subspaces. In particular, the feature subspace consists of a few key latent variables that carry most of the important data information, with each latent variable being a combination of original variables.…”

Section: Introductionmentioning

confidence: 99%

A Multigroup Fault Detection and Diagnosis Scheme for Multivariate Systems

Liu

Peng

Tong

et al. 2020

Ind. Eng. Chem. Res.

Self Cite

View full text Add to dashboard Cite

A fault in a multivariate system is usually attributed to abnormal changes of only a small part of variables. For such a fault, the fault detection index that is defined using all variables may not have a good detection performance, due to the amplification and masking effects caused by fault-free variables. To overcome this problem, this paper proposes a multigroup fault detection and diagnosis (FDD) scheme for multivariate systems. This scheme consists of two main parts: A method for the grouping of variables, and a method to use variable groups for online FDD. In the variable grouping method, the closely correlated variables are grouped together, because the close correlations among variables are proved to be advantageous to FDD. In the online FDD method, a key group to FDD is adaptively selected for every new sample, and then FDD is performed in the key group using two types of fault detection indices that take into account the intragroup and intergroup variable correlations, respectively. Because online FDD is carried out only in one variable group, the multigroup FDD scheme has two advantages. First, the fault detection capability is improved by reducing the amplification and masking effects caused by variables in other groups. Second, fault diagnosis becomes easier because the search scope of faulty variables is narrowed down to members of the key group. These two advantages are illustrated with two case studies.

show abstract

Distributed Statistical Process Monitoring Based on Multiblock Canonical Correlation Analysis

Cited by 11 publications

References 34 publications

Multiblock Adaptive Convolution Kernel Neural Network for Fault Diagnosis in a Large-Scale Industrial Process

Multiblock Adaptive Convolution Kernel Neural Network for Fault Diagnosis in a Large-Scale Industrial Process

A multimode process monitoring strategy via improved variational inference Gaussian mixture model based on locality preserving projections

A Multigroup Fault Detection and Diagnosis Scheme for Multivariate Systems

Contact Info

Product

Resources

About