Abstract
Quality-relevant fault detection aims to reveal whether quality variables are affected when a fault is detected. For current industrial processes, kernel-based methods focus on the nonlinearity within process variables, which is insufficient for obtaining nonlinearities of quality variables. Alternatively, neural network is an option for nonlinear prediction. However, these models are driven by predictive errors on samples. For quality-relevant tasks, the key is to capture the trends of quality variables. Therefore, this study proposes a new model, namely, maximizing correlation neural network (MCNN), to predict the quality-relevant information intuitively. The MCNN is trained to maximize the linear correlation between quality variables and the combinations of nonlinear representations mapped by a multilayer feedforward network. As such, fault detection can be implemented in the quality-relevant and irrelevant subspaces on the basis of the deep most correlated representations of process variables. Considering that different variables have different sensitivities to quality at various locations due to their nonlinear relationship, fault backpropagation is designed in the MCNN to isolate the faulty variables on the basis of real-time faulty information. Finally, numerical example and Tennessee Eastman process are used to evaluate the proposed method, which exhibits a competitive performance.
Similar content being viewed by others
References
Qin SJ (2012) Survey on data-driven industrial process monitoring and diagnosis. Annu Rev Control 36(2):220–234
Ge Z, Song Z, Gao F (2013) Review of recent research on data-based process monitoring. Ind Eng Chem Res 52(10):3543–3562
Wang Y, Si Y, Huang B, Lou Z (2018) Survey on the theoretical research and engineering applications of multivariate statistics process monitoring algorithms: 2008–2017. Can J Chem Eng 96(10):2073–2085
Yin S, Li X, Gao H, Kaynak O (2015) Data-based techniques focused on modern industry: an overview. IEEE Trans Ind Electron 62(1):657–667
Zhang K, Hao H, Chen Z, Ding SX, Peng K (2015) A comparison and evaluation of key performance indicator-based multivariate statistics process monitoring approaches. J Process Contr 33:112–126
Yan S, Huang J, Yan X (2019) Monitoring of quality-relevant and quality-irrelevant blocks with characteristic-similar variables based on self-organizing map and kernel approaches. J Process Contr 73:103–112
Zhou D, Li G, Qin SJ (2010) Total projection to latent structures for process monitoring. AIChE J 56(1):168–178
Qin SJ, Zheng Y (2013) Quality-relevant and process-relevant fault monitoring with concurrent projection to latent structures. AIChE J 59(2):496–504
Wang G, Luo H, Peng K (2016) Quality-related fault detection using linear and nonlinear principal component regression. J Frankl Inst 353(10):2159–2177
Huang J, Yan X (2017) Quality relevant and independent two block monitoring based on mutual information and KPCA. IEEE Trans Ind Electron 64(8):6518–6527
Zhou J, Ren Y, Wang J (2018) Quality-relevant fault monitoring based on locally linear embedding orthogonal projection to latent structure. Ind Eng Chem Res 58(3):1262–1272
Peng K, Zhang K, Li G (2013) Quality-related process monitoring based on total kernel PLS model and its industrial application. Math Prob Eng. https://doi.org/10.1155/2013/707953
Jiao J, Zhao N, Wang G, Yin S (2017) A nonlinear quality-related fault detection approach based on modified kernel partial least squares. ISA Trans 66:275–283
Wang G, Jiao J (2017) A kernel least squares based approach for nonlinear quality-related fault detection. IEEE Trans Ind Electron 64(4):3195–3204
Deng X, Tian X, Chen S, Harris CJ (2019) Deep principal component analysis based on layerwise feature extraction and its application to nonlinear process monitoring. IEEE Trans Contr Syst Tech 27(6):2526–2540
Jiang Q, Yan S, Yan X, Chen S, Sun J (2020) Data-driven individual–joint learning framework for nonlinear process monitoring. Contr Eng Pract. https://doi.org/10.1016/j.conengprac
Yuan X, Huang B, Wang Y, Yang C, Gui W (2018) Deep Learning-based feature representation and its application for soft sensor modeling with variable-wise weighted SAE. IEEE Trans Ind Informat 14(7):3235–3243
Dong J, Sun R, Peng K, Shi Z, Ma L (2019) Quality monitoring and root cause diagnosis for industrial processes based on Lasso-SAE-CCA. IEEE Access 7:90230–90242
Li G, Qin SJ, Ji Y, Zhou D (2009) Total PLS based contribution plots for fault diagnosis. Acta Autom Sinica 35(6):759–765
Lv Z, Yan X, Jiang Q (2016) Batch process monitoring based on multiple-phase online sorting principal component analysis. ISA Trans 64:342–352
Downs JJ, Vogel EF (1993) A plant-wide industrial process control problem. Comput Chem Eng 17(3):245–255
Ricker NL, Lee JH (1995) Nonlinear model predictive control of the Tennessee Eastman challenge process. Comput Chem Eng 19(9):961–981
Acknowledgements
This work was supported by the National key research and development program of China (2020YFA0908303) and National Natural Science Foundation of China (21878081).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix 1: Derivation of MCNN gradient
Appendix 1: Derivation of MCNN gradient
In order to perform the mini-batch back propagation algorithm to train MCNN, gradient of the objective \(\frac{\partial \rho }{{\partial {{\varvec{\uptheta}}}_{k} }}\) is needed, where \({{\varvec{\uptheta}}}_{k}\) is one of the trainable parameters of MCNN. Based on the chain rule, we can get that
Since \(\rho = \left( {{\mathbf{K}}^{T} {\mathbf{K}}} \right)^{ - 1/2} = \left[ {\left( {{\mathbf{S}}_{xx}^{ - 1/2} {\mathbf{S}}_{xy} } \right)^{T} \left( {{\mathbf{S}}_{xx}^{ - 1/2} {\mathbf{S}}_{xy} } \right)} \right]^{ - 1/2},\) we firstly show that
Proof of (32) and (33) is illustrated as follows.
Next, \(\frac{{\partial \left( {S_{xx} } \right)_{ab} }}{{\partial {\mathbf{F}}_{ij} }}\) and \(\frac{{\partial \left( {S_{xy} } \right)_{a} }}{{\partial {\mathbf{F}}_{ij} }}\) are computed as follows.
where \({\mathbf{J}}^{ij}\) is the single-entry matrix, 1 at \(\left( {i,j} \right)\) and zero elsewhere, and \({\mathbf{A}} = \left[ {{\mathbf{I}} - \left( {1/n} \right){\mathbf{1}}_{n} {\mathbf{1}}_{n}^{T} } \right]^{T} \left[ {{\mathbf{I}} - \left( {1/n} \right){\mathbf{1}}_{n} {\mathbf{1}}_{n}^{T} } \right],\;{\mathbf{B}} = \left[ {{\mathbf{I}} - \left( {1/n} \right){\mathbf{1}}_{n} {\mathbf{1}}_{n}^{T} } \right]^{T} {\mathbf{y}}\)
Since \(\frac{{\partial {\mathbf{F}}_{ij} }}{{\partial {{\varvec{\uptheta}}}_{i} }}\) can be derived in a traditional way in traditional NNs, it is not repeated here and thus the gradient of MCNN is complete for the backpropagation training.
Rights and permissions
About this article
Cite this article
Yan, S., Yan, X. Nonlinear quality-relevant process monitoring based on maximizing correlation neural network. Neural Comput & Applic 33, 10129–10139 (2021). https://doi.org/10.1007/s00521-021-05776-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-021-05776-3