Fault detection and diagnosis of non-linear non-Gaussian dynamic processes using kernel dynamic independent component analysis

Fan, Jicong; Wang, Youqing

doi:10.1016/j.ins.2013.06.021

Cited by 140 publications

(61 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…, where subscript ij denotes the element of i-th row and j-th column of a matrix, operator · denotes the inner product of vectors, and k(·,·) denotes the kernel function (Fan and Wang, 2014). Kernel function allows us to compute the inner product of vectors in feature space without considering the explicit form of nonlinear mapping.…”

Section: Data Whitening In Feature Spacementioning

confidence: 99%

“…Lee et al (2007) proposed a modified KICA through combining a modified ICA with the kernel technique. Fan and Wang (2014) proposed a kernel dynamic independent component analysis (KDICA) method and a non-linear contribution plot for monitoring a non-linear non-Gaussian dynamic process. Kaneko and Funatsu (2015) proposed a new index to diagnose the process variables that contribute to process faults using a data density-based MSPC model.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Nonlinear Processes Fault Identification with Application to PCFBP

Zhai

Zhang

et al. 2018

J. Chem. Eng. Japan / JCEJ

View full text Add to dashboard Cite

This study focuses on developing a nonlinear fault identi cation approach whose goal is to nd the fault variables after a fault is detected. For nonlinear processes monitoring, it is a challenging problem to identify the fault variables since the monitoring statistics are usually an implicit function with respect to the process variables. In such a case, a traditional contribution plot-based fault identi cation approach that decomposes the monitoring statistics into the summation of variable contributions will be invalid. To solve the issue mentioned above, in this paper, a new nonlinear processes monitoring technique including fault detection and identi cation approaches is proposed based on kernel independent component analysis (KICA). On the basis of KICA, two monitoring statistics are de ned for fault detection, and the contribution of each variable to the monitoring statistics is de ned for fault variable identi cation. After a fault is detected, one can use the proposed fault variable identi cation method to judge which variable has the largest impact on the abnormal event and demonstrate identi cation. At last, the pulverized coal red boiler process (PCFBP) is taken to evaluate the validity and e ectiveness of the proposed approaches. Experiment results show that the proposed methods have satisfactory monitoring performance in the application to PCFBP.

show abstract

Section: Data Whitening In Feature Spacementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Nonlinear Processes Fault Identification with Application to PCFBP

Zhai

Zhang

et al. 2018

J. Chem. Eng. Japan / JCEJ

View full text Add to dashboard Cite

show abstract

“…Para evaluar estas potencialidades el enfoque kernel-MEWMA con dinámica reforzada fue aplicado en el problema de prueba TEP, mostrando una alta capacidad para detectar, con pocas falsas alarmas y bajos tiempos de latencia, todos los fallos considerados, independientemente de su naturaleza o magnitud. En comparación con otros enfoques kernel previamente reportados en la literatura de diagnóstico de fallos [14][15][16][17], esta propuesta demostró que es posible mejorar el desempeño del sistema de detección ante fallos de pequeña magnitud al considerar la dinámica anterior del proceso, y realizar un ajuste adecuado de las herramientas involucradas.…”

Section: Capítulo 2: Métodos Kernel Para Detectar Fallos De Pequeña Munclassified

Diagnostico de fallos en procesos industriales empleando tecnicas de aprendizaje basadas en metodos kernel

Bernal¹

2016

INTARTIF

View full text Add to dashboard Cite

Resumen: Este artículo resume las contribuciones principales de la tesis de doctorado con título: "Application of learning techniques based on kernel methods for the fault diagnosis in industrial processes". Esta tesis se centra en el análisis y diseño de sistemas de diagnóstico de fallos (DDF) basados en datos históricos. Específicamente esta tesis ofrece: (1) nuevos criterios para el ajuste de los métodos kernel que se emplean en la selección de características con alto poder discriminante para el diagnóstico de fallos, (2) una propuesta de enfoque de monitorización de procesos mediante técnicas de estadística multivariada que incorpora información reforzada sobre la dinámica de los estadísticos T 2 de Hotelling y SPE, cuya combinación con métodos kernel mejora la detección de fallos de pequeña magnitud; (3) uníndice de robustez para comparar el desempeño de herramientas de clasificación teniendo en cuenta su insensibilidad ante posibles ruidos y/o perturbaciones en los datos históricos.Palabras Clave: Diagnóstico de fallos, Métodos kernel, Robustez, Metaheurísticas.

show abstract

“…Among the methods based on MSPM, fault diagnosis methods mainly include two types: one is contribution plot method, another is fault reconstruction method. The main idea of contribution plot method is when monitoring statistics detecting a fault, these variables which have the largest contribution to the statistics are considered to be the fault variables (Dunia et al 1998； Fan et al 2014). Contribution plot method does not require prior knowledge, and is easy to be used for online fault diagnosis.…”

Section: Introductionmentioning

confidence: 99%

On - line Fault Detection and Diagnosis of Sequencing Batch Reactor Using MKICA

Zhang¹,

Wang

Gao

et al. 2015

IJEPR

View full text Add to dashboard Cite

Considering the data of Sequencing Batch Reactor (SBR) having the characteristics of non-Gaussian distribution and highly nonlinearity, this research applies Multi-way Kernel Independent Component Analysis (MKICA) to the on-line process monitoring of SBR. Meanwhile, a novel contribution analysis scheme named bar plot is developed for MKICA to diagnose faults. Above all, the three-dimensional data of SBR is expanded into two-dimensional by a new data expanding method; then, Kernel Principal Component Analysis (KPCA) is utilized to map the two-dimensional data into a high dimensional feature space, and make use of Independent Component Analysis (ICA) to extract Independent Components (ICs) in feature space; finally, if MKICA detects a fault occurs during on-line monitoring stage, the bar plot is used to identify the variables causing the fault. The method is successfully applied to an 80L lab-scale SBR. The experimental results demonstrate that, compared with traditional MICA, the proposed method exhibit better performance in fault detection and diagnose.

show abstract

Fault detection and diagnosis of non-linear non-Gaussian dynamic processes using kernel dynamic independent component analysis

Cited by 140 publications

References 38 publications

Nonlinear Processes Fault Identification with Application to PCFBP

Nonlinear Processes Fault Identification with Application to PCFBP

Diagnostico de fallos en procesos industriales empleando tecnicas de aprendizaje basadas en metodos kernel

On - line Fault Detection and Diagnosis of Sequencing Batch Reactor Using MKICA

Contact Info

Product

Resources

About