Fault detection and diagnosis of chemical process using enhanced KECA

et al. 2020

IEEE Access

Control charts are significant diagnostic tools to detect and identify the quality fluctuation of the complex industrial process. In the practical production process, attention is being paid to the monitoring of mixture control charts, which usually coupled by two or more basic control charts modes. This research is to present a hybrid pattern recognition method for mixture control charts. The proposed method mainly covers the feature fusion extraction (FFE) and kernel extreme learning machine (KELM) with modified grey wolf optimizer (MGWO). The FFE module applies the original data and their shape and statistical features as the features, then uses kernel entropy component analysis to reduce the feature dimension and extract valid features. One significant difficulty of KELM is to get suitable parameters like the penalty parameter and the kernel function parameter value. MGWO is established to the optimal tuning of KELM parameters, which improves the population initialization and nonlinear convergence factor of traditional grey wolf optimizer. The proposed methodology is promising to obtain a better classification recognition rate, less computational time and achieves more stable results in the pattern recognition problem of mixture control charts. INDEX TERMS Control chart patterns recognition, feature fusion extraction, kernel extreme learning machine, kernel entropy component analysis, modified grey wolf optimizer.

Section: Keca Feature Reduction Algorithmmentioning

confidence: 99%

Features Fusion Exaction and KELM With Modified Grey Wolf Optimizer for Mixture Control Chart Patterns Recognition

Zhang

et al. 2020

IEEE Access

“…Consider a dataset

with the probability density function

. The Renyi quadratic entropy is defined as [ 28 ]:

and the Parzen window density estimator is described as:

where

is a Mercer kernel function, and σ is the parameter of the kernel function, or Parzen window. Using the sample mean approximation of the expectation operator, we can express the Renyi quadratic entropy as follows:

which can be re-expressed as follows:

where K is the kernel matrix and I is an ( n × 1) vector where each element equals one.…”

Section: Kernel Entropy Component Analysismentioning

confidence: 99%

State Clustering of the Hot Strip Rolling Process via Kernel Entropy Component Analysis and Weighted Cosine Distance

2019

Entropy

In the hot strip rolling process, many process parameters are related to the quality of the final products. Sometimes, the process parameters corresponding to different steel grades are close to, or even overlap, each other. In reality, locating overlap regions and detecting products with abnormal quality are crucial, yet challenging. To address this challenge, in this work, a novel method named kernel entropy component analysis (KECA)-weighted cosine distance is introduced for fault detection and overlap region locating. First, KECA is used to cluster the training samples of multiple steel grades, and the samples with incorrect classes are seen as the boundary of the sample distribution. Next, the concepts of recursive-based regional center and weighted cosine distance are introduced. For each steel grade, the regional center and the weight coefficients are determined. Finally, the weighted cosine distance between the testing sample and the regional center is chosen as the index to judge abnormal batches. The samples in the overlap region of multiple steel grades need to be focused on in the real production process, which is conducive to quality grade and combined production. The weighted cosine distances between the testing sample and different regional centers are used to locate the overlap region. A dataset from a hot steel rolling process is used to evaluate the performance of the proposed methods.

“…The FDD framework using binary classifiers is applied to detect and diagnose faults in one step, such as adaptive neuro fuzzy inference system (ANFIS) and kernel entropy component analysis (KECA) . The number of binary classifiers to be constructed is equal to the number of faults in the process.…”

Section: Introductionmentioning

confidence: 99%

Amplitude‐frequency images‐based ConvNet: Applications of fault detection and diagnosis in chemical processes

Zhang

Journal of Chemometrics

Gao

et al. 2019

Self Cite

Fault detection and diagnosis (FDD) have been major concerns in abnormal event management of chemical processes for decades. Frequency-wise variations in chemical processes are not considered in most traditional methods, which affects the monitoring performance. An amplitude-frequency imagesbased convolutional neural network (ConvNet) is proposed for FDD in chemical processes. The fast Fourier transform (FFT) is first performed on data slice collected within a period to extract both amplitude-wise dynamics and frequency-wise variations, with the results in images. Then, the amplitudefrequency images are fed into ConvNet for FDD. ConvNet is applied as a binary classifier, in which each classifier corresponds to only one fault. Thus, an expandable framework is provided to incorporate a new fault. The performance of the proposed amplitude-frequency images-based ConvNet in FDD is demonstrated in a numerical case and the Tennessee Eastman process. KEYWORDS amplitude-frequency image, chemical process, convolutional neural network, fast Fourier transform, fault detection and diagnosis