An effective fault diagnosis approach based on optimal weighted least squares support vector machine

Abstract

Fault diagnosis is always a vital technology in the chemical industry and is influenced by a large number of process variables in the practical manufacturing process. Thus, an effective diagnosis method is crucial in practical chemical processes. A novel fault diagnosis method is proposed based on the weighted least squares support vector machine (WLSVM) to deal with the small sample and non‐linear partition data. It provides the strong potential in predicting faults, especially by further employing a suitable particle swarm optimization (PSO) algorithm to determine the most important parameters of the punishment factor and the Gaussian RBF kernel, which has the advantages of high accuracy and low false alarm rate. PSO‐WLSSVM is further accomplished with the data from the classic benchmark test set Tennessee Eastman (TE) process. Also, the models SVM, PSO‐SVM, and PSO‐LSSVM are employed on the TE process to compare the performance of PSO‐WLSSVM. The results show the validity of the proposed approach on the TE process.

Deep Forest-Based Fault Diagnosis Method for Chemical Process

“…Window setting Testing average diagnostic accuracy (%) Training time (min) Testing time for one sample (ms) [15,30] 56.3 42.1 0.7 [20,45] 68.7 38.3 0.5 [13,30,42] 72.4 70.6 1.3 [18,36,45] 75.6 61.5 1.1 [16,27,35,42] 74.8 79.7 1.4 [18,25,36,47] 75.9 83.4 1.6 From the general confusion matrix in Table 3, the formulas for calculating recall and precision are as follows:…”

“…With the increase of storage capacity and computing power, data-driven fault diagnosis methods have been widely used in chemical processes [4,5]. Among these methods, the multivariate statistical method, mainly including principal component analysis (PCA) [6,7], partial least squares (PLS) [8,9], independent components analysis (ICA) [10,11], Fisher discriminant analysis (FDA) [12,13], random forest (RF) [14], canonical correlation analysis (CCA) [15], exponential discriminant analysis (EDA) [16], and their derivatives [17][18][19][20][21][22], have also made a rapid progress. Although certain effects have been achieved by these data-driven methods, there are still two shortcomings: On one hand, most of these methods rely on an assumption of a single data distribution (e.g., Gaussian distribution) [23,24].…”

Fault diagnosis in industrial chemical processes using optimal probabilistic neural network

“…First, we sequentially input the normal training set data and the training set data of each type of fault into the RFtb method, as with the second experiment. Next, we input the normal training samples and fault training samples corresponding to various fault characteristics into the MCS‐PNN model and common diagnosis models such as multi‐layer perception (MLP), linear discriminant analysis (LDA), Naïve Bayes (NB), K nearest neighbours (KNN), and support vector machine (SVM), given their extensive use in the literature . In addition, the MCS algorithm parameters are changed to 20 in addition to the number of iterations, and the remaining parameters are consistent with the parameters set in the first experiment.…”

“…The PNN is a radial basis function feedforward neural network based on Bayesian decision theory, which can solve nonlinear regression and pattern classification problems well. In addition, He et al used the particle swarm optimization algorithm (PSO) to optimize parameters in the radial basis function (RBF) and improved the SVM to improve its model fault diagnosis ability. The optimization of classifiers such as neural networks can be transformed into convex optimization problems, which can be solved by methods such as the least squares method .…”

Chemical process fault diagnosis based on enchanted machine‐learning approach

“…In order to verify the performance of the proposed model in fault diagnosis, first we use genetic algorithm (GA), particle swarm optimization (PSO) algorithm, and MBA algorithm to optimize PNN for diagnosis. In addition, we input normal training samples and fault training samples corresponding to various fault characteristics into the MBA‐PNN model and some common diagnostic models, such as support vector machine (SVM), linear discriminant analysis (LDA), K nearest neighbour (KNN), naive Bayes (NB), and multilayer perceptron (MLP) . In terms of the parameters of the optimization model, the number of iterations is 50 and the optimal number of algorithms in the GA‐PNN, PSO‐PNN, and MBA‐PNN models is 20.…”