Xiaofeng Yuan

A Probabilistic Just-in-Time Learning Framework for Soft Sensor Development With Missing Data

Just-in-time learning (JITL) is one of the most widely used strategies for soft sensor modeling in nonlinear processes. However, traditional JITL methods have difficulty in dealing with data samples that contain missing values. Meanwhile, data noises and uncertainties have not been taken into consideration for relevant sample selection in existing JITL approaches. To overcome these problems, a new probabilistic JITL (P-JITL) framework is proposed in this brief. In P-JITL, variational Bayesian principal component analysis is first utilized to handle missing values and extract Gaussian posterior distributions of latent variables. Then, symmetric Kullback–Leibler divergence is creatively employed to measure the dissimilarity of two distributions for relevant sample selection in the JITL framework. Finally, a nonlinear regression model, Gaussian process regression, is carried out to model the nonlinear relationship between the output and the extracted latent variables. In this way, the proposed probabilistic JITL (P-JITL) is able to deal with missing data and select relevant samples more accurately. To evaluate the effectiveness and flexibility of P-JITL, comparative studies between P-JITL and traditional deterministic JITL (D-JITL) are carried out on a numerical example and an industrial application example, in which missing data are simulated with percentages from 0% to 50%. The results show that P-JITL can provide more accurate prediction accuracy than D-JITL in each scenario considered.

Soft Sensor Modeling of Nonlinear Industrial Processes Based on Weighted Probabilistic Projection Regression

Probabilistic principal component regression (PPCR) has been introduced for soft sensor modeling as a probabilistic projection regression method, which is effective in handling data collinearity and random noises. However, the linear limitation of data relationships may cause its performance deterioration when applied to nonlinear processes. Therefore, a novel weighted PPCR (WPPCR) algorithm is proposed in this paper for soft sensing of nonlinear processes. In WPPCR, by including the most relevant samples for local modeling, different weights will be assigned to these samples according to their similarities with the testing sample. Then, a weighted log-likelihood function is constructed, and expectation-maximization algorithm can be carried out iteratively to obtain the optimal model parameters. In this way, the nonlinear data relationship can be locally approximated by WPPCR. The effectiveness and flexibility of the proposed method are validated on a numerical example and an industrial process.

Weighted Linear Dynamic System for Feature Representation and Soft Sensor Application in Nonlinear Dynamic Industrial Processes

Industrial process plants are instrumented with a large number of redundant sensors and the measured variables are often contaminated by random noises. Thus, it is significant to discover the general trends of data by latent variable models in the probabilistic framework before soft sensor modeling. However, traditional probabilistic latent variable models such as probabilistic principal component analysis are mostly static linear approaches. The process dynamics and nonlinearities have not been well considered. In this paper, a novel weighted linear dynamic system (WLDS) is proposed for nonlinear dynamic feature extraction. In WLDS, two kinds of weights are proposed for local linearization of the nonlinear state evolution and state emission relationships. In this way, a weighted log-likelihood function is designed and expectation-maximization algorithm is then used for parameter estimation. The feasibility and effectiveness of the proposed method is demonstrated with a numerical example and an industrial process application.

Semisupervised JITL Framework for Nonlinear Industrial Soft Sensing Based on Locally Semisupervised Weighted PCR

Just-in-time learning (JITL) is a commonly used technique for industrial soft sensing of nonlinear processes. However, traditional JITL approaches mainly focus on equal sample sizes between process (input) variables and quality (output) variables, which may not be practical in industrial processes since quality variables are usually much harder to obtain than other process variables. In order to handle unequal length dataset with only a few labeled data, a novel semisupervised JITL framework is proposed for soft sensor modeling for nonlinear processes, which is based on semisupervised weighted probabilistic principal component regression (SWPPCR). In the new semisupervised JITL framework, traditional Mahalanobis distance and a new proposed scaled Mahalanobis distance are used for similarity measurement and weight assignment. By selecting the most relevant labeled and unlabeled samples and assigning them with the corresponding weights, a local SWPPCR can be built to estimate the output variables of the query sample. Case studies are carried out to evaluate the prediction performance of the proposed semisupervised JITL framework on a numerical example and an industrial process. The effectiveness and flexibility of the proposed method are demonstrated by the prediction results.

Retrofit Self-Optimizing Control: A Step Forward Toward Real Implementation

After 15 year development, it is still hard to find any real application of the self-optimizing control (SOC) strategy, although it can achieve optimal or near optimal operation in industrial processes without repetitive real-time optimization. This is partially because of the misunderstanding that the SOC requires to completely reconfigure the entire control system, which is generally unacceptable for most process plants in operation, even though the current one may not be optimal. To alleviate this situation, this paper proposes a retrofit SOC methodology aiming to improve the optimality of operation without change of existing control systems. In the new retrofitted SOC systems, the controlled variables (CVs) selected are kept at constant by adjusting setpoints of existing control loops, which therefore constitutes a two-layer control architecture. CVs made from measurement combinations are determined to minimize the global average losses. A subset measurement selection problem for the global SOC is solved though a branch and bound algorithm. The standard testbed Tennessee Eastman process is studied with the proposed retrofit SOC methodology. The optimality of the new retrofit SOC architecture is validated by comparing two state of art control systems by Ricker and Larsson etxa0al., through steady-state analysis as well as dynamic simulations.

Distributed defect recognition on steel surfaces using an improved random forest algorithm with optimal multi-feature-set fusion

Inspecting steel surfaces is important to ensure steel quality. Numerous defect-detection methods have been developed for steel surfaces. However, they are primarily used for local defects, and their accuracy in detecting distributed defects is unsatisfactory because such defects are difficult to locate and have complex texture characteristics. To solve these issues, an improved random forest algorithm with optimal multi-feature-set fusion (OMFF-RF algorithm) is proposed for distributed defect recognition in this paper. The OMFF-RF algorithm includes the following three aspects. First, a histogram of oriented gradient (HOG) feature-set and a gray-level co-occurrence matrix (GLCM) feature-set are extracted and fused to describe local and global texture characteristics, respectively. Second, given the small number of samples of distributed defect images and the high dimensionality of the extracted feature-sets, a random forest algorithm is introduced to perform defect classification. Third, the feature-sets vary greatly in performance and dimensionality. To improve the fusion efficiency, OMFF-RF merges the HOG feature-set and the GLCM feature-set through a multi-feature-set fusion factor, which changes the number of decision trees that correspond to each feature-set in the RF algorithm. The OMFF factor is found by optimizing the fitting curve of the classification accuracy of the test set using a stepping multi-feature-set fusion factor. In experiments, the effectiveness of the proposed OMFF-RF was verified using 5 types of distributed defects collected from an actual steel production line. OMFF-RF achieved a recognition accuracy of 91%, a result superior to support vector machine (SVM) and conventional RF algorithms.

Multi‐similarity measurement driven ensemble just‐in‐time learning for soft sensing of industrial processes

Just‐in‐time learning (JITL) technique has been widely used for adaptive soft sensing of nonlinear processes. It builds online local model with the most relevant samples from historical dataset whenever a query sample comes. Hence, the prediction performance greatly depends on the similarity measurement for relevant sample selection. Different similarity measurements have been developed for sample selection in the past decades. However, each method only focuses on one aspect of sample similarity and has its own limitations. Moreover, it is difficult to obtain the similarity mechanism of real process data. A single similarity measurement does not always provide satisfactory prediction performance. To deal with this problem, a novel ensemble just‐in‐time learning (E‐JITL) framework is proposed in this paper. In E‐JITL, different similarity measurements are adopted for sample selection. Then, local prediction models are constructed and trained to estimate the output of the query data with different groups of relevant samples corresponding to the similarity measurements. At last, a final prediction can be obtained by an ensemble strategy on each local model. The effectiveness of the E‐JITL is validated on two industrial applications.

Parameter identification for breakage distribution function of bauxite based on state transition algorithm

The breakage distribution function of bauxite plays an important effect in the ball milling process, which can be used to predict particle size distribution of the product at the outlet of the ball mill. The structure of the breakage distribution function can be determined by the experience knowledge. Parameters of the distribution function are usually identified by using the grinding test data. The parameter identification problem is a complex constrained nonlinear problem in the grinding process. In this paper, a two-stage modeling method is proposed, which divides multiple single models into two parts that are symmetric about the line. It can quickly search the global optimal solution, which can avoid the trap of local optimum. At the same time, the state transition algorithm is applied to optimize the parameters of the breakage distribution function in the two-stage model because of its simplicity and good numerical results. Experimental results show that the optimization of the two-stage model with state transition algorithm can reduce the error than the single model, and it can find the optimal solution more quickly and accurately than the Powell algorithm.

Output-related feature representation for soft sensing based on supervised locality preserving projections

Locality preserving projections (LPP) is a useful tool for learning the manifold of high dimensional data, which is a linear approximation of nonlinear Laplacian Eigenmap (LE). However, the original LPP algorithm is an unsupervised method that extracts features without any reference to the output information. In this paper, a supervised LPP (SLPP) framework is proposed for output-related feature extraction in soft sensor applications. In the SLPP framework, the output information is utilized to guide the procedures for constructing the adjacent graph and calculating the weight matrix, with which the intrinsic structure of the data can be better described. Two specific SLPP algorithms are described. For performance evaluation of the proposed methods, experiments on a numerical example and an industrial iromaking process are carried out. The results show the effectiveness of the proposed framework.

Soft Sensor for Multiphase and Multimode Processes Based on Gaussian Mixture Regression

Abstract For complex industrial plants with multiphase and multimode characteristic, traditional multivariate statistical soft sensor methods are not applicable as Gaussian distribution assumption of data is not met. Thus Gaussian mixture model (GMM) is used to approximate data distribution. In the previous GMM-based soft sensor modeling researches, GMM is only used to identify operating mode, then other regression algorithms like PLS are used for quality prediction in different localized modes. In this article, an existing method—Gaussian mixture regression (GMR) is introduced for soft sensor modeling, which has been already successfully applied in robot programming by demonstration (PbD). Different from past GMM-based soft sensors, GMR is directly used for regression. In GMR, data mode identification and regression are incorporated into one model, thus there is no need to switch prediction model when data mode has changed. Feasibility and efficiency of GMR based soft sensor are validated in the fermentation process and TE process.