Yuqing Chang

Complex process operating optimality assessment and nonoptimal cause identification using modified total kernel PLS

Due to process disturbances and some uncertainties, the process operating performance will deviate from the optimal operating point along with time, so it is very important to develop strategies for online operating performance assessment on optimality. However, a little work has been published in this research area to our knowledge. In this study, a new online operating optimality assessment method for complex industrial process is proposed. Firstly, a new version of T-KPLS algorithm combined with the modified KPLS is developed and used to establish the assessment models for each of the performance grades. By calculating the similarities between the online data and each performance grade, one can obtain the assessment result online, and the assessment results not only include the certain performance grade but also the performance grade conversion. For nonoptimal operating performance, a method of nonoptimal cause identification based on variable contribution is developed. Finally, the proposed online assessment method is applied in a gold hydrometallurgical process, which indicates the efficiency of the proposed approach.

Nonlinear dynamic quality-related process monitoring based on dynamic total kernel PLS

Total projection to latent structures (T-PLS) has been used for quality-related process monitoring. Compared to PLS, the T-PLS is more effectively in detecting the quality-related abnormal situations for linear and static processes. To describe the nonlinear and dynamic process characteristics, a new monitoring approach, dynamic total kernel projection to latent structures (DT-KPLS), is proposed in this paper for the nonlinear dynamic quality-related process monitoring. DT-KPLS consists of two parts: (i) T-KPLS decomposes the process data X into four subspaces in a high-dimensional feature space; (ii) the time-lagged extension of data matrix is performed before applying T-KPLS to capture process dynamic. Finally, the effectiveness of the proposed method is demonstrated by a cyanide leaching process.

Multivariate process monitoring based on the distribution test of the data

In most industry processes, there is no prior knowledge of the data distribution. If the process monitoring method is chosen without considering its constraints, it will get wrong conclusions and increase the rate of high leaking and false alarm. To solve the problem, a method of multivariate process monitoring and fault diagnosis based on distribution test of the data is proposed. First, the improved F-straight method is used to test the distribution of the process data. According to the test result, appropriate modeling method is chosen automatically to monitor the process and conduct fault diagnosis, which solves the constraints of PCA, ICA in application. Feasibility, efficiency and accuracy of the method are evaluated by the case study.

PCA-SDG based fault diagnosis on CAPL furnace temperature system

PCA-SDG based fault diagnosis method and its application on Continuous Annealing Process Line (CAPL) furnace temperature system are mainly discussed. Principle component analysis (PCA) method is applied to build the process monitoring model with a large number of historical data under normal operation conditions. High-dimension process data with noise and linear correlation are projected onto low-dimension and orthogonal sub-space. Real-time monitoring of furnace temperature system is carried out through online calculating T2 and SPE statistics of PCA model. When a fault is detected, the signed directed graph (SDG) model of furnace temperature system is used to interpret the residual contributions of PCA model, and then perform fault diagnosis with the rules of SDG. PCA-SDG method combines the advantages of both PCA and SDG methods. The effectiveness and reliability of the proposed PCA-SDG method are verified by the simulations.

Phase-Based RMFDA Fault Diagnosis Method Using Bootstrap Technique

The fault diagnosis performance of Fisher Discriminant Analysis (FDA) method is superior to Principle Component Analysis (PCA) by taking into account both normal and fault data for modeling. For the cases with insufficient fault data, a diagnosis strategy is developed based on Bootstrap and phase-based Recursive Multi-way Fisher Discriminant Analysis (RMFDA). By this method, modeling data was constructed by Bootstrap. Besides, the diagnosis information of the previous phase was introduced in the next phase for MFDA modeling by combining recursive method. The effectiveness of the proposed method is demonstrated by applying it to the hydraulic tube tester.

Plant-wide optimization for gold hydrometallurgy based on the fuzzy qualitative model and interval number: Plant-Wide Optimization for Gold Hydrometallurgy

Since the difficulty of obtaining accurate online‐measurement of some key variables in hydrometallurgy plant‐wide production process causes that the quantitative models of some procedures are difficult to establish and the plant‐wide optimization based on quantitative model is difficult to realize, a plant‐wide optimization method based on interval numbers is proposed in this work. First, based on the information of expert knowledge and the experience of field workers, a fuzzy qualitative model is constructed, and outputs of the qualitative model are reasonably divided into multiple modes simultaneously. By analyzing the process properties, an optimal control problem for hydrometallurgy plant‐wide production process in the steady state is proposed to achieve process requirements which is to obtain the lowest cost of sodium cyanide and zinc as well as high gold quality. Then, by using interval numbers to represent the key variables that cannot be measured, an optimization method based on interval numbers is proposed for every output mode of the qualitative model. Finally, the hydrometallurgy process is carried in industrial simulation experiment. The results demonstrate that the proposed optimization scheme has much wider applicability than conventional optimization methods, especially for its improved performance of solving optimization problem with uncertainty.

Phase-based recursive regression for quality prediction of multiphase batch processes

Batch processes have been widely applied in industry. Quality prediction plays an important role in batch process quality control. Multiple phases within a batch have correlated contributions to final qualities. So, previous phases should be considered in the quality-regression modeling for the current phase. In this paper, a new phase-based recursive statistical quality regression method is proposed for the quality prediction of multiphase batch processes. First, within each batch, main phases are obtained by a basic process analysis. Second, by analyzing phase characteristics, phases which have significant impacts on final qualities are identified as critical-to-quality phases. Then, a recursive quality regression algorithm is proposed using each quality residual of the regression models built in those critical-to-quality phases. Besides, phase characteristics are represented by the average trajectories of process variables within each phase. Quality prediction is performed at each sample point for online prediction. The application to a typical multiphase batch process, injection molding process, illustrates the feasibility and performance of the proposed algorithm.

Multimode analysis and online monitoring for injection molding processes

Injection molding process is a typical batch processing technique to manufacture plastic products. In such process, plastic particles are heated to melt, injected to a mold with a certain configuration, and cooled down to solid state. In recent years, because of the more pressing market demand for modern multi-species, multi-standard and high-quality products, more attention have been paid on batch processes which can produce small-quantity, high-added-value products. In order to improve the safety of multimode production process, inter-mode analysis is an urgent need to establish monitoring for multimode production process. Besides inherent process variation along the time direction within each batch, multimode characteristic along the batch direction widely exists in injection molding processes. Different operation modes along the batch direction may not be well captured by a single model. In this work, multiple modes are modeled separately for online process monitoring along the batch direction. First, after a batch cycle is divided into multiple phases, normality test is conducted to analyze the data characteristic. Then, ICA-PCA models are constructed for each mode within each phase according to their different characteristic considering both the Gaussian and non-Gaussian information of the process data. Then online monitoring model is properly chosen by identifying which mode the new sample point belongs to. The application to a real injection molding process illustrates the feasibility and performance of the proposed algorithm.

Industrial process operating optimality assessment based on Gaussian mixture model

As a new research issue, the operating optimality assessment for industrial processes has received growing interests in recent years. In this study, a novel online operating optimality assessment method based on Gaussian mixture model (GMM) is proposed for industrial processes. The offline training data can be automatically divided into several data sets by the proposed performance grade division method, and it lays the foundation for establishing the assessment models. Then the assessment models are developed not only for each stable performance grade but also for the performance grade conversions for the first time, which provides important guidelines for online assessment. In online assessment, the multiple hypotheses testing technology is used to ensure the reliability of the assessment results through controlling the false discovery rate. The validity and superiority of the proposed operating optimality assessment strategy are then validated through case study on gold hydrometallurgical process.

Process operation performance optimality assessment and cause identification based on PCA-DCD

Making precise assessment on the optimality degree for process operation performance and giving correct operating instructions for non-optimality, contributes to the adjustment of production to keep the process stay at optimal states. Better operation performance helps the factory obtain more benefit. Traditional monitoring approaches determine if the process is working normally, but cannot tell if it is running well. Therefore, establishing effective operation performance optimality assessing mechanism is of great significance. Data driven method, Principal Component Analysis (PCA), deals with linearly related variables and extracts the main information from data. However, PCA lacks in interpretability. Process knowledge based technique, Dynamic Causality Diagram (DCD), intuitively reveals causal relations between variables. But the discretization in DCD decreases calculation precision. A novel approach to realize process operation performance assessment and corresponding cause identification for non-optimal performance is proposed in this article. The proposed method is finally utilized in gold hydrometallurgy process operation performance assessment to illustrate its effectiveness.