Shi-Shang Jang

Constrained optimization of combustion in a simulated coal-fired boiler using artificial neural network model and information analysis☆

Abstract Combustion in a boiler is too complex to be analytically described with mathematical models. To meet the needs of operation optimization, on-site experiments guided by the statistical optimization methods are often necessary to achieve the optimum operating conditions. This study proposes a new constrained optimization procedure using artificial neural networks as models for target processes. Information analysis based on random search, fuzzy c-mean clustering, and minimization of information free energy is performed iteratively in the procedure to suggest the location of future experiments, which can greatly reduce the number of experiments needed. The effectiveness of the proposed procedure in searching optima is demonstrated by three case studies: (1) a bench-mark problem, namely minimization of the modified Himmelblau function under a circle constraint; (2) both minimization of NOx and CO emissions and maximization of thermal efficiency for a simulated combustion process of a boiler; (3) maximization of thermal efficiency within NOx and CO emission limits for the same combustion process. The simulated combustion process is based on a commercial software package CHEMKIN, where 78 chemical species and 467 chemical reactions related to the combustion mechanism are incorporated and a plug-flow model and a load-correlated temperature distribution for the combustion tunnel of a boiler are used.

Performance Analysis of EWMA Controllers Subject to Metrology Delay

Metrology delay is a natural problem in the implementation of advanced process control scheme in semiconductor manufacturing systems. It is very important to understand the effect of metrology delay on performance of advanced process control systems. In this paper, the influences of metrology delay on both the transient and asymptotic properties of the product quality are analyzed for the case when a linear system with an initial bias and a stochastic autoregressive moving average (ARMA) disturbance is under an exponentially weighted moving average (EWMA) run-to-run control. Tuning guidelines are developed based on the study of numerical optimization results of the analytical closed-loop output response. In addition, effective metrology delay of a variable time delay system is analyzed based on the resampling technique implemented to a randomized time delay system. A virtual metrology technique is a possible solution to tackle the problem of metrology delay. The tradeoff between additional error of virtual metrology and reduction in time delay is studied. The results are illustrated using an example of control of the tungsten deposition rate in a tungsten chemical-vapor deposition reactor. The basic conclusion is that metrology delay is only important for processes that experience nonstationary stochastic disturbance. In such a case, use of virtual metrology is justified if the error of the virtual metrology method is less than the error caused by stochastic process noise. The accuracy of the virtual metrology noise with respect to the traditional metrology is not critical, provided that the error due to metrology is much less than that due to process disturbances.

Fault Detection Based on Statistical Multivariate Analysis and Microarray Visualization

In this work, a statistical method is proposed to mine out key variables from a large set of variables recorded in a limited number of runs through a multistage multistep manufacturing process. The method employed well-known single variable or multivariable techniques of discrimination and regression but also presented a synopsis of analysis results in a colored map of p-values very similar to a DNA microarray. This framework provides a systematic method of drawing inferences from the available evidence without interrupting the normal process operation. The proposed concept is illustrated by two industrial examples.

Development of adaptive soft sensor based on statistical identification of key variables

An adaptive data-driven soft sensor is derived based on systematic dynamic key variables selection of a process system. The key variables are captured using statistical approaches. The on-line plant measurements can be directly selected as key features to estimate the tardily-detected quality variables. The statistical method adopted is the standard stepwise linear regression. The linear model is adapted as the on-line/off-line quality data becomes available. The adaptation of the model is implemented by standard Kalman filtering theory. The key variables are re-selected in case of new scenarios arrive and are detected by the soft senor. The real time data from an industrial O-xylene purification column is implemented to demonstrate the validity of the approach. Many different scenarios are simulated through an industrial standard dynamic simulator. The simulation results also showed the approach is adequate for the industrial applications. Copyright

Developing a robust model predictive control architecture through regional knowledge analysis of artificial neural networks

Abstract Chemical processes are nonlinear. Model based control schemes such as model predictive control are highly related to the accuracy of the process model. For a highly nonlinear chemical system, it is clear to implement a nonlinear empirical model, such as artificial neural network model, should be superior to a linear model such as dynamic matrix model. However, unlike linear systems, the accuracy of a nonlinear empirical model strongly depends on its original data or training data based on how the model is built up. A regional-knowledge index is proposed in this study and applied in the analysis of dynamic artificial neural network models in process control. New input patterns that imply extrapolations and thus unreliable prediction by an artificial neural network model can be recognized from a significant decrease in the regional-knowledge index. To tackle the extrapolation problem and assure stability of the control system, we propose to run a neural adaptive controller in parallel with a model predictive control. A coordinator weights the outputs of these two controllers to make the final control decision. The present state of the controlled process and the model fitness to the present input pattern determine the weightings of the controllers output. The proposed analysis method and the modified model predictive control architecture have been applied to a neutralization process and excellent control performance is observed in this highly nonlinear system.

A Virtual Metrology System for Predicting End-of-Line Electrical Properties Using a MANCOVA Model With Tools Clustering

The ability to predict end-of-line electrical properties of wafer in semiconductor manufacturing processes is critical to developing and maintaining a high yield. However, this is difficult because an advanced wafer manufacturing process consists of 300-400 steps, and in-line metrology data is only available for a few steps and for infrequently sampled wafers. Although a large amount of equipment sensor outputs are readily available for most wafers, most of the sensor variables may not be related to the end-of-line properties. Further, differences in end-of-line properties of wafers processed by tools of the same stage do not imply differences in the values of sensor variables between these tools. Thus, it is important to develop a reliable screening and model building procedure to construct a robust virtual metrology model with good generalization capability. Despite its simplicity, this approach is found to have significantly better generalization capability than nonlinear models, as well as substantial improvement in modeling and prediction capabilities of linear models that use only in-line metrology. The proposed method is also evaluated by an industrial application in a local fabrication unit.

Performance assessment of run-to-run control in semiconductor manufacturing based on IMC framework

The objective of this paper is to propose a universal methodology for performance assessment of run-to-run control in semiconductor manufacturing. The slope of the linear semiconductor process model is assumed to be known or subjected to mild plant/model mismatch. Based on an internal model control framework, analytical expressions of minimum variance performance (MVP) and best achievable performance (BAP) for a series of run-to-run control schemes are derived. In the methodology, closed-loop identification is utilised as the first step to estimate the noise dynamics via routine operating data, and numerical optimisation is employed as a second step to calculate the best achievable performance bounds of the run-to-run control loops. The validity of the methodology is justified by examples of performance assessment for EWMA control, double EWMA control and RLS-LT control, even under circumstances where the processes encounter model mismatch, metrology delay and more sophisticated noises. Several essential characteristics of run-to-run control are discovered by performance assessment, and valuable advice is offered to process engineers for improving the run-to-run control performance. Furthermore, a useful application example for online performance monitoring and optimal tuning of run-to-run controller demonstrates the advantage of the methodology.

Optimal energy management integration for a petrochemical plant under considerations of uncertain power supplies

The electric power demands of many petrochemical plants are matched by supplies from an in-house cogeneration system and from the electric grid. However, due to the fluctuations of fuel costs, production, and electricity rates, it is necessary to balance electric supply between these two sources. In reality, uncertain effects play a very important role in this decision-making problem. One of the most important uncertainties is the occurrence of power interruptions from either one of the supply sources, which could endanger operability and reliability of plant operations. To minimize the total energy cost under consideration of unexpected power failures, we break up the solution of the problem into two layers. The outer layer is to determine the optimum contracting of three-section time-of-use rate. We use an artificial neural network regression model as a meta-model to simulate the contour plot of a nonconvex cost function. The occurrences of incidental power failures are simulated by the Monte Carlo method. The inner layer is to determine the optimum operation of the cogeneration system. Since the searching space is huge in the outer layer and the Monte Carlo simulation in the inner layer is time consuming, we implement an interactive sampling search approach to find the optimal contract capacity in this multi-local-optima problem.

An EWMA Algorithm With a Cycled Resetting (CR) Discount Factor for Drift and Fault of High-Mix Run-to-Run Control

Run-to-run controllers based on the exponential weighted moving average (EWMA) statistic are probably the most frequently used for the quality control of certain semiconductor manufacturing process steps. The threaded-EWMA run-to-run control is an important stable control scheme. However, the process outputs will deviate largely in the first few runs of each cycle if the disturbance follows an IMA(1,1) series with deterministic linear drift and the thread has a long break length. In this paper, the output of the threaded-EWMA run-to-run control is derived, stability conditions are given, and the causes of large deviations in the first few runs of each cycle are found. Based on the analysis of system performance, a cycled resetting (CR) algorithm for discount factor is proposed to reduce the large deviations, as well as to achieve the minimum asymptotic variance control. Furthermore, how to deal with step fault is also discussed in this paper. By analyzing the influence of the fault, a discount factor resetting fault-tolerant (RFT) approach is proposed. Simulation study shows both the mean square error (MSE) and variance of the output by the proposed algorithm is about 30% to 50% lower than that of the algorithm with fixed discount factor in the process with and without oscillation. This verifies the effectiveness of the proposed approach.

Threaded EWMA Controller Tuning and Performance Evaluation in a High-Mixed System

The exponentially weighted moving average (EWMA) controller is a very popular run-to-run (RtR) control scheme in the semiconductor industry. However, in any typical step of semiconductor process, many different products are produced on parallel tools. RtR control is usually implemented with a ldquothreadedrdquo control framework, i.e., different controllers are used for different combinations of tools and products. In this paper, the problem of EWMA controller tuning and performance evaluation in a mixed product system are investigated by simulation and time-series analysis. It was found that as the product frequency changed, the tuning guidelines of a threaded EWMA controller were different for different types of tool disturbances. For a stationary ARMA(1,1) noise, the tuning parameter lambda should be decreased as product frequency decreases. If the tool exhibits nonstationary tool dynamics, e.g., ARIMA(1,1,1) noise, the tuning parameter should increase as the product frequency decreases.