Neural Network Modeling for Advanced Process Control Using Production Data

Mevawalla, Z. N.; May, G.S.; Kiehlbauch, Mark

doi:10.1109/tsm.2011.2115261

Cited by 19 publications

(5 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…At the same time intelligent (AI) strategy based control [8][9][10][11][12][13] is now a trend as process system is becoming complex which is difficult to cope with conventional control strategy that requires accurate mathematical model. Many of them use neural network based control strategy [14][15][16][17][18]. Distributed implementation is also necessary to make the software available across several machines [19].…”

Section: Introductionmentioning

confidence: 99%

Advanced control software framework for process control applications

Hossain¹,

Hussain²,

Omar³

2014

IJCIS

View full text Add to dashboard Cite

Industries are now moving towards PC-based control from PLC (Programmable Logic Control) based control as the PC (personal computer) is easily available and capable of implementing various control strategies to improve productivity. Advances in both hardware and software technology are expediting this move. Absence of a user friendly, cost effective and easy to apply advanced control strategy based software is the motivation for this work. In this paper we reviewed issues of software application in process control systems. We present the development of an advance process control software tool and its application on a pH neutralization plant. Neural network model based DIC (Direct Inverse Control) strategy is implemented to control the process. The application and results shows the viability and robustness of the advanced control software and its superiority over the conventional PID (proportional integral derivative) controller.

show abstract

Section: Introductionmentioning

confidence: 99%

Advanced control software framework for process control applications

Hossain¹,

Hussain²,

Omar³

2014

IJCIS

View full text Add to dashboard Cite

show abstract

“…In general, data-based control methods can be divided into two categories: the indirect approaches and the direct approaches. For the indirect approaches, people need to build an equivalent model using measured data and modeling techniques, such as system identification neural networks [9][10][11] and support vector machines [12][13][14], and then design the controller according to this equivalent model.…”

Section: Introductionmentioning

confidence: 99%

Data‐based output feedback control using least squares estimation method for a class of nonlinear systems

Wang

Liu

2013

Intl J Robust & Nonlinear

View full text Add to dashboard Cite

This paper develops a data-based output feedback control method for a class of nonlinear systems, which have unknown mathematical models. The dynamic model of the system is assumed to be smooth, while the corresponding Jacobian matrices are constant matrices in each sampling period. We employ a zero-order hold and a fast sampling technique to sample and measure the output signal. When these measured data contain white noises, we use the least squares method to estimate the corresponding Jacobian matrices. The feedback gain matrix is calculated and adjusted adaptively in real time according to them. Theoretical analysis on the convergence condition is provided, and simulation results are used to demonstrate the feasibility of this method. methods need some prior knowledge about the system characteristics. There are some typical direct data-based control methods, such as iterative learning control [15][16][17], model-free adaptive control [18,19], virtual reference feedback tuning (VRFT) [20,21], and unfalsified control [22,23]. The iterative learning control method requires the system to perform the same task over a finite time interval repetitively. The model-free adaptive control method is suitable for the nonlinear systems, which have continuous partial derivatives with respect to control inputs, and it needs the systems to satisfy some generalized Lipschitz conditions. The VRFT method is usually applied to control discrete-time SISO systems, whose controller design problem is transformed into a parameter identification problem with the help of a virtual reference signal. The VRFT method assumes that the structure of the controller is known. The unfalsified control method recursively "falsifies" controllers that fail to satisfy a performance specification by using online I/O data.This paper develops a data-based output feedback control method for a class of nonlinear systems, which have unknown mathematical models. The dynamic model of the system is assumed to be smooth, while the corresponding Jacobian matrices are constant matrices in each sampling period. We employ a zero-order hold (ZOH) and a fast sampling technique to sample and measure the output signal. When these measured data contain white noises, we use the least squares method to estimate the corresponding Jacobian matrices. The feedback gain matrix is calculated and adjusted adaptively in real time accordingly. This control method only needs some basic prior knowledge about the system, while requiring no repeatability of the system behaviors. Besides, this method is also applicable to MIMO systems.Moreover, most existing nonlinear adaptive control methods generally fall into two broad categories: methods based on explicit knowledge of mathematical model structures and methods based on intelligent control theory. The first category consists of control methods with special requirements on the model structure [24][25][26], for example, lower triangular structure, nonlinear affine structure with respect to the input, and the structure of controlled pla...

show abstract

“…6 ML-based techniques for PV manufacturing have been explored for solar cell material design, 7 optimizing individual processes, 8 and a combination of processes. 9 It has also been explored with regard to DoE optimization, 10 quality control, 11 and troubleshooting with access to wafer tracking. 12 However, simultaneous optimization of the entire fabrication process has not yet been done.…”

mentioning

confidence: 99%

Optimization of Solar Cell Production Lines Using Neural Networks and Genetic Algorithms

Buratti

Eijkens

Hameiri

2020

ACS Appl. Energy Mater.

View full text Add to dashboard Cite

To keep improving the efficiency-to-cost ratio of photovoltaic solar cells, manufacturing lines must be continuously improved. Efficiency optimization is usually performed process-wise and can be slow and time-consuming. In this study, we propose a machine-learning-based method to perform simultaneous multiprocess optimization. Using the natural variation of a production line, we train machine learning models to investigate the relationship between process parameters and cell efficiency. We employ genetic algorithms to identify new process parameters in order to maximize cell efficiency. The proposed method is demonstrated on a simulated production line of monocrystalline aluminum-back surface field solar cells. Using neural networks, an accurate model is built to predict cell efficiencies from input process parameters with errors of <0.03% absolute efficiency. In five iterations, the mean cell efficiency increases from 18.07% to 19.45%. Provided strong process monitoring and accurate wafer tracking, the proposed method is directly applicable to production-type datasets, enabling the photovoltaic industry to build smart factories and join the fourth industrial revolution.

show abstract

Neural Network Modeling for Advanced Process Control Using Production Data

Cited by 19 publications

References 24 publications

Advanced control software framework for process control applications

Advanced control software framework for process control applications

Data‐based output feedback control using least squares estimation method for a class of nonlinear systems

Optimization of Solar Cell Production Lines Using Neural Networks and Genetic Algorithms

Contact Info

Product

Resources

About