Michael Joseph Piovoso

Multi-way PCA applied to an industrial batch process

Batch and semi-batch processes are common in most chemical companies. These processes are characterized by a prescribed processing of materials for a finite duration of time. Feedback control often cannot be applied to correct for disturbances in a timely manner during the batch. Techniques which can provide insights into correlations among variables and their relationships to product quality will provide insights in the design of a control strategy that may improve product quality and minimize batch to batch variations. In this paper, the authors apply the statistical technique of multi-way principal component analysis to analyze the data from an industrial batch process. Using this technique, the authors were able to associate several significant causes of variability with the recipe imposed by the process. This information gave rise to a different control strategy which is presented. Subtle effects among batches were also uncovered and identified.

A multivariate statistical controller for on-line quality improvement

Abstract Producing good quality products is an important process control objective. However, achieving this objective can be very difficult in a continuous process, especially when quality measurements are not available on-line or they have long time delays. In this paper, a control approach using multivariate statistical models is presented to achieve this objective. The goal of the control approach is to decrease variations in product quality without real time quality measurements. A PCA model which incorporates time lagged variables is used, and the control objective is expressed in the score space of this PCA model. A controller is designed in the model predictive control (MPC) framework, and it is used to control the equivalent score space representation of the process. The score predictive model for the MPC algorithm is built using partial least squares (PLS). The proposed controller can be developed from and implemented on top of existing PID control systems, and it is demonstrated in two case studies, which involve a binary distillation column and the Tennessee Eastman process.

Nonlinear adaptive control with parameter estimation of a CSTR

Abstract The goal of this paper is to describe a linearizing feedback adaptive control structure which guarantees high quality regulation of the output error in the face of unknown parameters. The effectiveness of this control structure is demonstrated on a continuous stirred tank reactor in two instances. The first is when there is full state feedback and the second when only temperature measurements are available. In the latter a nonlinear observer is constructed to infer conversion. In both cases conditions for asymptotic stability are presented and discussed.

Operating regime-based controller strategy for multi-product processes

Abstract A hybrid, dynamical supervisory control strategy is introduced to control multi-product processes. The scheme involves the selection of the best setpoint controller, from an existing family of feedback controllers, to be placed into feedback with the process so as to cause the output to track the desired setpoint. This task is performed without exhaustive trial and error, by minimizing a specific normed output estimation error that relates the process output to a shared controller state. The performance of the supervisory control strategy is demonstrated on a continuous-stirred tank reactor that can be operated at three different conversion levels. However, the most desirable operating point is open-loop unstable. Linear controllers are designed for each operating region and the role of the supervisory control strategy will be to determine the switching logic as different scheduling policies are demanded.

Statistics on reliability of neural network estimates

It has been demonstrated that artificial neural networks can be used to infer estimates of variables infrequently measured in many applications. These estimates have been used in closed-loop control applications, however, the reliability of the estimates have not been used to improve the controllers performance. This work focuses on the generation of an appropriate statistic for the reliability of the estimate, the confidence interval. This statistic is calculated from an empirical sampling distribution obtained using the bootstrap technique. A demonstration of the bootstrap method in the context of an ANN to estimate of a distillation process bottoms composition is provided. A discussion on the use of the bootstrap estimate and its confidence interval to the practical problem of controller tuning and process performance follows.

Poisson Wavelets Applied to Model Identification

Abstract A novel wavelet transform is introduced based on the backward difference of the Poisson probability density function. This family of wavelets is a function of one discrete and two continuous variables. The Poisson wavelet transform is useful for system identification, parameter estimation and model validation. It is particularly well-suited for linear time-invariant systems that are modelled as combinations of decaying exponentials with a single time delay. This is demonstrated on a three-tank system modelled as a first-order plus deadtime functional form.

Control of a Batch Reactor Using a Multivariate Statistical Controller Design

Abstract Batch processes play an important role in the chemical, pharmaceutical, and agricultural industries. However, the main characteristics of batch processes, flexibility, finite duration, and nonlinear behavior are associated with both their success and their incompatibility with the usual techniques for control. The batch reactor studied indirectly controls the extent of reaction by using reactor pressure to control temperature which is as an inferential measurement of extent of reaction. The manipulated variable is the flow of the heating or cooling medium through a split range valve. However, just controlling to a reactor temperature profile will not necessarily reduce product nonuniformity and batch-to-batch variability. In this paper, it is proposed to control the reactor via control actions calculated in a reduced representation space of a principal component representation, the score space. This space is determined by the interrelationships among the process variables using Multi-way Principal Component Analysis (MPCA); the control action that will regulate the scores to the nominal score trajectories are calculated. The development of the MPCA representation of the process and the associated controller based on score space data are presented. The performance of this controller formulation in the face of a typical heat transfer disturbance is illustrated.

OPTIMAL AVERAGING LEVEL CONTROL FOR MULTI-TANK SYSTEMS

Abstract In this paper earlier results on optimal averaging level control (McDonald et al, 1986) are extended to cover the case where two or more tanks are connected in series. It is shown that the optimal averaging level control approaches can be derived for a multiple tank system if the total volumetric capacity of the system is taken into account. Intermediate flows between tanks are arbitrary, provided they satisfy the level constraints. The optimal solutions that consider tanks together produce beuer flow filtering than the optimal averaging level control approaches where each Lank is treated individually.