Willy Wojsznis

Practical approach to tuning MPC.

This paper presents the results of a heuristic approach for developing model predictive control (MPC) tuning rules. The tuning has been applied and tested in easy-to-use MPC. Process modeling in this MPC uses normalized input/ output range. As a result there is no need for tuning outputs, a procedure known as adjusting equal concern error. Penalties on moves are set as a function of process dead time as the primary factor, with some correction from process gain. The default calculation delivers robust control, which tolerates up to triple increase in process static gain. If control is too aggressive, further on-line adjustment can be done by set point reference trajectory. Test results show that this tuning is robust for process gain change, however, it is much less efficient in compensating for process dead-time changes. It was found that dead-time mismatch is much better compensated with the model correction filter. Combining the three handles, i.e., penalties on moves, reference trajectory, and model filter, easy and intuitively understandable MPC tuning was achieved. The findings are illustrated by numerous MPC simulated tests.

Evaluating PID adaptive techniques for industrial implementation

This paper outlines several adaptive techniques considered for implementation in the PID controller. It presents frequency domain interpolators for tracking ultimate gain and ultimate period, model-free tuning based on balancing P and I terms, and the adaptation of controller switching and model switching with the interpolation of parameters. From the perspective of the authors, a model switching technique using the interpolation of parameters offers a clear advantage, particularly for feedforward/feedback loops and multivariable control.

PID control using wireless measurements

This paper presents a wireless PID controller, known as PIDPlus. The same quality control as wired PID can be provided by PIDPlus using a wireless measurement, despite slower wireless measurement update and communication interruptions. The PIDPlus novel algorithm is based on a modification of the PID reset and rate calculation to account for non-periodic measurement updates. The paper presents and evaluates an alternate approach of PID control accounting for non-periodic measurements with a Kalman filter observer that has been modified for use with a wireless measurement. A second alternative shows how the Smith Predictor may be modified to work with a wireless measurement. Test results are presented that compare the PIDPlus performance to that achieved using these alternate approaches for a variety of operating conditions that may be encountered when using wireless transmitters.

Data cleaning in the process industries

Abstract In the past decades, process engineers are facing increasingly more data analytics challenges and having difficulties obtaining valuable information from a wealth of process variable data trends. The raw data of different formats stored in databases are not useful until they are cleaned and transformed. Generally, data cleaning consists of four steps: missing data imputation, outlier detection, noise removal, and time alignment and delay estimation. This paper discusses available data cleaning methods that can be used in data pre-processing and help overcome challenges of “Big Data”.

Process Control over Real-Time Wireless Sensor and Actuator Networks

Wireless technologies have been successfully applied in the process industry since the creation of the first international standard IEC62591 WirelessHART. Applications started in areas where wireless sensors provide rich process information to the automation systems. Although real and demonstrated control applications are advertised, wireless-for-control is still in the initial stage and faces a lot of challenges. In particular, feedback latency and battery longevity, which are also problems for wireless-for-sensing, are even more critical when wireless actuators are applied. There are additional challenges in using wireless actuators because they actively affect the process. This paper lays the foundation for control using a wireless actuator. It demonstrates how traditional control methodologies can be modified to effectively work with general wireless communication. The innovations are tested with simulations and experimentations, both on commercial distributed control systems.

Fieldbus support for process analysis

Abstract Fieldbus support for the analysis of process and control system operation is addressed in the Fieldbus Foundations specification. Devices based on this specification will, as an integral part of their design, allow process inputs and outputs to be precisely sampled by the device without aliasing or skewing. This information may be accessed over an H1 or H2 fieldbus in an efficient manner without impacting control distributed between field devices. An overview of the features defined by the Fieldbus Foundations specification for the support of process analysis is presented. Considerations given to filtering, variable sample rates, and communication of process measurements are discussed. Also, the manner in which fieldbus may influence future implementations, such as process identification for controller self-tuning, is reviewed.

Neural network assisted control loop tuner

Explores the application of nonlinear tuning rules estimators to a known relay-oscillation tuner. Two approaches were tested. One uses nonlinear functions to approximate the desirable controller parameters. The other incorporates a neural network for computing the process model and controller parameters. As a basis for computation, the ultimate gain, ultimate period, and process dead time are defined during the tuning experiment. The neural network is trained in simulation using these process parameters as inputs and known process model parameters and desired PID controller tuning parameters as outputs. The PID tuning parameters are defined from the simulation process model using IMC or lambda tuning rules. This concept was implemented in a scalable industrial control system. Simulation test results show a vast improvement in model identification and control loop performance as compared to previous relay-oscillation based tuning approaches.

Flexible gain scheduler

Abstract A gain scheduler for PID controllers is discussed, which combines the simplicity of a rigid gain scheduler with continuous updating of controller parameters, specific to adaptive systems. The algorithm distinguishes the ranges of a linear process like a rigid scheduler and, additionally, assumes nonlinear process behavior in some proximity of the range limits. The algorithm applies fuzzy interpolation in the bands of process nonlinearity. In the implementation, available features include disabling fuzzy control while preserving a hysteresis between ranges.

Multistate PLS for continuous processes

Batch process monitoring methods, such as multiway PCA and multiblock multiway PLS, make use of time profiles to define expected process variable trajectories for statistical process control. Nevertheless, continuous process counterpart methods of desired process variable profiles have not been developed, nor addressed in the literature. This work presents a novel methodology to define multiple operating points around which continuous processes operate. Process operating regions are divided into multiple states of operation and shifts in operating conditions are captured by special variables, named state variables. Transition trajectories between states are calculated to determine the most likely path between states. This methodology can be implemented in the context of empirical monitoring methods, named Multistate PLS. A case study shows how this methodology enhances fault diagnostics and statistical monitoring of continuous processes.

Extending LP Optimizer Functionality for Model Predictive Control

The subject of this paper is linear programming (LP) optimizer application with Model Predictive Control (MPC). This extremely successful merger of two major control technologies is enabled as a consequence of the MPC feature of providing a prediction of the process outputs up to steady state, thus creating the required conditions for optimizer operation. However, the standard LP algorithm which finds a solution only within acceptable limits, does not perform properly when some of the predicted process outputs are out of limits. On the other hand, the optimizer applied with the MPC controller must always find a solution and thus there is a need to extend the original optimization formulation. This paper presents robust and reliable ways of handling optimized process outputs that are out of the limits. The technique is based on the priority structure, penalizing slack variables, and redefining the constraint model. In addition LP functionality is extended by defining one- or two-sided ranges around the control variables set points and preferred settling values for the manipulating variables. This technique has been implemented in an industrial control system and will be presented interactively by simulating the optimization and control of a distillation column.