Model-Free Adaptive Control: Impact on the Automation Industry
05 February 2010
MFA technology has been fulfilling its promise as a ‘next-generation mainstream controller.’ Since its introduction to the automation industry in the late 1990s, the technology has been implemented by a number of end-users and embedded in OEM equipment by a host of process control, equipment control, and building automation vendors.
CyboSoft key members (left to right): Co-founder Dr. Carl Hsu, Director of Engineering Steve Mulkey, and CEO Dr. George Cheng.
The February 2001 issue of Control Engineering Europe reported on a new development in process automation, Model-Free Adaptive (MFA) control, which offered a ‘next-generation mainstream controller’ capable of regulating time-varying, multivariable processes without the need for complex mathematical models or manual tuning procedures. (See ref. 2; for a web version of this article click here.)
Developed by CyboSoft, General Cybernation Group, Inc., a California-based control technology firm, MFA promised significantly improved control performance and closed-loop stability for a wide range of industrial plant operations.
The following article examines the impact of MFA control since its introduction to the automation industry in the late 1990s. The technology has been implemented by a number of end-users around the world, and embedded in OEM equipment by a host of leading process control, equipment control, and building automation vendors.
CyboSoft emerged onto the global automation market in September 1997, unveiling its patented MFA technology at the ISA Tech show in Anaheim, California. The company’s CEO, Dr. George Cheng, heralded MFA as a ‘breakthrough for automatically controlling simple to complex physical processes in industrial environments.’
According to Dr. Cheng, no other comparable control technology possessed all the attributes of MFA, which he described as ‘adaptive, robust and easy to use, and requiring minimal investment and risk.’
Since that time, CyboSoft’s mission has been to deploy the appropriate MFA controllers in instruments, equipment, tools, software platforms and automation systems of brand-name suppliers in targeted industries and market segments.
The company believes MFA can replace Proportional, Integral, Derivative (PID) technology on a very large scale, and provide an effective solution for controlling existing loops operating in manual control.
Dr. Cheng commented, ‘We developed MFA as a ‘dream controller,’ which would be a superior alternative to PID controllers. PIDs are unable to control complex systems—no matter how much tuning is performed.
‘MFA technology also has a much wider robust range than PID. Additionally, it eliminates the need for complex and costly process models required with model-based controllers.’
In many industrial operations, a model-based adaptive control approach can be difficult and costly to implement. For dynamic modeling-based adaptive control, there may not be enough time and data to learn a new model. Control experts typically design a high-speed, complex control system with precise process models requiring significant financial and time investments.
Despite the performance benefits of MFA control, Dr. Cheng acknowledges the challenges facing the technology.
‘In the automation industry, there is a strong mentality of PID being ‘good enough.’ Some people are afraid of making any changes, even though their installed technology and products might be outdated and wasting a lot of energy and materials.’
He added, ‘We are persistent in our efforts to work with companies having the vision and desire to stay ahead by applying new automation solutions.’
Figure 2. MFA advantages and suitability.
Today, many industrial processes are still being controlled manually or by 70-year-old PID technology. PID is a general-purpose automatic controller useful for controlling simple processes. However, it has significant limitations.
PID works for linear and time-invariant processes, but cannot effectively control tough loops that are nonlinear, time-variant or coupled, or having large time delays, major disturbances and uncertainties.
Also, PIDs cannot adapt or tune themselves with changes in operating and environmental conditions. To overcome this limitation, self-tuning PID controllers were introduced but they, too, suffer from the inability to control complex systems.
Why is MFA better? we ask Dr. Cheng.
‘MFA isn’t just one controller, but a series of controllers—each of which solves a difficult control problem. Users can simply choose the appropriate MFA controller, do some basic configuration, and launch.’
He says the MFA controller portfolio includes
* SISO MFA controller to replace PID and control simple to complex processes;
* MIMO MFA controller to control multi-variable processes;
* Nonlinear MFA controller to control extremely nonlinear processes;
* Anti-delay MFA controller to control processes with large time delays;
* Robust MFA controller to force the process variable to stay within defined bounds;
* Feedforward MFA controller to deal with measurable disturbances;
* MFA pH controller to control pH processes;
* Anti-delay MFA pH controller for pH processes with large time delays;
* Time-Varying MFA controller for processes with large process time constant and delay time changes;
* Flex-phase MFA controller to control open-loop oscillating processes;
* MFA XRT controller to control eXothermal Reactor Temperature; and
* MFA Opminiser for dynamic optimisation applications.
For example, pH control is needed in just about every process plant, and yet a large percentage of pH loops perform poorly. This results in inferior product quality, environmental pollution and material waste. MFA pH controllers have proven to be beneficial in controlling pH loops for industrial wastewater treatment, especially for pH processes with significant time delays, resulting in significant chemical reagent savings and quick ROI.
Suitablity of PID, MFA, and Model-Based control
Figure 2 illustrates the suitablity of PID, MFA, and Model-Based control.
PID is a controller for a large variety of processes where no detailed process information is required (‘Black Box’). However, since PID is a fixed controller, it cannot deal with process dynamic changes.
Model-based control is well suited for controlling a process where detailed knowledge is available (‘White Box’). For instance, a perfect control system can be designed using model-based methods to control an airplane, since the mathmetical models are readily available.
MFA is suitable for controlling processes with qualitative process knowledge, but where no detailed process models are available. In addition, the process dynamics can have significant variations (‘Grey Box’). Many industrial processes are grey boxes that have frequent load, fuel and dynamic changes.
Unlike PID, MFA is inherently designed to adapt when process dynamics change. As shown in Figure 3, starting from the same oscillating control condition, the control loop will continue to oscillate under PID control, while the MFA system will quickly adapt to an excellent control condition. When the setpoint changes, the MFA loop shows no oscillation.
Figure 3. Refining furnace with multiple temperature zones – a typical application for the MIMO MFA controller
MFA has been widely deployed in most types of continuous and semi-continuous plants controlling critical process variables and quality variables, including temperature, pressure, level, flow, density, concentration, pH, moisture, product dimensions, etc. For instance, in a large-scale continuous process like a distillation column chain, robust MFA controllers can be used to control the levels to allow smooth material and energy transfers between operating units, and also protect against overflow during abnormal conditions.
MIMO (Multi-Input-Multi-Output) MFA controllers can be used to control various multivariable processes such as evaporators, distillation columns, and industrial furnaces. As shown in Figure 3, an MIMO MFA control system was launched to control a refining furnace with four temperature zones in an oil refinery. MFA tightly controls furnace outlet temperature and minimises deviations in zone temperatures; decouples loop interactions and minimises chain reactions among the columns and furnaces; improves feed throughput; and minimises over- and under-heating. Model-based control could also be used here, but model-building and maintenance will be labour intensive, which discourages wide range deployment.
Path of Development
Since its founding in 1994, CyboSoft has developed an expanding set of MFA controllers to solve difficult control problems. Today, MFA controllers are available on platforms offered by various OEM vendors supplying building controllers, single-loop controllers, Programmable Logic Controllers (PLCs), process automation controllers (PACs), and control software.
For instance, CyboSoft recently developed a special MFA XRT controller to control the temperature of exothermal reactors. This innovative controller handles delayed ‘run-away’ processes with fast and intelligent movements that force the system to reach a dynamic balance.
CyboSoft originally implemented MFA controllers in CyboCon, a software product designed to run on a PC interfaced to a PLC or DCS through an API-based interface driver or OPC. CyboSoft also developed a number of hardware control products including its CyboCon CE advanced control instrument, CyboCon Dragon Micro DCS, and MFA Loop Controller.
MFA controllers have a small footprint, use very little CPU time, and can be embedded in a wide range of CPU and programming platforms. CyboSoft has deployed embedded MFA software with a growing list of strategic partners. For instance, users of National Instruments’ LabVIEW™ software can use MFA Control Toolset for LabVIEW software to help solve difficult control problems quickly during their equipment design, proto-typing, and testing phases. This enables equipment developers to significantly reduce R&D risks and cost, and shorten their time to market.
In the academic and engineering design areas, the MFA Control Toolbox for MFATLAB software is very useful. The product enables users to simulate, test and evaluate MFA controllers on sophisticated process models built in the MATLAB/Simulink environment. Professors can teach and students can learn MFA control methods through their familiar MATLAB/Simulink environment.
Energy, building automation
In the energy industry, Nabors Industries through its subsidiary Canrig Drilling uses embedded MFA controllers in its DrillSmart™ drilling systems for oil and gas exploration and production. Due to MFA's adaptive and robust capabilities, challenging jobs such as horizontal and lateral drilling are much easier. MFA controls the rate of penetration, weight on bit, and differential pressure.
Since 2005, MFA control technology has been embedded as part of the Siemens APOGEE® Building Automation System for controlling heating, ventilation and air conditioning (HVAC) systems. It is estimated that 100,000 new MFA control loops are launched annually for this application. The MFA software improves energy efficiency and comfort level since it optimises control of temperature, pressure, airflow and humidity regardless of load and seasonal changes.
Jeff Wills, Product Portfolio Manager, Siemens Industry Inc., Building Technologies Division, said, ‘Thanks to MFA technology, which is a leading-edge and well-established industrial control algorithm, APOGEE system users can achieve optimal control performance.
‘Benefits include increased valve and actuator life expectancy. The reduction in cycling-induced wear and tear on valves and actuators increases their lifespan and reduces repair, replacement and maintenance costs for end devices. This control solution also eliminates the need for seasonal control loop retuning because it continuously and automatically adjusts to system changes.’
Major Company Milestones
CyboSoft has has received multiple grants from the U.S. Department of Energy (DOE) for R&D involving power boiler control, actuation control, and industrial furnace control using MFA control technology.
The boiler control project is aimed to develop intelligent control systems for once-through supercritical boilers and circulating fluidised-bed (CFB) boilers, both of which are much more efficient but difficult to control. Since coal is and will be the main energy source for electricity generation, clean coal technologies are critical in order to achieve DOE’s ultimate goal: delivering maximum-energy-efficiency, near-zero-emissions, fuel-flexibility, and multi-products.
Looking to the Future
Basic architecture of an MFA controller -- only a portion of the neural network diagram is reproduced here.
Recent MFA innovations are addressing the most difficult automation challenges, including problems associated with climate change and global warming, where enhanced control capabilities are a major factor in emissions reduction, pH control, oil exploration and production, and renewable energy.
CyboSoft is currently focused on developing multiple new MFA control solutions and products for the power generation and renewable energy markets. This includes products to solve actuation-related problems in power plants and other industrial facilities. Studies show that as many as two-thirds of all control loop oscillations are caused by control valve or damper problems. Effective combustion of fossil fuels such as oil, coal and gas requires good fuel-air ratio control. Poor combustion results in wasted energy and increased plant emissions.
Dr. Cheng stated, ‘Since fossil fuel will remain the world’s major energy source for many years to come, we believe improved combustion control has strategic importance to our nation and the world.’
How the Technology Works
Model-free Adaptive Control, as its name suggests, is an adaptive control method that does away with the tedious and time-consuming task of process modelling. This, in turn, means lower development costs and faster time to market for OEM suppliers, and lower operating costs and longer uptime for automation end-users. Thanks to MFA technology:
* Precise quantitative knowledge of the process is not necessary;
* Process identification mechanisms or identifiers are not included in the system;
* Controller design for a specific process is not needed;
* Manual tuning of controller parameters is not required; and
* Closed-loop system stability analysis and criteria are available to guarantee system stability.
The figure at the left illustrates the basic architecture of an MFA controller. An artificial neural network is used as a key component of the controller, but not as a model (NOTE: due to resolution limitations, only a portion of the neural network is reproducible in this web version.) It consists of one input layer, one hidden layer, and one output layer. Inside the neural network there is a group of weighting factors (wij and hi) that can be updated to vary the behaviour of the controller in real-time. The algorithm for updating the weighting factors is towards the goal of minimising the control error, which is the difference of the setpoint and process variable. Since this effort is the same as the control objective, the adaptation of the weighting factors can assist the controller in minimising the control error while process dynamics are changing.
The artificial neural network-based MFA controller also saves and processes a moving window of data that has valuable information for the process dynamics. In contrast, a digital version of the PID controller saves only the current and previous two samples. In this regard, PID has almost no memory and MFA possesses the memory that is essential to a ‘smart’ controller. [see reference 4]
1. Cheng, G. S., MFA in Control with CyboCon, CyboSoft, General Cybernation Group, Inc., September, 2008.
2. VanDoren, V., ‘Model Free Adaptive Control—New Technique for Adaptive Control Addresses a Variety of Technical Challenges,’ Control Engineering Europe, March 2001. For a web version of this article click here.
3. Cheng, G. S., ‘Model-Free Adaptive (MFA) Control,’ IEE Computing and Control Engineering, June/July 2004.
4. Liptak, B., Instrument Engineers’ Handbook - Process Control and Optimization, (Section 2.15, MFA Control by George Cheng). CRC Press LLC, October 2005.
Most Viewed Articles...