To get the desired flexibility, suitable controllers are required to facilitate tuning and optimization not just during installation and startup for various equipment configurations and loads. A controlled plant in which dynamics are initially uncertain or vary over time needs self-tuning adaptive controllers. Real-time adaptive control can achieve near-optimal performance for various equipment configurations and changing loads. This is different from traditional PID control, where controllers are designed—perhaps using auto-tuning—for a specific operation and then remain fixed during operation. It is also different from robust control, where controller design considers a priori information about plant uncertainties, often at the cost of performance.
Certain applications and niche markets have used adaptive control methods since the 1970s. Gain scheduling is one well-known method, where predesigned linear controllers are activated for various plant operating points. This method is used, for example, for aircraft control. Special applications in process control use adaptive features. However, adaptive control has yet to be embraced by the broader control industry, despite potential benefits.
Possible concerns relate to guaranteed system stability and performance when the controller changes without human supervision. Also, the fact that advanced technology can match, and even outperform, experienced engineers in tuning and optimizing complex machinery is not yet widely known.
Several adaptive methods have been developed and successfully tested in academia and in industry, in addition to those listed above. Collaborative efforts between industry and academia can help to identify and productize suitable adaptive control methods. Initially embedding adaptive features with familiar PID-type algorithms, implemented on field-proven controller platforms, can facilitate market introduction. This would increase the market acceptance for other, perhaps more advanced controller products, particularly in the machine control, process control and building control markets, among others.
There are encouraging examples. Commercial software products with good GUIs enable engineers to design, implement and analyze adaptive controllers for both continuous-time and discrete-time plants, including needed system identification capabilities. The idea is to perform and test many scenarios during off-line simulations for designing adaptive controllers prior to their installation and test on the physical system.
Adaptive, real-time servo control systems are commercially available for further improving CNC machine tools. Such improvements rely on the use and refinement of adaptive control, i.e., the automatic monitoring and adaptation of the machine in response to changes in operating conditions. There are control platforms for CNC machines that monitor operating conditions in real time and adapt critical cutting parameters to maintain consistent product quality and to protect machine tool and product from damage. For instance, the cutting feed rate can be adjusted in response to the measured spindle load.
The use of adaptive control is promising in cases where traditional controllers are insufficient to solve difficult control problems, or too costly or time-consuming to install and maintain such control systems manually. Ultimately, when the technical and commercial potential of adaptive control is widely accepted, engineers might no longer have to design and tune their control solutions for complex problems and machines. Instead, they could design controllers that find the best solution through adaptation. The potential cost savings and performance improvements are compelling.
Karl Mathia is principal engineer at Zitech Engineering (www.zitechengineering.com), an engineering consulting company in Menlo Park, Calif. He is also chair of the Santa Clara Valley Chapter of the IEEE Control Systems Society.
Controllers that find the best solution through adaptation offer compelling potential cost savings and performance improvements.
