Introduction
Imagine this, a system which behaves depending on the variables surrounding it. A system that can modify its internal parameters to give the best possible output. A system that can maintain stability in the midst of constantly changing and unpredictable factors around it?
Sounds a lot like a Machine Learning problem doesn’t it? Well, what if I told you that this technology was put on paper way before Machine Learning was even a thing?
Adapt
It means to change a behaviour to better accommodate the surrounding circumstances. The idea for Adaptive Control was sparked in the early 1950s, during research and design for autopilot mechanisms, for supersonic aircrafts. Since then it has been implemented in various areas like Robotic systems, Navigation and altitude control for satellites, Autopilot, etc.
A Robotic Arm
A Supersonic Aircraft
Just like how a Neural network model learns parameters based on feedforward and feedback functionalities, an Adaptive control system detects changes in characteristics of the process, and compensates by modifying the controller parameters.
An Adaptive control system consists of a feedback loop and a parameter adjustment loop. The feedback loop contains the controller and the plant, and the parameter adjustment loop contains a parameter adjustment mechanism and a controller with adjustable parameters. The control differs based on the mechanism used to adjust the parameters. If you think about it, it is similar to Machine Learning!
Schematic Diagram of Adaptive Control System
In practical use, there are vastly 2 approaches:
Gain Scheduled Adaptive Control
Gain Scheduling is similar to a feed forward network because controller parameters are adjusted based on information prior to functioning. Due to this feedforward nature, the correctness of the adaptation cannot be measured.
Schematic Diagram for Gain Scheduled Adaptive Control
Self Adaptive Control
In this method, feedback loop performance is measured, and is optimized by adapting controller parameters. It is analogous to a feedback system.
This method is used in Model Reference Adaptive Control Systems (MRAC), Self Tuning Regulators (STRs), etc.
Schematic Diagram of MRAC
Schematic Diagram of STR
Applications
Early control methods were susceptible to the controller being unable to perform in nonlinear process conditions. They were also affected by natural disturbances, and changes in input nature. Adaptive control proves effective against these issues
Today Adaptive control has applications in various fields. Missiles make use of Self-oscillating adaptive controllers. STRs have industry related applications. Adaptive autopilots now exist, which outperform PID based autopilot systems. Adaptive control is used in high performance systems, for various industries and chemical plants.
Conclusion
If you think about it, Adaptive control is a lot like Natural Selection. Changes in natural variables lead to changes in the organism’s body and behaviour, which is what we call Evolution. Similarly, changes in process and environmental variables lead to change in controller parameters and behaviour. In a sense, Adaptive Control can be considered to be a “Darwinian” approach to control systems, a way of giving Systems a sense of “Evolution”.
Though this might be a stretch, it is still a nice thought, how much of today’s technology is inspired by, or resembles natural processes.
An article by Harish Bachu, 3rd Year Electronics and Communications Engineering.
