Researchers: Karl Johan Åström, Tore Hägglund, Jonas Hansson, and Kristian Soltesz.

Methods for automatic tuning of PID controllers were developed in the early eighties, and implemented in industrial single-station controllers and DCS systems. A main reason was the technology shift from analog to computer-based controllers and systems at that time, which made implementation of such tuning functions possible. These methods were limited by the computer power and the knowledge about PID design that were available at that time. Since then, the computational  power and the knowledge about PID design has increased, which provides the possibility to develop new tuning functions with better performance.

Within process industry, a large number of processes can be accurately modeled using simple models, i.e. SISO FOTD or SOTD, and there are efficent tuning rules for PID controllers that are based on these model structures.

We aim at developing a methodology for automatic tuning of PID controllers, using nonlinear feedback for idenfication input generation and optimization based methods for both process parameter identification and controller synthesis.

The main components of the auto-tuning algorithm are the following:

1. Generate identification input with little or none a priori system information
2. Transfer function parameter identification through optimization
3. Model verification
4. PID synthesis
5. Performance evaluation

As a case study, a modified version of the method has been applied in closed-loop controlled anesthesia. The closed-loop anesthesia project is described here.

Developed autotuners

In the PhD Thesis by J. Berner, three different versions of the autotuner are developed and described.

One version is called the Tau-tuner and conducts a simple experiment, using an asymmetric relay function as feedback. From the experiment the static gain and the normalized time delay of the system can be estimated, and from this a FOTD model can be achieved from analytical formulas. From the achieved model the parameters of a PI or PID controller is tuned by existing tuning rules.

The second version is called the NOMAD-tuner and it uses a similar but even shorter experiment, and then uses the experiment data to find a first or second order model with time delay using numerical parameter estimation methods. The PID controller parameters are then obtained by optimization methods. This version requires more calculations, but is more robust to disturbances, gives more accurate models and can be started without waiting for steady-state.

The third version is called the multi-NOMAD and is an extension of the NOMAD autotuner to multivariable (two-input two-output) systems.

All versions have been evaluated in simulations and on physical processes. The Tau-tuner has also been tried on an industrial air handling unit. Matlab/Simulink implementations of all three autotuners are available in this git-repo, and the Tau-tuner has also been implemented and tested in Modelica (link to MSc thesis) and JGrafchart (link to paper).

Two master thesis projects have been performed in collaboration with ABB, where versions of the NOMAD autoutuner has been developed and implemented in one of their control systems. The implementation has been tested both in simulations and on laboratory processes.