Autotuning of PID and Learning Feedforward

Autotuning of PID controllers

The manual optimal tuning of PID control loops is a relatively labour intensive job, and requires expertise and experience. In practice PID loops are rarely tuned optimally. Moreover, to maintain optimal settings, the PID controller must regularly be retuned, due to changes in the system it controls.

DotX has developed code that automatically tunes PID controllers. The method is based on the auto-relay method.

When the PID Autotuner is activated, a feedback relay controls the output by sending out either a fixed positive value or negative value, depending on the control error. As a result, the controlled output (y) will show a stable oscillation with a limited amplitude. After only a few oscillations, the relay is switched off again and new PID settings are calculated from the oscillation’s period and amplitude, and these are applied to the PID controller.

DotX PID Autotuner has been fully tested in industrial practice.

Learning feedforward

Control errors can be reduced considerably by feedforward, but the design of a conventional feedforward controller can be complicated, and retuning may be necessary if the controlled system changes with time. If the disturbance cannot be measured, the application of a conventional feedforward controller is impossible.

DotX has developed Learning Feedforward code on the basis of Iterative Learning control. In this control technique, the feedforward controller learns from control errors made in the past. Due to the continous learning, the feedforward controller adapts to changes in the system it controls. It does not need a measurement of the disturbance signal itsself: it only needs a trigger bit (signal that is either 0 or 1). It should be noted that Learning Feedforward can also be applied to improve setpoint tracking.

The DotX Learning Feedforward controller has been fully tested in industrial practice.

The figure above shows measured trends of 16 occasions where the system was disturbed. The disturbance activity (yellow line) indicates when the disturbance is active (in which case it goes to 1), and is used by the Learning Feedforward controller for triggering. Hence, as soon as disturbance activity signal goes to 1, the Learning Feedforward starts 'to fire'. The controlled output is indicated by ‘y’, the manipulated input by ‘u’. Black lines show the average of 16 measurements. The left graphs (blue lines) show the result when Learning Feedforward was switched ON. The right graphs (red lines show) the result when Learning Feedforward was switched OFF.