Design of Plant Experiments

If the dead time is estimated inconeetly, the residuals will show an autocorrelation that falls outside the confidence limits. In addition, the residuals will not average around zero and the model fit will be poor. 

Identification results are shown in Fig. 24.11. Agreement with the original data is good and the three identified models give good predicted values. The model fit is 81.5%. 

The identification can also be done using the MATLAB graphical identification user interface. Different model orders and dead times can quickly be tried and results plotted. The models can be viewed using the LTl viewer. 

As a result, it is in most cases necessary to design a plant experiment to ensure that there is sufficient information in a collected data set. 

To generate an informative data set of a process, it will be necessary to vary the inputs of the process, thereby generating changes in the process output(s). 

---