When is a D-optimal design appropriate

There are situations when a D-optimal design is the preferred choice  

The available domain is truncated by constraints. A two-level design of any kind would necessitate the shrinking of the explored domain to accomodate the constraints. This would leave interesting parts of the domain uncovered by the design. A D-optimal design would select experiments to cover the whole domain of interest. 

Selection of complementary experiments to improve a model. When a series of experiments has been run to establish an initial model it is sometimes found that the model shows a lack of fit. If the experimenter can give a reasonable explanation for this, and he/she wishes to improve the model by adding some corrective terms, a D-optimal design can be used to select those complementary experiments which should be run to obtain maximum precision estimates of the parameters of the corrected model. 

It should be remembered that factorial, fractional factorial and Plackett-Burman designs are D-optimal designs. These designs can easily be constructed by hand. In 

A review on D-optimal regression designs has been given by St John and Draper.[5] 

---