Designing industrial experiments with restricted experimental resources
This thesis presents two procedures for designing industrial experiments with restricted experimental resources. First, a two-stage approach is developed using the D- and Ds-optimality criteria in combination, to focus information into the effects of most relevance to the robust engineering design of a product. The method is demonstrated by finding a design for a study on gas sensors. The second approach forms the major contribution of this thesis. Semicontrolled experiments are developed for use in situations where the factor values cannot be pre-determined, although the way in which they are combined with other factors can be controlled. This situation arises when the costs of measuring components in a product, or of making one-off components, prevents the use of a conventional factorial experiment. In the method, samples of the components are obtained and features of interest are measured. The components are then combined in products for testing to give combinations of the factor values which will maximise the information on the factorial effects under investigation. This method is particularly useful when derived factors, that is, factors defined as functions of variables from one or more components, are of engineering interest. An exchange algorithm is presented for finding optimal designs for semi-controlled experiments and demonstrated on a small pilot study on a gear pump. An extension of the algorithm is provided to design experiments when some factors can be pre-set in the conventional way, but others cannot. The issues of power and the size of experiment for a semi-controlled study are also addressed through simulated experiments.