Implementation and design issues in large factor experiments
MetadataShow full item record
The Design of experiments can be described as a systematic approach used for investigating a system or a process from a stochastic point of view. These systems or processes often involve many factors or independent variables, and while designing experiments, changes are made to these systems or processes by changing the level of magnitude of the inputs. The "effects" of these input changes are observed on the outputs. The major step in a DOE is to identify the input and the response variables. There is always a trade off between the number of factors, the number of levels, the desired level of insight into the results, and the cost of carrying out the experiments. Thus the larger the number of factors the harder, longer and costlier is the DOE. The goals of a large factor DOE are to lower the number of trials with variations of multiple factors, finding the separation of effects due to individual factors and interactions and also keeping a vertical balance. There are many methods that help solve these large factor DOE problems. In the following thesis an attempt has been made to develop a detailed understanding of methods such as Latin Square designs, Plackett-Burman designs, Taguchi designs, ANCOVA, Central composite designs and Split plot designs which solve large factor designs very economically. These methods are used in varied industries to solve 14 to 400 factors, at a single level or in sub-levels according to the design objectives. This thesis is an extensive approach to study these methods and provide a recommendation based on the design objective, number of factors and level of complexity to solve the design. A further classification has been proposed for rating the design methods based on their advantages and disadvantages. Also, a comparative study has been carried out to find, analyze and describe the various software tools that help solve these large factor DOE challenges.