A framework for establishing an experimental design approach in industrial data mining
Abstract
This research was conducted due to a need in a specific industrial environment to provide a structured problem solving approach, which accommodate DOE within a framework, assisting analysts and management for strategic decision making for process improvement. Building an experimentation model within an industrial manufacturing environment is the base for the developed framework that consists of two main components; one being methodologies, showing the high level non-analytical portion of the framework, and two; selected statistical methods, including DOE, which represent statistical techniques for scientific data analysis following a sequential statistical technique selection for the data analytical process. The proposed framework and the sequential data analytical process is presented in the diagram below.
This framework is the blueprint for this research which was applied to a case study to evaluate the pragmatic relevance of this framework. Specific goals were set for the research that were aligned with the proposed framework. These goals are; To accommodate DOE as a Data Mining Technique in an Industrial Data Mining environment.
To enhance the awareness of expanding DOE as a statistical approach to complement existing methods and methodologies used for Data Mining.
To validate the integrity of captured data through the refining process to determine upper and lower operating conditions required by DOE, any abnormal data points will be exposed.
To focus on Industrial Data Mining, and concentrate on process data, applying DOE rather than generic, traditional Data Mining techniques.
To develop a methodology to accommodate the use of DOE as a Data Mining technique to determine impacts of variables on process outcomes through experimenting with data within current databases.
The main purpose by meeting the above goals at the end of the study shows that the analytical process illustrates:
That the proposed framework is generic, applicable to this case study and for any data analytical process.
That the focus is on process improvement with experimentation as a process improvement basis.
Shows an alternative perspective to data analytics by utilizing historic data within databases by applying experimentation to reduce the impact of experimentation cost.
Applying the proposed framework for process optimisation studies in any company where needed should enhance process improvement, because this research is about following a new experimental analysis design approach that is generic for any process development and improvement, irrespective of the product rendered. The framework and techniques used in this research are applicable within any processing plant where multiple variables affect product quality. The proposed model for process development could not be tested because the company has shut its operations in South Africa but the concept for the proposed DOE methodology proved to be representative for the period upon which the model was developed and tested, based on all the different the comparative results between the predictive model and the validation period.