An approach to generalized analysis of automated manufacturing systems through classification

Catherine Mary Harmonosky, Purdue University

Abstract

This research considers a general approach to the analysis of automated manufacturing systems, using the concept of a system classification scheme as an underlying structure for generalized analysis. Generalized analysis will allow results to be transferred and applied successfully in similar system environments, which would provide important information applicable early in the design of systems. Previously, results of analysis have been linked directly to unique systems. This work develops analytical techniques that facilitate determining when analysis results may be generalized from the study of one system to some other system. A hypothesized classification scheme is presented as a framework for development of generalized analysis techniques. A computer program was developed to generate testbed systems belonging to a particular class. This system information is used in a simulation program to obtain performance measure data for analysis. The similarity of several systems of the same class is tested on two aspects, the mean system performance and the system response to changes in the environment. To compare mean performance measures among systems, a simulation was allowed to run until the designated performance measure was within 2.5% of its true value, based on confidence interval construction using independent batch means. The range of performance measure values among eight systems in each of three conjectured classes indicates similarity does exist. To study system response behavior to changes in designated system parameters, an analysis method is developed utilizing covariance analysis. The regression equations discerned for each system also provide insight into parameters' influences upon the performance measures. Two hypothesized system classifications were used to discuss this method of testing system response. For one class with low congestion, system similarity was indicated suggesting an acceptable classification definition. For a more congested class, some variability among system responses was observed, suggesting refinement of the classification in this case may be appropriate. It was also noted that different parameters influenced different performance measures, stressing the importance of early knowledge of the selected performance measure by which a system's operational success will be judged.

Degree

Ph.D.

Advisors

Sadowski, Purdue University.

Subject Area

Industrial engineering

COinS