Search

This thesis explores the use of statistically valid process improvement tools in low-volume applications. Setting out the following research questions: How can the Six Sigma Measure and Analyse phases of a chronic quality problem be statistically validated in a low-volume process? How can a statistically valid approach for process control be implemented in a low-volume process? And how can this tool be extended to fit multivariate processes and can the calculation of control parameter adjustments be automated? In answer, the thesis presents an enhanced PROcess VAriation Diagnosis Tool (PROVADT) method, driving a Six Sigma improvement project through the Measure and Analyse phases. PROVADT provides a structured sampling plan to perform a Multi-Vari study, Isoplot, Gage R&R and Provisional Process Capability in as few as twenty samples and eighty measurements, making the technique suited to low-volume applications. The enhanced PROVADT method provides a Gage R&R without confounded variation sources, as was the case in the original method, and its practical application was demonstrated through two case studies. Process control tools for low-volume, high-variety manufacturing applications were developed. An adjustable traffic-light chart, with control limits linked to tolerance and simple decision rules, was used for monitoring univariate processes. This tool, the Set-Up Process Algorithm (SUPA), uses probability theory to provide 98% confidence that the process is operating at a pre-specified minimum level of Cp in as few as five samples. SUPA was extended to deal with high-complexity applications, resulting in multivariate SUPA (mSUPA). mSUPA maintains SUPA’s principles, but presents the information about multiple process features on one chart, rather than multiple univariate charts. To supplement the mSUPA tool, a theoretical method for calculating optimal process adjustment when a multivariate process is off-target was introduced, combining discrete-event simulation and numerical optimisation to calculate adjustments.

The work develops the application of evolutionary algorithms in the domain of automotive heat exchanger design. The principal employed is that of computer regulated changes to a physical model which attempts optimisation using methods analogous to biological evolution. It shows that the use of airside fins with differing louvre angles can enhance the performance of automotive heat exchangers by a useful margm. This has been achieved with a wind tunnel model that allows automatic configuration of the louvre angles, and novel instrumentation allowing heat transfer performance to be assessed in terms of shear and drag forces acting on the louvres. During the investigation an important coupling between the behaviour of adjacent louvres was discovered, manifested as a loss in useful shear force at around ±12° relative angle. The work as a whole shows that specific louvre angle selections and quantitative estimates of the potential performance gains could be made with the following improvements to the physical model and search algorithms. The number of louvre rows should be doubled (to 16) to better represent typical matrices and the instrumented louvres should be centrally positioned in the air stream. Improved data filtering is required for reliable operation and the specific figure of merit has been shown to be an important factor in the optimisation process. A parallel area of application for the optimisation strategies was the solution of the Wilson plot problem. This represents a novel approach to the analysis of heat exchanger experimental test data where an alternative curve fitting and visualisation format allows more accurate models to be established. By these methods functions defining heat transfer coefficients for both sides of a heat exchanger may be determined that give a fit to experimental data to within less than 1. 5% on measured overall heat transfer coefficient.