Statistical process control of manufacturing systems with correlated stages

Paul Francis Zantek, Purdue University

Abstract

Manufacturing systems typically contain processing and assembly stages whose output quality is significantly affected by the output quality of preceding stages in the system. Part One of this thesis proposes and empirically validates a procedure for: (1) measuring the impact of each stage's performance on the output quality of subsequent stages including the quality of the final product, and (2) identifying the stages in the manufacturing system where management and engineering should allocate process quality improvement efforts. A distinguishing feature of the procedure is that it utilizes the information provided by the correlations between product quality measurements within and across the stages of the system. The starting point of the procedure is a computer executable network representation of the statistical relationships between the product quality measurements. Execution automatically converts the network to a simultaneous-equations model and estimates the model parameters by partial least squares. We report some of the results from an extensive empirical validation of the procedure using circuit board production fine data from a major electronics manufacturer. The findings from this empirical analysis show that: (1) management and engineering can overlook important opportunities for quality improvement if they ignore the presence of interdependencies between stages, and (2) the use of conventional statistical process-monitoring procedures in multistage systems can lead to erroneous process-control decisions. Part Two proposes a new multivariate procedure for monitoring process and product quality in manufacturing systems with correlated stages. The procedure can be used not only to detect the presence of out-of-control process conditions but also to identify the stages in the system responsible for such departures. Another important feature of the procedure is that it is well suited for monitoring a large number of process performance and product quality measurements. We illustrate the procedure empirically using circuit board production line data from a major electronics manufacturer. Part Three generalizes the partial least squares algorithm utilized in Part One to allow for contemporaneous correlation between the disturbances of structural equations. This section also describes how the information provided by contemporaneous correlation among disturbance terms can help manufacturers identify sources of variation in product quality.

Degree

Ph.D.

Advisors

Wright, Purdue University.

Subject Area

Industrial engineering|Management|Statistics

COinS