REPOSITORY > RESULTS

Abstract

Unexpected production breakdowns and unpredictable defects in product quality remain
critical problems in modern manufacturing systems. Usually occuring by no obvious
exterior cause and often during periods of smooth production, they lead to production
losses and reduced process availability, which results in lower profit rates. Conventional
statistical methods of process control are completely unable to provide adequate warnings
and examine the causes, since they are focused on individual output variables and largely
ignore the inherent relationship between process variables.
In quest of improved competitiveness, the European manufacturing sector is
attempting to shift the production and business paradigm towards innovative model-based
solutions that make use of the large amount of process data generated daily. Improved
insight enables more predictable performance of large scale manufacturing processes.
In this spirit, the aim of this project is to build up a prototype system for surveillance
and control of product quality in complex manufacturing processes. The novelty of the
approach is based on the integration of modern e-manufacturing solutions for accurate
acquisition of production data and advanced regression and classification methods for
model synthesis, data interpretation and decision support. The modeling approach is
based on statistical regression with correlated data. Since the outcomes of some of the
performance indicators (product quality, availability of equipment) reside inside {0, 1}
interval, the present dissertation is concerned with the logistic regression within the
context of correlation between starting materials and technological parameters of the
production process. Developed statistical models are used for the perception of disparities
in measured data, which is the basis for timely problem detection. The concepts shown
in the underlying thesis were tested on simulated examples and real production data.