THESIS
2016
xiv, 94 pages : illustrations ; 30 cm
Abstract
In recent years, quality has been one of the most influential factors in modern business. There is an increasing demand on quality products from customers, and quality
improvement becomes the pursuit of any successful company. In the past few decades,
various approaches have been proposed for quality improvement. Among them, Statistical
Process Control (SPC) serves as the most useful tool, and it is widely applied in modern
manufacturing and service processes.
With the rapid development of technology, people are seeing an explosive growth on
the complexity of modern processes. Also, limited by inherent mechanism, some processes
present certain unique features. Both process complexity and unique features result in
an increasing variety of data types (data hybridity), as well...[
Read more ]
In recent years, quality has been one of the most influential factors in modern business. There is an increasing demand on quality products from customers, and quality
improvement becomes the pursuit of any successful company. In the past few decades,
various approaches have been proposed for quality improvement. Among them, Statistical
Process Control (SPC) serves as the most useful tool, and it is widely applied in modern
manufacturing and service processes.
With the rapid development of technology, people are seeing an explosive growth on
the complexity of modern processes. Also, limited by inherent mechanism, some processes
present certain unique features. Both process complexity and unique features result in
an increasing variety of data types (data hybridity), as well as the difficulties in process
control and monitoring. For example, in some complex processes, rather than single
quality index, multiple quality information are recorded; while some other processes can
only generate incomplete data, in which only part of the collected data are accurate. A
natural question is raised that how to make use of the multiple quality information or
inaccurate data for process monitoring. This thesis devotes some discussions for process
control under data hybridity, and it mainly considers two commonly seen processes - the
multivariate process and lifetime process.
Although some previous research works have been devoted for monitoring multivariate processes and lifetime processes, the problem associated with designing robust and flexible control schemes for both processes have yet to be fully addressed. In multivariate
processes, the challenges such as non-normality and multiple data types prevent conventional methods from working efficiently; and some existing methods for lifetime process
monitoring cannot deliver satisfactory in control (IC) and our of control (OC) performances. To overcome the above limitations, this thesis provides some novel methods as
well as useful insights for monitoring both processes.
In the thesis, two useful methods are suggested for monitoring multivariate process.
Both methods utilize some data mining techniques and make use of the historical OC
information. First, a support vector machine (SVM) model is set by combining the IC
and OC data, and the working status of the process is indicated by the probabilistic
output of the SVM model. The other method appears statistically appealing. A K
Nearest Neighbor (KNN) method is employed to transform the multivariate data, and a
CUSUM statistic is constructed based on the density of the transformed variable. In the
context of lifetime process detection, a newly developed weighted likelihood ratio test is
utilized, and a novel monitoring strategy that automatically combines the likelihood of
past samples with the EWMA scheme is suggested for building the control chart.
The three methods compose a systematic methodology for multivariate and lifetime
processes control. Some numerical simulations are provided to compare each method
with corresponding counterparts as well. In addition, real-data examples have demonstrated the effectiveness of these techniques. They can be implemented into practice with
reliability.
Post a Comment