TY  - GEN 
ID  - moewes_multivariate_2007
T1  - Multivariate data analysis of pharmaceutical spectrometric and process data
A1  - Moewes, Christian
Y1  - 2007
N1  - Internship Report, School of Computer Science, University of Magdeburg, Germany
KW  - Chemometrics
KW  - Multivariate Data Analysis
KW  - Partial Least Squares
KW  - Principle Component Analysis
N2  - This report is related to my project which I was working on during my internship with the Intelligent Vision and Reasoning Department at Siemens Corporate Research (SCR) in Princeton, NJ from March 6th until September 6th, 2006. Real world systems like the manufacturing of of pharmaceutical drugs, semiconductors, petroleum chemicals or metallurgy are much too complex to be understood down to the ground. That is why experiments with batteries of sensors are a necessity. Those detectors (spectrometer, thermometer, barometer, ...) are used for any kind of physical measuring value like spectra, temperature, pressure etc. Nowadays, the modern measurement technique makes it possible to collect a huge amount of data over time for every sensor. The need of model that is as easy as possible describing the behavior of this multivariate system is big. Nearly every theoretical approach to model a complex system has its limits. The idea of an empirical model is to observe intrinsic, latent variables in an indirect way. Every measurement contains defects like noise, uncertainties, vagueness. With the help of statistics one can induce knowledge from noisy data. So it is natural to use statistical methods to tackle noisy real-world applications. Multivariate models like PCA or PLS consider the joint effect of all measured variables that maybe reveals unrecognized features. They are also more resistant to noise than classical multivariate regression methods. My task was to extend the software CAP (System for Condition Assessment & Prognosis) in order to handle multivariate pharmaceutical spectral and process data which were ascertained by sensor measurement of bioreactors. I will give an insight into the software architecture of CAP and describe the utilised methods for multivariate data analysis (MVDA). Moreover I will outline my contributions for CAP which cover graphical user interfaces, hardware interfaces and data mining algorithms as well.
ER  -