(531c) Integrating Process Databases: Image Analysis in Data Analysis Perspective
It has been proven that latent variable methods, such as Principal Component Analysis (PCA) and Projection to Latent Structures (PLS), are powerful techniques for solving problems with conventional process databases (ones with numeric and/or nominal data). They have been successful in innumerable topics and applications, including process analysis for trouble-shooting, process monitoring (MSPC), process modeling, process control, and product design, to list just a few. Due to recent advances in sensor technologies however, the need for analyzing new types of data ? for example, images, spectra and acoustic and/or vibration signals ? and integrating them with conventional databases has emerged. Advanced signal processing techniques, such as wavelet and wavelet packet transforms, can extract useful information from such data and novel combination of them with latent variable methods enables us to integrate databases with different types of data easily. This article presents an overview of methods and applications for the following: (1) extraction of useful information from large process databases that consist of numeric, nominal and image data, and (2) use of the information for analysis, monitoring, control and optimization of processes and/or products. The methods will be illustrated with industrial applications in the following topics: (1) product grading ? (Dofasco and DuPont), (2) process monitoring and control (Agnico-Eagle), and (3) predictive modeling and optimization (GE Advanced Materials). In each of these problems, a unified framework is used to integrate conventional process databases with image data, and the framework can be extended to handle other types of data ? acoustic and/or vibration signals and spectra.