Data Mining - Mehmed Kantardzic [285]
B.5 DATA MINING IN SCIENCE AND ENGINEERING
Enormous amounts of data have been generated in science and engineering, for example, in cosmology, molecular biology, and chemical engineering. In cosmology, advanced computational tools are needed to help astronomers understand the origin of large-scale cosmological structures as well as the formation and evolution of their astrophysical components (galaxies, quasars, and clusters). Over 3 terabytes of image data have been collected by the Digital Palomar Observatory Sky Survey, which contain on the order of 2 billion sky objects. It has been a challenging task for astronomers to catalog the entire data set, that is, a record of the sky location of each object and its corresponding classification such as a star or a galaxy. The Sky Image Cataloguing and Analysis Tool (SKICAT) has been developed to automate this task. The SKICAT system integrates methods from machine learning, image processing, classification, and databases, and it is reported to be able to classify objects, replacing visual classification, with high accuracy.
In molecular biology, recent technological advances are applied in such areas as molecular genetics, protein sequencing, and macro-molecular structure determination as was mentioned earlier. Artificial neural networks and some advanced statistical methods have shown particular promise in these applications. In chemical engineering, advanced models have been used to describe the interaction among various chemical processes, and also new tools have been developed to obtain a visualization of these structures and processes. Let us have a brief look at a few important cases of data-mining applications in engineering problems. Pavilion Technologies’ Process Insights, an application-development tool that combines neural networks, fuzzy logic, and statistical methods has been successfully used by Eastman Kodak and other companies to develop chemical manufacturing and control applications to reduce waste, improve product quality, and increase plant throughput. Historical process data is used to build a predictive model of plant behavior and this model is then used to change the control set points in the plant for optimization.
DataEnginee is another data-mining tool that has been used in a wide range of engineering applications, especially in the process industry. The basic components of the tool are neural networks, fuzzy logic, and advanced graphical user interfaces. The tool has been applied to process analysis in the chemical, steel, and rubber industries, resulting in a saving in input materials and improvements in quality and productivity. Successful data-mining applications in some industrial complexes and engineering environments follow.
Boeing
To improve its manufacturing process, Boeing has successfully applied machine-learning algorithms to the discovery of informative and useful rules from its plant data. In particular, it has been found that it is more beneficial to seek concise predictive rules that cover small subsets of the data, rather than generate general decision trees. A variety of rules were extracted to predict such events as when a manufactured part is likely to fail inspection or when a delay will occur at a particular machine. These rules have been found to facilitate the identification of relatively rare but potentially important anomalies.
R.R. Donnelly
This is an interesting application of data-mining technology in printing press control. During rotogravure printing, grooves sometimes develop on the printing cylinder, ruining the final product. This phenomenon is known as banding. The printing company R.R. Donnelly hired a consultant for advice on how to reduce its banding problems, and at the same time used machine learning to create rules for determining the process parameters (e.g., the viscosity of the ink) to reduce banding. The learned rules were superior to the consultant’s advice in that they were more specific to the plant where the training data was collected and they filled