This page describes work done for Dinah Maintenance of Way, Inc. in the summer of 1997 as part of an intership with the MCIM. The applets and documentation here describe a method for quickly identifying a rail shape and choosing the best combination of grinding patterns. This project was supervised by Dr. Fadil Santosa of UMN and Kenneth Iverson of Dinah. Note: To avoid random web searches by competitors, the name of the company has been changed to "Dinah". The actual name of the company can be seen in the top image at left.

Shape Matcher
Describes the ShapeMatcher class written for the Dinah C++ library.

Pattern Selector
Describes the PatternSelector class written for the Dinah C++ library.

The Genetic Algorithm
Describes the workings of the genetic algorithm. The javascript used to load the mathematical notation may not work on some machines.

Pattern Recognition and Decision-Making in Railroad Maintenance
A copy of my master's thesis in Microsoft Word 98 format. It's about 560 kB.

The Moments Applet (Old Version)
This is the old and uninteresting version of the moments applet. The newer version below does not seem to work on old versions of Netscape.

The Moments Applet
This applet tries to identify which representative rail shape best matches a given head profile. Works by calculating various moments and the curvature histogram.

The Genetic Algorithm
This applet illustrates the (in)effectiveness of the genetic algorithm in deciding which combination of pattern MR curves best matches a given MR curve.

The Comparison Applet
This applet compares the performance of four different algorithms in trying to determine which combination of patterns best matches a given MR curve. The four algorithms are the Brute-Force Search for 1 pattern (BF1), a successive BF1 or Greedy approach (Greedy), a Brute-Force Search on all combinations of 1, 2, or 3 patterns (BF123), and the aforementioned Genetic Algorithm (Genetic).