1999 Villanova University Computer Science Masters Program Expert Systems Course Project

by: E.J. Dougherty III :: www.august-design.com

Intermodal transportation involves the use of the multiple shipping modes (container vessel, train, airplane and truck) for cargo. There are three major intermodal equipment types: containers, chassis and trailers. A container (CN) is a structural unit for transporting commodities. A chassis (CH) is basically a frame with has wheels. A trailer (TL) is a container that has wheels. A container differs from a trailer in that it has corner fittings for securing it to chassis, railcars and ocean vessels. To move a container by truck a chassis needs to be mounted underneath the container. When intermodal equipment is shipped it is described as either moving COFC (container on flat car) or TOFC (trailer on flat car). A container by itself is COFC, where as a chassis, trailer and container mounted to a chassis are referred to as TOFC. Intermodal equipment can vary in length, width and height.

Intermodal railcars are special cars that can handle intermodal equipment. Intermodal railcars are defined by there AAR (Association of American Railcars) type. Currently there are approximately 150 AAR car types. The AAR car types are divided into 3 types: P (conventional), Q (lighter weight, low profile) and S (stack). The AAR car type defines the railcar’s physical characteristics, such as number of hitches, length, width and payload. A hitch can be thought of as a location on the railcar where equipment can be secured. For my project I focused on the most common AAR car types. I was able to cover 17 of the 150 AAR car types.

The process of loading equipment to railcars is time consuming and can be a complicated task. The relevant information that is needed to develop a railcar-loading pattern for equipment is: equipment major type (TL, CN, CH), length, width, weight, mode of the shipment (TOFC, COFC), and if the unit has a nose-mounted refrigeration unit (reefer) which might protrude beyond the profile of a standard CN or TL. For railcars one need to know: the railcar hitch count, length, width, maximum weight limit and equipment type that the car can handle (TOFC, COFC, or both). The person who is generally assigned this job is someone who has at least one year of experience at this position. Since there are about 150 different car types and multiple pieces of various equipment that can be loaded; there are many possible loading patterns.

An expert system (ES) would greatly improve the process of loading railcars in a number of ways. Intermodal terminals operate 24x7; an ES could provide an increased availability of the expertise around the clock. Once the knowledge base (KB) is created, the expertise is permanent; which ensures that the knowledge will always be around even if the terminal’s expert leaves. The ES could provide a faster and reliable response than that of the human expert to produce the railcar loading patterns. This would allow more time and less loading errors for the hostler and lift operators and would reduce the possibility of any train delays. Multiple expertise can be captured to cover all of the 150 different AAR car types and their loading patterns. The ES would reduce the cost to run the terminal by allowing less experienced workers use the ES to generate the proper loading patterns.

The ES could also make the process automated by interfacing to a database and retrieve the available railcars and equipment to be loaded. The same ES could also act as an intelligent tutor to train new hires and other terminal personnel such as the track inspectors. The ES could be expanded to be used as a simulation tool to project the amount work that is needed to be done under a certain set of circumstances. For example UPS traffic will increase 40% during the holidays, how more resources will be needed to handle the new increase in UPS service without delaying the trains?

The problem domain of loading intermodal railcars is an appropriate paradigm for an ES. The domain is well bounded; the only concern is loading equipment to intermodal railcars. The problem cannot be readily solved algorithmically; the solution more heuristic. There is a definite need for this technology. And there are experts who are willing to cooperate and can explain the knowledge explicitly.

A rule-based ES seems to complement the process of loading intermodal railcars. The loading process can be broken down into two groups of rules: car type rules and hitch rules. The car types rules contain the broad loading rules such as width and weight limitations and if the car type only accepts TOFC or COFC equipment. The hitch rules are more specific; they involve length limitation, whether a nose-mounted reefer is allowed, kingpin limitation if the hitch accepts TOFC equipment and if an adjacent hitch needs to be covered. Even during the interviews with the expert, the knowledge was being conveyed as declarative and procedural statements.

An enhancement to the ES might be to provide optimal loading patterns for multiple railcars. Additional features could be added to include customer priorities, blocking and route information. For example, UPS is CSX’s most important customer, so the terminal manager would want as many UPS units included in the loading pattern for the current train. Blocking is grouping units together with similar destinations based on the train profile block. For example, the Chicago block contains city destinations such as Chicago, IL and St. Louis, MO; the ES would try to group equipment destined to these cities together. Blocking the train eliminates the time and effort needed to switch the railcars around at intermediate terminals. Route information would allow to ES to be aware of any height and length restrictions for a particular train route. For example, the Philadelphia terminal receives stack cars but because of the railroad tunnels height restrictions, equipment cannot be loaded to the top tier of the stack cars.

In conclusion, the ES developed under this project is an excellent example of an ES that can be implemented for a practical application and provide significant improvement over the exiting process.