David Martinelli

Optimization of railway operations using neural networks

Railroad operations involve complex switching and classification decisions that must be made in short periods of time. Optimization with respect to these decisions can be quite difficult due to the discrete and non-linear characteristics of the problem. The train formation plan is one of the important elements of railroad system operations. While mathematical programming formulations and algorithms are available for solving the train formation problem, the CPU time required for their convergence is excessive. At the same time, shorter decision intervals are becoming necessary given the highly competitive operating climates of the railroad industry. The field of Artificial Intelligence (AI) offers promising alternatives to conventional optimization approaches. In this paper, neural networks (an empirically-based AI approach) are examined for obtaining good solutions in short time periods for the train formation problem (TFP). Following an overview, and formulation of railroad operations, a neural network formulation and solution to the problem are presented. First a training process for neural network development is conducted followed by a testing process that indicates that the neural network model will probably be both sufficiently fast, and accurate, in producing train formation plans.

Universal Pavement Distress Evaluator Based on Fuzzy Sets

Setting priorities for pavement maintenance and rehabilitation depends on the availability of a universal scale for assessing the condition of every element in the network. The condition of a pavement section has traditionally been assessed by several condition indexes. The present serviceability index (PSI) is one common evaluator used to describe the functional condition with respect to ride quality. Pavement condition index is another index commonly used to describe the extent of distress on a pavement section. During the decision-making process, both classes of indexes are needed to evaluate the overall status of a pavement section in comparison to other sections in the network. Traditionally, regression techniques were used for the development of functions that relate condition indexes to the information recorded in the pavement management database. This approach produces mathematical functions that are limited to a particular database. The functions so developed may also suffer from inaccuracies due to errors in data collection and recording. There is a need for a more generalized approach for the evaluation of pavement conditions to enable efficient management of large transportation networks. The development of a universal measure capable of formally assessing the condition of a pavement section within the universe of pavement conditions is described. This is accomplished by the fusion of a set of fuzzy membership functions that describe different parameters in the database with the perception of each parameter’s significance. The model output is the fuzzy distress index (FDI), which combines the extent of structural distress with traditional performance parameters such as roughness to describe the overall status of the pavement section. The behavior of FDI over time is examined for a random sample of pavement sections and is compared with trends in the corresponding PSI values (PSI was used only because it was readily available in the database). The results indicate that the flexible, universal FDI is a consistent and accurate measure of the overall pavement condition. The set of generated membership functions describing the different extents of every distress type can be easily standardized over the 50 states, allowing the model to be implemented on any pavement at any location. Also, the parameter weights used in the assessment may be easily adjusted (increased or decreased) to reflect changes in maintenance policies or budget availability at the local, state, or national decision-making level. Moreover, the concept allows for the omission of any number of parameters that might not be available in a particular pavement management database.

Dispelling the Myths of Holistic Engineering

Over the past year, articles by Grasso and others – including Holistic Engineering and Holistic Engineering and Educational Reform – have followed former IEEE President Joe Bordogna in adopting the term “holistic engineering” to describe a more cross-disciplinary, whole-systems approach to engineering education. It is an approach that emphasizes contextualized problem formulation and encourages innovative changes to traditional engineering training.

Developed Incident Detection Algorithm Compared with Neural Network Algorithms

The CUSUM (cumulative sum of log-likelihood ratio) algorithm is an optimization-based algorithm that is attractive for many applications because it can minimize detection delay and can explicitly incorporate the characteristics of processes before and after changes. One such application is freeway incident detection, where field-measurable traffic-flow parameters are used to flag incidents in real time in an expedient and reliable manner. In the presented study, the special characteristics of traffic processes associated with incidents are incorporated into the CUSUM algorithm for freeway incident detection. In the algorithm evaluation, the most recently developed neural networks are compared with an enhanced CUSUM algorithm. The neural network algorithms are systematically evaluated first among themselves, and then the best of them is compared with the CUSUM algorithm. The results demonstrate that the CUSUM incident detection algorithm can perform better than the neural network algorithms. The neural network algorithm may show inferior performance because it cannot adjust its decision threshold in real time.

Assessment and mitigation measures for graffiti on highway structures

Highway structures are public works facilities that are inherently accessible, to a certain degree, to the general public at all hours of the day and every day of the year. As a result, some highway structures are susceptible to graffiti. Graffiti on highway structures is a significant problem throughout the United States. Not only is graffiti an eyesore to the traveling public, it presents a hazard to the perpetrator and a liability exposure for transportation agencies because highway structures span high elevations and are in close proximity to motor vehicle traffic. The most common methods for combating graffiti include washing the surface of the structure with high-pressure water sprays, repainting the surface, and sandblasting. Although each of these methods can, in most cases, effectively remove the graffiti, the solution is often temporary; more graffiti is likely to appear in the future at the same site. Further, these measures can be quite costly, especially if they have to be repeated on numerous occasions to remove recurring graffiti. Results of a comprehensive survey of transportation agencies are presented and analyzed. The survey was designed to assess the nature and extent of the graffiti problem as well as to identify some solutions to the problem and identify various preventive as well as removal techniques. The study focuses on current graffiti prevention and removal policies and various other graffiti-removal techniques that are undertaken by different state departments of transportation to mitigate graffiti problems in their states.

Performance Evaluation of Neural Networks in Concrete Condition Assessment

A neural network modeling approach is used to identify concrete specimens that contain internal cracks. Different types of neural nets are used and their performance is evaluated. Correct classification of the signals received from a cracked specimen could be achieved with an accuracy of 75 percent for the test set and 95 percent for the training set. These recognition rates lead to the correct classification of all the individual test specimens. Although some neural net architectures may show high performance with a particular training data set, their results might be inconsistent. In situations in which the number of data sets is small, consistent performance of a neural network may be achieved by shuffling the training and testing data sets.

Dynamic performance of composite pavements under impact

The importance of developing a deep understanding of the behavior of pavement layers under the action of dynamic loads, and the availability of cutting-edge computational and visualization technologies, led to the study presented in this paper. Explicit finite-element analysis was used to investigate the propagation of dynamic displacements induced in pavement layers under the action of an impact load similar to the one applied in a falling weight deflectometer test. The time-dependent dynamic response of a rigid pavement with straight asphalt concrete overlay was studied for two cases of unbonded and fully bonded interfaces between different layers. Significant differences in behavior were observed. Three-dimensional computer graphics and animation of the deformed model were used to display the propagation of vertical dynamic displacements through pavement layers. It was found that in the absence of a perfect bond between all pavement layers, the displacements measured on the top surface correlated little with the deformation measured in subsequent layers. In this case, a complicated pattern of behavior took place between the asphalt overlay and the concrete. The time histories of vertical displacements at selected surface locations and on the top and bottom of every layer were plotted. The plots revealed the existence of time shifts between the maximum displacements experienced by each layer, irrespective of the type of bond assumed between the interfaces.

Communication Strategies for State Transportation Research Programs

Transportation research is often open ended and difficult to measure in that its beneficiaries may not know how they have been served by research activities and results. This research project sought to (a) obtain feedback from Ohio Department of Transportation (Ohio DOT) constituents; (b) develop a strategic communication plan that supports Ohio DOTs overall mission and goals, keeping in mind research office resources; and (c) develop a communication template that other departments of transportation could model or use. To accomplish those objectives, existing knowledge, attitudes, and behaviors concerning Ohio DOTs research office were obtained through a number of surveys. Those surveyed included the Ohio general public, internal DOT constituents (technical liaisons, administrators and directors, district deputy directors, district research contacts, and FHWA regional center directors), and external DOT constituents (consultants, contractors, transportation committee legislators, and college civil engin...

Incorporating Neural Network Traffic Prediction into Freeway Incident Detection

Because of their superior capabilities in emulating nonlinear systems, neural network models have been applied to traffic prediction with various degrees of success. However, these neural network-based traffic prediction models have not been used for incident detection. On the other hand, it is expected that the performance of an incident detection algorithm can be improved if an advanced prediction model is used. The development of several traffic prediction models that were then integrated into incident detection algorithms is documented. The traffic prediction models were developed on the basis of three different choices of independent variables, whereas the incident detection algorithms used different decision functions. The results show that a good prediction model can improve the performance of an incident detection algorithm only when the decision function of the algorithm is appropriately chosen.

SPECIFICATIONS FOR AUTOMATED LICENSE PLATE READING EQUIPMENT FOR TRANSPORTATION PLANNING

Recent technological advances in computer hardware, software, and image processing have led to the development of an automated license plate reader (ALPR). This equipment was developed primarily for enforcement and security applications, such as monitoring parking garages or border crossings. Because license plate data are used in several transportation planning studies, ALPRs have the potential to increase the quality and efficiency of many typical activities of transportation planning agencies. The key performance attributes of an ALPR with respect to the specific needs of transportation planning are determined. The following general needs are investigated: (a) the specific license plate data requirements of transportation planning studies; (b) the effect of the equipment on traffic operations and safety; and (c) special equipment characteristics required because of the temporary nature of transportation planning and the constraints of transportation planning agencies. In addition, an existing ALPR was tested in situations likely to be encountered in transportation planning applications. Technical specifications for an ALPR for transportation planning are developed. These specifications can be used to provide motivation and direction for the future development of an ALPR for transportation planning. Finally, the technical challenges to developing the ALPR are discussed.