Suggests Corrective Options
University Park, Pa. -- 150 Penn State engineers have developed a computer-based technique that not only reliably predicts the probable number of accidents due to wet pavement on a particular section of highway, but also can suggest corrective actions to improve safety.
Dr. Bohdan Kulakowski, professor of mechanical engineering and director of the University's Pennsylvania Transportation Institute (PTI), says, "With this approach, state departments of transportation may eventually have a tool to help identify which road sections to fix first in order to get the best return on safety and economic investment. They could focus resources on eliminating problems where the risk is highest."
Currently, highway departments most often resort to fixing only the road sections with the highest accident toll. With the new Penn State approach, accident-prone locations could be identified early before the road section climbed into the "most accident" category.
In addition, the new approach can identify the most effective remedial strategy for a particular problem location.
The technique was detailed in a paper, "Prediction of Risk of Wet Pavement Accidents Using Fuzzy Logic," presented today (Jan. 13) at the Transportation Research Board annual meeting in Washington, D. C. Jie Xiao, doctoral candidate in mechanical engineering, presented the paper. Her co-authors are Kulakowski, who is her adviser, and Dr. Moustafa El-Gindy, director of PTI's Crash Safety Program.
Xiao explains that actual accident data from 123 locations on the Pennsylvania Highway system between 1984 and 1986 were used in the study. Accident predictions made by standard techniques and those made via the Penn State researchers' new approach were compared with the actual accident data.
The researchers found that standard mathematical techniques, including linear and nonlinear regression models as well as probabilistic models, often failed to predict the number of accidents with any accuracy. The new approach, they say, is clearly superior.
"Significant social and psychological factors, for example, driver inattentiveness and road rage, as well as random events, affect the number of accidents," Kulakowski says. "Approaches to prediction that are totally quantitative, such as the regression and probabilistic models, can't take into account subjective factors."
The Penn State group based their new approach on "fuzzy logic," which provides a way to translate a verbal description about accident risk into a mathematical framework that can be used as the basis for computerized decision-making. For example, using fuzzy logic, the researchers can base a computerized decision making process on if/then statements such as "If skid resistance is high and pavement wet time is short, then the risk of skidding accidents is low. Or, "If driving difficulty is high and vehicle speed is high, then the risk of skidding accidents is high."
The new fuzzy logic approach doesn't exclude quantitative information. It does use skid number, posted speed, average daily traffic, percentage of wet time and driving difficulty as inputs to predict the number of wet pavement accidents. However, the new approach is the first to also take into account if/then inferences based on quantitative data.
To use the new computerized approach to develop recommendations for the most efficient and effective ways to improve safety, the researchers vary the inputs on roadway or traffic conditions one variable at a time and observe the outputs from the computerized process. For example, at one location the researchers analyzed, the safety conditions could be improved by almost 60 percent if the skid number was increased from 33.4 to 48. The same location could be improved by almost 90 percent, according to the researchers' calculations, by decreasing the driving difficulty.
The Penn State researchers note that after examining each of the variables, "it is not difficult to choose an effective and efficient way to improve the safety condition at that location."
Kulakowski notes that the predictions based on the fuzzy logic approach did not always exactly match the recorded accident data. "At one or two sites, our approach predicted a higher number of accidents than were actually recorded. However, when we checked the history of that site, we found that there had been a higher number of accidents in the past," he says.
The researchers conclude, "Fuzzy logic is a promising methodology for predicting and preventing the occurrence of wet pavement accidents." They suggest specific lines of further research to maximize the approach.
EDITORS: Dr. Kulakowski is at 814-863-1893 or btk1@psu.edu by e-mail.