Leveraging Machine Learning and Artificial Intelligence to Improve Roadway Grip Sensing During Winter Weather

Abstract

Road grip, or friction, is an important variable used to assess numerous implications of winter weather impacts on surface transportation infrastructure and operation. “Grip” represents how well a vehicle’s tires adhere to the road surface and can be a function of road surface temperature and road surface condition, among other variables. Moreover, grip influences vehicle handling, maneuverability, and stopping distance. Thus, grip can be used as a proxy to assess safety and mobility on roadways during adverse weather as well as to provide performance measures for transportation agencies about their winter maintenance operations. An important caveat is that grip observations can be temporally and spatially limited. The objective of this research was to model road grip during winter weather conditions to predict road grip where observations may not be available. This was accomplished using atmospheric and road-based observations in Colorado and Minnesota in the U.S., to deduce road grip conditions to develop state/location-specific grip machine-learning-based models. These algorithms can produce an estimate of grip at locations that are not already equipped with grip sensors. The prediction accuracy is improved when road state and precipitation information are available. Road weather observations—including air temperature, road surface temperature, dew point temperature, relative humidity, road surface condition, road surface state, and precipitation information—can be used to derive an accurate grip model. This framework may further leverage mobile and connected vehicle grip observations to advance grip modeling between fixed locations.