Contents lists available at ScienceDirect Automation in Construction journal homepage: www.elsevier.com/locate/autcon Detect and charge: Machine learning based fully data-driven framework for computing overweight vehicle fee for bridges Osman Erman Gungor a, ⁎ , Imad L. Al-Qadi a , Justan Mann b a Illinois Center for Transportation, University of Illinois at Urbana-Champaign, 1611 Titan Drive, Rantoul, IL 61866, United States of America b Illinois Department of Transportation, Hanley Building, 2300 S. Dirksen Parkway, Springfield, IL 62764, United States of America ARTICLEINFO Keywords: Overweight fee Bridge Machine-learning National bridge inventory Weigh-in-motion data ABSTRACT Thisstudydevelopsafullydata-drivenframeworkforcomputingoverweightvehiclefeethatcombineshistorical bridge data from National Bridge Inventory (NBI) and weigh-in-motion (WIM) data. In this framework, in- formation regarding vehicle weight distribution on bridges was obtained using Gaussian mixture model (GMM) based interpolation. Using this interpolation approach, the vehicle weight distribution on each bridge could be estimated from WIM data based on their location. Later, these estimated distributions were combined with the NBI for developing a machine learning-based prediction model that inputs bridge characteristics (e.g., age and trafc) and outputs deck condition. The model was employed to calculate the expected bridge service life under two scenarios to compute a bridge life reduction per damaging load. Finally, the bridge life cycle cost was conducted to convert the calculated service life diference into a fee. Integration of this framework with existing geographicalinformationsystembasedonlinepermitissuingtoolswillallowfordetectionofbridgesonvehicles' routes and charge them a fee considering their weight and the load capacity of the bridges they will pass over. Therefore, fees will be calculated more accurately and efciently. Additionally, the proposed framework has the fexibility of being converted into a table for conforming to the conventional permit fee calculation scheme. 1. Introduction The U.S. population and economy exhibited a signifcant growth between 2000 and 2014. According to the U.S. Department of Transportation Freight Fact reports [1], while the population grew by 13% during that time, climbing to an estimated 319 million in 2014, gross domestic product increased by 24.9% in real terms (infation adjusted), reaching $15,773,516 (millions of chained dollars). This expansion in the economy and population caused a concurrent increase in truck freight transportation which carried 69.6% (by ton) of total goods moved in 2013. Moreover, a total of 13,732 million of tons of goods valued at $11,444 million shipped by trucks in 2013, re- presenting9.21%and6.16%increaseovertheestimatesof2007byton and value, respectively [1]. Well-maintained and functional transportation infrastructures are instrumental to sustain this growth in economy and to provide safer mobility for the increasing truck trafc. Nevertheless, with 20% of roadway miles in poor or mediocre conditions and 9.1% of bridges being structurally defcient or functionally obsolete [2], state depart- ments of transportation (DOTs) face a major challenge in meeting their infrastructure needs. Therefore, DOTs have become more interested in initiating and supporting research projects that try to address chal- lenges in transportation infrastructure management. This research addresses one of these challenges under a project supported by Illinois Department of Transportation: quantifcation of the damage on bridges caused by overweight vehicles. In this research, a fully data-driven framework, “Detect and Charge”, is developed to assesstheeconomicimpactofthevehiclesthatviolatefederallydefned weight limits. Thereby, a permit fee that compensates for the damage imparted on bridges by such vehicles can be established. In the State of Illinois, the overweight limits for a group of two or more consecutive axles are calculated by the following formula (Eq. (1)): = + + W LN N N 500 1 12 36 (1) where W, overall gross weight of any group of two or more consecutive axles, to the nearest 500lb. L,distanceinfeetbetweentheextremeofanygroupoftwoormore consecutive axles https://doi.org/10.1016/j.autcon.2018.09.007 Received 17 April 2018; Received in revised form 5 September 2018; Accepted 11 September 2018 ⁎ Corresponding author. E-mail address: gungor2@illinois.edu (O.E. Gungor). Automation in Construction 96 (2018) 200–210 Available online 28 September 2018 0926-5805/ © 2018 Elsevier B.V. All rights reserved. T