Improving B-WIM performance with big data and Artificial Intelligence

The increasing number of heavy goods vehicles (HVG), such as trucks, buses and the upcoming road trains and electronically controlled convoys, has a significant impact on road safety and infrastructure management. Overloaded vehicles are particularly problematic as they increase safety risks and accelerate the destruction of road infrastructure. Therefore, preventing and efficiently enforcing illegal use of the roads benefits many who use or manage the road infrastructure.

Preventing congestion starts with traffic monitoring, which involves collecting data on axle loads and gross vehicle weights, vehicle types and frequency of occurrence, etc. These results are key inputs for traffic studies and for designing and assessing the condition of existing road structures. Traditional static weighing, while providing the most accurate results, is expensive and inefficient for heavy traffic. As a result, we use Weigh-in-Motion (WIM) systems to collect information on traffic loads.

Most WIM systems are difficult to install, require traffic closures and usually damage the road surface. The only technology that avoids cutting sensor grooves in the road surface and does not require roadblocks during installation and maintenance of the equipment is the Bridge-Weigh-In-Wake (B-WIM) system. Unfortunately, the accuracy and reliability of the current B-WIM results do not meet the legal requirements set by the OIML (International Organisation for Metrology). As a consequence, they cannot be used for the purpose of penalising offenders and are limited to the collection of load statistics and only the detection of possibly overloaded vehicles for static weighing.

The primary source of error of B-WIM systems is the measurement principle: axle loads are not measured from the contact between the wheels and the sensors, but are calculated indirectly from the bridge deflections. The errors increase rapidly with the length of the bridge and with the presence of more axles on the bridge. In such situations, the B-WIM system can accurately measure the total mass of the vehicle, but often cannot correctly distribute it to the individual axles. Reliable and accurate measurement of the axle loads of all vehicles on most types of bridges therefore remains an open research question. The B-WIM community agrees that the conventional research approaches used over the last 30 years are exhausted and cannot deliver the significant improvements needed to prosecute offenders.

The project aims to address these shortcomings by using an alternative approach - big data and advanced artificial intelligence (AI) methods. We plan to:

1. Develop AI-based evaluation methods to obtain reliability estimates of current B-WIM heuristics. The use of a fully automated quality assurance (QC) system for 8-WIM results will significantly expand the possible applications of B-WIM systems.

2. Design and evaluate an innovative AI-driven approach (AI-BWIM) that will seek to improve the accuracy of measurements. The key to improvement is to determine more reliably the number and position of axles and the number and position of vehicles on the bridge. We intend to do this by combining deformation measurement data, B-WIM results and traffic camera photos. We expect to develop the first OIML-compliant B-WIM system that will identify at least 60% of the measurements with the required accuracy, a percentage that no other WIM system worldwide has achieved to date.

3. to validate the proposed methods with real measured data from the Slovenian road network and to create a public repository of B-WIM data that will allow other researchers to validate our findings and improve the B-WIM algorithms. The research community will consequently benefit from the opening of a new research area in B-WIM technology.