Traffic control centre optimisation through applied intelligent weigh-inmotion

Abstract

Transport logistics are of utmost importance in all countries because they ensure that goods are efficiently delivered to businesses and thus assist in trade facilitation. A country such as South Africa, where the primary economic hub is landlocked, requires road freight transport of these goods in the forms of trucks. Freight transport by road is regulated to ensure safe transport and to protect the road infrastructure because it is one of the largest resources a country requires for GDP growth (it is a South African asset valued at South African rand (ZAR) 1 trillion. Mass regulations specify maximum permissible axle loads, total gross vehicle mass and a minimum drive axle mass. Vehicle compliance is tested by means of Overload Control Centre (OCC)s, primarily known as Traffic Control Centre (TCC)s, strategically situated on the freight corridors. A typical setup has a weigh-in-motion (WIM) scale that directs a vehicle to a static scale if its mass falls within a threshold. The static scale determines whether the vehicle is overloaded and, if so, the vehicle owner is prosecuted. Non-overloaded vehicles are allowed to return to the corridor and proceed on their journey. There are currently inefficiencies in this system because each TCC operates in isolation, not sharing information with the previous or next TCC. As a result, the same vehicle can be weighed multiple times on a journey, even if it is loaded legally. Current annual reports indicate that 75,95% to 78,05% of vehicles that are statically weighed are not overloaded. Being weighed increases delays and costs in their logistic supply chains. The proposal was to design a system that would remove the isolation of data between TCCs and implement an artificial intelligence model to predict whether a vehicle is overloaded and should be statically weighed or not. Successful implementation will result in much fewer statically weighed vehicles per annum, and decrease supply chain

turnaround times. A case study was conducted in the South African TCCs at Mantsole and Heidelberg. Layout and operational flows were investigated to ensure full insight into current operations. Interventions in the current system are proposed to ensure data sharing between stations. These interventions include collecting data from the current systems to use as inputs to the proposed artificial intelligence (AI)-based algorithm. Implementation of data sharing between stations will ensure that an intelligent decision can be made about whether a vehicle should be statically weighed while travelling along the freight corridor. In this discussion, the first TCC where the vehicle is statically weighed is called Station 1, while the second station, where an improved decision must be made, is called Station 2. The decision will be made using normalised to legal threshold mass from the WIM, actual vehicle combination mass (AVCM), and static scale measurements at Station 1. Algorithm values are normalised to legal thresholds to ensure the models will remain relevant if legislation changes or when applied in another country with different legislation. The next station uses a normalised travel time taken from benchmarked travel times combined with the WIM mass. As all the variables available to make the decision (overloaded or not) are noisy, a simple rule-based decision would make many mistakes. Since AI techniques have been found to improve the quality of decision-making when more than one noisy variable is available, we applied AI techniques to improve the quality of decisions resulting from a simple rule-based approach. Data was collected between Mantsole and Heidelberg TCC, accurately linked, and used as training data sets. The use of a random forest AI model reduced the percentage of non-overloaded vehicles that were statically weighed in the worst case from 76,64% (average) to 0,03% while sending 0% of overloaded vehicles incorrectly to the corridor. This model had slightly better performance when compared with the artificial neural network (ANN) implementation. Implementation of the model will therefore drastically reduce the number of legally loaded vehicles that are weighed statically.

The financial benefits of implementation would be considerable for several stakeholders. There would also be operational benefits for cargo owners, transporters, road users, clients, road agencies, cross border operations, TCCs, toll concessionaires, heavy vehicle operators and the National Treasury of South Africa. The potential benefits are compared to the operational costs of the proposed system for funding by the Treasury. The novel contributions of this work are, firstly, the operational and system design for data sharing between TCCs in South Africa. Secondly, the development of a data simulator and AI techniques for training with historic data to reduce the number of statically weighed vehicles without increasing the risk of vehicle overloading represent a major innovation in this field. Thirdly, an estimated economical cost-benefit model was developed to indicate the potential benefits to the sector if the concept is implemented.