Model to predict dynamic performance of a tractor semi-trailer car-carrier

Abstract

A performance-based standards (PBS) framework evaluates actual on-road performance of a vehicle, allowing the length and mass of a vehicle to exceed prescriptive legislation, without compromising on vehicle safety and dynamic stability. This PBS approach is currently being piloted as a demonstration project in South Africa. As of June of 2018, 270 PBS vehicles are operational with a recorded 39% lower crash rate relative to conventionally-designed vehicles; testament to their improved safety. The PBS framework defines the safe performance envelope of vehicles but does not optimise their safety and productivity. The design process to achieve the optimal productivity of PBS vehicles is highly iterative. An initial design is evaluated using multi-body dynamics simulation. If the required PBS performance is not achieved, design iterations are made until the required PBS performance is achieved. The process is costly, time-consuming and computationally expensive. In this study, we simulate a range of tractor semi-trailer car-carriers representative of possible design configurations. Supervised machine learning techniques within H2O.ai driverless AI are used to develop prediction models for the low and high-speed PBS performance of a tractor semi-trailer car-carrier The vehicle design parameters that form the feature vector for each vehicle combination are chosen according to the results of previous studies which evaluated the impact of vehicle design parameters on vehicle dynamic performance. The number of design parameters is minimised to simplify the amount of input data required to train the vehicle performance models. The machine learning models for SRT, RA, HSTO, TASP, LSSP, TS, FS and STFD (PBS measures used to quantify vehicle safety) were accurately predicted for all configurations in the test dataset. The models for MoD, DoM and YDC (further PBS measures) were less accurate but produced a negligible number of false pass results where the absolute percentage errors were significant. It is envisioned that with further development and validation the simplified machine learning model will be used by the car-carrier industry to determine the preliminary PBS performance of their combinations before submitting the design for the final PBS performance assessment. Reducing or eliminating the iterative design process for optimal PBS vehicles will accelerate the design process of safer and more productive vehicles; leading to a reduction in the cost of transport in South Africa.