English, 17 pages, 22 figures, 44 references

Collision reconstruction aims to represent the physics of vehicle dynamics and structures, along with driver input and occupant biomechanics, using a range of tools from simple physics calculations through to complex mathematical models. While vehicle design is increasingly Computer Aided Engineering (CAE) driven, the advanced modelling methods developed for CAE are not necessarily straightforward to employ in general collision reconstruction. CAE models are often dedicated to specific (laboratory) test conditions for pre-impact driving or purely for impact, while collision reconstruction often addresses complex real world cases requiring integrated simulation of pre-crash and crash events. Furthermore vehicle and restraint models developed for design may not be accessible to collision investigators.
The paper describes the recent development in our CAE simulation methodology: the use of DOE (Design of Experiments) and Multi-Objective Optimisation techniques together with MADYMO modelling in order to generate an understanding of a complex real world collision scenario. This case study involves a dual ejection single vehicle incident which was investigated by Advanced Simtech in the UK for Avon and Somerset Police during 2007. There was insufficient data to allow the case to be solved using other well-established police methods but our methodology and toolsets allowed the incident to be successfully reconstructed, securing a conviction as a direct result of our analysis. The reconstruction was conducted in two separate phases. The purpose of the first phase was to obtain an accurate motion for the vehicle during the event, using a vehicle dynamics reconstruction simulation that was correlated to the physical evidence and tyre markings left at the scene. This was followed by an occupant analysis phase which focussed on the opportunity of ejection for the people involved.
Both phases incorporate DOE techniques allowing us to fully understand the effect and influence of each of the input parameters involved.