Car-pedestrian Collision Reconstruction On The Basis Of Finite Element Simulation And Genetic Neural Network

The wide popularization of such new automotive technology as anti-lock braking system(ABS) and electronic stabilization program(ESP), lack of protective measures as well as other situations might lead to difficulty in the collection of relative evidences and traces at the scene of a car-pedestrian accident. Traditional reconstructions based on the braking marks and pedestrian throw distances are increasingly under restriction. In the face of enhanced public legal awareness and the great number of car-pedestrian accidents, how to make use of advanced technology and potentially useful information like pedestrian injuries and event data recorder(EDR) in collision reconstruction become practically valuable and theoretically significant in those ways such as improving analysis ability, ensuring the impartial handling and instructing the prevention of accidents.Studies have shown that injuries associated with car-pedestrian accidents can provide such kind of reference information as contact position and impact for the reconstruction. The injury information of the pedestrian is more easily reserved than traces at the scene. Therefore, body injury is more and more likely to become important evidence for an accident reconstruction. This thesis aims to explore in a new method of car-pedestrian collision reconstruction on the basis of a systematic research and analysis based on the characteristic of head and thoracic injury.Main research methods and conclusions:(1) Characteristic analysis of pedestrian head and thoracic injury under different impact parametersCharacteristics of real pedestrian head and thoracic injury are studied with epidemiology and orthogonal impact simulation from different contact angles(back, left, front, right) at different speeds(25, 32.5, 40, 47.5, 55 km/h) is analyzed through using Hyperworks and LS-DYNA on the basis of THUMS4.0 pedestrian models. Confirmatory analysis of the results of above researches is conducted to discuss the characteristics of pedestrian head and thoracic injury in biomechanical response, injury severity and risk under different impact parameters. The research outcome indicates that:In a car-pedestrian collision, the higher the speed of the car is, the more severe the head and thoracic injury of the pedestrian is inclined to be and the pedestrian is more likely to suffer skull fractures, brain injuries, rib fractures and soft tissue injuries of chest; when the pedestrian is hit from his/her back, the severity and risk of skull fractures and brain injuries are greater than when he/she is hit from the side, which is in turn greater than from the front; when the pedestrian is hit from his/her front, the severity and risk of soft tissue injuries of chest are greater than when he/she is hit from the back, which is in turn greater than from the side; when the pedestrian is hit from his/her side, the severity and risk of rib fractures are greater than when he/she is hit from the front and the back.(2) Collision reconstruction method on the basis of neural network and genetic algorithmIn this thesis, a reconstruction method---the combination of the head and thoracic injury of the pedestrian, neural network and genetic algorithm based on finite element simulation of car-pedestrian collision---is proposed. In this method, the orthogonal method is employed to design the simulating experiment program. The quantification of the simulation results is achieved with evaluation indicators for injuries. Injury information after quantification is input as neural network models and the impact parameter as output to construct BP neural networks applied in the recognition and prediction of impact parameters; parameter optimization is worked out by optimization of BP neural networks after training is finished with genetic algorithm.According to analysis of the specific numerical examples, the maximum of relative error of neural network in the recognition and prediction of impact parameters is 21.48%, which is worked out on the basis of the head and thoracic injury information of the pedestrian. This value decreases to 9.90% after optimization. Furthermore, the applicable range of speeds for this model is 20~58 km/h.(3) Confirmatory analysis of reconstruction method and genetic neural network modelReconstruction method and genetic neural network model proposed in this thesis are employed in the analysis of two real car-pedestrian accidents with detailed injury information and accurate impact parameters. The relative errors of the two accidents in reconstruction of speed are respectively 5.56% and 12.24% while those in reconstruction of impact angle are respectively 2.22% and 6.11%. The reliability of the method proposed in this thesis is verified through finite element simulation with results of impact parameters as boundary condition and visual analysis of collision process in combination with contrastive analysis of qualitative and quantitative injury.