|
Citation
|
Tamai S, Miyazaki Y, Yamamoto H, Yoshii K, Amamori I. 27th International Technical Conference on the Enhanced Safety of Vehicles (ESV); April 3-6, 2023; Abstract #: 23-0239, pp. 10p. Washington, DC USA: US National Highway Traffic Safety Administration, 2023 open access. |
|
Abstract
|
27th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Enhanced and Equitable Vehicle Safety for All: Toward the Next 50 Years
https://www-esv.nhtsa.dot.gov/Proceedings/27/27ESV-000239.pdf
An effective brain injury risk assessment is required to minimize the risk of brain injury from traffic accidents. Thus, anthropomorphic test devices (ATD) have been used for overall vehicle safety evaluation. The authors developed a novel ATD head that incorporates detailed intracranial structures with the brain and can measure the relative displacement between the brain and the skull. However, the strain inside the deep brain of the head surrogate cannot be measured or estimated. Although one can simulate the brain strain waveform using the finite element model, the computational cost is high, and the real-time evaluation of brain strain during crash tests is difficult. To compute the brain strain response in real-time, deep learning (DL) methods can be used to predict brain strain behavior. Therefore, this study aims to propose a method to predict the waveforms of maximum principal strain in the brain using a DLmethod called long short-term memory (LSTM). Reconstructed simulations for impact tests using a finite element head model were conducted to obtain the principal strain waveforms of the brain and construct a dataset for machine learning. The impact tests included 125 occipital head impact tests, 7 frontal sled tests, 35 vehicle frontal crash tests, and 53 American football impact tests, constituting a total of 220 head impact tests. Furthermore, the LSTM model wastrained on triaxial angular velocity and acceleration waveforms, and the models were constructed to predict the principal strain waveforms in the cerebellum, brainstem, and right and left cerebrums. Subsequently, to validate the predictive model of brain strain, CORAwas calculated as an index of the prediction error. The average CORA score between the brain strain waveforms predicted by LSTM and those of the dataset was 0.963 for occipital head impact tests, 0.928 for frontal sled tests, 0.898 for vehicle frontal crash tests, and 0.875 for American football tests. The occipital head impact tests, vehicle frontal crash tests, and frontal sled test cases were predicted with high accuracy. However, the football impact test cases were inferior to the other three test cases. The football impact cases included more multidirectional impact patterns and failed to learn similar collisions. However, an error in the waveform was observed in the rebound phase of the head impact in the latter half of the brain strain waveform. Therefore, the impact test datasetshould be expanded, including cases with rebound behavior of the head, and the set of features that can reflect the rebound behavior of the head should also be examined further. In conclusion, the maximum principal strain waveforms of the brain can be rapidly and accurately predicted from the angular acceleration and angular velocity of the ATD head in occipital head impact, frontal sled, and vehicle crash test cases using LSTM. This method enables real-time evaluation of brain strain waveforms during impact tests.
Language: en |