ISSN: (Print) X (Online) Journal homepage:

Traffic Injury Prevention ISSN: 1538-9588 (Print) 1538-957X (Online) Journal homepage: https://www.tandfonline.com/loi/gcpi20 Development and Validation of the Total HUman Model for Safety (THUMS) Toward Further Understanding of Occupant Injury Mechanisms in Precrash and During Crash Masami Iwamoto, Yuko Nakahira & Hideyuki Kimpara To cite this article: Masami Iwamoto, Yuko Nakahira & Hideyuki Kimpara (2015) Development and Validation of the Total HUman Model for Safety (THUMS) Toward Further Understanding of Occupant Injury Mechanisms in Precrash and During Crash, Traffic Injury Prevention, 16:sup1, S36-S48, DOI: 10.1080/15389588.2015.1015000 To link to this article: https://doi.org/10.1080/15389588.2015.1015000 Masami Iwamoto, Yuko Nakahira, and Hideyuki Kimpara. Published with license by Taylor & Francis Masami Iwamoto, Yuko Nakahira, and Hideyuki Kimpara View supplementary material Published online: 01 Jun 2015. Submit your article to this journal Article views: 2302 View Crossmark data Citing articles: 16 View citing articles Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalinformation?journalcode=gcpi20

Traffic Injury Prevention (2015) 16, S36 S48 Published with license by Taylor & Francis ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2015.1015000 Development and Validation of the Total HUman Model for Safety (THUMS) Toward Further Understanding of Occupant Injury Mechanisms in Precrash and During Crash MASAMI IWAMOTO, YUKO NAKAHIRA, and HIDEYUKI KIMPARA Toyota Central R&D Labs., Inc., Aichi, Japan Received 14 November 2014, Accepted 31 January 2015 Objective: Active safety devices such as automatic emergency brake (AEB) and precrash seat belt have the potential to accomplish further reduction in the number of the fatalities due to automotive accidents. However, their effectiveness should be investigated by more accurate estimations of their interaction with human bodies. Computational human body models are suitable for investigation, especially considering muscular tone effects on occupant motions and injury outcomes. However, the conventional modeling approaches such as multibody models and detailed finite element (FE) models have advantages and disadvantages in computational costs and injury predictions considering muscular tone effects. The objective of this study is to develop and validate a human body FE model with whole body muscles, which can be used for the detailed investigation of interaction between human bodies and vehicular structures including some safety devices precrash and during a crash with relatively low computational costs. Methods: In this study, we developed a human body FE model called THUMS (Total HUman Model for Safety) with a body size of 50th percentile adult male (AM50) and a sitting posture. The model has anatomical structures of bones, ligaments, muscles, brain, and internal organs. The total number of elements is 281,260, which would realize relatively low computational costs. Deformable material models were assigned to all body parts. The muscle tendon complexes were modeled by truss elements with Hill-type muscle material and seat belt elements with tension-only material. The THUMS was validated against 35 series of cadaver or volunteer test data on frontal, lateral, and rear impacts. Model validations for 15 series of cadaver test data associated with frontal impacts are presented in this article. The THUMS with a vehicle sled model was applied to investigate effects of muscle activations on occupant kinematics and injury outcomes in specific frontal impact situations with AEB. Results and Conclusions: In the validations using 5 series of cadaver test data, force time curves predicted by the THUMS were quantitatively evaluated using correlation and analysis (CORA), which showed good or acceptable agreement with cadaver test data in most cases. The investigation of muscular effects showed that muscle activation levels and timing had significant effects on occupant kinematics and injury outcomes. Although further studies on accident injury reconstruction are needed, the THUMS has the potential for predictions of occupant kinematics and injury outcomes considering muscular tone effects with relatively low computational costs. Keywords: finite elements, modeling, occupant kinematics, injury outcome, biofidelity, muscle activation Introduction In Japan, the number of fatalities due to automotive accidents tends to decrease during the last 2 decades, whereas the numbers of injured persons and accidents are almost the same as those 20 years ago. As for the fatalities worldwide, the number of fatalities has changed little in the United States in these 6 decades, whereas the number of fatalities dramatically increased in India and China with an increase in the Masami Iwamoto, Yuko Nakahira, and Hideyuki Kimpara Managing Editor David Viano oversaw the review of this article. Address correspondence to Masami Iwamoto, Toyota Central R&D Labs., Inc., 41-1 Yokomichi, Nagakute, Aichi 480-1192, Japan. E-mail: iwamoto@mosk.tytlabs.co.jp number of vehicles. Active safety devices such as automatic emergency brake (AEB) and precrash seat belt have the potential to accomplish further reductions in the number of the fatalities due to automotive accidents. However, their effectiveness should be investigated by more accurate estimations of their interaction with human bodies. Although crash dummies with higher biofidelity have been developed and used for injury assessment, detailed investigations of the locations and number of bone fractures and the severity of soft tissue injuries in the brain and internal organs as well as effects of active muscles in precrash on injury outcomes are needed for injury mitigation of real human bodies. Computational human body models are useful to solve these demands. During the last 2 decades, many researchers have developed computer simulation models of the human body with a variety of modeling approaches using multibody method or finite element (FE)

Occupant Injury Mechanisms 37 method as well as representing human anatomy by simplified or detailed structures. However, each modeling approach has advantages and disadvantages in computational costs and injury predictions. Meijer et al. (2013) developed a multibody human model with active muscles to investigate the muscular tone effects on occupant motions during frontal impacts. Each muscle was modeled using bar elements with Hill-type muscle material properties. Each joint was modeled by mechanical joint structure and each human body part was modeled as rigid bodies. Their model has a benefit in computational costs and in predicting overall occupant motion with muscle activation. However, the model cannot be used for injury predictions in bones, ligaments, and internal organs. Osth et al. (2012) developed a human body FE model with some muscles in the neck and trunk. They incorporated muscle models with Hill-type material properties into a human body FE model THUMS Ver. 3.0 and applied the model to prediction of occupant behaviors during deceleration with AEB. However, their model did not include internal organs. Additionally, in their model, the muscles were modeled with several one-dimensional Hilltype elements attached directly, without any tendon elements, to nodes on the skeletal structure. Therefore, it is not suitable for accurate prediction of bone fractures and internal organ injuries. Gayzik et al. (2012) developed an FE model of an adult mid-size male occupant, which was called the Global Human Model Consortium model, and validated the model against several cadaver test data. The model showed good or acceptable impact responses. However, the model has disadvantages in computational costs and in considering muscular tone effects because the number of elements in the whole body model was about 2 million, and each joint was not flexible. The objective of this study is to develop and validate a human body FE model that can be used for predictions of occupant kinematics and injury outcomes considering muscular tone effects in the whole body with relatively low computational costs. In this study, we developed a human body FE model called THUMS including a head/brain, spine, shoulder, internal organ, and whole body muscles and validated the THUMS against 35 series of cadaver or volunteer test data on frontal, lateral, and rear impacts. In this article, we show how to develop and verify the 5 models of long bones in the lower extremities, thoracic and lumber spines, shoulder, internal organ, and whole body muscles in addition to refinement of the head/brain model. Descriptions on development and verification of the new 5 models include 6 validation results using cadaver test data among the 35 validations. We also showed how to validate the THUMS against 9 series of cadaver test data associated with frontal impacts among the 35 validations. In addition, head neck responses predicted by the THUMS with and without muscle activation were compared with those of volunteer test data conducted with low-speed frontal impact, and brake pedal force and steering force predicted by the THUMS with muscle activation were compared with those obtained from experimental test data on a specific driver s bracing condition using a male volunteer. Finally, we applied the THUMS to investigate the effect of muscle activity on the occupant kinematics and injury outcomes during frontal impact situations with the AEB. All simulations were performed using an explicit finite element code MPP Fig. 1. A new human body FE model. LS-DYNA v971 Revision 6.1.2 single precision (LSTC) on a FUJITSU PRIMERGY BX922 S2 (3.33 GHz Xeon X5680 processors with 48 GB of installed RAM utilizing Infiniband nodal interconnects) running on a Red Hat Enterprise Linux 5.6. Methods Updated Modeling of a Human Body FE Model THUMS Ver. 3.0 We developed a new FE model of human body with the body size of a 50th percentile adult male (AM50) whose height and weight were 175 cm and 77 kg, respectively. Figure 1 shows the developed human body FE model, including a head/brain, spine, shoulder, internal organs, long bones in the lower extremities, and whole body muscles. Previously we developed the LS-DYNA version of THUMS Ver. 3.0, which included the head/brain FE model (Iwamoto et al. 2007). A detailed description of the head/brain FE model can be found in Kimpara et al. (2006).

38 Iwamoto et al. In the new model, the head/brain model was refined for representation of more accurate anatomical structure and improvement of the computational stability. In addition, 5 major characteristic features of long bone models in the lower extremities, thoracic and lumbar spines, shoulder, internal organs, and whole body muscles as shown in Fig. 1 were newly implemented in the THUMS Ver. 3.0. Refinement of the Head/Brain Model The original head/brain model of the THUMS Ver. 3.0 consisted of the skull, brain, and skin. The brain model consists of all hexagonal solid elements representing the cerebrum, cerebellum, brainstem with distinct white and gray matter, and cerebral spinal fluid (CSF). The brain was modeled as a viscoelastic material and the CSF was modeled as an elastic material with fluid-like behavior. In the refined head/brain model, a gap between the skull and the brain was eliminated in the base of skull to represent the anatomical structure of the head and brain more accurately. In addition, the skull and the brain were connected with shared nodes, although a tied contact definition was used for the interface between the skull and the brain in the THUMS Ver. 3.0, which sometimes caused computational instability. Modeling and Verification of Long Bone Models in the Lower Extremity Cortical bones of long bones in the lower extremities that is, the femur, tibia, and fibula were modeled by using solid elements with the corresponding thickness obtained from computed tomography image data (Visible Human Project Data, NIH) to make more accurate bone fracture prediction (Fig. A1, see online supplement). The femur, tibia, and fibula models were verified for quasistatic 3-point bending test data of wet long bones obtained from human cadavers with ages ranging from 20 to 39 years reported by Yamada (1970). Simulation setups for the femur, tibia, and fibula were used to reproduce the test conditions. Load deflection curves predicted by the model showed good agreement with those of the test data in the femur, tibia, and fibula. The developed long bone models were integrated with the THUMS Ver. 3.0. Modeling and Verification of the Thoracic and Lumber Spines Thoracic and lumbar spines were modeled as deformable bodies including intervertebral discs and major ligaments with shared nodes to estimate spine fracture risks (Fig. A2, see online supplement). An L3 L4 unit model extracted from the spine model was verified for human cadaver test data on quasistatic flexion, extension, lateral bending, anterior shearing, torsion, and compression of functional spinal units (FSUs) from lumbar spines reported by Schultz et al. (1979) and Begeman et al. (1994). The simulation condition almost reproduced the test setup, in which L3 was loaded while L4 was fully fixed. Simulation results showed good agreement with cadaver test data in all loading conditions. The material properties of the L3 L4 unit were used for other FSU composing the lumbar spine. The FSU in thoracic spine was also verified for the same kind of cadaver test data reported by Panjabi et al. (1976). The developed thoracic and lumbar spines were integrated with the THUMS Ver. 3.0. Modeling and Verification of the Shoulder Based on one of the author s papers (Iwamoto et al. 2000), the shoulder complex was modeled as 4 articulated joints of the glenohumeral joint, acromioclavicular joint, sterno-clavicular joint, and scapulo-thoracic articulation with major ligaments and muscles to estimate shoulder injury risks (Fig. A3, see online supplement). The shoulder model was verified for cadaver test data on lateral impacts for the shoulder reported by Bendjellal et al. (1984). In the cadaver tests, a cylindrical impactor with a mass of 23.4 kg and a diameter of 150 mm hit the left shoulder with an initial velocity of 4.5 m/s. The developed shoulder model was integrated with the THUMS Ver. 3.0. Simulations using the updated THUMS were performed under conditions similar to the test condition. The simulation results fell within the cadaver test corridor. Modeling and Verification of the Internal Organs Internal organs such as the lungs, heart, liver, kidney, and pancreas were modeled to estimate internal organ injury risks (Fig. A4, see online supplement). Basic geometry data of the internal organs were obtained from Visible Human Project data (NIH). The lung and liver models extracted from the internal organ model were verified for porcine test data on direct impacts for the organs reported by Hayamizu et al. (2003) and Kemper et al. (2011), respectively. Simulation setups almost reproduced the test conditions. In the impact simulation for the lung model, the model predicted higher force at the initial deformation than test data. In the impact simulation for the liver model, the model reproduced the force deflection curves obtained from Kemper et al. (2011) until a deflection of 0.05, but the model did not reproduce the curves after the deflection of 0.05. Although the models did not reproduce the test data, the tendency of the force deflection curve predicted by the model was almost the same as that of test data. The developed internal organ model was integrated with the THUMS Ver. 3.0. Modeling and Verification of the Whole Body Muscles Skeletal muscles of the whole body except the head and face were modeled with their corresponding tendons and were attached to the THUMS Ver. 3.0 to simulate a braced occupant (Fig. A5, see online supplement). The model includes 262 skeletal muscles in the whole body. The model consists of 800 parts for muscles and 296 parts for tendons. Each muscle was modeled using truss elements with Hill-type muscle material, and each tendon was modeled using seat belt elements to represent tension-only nonlinear elastic properties. The geometry of each muscle tendon complex was represented as a series of line segments that connected the origin and the insertion. The muscle models were attached to bone models at their origin, insertion, and via points by using an interpolation constraint method and the via points were modeled using the element beam pulley, which is available in LS-DYNA v971 Revision 6.00 or newer revisions and has been developed to achieve continuous sliding of muscle elements in the same manner as

Occupant Injury Mechanisms 39 a seat belt slip ring. The mass of the whole muscle model was around 0.9 kg. The material properties of muscles and tendons were obtained from the literature (Gans 1982; Thelen 2003; Winters 1990; Yamada 1970). The muscle model was verified for muscle moment arm angle relationship and torque angle relationship in each joint of the upper and lower extremities obtained from the literature. Torque angle relationships in the elbow and knee joints were compared between simulation results predicted by the updated THUMS and volunteer test data (Amis et al. 1980; Anderson et al. 2007; Buchanan et al. 1998; Murray et al. 1980; van Eijden et al. 1987). The torque and angle were calculated during motions of the elbow joint and knee joint. The torque angle relationships predicted by the model agreed well with those of test data in the flexion and extension of the elbow joint and the flexion of the knee joint. In addition, some body shapes of the THUMS Ver. 3.0 such as the buttock, thigh, and shoulder were modified with remesh to represent more realistic human body shape by referring to anthropometric and geometry data for mid-sized adult male reported by Schneider et al. (1983). The total numbers of elements and nodes for whole body of the updated THUMS are 281,260 and 184,242, respectively. The smallest time step of the updated THUMS is 2.7e-07, which would realize relatively low computational costs comparing to some more detailed human body FE models in which the total number of elements for whole body is about 2 million. Validation of the Updated THUMS Each body part of the updated THUMS was validated against 9 series of cadaver test data obtained from the literature. The simulations for the model validations were performed without muscle activation. In the model validation, a quantitative evaluation method was used to compare time history curves between simulation results and cadaver test data. In this study, the correlation and analysis (CORA) method reported by Gehre et al. (2009; Gehre and Stahlschmidt 2011) was used for quantitative evaluation of the model accuracy. The CORA method provides an objective evaluation of whole response curves obtained from simulation results and test data. The method combines 2 independent submethods, a corridor rating and a cross-correlation rating. The corridor method evaluates the fitting of a response curve into user-defined or automatically calculated corridors. The cross-correlation method evaluates cross-correlation function, size, and phase shift. The use of both of these 2 submethods is essential to compensate for the disadvantages of each submethod by the other method. In this study, we used CORA release 3.6, in which we used all defaults for parameters that control the evaluation except A EVAL to enlarge the evaluation interval for adjustment to each test. Thus, in the cross-correlation method, the weights are 0.50, 0.25, and 0.25 for cross-correlation function, size, and phase, respectively, and in total evaluation, the weights are 0.50 and 0.50 for the corridor method and cross-correlation method, respectively. The evaluation using the CORA method is performed by rating between 1 and 0. The following sliding scale is defined by the technical report ISO/TR 9790: CORA ratings of 0.86 to 1.0, 0.65 to 0.86, 0.44 to 0.65, 0.26 to 0.44, and 0.0 to 0.26 are evaluated as excellent, good, fair, marginal, and unacceptable, respectively. Because the CORA method can be used for time history curves, the CORA method was used for 5 series of cadaver test data except the validations on frontal impacts for the thorax and frontal impacts for the abdomen, for which force deflection curves are presented. Frontal Sled Impacts Vezin et al. (2001) conducted a series of sled tests using 4 unembalmed cadavers to see head and thorax responses of occupants in frontal impact. Rigid flat seats with geometry close to that of a standard mid-size car were used in the tests. The feet of the cadavers were fixed on the footrest and the hands and the head were maintained in the natural driver posture, with 2 nylon wires, which were released at impact. The seat back was tilted at a 20 angle. The subjects were restrained by separate shoulder and static pelvis belts. The shoulder belt was equipped with a force-limiting system. The nominal force limit was 4 kn and the pretension was made manually before the crash. Simulation setups with a force limit of 4 kn using a whole body model of the updated THUMS carefully reproduced the abovementioned experimental setups. Figure 2a shows a simulation setup for frontal sled impacts. Resultant accelerations of head, T1, T8, and pelvis were obtained from the simulation and were compared with those of cadaver test data. The computational time of this simulation using the whole body model for a period of 200 ms was 15 h 59 min with 4CPU. Axial Impacts for the Foot Kitagawa et al. (1998) conducted dynamic axial impact tests using 16 specimens from 8 pairs of human cadaveric lower legs to investigate the combined effect of muscle preloading by emergency braking and the associated pedal impact force. In this study, dynamic axial impact simulations were performed to validate the right foot and ankle region of the updated THUMS against dynamic cadaveric responses with muscular tension using the cadaver test data from by Kitagawa et al. (1998). Figure A7a (see online supplement) shows a setup for the axial impact simulations. The boundary conditions for the simulation reproduced those for the cadaver tests. A pendulum weighing 18 kg with a diameter of 70 mm was modeled by a pad and a small portion of the rigid impactor. The Achilles tendon force was applied to a node of the calcaneus as a constant force of 1 kn. The pendulum hit the forefoot at 3.5 m/s in the tests. The impactor force and tibial force time histories were compared between simulation results and cadaver test data. The computational time of this simulation using the whole body model for a period of 50 ms was 13 h 51 min with 4 CPU. Frontal Impacts for the Knees Rupp et al. (2008) conducted cadaver tests on frontal knee impact using 5 cadavers. The test apparatus with 2 load cells accelerated a 255-kg weighted platform to a velocity of 1.2, 3.5, or 4.9 m/s, and the platform traveled along a linear track until padded surfaces located on the front of the platform contacted the knees of a seated cadaver. Load cells located be-

40 Iwamoto et al. Frontal Impacts for the Abdomen with a Simulated Steering Wheel Nusholtz et al. (1994) conducted cadaver tests to investigate abdominal responses during anterior posterior impacts by a steering wheel. In their test, a rigid impactor with the geometry similar to that of the lower half of a steering wheel and a mass of 18 kg hit abdominal regions of 3 male subjects in the anterior posterior direction with an initial velocity of 10 m/s. The boundary conditions for the simulation reproduced those for the cadaver tests are shown in Fig. A9a (see online supplement). Force defection curves predicted by the updated THUMS were compared those of cadaver test data. The computational time of this simulation using the whole body model for a period of 29 ms was 7 h 54 min with 4 CPU. Belt Impacts for the Abdomen Hardy et al. (2001) conducted a series of cadaver tests on abdominal impacts and presented abdominal responses during impacts using rigid bar, seat belt, and airbag loadings. This study selected seat belt loading tests among a series of cadaver tests. Three cadavers were used for the tests. By pulling standard seat belt webbing into the cadavers from behind with a peak loading rate of approximately 3.5 m/s, the cadavers were loaded about the mid-abdominal region, approximately at the level of the umbilicus. Penetration, which corresponds to the difference between the motion of the seat belt and the displacement of the spine with respect to the laboratory, and load in the seat belt webbing were measured. Simulation setups reproduced the abovementioned experimental setups as shown in Fig. A10a (see online supplement)). Force penetration curves predicted using the updated THUMS were compared with those of cadaver test data. The computational time of this simulation using the whole body model for a period of 74 ms was 22 h 2 min with 1CPU. Fig. 2. Validation of whole body model for cadaver test data on frontal sled impacts. hind each impact surface independently measured force histories applied to the left and right knees. Simulation setups using the updated THUMS carefully reproduced the abovementioned experimental setups as shown in Fig. A8a (see online supplement). In the simulation, 2 impactors with a mass of 255 kg of the weighted platform hit the left and right knees of the seated human models. The average forces of both knees were obtained from the test data and the simulation and were compared. The computational time of the simulation using the whole body model of 4.9 m/s for a period of 70 ms was 18 h 52 min with 4 CPU. Frontal Impacts for the Thorax Kroell et al. (1971, 1974) conducted cadaver tests to investigate thoracic responses during frontal impacts. A hub impactor with a diameter of 152 mm and a mass of 23.4 kg was applied to the frontal surface of human thorax in the anterior posterior direction with an initial velocity of 6.9 m/s. The boundary conditions for the simulation reproduced those for the cadaver tests are shown in Fig. A11a (see online supplement). Thoracic deflection was obtained from the distance between an impactor surface and a thoracic posterior surface. Force displacement curves predicted by the updated THUMS were compared with those of the cadaver test data. The computational time of this simulation using the whole body model for a period of 50 ms was 13 h 57 min with 4 CPU. Belt Impacts for the Thorax Kent et al. (2004) presented thoracic response corridors developed using 5 postmortem human subjects subjected to 4 loadings of single and double diagonal belt, distributed, and hub loading on the anterior thorax. This study selected single seat belt loading tests among a series of cadaver tests. Subjects were positioned supine on a table and a hydraulic master slave cylinder arrangement was used with a high-speed materials testing machine to provide controlled chest deflection at a rate similar to that experienced by restrained postmortem human subjects in a 48 km/h sled test. Thoracic response was characterized using the deflection at the midline of the sternum and a load cell mounted between the subject and the loading table.

Occupant Injury Mechanisms 41 Simulation setups using the updated THUMS carefully reproduced the abovementioned experimental setups as shown in Fig. A12a (see online supplement). Predicted force deflection curves were compared with cadaver test corridors. The computational time of this simulation using the whole body model for a period of 265 ms was 70 h 12 min with 4 CPU. Head and Neck Flexion During Frontal Impacts Wismans et al. (1987) compared head neck responses in frontal flexion between human volunteers and cadavers. They used sled test data using 5 cadavers and 5 volunteers under the same boundary condition and reported that head center of gravity rotational angles obtained from volunteer tests were lower than those obtained from cadaver tests. The difference in the angles between volunteer and cadaver was about 20. In this study, we performed simulations using the updated THUMS to predict head neck responses during frontal impact with a velocity of 16 km/h. Figure A13a (see online supplement) shows a simulation setup. The acceleration time history of the X direction (anterior posterior) and time history of rotational angle around the Y direction (left right) at T1 obtained from the cadaver tests were given to the T1 in the model. The rotational angle of head center of gravity was obtained from the simulation and was compared with volunteer test corridor reported by Wismans et al. (1987). The computational time of this simulation using the whole body model for a period of 250 ms was 66 h 2 min with 2 CPU. Brain Motion During Occipital Head Impact Hardy et al. (2001) conducted a series of experimental tests using 3 cadavers to investigate brain displacements in human cadaver head during occipital and frontal head impacts. In their cadaver tests, a flat disk impactor of 150 mm diameter including an Ensolite padding of 50-mm thickness hit the occipital and frontal region of the head with impact velocities from 2.5 to 3.5 m/s. The neutral-density targets (NDTs) were implanted in the brain along 2 vertical columns located in the occipitoparietal region (labeled p for posterior column) and in the temporoparietal region (labeled a for anterior column) with the spacing between the centers of the NDTs in the columns ranged from 7 to 12 mm. Three-dimensional trajectories of the NDTs were measured by a high-speed biplanar x-ray system to obtain the relative motion between the skull and the NDTs, which moved synchronously with the brain tissue. In this study, simulation setups using the head/brain model extracted from the updated THUMS carefully reproduced the abovementioned experimental setups (Fig. A6, see online supplement). In this article, only simulation results for the test C755 T2 of the occipital head impacts are presented. The translational and angular accelerations of the head center of gravity obtained from the test data were input into the head/brain model, and the NDTs of a1 to a5 and p1 to p5 were compared between simulation results and cadaver test data. The computational time of this simulation using the head/brain model for a period of 60 ms was 1 h 29 min with 4CPU. Effect of Muscle Activation on Head Neck Impact Responses and Reaction Forces of a Braced Driver In this section, the update THUMS was used to investigate effect of muscle activation levels and the activation timings or occupant postures on the head neck responses of occupants and reaction forces of a braced driver. Head Neck Responses During Low-Speed Frontal Impact Arbogast et al. (2009) conducted a series of volunteer tests on low-speed frontal impacts with peak sled accelerations of 3.6 and 3.8 g using 20 pediatric males and 10 adult males. They obtained time histories of X and Z displacements and Y rotational angle at the T1 of each subject by using motion capture markers during the tests. Dibb et al. (2013) used their volunteer test data to validate 3 head neck musculoskeletal models of a 6-year-old, 10-year-old, and adult with and without muscle activation. They input the time histories at the T1 to the T1 of each head neck model and compared several time histories of head rotational velocity and X, Z displacements of the external auditory meatus (EAM) and the nasion between model responses and the volunteer test data. In this study, we also input the time histories at the T1 obtained from the same volunteer test data by Arbogast et al. (2009) to the T1 of the updated THUMS and compared the head rotational velocity and X, Z displacements of the EAM and the nasion between model responses and the adult volunteer test data. In the simulations, the activation level of each neck muscle was given with the following 3 cases. In case 1, activation levels of all neck muscles were assumed as 1% to simulate cadaveric human responses, shown as no activation in Fig. 3. In case 2, activation levels of all neck muscles were assumed to be a constant value of 20%, which was a mean value of electromyography (EMG) data, previously measured at a neck muscle (the sternocleidomastoid) of a male subject in a standing posture in our laboratory. In case 3, activation levels of neck extensors and flexors were assumed by referring to optimal muscle activation levels to reproduce the same volunteer test data (Arbogast et al.2009) reported by Dibb et al. (2013). The onset of activation began at 75 ms, increasing to a peak value of 42% at 180 ms and then decreased to zero at 400 ms for neck extensors; the onset of activation began at 250 ms, increasing to a peak value of 22% at 400 ms and then decreased to zero after 500 ms for neck flexors. Simulation results for these 3 cases were compared with the volunteer test data. The computational time of the simulation using the whole body model without muscle activity for a period of 500 ms was 78 h 27 min with 8 CPU. Reaction Forces of a Braced Driver The activation level of each muscle in human whole body is critical to simulate a driver s braced condition in precrash using a human body FE model with multiple muscles. Audrey et al. (2009) conducted a series of volunteer tests to analyze behaviors of 80 drivers during critical events using a driving simulator. They found that more than 67% of subjects moved backward with the right leg extended to the brake pedal and arms extended to the steering wheel to anticipate the crash. According to their findings, we selected a driver s braced

42 Iwamoto et al. Fig. 3. Parametric simulations to investigate effect of muscle activation on head-neck responses during low-speed frontal impacts. condition in which a volunteer subject pushed his right foot on the brake pedal and his hands on the steering wheel with the maximal voluntary force. In this study, a volunteer test with one healthy male subject, 33 years old, whose height was 176.5 cm and weight was 75 kg, similar to AM50, was conducted to obtain physiological information in the braced condition with braking under his informed consent based on the Helsinki Declaration. All procedures were approved by the institutional ethics committee. In this test, the subject was asked to push his right foot on the brake pedal and his hands on the steering wheel with his maximal voluntary force in the developed test apparatus fixed in the laboratory. However, the brake pedal did not allow for any strokes, and the steering wheel was different from an actual one and did not allow for any rotations. In the test, some data such as 3D motion of the subject, EMG from 24 skeletal muscles of right upper and lower extremities, pedal force, right and left separated steering reaction forces, and reaction force on seats were obtained. Detailed information on the volunteer test were described in Iwamoto et al. (2012). In this study, we performed a simulation to reproduce reaction forces in the braced condition using the updated THUMS. In the simulation setup, the THUMS was set to a sitting position with rigid seats; the right foot was positioned on a brake pedal and the hands were positioned to grip the steering wheel in order to reproduce the volunteer test setup. In this study, all muscles in the whole body were activated to investigate whether we could reproduce the pedal force and steering forces in the braced condition using the following 3 activation level sets. In case 1, activation levels of all muscles were given by using EMG data measured in our laboratory. The normalized EMG activation levels of 24 muscles measured in the right lower extremity and right upper extremity in the abovementioned braced condition were directly input into the corresponding muscle models. Because muscle activation levels of other muscles in the right lower extremity and the upper extremities were not measured in the test, they were estimated based on the same method as described in Iwamoto et al. (2012). Muscle activation levels of the left lower extremity, trunk, and neck were also not measured in the test. Therefore, we assumed activation levels of most muscles in the neck and trunk as a constant value of 0.2 (20%), which was the mean value of activation levels of some muscles such as sternocleidomastoid in the neck and rectus abdominis in the trunk in other experimental volunteer tests using the same male subject in a standing posture. The activation level of each muscle in the left lower extremity was assumed to be 0.01 (1%), because there were no EMG data on muscles in the left lower extremity during braking motion of the same subject. In case 2, parametric simulations with 3 different activation levels were performed to investigate effect of muscle activation levels on pedal force and steering force. In the activation level set called as braking1, activation levels of extensors for upper extremities and right lower extremity were multiplied by 1.5 to 3. In the activation level set called braking2, the activation levels for extensors of the shoulder joint and extensors of the hip and knee joints were 1.5 2 times as large as those in braking1, and the activation levels of trunk extensors (back muscles) were 2 3 times as large as those in the braking1. In the activation level set called braking3, the activation levels of extensors for the upper extremities were almost the same as those in braking1 and the activation levels of extensors for the right lower extremity were almost the same as those in braking2, whereas activation levels of the neck extensors were 4 times as large as those in braking1 but the activation levels of trunk extensors were the same as those in braking1. In case 3, the initial posture of the right lower extremity was set to have a knee angle of 110 and the initial posture in cases 1 and 2 was set to have a knee angle of 120 to investigate the effect of initial postures on the reaction forces of the brake pedal. In this case, the activation levels in braking3 were used. The steering force was calculated as a mean value of reaction forces obtained from the right hand and left hand. The computational time of the simulation for a period of 900 ms was 50 h 15 min with 8 CPU. Prediction of Driver s Kinematics and Injury Outcomes in Precrash and During Crash The updated THUMS was applied for reproduction of a driver s braced condition and prediction of a driver s kinematics and injury outcomes in an assumed precrash event and during an assumed crash event. The simulations under a frontal crash situation were performed in 2 cases of no muscle activation and a driver s braced condition. In this study, a frontal crash situation with AEB was selected to determine the differences in an adult male driver s

Occupant Injury Mechanisms 43 behaviors in precrash and during crash between the human body with and without muscle activation. The driver was assumed to sustain deceleration of 0.8 g for 300 ms in precrash with AEB and then sustain a frontal impact deceleration with a speed of 50 km/h. In the selected crash situation, the driver had the potential to brace his body with his muscle activity. However, we do not have experimental volunteer test data on emergency braking using a vehicle or a simulated vehicle, which should include a driver s motions, EMG data on some muscles, and reaction forces of a brake pedal, a steering wheel, seats, and so on for reconstruction of a driver s kinematic and kinetic responses. These kinds of experimental volunteer tests are not easy to conduct due to the risks to volunteers. Therefore, this study assumed the driver s braced condition to be similar to that in emergency braking motion conducted in our laboratory described previously. The frontal crash simulation was performed with a vehicle sled FE model, which was originally developed by the National Crash Analysis Center of the George Washington University under contract with the Federal Highway Administration and the NHTSA and was modified by JSOL Corporation. The vehicle sled model includes a Ford Taurus vehicle body, an automotive seat, a 3-point seat belt, and an airbag for drivers. Four simulations were performed to investigate the effect of driver positions and muscle activations on injury outcomes as shown in Fig. 5a. Case 1 was performed without muscle activation during all simulation periods. Case 2 was performed with the driver s braced condition during all simulation periods. Case 3 was performed under a condition in which a driver braced his body before impact and had no muscle activation after impact. Case 4 was performed under a condition in which a driver with a different initial position similar to a driver s posture just before impact in case 1 without muscle activation had no muscle activation after impact. These 4 simulation results were compared to investigate the effect of driver position and muscle activation on injury outcomes. The computational time of the simulation without muscle activity for a period of 500 ms was 101 h with 4CPU. Result Validation in the Updated THUMS The results obtained from quantitation evaluation using the CORA method for each validation are listed in Table 1. Frontal Sled Impacts Figures 2b 2d show a comparison of resultant accelerations of the pelvis, first and eight thoracic vertebrae, and head between simulation results using the updated THUMS and cadaver test data. In Fig. 2, the curve of the cadaver test plotted in each graph represents the average of test data using 2 cadavers. Simulation results show good agreement with test data except the head acceleration. According to the quantitative evaluation using the CORA method, the evaluations for the accelerations of the head, T1, T8, and pelvis were fair, good, fair, and good, respectively. The evaluations indicate that whole body model of the updated THUMS has somewhat good biofidelity Fig. 4. Parametric simulations to investigate effect of muscle activation levels and knee angles in a driver s bracing conditions. (a) Simulation results. (b) Volunteer tests. (c) Maximal reaction forces on brake pedal and steering wheel. for whole body behaviors during high-speed frontal impacts. However, in particular, the head responses had lower ratings of 0.612. The lower biofidelity in the head responses is probably due to modeling of the spine disc, in which the disc was modeled by a kind of viscoelastic or rubber material that did not represent asymmetry of tension and compression in material properties of the disc. When fractures of rib cartilages were evaluated using an ultimate stress of 4.4 MPa (Yamada 1970), 5 fractures at the right ribs and rib cartilages and 13 fractures at the left ribs and rib cartilages were predicted, and 1 to 2 right rib fractures and 8 to 18 rib fractures and sternum fracture were observed in the cadaver tests. Therefore, the predicted injury outcomes were similar to those observed in the cadaver tests.

44 Iwamoto et al. Table 1. Evaluation results on model validations using CORA method Correlation method Correlation method Test data (physical value) Corridor method rating Cross correlation Size Phase shift Rating Total rating Evaluation Frontal impact sled (Vezin et al. 2001) Axial foot impact (Kitagawa et al. 1998) Frontal knee impact (Rupp et al. 2008) Frontal neck flexion 15 g (Wismans et al. 1987) Head rotation C755-T2 (Hardy et al. 2001) Head center of gravity 0.567 0.658 0.646 0.665 0.657 0.612 Fair acceleration T1 acceleration 0.786 0.846 0.883 0.949 0.881 0.833 Good T8 acceleration 0.580 0.566 0.935 0.802 0.717 0.649 Fair Pelvis acceleration 0.695 0.640 0.933 1.000 0.803 0.749 Good Impact force 0.537 0.699 0.670 0.807 0.719 0.628 Fair Tibia force 0.727 0.864 0.729 0.838 0.824 0.775 Good Impact Force (1.2 m/s) 0.986 0.996 0.926 0.990 0.977 0.981 Excellent Impact Force (3.5 m/s) 0.892 0.996 0.829 0.970 0.948 0.920 Excellent Impact Force (4.9 m/s) 0.888 0.995 0.864 0.815 0.917 0.902 Excellent Head rotational angle 0.721 0.990 0.504 0.909 0.849 0.785 Good X displacement (a1) 0.578 0.652 0.743 1.000 0.762 0.670 Good Z displacement (a1) 0.218 0.000 0.865 0.931 0.449 0.334 Marginal X displacement (a2) 0.661 0.763 0.665 1.000 0.798 0.729 Good Z displacement (a2) 0.306 0.169 0.996 1.000 0.583 0.445 Fair X displacement (a3) 0.277 0.375 0.073 1.000 0.456 0.366 Marginal Z displacement (a3) 0.401 0.350 0.902 1.000 0.650 0.526 Fair X displacement (a4) 0.178 0.000 0.355 0.772 0.282 0.230 Unacceptable Z displacement (a4) 0.297 0.158 0.845 1.000 0.540 0.419 Marginal X displacement (a5) 0.278 0.467 0.307 0.923 0.541 0.410 Marginal Z displacement (a5) 0.255 0.055 0.869 1.000 0.495 0.375 Marginal X displacement (p1) 0.551 0.608 0.763 1.000 0.745 0.648 Fair Z displacement (p1) 0.517 0.357 0.983 1.000 0.674 0.596 Fair X displacement (p2) 0.422 0.582 0.946 1.000 0.778 0.600 Fair Z displacement (p2) 0.519 0.176 0.882 1.000 0.559 0.539 Fair X displacement (p3) 0.372 0.697 0.611 1.000 0.751 0.562 Fair Z displacement (p3) 0.519 0.349 0.661 1.000 0.590 0.554 Fair X displacement (p4) 0.245 0.137 0.459 1.000 0.433 0.339 Marginal Z displacement (p4) 0.518 0.321 0.458 1.000 0.525 0.522 Fair X displacement (p5) 0.258 0.029 0.687 0.604 0.337 0.297 Marginal Z displacement (p5) 0.521 0.472 0.799 1.000 0.686 0.603 Fair Axial Impacts for the Foot Figure A7b and A7c compare the impactor force and the tibial axial force of simulation results with those of test data. The simulation results generally showed good agreement with the test data. According to the quantitative evaluation using the CORA method, the evaluation for impact force was fair and that for the tibia force was good. This indicates that the foot and ankle region of the updated THUMS has somewhat good biofidelity for foot impact responses. Frontal Impacts for the Knees Figures A8b 8d show a comparison of impactor force time histories between simulation results and cadaver test data for 3 velocities. In Fig. A8, the curve of the experimental test plotted in each graph represents the average of test data using 5 cadavers under the same condition. Simulation results showed perfect match with test data in the velocity of 1.2 m/s and peak forces predicted by the model were a little bit higher than those of test data in the velocities of 3.5 and 4.9 m/s. According to the quantitative evaluation using the CORA method, the evaluations for impact forces at 1.2, 3.5, and 4.9 m/s were all excellent. This indicates that the knee region of the updated THUMS has excellent biofidelity for frontal knee impact responses. Frontal Impacts for the Abdomen with a Simulated Steering Wheel Figure A9b shows comparison of force deflection curves between simulation results and 3 cadaver test data (86M006, 86M042, 86M052). Simulation result showed good agreement with test data. Belt Impacts for the Abdomen Figure A10b shows comparison of force penetration curves obtained from belt impacts for the abdomen between simulation result and 3 cadaver test data. The simulation result presented a force penetration curve similar to those of cadaver test data. These 2 simulation results on steering and belt impacts for the abdomen indicate that the updated THUMS has good biofidelity to simulate abdominal responses during frontal impacts. Frontal Impacts for the Thorax Figure A11b shows comparison of force deflection curves obtained from frontal impacts for the thorax between simulation

Occupant Injury Mechanisms 45 results and cadaver test data. Simulation results fell almost within the test data. Belt Impacts for the Thorax Figure A12b shows simulation results of the posterior reaction forces and chest deflection compared with test corridors. The simulation result perfectly fell within the test corridors. These 2 simulation results on hub and belt impacts for the thorax indicate that the updated THUMS has good biofidelity to simulate thoracic responses during frontal impacts. Head and Neck Flexion During Frontal Impacts Figure A13b compares head center of gravity rotational angle time histories predicted by the model with those of volunteer test corridor and cadaver test data. The predicted peak angle was lower than those of cadaver test data, although it was close to the peak angle of the volunteer test corridor. According to the quantitative evaluation using the CORA method, the evaluation for head rotational angle was good. Although the head neck region of the updated THUMS has good biofidelity for the head neck responses during frontal impact, the size had a lower rating of 0.504. This is because the neck model was stiffer in flexion comparing to the cadavers. Brain Motion During Occipital Head Impact Figure A6 compares X and Z displacements of the target makers of a1, a5, p1, and p5 in the brain with respect to the head center of gravity between simulation results and cadaver test data. Simulation results are indicated by solid lines and cadaver test data are indicated by dotted lines. Simulation results generally show good agreement with test data. According to the quantitative evaluation using the CORA method, the evaluations for X displacements at a1 and a2 were good, and those for X displacements at p1, p2, p3 and Z displacements at a2, a3, p1, p2, p3, p4, and p5 were fair. On the other hand, the evaluations for X displacements at a3, a5, p4, and p5 and Z displacements at a1, a4, and a5 were marginal, whereas that of X displacement at a4 was unacceptable. These evaluation results showed that the head/brain model of the updated THUMS has relatively higher biofidelity in the center of the brain but relatively less biofidelity in the outer side of the brain. The reason for the relatively less biofidelity is probably due to the modeling method of CSF, in which CSF was modeled by a kind of elastic material that did not include effect of fluid pressures. Effect of Muscle Activation on Head Neck Impact Responses and Reaction Forces of a Braced Driver Head Neck Responses During Low-Speed Frontal Impact Figure 3 compares time histories of head rotational velocity and X and Z displacements of the EAM and nasion between model responses and volunteer test data. The model without muscle activation in case 1 had the largest maximal values of the head rotational velocity and X and Z displacements of the EAM and nasion among the 3 cases. The model with a constant muscle activation level of 20% in case 2 had a decrease in the maximal values, and the model with the activation level set in case 3 had the smallest maximal values and the values were close to the volunteer test corridors. These parametric simulation results suggest that not only the activation levels but the activation timing could have significant effects on head neck responses. Reaction Forces of a Braced Driver Figure 4 shows a comparison of a male driver s kinematics during the braced condition and a comparison of the maximal reaction forces on brake pedal and steering wheel between simulation results and volunteer test data. The comparison demonstrated that the predicted driver s kinematics in bracing motion were comparable to those of volunteer test data. The maximal pedal forces were predicted as 269, 588, 620, 603, and 532 N in the activation level set of EMG base, braking1, braking2, braking3 with a knee angle of 120, and braking3 with a knee angle of 110, respectively. The maximal steering forces were predicted as 77, 137, 114, 153, and 172 N in the activation level set of EMG based, braking1, braking2, braking3 with a knee angle of 120, and braking3 with a knee angle of 110, respectively. On the other hand, the maximal pedal force and the maximal steering force were 640 and 220 N in the test, respectively. The parametric simulations suggest that extensors of the hip and knee joints and the trunk could contribute to an increase in braking force, and extensors of the elbow joint and the neck could contribute to an increase in steering force. In addition, a decrease in the knee angle contributed to the decrease of pedal force and a slight increase in steering force when the same activation level set was used. These simulation results suggest that the activation level and the combinations of levels in different muscles of the whole body as well as the initial posture could have a significant effect on pedal forces and steering forces of braced drivers. Prediction of Driver s Kinematics in Precrash and During Crash Figure 5b shows a comparison of drivers postures in precrash and injury outcomes among 4 cases. Just before impact, the driver s posture in case 1 without muscle activation was different from that in cases 2 and 3 with a braced condition. The driver had more flexion of the trunk in case 1 without muscle activation compared to cases 2 and 3 of the braced condition. As for the injury outcomes, HIC15 and femur load were the smallest in case 4 among 4 cases, wheras chest deflection was the smallest in case 2 among 4 cases as shown in Fig. 5c. However, HIC15 and femur load were less than their criteria in all cases. On the other hand, the maximal first principal strain of brain tissue in cases 1, 2, 3, and 4, which were found around the center of the brain, were 15.2, 8.8, 7.3, and 16.7%, respectively, and the number of rib fractures in cases 1, 2, 3, and 4 were predicted to be 8, 5, 8, and 8, respectively. A comparison of 4 simulation results showed that muscle activation had a significant influence on injury outcomes although the driver s posture just before crash did not have significant influence on injury outcomes in the specific frontal crash situation with AEB.

46 Iwamoto et al. Discussion Fig. 5. Parametric simulations to investigate effect of muscle activations on driver s postures and injury outcome in a frontal collision with AEB. (a) Sled acceleration and simulation conditions. (b) Driver s posture just before impact (0.35 sec). (c) Injury outcome. In this study, we developed a new human body FE model of AM50 by incorporating 5 features of long bones in the lower extremity, thoracic and lumbar spines, shoulder, internal organs, and whole body muscles into a THUMS Ver. 3.0 AM50 occupant model including a head/brain model. In addition, the head/brain model was refined for representation of a more accurate anatomical structure and improvement of the computational stability. This updated THUMS was developed for the purpose of detailed investigation on the interaction between human bodies and vehicular structures, including some safety devices in precrash and during a crash with relatively low computational costs. The updated THUMS without muscle activation was validated against several cadaver test data on axial impacts for the foot and ankle, frontal impacts for the knees, belt impact for the thorax and the abdomen, head neck responses during frontal impacts, as well as cadaver test data on frontal sled impacts. Simulation results predicted by the updated THUMS showed good or acceptable agreement with test data. However, the head/brain model of the updated THUMS has relatively less biofidelity in displacements at the outer side of the brain, and the neck model of the updated THUMS had relatively less biofidelity in head responses during frontal impacts, as mentioned previously. Therefore, further study is needed for modeling of CSF and intervertebral discs. The updated THUMS was used to investigate effect of muscle activation levels and the activation timings or occupant postures on the head neck responses of occupants and reaction forces of a braced driver. The parametric simulation results on head neck responses of occupants suggest that not only the activation levels but also the activation timing could have significant effects on head neck responses. The parametric simulation results on reaction forces of a braced driver suggest that the activation level and the combination of levels in different muscles of the whole body as well as the initial posture could have a significant effect on pedal forces and steering forces of a braced driver. However, we did not investigate effect of body size on head neck responses of occupants and reaction forces of braced drivers. Therefore, further comprehensive studies are needed for more detailed investigation on how activation levels and timings of muscles, combination of levels in different muscles of whole body, initial postures, and body size of occupants could affect occupant motions and reaction forces using the updated THUMS. The updated THUMS with and without activation of all muscles was applied to simulate a driver s kinematics and injury outcomes during an assumed frontal crash situation with AEB. From the simulation results, we found that a driver s kinematics just before impact were different between 2 cases of an unbraced condition and a braced condition. We also found that muscle activation had significant effects on injury outcomes although the driver s posture just before a crash did not have any significant effects on injury outcomes in the specific frontal crash situation with AEB. Further studies are necessary to verify the accuracy of injury outcomes predicted