Kinematic analysis of traumatic brain injuries in boxing using finite element simulations

IN

DEGREE PROJECT MEDICAL ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2016 ,

Kinematic analysis of

traumatic brain injuries in boxing using finite element simulations

XUELONG FAN

Kinematic analysis of traumatic brain injuries in boxing

using finite element simulations

X U E L O N G F A N

Master of Science Thesis in Medical Engineering Advanced level (second cycle), 30 credits Supervisor: Victor Str¨ omb¨ ack Alvarez Examiner: Svein Kleiven School of Technology and Health (STH) Royal Institute of Technology (KTH) TRITA-STH, 2016:100

STH, KTH

SE-141 86 Flemingsberg, Sweden

http://www.kth.se/sth

Abstract

The purpose of the thesis was to analyze and evaluate the head injuries due to a striking in a boxing match by LS-DYNA.

A simplified arm model was built up and was equipped with three segments which were linked with two spherical joints. The strain-stress curves of the boxing glove foam and glove leather were measured in the Neuronic Lab in School of Technology and Health, KTH. The dimension and weight of the model was also set as adjustable to fulfill various require- ments in different cases. Then a method was developed to facilitate the simulation. Finally, 39 video clips from the database were processed and the 13 cases were chosen to test the method and to perform the simula- tions. Additionally, the reliability of the model was assessed by comparing the outcome of the simulations with the results of the visual analysis from a previous study.

The outcome showed that the model was able to restore the scenario from the videos both quantitatively and qualitatively, but it also suggest a high sensitivity of the model to the data artifacts from the video analysis.

Interpretations and suggestions for the future work were also discussed.

Keywords: Finite element simulation, kinematic analysis, human arm

modeling, boxing

ACKNOWLEDGEMENTS

I would like to express my gratitude to Dr. Svein Kleiven and Victor S. Alvarez

for their thorough and professional supervision and support in the thesis. I

also would like to thank Dr. Mats Nilsson, Dr. Massimiliano Colarieti-Tosti

and Dr. Dmitry Grishenkov for their guidance and assistance which encouraged

and helped me to accomplish the whole process. Specially, I want to thank

Dr. Colarieti-Tosti for his help with the administration work so that I can

finish my thesis on time. Finally, I would like to thank Julianne Okan for being

a responsible and considerable opponent who inspired me and helped me to

improve my work.

Contents

1 INTRODUCTION 1

2 BACKGROUND 3

2.1 Injuries in boxing . . . . 3

2.2 Head injury evaluation . . . . 3

2.3 Anatomy of the arm . . . . 4

2.3.1 The structure and function . . . . 4

2.3.2 The distribution of mass . . . . 6

2.4 Previous Study . . . . 6

2.4.1 The model of head . . . . 7

2.4.2 The data from video analysis . . . . 8

2.5 The goal . . . . 8

3 METHODOLOGY 9 3.1 Case analysis . . . . 9

3.2 Geometry of the arm . . . . 10

3.3 Material of the model . . . . 13

3.3.1 Foam . . . . 13

3.3.2 Leather Cover . . . . 14

3.3.3 Extrapolation . . . . 15

3.3.4 Material registration . . . . 16

3.4 The simulation . . . . 17

3.4.1 Mass control . . . . 17

3.4.2 Initial velocity control . . . . 18

3.4.3 Coordinate system transformation . . . . 19

3.4.4 Control of the simulation . . . . 22

4 RESULTS & DISCUSSION 25 4.1 Materials . . . . 25

4.2 Qualitative outcomes from a visual perspective . . . . 27

4.3 Kinematic analysis . . . . 29

4.3.1 Restoration . . . . 29

4.3.2 Application . . . . 30

5 CONCLUSION & FUTURE WORK 33

Appendix A The arm model 38

List of Figures

1 The skeletal structure of the arm . . . . 4

2 The structure of the shoulder joint . . . . 5

3 The structure of the forearm and the elbow joint . . . . 5

4 The two joints inside the elbow join . . . . 6

5 The head model from KTH . . . . 7

6 The linear velocity of the different points on the arm and the angular velocity of the head . . . . 10

7 The motion of the arm during a striking . . . . 11

8 The construction of the wrist in the model . . . . 11

9 The model of the glove . . . . 12

10 The elbow joint in the model . . . . 13

11 The shoulder joint in the model . . . . 13

12 The foam of the glove . . . . 14

13 The leather of the glove . . . . 15

14 The extrapolation of the stress-strain curve . . . . 16

15 The stress-strain curves of the foam samples . . . . 16

16 The reference nodes for transformation . . . . 21

17 The accelerometer in the head model . . . . 22

18 The stress-strain curves of foam samples . . . . 26

19 The stress-strain curves of leather samples . . . . 27

20 The linear velocity of the fist at the f rame rate

s from the beginning of the simulation . . . . 29

21 The rotational acceleration of the head . . . . 31

22 The linear acceleration of CG of the head . . . . 31

23 The material card of the arm 1 . . . . 38

24 The material card of the arm 2 . . . . 39

25 The card of the joints . . . . 40

List of Tables

1 The suggested thresholds for MTBI . . . . 4

2 The average ratio of weight of segments as a percent of total body weight . . . . 6

3 The parts in the head model . . . . 7

4 The information required for the simulation . . . . 17

5 The mechanic information required for the simulation . . . . 18

6 The setup of the initial velocity . . . . 19

7 The registration sequence . . . . 23

8 The dimension of samples . . . . 25

9 The parameters of fit functions . . . . 25

10 The dimension of leather samples . . . . 26

11 The parameters of fit functions . . . . 26

12 The result of qualitative comparison between the video and the simulation . . . . 28

13 The linear velocity of the fist at the f rame rate

s from the beginning of the simulation . . . . 29

14 The velocities of the head at the f rame rate

s after struck by

the fist . . . . 30

1 INTRODUCTION

Traumatic Brain Injury (TBI) is one of the leading public health issues around the world, especially in children and young adults. In 2009, around 2.4 million TBI-related cases are reported in the United States [1]. Unfortunately, the real number could be even higher. The number of TBI cases that has been ignored in the official estimates could even be five-fold larger than the number that is recorded [2]. It is mainly due to that most of TBI cases (70-90%) are mild traumatic brain injury (MTBI) [3]. Symptoms that follow MTBI can be very mild neurometabolic brain changes that are rapidly recovered [4], but it can also be severe and lead to long-term damage in the brain. Without a following observation on the patients, it is easy to ignore the potential development of MTBI [3]. Therefore, only a small part of them are recorded and confirmed as TBI in hospitals. The lack of adequate and solid statistics impeded the development of research on MTBI.

On the other side, every year many physical injuries are reported and well documented in various intensive sports like American football, ice hockey and boxing. The injuries varies in different sports but for each sport, particular injuries occur more frequently than the other. For example, in the boxing match, head (including face) is the region where the injuries mainly happen due to constantly striking during the game. Specifically, concussion (also known as MTBI) is the second most common injury in those head injuries (after injuries in the eye area) [5]. Besides, the health situation of a professional boxing player is easy to follow. Therefore, boxing match can offer plenty of sound medical data that can be used in studies. The study of injuries in the sports may not only help to improve the protection technology in the game but also broaden and deepen the understanding of impacts of MTBI.

This thesis was aimed to set up a simulation to evaluate the reliability of the

video analysis and offer an alternative choice for kinematic research on intensive

sports such as boxing.

2 BACKGROUND

2.1 Injuries in boxing

In a boxing match, one of the way to end a game is to strike intensively on the opponent ’s head to cause a knockout or a technique knockout. A knockout (KO) is counted when a player is no longer able to continue the match due to the striking. It means he may fall down to the canvas and cannot stand up on his feet for more than a specific time period. The reason can be due to exhaustion, pain, unconsciousness and disorientation. Except for this,there are also other situations that can result in an end of the game. When the player is standing on his feet but has no longer the ability to defense oneself, the referee ends the game and counts this case as a technique knockout(TKO).

When the player is knocked out, especially suffering symptoms such as change of sensation or consciousness, it is highly possible due to a MTBI. MTBI may cause neurometabolic changes temporarily or permanently. Some of the damages can develop as the time goes and long-time exposure to MTBI may lead to disease like Dementia Pugilistica(DP), known as punch-drunk syndrome.

DP is a neurodegenerative disease that has also been found undergoing very sim- ilar pathological progress as Alzheimer ’s disease [6]. The study of MTBI in the boxing matches may lead to a deeper apprehension of how the DP was devel- oped, which may also help to better understand the pathology of Alzheimer ’s disease.

2.2 Head injury evaluation

The impacts and types of head injuries are highly related to the kinematics of

the head during the incidents. The relationship has been studied over decades

and many head injury criteria have been reported to evaluate the damage of and

predict the thresholds for different types of head injuries [7]. Although different

criteria have different interpretations of limits for causing head damages, most

of them include rotational acceleration as a significant parameter to evaluate

the possibility of an injury. Among the criteria, peak linear acceleration(PLA)

method [7] and Head Injury Criterion (HIC) focus on translational acceleration

[8] while methods like Rotational Injury Criterion (RIC) [9] are based on the ro-

tational acceleration. Moreover, criteria like Head injury power (HIP) considers

impacts of both translational and rotational acceleration [7, 10]. However, this

method is also less commonly used [7]. Table 1 shows human brain tolerance to

translational and rotational acceleration that is applied on the head during an

abrupt movement of the head.

Table 1: The suggested thresholds for MTBI

Acceleration Probability acceleration

Translational (PLA) [11]

25% a = 559m/s

50% a = 778m/s

75% a = 965m/s

95%[12] a = 1131m/s

Angular[11]

25% α = 4384rad/s

50% α = 5757rad/s

80% α = 7130rad/s

2.3 Anatomy of the arm

By bones, muscles and joints, arms are equipped with a flexible mechanic system to accomplish a series of motions. The mechanic system allows the player to hold the fist, to aim at the opponent and to throw out a powerful strike.

2.3.1 The structure and function

The skeletal structure of the arm (apart from hand and shoulder) contains three long bones: humerus, ulna and radius. As shown in Figure 1, humerus is in the upper-arm and is connected by the shoulder joint to the trunk. Ulna and radius are in the forearm and are connected by the elbow joint to the upper-arm.

Hands are linked by the wrist to the distal end of the forearm.

(a) Anterior aspect of the shoulder

(b) Lateral aspect of the fore- arm

Figure 1: From Primal Picture 2015

The shoulder joint, namely glenohumeral joint, is a multiaxial synovial ball

and socket joint (see Figure 2). The joint is controlled by a rotator cuff which

consists of four muscles: supraspinatus, infraspinatus, teres minor and sub-

scapularis. Together with deltoid, teres major, coracobracialis, those muscles

can hold the arm inside the socket joint and enable shoulder flexion, extension,

abduction, adduction, medial and lateral rotation.

(a) Muscles of the scapular re- gion anterior aspect

(b) Muscles of the scapular re- gion posterior aspect

Figure 2: From Primal Picture 2015

The elbow joint is a synovial hinge joint that connects humerus with ulna and radius. The joint consists of three sections that have been capsuled in one joint capsule. As shown in Figure 3c, the joint between trochlear notch of the ulna and trochlea of the the humerus, namely humeroulnar joint, is the main part of the elbow joint. It provides the elbow with flexion and extension.

That is also where ”hinge-joint” comes from. Under the control of brachialis, brachioradialis, biceps brachii and triceps brahii, the flexionextension range can reach from 0

to 145

actively and to 160

passively [13].

(a) (b) (c)

Figure 3: From Primal Picture 2015. (a) Deep extensor muscles of the forearm;

(b)Deep flexor muscles of the forearm; (c) Bones of the elbow joint sagittal section.

The other two sections inside the elbow joint capsule are located at the

head of the radius: one is attached to capitulum of the humerus and the other

to radial notch of the ulna. Those two joints enable two more movements for

forearm: pronation and supination. The normal range of motion for those two

movements are usually from -50

to 50

[14].

Figure 4: From Primal Picture 2015. It shows the postion of the two joins that are located at the head of radius.

2.3.2 The distribution of mass

In 1969, Charles E. Clauser et al. measured samples of 14 sections of a human body which were dissected from 13 male cadavers [16]. The ratio of weight of hand, forearm and upper-arm respectively as a percent of total the body were calculated. The study allows a reliable estimation of arm weight to later research where only the total body weight is known. Table 2 shows the results from the study.

Table 2: The average ratio of weight of segments as a percent of total body weight

Segments Ratio

Hand + forearm 2.27%

Upper arm 2.63%

Body except for one total arm* 95.10%

*It was calculated in this thesis

Since it was impossible to measure the exact dimension and weight of the arm of every case in this thesis project, an estimation of the player ’s mass dis- tribution information were made based on Charles E. Clauser et al. ’s research.

2.4 Previous Study

This thesis used a head model and a set of velocity data of boxing matches

from two other researches that have been accomplished before in the Unit of

Neuronic in the Department of Medical Engineering in the School of Technology

and Health (STH) of the Royal Institute of Technology (KTH)[15][17].

2.4.1 The model of head

The head model used in the thesis is provided by Department of Medical Engi- neering in STH, KTH, shown as Figure 5.

(a) Frontal view (b) Side view

Figure 5: The head model from KTH

The model is consisted of six parts which build up the major skeletal struc- ture of the skull. The six parts are shown below in the Table 3. Part 102 (solid) and part 104 (shell) consist of the cranial bones and upper part of the facial bones; part 141 represents the rest part of the facial bones; part 111 is used as the skin covering the whole head; part 1011 is a shell that used to control the total mass; part 199 represents the ring around foramen magnum, which is a rigid part and is used as a foundation that all other parts can be attached to.

Table 3: The parts in the head model

Part 102 Part 104 Part 111

Part 141 Part 199 Part 1011

2.4.2 The data from video analysis

The previous study in Dept. of Medical Engineering presented an analysis of 25 KO cases and 10 non-KO cases in professional boxing matches [17]. By using video analysis technology, the kinematic data has been acquired and the difference between KO and non-KO was evaluated. The kinematic data contain position, velocity and acceleration information of both the arm and the head during a strike. The study only focused on the kinematics of the head while the data of the arm was not considered.

2.5 The goal

In this thesis, a simplified model of arm will be established, to which the kine-

matic information of the arm will be applied. Then the simulation of striking

would be carried on to the head model from KTH. Finally, the outcome would

be monitored and compared to the outcome from the previous study.

3 METHODOLOGY

The aim of the thesis was to use LS-DYNA to restore a striking in a real boxing match scenario and to analyze the kinematics during this process. Also, to spec- ify the terminology that was used in the project, two pronouns were introduced.

A striking player was referred to the boxing player who threw out a punch, and a struck player represented the player who suffered the punch. According to the response of the struck player after the punch, the video clips in the database included two different cases: the ones where the struck player was knocked out (KO) and the ones where the struck player was judged as lack of ability of self defense by the referees (TKO). The two different cases were treated with no difference in this thesis. A kinematic analysis was then performed on all the cases and outcomes were also compared with the video analysis on the same video clips in a previous research[17].

3.1 Case analysis

During a striking, three phases were identified, which were defined as accelera- tion, interaction and re-bounce.

In a typical case (see Figure 6), the punch was started with an acceleration period in which the shoulder, elbow and the fist were accelerated accordingly to maximize the velocity of the fist. The duration varies from cases to cases on the 100 ms scale. In this period, the waist spun and the body turned, with the help of which the whole arm was quickly thrown towards the target. Then, the upper-arm and lower-arm were adjusted that the fist was able to precisely reach the target. Since the position of forearm and arm varied in different phases and cases (see Figure 7), adequate flexibility of the model was required to properly control the motion of the fist in the simulation.

The following period was the interaction period, which was a very short pe- riod usually lasting for 30-50 ms. In this period, the fist punched the head.

During the impact, the fist transferred most the energy to the head; the head passively suffered and absorbed the energy. Occasionally, the velocity of the fist was still increased in the very beginning because the striking player would contract the muscle in the upper-arm to produce an extra acceleration to max- imize the striking impact on the struck player’s head. However, the increase of velocity at this moment was insignificant comparing to the previous period.

Therefore, in this short time window, the striking was simplified into a model where a free weight object, the fist, with an initial velocity hit a static weight subject, namely, the face. This was done to lower the complexity of the system.

The last period was the re-bounce period. In this period the punch was

finishing and the striking player’s arm were pulled back from the hit and was

set up in the original position in case another striking was needed. The velocity

of the arm was decreased and became more irregular. However, on the other

side, the head, after pushed away from the original position, was translated

along the direction of the strike and rotated around the axis of neck. In the

beginning of this period, the head was treated as a free subject to lower the

Figure 6: The yellow, green and red curve shows the change of linear velocity of the striking player ’s fist, elbow and shoulder respectively in 3D space ; the blue curve shows the angular velocity of the struck player’s head in 3D space;

the yellow vertical line is the marker that marks the frame when the striking occurs.

complexity of the system since the duration of interaction was too short. Then, when the ligament, muscle and skeleton structure between head and neck was stretched/compressed, the constrains from the neck increased until a limitation was reached and the head stopped. It was then bounced back like a spinning fan hitting an object. During this short period, however, the head was no longer a free subject and reaction between the neck and the head should be taken into consideration.

Finally, to simplify the simulation further, Only few frames would be taken into consideration. The simulation would start one frame before the striking happened and stop at one frame after the striking was done. Therefore, the size of the simulation was greatly decreased down. Also several assumptions were built up, including that the head of the struck player was static before he was punched; Also, the body of the striking player was assumed to be initially static, and the constrain from the neck was ignored.

3.2 Geometry of the arm

To conduct the simulation, an arm model was created. The geometry of the arm in this project was designed in a way that not only satisfied the requirements for the flexibility of the arm, but also required low complexity.

The model of the arm was consisted of six parts and two joints. The parts included a fist-forearm combination, a glove, an upper arm and a body mass ball. The parts were connected either by sharing common nodes or by a joint.

There were two joints in this model: the elbow joint and the shoulder joint.

The fist-forearm combination, as the main part of the model, contained a

forearm part and a fist part. Considering the fact that the wrist of the striking

(a) (b)

Figure 7: (a) and (b) are two frames in the acceleration period of a striking.

The change of the relative positions of three points suggests independent degree of freedom for each joint.

player was barely bent during the match, the fist was merged into the fore- arm as one part (red part in Figure 8) and the total mass of those two objects were controlled as 2.27% of the total weight of the striking player[18]. The geometry of the forearm part was a solid cylinder. The length of the cylinder was decided by measuring the distance between the elbow and the hand of the striking player in the video clips. The geometry of the fist was created from the geometry of a real boxing glove. As shown in Figure 9, a 3D model of the glove

Figure 8: The fist and the forearm is merged into one part

was constructed based on the photo of the real glove, and the thickness of the foam was measured. Then, the yellow part was cut out as the foam according to the measurement; the rest part became the fist. In this way, the foam was connected to the fist by sharing a common layer of nodes. To do so, when the fist was moved the foam would follow tightly and naturally.

The glove was covered by a layer of leather. The leather cover was generated from the volume of the foam and the fist to enhance the stiffness of the foam. It shared common nodes of surface of those two parts and was assigned as a shell with a thickness of 0.8 mm, which was measured from the real boxing glove.

The upper arm was built up following the same strategy of the forearm.

(a) Blueprint of the glove (b) The mesh of the blueprint (c) The final model

Figure 9: (a) The side of glove model is sketched up first based on the mea- surement of the real boxing glove. The red frame shows the outline of the side of the model.(b) The side of the glove model is meshed up and the volume is generated by extruding this 2D meshed surface in the third dimension.(c) is the final model of the glove. Yellow part is the foam and brown part is the fist.

The geometry was simplified into a cylinder; the length was measured from the video clips; the weight was controlled as 2.63% of the total weight of the striking player[18] .

The last entity in this model was the body mass ball. It was an abstract solid ball representing as the torso. The shoulder was usually considered to include only a part of the chest. However, in this model, the initial velocity of the shoulder was not considered. Therefore, there should be a mass object to control the free end the upper arm while the mass of this object should reflect the active mass of the shoulder. In a prototype test, it was found out that the static mass of the shoulder was not enough to control the arm. Therefore in the end, the mass ball was assigned with the weight of a body without the arm [18].

Except for the foam and the fist-forearm combination, the rest of the parts were connected by joints. The joint between the forearm and the upper arm was the elbow joint. The elbow joint is a synovial hinge joint. The join offers one degree of freedom for the forearm; however, the limited movement of the remote ends of radius and ulna can also offer another degree of freedom for the arm.

Therefore, the arm can accomplish flexion, extension, pronation and supination.

As a result, a CONSTRAINED JOINT SPHERICAL card, instead of a Rev- olute card, was used to function as a combination of those structures.(Figure 10)

The shoulder joint connected the upper arm and the mass ball. The shoulder

joint is a socket joint that allows the arm rotating freely in three directions

around the joint node. Therefore, a CONSTRAINED JOINT SPHERICAL

card was applied.

(a) Elbow joint (b) Elbow joint (top)

Figure 10: Two sets of four nodes form a revolute joint connecting the two arms.

(a) Shoulder joint (b) Shoulder joint (Wireframe)

Figure 11: One set of two nodes forms a spherical joint functioning as the socket joint of shoulder

3.3 Material of the model

The glove is composed of leather and PVC foam. Leather is the cover and the PVC foam is the fillings. The mechanic properties of the PVC foam and leather cover were tested in the Neuronic Engineering Laboratory. Experiments were designed to find out the Young’s modulus and the stress-strain curve of the materials.

3.3.1 Foam

The foam of gloves were divided into three isolated parts: the main part is the thickest and largest piece of foam that protects the upper side of the whole hand; the thumb part is a smaller part at the side of the gloves protecting the thumb; the inner part is a thin layer of foam under the palm that offers support when the fist is holding and protects the whole palm. The main impact during a striking happens on the front surface, so only the front part of the foam was considered in the simulation.

The foam was treated as a homogeneous material in the model. However,

the real foam was molded with curved surface. The density and other mechanic

properties of the foam may vary radially. To simplify the model, the samples were taken along the radial direction so that the mechanic property parameters that was measured was approximately equal to the average value of the foam.

Figure 12 shows how the samples were selected and cut.

(a) Foam(side view) (b) Foam Samples

Figure 12: (a) A,B,C are three areas where the samples are taken.For each area, three square columns are cut out.(b)Samples are shaped properly and samples with poor conditions are excluded.

In the test, each sample was tested axially along the direction which was vertical to the curved surface. Since the material was not non-Newtonian ma- terial, the stress was not depended on the shear rate. Therefore the proceeding velocity of the test was set up less than 0.01 m/s which was easy to manage although it was far slower than the impacting velocity in the reality.

3.3.2 Leather Cover

The cover of the gloves consists of two materials: leather, for the main part

of the gloves where the main impact occurred, and synthetic leather, for other

insignificant area. As shown in Figure 13, sample A (A1, A2) were synthetic

leather taken from other insignificant area while B, C, D, E were leather which

were taken from the main part. Since the synthetic leather covered only palm

or thumb which was not considered as important during the simulation, only

the leather samples B, C, D and E were tested.

(a) Leather(front view) (b) Leather Samples

Figure 13: (a) B,C,D,E are four areas where the samples are taken.(b) Samples are shaped and samples with poor conditions are excluded.

To perform the test, the two ends of the leather sample were fixed between two forceps which pulled the sample apart, and both the stretching distance and force against the forceps were recorded and plotted by the computer.

3.3.3 Extrapolation

Since the surface of the foam sample was spherical instead of flat, the sample was compressed to reach a stable state before the measurement started. To get a complete curve, an extrapolation was used. A normal stress-strain curve of the foam contained three part: linear elasticity, plateau and densification. Since the missing part was in the very beginning near the linear elasticity. After several trials, an exponential function was used. 20 data from the beginning of the raw curve were chosen to perform the extrapolation in an online data processing website, Plotly. As shown in the Figure 14, a fitting function was produced that extrapolate the curve to find the theoretical origin of the curve. Then the curve was translated along the x-axis so that the theoretical curve was coincided with the coordinate origin. Such process was applied to stress-strain curve of each sample. Each sample gave a stress-strain curve. When overlapping them in the same figure (see Figure 15), it shows that the curves vary slightly when the strain is low and the variation dramatically increases after the plateau region.

Alternatively, a fitting function should be calculated from all the Stress-strain

curve. However, it requires a huge amount of computation and the improvement

to the final simulation was considered as insignificant comparing to the time

cost, so this step was not performed. Instead, a curve that is approximately in

the middle among the curves was chosen as the one for the model. Similarly,

the extrapolation and calibration of origin was also applied to the stress-strain

curve of the leather samples.

Figure 14: The blue part is the raw data; the orange part is the data from extrapolation. The fitting function is marked aside.

Figure 15: The stress-strain curves of different foam samples. The curves vary slightly when the strain is low.

3.3.4 Material registration

To build up the model, certain material must be assigned to each part. The material for the human body, the foam and the leather is *MAT RIGID TITLE,

*MAT LOW DENSITY FOAM TITLE and *MAT PLASTICITY POLYMER TITLE,

respectively. Density is another factor that was restricted accordingly. For the

part of human body, density was calculated to control the total mass of the part and stiffness was chosen from the KTH model. For the foam and the leather, density was also used to control the total mass, which was based on the previ- ous research as Section 2.3.2; stiffness was assigned with the stress-strain curve measured in the experiment.

3.4 The simulation

The simulation was performed in LS-DYNA, with the head model from KTH.

Each simulation was based on a real striking case in the boxing match. The kinematic information of the cases such as position, velocity, acceleration of the subject have been obtained from the study by using SkillSpector. The data is from Enrico ’s MSc thesis in 2011[17]. It contains the position and velocity information of shoulder, elbow, fist(finger in the analysis) of the striking player, and same information of forehead, ears, nose and chin of the struck player.

3.4.1 Mass control

18 cases from KO group and 6 cases from non-KO group were first pre-tested to evaluate the availability for the simulation. Eventually, 7 KO cases and 3 non- KO cases were categorized as ”available”. The rest were abandoned for reasons as missing of important coordinate information and wrong interpretation of coordinate during the 3D reconstruction in the previous study. Table 4 showed the information that was required for simulation of each available case.

Table 4: The information required for the simulation

Case

Frame rate (fps)

The striking player (A)

The struck player (B)

Body weight A (kg)

Body weight B (kg) KO group

01 50 Johnson Guthrie 78.0 78.9

02 50 Whitaker Lomeli 61.0 61.0

03 60 Trinidad Zulu 66.7 66.7

04 60 Jackson Cardamone 72.6 72.6

05 90 Tackie Garcia 62.1 61.7

06 90 Tua Nicholson 112.0 101.2

07 90 McCallum Curry 67.7 69.9

non-KO group

08 50 Taylor Gonzales 64 64

09 60 Trinidad Vargas 69.9 69.9

10 60 Tua Maskaev 101.2 105.6

Before the simulation, the length and the weight of the arm was registered

accordingly. The raw data from the video that were required for this process

included: the position of the marker at finger (fist), elbow and shoulder of the striking player; the position of the marker at nose, two ears, forehead and chin of the struck player. Table 5 showed the mechanic information of each case that was calculated from the raw data.

Table 5: The mechanic information required for the simulation

Case

Forearm Length (mm)

Upper- arm Length (mm)

Forearm weight (kg)

Upper- arm weight (kg)

Body weight*

(kg) KO group

01 447.25 324.98 1.7706 2.0514 74.178

02 304.44 433.73 1.3847 1.6043 58.011

03 330.97 305.62 1.5141 1.7542 63.432

04 348.03 324.54 1.6480 1.9094 69.043

05 467.65 358.58 1.4097 1.6332 59.057

06 489.57 280.16 2.5424 2.9456 106.51

07 373.04 336.68 1.5822 1.8331 66.285

non-KO group

08 297.70 191.16 1.4528 1.6832 60.864

09 382.96 327.84 1.5867 1.8384 66.475

10 362.26 277.07 2.2972 2.6616 96.241

*It refers to body weight apart from the whole arm

Moreover, the weight of the glove was also controlled. According to World Boxing Federation Rules & Regulations Of Championship Contests[19], the weight of the glove was regulated by the weight level of the player. There- fore, the total weight of glove (including the foam and the leather cover) was controlled by the conditions below:

m

=

( 226.8g if m

> 66.68kg 283.5g if m

< 66.68kg

3.4.2 Initial velocity control

Finally, linear velocity of all the markers and the angular velocity of the upper-

arm segment and the forearm segment of the striking player were collected .

They were then transformed from the video coordinate system to the LS-DYNA

coordinate system. Table 6 showed the initial velocity set up for each case. It

included the angular velocity of upper-arm and forearm and final linear velocity

of the fist in the simulation after the compensation. Due to problems described

before, some velocity were adjusted following the rules in section 3.4.4.

Table 6: The setup of the initial velocity

Case

Angl. V

(Fore-

arm)(rad/ms)

Angl. V

(Upper- arm)(rad/ms)

Linr. V

(Fist)(mm/ms) KO group

01 0.0232819 0.0169636 6.759991

02 0.0239895 0.0126597 5.007398

03 0.0256748 0.0359801 5.239075

04 0.05018 0.03545 13.34703

05 0.0510569 0.0216838 23.57395

06 0.0836251 0.0839471 46.98307

07 0.0172138 0.0192819 8.296985

non-KO group

08 0* 0.0463052 6.285333

09 0.0182121 0.0234484 2.759190

10 0.091751 0.0327514 7.052865

*The angular velocity of the forearm was not reliable. There- fore Rule 5 was applied and the forearm was only assigned with the linear velocity.

The initial linear velocity of the fist in case No.19 was comparatively high.

It may have to do with an artifact in the video analysis in the previous study, where the software failed to identify the right position of the marker in the fist.

3.4.3 Coordinate system transformation

The raw data was measured in the coordinate system that was created based on the video. To use those data, a coordinate system transformation was required.

The transformation was a rigid transformation; therefore, spacial relationship, distance, shape of objects were remained the same. LS-DYNA contained a function to perform such a transformation; however, it required of plenty of manual work. Therefore, a simple method was used instead. Since the head was similar and relatively static before the striking, three reference points were taken on the nose, forehead and chin from both the model and the video (see figure 16a and 16b). Once registered those two sets of three nodes together, any nodes can be transformed between the coordinate system in the video and the one in the model. Furthermore, if the introduced nodes were actually components that can define a coordinate system, the coordinate of any node from the video in the model can be calculated directly by the definition of the coordinate. Shown as below.

If r

is a vector in coordinate system (1), then we have r

= xˆ i

+ yˆ j

+ zˆ k

Then, if ˆ i

,ˆ j

,ˆ k

were transformed first into the coordinate system (2),

we have

ˆ i

= a

1ˆ i

+ a

2ˆ j

+ a

3ˆ k

ˆ j

= a

1ˆ i

+ a

2ˆ j

+ a

3ˆ k

ˆ k

= a

1ˆ i

+ a

2ˆ j

+ a

3ˆ k

Therefore, we have

r

= x y z



 ˆ i

ˆ j

ˆ k



 = x y z





a

a

a

a

a

a

a

a

a







 ˆ i

ˆ j

ˆ k



 = r

Therefore, to get the matrix, a set of nodes was introduced into LS-DYNA:

A(1000,0,0), B(0,1000,0), C(0,0,1000), O

(0,0,0) and P(1000,1000,1000) (see Figure 16c). Choosing 1000 instead of 1 was due to the requirement of precision.

Then, the transformation tool in LS-DYNA was applied to get the new coor- dinate of the origin O

and the matrix. Finally, if a node instead of a vector was needed, one would treat the node as a vector first when it was introduced in LS-DYNA and add the transformed node(vector) with the coocrdinate of the new origin.

n

= r

+ O

With this method, all the velocity and position information can be easily

transformed from the video to the model (see Figure 16d). The processed in-

formation was then used to set up the initial position and velocity of the arm

for the simulation.

Figure 16: a) and b), reference nodes include three main nodes on the nose, forehead and chin and two optional nodes on the left and the right ear. The nodes in LS-DYNA are based on the ones in the video; c) shows the trans- formation coordinates and the four nodes for calibration; d) shows the results of transformation, in which the position of the nodes for the arm (91000201, 91000202, 91000203)has been calculated and transformed.

(a) (b)

(c)

(d)

3.4.4 Control of the simulation

The main goal of the project was to evaluate the reliability of the model. There- fore, a kinematic analysis was carried on to see if the data match the criterion of causing a concussion, which was the main reason of the knockout. In the kinematic analysis, the rotational acceleration of the head were calculated and compared to the previous study [17]. It was hard the measure the kinematic information of the whole head. Instead, three nodes were added on and around the center gravity of the head, and they were also constrained with the rigid part of the head. Then the rotation and translation of the head can be measured and recorded by measuring and monitoring the movement of those three nodes.

As shown in Figure 17, 90000016 was on the center gravity point of the head, and the line that went through 90000016 and 90000017 was perpendicular to the line between 90000016 and 90000018.

(a) (b)

(c)

Figure 17: Nodes on and around the center of the head. 17a shows the position of the three nodes in the head; 17b shows the relative position of the three nodes; 17c shows an accelerometer is applied on the three nodes.

The model was registered with parameters sequentially to make sure the fist

was moving with the right velocity in the beginning as the one in the video. The

necessity of the 3rd step is that the ball that represents the body was massive and had no measurement of velocity. It slowed down the whole arm once the simulation started. Therefore an correction was required to compensate the significant decrease in the velocity of the fist.

Table 7: The registration sequence Sequence Parameters Description

1

Magnitude of angu- lar velocity of each segment

The absolute value of the angu- lar velocity of upper arm and forearm-hand combination.

2 The rotational axis of each segment

The origin was chosen as the elbow for the forearm and the shoulder for the upper-arm. The direction of the axis was calcu- lated and transferred to the co- sine of the axis to x, y and z axis.

3 Linear velocity

Usually, there is a difference of the linear velocity of the fist between the one in the video and the one after the first two steps above in the simulation.

Therefore, an extra the linear ve- locity would be applied to the forearm, the foam of glove and the cover of glove as a com- pensation. It was calculated by v

= v

− v

However, due to the low quality of the video, limited angle of view or different size of the head, some inaccurate or wrong interpretation of markers during the 3D reconstruction was identified from the data from the previous study. It can lead to problems like overlap of the head and the arm in the initial state, wrong position of the nodes, wrong direction of speed and etc. Therefore, a manual adjustment was also performed in some cases.

Additionally, the sampling rate of the simulation was set 10 times higher

than frame rate in the video so that the precision of the simulation can be

controlled and the results of the simulation were compatible with the results of

the kinematic analysis.

4 RESULTS & DISCUSSION

4.1 Materials

The foam The results were recorded as a file containing data of time, stretching distance and force against the sensor. Table 8 shows the dimension of different samples. The samples were cut into a square column with a and b as the width of each edge at the bottom and h as the height of the column.

Table 8: The dimension of samples

Foam Samples Side a(mm) Side b(mm) height h(mm) Section A

1 205.0 205.5 21.35

2 207.5 221.0 21.65

3 229.0 222.0 17.30

Section B

1 211.0 249.0 48.20

2 183.0 239.5 52.60

3 215.5 228.5 50.50

Section C 1 146.0 202.5 36.75

2 140.0 234.5 35.40

Since the data of the beginning of the curve were missing. To complete the curve, an extrapolation was performed.

Table 9 showed the extrapolation result of the raw data. The parameters in the table were defined as in the predefined fit function below:

f (x) = a + b exp(cx)

Table 9: The parameters of fit functions

Sample a b c R

A1 0.01124 -0.005091 -27.16 0.8826 A2 0.01779 -0.008037 -18.91 0.9611 A3 0.01659 -0.008739 -16.25 0.9703 B1 0.01570 -0.007344 -18.10 0.9467 B2 0.02188 -0.01664 -9.362 0.9680 C1 0.01687 -0.01013 -23.17 0.6729 C2 0.01289 -0.006984 -36.38 0.9055

After calibration of the origin, the curves of samples were plotted as Figure

18. Since the simulation did not require a high precision of the curves, especially

of the part where the stress is higher than 60%, the curve in the middle of the

cluster of all the curves were chosen to be registered as the material stress curve

in the simulation.

Figure 18: The stress-strain curves of foam samples

The leather The dimension of the leather samples were shown in Table 10. Length represented the length of the tested part.

Table 10: The dimension of leather samples

Sample Width(mm) Length(mm) Thickness(mm)

B 26.05 66.45 0.80

C 27.50 65.45 0.80

D 26.50 67.00 0.80

E 25.75 66.70 0.80

Table 11 showed the parameters of fitting functions in the extrapolation.

The predefined fit function was also the exponential function Table 11: The parameters of fit functions

Sample a b c R

B 0.2295 -0.1242 -246.3 0.9242 C 0.3005 -0.1792 -153.4 0.9303 D 0.3992 -1.280 -97.39 0.9707 E 0.3087 -0.1886 -298.8 0.9467

Figure 19 illustrates the stress-strain curves of the leather samples. How-

ever, the similarities of the curves of different samples were relatively low. One

possible explanation was that the sample was not perfectly flat and the center

of the sample was protruding out which changed the mechanic structure of the

sample. Then, since the samples were cut out from different parts of a irregular

spherical surface, the curvature of the center of different samples was various.

Therefore, the response to the stress (stretch) varied.

Figure 19: The stress-strain curves of leather samples

Since the striking happened mainly within the area where sample B, C and D were taken and sample C was in the center of the area, the stress-strain curve of sample C was chosen to registered as the material stress curve in the simulation.

The material curve could be more precise by calculating the average curve.

It can be improved in the future study.

4.2 Qualitative outcomes from a visual perspective

The simulations were based on the realistic boxing match videos; therefore, a comparison of visual outcomes between the videos and the simulations were performed to give a qualitative evaluation of the performance of the model.

The table shows that the trace of the motion of the arm in the simulation

match with the one in the video. However, the head in the simulation moves in

a way that differs from the one in the video. It is due to the fact that the head

in the simulation was not restricted by a body since it is not the interest in this

project.

Frame1 Frame2 Frame3 Frame1 Frame2 Frame3

Skillspector

LS-DYNA

Skillspector

LS-DYNA

Skillspector

LS-DYNA

Skillspector

LS-DYNA

Skillspector

LS-DYNA

Skillspector 49.95

49.95 01

02

59.94

89.91

89.91 04

05

06

89.91 07

Table 12. The result of qualitative comparison between the video and the simulation

Case fps

Platform View 1 View 2

4.3 Kinematic analysis

4.3.1 Restoration

Table 13 showed the linear velocity of the fist in the video and in simulation respectively at the f rame rate

s from the beginning of the simulation. Also, Figure 20 illustrated Table 13

Table 13: The linear velocity of the fist at the f rame rate

s from the beginning of the simulation

Case Video (m/s) Simulation (m/s)

Difference

(based on

Video) KO group

01 8.479 6.195 -26.94%

02 4.848 3.176 -34.48%

03 9.699 4.380 -54.84%

04 8.415 12.419 47.59%

05 13.652 19.737 44.57%

06 28.776 19.751 -31.36%

07 2.203 3.995 81.33%

non-KO group

08 7.878 1.356 -82.79%

09 2.024 1.878 -7.19%

10 3.323 3.989 20.05%

Figure 20: The linear velocity of the fist at the f rame rate

s from the begin- ning of the simulation

A similarity in the pattern of data can be observed from Figure 20 between

the velocity of the simulation and the results from the video analysis. It qualita-

tively supports the validity of the model. However a further study is suggested

to quantify the similarity of results and validate the model quantitatively. On the other hand, the difference of two sets of results implies a poor precision of the model. Nevertheless, the precision of the results from the video analysis was not validated. Therefore, it can also be improved and verified in the future work.

4.3.2 Application

Table 14 showed the outcome of the kinematic analysis of the head. The analysis was performed on the rotational velocity (angular velocity) of the head and the linear velocity of the center gravity of the head in the simulation. The results were also illustrated in Figure 21 and 22

Table 14: The velocities of the head at the f rame rate

s after struck by the fist

Angl. A* (krad/s

) Linr. A** (g)

Case Simulation Video[17] Difference

(%) Simulation Video[17] Difference

(%)

KO group

01 18.3 4.78 283.4 130.8 19.5 571.5

02 9.79 7.59 29.0 72.0 27.0 166.6

03 4.04 7.59 -46.8 92.5 34.9 165.0

04 18.9 6.81 177.8 92.6 20.9 342.3

05 69.3 5.18 1236.0 467.3 46.1 912.8

06 218.3 8.22 2556.7 1547.4 64.2 2308.3

07 3.44 1.31 163.0 29.3 43.6 -32.9

non-KO group

08 40.2 3.91 929.9 236.0 26.8 571.5

09 1.82 5.35 -66.1 9.95 25.0 166.6

10 29.7 2.64 1024.1 198.3 15.6 165.0

*Angl. A stands for Angular Acceleration

*Linr. A stands for Linear Acceleration

Figure 21: The rotational acceleration of the head

Figure 22: The linear acceleration of CG of the head

As mentioned before, the data from video analysis were not always reliable.

Therefore, the small difference may cause dramatic artifacts. For example, when the position of the fist was mis-interpreted, the direction and magnitude of the velocity would be changed. Furthermore, the interaction between the fist and face was therefore changed as well. That may explain the abnormal artifact in case KO 19.

Another problem was that the head in the simulation was designed as a free

object, which meant that there was no constrain of the neck on the head as it was in the reality. Therefore, when the head was struck, it would spin faster than it was supposed to be. It may also contribute to high acceleration.

On the other side, the most results of kinematic analysis of group KO were

satisfied with the criterion for concussion. However, since sample number in

the control group, group non-KO was too small, it was not adequate to show

enough evidence that the result supports the criterion.

5 CONCLUSION & FUTURE WORK

A simplified 3D arm model was constructed and parameters were designed to restore the scenario in a real boxing match. The arm model offered two joints with three degrees of freedom each, adjustable dimension and realistic material property. The model was tested and proved to be able to restore the kinematic data from a third source.

However, the system was sensitive to the artifact from the data source. The

artifacts came from different reasons. For example, two cameras were occasion-

ally 180

to each other within the same plane, which can barely contribute to

the 3D reconstruction and finally lead to low accuracy of the data. Another

example could be that the fist or face was sometimes blocked by the opponents

or one ’s back. The missing of data at crucial moment can directly cause the

failure of analysis. Moreover, the low quality of videos decreases the accuracy

and precision of the data which can also result in artifacts. To overcome this

disadvantage, the measurement of motion capture on the live boxing cases can

be performed so that the quality of data source can be guaranteed. Addition-

ally, a further study can also focus on improvement of model structures and

materials to increase the model reliability. For instance, the installation of the

neck in the FE-model could be a beneficial direction to work with for it can

increase the reliability and accuracy of the simulation.

References

[1] Coronado, Victor G., et al. Trends in traumatic brain injury in the US and the public health response: 1995-2009. Journal of safety research 43.4 (2012):

299-307.

[2] Feigin, Valery L., et al. Incidence of traumatic brain injury in New Zealand:

a population-based study. The Lancet Neurology 12.1 (2013): 53-64.

[3] Feigin, Valery L. Handbook of clinical neuroepidemiology. Nova Publishers, 2007.

[4] Iverson, Grant L., and Rael T. Lange. Mild traumatic brain injury. The Little Black Book of Neuropsychology. Springer US, 2011. 697-719.

[5] Zazryn, Tsharni R., Caroline F. Finch, and P. McCrory. A 16 year study of injuries to professional boxers in the state of Victoria, Australia. British journal of sports medicine 37.4 (2003): 321-324.

[6] Pedro Pineda & Douglas J. Gould, The Neuroanatomical Relationship of De- mentia Pugilistica and Alzheimer’s Disease, 9 NEUROANATOMY 5 (2010), available at http : //www.neuroanat − omy.org/2010/005 007.pdf .

[7] Fernandes, Fbio AO, and Ricardo J. Alves de Sousa.Head injury predictors in sports trauma: A state-of-the-art review. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine 229.8 (2015): 592-608.

[8] J. Versace. A review of the Severity Index. SAE Technical Paper 71081, 1971.

[9] Kimpara H and Iwamoto M. Mild traumatic brain injury predictors based on angular accelerations during impacts. Ann Biomed Eng 2012; 40(1): 114-126.

[10] Newman JA, Shewchenko N and Welbourne E. A new biomechanical head injury assessment function: The maximum power index. Stapp Car Crash J 2000; 44:215-247.

[11] King A, Yang K, Zhang L, Hardy W and Viano D. Is head injury caused by linear or angular acceleration In: 2003 IRCOBI conference, Lisbon, Por- tugal, September 2003, pp. 1-10. Zurich, Switzerland: IRCOBI

[12] Newman J, Barr C, Beusenberg M, Fournier E, Shewchenko N, Welbourne E, et al. A new biomechanical assessment of mild traumatic brain injury.

Part 2 Results and conclusions. In: 2000 IRCOBI conference, Montpellier, France, 20-22 September 2000. pp. 223-233

[13] Kapandji, Ibrahim Adalbert (1982). The Physiology of the Joints: Volume

One Upper Limb (5th ed.). New York: Churchill Livingstone.

[14] Morrey, B. F., L. J. Askew, and E. Y. Chao. A biomechanical study of normal functional elbow motion. The Journal of Bone & Joint Surgery 63.6 (1981): 872-877.

[15] Kleiven, Svein. Predictors for traumatic brain injuries evaluated through accident reconstructions. Stapp Car Crash J 51 (2007): 81-114.

[16] Clauser, Charles E., John T. McConville, and John W. Young. Weight, volume, and center of mass of segments of the human body. ANTIOCH COLL YELLOW SPRINGS OH, 1969.

[17] Pellegrini, Enrico. Studio Della Cinematica Di Traumi Al Cervello Riscon- trabili Durante Incontri Professionistici Di Boxe. Thesis. The University of Padova, 2011.

[18] Clauser, Charles E., John T. McConville, and John W. Young. Weight, vol- ume, and center of mass of segments of the human body. ANTIOCH COLL YELLOW SPRINGS OH, 1969.

[19] (WBF)World Boxing Federation. WORLD BOX-

ING FEDERATION RULES & REGULATIONS OF

CHAMPIONSHIP CONTESTS. Web. 05 Sept. 2015.

¡http://www.worldboxingfederation.net/wbfrulesandregulations.htm¿.

37

A The arm mo del Figure 23: The material card of the arm 1

Figure 24: The material card of the arm 2

Figure 25: The card of the join ts