Design and manufacture of a novel system to simulate the biomechanics of basic and pitching shoulder motion

Design and manufacture of the testing system

In order to simulate the normal range of movement and pitching motion of the shoulder, a robotic system consisting of a lower frame (moving the torso) and upper frame (moving the arm) was designed to provide linear and rotational motion along seven axes. The lower frame was able to linearly move the torso in 3D and around its long axis (Z) (along X_TORSO, Y_TORSO, Z_TORSO, around Z_TORSO-θ) while the upper frame was able to linearly move in 3D the end point of the arm (along X_HAND, Y_HAND, Z_HAND) (Fig. 1). This setup allows for the free 3D linear motion of the hand and torso, in addition to the rotation of the torso about its long axis. The testing apparatus utilised high-voltage Parker stepper motors (Parker-Hannifin, Cleveland, Ohio) to affect positioning of the actuators and gearboxes in order to improve output (Table I).

Fig. 1

Schematic view of the testing system, showing the torso and hand frames along with the reference system for both frames.

Table I

The robotic system actuators’ axes with their corresponding travel lengths and normal loads. The prime notation indicates a parallel track connected by a link shaft. ‘Actuator type’ includes both the actuator axis and Parker-Hannifin model. The Z-axis actuators act along the normal axis and thus do not have a normal load. The theta-axis actuator has 360° of rotational capability. All actuators were manufactured by Parker Automation (Cleveland, Ohio)

Actuator model (Parker Automation)	Range of movement (mm)	Normal load (kg)
Upper (hand)
XX’-axis Dual ERV5	1832	114.8
Y-axis AAD ERV	1078	114.8
Z-axis ET50	582	N/A*
Lower (torso)
XX’-axis Dual ERV5	1513	114.8
YY’-axis AAD ERV	231	114.8
Z-axis ET50	151	N/A
Theta-Axis 200-series	2 radians	90.7

1. * N/A, not applicable

An adjustable posterior restraint was developed to generate the degree of external rotation required by the throwing motion (Fig. 2a). This device allows for proper positioning of the elbow during the late cocking phase of the throwing simulation and is easily manipulated to suit cadaveric specimens of heights between 1.60 m and 2.10 m.

Figs. 2a - 2c

Photographs showing a) a saw-bone skeleton mounted onto the system with the arm fixed to the upper frame actuator, b) the testing apparatus with the cadaveric torso and the five-camera setup, with two cameras located on the side, two above, and one on the back wall behind the torso, and c) the placement of the five marker clusters on a human torso, each with three or four individual markers, on the sternum, clavicle, humerus, scapula and forearm.

A sensor system was implemented to monitor the motion of all axes and robotic drives. This feedback system facilitates communication between the computer and the testing apparatus. The system includes Hall Effect end of travel limits (SMC-1N) and home sensors (SMH-1N) (Parker-Hannifin). The home sensors define points of reference and create the coordinate system for the apparatus. The travel limits protect the specimen or the apparatus by defining the safe range of motion. An additional encoder system was included in the design to provide closed loop negative feedback for the stepper motors to ensure precision and accuracy. In this way, each axis was equipped with an E5 optical rotary encoder (US Digital, Vancouver, Washington) allowing for a resolution of 1000 counts per revolution.

Motion analysis component

Motion data were collected using five Qualisys ProReflex (Qualisys AB, Gothenburg, Sweden) high-speed cameras surrounding the testing apparatus (Fig. 2b). The cameras sampled the motion at 120 frames per second and were calibrated to discern motions as small as 0.5 mm. A passive retro-reflective cluster with three or four markers was pinned directly into the each of the bones of interest (humerus, clavicle, scapula and thorax (sternum)). Marker clusters were arranged to decrease the possibility of contact between the clusters during movement and to avoid overlap and mislabeling during the analysis (Fig. 2c). Anatomical landmarks selected by the International Society of Biomechanics¹¹ were calibrated with respect to the bone-embedded marker clusters as described in Cappozzo et al.¹² The centre of rotation of the glenohumeral joint in the scapular reference system was determined using the method described by Meskers et al.¹³ The centre of the glenohumeral joint was assumed to be coincident to the centre of the humeral head in a given reference posture (that of a hanging arm). Therefore, glenohumeral translation could be determined in the scapular reference system. All joint kinematics were described in the reference system of the relevant proximal segment.

Validation of robotic system

The precision and accuracy of the axes of the system was determined by comparing the measured displacement of each axis with the relevant expected displacement. The expected displacement was calculated based on the average number of revolutions of the actuator motor during three independent trials. To ensure accuracy over the entire range, each axis was tested from homing sensor (Parker SMC-1H) to 10%, 50% and 90% of the total possible distance. In order to demonstrate that loading the actuators during the experiment did not affect their validity, the lower frame was subjected to a 57 kg (125 lb) weight and validation process was repeated.

Validation of combined axes motion

The reference frame of the motion capture system was defined by the calibration object provided by the manufacturer. The reference frame of the robotic system coordinate system was aligned with the upper and lower frames. In order to minimise the potential effect of any misalignment between the reference frames, the magnitude of displacement was reported for a number of movements of the upper frame end effector. The magnitude of the displacement of the end effector of the upper frame through trajectories diagonally crossing the calibrated volume, as recorded by the robotic system, were compared with the displacement recorded by the camera system (four diagonal trajectories: A, B, C and D; Fig. 3). Similarly, two diagonal motions were used in the lower frame (E and F; Fig. 3). Each trajectory was tested three times. During validation, a separate retro-reflective marker cluster was placed on the hand and torso actuators (base plate) and their motion was captured.

Fig. 3

Diagram showing the diagonal trajectories generated for the hand and torso actuators (Table IV).

Table IV

The precision and accuracy of the camera system to capture magnitude of displacement created for upper (hand actuator) and lower frames (torso actuators)

		Actual magnitude			Precision			Accuracy
Diagonal	Expected magnitude	Mean (sd)	CV* (%)		ICC†	p-value		Error	Error (%)	CC‡	p-value
Upper frame
A (mm)	863.8	858.7 (0.0)	0.0		1.0	0.001		5.2	0.0	1.0	0.001
B (mm)	863.8	863.1 (0.0)	0.0					0.7	0.0	1.0	0.001
C (mm)	863.8	865.2 (0.0)	0.0					1.4	0.0	1.0	0.001
D (mm)	863.8	868.2 (0.3)	0.0					4.3	0.0	1.0	0.001
Lower frame
E (mm)	1027.4	1022.8 (0.1)	0.0		1.0	0.001		4.6	0.0	1.0	0.001
F (mm)	1027.4	1017.9 (0.6)	0.0					9.5	0.0	1.0	0.001

1. * CV, coefficient of variation † ICC, intraclass correlation coefficient ‡ CC, Pearson correlation coefficient

Cadaveric shoulder model

In order to simulate the pitching motion, motion data from professional Major League Baseball (MLB) pitchers (courtesy of Dr. G. Fleisig, American Sports Medicine Institute (ASMI)), were used to define the motion of the robotic system axes. The velocity of the simulated motion was slowed to 1/750th of the real-time throwing motion in order to ensure a smooth motion without damage to the cadaveric torso. As the majority of shoulder injuries happen during the initial phases of throwing motion,^14,15 an abbreviated throwing motion (ATM) was defined with the data from the ASMI. It begins at maximal external rotation (late cocking) and continues through acceleration to ball release. The ‘mid-wrist’ coordinates were calculated by averaging the coordinates of the medial and lateral wrist markers in the throwing hand. The ‘mid-hip’ coordinates were calculated from the throwing and leading hip markers. In this model, the relative motion of the wrist to the body determines the shoulder‘s motion, the mid-wrist coordinates were re-calculated relative to the mid-hip. This treatment resulted in a described pattern of motion, with three degrees of freedom, which was then used to model the ATM motion using the hand actuator, while keeping the torso still.

The right and left shoulder of a 55-year-old fresh-frozen male cadaver (Medcure, Naples, Florida) were used for this study. A stainless steel frame was placed inside the thorax and secured with expandable polyurethane foam (Convenience Products, Fenton, Missouri) (Fig. 2a). The stainless steel frame was then mounted on the base plate. The hand was amputated at the wrist and a 5 mm Steinmann pin was drilled perpendicularly through the distal radius. An osteotomy of the distal ulna was performed to allow free rotation of the radius with respect to the proximal ulna, as performed in a Sauvé-Kapandji procedure in hand surgery.¹⁶ The specimen was then tested over the ATM, and its motion patterns were recorded.

Statistical analysis

The mean, standard deviation (sd) and coefficient of variation (CV) were calculated for the displacement of each axis in both loaded and unloaded conditions, and the expected (input data) travel for each actuator was compared with the directly measured observed (camera recorded data) travel to calculate the absolute and percentage error in movement. In order to determine precision, the intraclass correlation coefficient (ICC) for individual measurements was used for 10%, 50%, and 90% travel of each actuator. The Pearson correlation coefficient (CC) of the observed versus expected displacements for each actuator was used as a measure of accuracy. The same descriptive statistics were used for magnitudes of travel captured by the camera system. The ICC for individual measurement was used as a measure of precision, and the CC of the observed versus expected diagonal magnitude was used as a measure of accuracy.

The right and left shoulder of a cadaver were subjected to three trials of the ATM motion. Error bars representing the sd of glenohumeral translations and scapulothoracic rotations were reported.

The data analysis was performed using SPSS statistical software (version 17.0; SPSS Inc., Chicago, Illinois), and the level of statistical significance was set at p < 0.05.

Robotic system

The testing system demonstrated high precision and accuracy based on the expected versus observed displacements of individual axes. The maximum CV for displacement of unloaded axes was less than 0.5% for all axes (Table II). Individual axes demonstrated displacement precision as evidenced by ICC values greater than 0.99 (p = 0.001) for three repeated trials of 10%, 50%, and 90% of the total travel. The expected and observed actual displacements had high level of correlation with coefficients of determination of 1 for all axes (p = 0.001). The absolute and percent errors in displacement of unloaded axes were 0 and less than 0.9%, respectively (Table II).

Combined axes motion

The diagonal motion was reproducibly captured by the high speed cameras showing a maximum standard deviation of 0.6 mm between the trials, in any direction (Table IV). The camera-based measurement of the six diagonal motions (four from the upper frame and two from the lower frame) showed high precision and accuracy. The absolute differences between robotic system and motion capture system estimates in observed magnitude of diagonal trajectories were a mean of 2.9 mm (sd 2.2) for the upper frame and 7.0 mm (sd 4.0) for the lower frame.

Cadaveric shoulder model

The sd of the ATM motion was < 0.2 mm. The precision of the camera system in capturing three trials was very high with an ICC of 1 (p < 0.001). The observed and expected ATM magnitudes highly correlated with a correlation coefficient of 0.99 (p < 0.001). The absolute error in observed ATM magnitude was a mean of 3.9 mm, corresponding to a relative error of 0.8%. The glenohumeral (GH) translations and scapulothoracic (ST) rotations in the planes of X, Y and Z during ATM motion were determined for the right and left shoulders (Fig. 4). The maximum sdof the GH translations along the three axes were 2.2 mm, 3.4 mm and 1.7 mm, respectively, while the maximum sd of the ST rotations around the three axes were 0.9°, 1.2° and 0.7°, respectively.

Fig. 4

Graphs showing the glenohumeral (GH) translation (in mm) and scapulothoracic (ST) rotations (in °) for the right and left shoulder (red and blue, respectively) in the cadaveric shoulder model, versus time (in seconds). The GH translation is reported in the scapular reference system and ST rotation is reported in the thoracic reference system. Error bars showing the standard deviation are also reported for instants of time.