Automation of Bird Front Half Deboning Procedure: Design and Analysis

Transcription

1 Automation of Bir Front Half Deboning Proceure: Design an Analysis Debao Zhou, Jonathan Holmes, Wiley Holcombe, Kok-Meng Lee * an Gary McMurray Foo Processing echnology Division, AAS Laboratory, Georgia ech Research Institute, * School of Mechanical Engineering, Georgia Institute of echnology, Atlanta, GA Abstract Bio -material cutting, such as meat eboning, is one of the most common operations in the foo processing inustry. It is also the largest employer of people in the Unite States. hese tasks are currently manual processes with only limite use of fixe automation. he main ifficulty in this task is the natural variability of the prouct's size an iniviual anatomy. he inustry is looking to robotics to help solve these problems. his research has focuse on automating the cutting of chicken front halves to obtain high quality breast meat. In orer to specify the cutting locations an cutting trajectories on chicken front halves, in this paper, the anatomy structure of the chicken shouler joint was stuie first. hen a 2-DOF cutting mechanism was propose. hrough the formulation of the kinematics an ynamics of the mechanism, the cutting trajectory was simulate. Pneumatic actuators with position feeback sensors were selecte as the riven system. o verify whether the pneumatic riving system coul satisfy the trajectory following requirements (spee an response time), experiments were carrie out. he results show that the pneumatics riven system can marginally follow the esire trajectory with enough spee for the aaptation motion. he evice will be built an teste in our future research. Keywors: robotics, poultry eboning, pneumatic rives I. Introuction he proceure to manually harvest the chicken breast meat an wings (butterfly) is shown in Fig. 1. A front half is fixe on a cone as shown in Fig. 1(a). After the joint connections are severe, the wing an breast meat are separate from the carcass by pulling the butterfly (Fig. 1(b)) away from the carcass (Fig. 1(c)). he wing pulling motion is shown in Fig. 1(), where F is the pulling force irection. Consiering this process has to be manually repeate approximately 300,000 of chickens each ay, it is obviously very onerous. However, ue to the naturally eformable boies, size ifference an possible har bone chips in meat, it is very ifficult to be automate. One commercial solution is the automation eboning lines by the Stork Gamco Inc. [1]. However, their metho still belongs to the fixe automation category since they require the cutting motions be preset manually an thus, they cannot automatically ajust to changes in iniviual bir sizes. Meyn Inc. [2] also evelope a cutting evice. heir cutting evice has only one fixe motion for all the front halves eboning. In orer to aapt to the variations in chicken size, Delay et al. [3] propose a reference-point metho to estimate the locations of the cutting trajectory an these reference points were obtaine through the analysis of the computer images. Using a similar metho, Heck [4] propose to use water-jet cutting metho to cut chicken breast meat to obtain certain shapes accoring to the ientifie trajectory from computer images. Some research on eboning has also been carrie out on the pork or beef eboning such as [5] an [6]. he pork an beef eboning in these works is to cut through everything incluing har bones, while the eboning in chicken tries to avoi har bones in orer to obtain high quality butterfly. (a) Cone an front half. (b) Butterfly (wings + breast meat). (c) Carcass on cone. () Pulling away the butterfly. Fig. 1 Illustration to harvest chicken butterfly. Currently, no cutting evice is available for this automation with the capability to aapt to the size-change an boy-eformation. In this research, through the unerstaning of the anatomy of chicken shouler joints, a new processing metho associate with a simple mechanism was propose. Note that the scope of this paper is to esign a evice which has the ability to aapt the bio-material eformation uring eboning. he aaptability will be further stuie through the motion control an force control. In the following, the anatomy of the eboning relate chicken boy was stuie an the cutting trajectories were specifie in Section II. Accoring to the front half transport ation metho in inustry, the cutting system was specifie an the simulate trajectory following motion using the cutting mechanism was shown in Section s III an IV. he riven evice was selecte an verifie in Section V. After the iscussion in Section VI, remarks were specifie. F F

2 II. Deboning Relate Chicken Shouler Anatomy A. Chicken Shouler Anatomy he eboning relate components in a chicken boy s inclue the front half skeleton an the connection anatomy between the wing an the carcass. Fig. 2 shows the skeleton of a chicken front half. Reference [7] has shown the etaile stuy. However, their anatomy stuy cannot be irectly applie to the cutting evice esign. Fig. 2 Front half skeleton. Fig. 4 op ligament an meat. Fig. 3 Outsie view. Fig. 5 Humerus an ligaments. (a) op view. (b) View from wing irection. Fig. 6 Joint ligaments an tenon connection area. (a) op view. (b) View from the breast sie. Fig. 7 Front half cutting area ientification. As shown in Fig. 2, the scapula, coracoi an clavicle bones form a shouler bone girle. he right an left girles connect to the vertebrate through the ribs. he coracois connect together through the keel bone. In fresh chickens, the wish bones (clavicles) connect with the keel bone through soft tissues. he location an orientation of the humerus an breast meat relative to the skeleton is shown in Fig. 3. he area below the breast meat an between the coracoi an humerus is the starting portion of the tener meat. A ball-socket joint is forme between the humerus an the carcass by the ligaments, tenons an meat. When a fresh chicken wing is pulle, a small gap (about 2-5 mm) is forme between the humerus an the shouler bone girle. his gap provies the space for a blae to enter the joint through cutting the connecting ligaments. he gap increases when the top ligament is cut as shown in Fig. 4. here are 5 main ligamentstenons connecting the humerus to the carcass as shown in Fig. 5. here are three ligaments connecting the humerus an the coracoi, one tenon connecting the breast meat with the humerus an one ligament connecting the scapula an humerus. hese ligaments an tenons occupy about ¾ of a circle perimeter aroun the joint an leave about ¼ joint near the neck empty, as shown in Fig. 6(b). he carcass an bone structure without the breast meat an wings are shown in Fig 7(b). he joint length from top is about 25mm. As shown in Fig. 7(a), the istance between ruler an clavicle is about 25mm an the thickness of the top meat (BJ) is about 5 mm. he angle between the ruler an the upwar irection of the cone is about 17. Note that for ifferent size chickens, the above imensions are ifferent. he cutting evice shoul have the ability to ajust to aapt to the variations. B. Cutting Area an Cutting rajectory A frame oxyz is efine on the chicken as shown in Fig. 7. he upwar irection of the cone is the positive z axis. wo points are ientifie as reference points: in the z irection, the highest coracoi points on the left an right, where there is a small ent. he x axis is forme by connecting these two points. In Fig. 7, point A is the location where the clavicle connect s to coracoi; G is the mile point of the joint gap; BC is the joint gap location; Ruler surface plane shown as BK in Fig. 7(b) or plane (BB KC C) in Fig. 6(b) is the joint cutting plane. he coracoi is uner the area of ABCD as shown in Fig. 7(a). In orer to pull away the butterfly, the connections with area ABCD through lines AB, BC an CD nee to be severe. hus the must cut areas are BB KC C in Fig. 6(b) an area AHJB in Fig. 7(b). he following parameters will change with chicken size: (1) length of AB, (2) length of BC, (3) shift istance of ABJ plane in the parallel irection to the AB position shown in Fig. 7(a) an (4) the shift istance of BB KC C plane in the parallel to the shown plane BB KC C. he must not cut areas are as follows: (1) the area from B to A but not passing point A ue to the clavicle bone, (2) from point C to B but not passing point B ue to the protection of the breast meat, (3) area ABCD ue to the coracoi, (4) BC to chicken wing ue to the humerus. he can be cut areas are as follows: (1) from A to B, but passing point B, (2) from B to C but passing point C, (3) region CF an (4) the region below B KC, where there is either emptycone or scapulacoracoi. he cutting path from A to B to C can be a smooth curve (such as a circle) or straight lines AB, BC. In this research, in orer to simplify the cutting evice, the planes ABJ an BB KC C are selecte as the cutting

3 trajectory. he requirement for the size aaptation is to eal with the size, location an orientation change of the two planes. III. Deboning System an Cutting Device It is assume in this research that a front half is fixe on cone an the cone moves together with the conveyor at 10ins. It is also assume that the cone can provie all the require roll, yaw an pitch motions. he other requirements for the cutting evice are the must cut trajectory must be followe, must not cut region cannot be entere, chicken size must be aapte an cutting force must be small enough to avoi amaging the carcass bone structure an to keep the eformation of the carcass very small. Five eboning stations are use to harvest breast meat. he first one is the vision station which is use to ientify the location of the chicken joint relative to the cone. he secon is the scapula cut station. he next two are for the left an right clavicle cuts. he last one is for the joint cut. he cutting stations, cone, front halves an conveyor are shown in Fig. 8. he cutting evice is shown in Fig. 9. Fig. 8 Cutting system iagram. Fig. 9 Cutting evice. (a) Initial location. (b) One cutting posture. Fig. 10 schematic illustration of the cutting evice. he current research focuses on builing a prototype to test the cutting metho instea of builing the whole cutting system. hus in the following, the cutting evice in Fig. 9 is iscusse only. he cutting tool hanle is fixe to the base through a universal joint. he hanle connects with two linear actuators: top actuator an sie actuator. he top actuator is mainly for cutting an the sie actuator is mainly for chicken size aaptation. he other sie of the actuators is also fixe to the base through a universal joint. he two actuators an the hanle can rotate freely relative to its universal joint. he hanle is riven by the translation motion of the two cyliner pistons an its schematic iagram is shown in Fig. 10, where the frame OXYZ is fixe to eh base (a space-fixe frame). In OXYZ frame, point O t (R t, 0, Z t0 ) is the base location of the top actuator, O s (R s, Y s0, 0) is the base location of the sie actuator, P t (R t, 0, 0) is the top actuatorhanle connection point, P s (R s, 0, 0) is the sie actuatorhanle connection point an P i (L, 0, 0) is the cutting trajectory in OXYZ frame. Parameters with subscript t are for the top cyliner an those with subscript s are for the sie cyliner. he moving istance in Y irection is y an in Z irection is Z, where y is for the chicken size ajustment an z is for the clearance ajustment an cutting motion. he initial position is at OX an the final position is at OX, where the plane O P i P i is parallel to OXZ plane, thus y = OO as shown in Fig. 10(b). he istance from point P i to XOY plane is Z. he coorinate of point P i is ( x, y, z ), where y < 0, z < 0 an L = x + y + 2 z. So there is = ( + ) x y y z z x. he coorinate of P t is R t L( x, y, z ). he vector O t P t is r t = R t L[ x, y, z ] [R t, 0, Z t0 ], where vector r express the piston en location in the OXYZ frame an [ ] = R +,, L r. t t y y x z z x y z he coorinate of P s is ( x, y, z )R s L. he vector O s P s an its erivative with time are r s = [ x, y, z ] R s L [R s, 0, Z s0 ], r s = R [ s y y x + z z x, y, z ] L. hus the motions of the two cyliners can be etermine by preefine y, z. In size aaptation motion, z = 0. here are L 2 2 = x + 2 y, = x y y. x he coorinate of P t is ( x, y, 0) R t L. he vector O t P t is r t = R t L [ x, y, 0] [R t, 0, Z t0 ]. So r = R,,0. [ ] L t t y y x y he coorinate of P s is ( x, y, 0) R s L. he vector O s P s an its erivative with time are r s = R s L [ x, y, 0] [R s, 0, Z s0 ], r = R,,0. [ y y x y ] L s s For the clearance of the ajustment an cutting motion, y? 0 an keeps constant ( y = 0). here are L 2 = x 2 + y 2 + z 2, =. x z z x he coorinate of P t is ( x, y, z ) R t L an the vector O t P t is r t = R t L [ x, y, z ] [R t, 0, Z t0 ], an its time erivative is r R [ ] t = t z z x,0, z L. he coorinate of P s is ( x, y, z ) R s L an the vector O s P s an its erivative with time are r s = R s [ x, y, z ] L [R s, 0, Z s0 ], r = R,0,. [ z z x z ] L s s

4 IV. Cutting rajectory Simulation he important parameters in this research are the moving spee of the front halves, 10 ins with 12 inch interval. For the AB an BC cut, the moving istance is about 1 inch. hus the cutting takes about 100ms. For AB cut, the blae plane (OX Z) shoul lies in the ABJ plane as shown in Fig. 11. Note the orientation ifference between the blae plane an its initial plane. It is assume that the two planes OX Z an OXZ in Fig. 10(b) are in the same orientation (plane normal vectors are the same). he problems (yiel rop an cutting position errors) brought by this assumption will be iscusse later. he maximum breast meat cut epth in AB cut is less than 0.5 inch. If BM in positive OX irection is 14 inch an BJ is 0.5 inch in OZ irection, it takes 25ms. Note Z irection moving spee is 0 at point J, thus the moving trajectory in positive X irection is 1.5 inch (1 inch for height ajustment an 0.5 inch for cut epth). It is accelerate from the contact point between the blae an the meat an e-accelerate to 0 at the moving istance in -Z is 1.5inch. acceleration. Fig. 15 shows the cyliners motion in orer to generate the trajectories shown in Figs. 13 an 14. he top cyliner piston moves about 0.03 inch an the sie cyliner piston moves about 0.8 inch. he maximum spee magnitue for the top cyliner is 20 inchs an the sie cyliner is about 30ins. (a) wo pistons motions (b) Resultant spee of each piston Fig. 15 cyliners trajectory in AB ajustment cut motion In cutting motion, y keeps constant at 1 inch an z changes from 0 to 2 inches. Since BC nees more moving istance in Z irection than AB cut, only BC cut is simulate for the cutting motion. he ratio between the acceleration an whole moving own motion is set as 12, which means that 1 inch is for acceleration of the blae in Z irection an 1 inch is for the joint cut (eceleration in Z irection). he Z motion trajectory is shown in Fig. 16. he corresponing X motion is shown in Fig. 17. Fig. 11 AB cut trajectory. Fig. 12 BC cutting evice. Fig. 16 BC cut, Z esire motion Fig. 17 BC cut, X motion Fig. 13 Desire Y motion. Fig. 14 Corresponing X motion. For BC cut, the blae plane (OX Z) shoul lie in the BB C C plane. he moving istance in OZ irection is about 1 inch an in OX irection is about 14 inch. In orer to realize slicing cut, rotary blae is use as shown in Fig. 12. Since moving 1 inch in X irection takes 100 ms, moving 14 inch in X irection takes about 25ms. At the same time, the Z irection moving istance is 1 inch. hus the Z-irection average moving spee of the cutter is about 40 ins from B to B an from C to C. At points B an C, the spee in Z irection is zero. he other simulation values are as follows: coorinate of O t as {4, 0, 4} inch, O s as {6, 4, 0} inch, S as 1 inch S z as 2 inch, L as 8 inch, moving time is 50ms an the acceleration is 12 of the whole moving time. In siz e-aaptation motion, the knife moves in the OXY plane, thus z = 0 an the final value of y is set as 1 inch. he esire motion in OY irection is shown in Fig. 13 an the corresponing motion in OX is shown in Fig. 14. Since the motion in Z irection keeps at 0, only Y an X motions are shown. In Fig. 13, the Y coorinate changes 1 inch an the acceleration is about 5g, where g is gravity Fig. 18 Cyliners trajectory in BC cutting motion. he acceleration for the blae is about 8g an moving istance is 2 inch in 50 secon. he X motion is about 0.3 inch an about 12 ins. In orer to generate this motion, the cyliners motion is shown in Fig. 18. From the results, it is observe that the moving istance of the top cyliner piston is about 1 inch an maximum spee is about 60 ins, an 0.3 inch an 40 ins for the sie cyliner. V. Applicability Verification From Section IV, it is seen that the blae nees to move 2 inches in 50ms. hus the most critical requirement s for the cyliner are extening 1 inch in 50s an extening 12 inch in 25s. he other locationtime

5 relations are not critical. It is esire to use a pneumatic system to realize this motion ue to its price an mechanical simplicity. he Enfiel LS-V15 valve, LS- C10 controller, an Numatics ACC M series cyliner 15AM 1-06A [8] were use for the verification. he experimental evice an the system connection iagram are shown in Fig. 19 (a) an (b), respectively. Fig. 20 Piston position feeback an actual piston position. Fig. 21 Step response to ifferent control gains. (a) Experimental evice. (a) Piston position. (b) Piston spee. (c) Piston acceleration. Fig. 22 Piston motion characters when valve is fully opene (b) Connection iagram. Fig. 19 Experimental evice an connection iagram. he relationship between the feeback volt age an the actual piston position of the system was calibrate first an the results are shown in Fig. 20. By ajusting ifference gains, the step response is obtaine as shown in Fig. 21. It is observe that when control gains are P gain (K p ) = 1950 an D gain (K ) = 300, the best control results are obtaine. When the valve is fully opene, the maximum acceleration can be reache. High voltage is provie to the LS-C10 controller to make sure the LS-V15 valve is fully opene. he piston position, spee an acceleration versus time, respectively, are shown in Fig. 22. It is observe that the maximum spee is 55 ins an the average spee is 23 ins. In these experiments, the commane moving istance was a step signal with 1 inch in magnitue. he open status of the valve was recore in the experiments to show the possibility to change to a bigger valve. he experimental results are shown in Fig. 23. It is seen from Fig. 23 that the valve is fully opene for the initial comman of the position change. It took about 40s for piston to reach commane position (1 inch). Further experiments showe that if it is fully opene, the response time was about 33ms to 50 ms with 60% overshoot. When more loa mass was ae to the piston, it was observe that the system was capable to rive to the 1 inch position in 50 ms. he results for the 1 inch step response with an without mass are shown in Fig. 24. he time elay is observe when more mass is ae. VI. Discussions Fig. 23 Response to a step input with valve status. Fig. 24 Step resp onse with an without loa mass. A. Response ime Response time is often efine as the interval from the instant when a user initiates a request to the instant at which the first part of the response is receive. In this stuy, the response time is efine as the interval from the moment when the comman signal (step signal) is sent out to the moment when the response of the piston position reaches 95% of the comman signal. In these experiments, it was observe that when the commane step trajectory was 1 inch, the response time was about 35ms to 50ms. In the esire trajectory, when the moving istance is about 1 inch, the response time is about 50ms. he maximum spee is about 55 ins, but the esire maximum spee is 60 ins. It can be preicte that at 25ms, the piston cannot reach t he 0.5in location. he response time mainly epens on the following parameters: (i) the magnitue of the signal, ( ii) the control

6 gain (i. e. valve open area, or mass flow rate), (iii) compression air pressure, (iv) the friction force on the piston an (v) the inertia of the piston an loa. With larger loa mass, the response time will be longer. In the experiments, the loa mass is 2.35 lb. he maximum acceleration can be realize when (a) the valve is fully opene anor (b) the maximum pressure is realize without fully open the valve. wo methos can be use to increase this maximum spee to 60 ins by either using to a bigger valve or changing to a smaller inner iameter cyliner. Using a bigge r valve generates bigger overshoot. he sampling time of the current control system is about 13 ms an the sampling frequency is obviously too slow. Using a shorter sampling time control system, the control results may be better in the sense of faster an smooth response an lower overshoot. It can be seen that the teste pneumatic system can marginally satisfy the esign requirement s. B. Orientation Error Analysis In the simulation, y was set as 1 inch. For AB an BC cut, by selecting a mile position of the chicken size, the value can be set as y = ±0.5 inch. If the blae hanle length is 8 inch, the angle to OXZ plane satisfy δ = Suppose the vision system can provie very goo results to make the plane ABJ an BB KC C are parallel with the moving irection an in the OXZ plane, it can be seen that the angle between the blae sie plane an cutting plane is Suppose the AB or BC cutting trajectory is 1 inch, the error generate in OY irection is y = inch. For AB cut, ue to this misalignment, the meat chunk in this regain is less than 5 mm in thickness. he meat ue to this error is very small when compare to whole breast meat. For the BC cut, the joint gap can be 2-5mm. If the blae knife intercepts with BB KC C an the mile of BC, ue to the roun shape of the ball part of the humerus, the influence of this error can be efinitely ignore. his will be further verifie using our final mechanism. VII. Conclusions he foo processing inustry is very epenent on manual labor ue to the iverse set of tasks that must be performe on a prouct that has no consistent physical parameters. his research work was focuse on the evelopment of an intelligent cutting system capable of using robotics to automatically aapt to variations in bir size an anatomy. he following conclusions were rawn: (i) he eboning relate anatomy of chicken front halves was escribe in etail. Simple cutting trajectories were ientifie. his work results in the selection of the cutting trajectories wit h simple motion. (ii) he cutting system was propose an a simple cutting evice capable of size aaptation was esigne. (iii) he kinematics of the cutting evice formulate an the cutting trajectory was simulate using ientifie parameters an the cutting mechanism. (iv) he pneumatic riven system with position feeback was ientifie. he experimental results showe that this riven system can marginally satisfy the esign requirements of the cutting motion. Base on this preliminary work, the research team will buil the ynamic moel of the pneumatic riven system to further verify this mechanism. his prototype, once it is built, will be teste on a bir shouler eboning. Acknowlegement his project is supporte by the State of Georgia through the Agriculture echnology Research Program at the Georgia ech Research Institute. he team woul also like to thank Enfiel Inc. for proviing the experimental pneumatic evice. References [1] Stork Gamco Inc., Jan [2] Meyn Foo Processing echnology B.V., inex.php?option=com_contenttask=viewi=123itemi=34, Jan [3] Daley, W., He,. Lee K-M; Sanlin, M., Moeling of the natural prouct eboning process using biological an human moels, in the Proceeings of 1999 IEEEASME International Conference on Avance Intelligent Mechatronics, pp: 49 54, Sept.19-23, [4] Heck, B., Automate chicken processing: machine vision an water-jet cutting for optimize performance, IEEE Control Systems Magazine, 26(3) 17-19, June [5] Dempsey P. G. an McGorry R. W., Investigation of a pork shouler eboning operation, Journal of Occupational an Environmental Hygiene, vol 1, pp , [6] McGorry R. W., Dow P. C. an Dempsey P. G., he effect of blae finish an blae ege angle on forces use in meat cutting operations, Applie Ergonomics, Elsevier publications, vol. 35, pp , [7] Chamberlain, F. Atlas of Avian Anatomy. Michigan State College, Agricultural Experiment Station, East Lansing, MI. Hallenbeck Printing Company, Lansing, MI, [8] Numatics Actuator Group, Accu M Series Position Feeback Actuators, ACCUMCAD, Feb