A hygienically designed force gripper for flexible handling of variable and easily damaged natural food products

Innovative Food Science and Emerging Technologies 12 (2011) 344–351

Contents lists available at ScienceDirect

Innovative Food Science and Emerging Technologies j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / i f s e t

A hygienically designed force gripper for flexible handling of variable and easily damaged natural food products A. Pettersson a,⁎, T. Ohlsson a, S. Davis b, J.O. Gray b, T.J. Dodd c a b c

SIK - The Swedish Institute for Food and Biotechnology, P.O. Box 5401, SE-402 29, Gothenburg, Sweden Italian Institute of Technology, (Fondazione Instituto Italiano di Tecnologia,)Via Morego, 30-16163 Genoa, Italy Department of Automatic Control & Systems Engineering, University of Sheffield, Sheffield, S1 3JD UK

a r t i c l e

i n f o

Article history: Received 30 March 2010 Accepted 11 March 2011 Keywords: Gripper Variable Food Universal Robot Flexible Hygienic Listeria Cross contamination Decontaminate Force sensor Automation Robot station Flexible production

a b s t r a c t To overcome present difficulties in robotized food handling a force sensing robot gripper for flexible production is presented. A magnetic coupling is used to completely encapsulate the actuator mechanism, improving hygiene and enabling a future hose-down proof design. Product location, orientation and product type and width are extracted by a vision system to aid the gripping process. Knowing the product type the grip force is set individually for each product. In the paper data of achievable grip strength, positioning accuracy and gripping times for force controlled gripping are presented. Grip times of 410–530 ms for grip forces of 50–700 g respectively are realized. An initial microbiology study on a model system showed that an intermediate decontamination can be used to reduce the cross contamination of Listeria innocua (SIK215) significantly. The gripper is further shown to be able to handle an in-feed mixture of tomatoes, apples, carrots, broccoli and grapes without intermediate adjustments. Industrial relevance: This paper covers the development and evaluation of a hygienically designed universal robot food gripper. The gripper enables an increased use of robots in the food industry and makes very flexible production with minimal changeover times possible. © 2011 Elsevier Ltd. All rights reserved.

1. Introduction Previously, robots in the food manufacturing industry have been used mostly in packaging. However, today robots are slowly entering the open product applications (Brumson, 2008; Christensen, Dillmann, Hägele, Kazi, & Norefors, 2009). This requires much tougher demands on the equipment, not least hygiene requirements. If robots could be used in all parts of the production it would be possible to increase productivity, increase consistency and quality and reduce the risk of product contamination and product cross contamination (Wallin, 1997). The risk of repetitive motion injuries can also often be avoided by replacing human labour with robots for monotonous tasks (Christensen et al., 2009). As natural products are highly variable a key to be able to use robots for open food product handling is to be able to grip them effectively and without damage. Today it is furthermore important to be able to rapidly change the production to meet the changes in customer preferences (Jennergren, 2004). If a universal

⁎ Corresponding author. Tel.: + 46 10 516 66 42; fax: + 46 46 18 87 65. E-mail address: [email protected] (A. Pettersson). 1466-8564/$ – see front matter © 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ifset.2011.03.002

gripper could be developed that was able to handle not only the variation within one type of products but also many different types, it would further increase the use and flexibility of the robot stations e.g. to meet consumer demands in higher variation in ready-to-eat (RTE) meals. Today a dedicated gripper can be developed for almost any product. This is nevertheless an expensive solution with very low flexibility. A universal gripper would enable the use of a robot station for a range of food production/assembly applications e.g. picking tomatoes for 2 h and, with zero changeover time, change to handling carrots and grapes. An automated intermediate cleaning step could be used to reduce the risk of contamination and cross contamination. However, this in turn implicates that the gripper in addition must be suited to withstand the rough wash downs faced in the food industry. By increasing hygiene in product handling the risk of product contamination is reduced. It is not only wet products (suitable for microorganism growth) such as meat, fish, dairy products that can be contaminated. Many cases of low moisture content product such as chocolate, nuts, etc. have been causes of e.g. salmonella by cross contamination where the bacteria is transferred from contaminated equipment surfaces (Podolak, Enache, & Stone, 2009). Fresh products

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

can be contaminated e.g. directly from manure used as fertiliser to vegetables, and these can in turn contaminate equipment. If a contaminating source is present the rate of cross contamination can depend on e.g. surface material, contact frequency, surface roughness, if the surface is dry or wet and on the product and contaminant themselves (Smith, 2007; Midelet & Carpentier, 2002; Rodrígues, Autio, & McLandsborough, 2007). It is important to try to reduce the amount of microbial pathogens to produce safe foods. The effect on the consumer differs. Some microorganisms have low infection doses such as verotoxigenic Escherichia coli (VTEC) that can be infectious with as little as 100 colony forming units (CFU), which is considered a very low dose, and others such as salmonella that have an infection dose of approximately 1 million CFU (Szanto et al., 2007). Production of RTE meals also increases the hygienic demands on the manufacturer as RTE products are often only heated to serving temperature, not cooked which would destroy present microbial contaminants, by the consumer or even consumed cold without further heat treatment. As a consequence of this the products are very sensitive to contamination after the final heat treatment step (Rosengren & Lindblad, 2003). It is therefore important to avoid contamination or cross contamination of microorganisms during all production steps to produce safe foods. Potentially robots can improve production hygiene, compared to humans that have a tendency to cough, shed e.g. hair, skin fragments and saliva. Many food manufacturers want to reduce the amount of manual labour where open foods are handled. This is of course both a practical and economical issue. As labour costs increase and legislation is making it more costly for the company if workers are injured (repetitive motion injuries) a robot alternative becomes more attractive (Brumson, 2008). Robots are considered to be able to maintain a high quality and throughput 24 h a day without the need for breaks, toilet facilities, parking lots or a cafeteria. However, humans are extremely flexible and dextrous and are difficult to replace with robots. 1.1. Grippers Many different approaches have been used when designing grippers intended to be able to handle a variety of product shapes. Lien and Gjerstad (2008) presented a cryo gripper that freezes the product to the gripper surface. This allows handling of various products. However, it would cause unacceptable freeze damage in many food products. Hirose and Umetani suggested a bicycle chainlike design with pulleys and string that grips the products by wrapping itself around the product applying a uniform pressure on every link (Hirose & Umetani, 1977). Non-contact Bernoulli grippers have been presented for flat products (Davis, Gray, & Caldwell, 2008; Erzincanli & Sharp, 1997) and for 3D products (Pettersson et al., 2010). Perovskii demonstrated how the hardening effect generated by applying vacuum on a pouch of particles could be used to enclose and grip products of any shape (Perovskii, 1980). Pettersson, Davis, Gray, Dodd and Ohlsson (2010) presented a similar approach, utilizing the phase shifting characteristics of a magnetorheological fluid attached to two parallel gripper arms. Various parallel arm grippers for food products have also been suggested in research projects. Friedrich, Lim, and Nicholls (2000) presented a sensor gripping system that provided grip force, product weight and slip information for flexible and variable products to enable minimal force for gripping. Naghdy and Esmaili (1996) developed an online algorithm to measure grip force and fruit ripeness from the variations in armature current at product contact. With compliant inflatable rubber pockets as grip surfaces Choi and Koc (2006) demonstrated controlled pick and place handling of as varied products as eggs (50 g), steel hemispheres (5 kg) and wax cylinders. However, one substantial problem with parallel arm grippers is the difficulty involved in sealing the mechanism and

345

making them washable. Pneumatic devices are often low in cost and simpler to make wash down proof. Stone and Brett (1995) describe an inexpensive finger gripper design based on pneumatic rubber bellows. Pneumatic grippers are, however, more difficult to use if a fast force regulation without overshoot is needed. Most of these approaches are promising and demonstrate a high degree of flexibility. A two fingered parallel arm approach seems promising for universal handling of delicate and variable products. However, the grippers presented are not, in their current design, suitable for use in food production. The grippers are often not designed to be hygienic and the grip times are often quite long. The aim of this paper is to present the development of a universal gripper with short grip times able to handle different products and their inherent natural variations in size, shape, frailness and texture. Such a gripper would enable a unique flexibility in e.g. RTE meal production. A new actuator encapsulation has been investigated and an initial study has also been performed on the effectiveness of an intermediate decontamination step. With a vision system, products are localized and identified, removing the need of mechanical aligners and allowing extraction of product parameters to facilitate and speed up the gripping. In the paper data of grip times, grip strength, positioning accuracy and decontamination effect are presented. In the next chapter gripper design and the experimental setup used are presented. Section 3 describes the methods used for evaluation. In Section 4 the results are discussed and conclusions and future work are presented in Section 5. 2. Gripper design and experimental setup 2.1. Designing a force gripper Various actuators and mechanisms can be used to generate a gripper arm closing motion. In this project a stepper motor has been used removing the need for a position encoder and allowing for open loop control. Stepper motors are a low cost and robust solution. Pneumatic actuators were rejected as fine and fast force control is problematic. There are unlimited mechanisms available to produce the closing motion of two gripper arms such as scissors closing, beak closing or linear motion (Lundström, 1973). For relatively low grip forces, small gripper size and large arm strokes the linear actuator is found to be the most suitable. Firstly, grip force will be independent of opening and, secondly, the gripper arms will be moving in a plane. Other mechanisms generally induce a curved closing motion of the gripper arms making gripper positioning dependent on product size (Li & Zarrugh, 1983). A belt and lead screw mechanism translates motor rotation to a linear motion, allowing a large stroke in a small space. One gripper arm is connected to the linear mechanism and the other gripper arm to the aluminium gripper frame, producing a parallel arm gripper. The use of only one moving arm simplifies the design and lowers the cost. However, this was shown to lead to some unexpected consequences as shown in Section 5.4. Two strain gauge sensors placed on the stationary gripper arm base are connected to a ½ Wheatstone bridge amplifier circuit to measure the grip force. A 10 bit A/D sampling of the force is implemented with an Atmel AVR atmega16, 8 bit, microcontroller. In the prototype the grip force is limited to 800 g and the resolution is approximately ±4 g due to signal noise. Strain gauge sensors are a low cost, robust and well tested technique. To increase the contact area a 3 mm thick compliant foam covered with a vinyl rubber material is used as a grip contact surface. The microcontroller also handles the serial communication between the robot station and the gripper, motor motion and also keeps track of the gripper arm position. A linear motion motor control and PD force controller algorithm have been implemented in the microcontroller. Motion and gripping commands are initiated from the robot station but autonomously performed by the microcontroller.

346

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

The grip force data are continuously sent to the PC and can be stored for traceability if needed. The total weight of the gripper is approx 850 g and it has a maximal stroke of 78 mm. Gripper arm position resolution is 0.032 mm. It is important to note that the resolution is critical as a product deformation as small as 0.3 mm, for Delicious apples, is sufficient to produce initial cell rupture and bruising (Fletcher, Mohsenin, Hammerle, & Tukey, 1965). The resolution reached in this project is considered to be sufficient for most products. Some products are likely to be even more sensitive to deformation, however, a compliant surface on the gripper arm will be used to further reduce the risk of bruising. 2.2. Improving cleanability by gripper actuator encapsulation When designing parallel force grippers it is common to use linear bearings, ball screws or other mechanisms, however, these are seldom cleanable. In this project a magnetic coupling has been implemented. Magnetic couplings are used e.g. to transfer movement into vacuum chambers (Ackeret, 2007). By using a magnetic coupling to transfer the force to the gripper arm, the actuator mechanism, ball screw, belt transmission, motor, can be completely encapsulated. In this project the actuator mechanism has been encapsulated in a plastic box. This is not a wash down encasing but it will be used here to prove the design concept that will allow a wash down design. The coupling is comprised of two sets of neodymium magnets, one set is placed inside the encapsulation on the linear sled and one is placed in the base of the moving gripping arm, outside the encapsulating box, see Fig. 1. The section of the box under the gripper arm path has been replaced by a dry linear bearing material, IGUS® DryLin,(Helsingborg, Sweden) and the gripper arm magnet set is covered with a designated low friction plastic material. In addition to transfering the vertical forces for moving the gripper arms the magnetic coupling also has to be strong enough to withstand the momentum forces exerted at gripping. The gripper was designed to be able to handle grip forces of 800 g. The attainable grip strength can be increased further by the use of stronger magnets and, if needed, a stronger motor. 2.3. Robot station To be able to test the gripper's versatility in pick and place manoeuvres the gripper has been tested in a robot station. The robot station consists of a KUKA KR5sixx 6 axis robot (Augsburg, Germany), a Cognex In-Sight 5400 camera (Natick, MA, USA), IP67 with PatMax software and a control PC. All communication between equipment is routed through the PC. This allows for simple development of GUI (Graphical User Interface), status indication, data logging and to implement product pick and place logistics. As an example a size limit

function for apples has been implemented in this project. With this feature it is possible to exclude too large or too small apples. In most non-food robot assembly applications the products to be handled are well known in advance regarding physical properties. Food products have natural variations in e.g. shape, size, weight, hardness, frailness and susceptibility to bruising depending on sort, humidity, maturity, season, etc. A priori knowledge of the product has been considered a key to aid and speed up the gripping and to avoid bruising and denting. Using a knowledge base for food handling has previously been tested for deboning of fish fillets and handling of nonrigid products (Malone, Friedrich, Spooner, & Lim, 1994; Stone, Brett, & Evans, 1998). In this project the vision system has been used not only to extract product location and angular orientation but also grip position, product width (at grip position) and product type. The vision system is able to sort and identify simultaneously a mixture of apples, tomatoes, carrots, broccoli, grapes in one image. When the products have been correctly identified the system can decide suitable gripping strategies and grip forces depending on product to be handled. With this data the gripper arm movements can also be minimized and thus cycle times shortened. 3. Evaluation methods 3.1. Decontamination test A microbial study was performed to investigate the benefits of an intermediate gripper cleaning/decontamination. For this test the grip surface used was built up by an extra soft compliant surface, to allow for good surface contact, and covered with a vinyl glove material. The grip contact surface area was approximately 4.5 cm2. Contaminated agar plates were used as a contaminated food product model. Listeria monocytogenes was set as the target organism as it is present in most environments, soil, animals, humans and insects (Lovett, 1989) and can be found on process equipment. The food poisoning cases caused by L. monocytogenes are few but it has a very high mortality rate (20–30%). Furthermore, the bacteria is not inhibited by, but will instead grow in, chilled environment if conditions are appropriate, making it a problem for RTE food (Rosengren & Lindblad, 2003). It is also the most heat resistant vegetative pathogen of significance in chilled foods (ECFF, 2006). Listeria innocua (SIK215) was chosen as a surrogate target organism for L. monocytogenes and was cultivated in BHI (brain heart infusion) broth at 30 °C overnight. The culture was diluted 1:100 and 0.25 ml of the solution was spread on trypton–soy–agar (TSA) Petri dishes to a diameter of approximately 30 mm, and was then allowed to dry. The overnight culture contained approx 8 log CFU/ml and each dish therefore initially contained approx 4.5 log CFU/cm2 on the grip contact surface area. In this initial study of gripper decontamination, a grip surface was manually dipped in either 95 °C water for approximately 1 s or in a 70% ethanol solution for approximately 1 s followed by a drying

Fig. 1. Left image: schematic description of the gripper. a: encapsulating box, b: actuator mechanism, c: dry linear bearing surface, d: magnet coupling set, e: gripper arms, f: grip surfaces. Middle image: Prototype with enclosed mechanism. Right image: Cover lifted to display the drive mechanism.

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

1. Gripping a contaminated agar plate once, to see the initial transfer to the grip surface 2. Gripping a contaminated agar plate and then successively gripping 5 clean membrane filters (PALL corporation) (placed on agar plates for support), to indicate cross contamination from sample to sample 3. Gripping a contaminated agar plate and then decontaminating the gripper in 95 °C water to see the effect of decontamination on the gripper surface. 4. Gripping a contaminated agar plate and then dipping the gripper in 95 °C water, then grip a membrane filter, then decontaminate in 95 °C water, then grip a membrane filter. This was done to see the effect of intermediate decontamination. 5. Gripping a contaminated agar plate and then dipping the gripper in 70% ethanol solution and allowing the gripper to dry for 20 s after dipping, to see the effect of ethanol decontamination. After each test the grip surface material was removed and placed in 100 ml of peptone water. The bacterial count was determined by treating the membranes/surfaces in a stomacher for 1°min to suspend the microorganisms. For the tests where a low bacterial count was expected the solution was filtered with a membrane filter that was put on a TSA plate and incubated at 37 °C for 2 days. In cases where a larger bacterial count was expected a smaller amount of the fluid was filtered or it was cultivated immediately from the diluted fluid. For each test setting three replicates were measured. 3.2. Grip strength To measure the lift capacity for various grip force settings and products the gripper was mounted in an INSTRON 4301 material testing device (Norwood, MA, USA). A product was placed between the gripper arms and fixed to the surface under the gripper. Then the product was gripped with grip forces of 50 g, 200 g or 500 g. As the test was started the gripper was slowly forced to move up by the INSTRON with a constant rate of 0.17 N/s. When the force passes the threshold for the static frictional holding force a peak in the force displacement plot is detected followed by a decrease in force and increased displacement, as described by Puchalski, Brusewitz, and Slipek (2003). During the test the force and displacement are recorded. The initial peak has been used as a measure of the gripper's lift capacity. Tomatoes, apples, strawberries and carrots were tested. The grip forces used were 50 g, 200 g and 500 g for all samples except strawberries where 25 g, 50 g, 75 g were used. 3.3. Force control and grip time The implemented PD controller is used to speed up the gripping process and to minimize the grip force overshoot. At the start of the test a product was placed between the gripper arms. The arms were separated to the product width plus an additional space of 20 mm, as space margin when positioning the gripper over a target with a robot. Initially the arms move toward the product until contact is made, then the PD controller continues the gripping and controls the grip force. Table 1 Product size variation. The minimum (Min), maximum (Max) and mean (MW) values of the grip width are presented. Product

Min(mm)

Max(mm)

MW(mm)

Apple Tomato Carrot Broccoli Potato Grape

71.1 64.0 22.5 30.8 28.9 15.6

81.8 73.9 31.2 42.3 51.9 19.7

75.5 68.8 26.5 35.3 41.0 17.3

3.4. Product positioning The vision system's product positioning accuracy was measured by taking 5 images of a product without moving the product between images. For each product type the width or diameter at the grip position was measured and the standard deviation (SD) recorded. The positioning error was finally recorded by letting the robot station perform an automatic pick operation on the products and placing them in a predefined position. A new image is taken and the final position is recorded using the same vision system. This procedure is repeated with 10 different samples of each product category, sample size variation spread is shown in Table 1. The positioning accuracy was measured for apples, potatoes, grapes, broccoli, tomatoes and carrots. All products samples were randomly collected from supermarket bins. 3.5. Determining grip position For most products the grip strategy is to grip the sample perpendicular to the lengthwise axis and at the sides of the centre of mass. However, it is difficult to know at what height the gripper should be gripped as only a two dimensional (2D) image is available. The vertical grip position is important to know as the distance from the gripper arm base to the product contact point influences the grip force measurement. To get the grip force measurements correctly a rule for deciding the vertical grip position was developed. For each product type the grip width (diameter for spherical products) was recorded with the vision system. Furthermore, the vertical position of the widest part of the product, the point of contact, on both sides was measured manually. Ten replicates were used for each product type. It was assumed that the ratio between the grip width and the vertical distance from the pick surface to the contact point would be typical for each product type and give a good indication of suitable grip height when used on the 2D vision data. 3.6. Handling of various natural products Apples, tomatoes, grapes, broccoli and carrots were randomly placed on the pick surface of the robot and the robot station was activated. The robot placed the mixture of products according to a predefined pattern. A visual evaluation of the performance was made. 4. Results and discussion 4.1. Decontamination test In Fig. 2 the results from test setting 2 are presented. The starting concentration on the contaminated model product was approximately 4.5 log CFU/cm2 on the grip contact surface area. Test setting 1 revealed 4 3.5 3

log(CFU/cm2)

period of 20 s. Five different tests were performed, a grip force of 440 g was used in all tests. The test settings were:

347

2.5 2 1.5 1 0.5 0

1

2

3

4

5

contact succession number Fig. 2. Logarithmic presentation of the decrease in transferred CFU's/cm2 with successive contacts. Error bars indicate the SD of 3 replicates.

348

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

that the number of transferred units to the gripper surface at first contact with the contaminated model product was 3.6 log CFU/cm2 [SD 0.3]. It can be seen that the numbers of transferred units to a product handled decreases for each successive product. Even if the decrease of bacteria transferred for each product gripped might seem large, it is a slow decrease from a hygiene safety point of view. Bacteria multiply exponentially, if proper growth conditions are available, which will lead to an extremely rapid increase in population in contaminated products. It should further be commented that even if much lower initial contamination concentrations might be more realistic, the limited reduction might still allow adequate bacterial numbers to spoil many product types. The tendency to “squeeze” micro water droplets from the agar model food during gripping has been noted, which likely can lead to higher transfer rates. However, as food products handled often have some surface moisture or a high moisture content it might be most relevant to look at wet/wet surface contact transfer. Rodrigues has shown that for products with a high moisture content a wetted contact has little impact on the transfer rate (Rodrígues et al., 2007). Both the treatment with 95 °C water and the 70% ethanol solution, test setting 3 and 5, reduced the number of CFUs, of the used bacterial strain, to below the detection limit. D and z values are often used to describe the time and temperature relationship for microbial inactivation. The D value defines the time to inactivate 90%, or 1 log unit, of the microorganisms at a given temperature. The z value states the temperature change needed to reduce or increase the D value by a factor of 10. The D and z values are strongly dependent on e.g. media, organism and treatment (Lovett, 1989). ECFF uses D70 °C = 0.3 min and z = 7.5 °C as typical D and z values for L. monocytogenes. These data indicate that it would be sufficient with a 0.05 s treatment in 95 °C water to reach complete inactivation (6 log unit reduction). In a study by Goulter, butcher knifes smeared with a meat-culture were shown to require at least 20 s at 82 °C for a 5 log unit reduction (Goulter, Dykes, & Small, 2008). Taking the temperature into consideration this would indicate treatment times of 0.37 s at 95 °C. Even if a complete inactivation was reached in the test further evaluation is needed for various conditions. The fruit and vegetables used for the other tests in this project might not be the most L. monocytogenes sensitive products. However, the gripper is intended to be able to handle chicken nuggets, meatballs, sausages or cakes as well. In general, the results indicate that an automated and fast decontamination process, of the gripper surfaces, will reduce the risk of product cross contamination. 4.2. Grip strength In Fig. 3 the achieved lift strength for various products and grip forces are presented. A higher grip force clearly leads to increased grip 6

grip strength (N)

5

strength. This is in accordance with the friction Eq. (1) (general form) and Eq. (2) for a parallel arm gripper situation. F ≤ μs ⋅N

ð1Þ

F ≤ μs ⋅N⋅2

ð2Þ

where F is applied force, μs is the static friction coefficient and N is the normal force (grip force). However, the result does not increase linearly with the grip force. This can depend on variables such as gripper surface deformation, possible product deformation and that a compliant elastomeric material is used as grip surface material. According to Cutkosky and Wright (1986), the contact area generally increases less rapidly than the contact pressure for elastomers, resulting in the coefficient of friction being higher for the low gripping forces. In this test the maximal grip force used for apples and carrots have been 500 g resulting in grip strength of 4.5 N for apples and 3.8 N for carrots, and 50 g grip force for strawberries resulted in 1.15 N lift force. Acceleration must be considered when handling objects using a robot. Newton's law of acceleration states that the force to move an object is equal to the mass multiplied by the acceleration. A result from this is that if an object is lifted vertically with a parallel arm gripper a high acceleration in that movement will increase the grip strength needed. Taking acceleration into account the products presented could be handled with 3.1 G, 3.9 G and 5.9 G for carrot, apple and strawberry respectively (using representative weights, carrot 100 g, apple 150 g, strawberry 20 g). In many fast pick and place operations higher accelerations might be used, even as high as 10 G to 15 G. If higher grip strength is needed, either the surface friction coefficient should be increased by choosing another grip surface material or the grip force can be increased. To be able to increase the grip force without bruising the product the contact area must be large enough to spread the force. This can be achieved by choosing a suitable compliant grip surface. A suitable grip force must be decided for each product to be handled to avoid bruising or product slip. During tests it was noted that a grip force of 100 g tended to produce a lasting visual mark on strawberries. Bruising will likely occur before denting thus limiting the applicable force further. For the larger products a higher grip force can likely be used to increase grip strength without bruising. For strawberries (Arbutus unedo L.) the friction coefficient has been shown to vary from 0.42 to 0.56 for galvanized steel, iron sheet, plywood and rubber (Özcan & Haciseferogullari, 2007). Apples have shown large differences of friction coefficient between varieties. Puchalski et al. (2003) presented values of 0.75 to 0.9 for McLemore and Gala apples respectively on rubber. To be able to better optimize the grip surface material and grip force for different products a database with product parameters such as surface friction coefficient needs to be developed.

Tomato

4

4.3. Force control and grip time

Strawberry Carrot Apple

3 2 1 0 25

50

75

200

500

grip force (g) Fig. 3. Grip strength achieved with the force gripper at various grip forces and various products. The error bars indicates the SD of 5 replicates. Due to the fragile structure of strawberries only 25, 50 and 75 g of grip force could be tested without visual lasting deformation and denting.

Fig. 4 shows representative graphs of grip force vs. time, for various set grip forces. The PD regulator controls the grip force and the overshoot is limited to b18 g for all grip forces tested. This level is considered to be adequate for the intended application but in future prototypes it should be reduced to b5 g. A source to the overshoot is the use of the magnetic coupling and its inherent vertical elasticity. When the grip force reaches the set value the regulator changes the direction of the sled, the magnetic coupling will come to a stand still until the sled has moved far enough to overcome the elastic region, resulting in a sort of backlash. This will also induce some oscillation in the force while the product is kept in the grip. The PD regulator used is not advanced enough to compensate for this backlash completely, resulting in an overshoot. It is considered that a stiffer magnetic

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

800

However, as it is only one arm that is moving the product can roll or lean slightly in the gripping procedure due to friction against the surface and the gripper arm motion. When the product is released it “rolls” back with some momentum, increasing positioning deviation. This problem can likely be solved by using two moving gripper arms or a low friction surface under the product to avoid the initial roll, allowing the product to slide instead. For the picking and placing of food products the accuracy is still considered adequate. It is important to note that optimal light conditions are crucial for correct identification and parameter extraction. These experiments have been conducted in a laboratory environment. In a real industrial setting it might be difficult to achieve correct lighting conditions.

700 Force 700g Force 500g

Gripforce (g)

600

Force 300g

500

Force 50g

400

A

B

349

C

300 200 100 0 0

100

200

300

400

500

600

700

800

4.5. Determining grip position

Time (ms) Fig. 4. The figure presents representative time/grip force plots for various grip force settings. The indexes indicate approximate values for: A) Delay time in communication system, B) transport time until gripper arms make contact with product, and C) time required for the controlled force gripping.

coupling should be constructed or that a more advanced control algorithm should be used to eliminate the overshoot. Grip forces up to 700 g have been tested with good results and performance of the magnetic coupling. The grip times are approximately 410–530 ms for gripping with 50 g–700 g of grip force respectively. 70–100 ms of this time is communication delay in the robot/PC/gripper routing, Fig. 4, A. Approximately 290–350 ms is transport time of the gripper arms before they make contact with the product (20 mm travel), Fig. 4, B. Communication, transport and force gripping times can likely be significantly shortened in a second prototype. More viscoelastic materials will need longer grip time, as the material will deform during force gripping. 4.4. Product positioning With a vision camera resolution of 480 640 pixels covering a picking area of 400 × 530 mm a resolution of 0.83 mm is theoretically available. The vision system was evaluated to be able to measure the product position to an accuracy of ±0.17 mm in both X and Y coordinates and to determine product width, diameter or length within ±0.7 mm. This enables the use of a small gap clearance between the gripper arms and the product at gripper positioning. This reduces the grip time, by reducing the transport time, Fig. 4. In Table 2 the X and Y placement error are presented for the various samples, 10 replicates were used for each product. It can be seen that all products, except the tomato (2.8 mm), are placed with a SD within 1.2 mm in Y direction. In the X direction most products are placed within 2.4 mm except for carrot (3.4 mm) and grape (4.2 mm). The reason for the larger deviation in X placement accuracy arises as the gripper arm is opening and releasing the product in the X direction, this allows some products to roll out of the grip and away from the release position. As products are at rest during gripping they are gripped from a stable position and transferred to a stable position when placed, minimizing movement after release. Table 2 Standard deviation (SD) for the X and Y placement error.

In Table 3 the product's grip width vs. grip height ratio is presented. When a product is identified the product's specific ratio can be retrieved from a database and multiplied with the extracted product grip width, to get the vertical grip height position. The data presented suggest that a reasonable estimate of a suitable grip height can be achieved by this method as the SD of the ratio measurements lies within 11% for all products except broccoli (14%). In the pick and place test performed it was observed that the products were successfully gripped and the point of contact was kept well within the gripper surface area. However, with a strain gauge sensor the force measured is dependent on correct identification of the contact point. A deviation in contact point localization results in incorrect force measurement, leading to an error in actual gripping force. In Table 3 data is presented on the calculated force deviation from set force resulting from the contact point deviation. It can be seen that for all products tested, except broccoli, the real force exerted will be kept within ±5.4% [SD b3.4] of the set value. From the observed positioning of the gripper according to the calculated height and the data in Table 3 the approach seems promising for all samples except broccoli which needs another strategy. More extensive studies are needed to verify if the approach is suitable, and for which products it is suitable. Additional sensor system or 3D vision could also be used but this will likely imply a larger cost. 4.6. Evaluation of robot station flexibility and performance If the products were identified correctly the gripper could successfully grip and release the products. The only problems observed are problems with correct identification and parameter extraction performed by the vision system. Variations in light conditions increase the risk of incorrect measurement of grip width. Also as the system uses size filtering to e.g. separate tomatoes and apples it is important that these two product types do not overlap in size sorting, otherwise an apple can be placed in a tomato position. Fig. 5A shows the products as presented to the robot and the resulting ordered output is shown in Fig. 5B. Even if the vision system currently is able to distinguish 8 different products (by size and elongation) it is Table 3 Mean values (MW) of the products' width/height ratio. In the two right most columns the mean values of the resulting deviation of the real grip force from the set grip force are presented.

Product

Pos. error X ± σ (mm)

Pos. error Y ± σ (mm)

Product

MW grip width/ grip height ratio

± SD

MW of resulting grip force error (%)

± SD

Apple Tomato Carrot Broccoli Potato Grape

1.1 2.0 3.4 2.4 2.4 4.2

1.0 2.8 0.6 1.2 1.2 0.7

Apple Tomato Carrot Broccoli Potato Grape

0.44 0.44 0.47 0.44 0.36 0.49

0.02 0.03 0.03 0.06 0.04 0.03

5.4 4.7 1.6 7.2 3.4 0.6

3.4 2.5 1.3 5.4 2.8 1.0

350

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351

extent is the result of the products' tendency to roll due to their spherical or cylindrical shape. For the majority of products the positioning is kept within b1.2 mm in Y and b2.4 mm in X. To improve gripping a novel approach to estimate grip location from vision system parameters was demonstrated with promising results for all products except broccoli. The resulting SD in exerted grip force was kept within ±5.5% (SD b3.4) of the set force, broccoli excluded. With the vision system 5 products (apples, tomatoes, grapes, broccoli and carrots), demonstrating a wide range of shapes and sizes, could be identified and handled with a grip force and vertical grip position individually set for each product. The concept of product identification and the use of a priori knowledge of these products were shown to enable the handling of very different products. The robot station was demonstrated to be able to carefully grip and place a mixture of products with sizes spanning from 15 to 81 mm. Improvements should be done to the prototype to reduce overshoot further and to decrease grip time. The suggested approach to calculate the vertical grip position should be further investigated. It is also considered important to look at more suitable vision system algorithms and to develop a product parameter database. Different grip surface materials should be investigated to find materials with suitable properties regarding hygiene, elasticity and friction. A more extensive investigation of important hygienic aspects for food grippers should be performed. Acknowledgment This study has been carried out with financial support from the Commission of the European Communities, Framework 6, Priority 5 ‘Food Quality and Safety’, Integrated Project NovelQ FP6-CT-2006015710. Special thanks to the Microbiology and Process Hygiene group at SIK for assistance and advice with the decontamination tests. References Fig. 5. A) presentation of mixed products to the robot, B) The result of automated handling of mixed products with the universal gripper.

suggested that new vision algorithms, better suited for open food products, should be developed. Furthermore it could be advantageous to use a colour system in combination with a blue conveyor belt to improve identification. 5. Conclusions and future work A new force gripper dedicated to food product handling has been developed and presented. A magnetic coupling is used to transfer the gripping motion to the gripper arms, this concept enables a complete encapsulation of the actuator mechanism and thus also the construction of a washable parallel arm electric force gripper. This design furthermore facilitates the decontamination to be performed on the gripper. In the paper it is shown that an intermediate decontamination of the gripper with hot water or ethanol can be used to reduce cross contamination of L. monocytogenes, eventually as a self cleaning cycle. It is further shown that without decontamination the gripper surface can cross-contaminate multiple samples after the event of contamination. Without cleaning the contamination will only be reduced slowly, allowing numerous products to be contaminated successively. The achieved grip strength is adequate for many handling applications and can be controlled to keep force overshoot to b18 g. Grip times for a PD controlled grip, range from 410 ms to 530 ms depending on grip strength. The robot station is able to position all products with a SD of b2.8 mm in Y and b4.2 mm in X. However, some products (apples, grapes) show a lower accuracy which to a large

Ackeret, M. (2007). Rare earth magnets provide an effective means to transfer motion into a vacuum system. Magnetics Business & Technology magazine, summer. Brumson, B. (2008). Food for thought: Robotics in the food industry. www.robotics.org visited 2008-08-04. Choi, H., & Koc, M. (2006). Design and feasibility tests of a flexible gripper based on inflatable rubber pockets. International Journal of Machine Tools & Manufacture(46), 1350−1361. Christensen, H. I., Dillmann, R., Hägele, M., Kazi, A., & Norefors, U. (2009). A white paper on the status and opportunities of the European Robotics Industry. European Robotics Forum (IFR ERF) & European Robotics Research Network (EURON), version: 20 2009. Cutkosky, M. R., & Wright, P. K. (1986). Friction, stability and the design of robotic fingers. The International Journal of Robotics Research, 5(20), 20−37. Davis, S., Gray, J. O., & Caldwell, D. G. (2008). A non-contact end effector based on the Bernoulli principle for handling of sliced tomatoes. Journal of Robotics and Computer Integrated Manufacturing, 24(2), 249−257. ECFF (2006). Recommendations for the production of prepackaged chilled food. The European Chilled Food Federation (ECFF). Erzincanli, F., & Sharp, J. M. (1997). Meeting the need for robotic handling of food products. Food Control, 8(4), 185−190. Fletcher, S. W., III, Mohsenin, N. N., Hammerle, J. R., & Tukey, L. D. (1965). Mechanical behaviour of selected fruits and vegetables under fast rates of loading. Transactions of the ASAE, 324−326. Friedrich, W., Lim, P., & Nicholls, H. (2000). Sensory gripping system for variable products. IEEE International Conference on Robotics and Automation (ICRA 2000), San Francisco, CA (pp. 1982−1987). Goulter, R. M., Dykes, G. A., & Small, A. (2008). Decontamination of knives used in the meat industry: Effect of different water temperature and treatment time combinations on the reduction of bacterial numbers on knife surfaces. Journal of Food Protection, 71(7), 1338−1342. Hirose, S., & Umetani, Y. (1977). The development of soft gripper for the versatile robot hand. Proceedings of I7th International Symposium on International Robots (pp. 353−360). Japan: Tokyo. Jennergren, L. (2004). Flexible assembly of ready-to-eat meals. Licentiate dissertation, Örebro University, Örebro, Sweden. Li, Y., & Zarrugh, M. Y. (1983). Kinematics and force control of robot grippers. http://hdl. handle.net/2027.42/6286 visited 2010-02-10. Lien, T. K., & Gjerstad, T. B. (2008). A new reversible thermal flow gripper for non-rigid products. Transactions of NAMRI/SMR, 36, 565−572.

A. Pettersson et al. / Innovative Food Science and Emerging Technologies 12 (2011) 344–351 Lovett, J. (1989). Listeria monocytogenes. In M. P. Doyle (Ed.), Foodborne bacterial pathogens (pp. 283−310). New York: Marcel Dekker inc. Lundström, G. (1973). Industrial robot grippers. The Industrial Robot, 72−82. Malone, D. E., Friedrich, W. E., Spooner, N. R., & Lim, P. P. K. (1994). Knowledge based control in the processing of highly varying products. Proceedings, 1994 IEEE International Conference on, 8–13 May 1994. Robotics and Automation, 1994, vol. 4. (pp. 2903−2908). Midelet, G., & Carpentier, B. (2002). Transfer of microorganisms, including Listeria monocytogenes, from various materials to beef. Applied and Enviromental Microbiology, 4015−4024. Naghdy, F., & Esmaili, M. (1996). Soft fruit grading using a robotic gripper. International Journal of Robotics and Automation, 11(3), 93−101. Özcan, M. M., & Haciseferogullari, H. (2007). The strawberry (Arbutus unedo L.) fruits: Chemical composition, physical properties and mineral constants. Journal of Food Engineering, 78, 1022−1028. Perovskii, A. P. (1980). Universal grippers for industrial robots. Russian Engineering Journal, 60(8), 3−4. Pettersson, J. A., Ohlsson, T., Dodd, T. J., Gray, J. O., Davis, S., & Caldwell, D. G. (2010). A Bernoulli Principle gripper for handling of planar and 3D (food) products. Industrial Robot: An International Journal, 37(60), 518−526. Pettersson, J. A., Davis, S., Gray, J. O., Dodd, T. J., & Ohlsson, T. (2010). Design of a magnetorheological robot gripper for handling of delicate food products with varying shapes. Journal of Food Engineering, 98(3), 332−338. Podolak, R., Enache, E., & Stone, W. (2009). Annex to control of salmonella in lowmoisture foods. GMA The Association of Food, Beverage and Consumer Products Companies.

351

Puchalski, C., Brusewitz, G. H., & Slipek, Z. (2003). Coefficients of friction for apple on various surfaces as affected by velocity. Agricultural Engineering International: The CIGR Journal of Scientific Research and Development Manuscript FP 03 002. Vol. V. December. Rodrígues, A., Autio, W. R., & McLandsborough, L. A. (2007). Effects of inoculation level, material hydration, and stainless steel surface roughness on the transfer of listeria monocytogenes from inoculated bologna to stainless steel and high-density polyethylene. Journal of Food Protection, 70(6), 1423−1428. Rosengren, Å., & Lindblad, M. (2003). Riksprojekt 2001 Listeria monocytogenes I kyld konsumtionsfärdig mat. Sweden: National Food Administration. Smith, D. (2007). Ranking of cross-contamination vectors of ready-to-eat foods: A practical approach.978 0 907503 41 5 Guideline No. 54. Stone, R. S. W., & Brett, P. N. (1995). A flexible pneumatic actuator for gripping soft irregular shaped objects. IEE Colloquium (Digest) Flexible Pneumatic Actuator for Gripping, 170, 13/1−13/3. Stone, R. S., Brett, P. N., & Evans, B. S. (1998). An automated handling system for soft compact shaped non-rigid products. Mechatronics, 8, 85−102. Szanto, E., Boqvist, S., Vågsholm, I., Lindqvist, R., Forshell, L. P., Andersson, Y., Löfdahl, S., Lindberg, A., & Bernekorn-Sandin, E. (2007). Verotoxinbildande E.coli VTECbakteriers smittvägar, förekomst samt risker för folkhälsan. Rapport från Livsmedelsverket, Statens Jordbruksverk, Statens veterinärmedicinska anstalt, Smittskyddsinstitutet, Socialstyrelsen och Naturvårdsverket. Wallin, P. J. (1997). Robotics in the food industry: An update. Trends in Food Science & Technology, 8, 193−198.