This paper discusses the design and implementation of a robotic gripper that uses compressed air to (a) orient the parts in the desired grasping position, (b) guide the parts inside a grasping mechanism and (c) feed the parts to a track conveyor with sufficient accuracy. The novelty of the approach lays in the ability to perform in-hand manipulation of the object by the gripper allowing to pick randomly placed objects that have a complex geometry. Unlike existing ‘pick and place’ operations which are mainly focused on flat objects that require minimal manipulation (rotation around vertical axis), the gripper can re-orient the parts itself, minimizing the robot’s motion. The major components of the gripper are 3D printed, allowing fast customization for different products. The manipulation and gripping mechanisms have been inspired by an application in the consumer goods industry involving the feeding of shaver handles to an assembly machine. The findings indicate that the proposed solution can be an alternative to part-dedicated, high-cost feeding equipment that is currently used. . . . . Keywords Robotics Pneumatic gripper Manufacturing Automation In-hand manipulation 1 Introduction manipulation capabilities, it has become possible to partially or fully automate these activities. In such applications, the Mass production of small-sized, lightweight parts is one of main development focus is on issues as follows: the most common activities encountered in different pro- duction sectors around the globe . Consumer goods, 1. Development and deployment of advanced vision sys- plastics, medical and healthcare products and small metal- tems that are able to identify multiple moving parts that lic parts are some indicative examples (Fig. 1). In most are either in bulk or moving on a conveyor surface . cases, hard automation has provided an efficient but very 2. Introduction of algorithms and communication frame- costly and inflexible solution [2, 3]. works  for the synchronized, collision-free and effi- The increased demand for higher product variety as well as cient operations of multiple robots inthe same the evolution of manufacturing processes and control systems workspace. This involves tools for the determination of  has favoured the development of robotic applications that the optimum sequence of operations (e.g. pick, place) [10, can undertake several tasks in the manufacturing chain either 11] as well as the real-time motion planning  and as standalone or as cooperating units [5, 6]. The sorting and operational reconfiguration  of each robotic arm to- feeding of the parts are indicative examples. Up to now, the wards minimizing the cycle time. need for loading parts from bulk has been covered by devices 3. Development of dexterous and flexible grippers that allow such as vibratory feeding bowls and conveyors as well as the simple and efficient pick and place operations empha- manual labour. With the advancement in robot sensing and sizing on hardware and software complexity minimiza- tion to achieve robust operation. A vast range of grippers have been proposed and classified so far and are also * Sotiris Makris discussed in the following paragraphs . firstname.lastname@example.org 1 An example of the most typical setups for robotized sorting Laboratory for Manufacturing Systems and Automation, Department and feeding is presented in Fig. 2. The parallel kinematic robot of Mechanical Engineering and Aeronautics, University of Patras, 26504 Patras, Greece on the right side is driven by a control logic to pick randomly 3736 Int J Adv Manuf Technol (2018) 97:3735–3750 Fig. 1 Lightweight non- symmetrical parts from multiple industrial sectors placed parts and form groups of four. The decision is driven by 4. The required re-orientation of the parts only takes place the output of a high-speed vision system. The coordination along the vertical axis rotation and thus, fast motions can software ensures that the batteries are picked in a sequence be achieved by the parallel robot on the expense of that optimizes the pick and place operation and also drives the constrained/2D manipulation. No further dexterity is second robot (on the left) to pick the group of parts without needed to manipulate the battery. colliding. In the specific system, although the robot provides the However, observing the items of Fig. 1, it is obvious that speed of motion, it is the gripper that provides the ability to these parts do not allow for most of the aforementioned sim- carry out the entire application. What is more important plifications. The main reason is that they have a complex 3D though is the fact that the problem at hand is somehow sim- geometry (not flat) that requires a more dexterous manipula- plified due to the following conditions: tion. Towards this direction, this paper presents the design of a novel gripper that can handle such parts. To efficiently present 1. The identification by the vision system is a 2D problem the design approach, the presented gripper is customized for and is solved rather easily by detecting the outlines of handling plastic shaver handles that are not symmetrical along each battery and the side with the battery poles (orienta- all axes. The ability of the gripper to overcome these limita- tion). There is therefore no need to identify the pose of the tions comes from the fact that it is designed to carry out in- item. hand manipulation of the part and correctly orienting them 2. The parts are flat objects that lay rigidly and do not move before grasping. The generalization of the proposed gripper with small forces due to the friction between them and the relies on the ability to use 3D printed components to meet the conveyor surface. geometry of the objects to be handled (see also Section 3.1: 3. The batteries always provide an area on their surface that Operating principle) and also the positioning of the com- is flat and large enough to allow picking using vacuum. pressed air nozzles to ensure that the required force vectors Picking is always performed on the part without the need for manipulating the object are achieved. to ensure that it is in a correct orientation. 1.1 Literature review The wide variety of objects manipulated in industrial process has resulted in an extensive development of new robot grip- pers and robotic hands, based on different geometries and working principles . The same work identifies 12 grasping principles namely friction, jaws, magnetic, suction, needle, electrostatic, Van der Waals, ice, acoustic, laser, Bernoulli and adhesive. In terms of actuation principle, air-driven grip- pers are in their majority, based on the Bernoulli and vacuum effects. However, they have been primarily applied in the case of planar products [17, 18 and 33 ] and in some limited cases in more complex 3D parts but without any manipulation ca- pability . An additional characteristic of few grippers is the ability of Fig. 2 High-speed sorting and pick and place operations  in-hand manipulation of parts. The grippers of [21, 22]use Int J Adv Manuf Technol (2018) 97:3735–3750 3737 two interlinked belts as active surfaces allowing the grasped The advantages of the discussed gripper, compared with object to be rotated in-hand. In-hand manipulation can also be other grippers and robotic hands (e.g. ), are that with the achieved with the use of omnidirectional driving gears use of four nozzles, it can achieve rotation and positioning in allowing the translation and rotation of the part . short period of time and it is lightweight as very few actuators The above-mentioned working principles as well as grip- are needed to perform the grasping and consists of parts that ping devices either are dedicated for manipulating simple geo- can be 3D printed, reducing the cost between product changes. metrical components or consist of complex devices that would The paper is organized as follows: Section 2 presents the require difficult and intricate control. Since the targeted appli- problem definition stemming from the analysis of require- cation poses requirements (please see Section 2) for fast ma- ments of the particular consumer goods industrial case. nipulation and accurate placement of components, the pro- Section 3 is dedicated to presenting the operating principle posed gripper cannot be classified in the above categories as of the gripper as well as the design of the constituting models. it differs from these examples in the following areas: The results of a fluid dynamics simulation analysis that was performed to validate the feasibility of the approach are also 1. It uses two different hardware modules to achieve in-hand presented in this section. Section 4 is dedicated to presenting manipulation and grasping. In this sense, it is closer to the the implementation of the actual gripper hardware and control Bernoulli or vacuum grippers (although no negative pres- systems and Section 5 outlines the outcome of experiments sure forces/suction are used) in the sense that it uses air conducted in laboratory. Finally, Section 6 draws the conclu- flow. No category in the literature uses compressed air to sions of this work and outlines areas of future research. directly manipulate the part. The final grasping is done by two parallel components and would otherwise classify it as a jaw gripper. 2 Problem definition 2. The manipulation is done using compressed air nozzles allowing the part to move freely before it is grasped. Up to The origins of this research work come from the consumer now, the non-contact manipulation can be partially goods industry and more specifically the production and as- achieved by using magnetic actuation which is nonethe- sembly of shavers. The presentation of the designed gripper is less applicable in a reduced set of metal parts . The in- therefore coupled to the specific application for simplicity of hand manipulation is not performed by the gripper explanation, but a generalized method to transfer the operation fingers/gears allowing a more simplified control ap- principles in gripper designs for different part types is also proach. No real-time sensing and correction is needed. provided in Section 3. Shavers are assembled by two compo- nents: the shaver handle and the razor head. The handle is a In the past, a somehow similar approach has been used in plastic moulded part weighing around 20 grams and its geom- vortex levitation grippers where the handling is performed by etry varies depending on the model. Its main characteristics blowing air into a vortex cup through a tangential nozzle. This however is that it is not symmetrical along all its axes. On the results in a swirling air flow which causes negative pressure opposite, it has a rather complex geometry as shown in Fig. 3. due to the developed centrifugal forces. The workpiece can be As most of the parts that are created by injection mould- picked up and held at an equilibrium position . Similar ing, it is produced in the orders of millions per day and has swirl-based handling effect has been replicated by means of to be allowed time for curing which is about 24 h. Given electrical actuation . In both cases however, only picking the large number of parts and its production method, the is possible and not direct manipulation can be achieved. produced handles are stored in bulk in large containers and In terms of the manipulation capabilities, the gripper devel- need to be fed into production after the plastic settling oped in  as part of the EU project AUTORECON is the period. More specifically, the handles have to be fed in most relevant. It belongs to the jaw gripper categories and an automated machine that assembles the handle with the incorporates linear actuators at the fingertips that allow the razor head. For this purpose, vibratory feeding bowls such rotation of the part. This gripper seems to perform very well as theoneshowninFig. 4 are used to sort and feed the in the case of parts that are balanced and have a circular cir- parts. The box with the bulk handles are emptied inside the cumference. The shaver handles required very good accuracy bowl and the vibrations drive the parts against the surfaces in the positioning of the device before grasping and this was of the bowl, making them to climb from the bottom to the affected by the accumulated errors by the robot and the con- top. During this process, the parts either obtain a correct veyor tracking or vision systems. Another fact is that the struc- orientation that allows them to finish their travel to the exit ture is heavy and very complex and thus inefficient for han- (left of the image) or fall inside the bowl and repeat the dling small lightweight parts as the handles. Vibrations during process until they are successfully oriented. At the exit of the acceleration/deceleration motion of the robot also affect the bowl, a track conveyor is attached, driving the handles the performance. to the entrance of the assembly machine by securing the 3738 Int J Adv Manuf Technol (2018) 97:3735–3750 front path of the handle inside an aluminium rail of a prop- er profile (Fig. 5). At the end of the conveyor, special opening slots are used to release the handle which is picked by means of an automated suction system. Feeding bowls have been very efficient in terms of produc- tion rates and robustness; however, they have the following critical drawbacks: 1. High investment costs. Each one starts in the order of tenths of thousands and may well surpass the hundreds of thousands. 2. Long development time. These devices are built by expe- rience people only and usually following a trial and error Fig. 4 Vibratory feeding bowl  approach that leads to months before the equipment can be delivered. All surfaces are created manually and it is reason is the fact that the handle and the production process almost certain that combines have to outsource its produc- requirements presents almost none of the features that would tion than manufacture it in-house. allow it to be manipulated by standard gripping solutions: 3. Product dedicated. These bowls are made for a specific product only. Even the slightest changes in the product 1. The handle can rest in many different poses, making it material, surface, weight or geometry will result in be- impossible to use a standard approach for grasping (as in coming it unusable. A new bowl has to be built every time the case of flat objects). a new product is introduced or an old one is modified. 2. It does not have large enough flat surfaces that could allow 4. Very high noise levels. Due to the vibratory motion, the it to be picked using small suction caps. Even if for one of generated sound levels are such that do not allow opera- the poses it could be possible to use suction caps, the pro- tors to safely work around these devices without using ear cess requirement for fast motion (acceleration) would not protection. Especially in the case where multiple bowls allow the part to be firmly grasped. operate inside the same room, the conditions become un- 3. The geometry is variable along almost all axes, thus re- bearable for the operators. quiring a very compliant grasping means (e.g. not rigid/ metal fingers). This however signifies very low grasping As a result, the consumer goods industry has initiated re- accuracy. The placement in the track conveyor on the search to identify other means of performing the feeding opera- other hand requires good positional accuracy to be suc- tion to the track conveyor. The H2020 VERSATILE Project cessfully achieved. which has funded this research has allowed to analyse the re- quirements towards the robotization of the process. However, The use of parallel, angular or suction grippers to pick and the investigation of ‘off the shelf’ solutions has led to the con- orient the part using direct contact is only possible in the case clusion that the case is much more complex than the simple ‘pick where the handle already has a correct pose allowing to grasp and place’ operations that have been presented in Section 1.The Fig. 3 Shaver handle resting poses on flat surface Fig. 5 Track conveyor Int J Adv Manuf Technol (2018) 97:3735–3750 3739 it directly. In the cases where the handle is laying on its side, DoF robots. Simple Cartesian mechanisms or SCARA the grasping will result in the position of the gripper being type robots seem more preferable for high-speed next to the handle and the fingers being on the top and bottom manipulation. side of it. As a result, this will not allow its release when the 2. Increased accuracy during the grasping, leading to accu- handle is correctly oriented. Considering to grasp it from the rate placement of the parts. rear end would fail as the centre of gravity is such that would 3. Small weight of the device removing the need for large require high clamping forces that may damage the outer sur- motors or mechanical structures to manipulate the object. face of the part. Even if this was resolved, the high accelera- The parts of interest are lightweight (several grams) and tion of the robot may result in forces that affect the certainty of therefore using a heavy device that has to be moved con- the grasping position (slipping, sliding, rotating at the grasp- tinuously in the 3D space is considered a waste of energy ing points), eventually leading in the part not being success- and efficiency. fully loaded to the machine. In addition, conventional grip- 4. Use of easily replaceable parts (3D printed) to accommo- ping devices, or grippers similar to the one proposed in , date different product types. result in deflections in the orientation of the manipulated han- dle once it has been grasped, leading to an inaccurate place- ment of the handle in the track conveyor. 3.1 Operating principle Summarizing the above, it is evident that a more efficient gripper design is required for performing the feeding process. Given the fact that the parts to be manipulated are lightweight, The main criteria are the ability to successfully pick the parts they can be displaced or rotated by small forces exerted on regardless of their orientation and the accurate positioning so their surface. As a result, the concept of manipulation by that they can be fed to the track conveyor. The following means of compressed air which is directed through small noz- section describes the design of the proposed gripper in terms zles in carefully selected locations was selected. Unlike the of operating principle as well as its detailed design. case of grasping/clamping devices, the part is manipulated without coming in contact with a moving object such as the fingers of a parallel gripper. In fact, the use of parallel or 3 Approach angular grippers is not advisable considering that the geome- try of the handle along its longitudinal axis is constantly To address the challenges that were described in the previous changing. As a result, the grasping with the use of rigid fixture paragraph, this section describes the design of a pneumatic would result in uncertainties of the final position when the part gripper that is able to pick randomly placed (in terms of posi- is grasped due to local sliding in the contact area. Even if the tion and 3D orientation) parts from a flat surface or conveyor fingers are shaped to match the object surface locally, the and feed them to a predefined location. The shaver handles positioning accuracy (vision system, robot accuracy and con- will be used as the basic example to explain the operation of veyor motion) could lead to a premature contact between the the device; however, it will be demonstrated in the next sec- gripper and the part, altering the calculated position between it tions that the same design can be used to derive grippers for a and the gripper fingers. range of similar products. Considering the analysis of the pre- Figure 6 depicts the overview of the methodology for the vious section, an efficient gripper design will need to lead to design and implementation of a pneumatic gripping device the following: able to manipulate and accurately place small complex shaped parts. The geometrical analysis of the handling surface will 1. A gripper that is able to correctly orient the part on its indicate the key features for the air stream impact points to own, eliminating the need of deploying and controlling rotate the part, while the analysis of the airflow will reveal the robots of multiple degrees of freedom (DoF), such as 6- precise placement of the nozzles as well as the required Fig. 6 General methodology 3740 Int J Adv Manuf Technol (2018) 97:3735–3750 geometry/surfaces of the gripping device. Based on the re- quirements of each application in terms of grasping and re- leasing the manipulated component, the final design of the gripper will be generated. The last step of the method deals with the validation of the proposed concept by conducting a fluid dynamics analysis for validating the orientation capabil- ities of the gripper, while the experimentation phase provides the feedback to the design phase for verifying and optimizing the device for each application (orientation, grasp, release/ Fig. 8 Selected nozzle arrangement place). The aforementioned methodology as well as its general applicability is described in the following sections. that the handle can be forced to rotate regardless of the position that it is laid upon the surface. Based on this anal- 3.1.1 Geometrical analysis ysis and the examination of the different part poses, the best configuration to achieve the rotation of the part is to The analysis of the handling surface has indicated some key have two nozzles pointed at the upper front and upper rear features of the surface that may be used to control the motion parts of the handle and one nozzle pointing at the bottom of the part under the exertion of air bursts on it. More specif- part of the middle section (Fig. 8). ically as illustrated in Fig. 7 (views of handles), the areas However, a critical question arises that is how to ensure that marked with A and B provide possible points where force the rotation of the handle is controllable and can be stopped at can be exerted to create a rotation moment around the longi- the desired orientation. The solution is found at the part ge- tudinal axis (forces in points A are vertical to the image and ometry itself. As it can be observed in Fig. 7, pose 1 is the only directed inwards while points B and C signify outwards one where empty space is observable at the bottom part of the forces, see also Fig. 8). In each case, the identified areas are middle section (area marked with C). By using the arrange- at the same height (h) from the flat surface, ensuring that the ment of nozzles that was described above, a single nozzle is force is applied in the same areas regardless of the handle’s directed to this area on one side of the handle. As a result, pose. In this sense, if the rotation around the other two axes is when the handle is found in this, and only this orientation, the physically constrained, a manipulation of the part in a desired air passes through this space, exerting no force and thus can- position is achievable. celling the force couple that creates the rotation. In other As it can be observed, the geometry of the handle is words, the arrangement of the nozzles in combination with such that allows identifying similar areas in every possible the part geometry creates a locking effect that does not allow pose that it may be resting on the flat surface. This ensures the part to further rotate. The inequality of the lateral forces results in the part being pushed against one of the side walls; however, this has no effect to the whole process as long as the desired orientation is achieved. The aforementioned analysis has led to the design of the manipulation module that is pre- sented in Section 3.2.1. 3.1.2 Orientation principle In addition to the above-mentioned geometrical analysis of the handle, and as soon as the points of interest on the surface of the component have been identified, an analysis on the airflow angle will ensure the efficient placement of the nozzles. As presented in Fig. 7, there are three possible areas where the air nozzles can be installed (namely upper front nozzle, upper rear nozzle and bottom middle nozzle) and there are six possible equilibrium positions of the shaver handle. As it is required to constrain the irregular movement of the shaver handle, once air streams are applied on its surface a concave design of the side walls of the gripper was selected. The geometrical anal- Fig. 7 Side view of possible handle poses and candidate points for force ysis of the object as described in the previous section allows to application Int J Adv Manuf Technol (2018) 97:3735–3750 3741 Fig. 9 Upper front nozzle design the side walls of the gripper according to the individual or fed to specific slots in a conveyor, a way to accurately geometry of the parts to be manipulated. position it and grasp it is required. The analysis of the angle of incidence coupled with the The main challenge is that no equipment can be added in location of each air nozzle follows. Considering the height the interior of the concave surface as it would disrupt the free of placement for the upper front and the rear nozzle as motion of the part or the air flow from the nozzles. To over- was described in Fig. 7, a placement of these nozzles come this issue, it was decided to add an additional module parallel to the flat surface would result in no significant which can actively grasp the part after it has been properly rotational effect to the handle. Taking also into account all oriented. Since the front and rear sides of the handle are not the six possible equilibrium poses of the handle, the air- needed for the orientation process, they were selected for the flow angle was extracted by identifying the minimum and removal of the part from the gripper. The addition of one extra maximum points of incidence to the manipulated compo- nozzle in the front part allows for driving the part towards the nent. Figure 9 shows that the selection of the upper front rear end of the gripper. The selection of the front part for nozzle placement at an angle between 35° and 55° from applying the air stream was due to the fact that: the flat surface will result in the same rotational effect. Similarly, the extraction of the angle of incidence was 1. The front part is more complex and thus would require accomplished for the upper rear and the bottom middle more effort to constrain. nozzles and their results are presented in Table 1. 2. The rear part involves a smooth geometry that can be also The previous concept is able to perform the in-hand used to fine tune the position of the handle upon grasping. manipulation/orientation of the part; however, it provides no means of accurately positioning of the part which is still free to The designed grasping module to achieve this operation is move within the concave surface. If the part is to be assembled presented in detail in Section 3.2.2. Table 1 Angle of incidence for Point Pose 1 Pose 2 Pose 3 Pose 4 Pose 5 Pose 6 Common area the upper rear and bottom middle nozzles Upper rear nozzle A φ =8° φ =10° φ =30° φ =25° φ =20° φ = 20° 30° ≤ φ ≤ 60° Α Α Α Α Α Α B φ =65° φ =60° φ =65° φ =80° φ =75° φ =75° Β Β Β Β Β Β Bottom middle nozzle A φ =0° φ =5° φ =35° φ =40° φ =25° φ =22° − 10° ≤ φ ≤ 5° Α Α Α Α Α Α B φ = -50° φ = -45° φ =-25° φ =-15° φ = -10° φ = -15° Β Β Β Β Β Β 3742 Int J Adv Manuf Technol (2018) 97:3735–3750 One might argue that once the correct orientation of the part is achieved a more conventional gripper can be used to pick up the part. However, this would mean that the robot will need to exchange the tools for every part that it handles (increasing drastically the cycle time) or more than one robots need to be used (essentially doubling the investment cost). For this purpose, the design was defined as a single device with two modules implementing the manipulation and grasping functionalities as described in the following section. Fig. 11 Medical parts 3.2 Design assembly of the manipulation module with the module Based on the analysis of the operating principle in the previ- that is responsible for the grasping of the part. In addition, ous section the operation of the gripper was assigned to two a fourth nozzle in front of the manipulation module has separate but cooperating modules, namely the manipulation been designed and integrated. This nozzle is responsible and the grasping module. for transferring the well-oriented component from the ma- nipulation to the grasping module and it is described in 3.2.1 Manipulation module the following section. The outer surface of the module has been designed to be flat in order to minimize the 3D In order to limit and not fully constrain the motion of the part, printing time and to provide space for the installation of a concave form was selected which is large enough to allow a components such as the valves for controlling air flow complete rotation of the part within it as shown in Fig. 10.The and the servomotors. handle is able to perform a full rotation along the longitudinal The manipulation module can be easily redesigned (e.g. axis and some small translation along all three axes. The ro- semi-circular, orthogonal) to meet a wider applicability on tations around the lateral and vertical axes are constrained by similar products (lightweight with complex geometries). the module walls. Some indicative examples of possible areas of applicability As already mentioned, the inner design of the manipulation of the proposed solution can be found in the medical, mold module is concave in order to meet the geometrical require- and consumer goods sectors. Small and lightweight parts are ments posed by the manipulated object. being extensively used in the medical sector (as the dosage Following the investigation of the previous section an spoon presented in Fig. 11). In this case and as the dosage arrangement of three nozzles on the side of the surface spoon has a simple orthogonal geometry, the manipulation has been designed. The exact position of each nozzle has module could follow a semi-circular concave geometry for been determined through the geometrical and orientation achieving the rotation of the component. analysis, while their feasibility was validated through the A more complex geometry of a component can be found in use of computational fluid dynamics simulation (please the mold industry and more specifically, the plastic compo- see Section 3.3). As will be explained in the following nent is depicted in Fig. 12. As the plastic part along its longi- sections, a pivot mechanism has been added to allow the tudinal axis has a wider front part, the manipulation module could be split into two chambers (front and rear). The front chamber of the manipulation module could follow a wider Fig. 10 Manipulation module CAD model Fig. 12 Plastic mold parts Int J Adv Manuf Technol (2018) 97:3735–3750 3743 Fig. 13 Consumer goods parts semi-circular geometry in order to provide enough space for the component to rotate, while the rear chamber has more narrow side walls that will allow a more controlled rotation Fig. 15 Parts of grasping module of the part while streams of air are applied on its surface. Several lightweight components with complex geometries can be found in the consumer goods industry. Fig. 18) that is able to transfer the well-oriented component The water scraper (depicted in Fig. 13) is an indicative towards the grasping module (Fig. 14). The geometry of the example of a complex lightweight part. In order to achieve part that slides against the surface upon which it lays, in com- the rotation of the scraper, the front part of the manipulation bination with the guiding surfaces within the manipulation module could follow a wider geometry for allowing the com- module, helps in the smooth transition from the manipulation ponent to rotate, while the rear part of the module could have a to the grasping module. more constrained geometry. The guiding surfaces inside the Following, a double-acting clamping mechanism is added manipulation module could help the part to be transferred to clamp the part from its sides. A servomotor is directly away from the module. At this point, it should be mentioned mounted on a rotating element which acts as a spacer between that the flexibility and reconfigurability of the manipulation the two clamping parts. As presented in Fig. 15, the servomo- module derives form the ability of using 3D printed compo- tor is placed on the top of the main body of the grasping nents. The applicability of the proposed solution under the module and with the help of two pairs of small linear guides, above-mentioned industrial sectors should be further investi- the open/close motion of the clamping parts is achieved. The gated to meet the individual requirements per case. assembled grasping module is presented in Fig. 16 while the open/close functionality of the module is presented in Fig. 17. The clamping elements that come in contact with the part 3.2.2 Grasping module are 3D printed to form a cavity that matches the shape of the handle. As a result, the closing action guides the handle in this The operation principle of the grasping involves the handle cavity and ensures that a known position is eventually as- being pushed by an air stream against a mechanical stop at the sumed by it. Again the design of this cavity can be easily end of the gripper. This ensures that its rear end is always located against a known surface. Another nozzle has therefore been added at the front of the manipulation module (see also Fig. 16 Assembled grasping module Fig. 14 Air steam of the front nozzle 3744 Int J Adv Manuf Technol (2018) 97:3735–3750 Fig. 17 Open/close functionality of the grasping module redesigned and 3D printed in order to meet the geometrical modifications in the design of the inner geometries of the characteristics of different components based on the several clamping parts (left and right) to meet the geometrical charac- different applications. teristics from the parts of the medical, mold and consumer Last but not least, it needs to be observed that the handle goods sectors in order to be graspable (Figs. 19, 20 and 21). needs to be somehow removed from the clamping mechanism which is however aligned with the body of the gripper. To 3.3 Fluid dynamics analysis accommodate this functionality, a pivot mechanism has been adopted allowing the grasping module to rotate along the lat- Following the design of the gripper and in order to validate the eral axis and thus reveal the front end of the razor by lowering assumptions of the operating principle and the feasibility of it. The arrangement is convenient as the front end is currently the design, a computational fluid dynamics analysis was per- used to introduce the part in the machines that assemble the formed. Since an analytic computation of the applied forces shaver handles with the razor heads. To ensure that the incli- on the complex geometry is a very complex task to model and nation angle is adjustable, a servomotor has been added in the calculate, simulation software was used to estimate and visu- design connected in a ‘piston-rod’ configuration with the alize the air flow inside the gripper. Moreover, it was possible grasping module. However, it can be easily converted to a to validate the motion of the handle when air burst is applied. pneumatic rotary joint in order to achieve higher speed and The initial analysis indicated that the generated forces are control simplicity. The assembled gripper CAD model is big enough to initiate the rotation of the handle; however, not shown in the following figure including the configuration all poses could be accommodated. For this reason, multiple for picking the part (grasping module aligned with the orien- positions and inclination angles of the nozzles where tested in tation module) and for releasing it (grasping module rotated) a ‘try and error’ approach which eventually led to identifying (Fig. 18). several placements that are suitable for the task. A similar The adaptability/reconfigurability of the grasping module analysis was also performed for the front nozzle which is derives from the ability of using 3D printed components. responsible to drive the handle towards the grasping module. Similar to Section 3.2.1, the following figures depict the This activity has allowed to prove that the assumptions of Fig. 18 Assembled gripper CAD model Int J Adv Manuf Technol (2018) 97:3735–3750 3745 Fig. 21 Consumer goods sector Fig. 19 Medical sector flow to the nozzles that are installed on the sides of the Section 3.1 are valid and that the gripper concept is techno- manipulation module while the front one (blue) is used to logically viable. activate the nozzle responsible for driving the handle to- Multiple runs confirmed the ability of the manipulation wards the grasping module. module to rotate the handle from every possible resting pose. A similar servomotor as the one that was used for the Figure 22 presents a screenshot of the analysis. grasping module was installed on the gripper body and at- With the use of the simulation experiment, the forces and tached to the grasping module through a revolute joint in a moments around the X, Y and Z axes of the part have been ‘piston-rod’ configuration as explained above. The rotation of calculated in the surfaces that are found opposite to each noz- the motor leads to the rotation of the grasping module and thus zle. The following table summarizes these values (Table 2). the lowering of the grasped part’s front end (Fig. 24)to achieve a configuration that allows the collision-free release of the handle. The total weight of the gripper prototype is 1.293 kg in- 4 Implementation cluding all the components shown in Fig. 18. Further reduc- tion of the weight and the resulting inertias that have to be 4.1 Gripper structure and hardware components handled can be achieved by more effective design of the alu- minium flange reducing drastically its size and mass (from Following the design of the previous section, the prototype in 340 to 250 grams) as well as by transferring the electrical Fig. 23 was developed. valves (total 360 grams) on the robot arm or even at the robot The manipulation module was printed as a single piece base. It is estimated that the reduction due to such changes will using a BfB 3000 3D printer and standard PLA plastic as be in the order of 0.75 kg. the build material. The grasping module was created by three printed parts of the same material. The linear guides and the servomotors were assembled with the use of small screws. An elastic band was attached at the back of the two moving parts in order to provide the grasping/closing force. An aluminium shaft was created providing a flange for attaching the gripper to the robot on the one side and a flat surface for attaching the manipulation module and the pe- ripheral components. More specifically, two electric valves operated by 24 V DC were installed and connected to the two sets of nozzles. The rear valve (black) regulates the air Fig. 22 CFD analysis model Fig. 20 Mold sector 3746 Int J Adv Manuf Technol (2018) 97:3735–3750 Table 2 Calculated forces and moments by the CFD simulation Nozzle 1 2 3 Total area (cm ) 8.77 2.91 5.32 −3 −3 −3 Fx (N) 5.31 × 10 20.61 × 10 70.4 × 10 −3 −3 −3 Fy (N) 92.10 × 10 95.79 × 10 228.92 × 10 −3 −3 Fz (N) 17.77 × 10 0.259489 × 10 0.497239 −3 −3 −3 Mx (N-m) 1.40 × 10 8.1 × 10 13.71 × 10 −3 −3 −3 My (N-m) − 1.32 × 10 − 24.51 × 10 − 18.80 × 10 −3 −3 −3 Mz (N-m) 6.45 × 10 7.80 × 10 11.94 × 10 4.2 Control system In order to control the device an ARM-based embedded PC (IGEP v2 with TI OMAP Cortex A8 CPU @ 1 GHz, 512 MB Fig. 24 Gripper at part placement configuration (right side view) NAND Flash memory and 512 MB RAM) was used running the UBUNTU 12.04 LTS distribution. A USB hub was at- tached allowing to connect a Denkovi 4 channel USB relay The integration between the gripper controller and the ro- that was used to control the two electrical valves that open and bot controller was implemented using a standard TCP/IP con- close air supply to the different nozzle sets (rotation and final nection. A server was implemented on the embedded PC and positioning). The relay is equipped with an FTDI chip through the wired connection the robot controller was able to supporting bitbang operation, thus allowing direct read/write send string messages (‘pick’ and ‘place’) to the gripper. These access to each relay on an 8-bit bi-directional. The open commands were programmed in the native PDL2 language of source library libFTDI  was used as the driver to interface the robot so that they are executed once the robot moves to the relay module through Linux. Moreover, two Phidget each relevant position. The receipt of each message by the PC Advanced Servo 1 controllers were attached to control the resulted in executing the relevant function for actuating the open/close and inclination functionalities of the grasping gripper modules. The following table summarizes the activi- module. A simple c program was used to implement the se- ties performed by each function (Table 3). quencing between the electric valves and the motors. Thanks The parametrization of the overall mechatronic system de- to the small size of the components, it was possible to fit them rives from the ability to control the individual degrees of free- in a small-sized casing that can be attached on the robot arm dom for the gripper, which in our case are the two sets of while the power requirements are kept very low (5 V DC for nozzles (rotation and positioning) controlled through the elec- the PC and the USB hub and 24 V for the electric valves). tric air valves and the rotation of two servomotors controlled Figures 25 and 26 depict the control structure and actual im- by the Phidget Advanced Servo controllers. Directly connect- plementation of the system respectively. ed to the robot controller through the embedded system (IGEP v2), the mechatronic system is a simple synchronization of I/ O’s that control the air valves and the rotation of the servomotors. 5 Experiments In order to evaluate the performance of the developed gripper, it was installed on a COMAU Smart 6 robot with an OpenC4G controller. In front of the robot, several handles were placed on a flat surface with different random orienta- tions. For the testing, ten different positions for handles to be picked were manually programmed to the robot in a sequential manner. However, in each test run, the handles were placed by hand in each pre-programmed position but in a random orien- Fig. 23 Gripper implementation (left side) tation and pose (see Table 4). This allowed to evaluate the Int J Adv Manuf Technol (2018) 97:3735–3750 3747 Fig. 25 Control system structure effectiveness under a small positional uncertainty. The setup is implementation under a common development is proposed in shown in Fig. 27. Section 6. One might argue that the use of a vision system for identi- After the first runs for fine tuning, ten experiments involv- fying the location and the pose of the shaver handle could be ing the manipulation of 100 handles were carried out. The also implemented. The current work though focuses on the results of the experiment are shown in the following table manipulation and placement of the components. Such relevant (Table 4, the numbers in italic indicate the part that failed to research studies have been conducted under [7, 30] and their be picked correctly). Out of the 100 handles, 90 were correctly oriented and successfully placed in the feeding track conveyor. The inabil- ity to orient the remaining ten parts is traced to the geometry of the manipulation module and the positioning of the nozzles against the handle. It seems that there is a sensitivity for poses 5 and 6 which means that the rotation is not always stable. The behaviour is also attributed to the fact that the air from the Table 3 Control functions of the gripper Function Description Pick 1. Ensure that grasping module is in open position—set servo value to 130° 2. Enable valve for rotation nozzles and disable after 150 ms 3. Close valves for 100 ms 4. Enable again rotation nozzles for 400 ms 5. Enable valve for front nozzle and disable after 400 ms 6. Close grasping module by setting servo value to 30° Set 7. Rotate the inclination servo at place position (45°) so that part can be inserted in the track Place 8. Open grasping module—setting servo value to 130 degrees Retract 9. Rotate the inclination servo at pick position (30°) so that next part can be grasped Fig. 26 Control hardware implementation 3748 Int J Adv Manuf Technol (2018) 97:3735–3750 Table 4 Experimental results Experiment ID Handle pose in each position (1–6as in Fig. 3) Correctly oriented handles 12 3456 78910 12 2 5 354 6 425 8 24 2 4 6 11 2331 10 3 64 1612 3 434 9 4 41 4446 3246 10 55 4 6 6 25 3535 9 63 3 4 6 5 56 2 4 5 8 72 5 6 1 5 14 3 3 1 9 8 44 5611 2164 10 9 2 1 2 534 6655 8 10 31 5265 436 6 9 Total 90 nozzles does not impact the razor surface only due to its dis- 6 Conclusions and future work persion. As a result, swirl effects also take place in the gripper, leading to a more stochastic behaviour. Further research in the In this paper, a novel gripper concept which uses compressed positioning of the nozzles is required to achieve 100% repeat- air streams to manipulate complex non-symmetrical parts has ability. All parts that were correctly oriented have been fed to been presented. The simulation experiments as well as the the conveyor, thus ensuring that the grasping module is very application in a case study stemming from the consumer efficient. The total time for executing each experiment was goods case has validated the ability of the gripper to perform around 3.5 min and is mainly attributed to the non- reliable in-hand manipulation and accurate positioning of the optimized picking positions (lower than the conveyor) and part. The design was carried out so that the part can be picked the robot trajectory. The gripper operating time is about 1 s from any resting pose, thus reducing the need for further ma- for rotating and grasping the part but it expected that with nipulation by the robot arm. The gripper is of very low weight further analysis, this time can be further reduced. The actual and can be easily integrated in any robot or feeding machine. productions requirements (around 100 parts per minute) may The use of 3D printed parts signifies low cost for maintenance be reached with the use of two robots but more work needs to and low downtime as well as modifiability in case of product be carried out on this topic in the coming years. changes. Fig. 27 Experimental setup Int J Adv Manuf Technol (2018) 97:3735–3750 3749 appropriate credit to the original author(s) and the source, provide a link The presented results indicate that the proposed solution is to the Creative Commons license, and indicate if changes were made. a step towards achieving flexible production in the consumer goods industry and can be easily extended to meet the require- ments that derive from similar applications in different indus- Publisher’sNote Springer Nature remains neutral with regard to jurisdic- trial sectors. Coupled with the identified enhancements that tional claims in published maps and institutional affiliations. are proposed hereafter, an industrialization phase is required in order to ensure the technical reliability of the proposed gripper, which at the moment presents a 90% of production References efficiency. Once the reliability of the gripper (as two cooperating modules) is ensured, then it is a matter of wear 1. Hu SJ (2013) Evolving paradigms of manufacturing: from mass of the main hardware components, namely nozzles, servomo- production to mass customization and personalization. Procedia CIRP 7:3–8. https://doi.org/10.1016/j.procir.2013.05.002 tors and electric valves. 2. Chryssolouris G (2006) Manufacturing systems: Theory and prac- There are several aspects that need to be considered for tice, https://doi.org/10.1007/0-387-28431-1 future research in order to allow introduction in actual produc- 3. Michalos G, Makris S, Papakostas N, Mourtzis D, Chryssolouris G tion environments. These involve the following: (2010) Automotive assembly technologies review: challenges and outlook for a flexible and adaptive approach. CIRP J Manuf Sci Technol 2:81–91. https://doi.org/10.1016/j.cirpj.2009.12.001 1. Further optimization of the manipulation module to 4. Krüger J, Wang L, Verl A, Bauernhansl T, Carpanzano E, Makris S, achieve 100% successful rotation of the handles. Fleischer J, Reinhart G, Franke J, Pellegrinelli S (2017) Innovative 2. Integration with vision system/depth sensors, as the one control of assembly systems and lines. CIRP Ann 66:707–730. suggested in , so that different nozzles can be activat- https://doi.org/10.1016/j.cirp.2017.05.010 5. Papakostas N, Michalos G, Makris S, Zouzias D, Chryssolouris G ed based on the actual pose of the part and minimize the (2011) Industrial applications with cooperating robots for the flex- manipulation time. ible assembly. Int J Comput Integr Manuf 24:650–660. https://doi. 3. Creation of more robust/metallic 3D parts of the gripper org/10.1080/0951192X.2011.570790 that are suitable for mass production conditions. Less 6. Makris S, Tsarouchi P, Surdilovic D, Krüger J (2014) Intuitive dual arm robot programming for assembly operations. CIRPAnn 63:13– wear and higher accuracy are needed. 16. https://doi.org/10.1016/j.cirp.2014.03.017 4. Application with a moving conveyor to evaluate accuracy 7. Tsarouchi P, Michalos G, Makris S, Chryssolouris G (2013) Vision under more dynamic conditions. system for robotic handling of randomly placed objects. Procedia 5. Elaboration of the grasping module so that the parts can CIRP 9:61–66. https://doi.org/10.1016/j.procir.2013.06.169 be released in open space to accommodate not only feed- 8. Makris S, Michalos G, Eytan A, Chryssolouris G (2012) Cooperating robots for reconfigurable assembly operations: review ing but also packaging scenarios. and challenges. Procedia CIRP 3:346–351. https://doi.org/10.1016/ 6. Optimization of compressed air consumption techniques j.procir.2012.07.060 such as the ones suggested in  for further improving 9. Michalos G, Makris S, Chryssolouris G (2014) The new assembly gripper design and efficiency. system paradigm. Int J Comput Integr Manuf 28:1–10. https://doi. 7. Replacement of the grasping module actuation with pneu- org/10.1080/0951192X.2014.964323 10. Bozma HI, Kalalıoğlu ME (2012) Multirobot coordination in pick- matic rotary/linear joints so that a single power source is and-place tasks on a moving conveyor. Robot Comput Integr needed, simplifying the control at the same time. Manuf 28:530–538. https://doi.org/10.1016/j.rcim.2011.12.001 8. Development of a HMI for easily modifying the input 11. Lin HC, Egbelu PJ, Wu CT (1995) A two-robot printed circuit parameters of the mechatronic system, such as the degrees board assembly system. Int J Comput Integr Manuf 8:21–31. of rotation of the servomotors. https://doi.org/10.1080/09511929508944626 12. Kaltsoukalas K, Makris S, Chryssolouris G (2015) On generating the motion of industrial robot manipulators. Robot Comput Integr Manuf 32:65–71. https://doi.org/10.1016/j.rcim.2014.10.002 13. Makris S, Michalos G, Chryssolouris G (2012) RFID driven robotic Acknowledgements The authors would also like to express their grati- assembly for random mix manufacturing. Robot Comput Integr tude to Mrs. Evita Bougiouri, Mr. Nikos Skounakis and Mr. Vasilis Davos Manuf 28:359–365. https://doi.org/10.1016/j.rcim.2011.10.007 for the valuable information and assistance they have provided. 14. Fantoni G, Capiferri S, Tilli J (2014) Method for supporting the selection of robot grippers. Procedia CIRP 21:330–335. https:// Funding information This research has been financially supported by the doi.org/10.1016/j.procir.2014.03.152 research project ‘VERSATILE – Innovative robotic applications for high- 15. URL: FANUC LR Mate 200iD and M1iA/5L Intelligent High ly reconfigurable production lines’ (Grant Agreement 731330) , Speed Battery Grouping (2015) http://www.fanucrobotics.com/ funded by the European Commission. cmsmedia/videos/LR%20Mate%20200iD%20and%20M1iA_ 5L%20Intelligent%20High%20Speed%20Battery%20Grouping_ Open Access This article is distributed under the terms of the Creative 458_684.mp4 Commons Attribution 4.0 International License (http:// 16. Fantoni G, Santochi M, Dini G, Tracht K, Scholz-Reiter B, creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Fleischer J, Kristoffer Lien T, Seliger G, Reinhart G, Franke J, distribution, and reproduction in any medium, provided you give Nørgaard Hansen H, Verl A (2014) Grasping devices and methods 3750 Int J Adv Manuf Technol (2018) 97:3735–3750 in automated production processes. CIRP Ann Manuf Technol 63: 25. Li X, Kawashima K, Kagawa T (2008) Analysis of vortex levita- tion. Exp Thermal Fluid Sci 32:1448–1454. https://doi.org/10. 679–701. https://doi.org/10.1016/j.cirp.2014.05.006 17. Stühm K, Tornow A, Schmitt J, Grunau L, Dietrich F, Dröder K 1016/j.expthermflusci.2008.03.010 (2014) A novel gripper for battery electrodes based on the 26. Li X, Kagawa T (2013) Development of a new noncontact gripper Bernoulli-principle with integrated exhaust air compensation. using swirl vanes. Robot Comput Integr Manuf 29:63–70. https:// Procedia CIRP 23:161–164. https://doi.org/10.1016/j.procir.2014. doi.org/10.1016/j.rcim.2012.07.002 10.065 27. Chen F, Cannella F, Canali C, Hauptman T, Sofia G, Caldwell D 18. Davis S, Gray JO, Caldwell DG (2008) An end effector based on (2014) In-hand precise twisting and positioning by a novel dexter- the Bernoulli principle for handling sliced fruit and vegetables. ous robotic gripper for industrial high-speed assembly. IEEE:270– Robot Comput Integr Manuf 24:249–257. https://doi.org/10.1016/ 275. https://doi.org/10.1109/ICRA.2014.6906621 j.rcim.2006.11.002 28. Sharma A, Noel MM (2012) Design of a low-cost five-finger an- 19. URL: RNAAUTOMATION Razor blade handle, (2014) http:// thropomorphic robotic arm with nine degrees of freedom. Robot www.rnaautomation.com/wp-content/uploads/2014/09/285_ Comput Integr Manuf 28:551–558. https://doi.org/10.1016/j.rcim. 351075-Razor-Blade-Handle.pdf 2012.01.001 20. Petterson A, Ohlsson T, Caldwell DG, Davis S, Gray JO, Dodd TJ 29. URL: libFTDI library (2015) https://github.com/df3xc/FTDI-dot- (2010) A Bernoulli principle gripper for handling of planar and 3D net-usb-relais/tree/master/RelaisCard (food) products. Industrial Robot: An International Journal 37:518– 30. Aivaliotis P, Zampetis A, Michalos G, Makris S (2017) A machine 526. https://doi.org/10.1108/01439911011081669 learning approach for visual recognition of complex parts in robotic 21. Read GR (2009) Robotic hand effector. Patent GB 2(459):723 manipulation, 27th International Conference on Flexible 22. Phillips LB, Jo H (1994) Lightweight, multi-purpose two roll grip- Automation and Intelligent Manufacturing, (FAIM2017) 27–30 per for part manipulation. Proceedings of the 5th World Conference June, Modena, Italy, Volume 11, pp. 423–430 on Robotics Research. Society of Manufacturing Engineers. SME 31. Ignjatović I, Komenda T, Šešlija D, Mališa V (2013) Optimisation MS94–243 of compressed air and electricity consumption in a complex robotic 23. Tadakuma K, Tadakuma R, Higashimori M, Kaneko M (2011) cell. Robot Comput Integr Manuf 29:70–76. https://doi.org/10. Finger mechanism equipped omnidirectional driving roller. IEEE 1016/j.rcim.2012.11.001 International Symposium on Micro-NanoMechatronics and Human 32. EU VERSATILE Project https://versatile-project.eu/ Science (MHS), 475–478 33. Dini G, Fantoni G, Failli F (2009) Grasping leather plies by 24. Roy D (2015) Development of novel magnetic grippers for use in Bernoulli grippers. CIRP Ann Manuf Technol 58:21–24. https:// unstructured robotic workspace. Robot Comput Integr Manuf 35: doi.org/10.1016/j.cirp.2009.03.076 16–41. https://doi.org/10.1016/j.rcim.2015.02.003
The International Journal of Advanced Manufacturing Technology – Springer Journals
Published: May 30, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera