Application experience of a robotic cell for automated adhesive dispensing

Application experience of a robotic cell for automated adhesive dispensing

IVIATHEMATICS AND COMPUTERS IN SIMULATION ELSEVIER Mathematics and Computers in Simulation 41 (1996) 419-427 Application experience of a robotic cel...

602KB Sizes 0 Downloads 39 Views

IVIATHEMATICS AND COMPUTERS IN SIMULATION ELSEVIER

Mathematics and Computers in Simulation 41 (1996) 419-427

Application experience of a robotic cell for automated adhesive dispensing Brian Davies a,,, S. Harris

a, A. Razban a, j. Efstathiou b

a Department of Mechanical Engineering, Imperial College, London SW7 2BX, UK b Department of Engineering Science, Oxford Universi~, Oxford, UK

Abstract

The need to have high integrity, structurally bonded adhesive and sealant joints throughout many industries, such as the automotive and aerospace industries, is increasing. To ensure process integrity, strict process control of the automated dispensing cell is essential. Many industries require the use of continuous beads of adhesive or sealant in applications where the integrity of the joint is critical, e.g., structural bonding or gasket sealing. A continuous and uniform bead gives required strength for good performance of the joints. For these reasons, the dispensed adhesive needs to be consistent. The consistency is a function of the amount of dispensed material and its displacement. The dispensed material is controlled through a closedloop system which is built around a six-axis robot carrying a digitally controlled dispensing gun, a single-axis linear table carrying the workpiece, and a vision system for on-line inspection and image processing. The bead parameters obtained by image processing are used by the controller to ensure that the bead stays in the desired bandwidth, whilst it is being laid down. Keywords: Automation; Dispensing system; Process control; Computer vision

1. I n t r o d u c t i o n

Many manufacturing processes in the aerospace and automotive industries involve laying down a continuous bead of adhesive or sealant, usually under manual control. Where the integrity of the joints is critical as, for example, in structural applications, automation is necessary to ensure the quality of bonded joints. It is, therefore, essential that the bead of adhesive is applied accurately along a path on the workpiece, and that the bead size remains within specified tolerance limits. The bead's dimensions m a y be required to vary along the path on the workpiece, with the possibility of specified breaks in the bead along portions of the path. As well as placing the bead accurately, the bead should be dispensed quickly to ensure manufacturing efficiency. The bead's dimensions need to be measured accurately on-line, so a vision system is used to check the dimensions of the bead on the workpiece and to adjust the dispenser flow where necessary. The above requirements apply equally to critical sealant joints such as in the "wet-wing" fuel tanks of aircraft. * Corresponding author. 0378-4754/96/$15.00 © 1996 Elsevier Science B.V. All rights reserved SSDI 0 3 7 8 - 4 7 5 4 ( 9 5 ) 0 0 0 8 9 - 5

420

B. Davies et al./Mathematics and Computers in Simulation 41 (1996) 419-427

For most modem applications the sealants and adhesives tend to be very viscous, to be able to stand on a vertical surface without slumping. They also have viscosities which are very temperature dependent and which thus lead to high variability of control. These factors, together with the need for high production rates requiring high speed dispensing, make the dispensing task very exacting. To assure the integrity for such critical processes, it is no longer adequate to monitor the robot and dispensing system performances. The quality of the actual workpiece must be checked and fed back to the controller as part of the on-line dispensing activity. Thus an automated system, which both monitors and controls, has been provided. Similar requirements exist for other processes where control is some distance removed from the desired effect on the workpiece, e.g. in water jet cutting, laser surface sputtering and arc welding. Thus the techniques described in this paper have a general applicability, in addition to their role in automated adhesive dispensing.

2. The automated dispensing cell The robotic cell has been developed [1,2] which comprises (Fig. 1): (i) A six-axis Fanuc robot for carrying a dispensing nozzle in a desired continuous path. The robot has a Pascal based controller and can adapt the robot speed during the operation. A single-axis table, capable of carrying large components at speeds up to 1.2 m/s, is linked to the robot and directly controlled from the robot controller. (ii) An automated adhesive dispensing system with a microprocessor controller which is addressed from an IBM-PC via a D / A board to provide control of a dispensing valve for flow rate adjustment. (iii) A custom made vision system which includes a CCD camera and laser diode mounted on the robot end-effector for image acquisition. A "structured light" system has been constructed to give the height and width of the adhesive bead. This is achieved by shining a line of light, at an angle, onto the bead and measuring the width and height of the resulting image using a video camera (Fig. 2). By using T800 transputer-based image processing software, the image was acquired and adhesive bead width was obtained and used as a feedback to adjust the adhesive flow rate while the bead is being laid on the workpiece. A typical bead image is given in Fig. 3.

Fig. 1. Automated dispensing cell.

B. Davies et al./Mathematics and Corn outers in Simulation 41 (1996) 419~127

421

.l:'2,-:tg0o

lens /

" "-~/ ~ t'~

! i

\~

~-~..

Workpiece

~ / ~

,~urfoce

/

~'~ °mfr~rl~°tob~°t

Fig. 2. Schematic diagram of the vision system layout.

Bead

Bead height

Workpiece

Bead width

Workpiece

Fig. 3. Typical image of bead. The adhesive used is the Elastosol M23 which is supplied by Evode Ltd. It has non-Newtonian characteristics with pseudoplastic behaviour, is pumped using a high pressure heated pump and is transferred to the gun using a heated hose.

3. Integration of the cell components To have a closed-loop system, the robot controller, flow controller and vision system have been linked together and an overall automated system with on-line process inspection has been developed [3]. The combination of the information from the robot controller, dispensing flow control and the sensing system makes the integration of the automated cell complete. The schematic diagram of the cell is shown in Fig. 4. Once the vision system has made its measurements, it is necessary for the controller to be able to act upon them. Results from the vision system should allow the controller to correct both the flow of adhesive

422

B. Davies et al./Mathematics and Computers in Simulation 41 (1996) 419-427

OV~RAI~ SYSTEII CONTROL (tAd6 l~)

lAIR

SUPPLY

PUMPING SYSTEM DIIUW I

/d~D ~ i g c o ~ a ~ c ~ o N (OFF-

ON-LD~ O0~LU~

Ln~ CONTROL)

1

T

tUACE PROCESSING SYSTEM [

I

~___~

Fig. 4. Schematic diagram of the overall cell. and also the time at which changes are made to its flow (for example, in order to reduce the flow when going round comers).

4. Problems with adhesive dispensing Adhesive dispensing is made difficult by many dynamic factors which change at different time-scales. S o m e of the variables affecting bead width are listed in Table 1. Standard algorithmic controllers, such as a PID or Smith-Predictor, can take care of controlling the bead size, at the millisecond time-scale, but other factors, such as adhesive age and ambient temperature, influence the viscosity of the adhesive and dimensions of the bead. These vary at longer time-scales and interact in a complex way. Surface tension, viscosity and shear rate are very important for determining the rate at which adhesive flows out of the nozzle and the shape that the bead assumes on the table. However, Table 1 Factors known and suspected to affect bead width Opening of the nozzle Robot speed Temperature of the dispensing system Nozzle pressure Pump pressure Control gains

Adhesive age Surface tension Viscosity Shear rate Ambient temperature Workpiece temperature

B. Davies et al./Mathematics and Computers in Simulation 41 (1996) 419-427

423

they cannot be measured on-line and are affected by other factors such as the age of the adhesive and the ambient temperature. Other factors, such as robot speed, are well understood, since changes to the speed of the robot's traversal will require changes to the nozzle opening to maintain a constant bead width. The robot's speed changes as the robot moves around comers, requiring changes to the bead width, so it is desirable to program the robot's path to minimize speed changes so as to minimize the changes to nozzle opening.

5. Requirements of the solution One of the main requirements of any solution was that it should execute satisfactorily in real time. The robot is moving around the workpiece at a speed of up to 400 mm/s with the vision system assessing the bead that has been deposited a few centimetres behind the nozzle opening. Any extension to the time to process the image of the bead and compute the change to the nozzle opening will lead to yet more faulty bead being deposited on the workpiece. Hence, there is a need for the control system of the integrated vision and controller to operate as quickly as possible. A further requirement is that the controller should not require expensive hardware on which to operate. In industry, the familiar PC is the preferred workhorse and a strong reason would be necessary to invest in special purpose hardware, since this would significantly increase costs. Although a transputer was used for the image processing, this also is becoming a standard piece of equipment for image processing applications. The industrial setting for this project imposes another stringent requirement on the cell controller. Panels in the aerospace and automobile industries can be expensive items by the time they are ready to be treated with adhesive. By that time, they will have already been blanked and pressed, and may have received other treatment such as painting and surface treating. Hence, a controller which causes loss of panels will not be favourably received. Repair actions should therefore take place before the bead dimensions go beyond specification, rather than afterwards, requiring the scrapping of an expensively produced workpiece. Further, it was demanded that the controller should be easily reconfigurable to other adhesives and not be constructed with implicit assumptions about the behaviour of the adhesive that was being used in the experimental investigations.

6. Self-calibration and learning Traditionally, when teaching a new part to an automated dispensing system a certain amount of trial and error is necessary to ensure that the nozzle is opened and closed the fight amount at the fight times. Many parts require a varying bead size (for example, reduced bead width near clinch flanges) to prevent excess adhesive from being squeezed out when the flange is closed, or for the bead to be switched off at times (for example, to avoid holes in the workpiece). Because there are timing delays in any controller, these must be taken into account when the path is taught. This trial and error process can be very lengthy. In our controller these delays are variable, depending upon whether the nozzle is already open and the amount of change required in the bead size. By using the vision system in conjunction with the robot and controller, a self-learning capability has been developed which can reduce the amount of time required to program a new part. This process works in the following way.

424

B. Davies et al./Mathematics and Computers in Simulation 41 (1996) 4 1 9 4 2 7

First, the robot is run around the workpiece, sending out a signal to the cell controller each time a node in its path is reached (nodes are used to define the path of the robot, and in this system also define points at which the adhesive width or robot speed may change). Each time a node signal is received the cell controller adds the time it occurs to a list. The cell controller also continuously records the actual robot speed over the course of the path. Once the path is completed, the cell controller uses the node times and the robot speeds to determine at what times the vision system should see a change in the bead width (the delay between a change in bead occurring at the nozzle and occurring on the table will be inversely proportional to the speed of the robot since the camera follows a constant distance behind the nozzle). A list of nozzle time signals is then generated. This list is simply a copy of the node times with an arbitrary delay time subtracted from each. Having generated a first approximation to the nozzle times, the robot is set to run through its path again. This time the vision system is triggered to start making measurements. During this path, as each nozzle time signal is reached, the controller changes the bead to the width required at that segment of the path. Once the path is completed, the vision system sends back a list of bead widths and the times at which measurements were made (usually spaced at 40 or 60 ms intervals). The controller then runs through the list of measurements, determining when changes occurred in the bead width. It then compares the times at which the vision system saw these changes with the times at which the vision system was expecting to see the changes. The differences between the two sets of times are used to update the nozzle time signals for the next run. This method therefore automatically compensates for delays in the adhesive dispenser and its controller without knowing anything about the internal workings of the controller. The next time the bead is laid around the path, the times at which the changes occur will be more accurate. This process is performed between every run, allowing for automatic adjustment for long term effects which may cause the dispenser timing to change. Typically the timing for a new bead is correct after three trials.

7. Performance of the timing correction algorithm To test the system's ability to correct the timing errors in the gun, a path was generated in the robot consisting of six nodes on which the bead size would be switched. The cell was calibrated and 10 runs made over the path with the vision system providing a full monitor (i.e. displaying a scrolling table of measurements, a scrolling graph of bead size and an annotated display of the input image) on line. When in this mode, the vision system can provide a new set of measurements every 60 ms. The errors in timing of each node were measured (the error was defined as the difference between when a change in bead size was expected and when it actually occurred) and an average error calculated for each run. The process was then repeated with the vision system set to a minimal on-line display (only showing a table of measurements) in which case it could provide a new set of measurements every 40 ms. The results are summarised in Table 2. In this set of experiments the first trial in each column has a much larger error than the rest because there is no information prior to this upon which to base corrections. The errors in these trials (approximately 100 ms) represent the delays within the adhesive dispenser between being given a demand signal and changing the output. Once one bead has been laid, corrections to the timing are generated based on it, and it can be seen from Table 2 that subsequent runs have much less error in the timing of changes in the bead size (averaging 29 and 17 ms for the whole of the corrected runs with the vision system working with a 60 or 40 ms sample interval, respectively). As would be expected, if the change in bead size can occur at any time during a video measurement, the mean error in the timing overall is approximately half-a-vision period.

B. Davies et al./ Mathematics and Computers in Simulation 41 (1996) 419-427

425

Table 2 Measurements of the mean errors in bead placement timing using the correction algorithm Trial number

Mean errors in switching time (ms) Vision system at 60 ms interval

Vision system at 40 ms interval

l

100

2 3 4 5 6 7 8 9 10

45 26 34 29 28 28 19 32 27

86 30 11 16 15 14 24 19 14 10

It should be noted that the speed of the robot may be up to 400 m m / s , so even with a 40 ms vision system interval, uncertainity in the placement of bead changes may be up to 4-16 mm around the node. Should more accuracy be needed, a faster image processing system could be used, capable of processing every frame from a non-interlaced camera, which would reduce the uncertainty to + 8 mm. To improve further on this, other imaging methods would need to be investigated such as the use of linear array cameras. These are capable of capturing a single line across an image at a high resolution (typically in excess of 1000 pixels) at high speeds (down to less than 1 ms per image). However, these have the drawback that vertical measurements are no longer available, removing the ability of the system to measure the height of the bead or the height of the robot above the workpiece.

8. V i s i o n b a s e d c o n t r o l

Another feature of the vision system is its ability to transmit bead measurements back to the controller on line, while the adhesive is being dispensed. The purpose of this is to maintain the bead size within the required tolerance between each programmed bead size change. The design and mathematical modelling of such a controller has been described in detail elsewhere [4]. For the purposes of this paper only a short overview will be given. Closed-loop feedback using the vision system is used to compensate for disturbances which arise in the system that may affect the bead size (for example, a drop-off in the pressure of the adhesive as it is dispensed). The closed-loop controller is a PI based system within the cell controller PC and is triggered when measurements from the vision system indicate that the bead is further than 5% away from its desired value. Closed-loop control consists therefore of the following steps: (1) The vision system measures the bead. (2) The vision system transmits the bead size to the cell controller. (3) The cell controller compares this size with the desired size. (4) If there is a discrepancy, it updates the setting of the dispenser valve. Note that because there are delays in the system (e.g. 4 0 - 6 0 ms caused by the calculations within the vision system, a variable delay (dependent upon the robot speed) caused by the distance the camera is focused behind the nozzle, and a delay in the gun itself), there is a lockout period on the on-line controller after each change in bead size is made during which the controller is disabled. This is necessary to prevent

B. Davies et a l . / M a t h e m a t i c s and Computers in Simulation 41 (1996) 419~127

426

the controller from trying to adjust a bead at one set-point based on measurements of the bead at a previous set-point.

9. T i m i n g d e l a y s

When using the Fanuc robot system, it was initially proposed that its high level 'Pascal' based controller act as the overall cell controller. However, it became quickly apparent that the control system performed real time adjustments in 32 ms 'blocks' and typically three blocks were used to communicate. To avoid this additional 96 ms delay and also to make the system as modular as possible, it was quickly decided to use a 486 PC computer as the overall cell controller. This had the advantage that the controller was cheap but robust and the cell could be applicable to a range of hardware configurations.Thus, if a user were to employ, e.g. only a simple robot controller, the cell controller would still be appropriate. The use of a transputer T800 vision system which was also based on a 486 PC means that the vision system could communicate readily with the cell controller. A CCD camera was placed just behind the dispensing gun. If it were too close, the bead of adhesive would be in an unstable and non-repeatable condition when viewed. If the camera were too far from the dispensing gun, the bead would be stable but the delay between the dispensing control and the result being seen by the vision system would be excessive, leading to only a few adjustments per workpiece. The compromise distance allowed the bead to be in a stable enough condition to be repeatable, but minimised the delay. The constant off-set results in a delay which varies with robot speed from 32.5 ms at 400mm/s to 130ms at 100mm/s. Fig. 5 shows a block diagram of the system, in which the above delay is shown as TD2. Further variable delays occur due to the dispensing rate. TD1 is the delay in achieving a change of flow when a voltage change is applied to the dispensing gun. At high robot speeds, for a particular size of bead the flow rate through the gun is naturally higher than at lower robot speeds for the same bead size. This change in flow rate means that, because the adhesive viscosity is dependent on shear rate, at high flows a lower effective viscosity is experienced by the gun. At 400 mm/s the delay is around 30 ms, whilst at the higher viscosities experienced at 100 mm/s the delay is 90 mm/s. The third delay shown in Fig. 5,

Wref

~

.

~

W ~ f = Desin~l bead width W = Bead widlh Vin = D i s ~ n s i n g gun input Q = Adhesive flow tale DP = Pressure variation

,---.,Q~+

I

T.D.I = 30.0 to 90.0 msec

T.D.2= 32.5 to

130 msec

T.D.3 = 60.0 msec

Fig. 5. Block diagram of the closed-loop showingtime delays.

w

B. Davies et al./ Mathematics and Computers in Simulation 41 (1996) 419~427

427

TD3, is due to the delays in the vision system. If the vision system simply transmits the data over the RS232 line, the delay is that of the camera frame rate, i.e. 40 ms. If, however, there is a desire to also show the graphical results, the delay will extend to 60 ms.

10. Conclusions The use of a vision system with an adhesive dispenser gives the ability to monitor the quality of the bead on the workpiece. The ability of the system to feed back measurements of the bead to the controller for the closed-loop correction based on the actual bead laid on the workpiece (rather than an estimate of the bead size obtained from dispenser parameters such as flow rate and nozzle pressure) allows regulation of the bead. The self-learning timing correction algorithm allows the correction of timings for changes in the bead size to be performed automatically. As can be seen from the results described above, this can reduce the time required to teach the cell the adhesive pattern for a new product since by the third run the bead is being switched at times very close to the nodes. However, the timing correction algorithm is limited by the speed of the vision system and the imaging time of the camera. To obtain more accuracy, a faster vision system, and possibly faster camera would be required. A linear array camera would provide this but some measurement parameters would no longer be available. The use of a vision system has enabled the automatic dispensing cell to give an ensured quality of adhesive bead even when viscous adhesives are dispensed at high speed. The ability to record the measurements means that verification of the process quality can also be achieved. A modular cell has been provided which will enable users to only implement those aspects they require. Thus, if only a process verification and recording system are required, then only the vision system and cell controller need to be used.

Acknowledgements This work was supported by grant GR/G64107 from the Engineering and Physical Sciences Research Council.

References [ 1] B. Davies, A. Razban, S.J. Harris and J. Efstathiou, A robotic cell for automated dispensing of adhesives using knowledge based systems, Proc. 24th Internat. Symp. on Industrial Robots, Tokyo, Japan, November 1993. [2] B. Davies, S. Harris, A. Razban and J. Efstathiou, Supervisory control of an automated adhesive dispensing cell, Proc. 2rid Internat. Conf. on Computer Integrated Manufacturing, Singapore, September 1993. [3] A. Razban, B.L. Davies and G.E Bryant, Control aspects of an adhesive dispensing cell, Proc. 12th World Congress International Federation of Automatic Control, Sydney, Australia, July 1993. [4] A. Razban, G.F. Bryant and B.L. Davies, Dynamic modelling and control simulation of automatically dispensed adhesive beads, Proc. Internat. Conf. on Industrial Electronics, San Diego, USA, November 1992.