This packing line has been built initially with one out of nine robots and one out of four lanes. The completed system (9 robots, 4 lanes), it will pack 250 - 400 trays per hour with only one worker. The fruit goes though the following processes; Softness measurement to give an equivalent penetromer reading for each fruit. Weighing to 0.1g Visual inspection to grade to Zespri standards including blemishes, shape and colour. Soft spot determination. Labelling better than 99 out of 100 labels to stick. Robotic placement in any of the standard trays, including automatic management of the plix for multilayer trays. Automated handling of empty and full trays. Overall control of the system using point and click menus. The complete system is due to run, doing repack, later this year. Image
This system allowed for extremely rapid inspection of glass panes (100 square centimeters per second) under clean room conditions(class 100). The system, with our proprietary lighting configuration, found defects down to 1 micron in size and classified the type of defect (scratch, chip, bubble etc.). Defect data (type and position) were stored for review by the operator. Review was either performed automatically, in which case the defects were displayed sequentially, or manually, in which case the operator selected the particular defect of interest and a microscopic view of it was displayed. This system was deployed for Corning Glass in Japan.Image
This system automatically made fine scratches on a sheet of glass held at a reference temperature. Thereafter, the glass passed through some thermal history during the flat panel manufacture process. It was then remeasured and the resultant compaction was reported to an accuracy of a one part in ten million (.03 microns over a typical gauge length of 300 mm). For this application, we developed a 400x microscope with a depth of focus of 4 mm.Image
These machines automatically ground the sharp edges off Rayban sunglass lenses. We installed 100 of these machines the U.S., Hong Kong, Brazil, Ireland and India.Image
This machine used large diamond wheels to grind the surfaces of 1000 lb graphite billets automatically. Automated materials handling was provided. The billets were used for arc furnaces.Image
Bausch and Lomb Rayban sunglass lenses of a particular style (defined by CAD drawing) were accepted and their 3D periphery was measured to a sigma of less than 0.0001 inch. This system used a very high quality CCD lens and 640x480 retina and then compensated mathematically for divergence and fisheye. It provided smoothing to deal with small peripheral chips and, of course operated at sub-pixel accuracies. In fact we were comfortable with edge finding to 0.05 of a pixel. The system was deployed in Hong Kong. (2 Joint Patents with Bausch and Lomb).
In order to temper sunglass lenses, they are heated and then air-quenched to provide a skin under compressive stress. All lenses are subjected to having 3/4 inch steel balls dropped on them from a height of 80 inches. Lenses which are imperfectly heated and quenched are likely to shatter, with significant financial consequences. We developed an oven which heated lenses to a very uniform temperature, in continuous pass-through mode. We then developed an air quench system which scoured the boundary layer and provided extremely uniform quenches. We reduced losses by 90% This system was developed for Bausch and Lomb in Rochester, NY.Image
A compact, turbine-driven machine was used to blunt the apex of Rayban sunglass lenses making them less susceptible to chipping. Because of intense space constraints of the station where this operation had to occur, we designed and developed the air turbine and gearing for these units. We built about 100 of them.Image
With sunglass lenses of large base curve, prism is built into the lens to provide a flat optical field. In grinding these lenses, it is necessary to orient them correctly so that the prism works appropriately. We shone a laser though the optical center of the lens and measured its refracted angle as the lens was rotated. This permitted lens orientation and the application of a paint dot.
Historically Bausch and Lomb used a glycerine based coolant which was feared to be carcinogenic. We conducted a study of the mechanisms by which diamonds, embedded in a bronze matrix, actually grind the glass and based on this, developed a suitable coolant. The coolant had to be tailored to provide the correct surface lubricity over a reasonable pH range. It also had to inhibit corrosion.
These machines took glass lens blanks and ground the edge to a desired sunglass lens shape. The edge was bevelled so that the lens could be retained in a grooved eyewire. Edging was held to a circumferential tolerance of 25 microns. An extremely user- friendly software interface allowed unskilled operators quickly to program new lens styles or to load a particular style from the existing library. These machines were deployed in the U.S. and Hong Kong. The project resulted in a patent with Bausch and Lomb.Image
This machine formed wire frames for sunglass lenses and used visual feedback to control and optimize the shapes of the eyewires. The eyewire around a sunglass lens is a complex three dimensional shape, defined from the shape CAD file and the known base curve of the lens. The wire from which the eyewire is formed has metallurgical properties which vary from roll to roll and from the beginning to the end of a roll. This system used an optical measuring system to assess the length and shape of the eyewire, it then modified the bending metallurgical model to make the final wire conform to specification. It was very important that the wire should not roll as it was bent and a novel bending technique was developed. The machine made an eyewire every two seconds. (2 Joint patents with Bausch and Lomb).
This system used sophisticated vision to examine the surface of float glass under extreme temperature conditions (as it came out of the tin bath) and an automated defect marking system to highlight the defective areas. The system distinguished between standard defects such as bubbles, tin, score, stones, etc. it also measured the extent and condition of the knurl using light in the infra red range.
This system operated on eight production lines where glass panes were scored and broken. The system measured the edges of the panes and discarded any whose edges or corners had defects larger than a set tolerance. In addition, the system detected markings from the previous process and discarded those panes.
When glass is scored and broken, the broken surface of the grass can be examined for indications of improper scoring or breaking. In the development of display panels where ultra clean surfaces are needed, bad breaks lead to vibration and a spray of very fine chips - up to about 100 microns in size. These are electrostatically attracted to the glass and, being pristine, will bond to the glass, making it useless for the laying down of semi-conductors and traces by printing. We developed a system which would optically scan the edge of the glass and determine hackle and chips. We could also measure the depth of the score and provided an RMS measure of smoothness.
Corning Glass developed glass hard-discs (for data storage) which were ultra flat and could therefore potentially store a very high density of data. We developed a system to examine the discs using Nomarsky optics in a microscope and could see sub-micron blemishes. We also developed a grinding technique to bevel the edges of the discs.
These units accepted the completed (or board level) controllers for Nissan HVAC automobile systems. The parts were then automatically run through a series of tests which required dials to be set (automatically), buttons to be pushed (automatically) and a range of electrical responses to be measured, including a Nissan proprietary serial coded LAN. In addition visual routines confirmed visual cues supplied by the unit and checked that all icons were unblemished and correctly positioned. All visual routines were written in-house, as primitives in Visual Basic. The units ran 24/7 with an out-of -service tolerance of 2 hours.
These instrument clusters contained speedometers, odometers, gauges and warning icons. We provided two robots which moved the cameras to learned locations and performed functions which included the following; icon checking (on/off) and quality and the determination of needle angle to within 15 minutes of arc. The application also required fairly subtle tests for acceptable color ratios. The robots were 2-axis Humbl's which we built.
This application used one of our robots to move parts through a test sequence, stamp them appropriately or discard them.
The parts were bowl fed and then visually checked for any malformation. They were then fed into a jig for insertion.
This system managed a set of five CNC routers, with their nesting software. Boards ranging in size from 8x4 feet to 10x5 feet by 3/4 thick were picked up from any of 5 dedicated stacks. They were then transported to the appropriate router and placed correctly on the vacuum table. At the end of the asynchronous routing process, the residue was picked up. By communication with the control software, the robotic system knew where the good parts were in the frame and what was scrap. It operated appropriate suction cups so that the scrap could be dropped and the good pieces bar coded and passed to the output conveyor. The robot envelope was 120 feet long by 20 feet wide.
Tongue and groove hardwood boards of random length up to seven feet and random orientation were received on a conveyor travelling at 100 fpm. Two widths of board were present and the boards were randomly distributed over a 48 inch wide conveyor. Some of the boards were inverted, signalling that they shouldn't be packed. Some of the boards were rotated so that they had to be re-rotated by about 180 degrees before packing. The boards were to be packed with tongues and grooves appropriately nested and such that each box had the desired total board length to +/- 2.5%. Further, the top layer of boards had to be inverted. At any robot cycle the best strategy had to be followed because the cycle could not be empty and the best pick had to be made from those planks within the robot's reach. Further, the assembly of the wood in the box was something in the nature of a 3D puzzle. This task was accomplished by using sets of two interacting cartesian robots under a Master Control which was capable of handling and queuing 16 sets of two robots. Of course, by the time a specific board travelled 150 feet to the last robots, about 90 seconds had elapsed. Further, the system fed and removed the corrugated boxes automatically and scheduled the time for such transfer, without letting robots stand idle unnecessarily. Provision also had to be made for accidents so that the system was self policing and could recover. This project took a year.
Oak boards up to 16 feet long and up to 22 inches wide were fed transversely on a chain conveyor to an inspection station. The boards were examined by 4 colour cameras which determined the curvature of the board and its dimensions and gross blemishes. The boards moved on oblique rollers to hold them against a fence. The control system could set the position of this fence. From a position against this fence, the boards were then fed longitudinally into a gang rip saw. The settings of the spaces between the blades were varied and represented acceptable widths of the rips. The control system set the fence so that the optimum yield was obtained from each board. This was a hostile environment, with temperatures from 0F to 100F, lumberjacks and a lot of dust.
After boards are ripped, the next stage was to cut them into acceptable lengths. There are standard lengths but they must be clear of knots, shake, wane, split and bark. We developed a system which performed acceptably on visual data alone. This competed successfully against a system which used XRays and multi-spectral visual analysis.
We developed a system to provide simulated breaths and coughs for the National Institute Occupation Safety and Health. It featured a large piston with programmable motion, with discharge through a venturi so that instantaneous velocity could be measured and integrated.
Steel gas bottles weighing up to 80 pounds were presented to our system as open ended steel tubes. A cartesian robot placed cylinders sequentially into one of three gas furnaces running at 2400F. It then selected the one with the longest residence time and placed it in the collet of a large lathe. Spinning blocks were manipulated under hydraulic control to form a hemisphere on the end of the tube. The completed part, with the end still yellow hot, was then removed by a second robot and slid over the stud of a bottoming press to form the dunt. The unit was then removed and discarded on a conveyor to cool. One robot had to work continuously in front of the blast furnace which required extreme measures to protect it.
Plastic tubes of several cross sections and diameters from 3/4 to 3 inches and length up to 20 feet were placed in a large hopper. The hopper had a feed system which fed one tube at a time to a conveyor which then fed into the machining unit. This cut tubes to a programmed length and drilled holes at programmed positions.Image
Plastic pipe is produced by continuous extrusion. We developed a saw which would cut lengths to an accuracy of +/- 0.063". The saw was intelligent in that it moved the cutting head by stepper motor while it measured the pipe motion by a wheel with encoder so that it knew how it was doing. It wasn't permitted rough motions but could learn a control system on the basis of previous performance. Obviously, the fairly heavy cutting head had to accelerate in such a way that it could travel alongside its mark on the pipe while it cut. Any deviation would rub the side of the blade against the pipe. This adaptive control worked extremely well.
Contact lenses are made using polypropylene molds. The molds are used once and then recycled. Lenses cannot be inspected very well but the molds are inspected very carefully. This involves determining the local radius of curvature of the mold very accurately. A system was developed which used fibre optics to provide stars of light in a flat plane above the mold, whose reflections could be observed by a CCD camera looking down on the axis of the mold. The mold was placed on a precision motorized stage capable of moving in increments of one micron. By moving the mold (rotation and translation) it was possible to obtain, from the perceived positions of the stars and the way they moved, a surprisingly accurate map of the mold. We could measure a curvature of 15 mm to better than 0.1 microns. This allowed the computation of prism and any astigmatism. This work was done for Bausch and Lomb.
This robot was developed for a clean room application where we handled flat panel displays for Corning Glass works. At that time (1994) there were no appropriate robots to function in a clean room and have a vertical stroke of one metre - which we needed to slide the panes into the slots of magazines. We developed and patented an appropriate robot. We also developed a full graphical user interface.
We also developed, primarily for our own deployment but also for outside sale, a SCARA built from third party components. This was an extremely inexpensive robot with comparable performance to the industry standards.
Having obtained commissions to load CNC lathes and mills, we developed and patented a hybrid SCARA with minimal overhead requirement above the output flange. This was capable of reaching into lathes and had approximately the ergonomics of a human.
West Virginia University had conducted extensive tests on emissions from diesel trucks. We were employed to model the data in terms of assumed emission relationships. We programmed in VB6, using the Nelder and Mead simplex protocol in 15-space.
We had a client who manufactured circuit boards for Nissan. We developed a vision system which identified the copper contact pads of the PCB and then identified any solder bridges between pads. We undertook to see bridge filiaments down to 0.001". We developed these routines in Visual Basic.
GM vehicles have an indicator module in the dash which has LEDS covered by a mask. Appropriate LEDS are lit in order to highlight appropriate areas of the mask to enunciate whether seat belts and air bags are as they should be. We developed testing machines to test the units electrically and to determine that the visual output was satisfactory.