Developing a flexible automated fixturing device.
THE DESIGN AND FABRICATION of parts-fixturing systems are time-consuming processes that add to the cost and decrease the responsiveness of a manufacturing system. A "fixture" is a device that performs the work-holding duties in a manufacturing fabrication or assembly operation. Fixture design, much of which is done before production parts become available, is more of an art than a design process. Only after real production parts and the real production process are available can the fixture be adequately tested and debugged.
Most fixtures, however, are not easily modified. This makes finding and correcting bugs an extremely time-consuming process. In addition, bugs that are not immediately apparent can develop over time, be a function of raw material tolerances, or appear randomly as the fixture is used.
For these reasons, much research has recently focused on parts fixturing. Some of this research is devoted to turning fixture design into a process--for example, by developing general fixturing design rules and CAD packages--and some has focused on new fixturing hardware and techniques. A "flexible fixture" is one that is readily programmed or adapted for a variety of parts or products, as opposed to a special-purpose or "custom" fixture.
The focus in this article is on a device that could be used in a flexible fixturing technique. The device uses closed-loop control and has a modular design to improve fixture flexibility and repeatability and simplify debugging and fixture modification. This apparatus is intended to be used in the adaptable fixturing of a welding cell but will be applicable to many processes that call for some manipulations.
The welding cell is one of the five main operations in the manufacture of a photocopier frame. The five stages consist of a punching station, two welding stations, a machining station, and an inspection station.
In the manufacture of a photocopier frame, angle iron steel stock is first cut and pierced in a punch press to form rails. The first welding operation takes place in a robotic welding cell, which is fed by two conveyor/pallet loops. Operators place the steel rails into stationary fixtures on the pallets that statically orient the rails in different configurations. After the rails are welded into any one of several two-dimensional configurations, the sub-assemblies are moved to the second welding operation, which is manual. An operator takes these subassemblies and loads them into a very complex fixture. The subassemblies are then welded into three-dimensional box frames and removed from the fixture.
In the next step, the three-dimensional frame is placed in a machining fixture that loads the frame into a large computer-numerical-controlled machine that performs several machining operations on the frame. After the machining is complete, the frames are inspected in a coordinate measuring machine, which measures the critical distance between feature points such as holes, flats, and pads.
The frames are the structural reference for all the other subsystems that comprise the photocopier. Because of this, very tight tolerances must be held regarding the distance between feature points. If these tolerances are not held, critical operations in other subsystems are all adversely affected. Thus, the manufacture of photocopier frames with correct tolerances is essential for the optimum operation of the photocopier.
Many of the problems associated with manufacturing photocopier frames are also common in other manufacturing fabrication processes. The fixtures used to hold the workpieces are very complicated, not adaptable, and expensive and time-consuming to design and build. The fixtures need to be very precise in order to achieve the tolerances required of the finished part. In addition, the tolerances on the workpieces are often tighter than necessary for the frame to function optimally because of limitations in the rigid fixtures. Since the design of the fixture usually cannot be started until the design of the part is nearly complete, there are long and costly lead times. Another problem is workpiece misalignment due to thermal stresses from welding; to solve this, the fixture has to be modified.
The goal of the current fabrication process is to manufacture frames that are in tolerance after the machining step. In theory, if each step creates parts that are in tolerance, the final part should be in tolerance. However, in actual application, the subassemblies created can be out of tolerance. Two subassemblies that are out of tolerance can be combined to make an assembly that is in tolerance.
The manufacturing process engineers at Xerox Corp. have traced many problems in the final part to the second welding operation. When subassemblies are welded too far out of tolerance, there is no compensating for them and the fixtures need to be adjusted. This requires finding out exactly what needs to be moved and how far to move it. Next, a toolmaker (who is not always on call in a manufacturing plant) has to adjust the fixture; this is a time-consuming process. Finally, to test the adjusted fixture, complete frames need to be assembled and inspected. All this downtime for the manufacturing line is very expensive.
Attempting to design an automated flexible fixture that would hold all the different subassemblies that make up a photocopier frame would be very difficult. We will address a simpler problem that can be used as a stepping stone to the ultimate design: Design, build, and demonstrate a flexible automated fixture that will hold and position two rails in a horizontal plane so that the distance between feature points meets a specified tolerance. Such a fixture will provide three advantages--faster process development, ease of adjustment, and improved process control--by allowing controlled manipulation of components during an operation.
SEARCHING FOR DESIGN IDEAS
The major difficulty in fixturing is that it is product- and process-specific; making totally flexible fixturing is difficult, if not impossible. Developing "dynamic fixturing" could improve manufacturing quality, however, because the fixturing could manipulate the parts during the process. This would provide an extra degree of control that was not available.
We wanted to determine if an existing technology could solve our design problem without substantial modifications. We found no such dynamic fixturing device; most research in precise positioning is directed toward robot end-effecters.
Two mechanisms served as the bases for our two candidate design concepts. The first is a five-bar linkage where the position of one joint is precisely controlled by driving two input links with dc servo motors. The position of the joint is determined by sensing the angular position of both input links. At Stanford University in Stanford, Calif., this mechanism, which was originally proposed in 1959, was attached to the end of a flexible beam and used to precisely position the beam's end point. Similarly, a direct-drive actively compliant end-effecter based on this five-bar mechanism was developed at the University of Minnesota in Minneapolis in 1987. This manipulator also has actuation and sensing at both input links. The position being determined in this case is a point located at the end of one of the output links. This linkage was designed so that for small displacements, one input link would cause movement in only the x-direction and the other input link would cause movement in only the y-direction. Thus, intelligent selection of link lengths effectively uncoupled the x and y motions of the end point. Combined with a force sensor and a milling tool, the linkage allowed for active normal and tangential compliance in a robotic deburring operation.
A closed-loop alternative to a serial six-degree-of-freedom (DOF) manipulator is a linkage known as a Stewart mechanism. This mechanism consists of six linear actuators connected to a base via six two-DOF universal joints and to an end-effecter via six three-DOF ball and socket joints. Position and orientation of the end-effecter are controlled by varying the lengths of the actuators. While the range of motion is more limited than with a serial manipulator, it can be made very stiff, a characteristic common to most closed-loop mechanisms. This mechanism demonstrates an interesting reversal in the nature of the forward and inverse kinematic solutions: the inverse solution is simple, the forward one typically complex. A two-dimensional version of this mechanism was the basis of our other candidate design concept.
Other robotic end-effecters are usually multiple-degree-of-freedom wrists that allow for positioning and orienting a point. However, these wrists are very complex and expensive and are not designed for use in a very small range of motion or in an industrial environment. Most other small devices for very precise positioning do not have the range of motion or force required by our problem; they were designed for the fine manipulations required in optics or electronics work.
There is also an automated fixture that uses actuators and position sensors; however, it was designed to move the clamp points and hard stop points so that different configurations could be formed to statically accept different part geometries. The design did not allow for movement of one feature point relative to another.
THE PROTOTYPE DESIGN
The two candidate design concepts were assessed using the key performance specifications: geometry and size, workpiece holding force, reliability, accuracy, cost, and possible extension to three-dimensional applications. Based on this evaluation, the concept selected for the flexible fixturing device consists of two hydraulic cylinders, each connected to ground with a revolute joint. The pistons of the two cylinders are also connected at the ends with a revolute joint, giving the overall mechanism plane-positioning capabilities (two degrees of freedom). The cylinder lengths are measured directly with linear incremental optical encoders and fed back to a microcomputer. The microcomputer then commands two three-position valves to control the cylinders.
The mechanical system is broken down into two main categories: mechanism geometry and mechanical hardware. Two geometries were considered for the mechanism; both are five-bar linkages.
Geometry 1 has five revolute joints. The point being controlled is located at the revolute joint connecting the second and third joint links. By placing a motor at each input link, the planar position of the control point can be manipulated. By sensing the angular position of these input links, the planar position of the control point can be uniquely determined through kinematic relations.
Geometry 2 has three revolute joints and two prismatic joints. In this geometry, the planar position of the control point can be manipulated by commanding the extension of the prismatic joints or again by controlling the rotation of the two input links. Sensing either the extension of the prismatic joints or the rotation of the base joints will kinematically determine the location of the control point.
Geometry 2 was selected for several reasons. Commanding the extension of a prismatic joint is analogous to a linear actuator. Thus, two actuators control all four moving links. Geometry 1 would have required two rotary actuators mechanically linked to two custom-made input links, as well as two custom-made follower links.
Directly sensing the extension of the prismatic joint in Geometry 2 was chosen over sensing the rotation of the two base links. Calculations showed that in order to attain the desired accuracy, the rotation of the base links needed to be geared up 44:1 when connected to a rotary encoder with a resolution of 1728 pulses per revolution (before quadrature decoding). This gear-up is necessary because of the kinematic nonlinearity of the mechanism. This was judged to be unfeasible. However, a linear position sensor placed directly on, or in, the linear actuator could achieve the desired resolution for all positions.
The hydraulic circuit for one positioning device (two cylinders) and the electrical hardware can be seen in Figure 1. The hydraulic circuit contains four ball valves. The purpose of these valves, and the reason for their placement directly between the flow-control valves and the cylinder ports, is that when open they allow the end point of the device to be moved around by hand. This is a feature designed into the mechanism that may help in manually positioning work parts in the fixture in the approximate location for manufacturing before computer-controlled fine positioning begins. This is one of the reasons hydraulics was selected over other actuation methods, including electromechanical linear actuators. The ball valves allow hydraulic fluid to flow from one side of the cylinder to the other and into the tank. By letting the fluid flow to the tank, the condition called "fluid lock" is avoided. Fluid lock develops because the two halves of the cylinder are not the same size; this is due to the presence of the rod on one side of the piston and not the other. Any movement of the piston displaces more fluid on one side, thus not allowing the piston to be moved unless the extra volume of fluid can be vented to the tank.
Four flow-control valves are used in the circuit. The valves are set up in a meter-out configuration, that is, they allow hydraulic fluid to flow into the cylinder ports unrestricted but restrict fluid flowing out of the cylinder ports. The meter-out configuration allows greater control of the piston.
Next in the circuit are two three-position four-port directional valves. The spool in these valves is spring centered. Two solenoids are used to shift the spool from side to side. The directional valves are affixed to a two-station aluminum manifold.
Providing pressure to the circuit is a power unit that consists of a 5-gallon tank, a single-phase 115-volt ac 1-horsepower motor, a gear pump, and a relief valve. For demonstration purposes the system will only be run for short periods of time. The power unit provided a cheap portable pressure supply. Since pressurized hydraulic fluid is available off a central line in most manufacturing facilities, the power unit is not needed for actual operation.
The control system is also broken down into two main categories: control schemes and electrical hardware. Affixed to each cylinder is a linear incremental optical encoder, which when measured in quadrature gives a resolution of 0.0001 inch. The quadrature signal is decoded by a control board inserted in the microcomputer performing the control. Quadrature can also be decoded in software using interrupts, but hardware decoding seemed a more robust and convenient solution.
Several control schemes were applied to controlling the cylinder lengths. These include on-off control, modified on-off control, and pulse-width modulation.
On-off control is very simple. A deadband, or tolerance, is set. If the error signal, the difference between the reference length of the cylinder and the actual measured length, is greater than the deadband, a signal is sent to decrease the error. Inside the deadband, no signal is sent and therefore no actuation takes place.
Modified on-off control is similar to on-off control except three additional parameters must be set: a pulseband, a pulse width, and a pulse period. When the error is larger than the pulseband (the pulseband is always greater than the deadband), the signal is constant, as in on-off control. When the signal error is between the pulseband and the deadband, a pulsing signal is sent. The signal period is determined by the pulse period and the portion of the period that the signal is on is determined by the pulse width. Modified on-off control is good for controlling inertial loads because it allows for a deceleration in the area of the set point. Another reason this type of control is good is that the magnitude of the actuation signal can be controlled near the set point. Controller response is therefore not dependent on how fast the controller can go from one computation to the next, as in on-off control.
Pulse-width modulation is a standard control technique. Outside of the deadband, pulse-width modulation sends variable-duty-cycle signals of fixed period. The portion of the period that these signals are on is proportional to the error. When the error is great, the signal may be on 99 percent of the period and off 1 percent. When the error is small, the signal may only be on for 1 percent of the time and off for the rest. This type of control is also good for inertial loads.
MODELING AND SIMULATION
During the concept-evaluation stage, a single hydraulic actuator with a three-position solenoid-operated on-off valve was modeled, and simulations were performed to determine the relative importance of the various parameters and to study the anticipated limit-cycle behavior.
Three sets of experiments were run with the prototype device, corresponding to the three different control schemes proposed. For each control scheme two cases are described: the deadband was selected so that, first, no limit cycling was observed, and then, the deadband was reduced, and limit cycling was observed. Settings for the flow-control valves were kept the same for all test runs. Therefore, the only load on the cylinder was the reaction forces from being connected to the other cylinder and ground, due to friction in the revolute joints.
For on-off control with a deadband of 0.015 inch, no limit cycling was observed. The deadband was then reduced to 0.0075 inch, and limit cycling was observed.
For modified on-off control, the pulseband was set at 0.1 inch, pulse period at 0.024 second, and the pulse width at 0.012 second. For a deadband of 0.0075 inch, no limit cycling was observed, an improvement over the simple on-off control. As the deadband was reduced to 0.002 inch, limit cycling was observed.
For the pulse-width modulated control, the pulse period was 0.024 second, and the pulse width was proportional to the position error. For a deadband of 0.002 inch, no limit cycling was observed, an improvement over the modified on-off control. As the deadband was reduced to 0.0002 inch, limit cycling was observed.
One of the primary areas of concern for the design was whether an inexpensive on-off valve rather than an expensive servo valve could attain the resolution called for in the design specifications. Piston positioning to within 0.001 inch exceeded expectations and requirements.
With the deadband as small as 0.0002 inch, valve leakage is readily observed and the system often compensates for the drift out of the deadband. Valve life may be greatly decreased with deadbands this small.
A demonstration for Xerox manufacturing engineers and managers consisted of a computer-animated dynamic solid model on a Silicon Graphics Iris workstation using solid modeling, dynamic analysis, and animation software from ICM Inc. of Reston, Va. The demonstration also allowed the customer to run the actual proof-of-concept device hardware.
The animated simulations were presented for two reasons: to show how computers can greatly aid the design process in a cost-effective manner and to show how two such devices could be used together to move a work part.
The control program initially ran an automatic demonstration that moves the device around the extremes of its workspace and then to the center of the workspace. During this automatic demonstration, cylinder lengths and operating point coordinates are displayed in real time on the computer screen. Next, the user is given the option of entering either cylinder lengths or operating point coordinates or quitting the program. If the values entered are not within the workspace, the program asks the user to reenter the values. The program then moves the device to the desired set point and controls it there until a key is pressed.
The main breakthrough of this design is the ability of the relatively inexpensive minimanipulator to achieve positioning to within 0.0002 inch, well above what is required.
The resulting manipulator provides a precise, modular, portable, adaptable, and durable device that can be used for any operation that requires manipulation. As a fixturing device, this precision will allow better-quality parts to be manufactured and may decrease some of the need for inspection. The modularity and portability will help in fixture design and setup. The adaptability will also help in fixture setup, and in debugging and making the best of bad parts. The durability will allow the device to be used in a harsh manufacturing environment.
The resulting manipulator achieves all the design goals. Most importantly, it proves that a device can be built to solve a manufacturing fixturing problem. Another benefit of the demonstration device is that it can be used as a laboratory experiment for the mechatronics course at Rensselaer Polytechnic Institute to illustrate the fundamentals of computer-controlled hydraulics.
The natural next step for this device is to attempt to apply it in an existing simple fixturing operation. This will allow the benefits of the device to be quantitatively examined and will justify actual implementation of the device in the factory.
Karl Kurz is an assembly line engineer at Xerox Corp. in Webster, N.Y.; Kevin Craig is an associate professor in the Department of Mechanical Engineering at Rensselaer Polytechnic Institute in Troy, N.Y.; and Barry Wolf is a principal scientist and Fred Stolfi is a member of the research staff at the Webster Research Center Mechanical Engineering Science Area of Xerox Corp. in North Tarrytown, N. Y.
|Printer friendly Cite/link Email Feedback|
|Author:||Kurz, Karl; Craig, Kevin; Wolf, Barry; Stolfi, Fred|
|Article Type:||Cover Story|
|Date:||Jul 1, 1994|
|Previous Article:||Agile manufacturing in a responsive factory.|
|Next Article:||Transferring solid model data.|