In this section we describe the components that make up the machining cell. The primary manufacturing elements are shown in figures 3-2, 3-3 or 3-4.
CNC machine tools:
· A vertical axis CNC milling machine. This machine tool should be equipped with closed- loop control of the position of the table and the spindle to provide positional accuracy on the order of +0.0005" or -0.0005".'The spindle speed should be software controllable. The mill should be equipped with a tool changer that will hold a large number of tools to accommodate work pieces requiring many different milling operations and to allow the processing of a variety of work pieces without pausing to install different tools in the tool changer. The power of the spindle drive motor is not critical for parts made from bar stock, such as the parts family discussed in section 4.1. Similarly the range of travel of the table and the spindle are not critical for these relatively small parts. Examples of a suitable machine tool include the Bridgeport BTC I and the Matsuura MC-SOOV vertical axis CNC milling machines.
· A horizontal axis CNC turret lathe. Like the milling machine described above, this machine tool should be equipped with closed-loop position control on all axes (in this case the longitudinal stroke or z-axis displacement and the cross stroke or x-axis displacement of the turret). This control will, like the control for the milling machine, be required to hold positional accuracy within the range of +0.0005" or -0.0005" . The spindle speed should be software controllable. The number of tool positions available on the turret should be as large as possible to promote flexibility in machining complex and/or varied work pieces. Any less than, say, eight tool positions would significantly reduce this flexibility. It is unfortunately inherent in the design of a turret lathe that fewer tool positions are available on the turret than would be found on a typical tool changer for a milling machine. Neither the power of the spindle drive motor nor the range of travel of the turret in the x and z directions is critical for the family of small work pieces under consideration. A typical example of a lathe of this type, size, and capability is the Miyano model CNC-6IK-3 turret lathe.
Parts handling devices:
· A lathe loading robot. The lathe loading robot is a specialized manipulator designed to be able to reach into a CNC turret lathe for loading and unloading parts from the chuck. The commercially available lathe loading robots are not particularly sophisticated machines. In general they consist of two arms suspended from a carriage riding in an overhead track, rather like an overhead crane. One arm is usually used for loading new pieces into the lathe and the other for unloading finished pieces. The arms are pneumatic or hydraulic and operate by means of limit switches. The grippers at the ends of the arms may be pneumatic or hydraulic. Only the movement of the carriage in its track is continuously programmable. This is accomplished by driving the track with IX servo motors and either ball screws or a rack and pinion. Examples of commercially available equipment are the Manca "Double Arm Gantry System" and the IS1 lathe loading system. A more flexible (but more expensive) alternative to the specialized lathe-loading robot is to use a 5 or 6 axis robot, such as the one described below, for tending the lathe. Another possibility is to mount the mill-loading robot on a track running between the mill and the lathe. These possibilities are illustrated in figures 3-3 and 3-4 and discussed below and in section 4.4.1.
· A mill-loading robot. This small robot requires at least five programmable axes to pick up a part from a flat surface and to orient it on the bed of a milling machine. With six axes the robot can place parts into fixturing that is arbitrarily inclined with respect to the horizontal or vertical. 'The robot should be quite accurate since it will be used for precision loading of parts into fixtures on the mill. A number of accurate, 5 or 6 axis robots are available. They include the Cincinnati Milacron T3-726, the ASEA IRb-6, the CRYO 820 and the Bcndix AA-160. The quoted repeatabilities of these machines are all within 20.008 inch. These robots are not large, but they will reach at least 30 inches and they will all handle at least 13 lb. Thirty inches should be enough to reach both the center of the linear table and the bed on the milling machine shown in figure 3-2. A 13 lb. limit for the robot arm effectively limits workpieces to 10 lb. once a gripper is mounted. This is a bit restrictive for pieces that are not made from aluminum although we note that pieces can enter the cellweighing more than 10 lb. if they first go to the lathe and weigh 10 lb. or less after turning. If 101b. seems too restrictive, then one of the larger robots in this group such as the CRYO 820 or the Bcndix AA-160 should be picked. If the robot is to be used for loading the lathe in addition to the mill (as suggested in figures 3-3 and 3-4) it will probably require a reach of more that 30 inches. This is particularly true for figure 3-4 where the reach of the robot determines how far apart the machines can be placed. If the larger robots in the group above do not have enough reach then it will be necessary to use a robot such as the Cincinnati Milacron T3-746 or the ASEA IRb-60. For a precision cell, it is not desirable to pick a larger robot than necessary since the increased reach and payload come at the expense of accuracy. Another Consideration is that CNC lathes generally offer less clearance for loading and unloading than CNC mills do: this may rule out any robot with a bulky arm. Figure 3-3 shows a very flexible arrangement in which the robot is mounted on a track to improve its reach. Some robots, however, have stringent mounting requirements and would not perform well on a track.
Robot controllers are much the same as machine tool controllers and the integration problems with both are explained below under controllers. Some robots are much easier to program than others. The better ones provide coordinate transformation packages, tool or "hand" coordinates, and easily specified inputs and outputs for working the gripper. This is a distinct advantage but the state of the art in robot software is changing so rapidly that if a particular robot does not provide such amenities today, it may offer them in a few months.
Linear table.
In figure 3-2 a linear table is used to transfer parts between the lathe loading robot and the mill loading robot. In addition, the table may stop beneath the camera of a vision system. The table carries a pallet with some simple fixtures to locate parts so that the robots can pick them up easily. The pallet should be big enough to hold several parts and will therefore offer a bit of in-process storage. The linear table must be quite accurate, but it need have only a few fixed stops. Linear transport tables are made by a number of manufacturers including SI Handling Systems.
Controllers.
Controllers for machines and parts handling equipment -- The controllers available in today's market provide good support for the operation of the individual machine tool. Our requirement, however, is to place a controller in an autonomous machining cell. This has two important consequences:
1. the controller must be able to send and receive complex messages from the cell host, and
2. the controller must allow the cell host to manage the took actions.
In particular, the controller must allow the cell host to do the following:
· Command the execution of programs and/or subprograms on the machine controller.
· Down-load parts programs to the machine controllers and receive up-loaded programs from the machine controller.
· Access the state of the machine controller. The access will consist of a message containing relevant information about the machine and its controller (including the robots). This is essential for the cell host to maintain a detailed account of the status of the cell.
There are additional functions the supervisor might want, but the list above already represents functions that no existing controller provides in a reasonable form. Most existing controllers do provide communications channels (generally some combination of parallel I/O ports and serial data lines) but few will allow a cell host to command the running or stopping of a controller program. Program up-load and down-load is provided on the more advanced controllers but it requires operator intervention. This is unacceptable for an autonomous cell. Solutions to this dilemma come in a couple of flavours, but basically involve either putting a small computer between the controller and the supervisor or re-writing the controller software. The people who make machine tool controllers will build controllers capable of supporting the functions above.
Vision system
Automated visual inspection will be used both to locate incoming parts and to perform final part inspection. Anumber of manufacturers produce what are essentially stand alone microprocessors combined with a great deal of special purpose hardware and software, all designed to process images from one or more video cameras. By process images we mean, roughly, the following: acquire an image (what amounts to a "snap-shot" of a particular field of view at a particular instant), convert the image to digital form, and perform a variety of computational tasks on this digital information. The computational tasks range from simply recognizing a distinct object to calculating its position and orientation within the field of view of the camera. Some of the more important differences between various vision systems are:
· Vision systems, like machine controllers, are based upon a microprocessor. It follows that the more advanced the microprocessor used in a particular system, the more desirable the system. Along these lines, a vision system utilizing a microprocessor with a 16 bit architecture is preferable to one using a microprocessor with an 8 bit architecture.
· Just as machine controllers must support two-directional communication with the cell host, so must vision systems. Furthermore, the need to allow the cell host to control the system's actions is much the same for a vision system as for a machine controller. Both of these characteristics are absent in some of the currently available commercial systems.
· All commercially available vision systems process pictures either as binary or as grey-scale images. Here the term binary implies that all images are seen as combinations of pure white and pure black dots or pixels. Grey-scale, on the other hand, implies that images arc seen as combinations not only of pure white and pure black pixels, but also of pixels having intermediate, or grey, levels of light intensity associated with them. Avision system which provides grey-scale capabilities is more desirable than a binary system if the intended application for the system can benefit from the additional information provided by grey- scale information. Similarly, if grey-scale processing is desirable for the application at hand, the resolution with which levels of light intensity can be distinguished by the system becomes important. Thus a system capable of distinguishing, say, 16 distinct levels of light intensity would be more desirable than a system which distinguishes only 8 grey-scale values. The software required to make use of this grey-scale information is often not commercially available and may have to be developed for specific applications. (see Soft warebelow).
· Another, frequently used criterion for comparing vision systems is the number and arrangement of pixels present in the imaging element of the video camera. Simply put, the more pixels there are in the imaging element, the more information will be available from each picture taken. In other words, the system is able to distinguish finer details in the image. The arrangement of the pixels is typically in the form of an (approximately) square matrix. For certain applications, however, the arrangement of the pixels may vary, such as in the case of linear array cameras intended to acquire information from a single axis (ie. A one-dimensional image rather than a two-dimensional image). Common rectangular imaging elements range from 128 X 128 pixels (rows X columns) to approximately 256 X 256 pixels, while a typical linear array may consist of a single row of 4096 or more pixels.
· Software. This is probably more important than all of the characteristics of a vision system described above. The software provided with a particular vision system will control, to varying degrees, the speed, accuracy, and overall utility of the system. Here software refers to everything from the lowest level image acquisition and processing routines to the highest level user interface. During the latter half of the 1970s a set of algorithms for binary image processing was developed at Stanford Research Institute (SRI). Since these algorithms arc efficient and in the public domain, virtually all commercially available vision systems utilize them for the very low level tasks of image processing. The most significant differences in the software provided with various vision systems are therefore found in the higher level sections of code, particularly the user interface. At one end of the spectrum, the user interface may be a fairly simple, user friendly, menu-driven routine which prompts the user for a command which in turn causes a particular action to be taken. Towards the other end of the spectrum the user interface is far more flexible, but less user friendly. The interface is an interpreter with a syntax reminiscent of a Pascal which enables the user to define, store, and execute complex functions.
An example of a typical vision system, featuring a 16-bit microprocessor, 16 distinct grey-scale levels, the ability to handle images of 244 X 248 pixels, and an advanced high-level programming language (RAIL) implemented as an interpreter. is the Automatix AutoVision II. This system might typically be combined with a General Electric modelTN2500 CID solid state video camera having 244 X 248 pixels sized 0.0014" X 0.0018".
Clamps and fixtures
Clamps and fixtures play an especially important role in a near-term flexible machining cell. This is because they are expected to locate and align parts in addition to holding them while they are machined or transported. These fixtures and clamps fall somewhere between the hard automation fixturing used for high volume production and the versatile clamps and tooling used in job shops, Like the fixturing in hard automation, they are designed to facilitate automatic loading, locating, and clamping of parts. At the same time, they must show some of the versatility of job shop tooling so that they can accommodate any of the pieces that the cell might produce. It is not possible to buy ready-made fixturing that fits this description. Instead, the fixturing will have to be built from a combination of components, some intended for automated production and others designed for manual production. The grippers used on robots constitute a special type of fixturing, less rigid than the clamps on machine tools but lighter and more versatile. Both the machine tool fixtures and the robot grippers will be provided with simple sensors, such as microswitches, to indicate whether workpieces have been properly located against them (see sensors). The robots can also be given Remote Center Compliance devices to reduce the contact forces occurring between parts and fixtures. These passive compliant units are manufactured by ASTEK INC. and Lord Corporation. They were actually designed to aid robotic parts assembly, but the act of sliding a pan into a precision fixture is much like an assembly problem. The mechanics of accurately loading parts into machine tool fixtures are discussed in more detail under section 4.4.
The ability to establish and maintain close tolerances on parts from one process to another depends entirely on the accuracy and repeatability of the fixturing in this cell. A more advanced cell, provided with force, pressure and slip sensors and advanced vision and gauging equipment, would be far less dependent on fixturing. Some of these more advanced sensors are discussed in section 4.7.
Host computer.
A number of manufacturers make Computers which are suited to cell supervision. A 16 or 32 bit internal architectureis sufficient, with the 32 bit models probably being more appropriate. A computer system built around the Motorola MC68000 series of computers or one of the small Digital Equipment Corporation Vax, say a Vax 11/730, would be excellent. With any particular choice though, the designer must keep several factors in mind:
· Considerable software development will be required to implement this cell. The correct choice of operating system, source language, and support environment can greatly influence the efficacy of the development effort the system will need the power and speed to service communication with all the machine controllers and must provide serious support to programmers building system programs. In particular:
o the cell supervisor scheme will need to be built largely from scratch. This program will be a major system level program.
o Software will need to be written to talk to the machine controllers. There are many excellent designs for communication protocols between computers, and the operating system should support their implementation.
The U N I X operating system is an excellent program development operating system, and it provides good services to system programmers. It is a time-sharing system, though, so the machine tool communications system must be designed very carefully.
· The cell host computer will initially be a stand-alone computer. It will only need to communicate to the machine controllers below it in the hierarchy. Later development may place the computer within a factory-wide network.
o The operating system should support an available network system. Alternatives abound, from Digital Equipment Corporation’s DECnet (phase III), to the proposed multi-company ether-net standard, to many others.
o The initial cell host, as a stand-alone computer, will need all the support equipment that a full fledged system requires, for example, disks and tape drives. Later, as a part of the factory, it may only need the processor and main memory. The latter configuration assumes that a factory level computer can down-load operating systems and applications programs. This particular decision has implications for the topology of the factory-wide computer system and should therefore be deferred for the time being.
Sensors
Any automated manufacturing cell depends on sensors. To a large extent, the quantity and the sophistication of the sensors in a cell determine its ability to function autonomously. The cell treated in this document is only semi-autonomous and therefore requires fairly simple sensors. These will be of several types:
· Optical sensors about the linear table to signal its location, and indicate the presence of parts on it. These optical sensors can be photodiodes and will work in much the same fashion as mechanical limit switches. They are distinct from the vision system, which will also be sensing the location of parts on the linear table, but which is a much more powerful and complex device and is treated separately.
· Microswitches mounted on machine tool fixtures and robot grippers. Some of the microswitchcs will indicate whether a gripper or clamp is open or closed. The machine controllers need signals confirming that the commands to open or close clamps and grippers have been successful. Otherwise, a robot might try to load a part into a clamp that never opened due to some equipment malfunction. Other microswitches will be mounted where they give a signal only if a part is correctly located within a gripper or a fixture. For example, if parts are supposed to rest snugly against the back plate of a gripper then a group of microswitches mounted at the back plate will all be depressed only if the part is in place. A particular arrangement of switches can only be expected to work for parts within a single family, which means that when the fixturing is modified the switches must also be rearranged.
· Pneumatic sensors can be incorporated into grippers and fixtures. The rate of air flow from these sensors, or the back pressure felt by these sensors, indicates whether the clamping surfaces are firmly contacting the face of a workpiece. Ideally, the air emanating from these sensors should also help to keep the contact areas free from chips and dirt.
· Finally there will be a number of sensors detecting things like air pressure, oil pressure, and motor temperature for the machines in the cell. These are wired to signal the machine controllers when something goes out of the normal operating range, but the cell host should have access (through the controller) to them as well.
Many other sensors are also available. For example, strain gages could be used to measure the clamping force of a fixture, piezoelectric accelerometers could measure the vibration during machining, and linear diode array cameras could measure the length and width of pieces. A compact optical array could also be used to detect the edge of a part in a fixture, and to indicate how far the part was loaded into the fixture. All of these devices require developing control algorithms to make use of the information they produce. They also cannot improve the fundamental accuracy of the cell until more advanced robots, grippers and fixtures become available to work with them. For these reasons. we omit them from our near term cell. Where examples of specific pieces of equipment have been given in the list above they reflect preferred choices. For instance, the 6 axis robots listed above all have coordinate conversion software and have a facility for down-loading and up-loading programs. If a robot is chosen that lacks these amenities then a considerable amount of development time will be spent in writing extra software.
