|Home Biography People Places Multimedia: Making It Work On the Water Writings/Presentations|
Automated Testing of HEPA Filters
May 7, 1987
Presented at the 33rd Annual Technical Meeting
Institute of Environmental Sciences
San Jose Convention Center
San Jose, California
May 7, 1987
By W. Tay Vaughan, III
Key Thinkers Incorporated
1832 Woodhaven Way
Oakland, CA 94611
During my undergraduate days I had an anthropology professor who succeeded in impressing several heretical ideas into the immature gray cells of my mind. One definitely "outre" idea was that a human child isn`t really human at all, by definition, until it is able to manage symbols and communicate; until then, the child is but an untidy and demanding animal without identity in the Linnean taxonomy.
Robots, I have discovered, follow a very similar developmental trajectory which includes a period of neonatal non-definition. Robots don't become robots until they begin to successfully communicate and perform their assigned tasks. Until they actually work, they are but an expensive assemblage of metal, plastic, and electronic parts.
As those of you in the audience who have conceived, designed, and built a robot or any other automated device will certainly agree, there is a well-defined period of difficult time between first applying power to the robot (which can be a very hushed and religious moment) to delivering the finished product, with invoice, to its new owner (an event usually satisfyingly secular).
During the last two weeks at KTI we have experienced this special time in the development of a robot, driven in part by a desire to have it ready for these IES meetings. Human baby parents get very little sleep during the neonatal period. Robot parents also find themselves running short on sleep. Fourteen all-nighters (counting all the parents involved) were given to our robot since mid-April. The software guys call it "Debugging"; the hardware guys call it "Tweaking".
Indeed, there is a certain satisfaction and sense of productivity in working the night through, pursuing problems and gleaning answers without mundane distraction. After a while, though, as I am sure many of you know, the all-night magic of the quiet hours (when watchmen come around with their funny little keys and round clocks and ask "How's it going?") drains away into the sands of dogged perseverance.
I tell you all this by way of setting the scene. As of Monday this week, our robot has come of age and is not only sweetly talking, it is responding precisely to commands. Quite a change from a week ago when we got so mad at its backtalk of ACKs and NAKs and buffer overflow that we electrocuted the recalcitrant critter and carried it back to its Maker for fundamental obedience training. It was thumbing its nose at Asimov's second law:
A robot must obey orders given by humans...
A few EPROMS later, however, its Maker, actually Mitch Weiss and his crew of wizards from ProgramMation in Paoli, Pennsylvannia, had it speaking proper Pygmalion English. The Robotics Industries Association defines an industrial robot as a
"...reprogrammable, multifunctional manipulator designed to move materials, parts, tools, or special devices through variable programmed motions for the performance of a variety of tasks."
The KTI/Flanders Autoscan™ robot is actually a straightforward system of sensors, analog to digital converters, software, operator controls, and a ProgramMation motion device designed to scan High Efficiency Particulate Air filters, searching for pinhole leaks and performance defects. Particle penetration, velocity, and pressure differentials are measured across the face of the filter in a programable pattern using one or several user-selectable data acquisition algorithms. In its total form, it is a software-driven cartesian coordinate robot with certain neat capabilities: It walks and it talks! I'll go into details in a moment.
The Autoscan™ project is sponsored by Flanders Filters of Washington, North Carolina. The Autoscan™ robot was designed to provide repeatable and accurate information as part of on-going quality control and quality assurance programs both at the Flanders manufacturing facility itself and in the field at the time of new filter installation and certification. It was also designed to to traverse the downstream face of HEPA filters deployed in plenum filter banks or in the special containment housings used in nuclear and bio-medical research environments, places where scanning by hand is either out of the question or very expensive in terms of human exposure to a hostile environment.
We had hoped to present to you today some real data in the form of a study of HEPA filter scanning techniques currently used in the industry. This will have to wait until our next meeting, as we are just now setting up the system for actual testing on the production line and in the lab at Flanders. Nonetheless, part of what I will discuss this morning concerns testing methodology according to the relevant IES Recommended Practices 001, 002, and 006 as well as the existing Federal Standard 209B and the ANSI/ASME Standard N510 ("Testing of Nuclear Air-Cleaning Systems"). As I go along, I will describe the Autoscan™ approach to operator control, instrument deployment, motion routines, and data archiving for statistical analysis.
The first obvious benefit from automation of the filter scanning process is the REPEATABILITY of test conditions and data acquisition. Locating and quantifying particle penetration and air velocity anomalies across the face of a HEPA filter is a nondeterministic, nonstationary random or stochastic process which readily lends itself to both parametric and nonparametric statistical study.
Occasional pinhole leaks may occur in the filter medium during manufacture, filter edges may be improperly sealed where the medium meets the housing, or filters may be physically traumatized during assembly or shipment. Identification and repair of these random imperfections is a normal part of quality control. While the customer may never see an imperfect filter, the filter may indeed have been repaired (following strict specifications and guidelines) at the manufacturer's quality control and rework station prior to shipment.
It is these repair data, of course, which are of greatest interest to a quality assurance program. Leak data, once acquired and archived, can easily be correlated with such variables as location (Are edges giving a problem?), media batch formulations (Are some formulas less prone to defects than others?), and even time of day (Is the Monday morning shift performing as well on Wednesday afternoons?). Once the data are gathered, statistical quality control software such as Northwest Analytical's "Quality Analyst" can help identify sources of VARIATION. While variation cannot be fully eliminated in any manufacturing process, for a company like Flanders Filters that rigorously pursues a zero-defect philosophy, it can be reduced and controlled and the consequences minimized. Thus automation of the scanning procedure, coupled with data storage, retrieval, and analysis, feeds directly back to the manufacturing process in a very positive way.
As Hill and Eassa wrote in Microcontamination last March about robots used for memory disk certification,
"Analysis of data generated and recorded by the robot stations has proven valuable [at Domain Technology] in the detection and control of problems that occur elsewhere in the production line."
Time after time we hear that good quality assurance programs with proper feedback channels are effective problem solvers.
Repeatibility of testing procedures provides also a second significant benefit perhaps more of interest to you in the audience this morning. Because filter scan conditions may be controlled or varied at the experimenter's will, very precise measurements may be undertaken while holding one or more variables constant. One experiment we are anxious to begin, for example, is a study of the relationship of scan rate (presently mandated by most standards at a maximum of 10 feet per minute), to the accurate detection of particle penetration. Holding all conditions equal, the Autoscan™ robot can traverse the filter face at any speed up to thirty feet per minute, time after time, precisely. Another experiment will be a study of the relationship of airflow velocities through a filter face to particle penetration and to the construction style and pleating of the filter, with the robot providing very fine positioning control.
I would welcome suggestions from any of you regarding other experimental ideas or test methodologies which might add to the general knowledge and further our understanding of air filtration phenomena, now that we have this tool to work with. As the Standards I mentioned above are continually revised and updated and as our understanding increases, we hope to be able to contribute a significant database to the standards-making process. As Mitch Weiss cautioned, however, in a 1984 article in Solid State Technology, probably less tongue-in-cheek and more tempered by the sometimes tart spices of reality: "A successful robot installation requires a thorough knowledge and understanding of the tasks to be performed. The gaining of this knowledge often takes longer than the implementation of the robot itself."
It seems that we are always on a steep learning curve! As we learn more as a result of higher-resolution tools, we will seek new applications, techniques, and methodologies for this system.
A second benefit realized by the Autoscan™ robot, in addition to REPEATABILITY, is that it can be programmed to do its job in hazardous areas where humans need to suit up and work under strict exposure time limits. Federal regulations limit whole-body radiation exposure among nuclear workers, for example, to a maximum of 5 rem per year. For your information, radiation received by the average person in the United States in one year due to natural background sources reaches a maximum of about 125 millirem per year. Scanning a single two- by four-foot filter face at ten feet per minute takes more than four minutes under the most ideal conditions. Keeping workers out of hot areas is encouraged.
The Autoscan™ robot can also be networked to a Facilities Monitor System such as that described in the IES Journal last December by Allart Ligtenberg of Hewlett Packard. Such monitoring, he stated about his facility efforts at the Cupertino IC Division, "...ensures industrial hygiene and environmental compliance," and [for his facility] ultimately resulted in significant cost savings.
These monitoring efforts and automated systems are certainly part of the factory of the future. But, you know, the future is now. As Steve Scheiber, Technical Editor of Test & Measurement World, says, "One drawback to studying the future is that, however long you wait, it never arrives." He hasn't given us a Confucian maxim here, but rather he has affirmed the fact that each incremental step forward drags the future into the present. For those of us in the trenches, this sort of makes the toil and the all-nighters worth it.
Now, before my time is up, let me give you some brief technical details about the Autoscan™ system.
We drive the robot with an IBM PC/XT clone. On board the clone are a Metrabyte Dash-8 12-bit A to D converter, a Votan continuous speech recognition card, and a 10-Meg hard disk. We use a CGA color monitor.
The clone daisy-chains several compiled programs according to the needs of the operator. A Main Menu provides custom set-up of test parameters and, for use at the QC and rework station, automatically sets up test parameters by Flanders filter type and size, greatly simplifying operator input requirements. A scanning program actually drives the robot through an asynchronous COM port and processes converted and calibrated data which are multiplexed in machine language at the A to D interface. Other programs then display data in 3-D profiles and manage statistical studies. All test results are stored both to hard disk and to floppy, and the unit is purged and polled every three days or so on the production line for archiving and record-keeping purposes. Some eight hundred test results can be stored on a 360 K floppy diskette.
While we were tempted to write these programs in C, we chose to go with a compiled BASIC environment using Microsoft's new QuickBasic compiler. We wished to leave the source code open to modification and change by any number of programmers, some of them unknown to us, some of whom were conversant in BASIC but not in C (or Pascal or Modula-2). While traditionally, BASIC compilers have suffered lack of respect among professional software developers, we felt justified in this choice and then later redeemed when Marty Franz wrote in PC Tech Journal a few months ago that "Several versions of BASIC have been announced for the PC, many of which are good enough to make serious programmers once again consider BASIC..." and gave the Microsoft compiler his highest rating.
Three channels of the A to D converter are used to input analog voltage signals from a Flanders Model 630 aerosol photometer, a custom-made TSI Model 1610 hot wire velocity probe, and a CELESCO differential pressure transducer. A fourth channel is used to interpret switch settings onboard the photometer which are automatically checked against the type of filter being tested and the rated efficiency of its medium.
A Votan continuous speech system is employed to allow the operator hands-free operation. This system requires voice training for each operator (speech templates are digitized for sixty-four command words such as Start, Pause, Menu, and the ordinal numbers 1 - zero), and as the system comes on line, we will study the practicality of voice actuated equipment in the workplace. You will hear more about this at our next meetings.
We worked up a natural command language with the ProgramMation team which allowed us to deal in English instead of Hexadecimal, and made the robot device independent of special software. We can drive it from any dumb terminal.
A special programming consideration was management of data acquisition at various speeds and at various sampling rates. The display screen posts anomalies (when any of the three measured parameters falls outside of acceptable limits) in one-inch increments. The program may be adjusted, however, to sample sensor data every eighth of an inch. For this, we developed selectable averaging or worst-case algorithms to summarize the data over one-inch increments. We were in the classic dilemma of digital versus analog; digital implies a sample taken during a slice of time and lost information inbetween slices. Having studied the motions of hand-scanning technicians, we concluded that one-eighth inch was more than sufficient resolution when dealing with the broad airstream fluctuations associated with pinhole leaks.
To relieve the main program of the processing overhead of timing and position counting, the robot provides "tick" signals to the controlling program every time the robot crosses a definable grid line. These signals trigger data acquisition and smoothing algorithms in the main program, allowing processing time to do computations and post the results to the screen. At 10 feet per minute, or two inches per second, there are sixteen rounds of A to D converter samplings per second which include the solution of fourth order equations and averaging computations, and each half a second, or at one-inch increments, data is posted to the screen, a "slow" process on IBM PC's.
I have attempted to summarize the developmental effort so far invested in the Autoscan™ system. Both Mitch Weiss from ProgramMation and George Cadwell from Flanders are here with us this morning, and perhaps among the three of us we might answer any questions you have. Without being commercial, and more in the venue of "proud parent", I would also add that the Autoscan™ system is on display at Booth 402.
Thanks very much.
1. Groover, M.P., Weiss, M., et al, "Industrial Robotics", McGraw-Hill, New York. 1986. Pg. 5.
2. Bendat, J.S. and Piersol, A.G., "Measurement and Analysis of Random Data", John Wiley & Sons, New York. 1966. Pg. 9.
3. Pierce, A., "Fundamentals of Nonparametric Statistics", Dickenson Publishing Company, Belmont, CA. 1970. Pg. 33.
4. Hill, J.E. and Eassa, K.D., "The Use of Robots for Memory Disk Certification in the Cleanroom", "Microcontamination", Vol. 4, No. 3. March, 1986. Pg. 47.
5. Weiss, M., "Automated Wafer Processing Using Robots", "Solid State Technology", Vol 27, No. 7. July, 1984. Pg. 167.
6. Fischetti, M., "The Puzzle of Chernobyl", "IEEE Spectrum", Vol. 23, No. 7. July, 1986. Pg. 38.
7. Ligtenberg, A., "Advantages of Facilities Monitoring System in Integrated Circuit Fabrication", "The Journal of Environmental Sciences", Vol. 29, No. 6. November/December, 1986. Pg. 41.
8. Scheiber, S., "The Factory of the Future is Here!", "Test & Measurement World", Vol. 5, No. 12. December, 1985. Pg. 24.
9. Franz, Marty, "Reconsidering BASIC", "PC Tech Journal", Vol. 4, No. 12. December, 1986. Pg. 143.