Equipment Reliability Institute
ERI News - your reliability newsletter
November 2004 - volume 17

Hello, readers -

We intend each issue of this Newsletter to help you with your problems. First: who among you is involved with the cooling of electronics? That’s the area in which Joel Newberger, now affiliated with ERI, consults and teaches. You might want one of his training classes (titles and brief description here) to meet at your facility.

Second, Chuck Wright writes about measurement uncertainty (he and I both prefer to emphasize certainty) and how you can deal with uncertainty requirements. You might want one of his training classes to meet at your facility.

Third is a contribution by myself, dealing with one of my favorite subjects: the simultaneous exciting of multiple resonances. That’s why we perform random vibration tests. You might want one of my training classes to meet at your facility. My “open” classes (to which you can come as an individual) are listed at both our Web pages and also here, at this issue.

Fourth are some additional “Test Lab Musings” by highly experienced test engineer Bob Renz of G/D Bloomington, MN. I had the pleasure of seeing Bob at the recent SAVIAC meeting. Comments or questions for Bob? Click here.

Best wishes,
Wayne Tustin

Cooling Board Mounted Components
Utilizing “Planes” and “Bridges” or
Maybe Passive Cooling of Printed Wiring Boards
by Joel Newberger

Thermally effective circuit board designs utilize embedded copper ground “planes” and via(s) (“bridges”) to minimize component thermal signatures.

Electrical vias are plated thru holes that interconnect signal, power, and ground board circuitry to embedded copper layers (planes). Sometimes, electrical vias behave as thermal vias . . . but not always! Thermal vias are also plated through holes, but are dedicated to conductively couple component heat to the dip side ground plane thermally conditioned by heat sinking to a chassis septum or flange(s). Copper layer(s) thickness range from ½ to 2 ounce copper. Thermal vias traverse, or bridge, the entire circuit board from component mounting surface to dip side ground plane. Electrical vias, whose principle function is electrical connectivity, do not necessarily traverse the entire board. Electrical vias are generally less effective since in most designs, they may not be coupled to the ground plane.

Unetched copper, at any layer on a circuit board, is useful since unetched copper spreads component heat in transverse and lateral directions . . . which is helpful in minimizing thermal signatures. Therefore, leave as much copper as the electrical design allows! Two-dimensional heat spreading takes place at every embedded copper plane. Thermal vias must be the principal “bridge” from component to circuit board embedded planes, especially, the dip side ground plane, in order to minimize component thermal profiles. Sometimes, thermal design difficulties may be eliminated if copper plane(s) thickness is increased.

At a board thickness of 1/16 of an inch, a 0.02 inch diameter plated thru hole thermal impedance is characterized as follows:

At 1/32 of an inch board thickness, via thermal impedance is one half the impedance levels shown above.

As an example, consider an IC dissipating 750 mw. If five thermal vias (.02”dia, 2 mil wall) are located below IC case foot print envelope, the effective thermal coupling across the circuit board is 55.2C/W/5 vias = 11.04C/W. Therefore, temperature rise across vias (board thickness) is .75 watts x 11.04 C/W = 8.28 C. With ten vias, the temperature rise is reduced to 4.14 C. Coupling heat from component case to board surface can be accomplished either by soldering case to vias, or utilizing easy to apply “filler” material (limited to below 8 mils thick if a non thermal conductor is used). The “filler” is used to bridge the void between component case and circuit board.

Circuit board materials are generally poor conductors of heat; however, when viewed through the prism of “planes” and “bridges,” consisting of embedded copper layers, (even when partially etched for circuitry) and smart thermal via placement, it is transformed into a cost-effective conduction-cooled, component mounting platform. This thermal design approach has been successfully exploited in all types of applications ranging from indoor/outdoor telecommunications equipment, industrial equipment, in-orbit space equipment, shipboard electronics, and aircraft avionics.

Over 6,000 years ago, the Ancient Sumerians knew the value of copper (they prized silver and gold too!). All three materials are excellent thermal conductors. Similar to the Ancient Sumerians, the circuit board designer should develop an appreciation of the thermal benefits associated with embedded copper “planes.” Once appreciated, use of thermal “bridges” will follow.

Joel is President of Thermalogics, Inc., and a principal in SNA Engineering. SNA specializes in mechanical design/packaging and in thermo/structural analysis of electronic equipment used in both commercial/industrial and military applications.

For a listing of available short courses (listing of 15+ short course descriptions) and more information about Joel, please visit his page at ERI. Joel invites readers to "mix and match" these courses and thus to identify electronics cooling training that can in 2005 meet at their facilities.

(back to the top)

Measurements Uncertainty?
I Prefer Measurement Certainty
by Charles "Chuck" Wright

 I used to think I was the only test measurements engineer that absolutely detested doing uncertainty analyses on my measurement system designs and test data. A legitimate uncertainty analysis can be a bookkeeping nightmare – bias or precision? Precision or bias? Have I got all the meaningful errors? How do I put it all together? Where should I focus my attention?

More than 80 uncertainty sources for a simple static surface strain measurement at room temperature have been identified. Each would have to be assessed in terms of bias and precision errors if a valid and complete uncertainty analysis were required. That’s more than 160 terms requiring investigation!

In teaching the Applied Measurements Engineering Short Course all over the country for the last 14 years, I’ve discovered that I am not alone. Most practicing measurements, instrumentation or data acquisition engineers, in my classrooms at least, simply do not do uncertainty analyses on their systems or their data. Some do, most don’t.

How can this be? How can an engineer sell his test data to the customer without defining the uncertainty in that data? An answer of 2120 Gpk maximum in a shock test is meaningless without an uncertainty statement and validation. Why would any smart structural dynamics analyst, paying for that test by the way, accept such an unsubstantiated answer? How can an engineer design a measurement system without using allowable uncertainty as a design requirement? Does any experimental mechanical engineering test laboratory have a legitimate written process in place demanding uncertainty analysis and stating a method for performing that analysis? Let me try to answer those questions for you from my experience teaching measurements across the country.

Before you can validate your test data, the experimental uncertainty must be known. Before you can acquire the data, a measurement system must be designed and implemented. You should have experimental uncertainty requirements before you design. In the absence of this knowledge, how do you know your measurement system will assure the uncertainties you need to meet the customer’s requirements? How many of us get uncertainty requirements with every test request from the customer for the data? When I ask this question in my courses, it’s usually followed by a roomful of silence. Even today, few test people ever get uncertainty requirements from their customers. Why is that?

The Customer’s Role
Here comes a crack analyst asking for an expensive test. All tests are expensive because they can only occur late in the product development cycle. You need hardware to test and the product owner has already made a large investment just to create that hardware. The test request defines the test conditions and the data requirements. If the customer cannot tell you the uncertainty requirement for every measured channel, then you should not run the test. The test problem has not yet been thought through in enough detail to justify the expenditure. The customer has to tell you how close to the truth you have to come (allowable uncertainty), or you can’t guarantee the validity of the data. If you can’t guarantee the validity of the data, why run the test? In the absence of the uncertainty allowable, in a rigorous world you simply should not test.

Why can’t test professionals get the needed information? A sensitivity analysis of the analytical model and any data reduction equations is required to define the allowable uncertainties in test. Perhaps the analyst was not taught how to do it in college. In this case, we blame the college engineering faculty. We need to educate the educators. Perhaps there is not enough funding on the program to do such an analysis. In this case, we need to educate the program manager. More likely, we test professionals did not demand the information from the customer and explain the reason why we need it. Here, the blame falls closer to home.

You Need the Uncertainty Requirements to Design and Test
You need the uncertainty requirements. Make sure you get them early. How do you get them? I only know of a single test laboratory in this country that has a written process for how uncertainty requirements are to be acquired from the customer. The process also defines a method for evaluating the uncertainty numbers and finishing the error propagation. The process owner told me that sometimes getting the number is painful, but they get them. I have some issues with how they deal with the numbers, but at least they have a process written down. If there are other labs that have written processes for doing this work, please let me know so I can share the information.

Over the last twenty years or so, the major argument in uncertainty analysis has been about what math to use. This argument occurred on the arcane playing field of statistics and partial differential equations. This argument has largely been settled. A good place to start is ISO’s Guide to the Expression of Uncertainty in Measurement, affectionately referred to as the GUM. You can find this at the NIST website. This international standard defines how to crunch the numbers.

The Real and Remaining Problem
The remaining and more important discussion is around what numbers to put in the analysis. Almost all of the numbers I see in published uncertainty analyses reflect only those available from the manufacturer’s specifications for the hardware and calibration data. The analysis shown below is my favorite, from an unnamed periodical’s public website on uncertainty analysis.

What is missing here? Reality is missing here. This analysis violates the most basic requirements of international standards. But worse, it clearly includes only data that was easy to get, as mentioned above. It totally disregards the much larger errors that will inevitably occur in the measurement due to environmental responses throughout the entire measurement system and interface to the phenomenon to be measured. The real system level uncertainty is probably one to two orders of magnitude larger than shown here. I assert that this analysis grossly understates the real uncertainties at the system level.

Most uncertainty analyses I see suffer from the same illness. They include the 0.03% FS errors given by the manufacturer and metrology, while neglecting the 5, 10 and 20% FS or larger errors that actually occur during the test. My course includes numerous examples of this unfortunate engineering myopia. Perhaps the best one is a temperature measurement in a rocket nozzle throat that carried a traceable calibration of 1.5% FS, but read 2500F in error – an “accurate” measurement of the wrong temperature! For a measurements professional interested in a fruitful career, this is not a good thing to do.

In a three decade career in measurement engineering, my organization only guaranteed a system level uncertainty less than 0.1% FS once. It took all the measurements knowledge and skill we could get together. And we could do it only in lab ambient conditions and only at DC. The three channel load measurement system lived in a velvet lined box when not in use and was never disconnected. The bill for each use of the system, including custom calibration, was about $25,000 -- and was worth it.

Some Rays of Hope
DoD flight test laboratories are getting very interested in the uncertainty of their measurements. At least the flight test centers at Edwards AFB in California and the Navy’s at Patuxent River, Maryland are working on this issue. The intent is to eventually have a standard approach to uncertainty across all government test facilities. That approach will probably end up in the public domain. By the way, my sense is that the Europeans are farther down this road than we are here in the states. If you have direct knowledge of this, please let me know so I can share the information.

If these issues are of interest to you, my short course is about the 1, 5, 20% or larger errors you are making, but don’t know you are making, how to avoid them by proper system design, and how to validate your data. This is the stuff they never taught you in college. Contact Wayne Tustin here at Equipment Reliability Institute for course information.

Charles "Chuck" Wright, developer of the Measurements Engineering Department for a leading satellite manufacturer, has three decades of direct and successful experience in the design and operation of advanced multi channel, computer-driven measurement systems for test and evaluation. To learn more about Chuck or contact him visit his page at ERI.

(back to the top)

Random Vibration contains all Forcing Frequencies
and thus excites all resonances
by Wayne Tustin

For many weapon and equipment suppliers to the US Navy, the 1979 Materiel Command (NAVMAT) document P-9492 “Navy Manufacturing Screening Program - Decrease Corporate Costs - Increase Fleet Readiness” was the first they’d heard of random vibration.

What is random vibration?
Let me commence by saying (to newcomers to vibration testing and screening) what random vibration is not. Random vibration is not like the familiar, classical, pure tone, one-frequency-at-a-time sinusoidal vibration of Figure 1. Note the time domain (as on an oscilloscope) trace at top. Note the one-line spectral display beneath with all vibratory energy at one specific frequency f. If that vibration were sensed by the multi-reed spectrum analyzer of Figure 2, one reed would respond strongly.

What is complex vibration?
Vibration that simultaneously exists at two or more frequencies, as in Figure 3, is often called complex or multi-toned. If the helicopter of Figure 4 had three blades, the resulting vibration would resemble Figure 3. With a multi-reed spectrum analyzer resembling Figure 2, two reeds would respond, at main rotor vibration frequency and at blade-passing frequency.

Along with basic vibration theory and learning to measure vibration in the field and in the lab, that’s about as far as we usually get on Day #1 of a 3-day course.

So what is random vibration?
A new kind of vibration appeared in the 1950’s. Early rockets failed at ignition or lift-off or flew off-course and had to be destroyed. This new vibration had none of the regularity of Figures 1 and 3. In theory, at least, random vibration time history never repeats and, as a result, has a spectrum that is quite wide. Rather than only one or two analyzer reeds responding, all respond, as you would see if Figure 5 were one of my video clips.

Do you recall, in high-school physics, how a beam of white light passing through a prism as in Figure 6 becomes an array of colors, showing that white light was a mixture of colors, of frequencies? Shift your thinking, now, from light to mechanics, to broad-spectrum random vibration.

Such broad-spectrum in-flight vibration creates hardware failures that never occur with sine or complex vibrations. Previously-useful sine vibration testing had little value in detecting weaknesses that would appear in flight. Random vibration tests had to be hurriedly developed, in order to identify those weaknesses, so that they could be remedied.

Conclusion
It is hoped that newcomers to random vibration testing and screening now feel comfortable in discussing random vibration. Applications include not only MIL equipment suppliers and users but also office high-rel equipment suppliers and users. All are finding that random vibration is useful.

Wayne Tustin, ERI's president, can be reached by e-mail or phone (805) 564-1260. Read more about Wayne at ERI's website.

(back to the top)

Test Lab Musings (part 6)
by Robert L. Renz

Even though accelerometers are designed for vibration and shock, they don’t appreciate being dropped off the side of the shaker onto the floor. If you drop an accelerometer, check it out before you use it again for a test. How? One way is to attach the suspect unit next to the control accelerometer on a bare fixture. If they don't agree, or if the suspect unit shows high noise levels, it may be time to repair if or replace that unit. Try it with another cable first, though.

While you’re at it, check out all doubtful cables. They are fairly delicate, and can get noisy very easily.

Accelerometers don’t last forever – and Murphy’s law says that they will die just when you need them the most. Put some away as spares. Keep extra accelerometer cables on hand.

If you have ever needed a connector or an adapter for an accelerometer cable, you know that the connectors are unique, and can be a challenge to locate. Check out the Microdot Connector Company’s 50-ohm screw-on coaxial connectors (S-50 series).

Robert L. Renz of General Dynamics - Advanced Information Systems at Bloomington, Minnesota.

(back to the top)

Steve Brenner recently conducted a three-day “open” course on vibration and shock test fixture design, at Pomona, CA. It ended with a visit to a “fixture factory” - Baughn Engineering at LaVerne, CA. I thought the course went well, and the critiques were favorable. I mention the event because the enrollment was very small, not nearly as large as I expected. May I please hear from you? Why were you not in the group? Should we try again? When? Where? Do you need special notification (besides this newsletter, our web pages, etc.)?

If you would like to request a free sample of Chapter 1 - "What are vibration and shock?", from Wayne's new book "(...) Random Vibration and Shock Testing", please visit our website. Fill out the quick form and submit it to us. We will then e-mail you a PDF file of Chapter 1.

Wayne recently attended the 74th Shock & Vibration Symposium, held this year at Virginia Beach, VA. Wayne’s contribution was a tutorial on SRS - the Shock Response Spectrum - and how it fits into shock testing. Visit the SAVIAC site for details of the 75th Symposium, to meet at New Orleans, LA October 30th, 2005.

ESTECH 2005
The 51st Annual Technical Meeting and Exhibition, Estech 2005, will meet on
May 1-4, 2005, at the Hyatt Regency Woodfield, in Chicago.

CEEES Conference
"Methods and Benefits of Environmental Testing and Engineering" will meet in Germany, on May 11 and 12, 2005.

The vet rolled his eyes, turned around and left the room, and returned a
few moments later with a black Labrador Retriever.

As the duck's owner looked on in amazement, the dog stood on his hind
legs, put his front paws on the examination table and sniffed the duck from top to bottom. He then looked at the vet with sad eyes and shook his head.
The vet patted the dog and took it out, and returned a few moments later with a cat. The cat jumped up on the table and also sniffed delicately at the bird. The cat sat back on its haunches, shook its head, meowed softly
and strolled out of the room. The vet looked at the woman and said, "I'm sorry, but as I said, this is
most definitely, 100% certifiably, a dead duck."
Then the vet turned to his computer terminal, hit a few keys and produced a bill which he handed to the woman.
The duck's owner, still in shock, took the bill.
" $150!", she cried, "$150 just to tell me my duck is dead?!!"
The vet shrugged. "I'm sorry. If you'd taken my word for it, the bill would have been $20, but with the Lab Report and the Cat Scan it's now $150.00...."

By David McLaughlin of Davco Weight Scales Inc.

ERI - Equipment Reliability Institute
1520 Santa Rosa Ave.
Santa Barbara - CA - 93109
Tel: (805) 564-1260
Our fax number:
(805) 966-7875

Wayne Tustin tustin@equipment-
reliability.com

Webmaster webmaster@equipment
- reliability.com

Websites
http://www.equipment-
reliability.com

http://www.vibrationand
shock.com

Copyright © 2000-2004 Equipment Reliability Institute. All rights reserved.

Subscribe
If you would like to subscribe to ERI News, go to either website, fill in the form "Free Newsletter" and hit the Submit button. Subscribe now!

Recommend
If you enjoy reading ERI News and want to recommend it to a friend, just hit "forward" on the menu of your e-mail program or tell your friend to subscribe at our website.

Previous issues
Missed the previous issues? It is not a problem. Just visit our newsletter archives section and find all ERI's News issues.

Unsubscribe
If you do not want to receive ERI's quarterly newsletter, please reply to the newsletter message with "remove" as subject or click here.