Oil and Gas Industrial zone,The equipment of oil refining,Close-up of industrial pipelines of an oil-refinery plant,Detail of oil pipeline with valves in large oil refinery.

The EXPAT

=====================================================

16. Design of Experiments (DOE)

=====================================================

          After three years in Indonesia, I was recruited by Intel as Training Manager at a new startup factory in Arizona. Most companies who make Integrated Circuits (IC’s) are done in off-shore locations to take advantage of low labor costs. Intel chose to buck that trend and find profit stateside. While at Intel, I increased my understanding of quality principles and began to sell my services, teaching quality principles, in Arizona at Motorola. Our son Daniel finished high school there and began his engineering studies. The Motorola experience connected me with different people in the business, at a time when quality was in huge demand. The Malcolm Baldridge National Quality award took us to New York and three years as Quality Manager at Standard Microsystems Corp., (SMC), on Long Island.

Malcolm Baldrige was Secretary of Commerce under President Ronald Reagan. During his tenure, Baldrige played a major role in carrying out Administration trade policy, taking the lead resolving difficulties with China and India. Baldrige held the first Cabinet-level talks with the Soviet Union, paving the way for increased access for U.S. firms. The Malcolm Baldrige National Quality award was created in his honor.
An important phase in my quality progress began while at Motorola. What is Experimental Design? Any industrial process is made up of many factors acting together. For example, in a foundry, variables might include the percentage of clay in the sand, compressibility of the sand mix, temperature of the iron as it is poured into the sand mold, the time the casting stays in the mold, and many other factors relating to iron chemistries.

          Question: How do we find the combination of all these factors together to obtain the best result? What is the best recipe? The same questions can be asked of the recovery process in gold mill operations; time/temp relationships in the roaster, time/pressure/temp relationships in the autoclave, chemical recovery from gold-loaded carbon, etc. Each process has many variables that impact process performance. What would a 5% improvement in yield be worth to a company over a year in one of these processes? Or a 15% reduction in cost of a reclamation process?
The long tradition (classical method) to address this question has been to hold all factors constant except one, then adjust that one to find the best level. Next, repeat for the next factor, and the next, and the next… This approach requires (assumes) “uniform conditions” for all factors except the one under study. This is bad for two reasons:

a. “Uniform conditions” are a myth. They never exist. Imposing artificial conditions create a process environment quite different from reality. The results cannot be repeated in the field.

b. It is usually not advisable to try to hold all factors constant except one, because there are often interaction effects between factors that are more important than any direct effect. This occurs whenever the effect produced by one factor is modified by changes in the other.

          A good designed experiment (DOE) is a series of trials, planned by smart people (note the plural), including all relevant factors (variables), sufficient in number to screen out system variation and noise, and arranged in a pattern to provide maximum information, whew! (note the difference between data and information).

U.S. companies face stiff competition in the global market. DOE is a strategic weapon to fight this technical battle. Many feel DOE is too costly and should remain in the laboratories. However, a simple DOE is the most cost effective way to gather this information. For instance, 5 factors (variables) can be studied in a 16-run fractional factorial experiment without disturbing the process. This will give us information on all five factors, as well as the 10 two-factor interactions present. It would require 80 runs to get the same precision from a one-at-a-time experiment, with no measure of interactions!

What is Experimental Design? For example, in a foundry, variables might include the percentage of clay in the sand, compressibility of the sand mix, temperature of the iron as it is poured into the sand mold, the time the casting stays in the mold, as well as numerous factors relating to the iron chemistries.

Question: How do we find the best combination of all these factors together to obtain the highest performance possible? What is the best recipe? The same questions can be asked of the recovery process in gold mill operations; time/temp relationships in the roaster, time/pressure/temp relationships in the autoclave, chemical recovery from gold-loaded carbon, etc. Each process has many variables that impact the process performance. What would a 5% improvement in yield be worth to the company over a year in one of these processes? Or a 15% reduction in cost of a reclamation process?

The classical answer to this question has been to hold all factors constant except one, then change that one factor to find the best level; then repeat for the next factor, and then the next, and the next… This approach requires (assumes) “uniform conditions” for all factors except the one under study. This is bad approach for two reasons:
“Uniform conditions” are a myth. They never exist!

Imposing artificial uniform conditions creates an environment quite different from reality. The results cannot be repeated in the field. It is usually not smart to try to hold all factors constant except one, because there are often interaction effects between factors, more important than any direct effect. This occurs whenever the effect produced by one factor is modified by changes in the other.

A good designed experiment (DOE) is a series of trials with a specific goal, planned by knowledgeable people (note the plural), including all relevant factors (variables), sufficient in number to screen out system variation and noise, and arranged in a pattern to provide maximum information.

The reason and purpose for DOE can be shown with a simple experiment. Our purpose is to show with an experiment how to test a simple wire-bond process. That is what we did at Intel, Motorola and now here at SMC.
The process, connecting wires to terminals inside an integrated circuit (IC), is shown here: The tool places the wire against the first bond-pad, secured (welded) by a burst of ultrasonic energy, then the tool is opened, pulling the wire out as it moves over to the second pad, repeating the bond to the second pad, then closing on the wire to break it off and pull it away from the package.

 

           Ball Bond (a)                     Wedge Bond (b)
(Notice the shape of the tool in each case)

                     Fig. 16.1 Wire-bond process.

          This process, done automatically, controlled by a computer, moves through the geometry of the IC with blinding speed. But how do we know if it was a successful operation? Similar automation tests a sample of the connections by pulling them to breaking point, to see if the bond produced was strong enough. Some sample size, conforming to the spec, says yea or nay. If sufficient bond pulls are within spec, the lot is passed.

(a) Wedge Bond                        (b) Ball Bond

Fig. 16.2 Wires Bonded within packages.

          Our youngest daughter Leslie graduated high-school on Long Island, her siblings having become engaged in marriage and school pursuits prior to the New York move.
To punch-up the importance of DOE, consider the following two-variable process is shown in Fig. 16.3: The output (Z) is some function of X and Y. Or, given the coordinates of X and Y, the output, Z is identified by the function,

Fig. 16.3 Two-Factor Process with 3-D surface output.

Fig 16.4 Pressure/Temperature Yield Control Plot – Actual Process.

          Two factors are thought to influence yield of a chemical process (Temperature and Pressure). Our aim is to find the best combinations of these two factors for highest process yield. If we approach this task, one-factor-at-a-time, let’s see what the process tells us. Fix the temperature, say at 220 degrees. Then change pressure, conducting experiments at different levels of pressure, say at 80, 90, 100, 110, and 120 psi. Identify the pressure that maximizes yield. Graphically, the optimum pressure is 100 psi.
Then leave the pressure at this value, 100 psi, and change the temperature above and below the default starting point at 180, 200, 220, 240, and 260. The conclusion that maximum yield occurs at 100 psi and about 220 degrees is obviously wrong. Of course, we could continue iterating one factor at a time, toward better conditions, but it would be painfully slow and expensive. Changing one-factor-at-a-time didn’t work here!

A better approach, using a designed experiment (DOE), with a factorial arrangement, for two factors, two levels. What does it tell us? It clearly tells us which way to move, for higher performance values.

Fig. 16.5   2-factor DOE

Fig. 16.6 Higher yield, with one additional data point.

 

Funditty #16

“How old would you be if you didn’t know how old you were?”
Satchell Paige

 

 

 

 

 

Bob Robertson

Bob Robertson

Bob Robertson is a retired professional quality engineer and educator with extensive experience in manufacturing environments throughout the world, including Singapore, Indonesia, Russia, and various locations throughout the United States. Besides all that, he Leslie Householder's admired and revered father, and she is pleased to spotlight his "Expat" stories here on her Rare Faith blog.