Planning for Design of Experiments: A Case Study
Preparing before experimentation saves time, money and frustration
by Dipak Thakur
I learned the value of planning when my company reached a plateau regarding customer satisfaction. The company, an electronics manufacturing services provider in India, achieved outstanding results using Deming's plan, do, check, act model. Our customers, however, demand continuous quality improvements (CQI), so when satisfaction ratings remained static we had to boost our CQI efforts.
We decided to strengthen several processes by implementing design of experiments (DOE)--a quality tool the company hadn't used before. During the experimentation process we discovered that had we benchmarked another organization or read a DOE case study, we could have avoided a lot of trouble from the start.
DOE planning begins
I led a cross functional team that was to prepare systems and staff for the company's DOE endeavors. Although the team had studied process steps and prepared experimental design layouts, members still had a few doubts about the process; therefore, we requested a DOE consultant. The services of an expert were secured, and the company flew him from New Delhi, India, to spend two days with us.
The team had planned to begin with the company's wave soldering process and intended to experiment with particular factors: flux density, preheater temperatures, solder bath temperature and conveyor speed. The team decided to exclude conveyor gradient. After the consultant arrived, however, he disagreed with the exclusion.
He explained that unless experiment results rendered a factor insignificant, it should not be left out. All factors need to be considered to derive maximum benefit from the experimentation.
As a result, the team tried to alter the conveyor gradient in the soldering machine. Unfortunately, we didn't know how, the ma- chine's manual was no help and we wouldn't be able to contact the manufacturer until the evening.
By the time we decided to eliminate the gradient factor from the experiment and finalized the experimental layout, we were two hours behind schedule. Had plans been shared with the consultant before his arrival, we would have included the conveyor gradient factor.
Existing soldering machine settings served as the first trial run combination. The second trial run involved randomly altering the machine's parameter settings. To our horror, the machine took 30 minutes to reach the desired temperature, and reaching subsequent trial run combinations took even more time. Because our time estimates were inaccurate, the soldering experiment needed to be continued the next day, consuming two more hours of the consultant's availability.
Time runs out
After the best possible combination of factors and their levels were obtained, we ran the new combination. Not impressed with the results, the consultant re- designed the experiment by reducing the variation of levels. Unfortunately, the machine could not control the factors of the newly reduced levels, and this important optimization stage of the experiment was called off--much to the discomfort of the consultant.
Next, the consultant felt it was necessary to change the preheating temperature from 70° to 100° C to get better soldering results.
The machine operator informed us that to increase the temperature, the preheater elements needed to cool and then be replaced with elements of a higher wattage. This would take two hours, however, and by that time, the expert would be gone.
If the team had determined the machine's capabilities beforehand, the time requirement could have been planned or the machine upgraded. But this wasn't the case, and the consultant needed to leave before the wave soldering experiment was completed.
We learned six lessons regarding DOE planning:
1. Do not exclude process factors from the experiment unless they are proven logically insignificant experimentally.
2. Be sure that experts on each piece of equipment being used during the experiments are readily available.
3. If using a consultant, discuss DOE plans with him or her before making estimates for the experimentation.
4. Because time estimates can go wildly wrong, allocate a sufficient time-factor cushion.
5. Ascertain machine controls and capabilities needed by experiments, and upgrade them if necessary.
6. Benchmark another company's attempt at a similar process.
Currently, our DOE endeavors are a success. And we came away learning one lesson in particular: Planning is everything.
DIPAK THAKUR is a quality manager at Deltron Ltd., in Chandigarh, India. He earned his degree in electronics engineering from the University of Poona in India. Thakur is an ASQ member.