Desensitizing to reduce variation - A Taguchi technique?


Steven Sulkin

Desensitizing to reduce variation

Any examples of how process desensitization has been used to reduce variation? Desensitization is the technique of making a process more robust to input variation. You may have heard it called something else? I believe this is a Taguchi technique. Any examples, even made up would be helpful.

Thanks in advance,


Don Winton

I had not heard of 'desensitization,' but I think I know what you may be talking about.

Controlling input variables (temperature, cycles times, feed rates, etc) generates a great deal of cost (less than sorting good from bad, however). But many companies fail to identify key inputs from the less critical ones. One particular case I know of was in chemical deposition. The bath had to start at a certain temperature, ramp up to another temperature within a given amount of time and end at another temperature. Typically, this ramp would look a little like a backwards 'Z' stretched vertically slightly.

According to theory, the start temperature had to be 'X' +/- 'A', the ramp had to be 'Y' +/- 'B' and the end temperature had to be 'Z' +/- 'C.' Any depositions that failed to meet this criteria were considered bad, but experimental test results of the finished product had failed to correlate this. So, basically they were controlling the process exactly as they wanted, but could not correlate it with results.

After reviewing the information, I asked to see the validation results from which they arrived at the process control variables. After review, they had verified the process with yield data (percentage passed) rather than variable data (how well they passed using LSL tolerances).

I designed an experiment and asked them to run it for me. Using the results of the experiment and the variable data (rather than attribute that was used before), I found that the only key input was the start temperature. Using ANOVA, it was determined that, in order of key, start temperature was first, ramp was second and end temperature was third (end temperature contributed nothing to the variable data). Based on these results, I requested they tighten the tolerance on start temperature, leave ramp where it was and remove the end temperature tolerance entirely. After some additional runs, tolerances were adjusted slightly, then made standard practice.

The results were they could use finished depositions for continued processing without reducing the quality of product (increase yield, decrease costs). It should also be noted that the finished product yield also increased (decrease costs).

By removing the tolerance on end temperature, control methods for this input could be removed (decrease costs).

Thus it was revealed that by allowing greater variation in end temperature, there was a decrease variation in quality of end product. Is this what you were asking?



Fully vaccinated are you?
Anyone heard anything about this recently? Or have anything to add?

Atul Khandekar

I have on my company site, one case study you may find useful:
Edit by Atul(May 28,2010): Invalid URL deleted. The case study has now been removed! sorry!!
Last edited by a moderator:
Top Bottom