Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our processes

[upliftpro]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

How can we adress the situation of rework? Lets say at the first process we have 100 pieces and two fail. The next process should start with 98, but because of our size we rework before the next process. How can we acount for 100 oppertunities for failure instead of 98?

 

[Bev D]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

again I'm not certain of what is troubling you with this situation?

simply use 100 in the denominator for the second operation.

calculate FPY for each station/operation as the number of units that passed the operation without requiring rework (regardless of where or when the rework is performed) divided by the number units processed at the station/operation.

 

[flyin01]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

again I'm not certain of what is troubling you with this situation?

simply use 100 in the denominator for the second operation.

calculate FPY for each station/operation as the number of units that passed the operation without requiring rework (regardless of where or when the rework is performed) divided by the number units processed at the station/operation.

upliftpro brings up some old headache of mine related to RTY calculations.

There are cases when RTY for an entire process (FTY step 1 * FTY step 2 ...* FTY step n) can be confusing. I used to try to interpret the metric from a probability point of view.

FTY for one station can be considered the probability of passing the station. That is rather intuitive. So if you start with 100 units and 98 pass FTY and probability to pass the station is 98%. For step 2, 98 units are tested and 95 pass on first try gives about 97% probability. Third step, 95 tested and 91 pass, 95.8%. In this case the RTY is also the probability for one unit to pass all three steps which is 91/100 or the product of the above three FTY´s.

However, I have worked in businesses where the FTY yield is multiplied to RTY and measured in discrete time intervalls of days/weeks just because it is convenient to follow up and compare over time. But the measurements where hard to interpret since the manufacturing site ran 24h a day the units may go into the first steps before midnight and into the last steps on the next day and also sometimes they went into storage for several days in the middle of the process.

This means that the number of tested units in each step may vary and sometimes fewer units where tested in early steps compared to later ones. Also it was not necessarily the same units that where compared when the FTY of each step´s where multiplied to the RTY. This made the RTY comparison to the probability of passing test impossible to do, at least for me. I remember people insisting on measuring the RTY even though the volumes of the test steps where totally inconsistent with each other and nobody could say for certain if the same unit in step 1 was also put into the last step on the same day :mg:

RTY can be such a mess. Anyone else have any interesting observations or experiences with strange RTY measurements?

 

[Bev D]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

This is a common occurence with RTY. You haven't yet said why this matters? How does the scenario you describe affect decisions that might be made? RTY IS probability. It is intended to not be a single snapshop but a trended time series to identify top problems.

I trend it on a weekly basis (using an I, MR chart which works pretty well)
then we do a pareto of top failure modes and their trends (usually a p chart or Xbar R type depending on the nature of the 'test') Using the correct control charts keeps us from nit picking on small fluctuations driven by sample size or other 'common' causes. we stay focused on solving the biggest problems.

We also have a pareto of each station/operation by FTY, with FTY trend charts for each station.

Of course we balance this with Cycle time data and cost of poor quality data (in the same trend - pareto - trend fashion).

We launch projects based on largest impact (given quality, cost and cycle time). We have made substantial gains in all 3 fronts over the five years adn no one has ever 'complained' or been 'fooled' by teh types of normal variation you describe. So I'm curious as to why this causes you concern??

 

[Bev D]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

perhaps the problem is that you are trying to 'reconcile' the failures to a simple fraction. Given the scenarios of different numbers of units at each station and the fact that a unit may start at station 1 on day 1 but not complete until several days later RTY is NOT SUPPOSED to be calculated as # total failrues / # units tested. It is intended specifically as a conditional probability. FPY1*FPY2*FPY3...*FPYi

It is related to the Poisson distribution and the fact that one unit may fail at several stations and another unit may never fail. IN fact a companion metric that many companies use is the %of 'no fail' units. the math is logistically a little difficult as you can only calculate the values once teh supbgrouped units (day, week or month of START) have ALL finished (either scrapped or shipped...

The key is to remember that there is really no value in the absolute number (that just tells you how awful things are). The value comes from the relative differences and changes that point you to the top failrues and show over time whether you are getting bettter or worse.

for example it didn't matter to me that the weather man said my town was 5 degrees, my LL Bean thermometer said it was 3.6 and my car said it was 7. It was COLD and I wore a coat and gloves and long pants. I didn't wear a skirt and blouse.

 

[flyin01]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

This is a common occurence with RTY. You haven't yet said why this matters? How does the scenario you describe affect decisions that might be made? RTY IS probability. It is intended to not be a single snapshop but a trended time series to identify top problems.

I trend it on a weekly basis (using an I, MR chart which works pretty well)
then we do a pareto of top failure modes and their trends (usually a p chart or Xbar R type depending on the nature of the 'test') Using the correct control charts keeps us from nit picking on small fluctuations driven by sample size or other 'common' causes. we stay focused on solving the biggest problems.

We also have a pareto of each station/operation by FTY, with FTY trend charts for each station.

Of course we balance this with Cycle time data and cost of poor quality data (in the same trend - pareto - trend fashion).

We launch projects based on largest impact (given quality, cost and cycle time). We have made substantial gains in all 3 fronts over the five years adn no one has ever 'complained' or been 'fooled' by teh types of normal variation you describe. So I'm curious as to why this causes you concern??

Your post was very informative. The problems that I have seen with RTY is actually more related to people interpreting it in the wrong context rather than understanding the logic behind the measurement. Since RTY is not as straightforward as only measuring FTY per station and using all the tools & charts that you mentioned.

I have seen problems causing confusion in companies that for example work project based. Different variants of similar products are developed in projects but the manufacturing process is pretty much the same. This makes decision making people - rescource owners - to compare measurements and performances between projects. Trying to put a common benchmark level for what RTY should and should not be. This leads to cross product targets that are expected to be met by all projects. When targets are set by someone who fails to grasp the meaning of RTY and the rationale behind the measurement. They may not understand how the measure can vary given certain circumstances in production. This leads to confusion in many cases.

FTY on one process step is allways easier to understand. But the product of several FTY´s like the RTY is not so easy to understand the dynamics behind it, especially if you are only interested in the "bottom line"-information of many things.

For example you can have great FTY performanace in say 10 out of 10 process steps this will give you a RTY of 90%. Then you have another RTY that consists of same 10 process steps but where 9 are perfect but one is for some reason really bad. It is very easy to think that there is a 1-1 relation ship with RTY-1 and RTY-2. They seem to be the exact same one figure summary of the total process, which of course is due to different causes. The difference can of course be explained in layman´s terms to anyone, but not all people listen to the experts.

In such I cases would prefer to stay away from RTY and keep the stations separated only observing the FTY of each station.

 

[flyin01]

Re: Normalized vs. Rolled Throughput Yield - First Pass Yield metrics for our process

perhaps the problem is that you are trying to 'reconcile' the failures to a simple fraction. Given the scenarios of different numbers of units at each station and the fact that a unit may start at station 1 on day 1 but not complete until several days later RTY is NOT SUPPOSED to be calculated as # total failrues / # units tested. It is intended specifically as a conditional probability. FPY1*FPY2*FPY3...*FPYi

It is related to the Poisson distribution and the fact that one unit may fail at several stations and another unit may never fail. IN fact a companion metric that many companies use is the %of 'no fail' units. the math is logistically a little difficult as you can only calculate the values once teh supbgrouped units (day, week or month of START) have ALL finished (either scrapped or shipped...

The key is to remember that there is really no value in the absolute number (that just tells you how awful things are). The value comes from the relative differences and changes that point you to the top failrues and show over time whether you are getting bettter or worse.

for example it didn't matter to me that the weather man said my town was 5 degrees, my LL Bean thermometer said it was 3.6 and my car said it was 7. It was COLD and I wore a coat and gloves and long pants. I didn't wear a skirt and blouse.

I must say that you provide only well written posts that provide valuable insights I am glad I joined this forum because now I know I am in the company of someone who knows what they are talking about!

I am no statistician, but I know my basic probability theory. I fully agree with you about a certain RTY value and that what matters is the change instead of a number. Unfortunately several former co-workers of mine have failed to grasp this simple issue.

I like to keep my measruements simple but never one dimensional. I have three indicators on my scoreboard in production.
- FTY - per station and RTY
- Cycle times. Median + min/max for outliers and also the distribution of it. If the tail is thick/small etc.
- Fault rate. This one is lagged since it consists of customer claims and it is too late to do any improvements here. But I always follow up this one with the other two.

By combining these I minimize drawing wrong conclusions on a RTY figure that may be inflationary for some reason or give the wrong impression of what is going on.