Cleanroom Control Limits: Cumulative or individual?

[BioBurden]

Hi everyone.

I'm looking for some quick help with control limits for the environmental monitoring programme for a Class VII cleanroom.

We have set our limits, based on historical data to +2 Standard Deviations for Alert and +3 Standard Deviations for action.

Now my question is this:

Should the limits be constantly changing every month or should I take the calculated limits after collecting the minimum of 12 months data?



I am using Excel and my own graphing, formulas, etc.

So should the limit change every month like the attachment below: EM Graph Capture.jpg

or should it be consistent, like EM Graph Capture2.jpg?

 

[bobdoering]

What kind if variation do you expect to see? Is it random? What is the reaction plan? From the charts you showed, there was no indication of process intervention. Is that true?

Have you done a CNX evaluation of your clean room process? What is the variable you adjust? Changing filters? Are you tracking that on your chart?

 

[Steve Prevette]

A set of control limits is innocent until proven guilty. Do not recalculate control limits unless you have at least a statistically significant shift in the data and know the reason that the shift occurred.

- on vacation in Switzerland.

 

[BioBurden]

These are counts for microbial growth. CFUs or yeasts/moulds per plate.
Eudralex limits are not strict enough so we are determining our own based on historical data.

The limits should be static then rather than constantly changing to reflect the data?

 

[bobdoering]

Do the counts increase until action is taken? Or do they randomly change on their own? When is any action taken?

If it was a particulate clean room, you would think there would be a slow increase (or maybe a dramatic one) until the signal indicates the filter needs changed. Then it would drop to a low level again. In that case you would have a sawtooth curve, and the charting is easy to pick out.

Just have not heard that your variation follows that - or any other predictable variation. Of course, there is a possibility that your true variation may be masked by measurement error. How repeatable are your measurements?

 

[Steve Prevette]

These are counts for microbial growth. CFUs or yeasts/moulds per plate.
Eudralex limits are not strict enough so we are determining our own based on historical data.

The limits should be static then rather than constantly changing to reflect the data?

Yes, static. As long as they predict future performance accurately, don't mess with them. You can inadvertently convince yourself a change is occurring when it really isn't if you turn the control limits into a version of moving averages.

Now, I should ask "what are the purpose of your control limits"? To determine if something is changing? If so, you need something fixed to compare it to.

 

[BioBurden]

The filter is changed annually and the room is wiped down daily and weekly.

Daily clean is performed with IPA. Weekly clean performed with Proceine/Qceine rotating sanitising agents.

We use air sampling (8 locations, TSA and SDA, 1000 litres), settle plates (8 locations, TSA and SDA) and contact plates (13 locations, TSA + L + Tw).

The analysis is performed by a third party, rather than on site.

We have had samples sent to 2 separate labs over a 3 month period for comparison and results have been consistent.

We have only ever had 1 set of results that exceeded the Eudralex limit (No weekly clean had been performed for 2 weeks)
Retested after a clean and all results were back in spec.

I have been using normal distribution with the data we have so far as Poisson distribution is far too strict. My attempts to form and identify histograms have been futile so far, as I am not a statistics expert.

 

[Bev D]

Steve's advice is spot on: control limits are not changed unless there is a deliberate improvement known to have been taken to improve performance.

Particle counts should follow a roughly Poisson distribution but may fit the I, MR chart better if counts are high enough.

I would recommend NOT using the 2sigma limits as warning limits - this is not just a mis-use of control limits, it provides no real added value to interpreting the results.

Can you post some of your data? Some of us here have experience with your exact situation with clean rooms and may be able to help you determine the best chart choice...

 

[BioBurden]

Apologies for the delay, It was a public holiday here.
Here are some of the examples

SDA Air Plate 1 (Yeasts/moulds)

JUN 2
JUL A 1
JUL B 1
AUG 22
SEP 0
OCT 0
NOV 0
DEC 0
JAN N/A
FEB 0
MAR 0
APR 0


TSA Air Plate 1 (CFUs)
JUN 2
JUL A 135
JUL B 0
AUG 140
SEP 2
OCT 1
NOV 2
DEC 1
JAN N/A
FEB 0
MAR 0
APR 0


As you can see there were spikes in the above results. This is when the daily and weekly cleans were not performed as scheduled, or incorrectly.
No results were obtained in January due to an issue with the equipment supplier.

It is a new facility so we only began to test in June. So I was working under the impression that the limits change continuously for the first 12 months to take the data into account, and after that, static limits would be applied. We have found that the poisson distribution (95th and 99th quartiles) was too strict for our cleanroom, Class VII, when low results were encountered. And obviously statistics cannot be applied when all results are 0.

I appreciate your input and defer to experts in this matter. I have no Statistics qualification and this is my first experience with setting the action and alert limits.

The company were just using eudralex limits and audits by some of our big multinational customers have made the observation that action and alert limits need to be applied.

 

[Steve Prevette]

Yes, except for the three obvious anomalies, the data are relatively low.

What does Jul A and Jul B mean?

Eyeballing it, the first shows a decreasing trend as the most recent 7 points are zero, so that will be seven in a row below average. The second is likely stable (again, once you remove the three outliers).

Sounds like you should be monitoring the performance of the cleaning cycles!