I was wondering if someone could give me a little direction with the requirements around Vision System data (in a GMP environment)?
We have a Vision System that is used to inspect and verify length and diameter of drug implants. The system works on a pass / fail basis, with automation incorporated to accept passing implants and reject failing implants. The Vision application (COTS) is installed on a B&R PC, with a Windows OS. The application is linked to a networked database which can be configured to save each inspection result. For the purposes of batch documentation a report is generated by the system at the end of each batch with a summary of accepted and rejected implants, also containing a min and max length and diameter result. This report is printed in PDF, is non-editable and is signed and attached to the batch file.
My questions are as follows:
Are the inspection results considered to be electronic records?
Are the reports considered to be electronic records?
What needs to be retained, the inspection results, the reports or neither since the PDF printout is signed and retained as part of the batch paperwork?
Any help or clarification on the above system would be greatly appreciated!!
Each vision is tuned for “confidence” or accuracy. A default value the software will typically have is 65% when installed, but should be tuned during the qualification stage. Example. The vision system is looking at an image and trying to determine what it is within the confidence parameter. This means that when the machine is trying to determine is the number (or letter, bar…) it sees is a particular number (per attributes), if it determines is “sees” that number within the parameter, it will register it as such, if it does not reach the confidence level, it will move to the next value…and repeats until it get a hit or nothing at all.
So when tuning the system to set the most appropriate value, you can run multiple samples past it with a known value and see (and build) a matrix of what the highest value should be set. Since a “4” and a “9” share many attributes, you want to be able to tune the 4 value in the system (confidence level) to accept the “4”, but reject the “9”, which would have a value close. Same thing for every other values. A “1” and a “7”, or a “7” and a “9”. You tune each value to the highest confidence level to accept what you want and reject everything else.
It’s easiest to create a matrix with the expected value and text all the number (or others) against it. The control system will give level of confidence (%value) that it will assign when comparing the value to what the expected is. You can then tune that parameter and do your testing to show that all other values are rejected.
Same thing holds true for the any measurement. Specify what you are looking for, compare it to values that are known not to meet it and get the data to support modifying the confidence level in the system.
For most vision systems, it is a quality gate, and any reports from the system are used only by engineering to maintain (re-tuned when necessary) or by Quality Engineering to evalute metrics for OE functions.
The system usually does not generate an e-record, since it can be considered (or argued) that is it qualified in the same way many instruments are. The goal of the vision system (CQA) is to ensure that only items that meet the specification pass through the system. The vision system is usually attached to a mechanical/pneumatic gate control to expel the components that do not meet the specification.
Configuration Management, however, should be a requirement. You want to be sure that you are aware of the system’s ability to evalute good product from bad. Should the data (from the vision system or another PLC) provide a metric that allows you to tune the system better, you
would use that to “improve” it’s capability.
So this would not be fall until PT 11 in the e-record world, if it is defined in the manner I reference here.
It is equivalent to a Go/No Go switch, maintained, evaluated and adjusted as necessary.