An interesting thread:
>From: Deborah Strauss - agilesys.com
Calibration of Software used for testing purposes is a big issue in my company. Any comments would be appreciated:
how to best do it? how often? R & D software - does it have to be done? (I say yes)
Subject: Re: Q: Cal. of Software Used in Testing/Strauss/Pfrang
Date: Fri, 4 Sep 1998 13:48:12 -0600 From: ISO Standards Discussion
From: (Doug Pfrang)
Every article used on the production floor to manufacture product -- including machines, instruments, chemicals, software-controlled devices, etc. -- has to be validated for its intended purpose. Many articles are validated by brute-force trial-and-error: you experiment with it in your process until you get the process to work the way you want it to; i.e., your products turn out the way you want them to.
Some articles perform a measurement function. For these articles, you have the option of validating the article by having it calibrated; i.e., calibration is one technique of validating the article for its intended purpose. Calibration is not the ONLY technique of validating an article which performs a measurement function, but it is a popular one, because it is often more cost-effective than brute-force trial-and-error validation. Calibration is mainly used when you have a process, formula or recipe that you already know will work (i.e., it has already been validated elsewhere, such as in your R&D department), and you are trying to make sure that your production floor is correctly following that process, formula or recipe. For example, if your process, formula or recipe says to heat something to X degrees for Y minutes, then you want a reliable way of knowing you have heated it to X degrees for Y minutes.
This same rule applies to software-controlled articles used in testing. The first question you ask is, "What is the intended use of this article?" Probably, the intended use of the article is to distinguish good product from bad product. OK, fundamentally, you need to demonstrate that the article has the capability to do this. You have two choices. You can validate the article by brute-force trial-and-error, or you can validate the article by calibrating it.
Validating the article by trial-and-error can be done by collecting many examples of good product and many examples of bad product, running them all through your software-controlled article, and adjusting the article until it passes all the good product and rejects all the bad product.
Validating the article by calibration can be done in the usual way -- comparing it to a recognized standard.
Choose whichever validation technique works and is cheapest in your given application.
How often you revalidate the article (whether it is software-controlled or not) depends on its robustness. First you must determine how long it takes the article to drift so far out of tolerance that it no longer performs its intended use. (This information is often obtained from historical experience with the article.) Once you have this information, you establish your revalidation cycle so you are able to readjust the article before it drifts too far out of tolerance.
Thus, the key to validating software used for testing purposes is to recognize that you do not validate the software, per se; you validate the software-controlled article, and this can be done by brute-force trial-and-error or by calibration.
-- Doug Pfrang --
>From: Deborah Strauss - agilesys.com
Calibration of Software used for testing purposes is a big issue in my company. Any comments would be appreciated:
how to best do it? how often? R & D software - does it have to be done? (I say yes)
Subject: Re: Q: Cal. of Software Used in Testing/Strauss/Pfrang
Date: Fri, 4 Sep 1998 13:48:12 -0600 From: ISO Standards Discussion
From: (Doug Pfrang)
Every article used on the production floor to manufacture product -- including machines, instruments, chemicals, software-controlled devices, etc. -- has to be validated for its intended purpose. Many articles are validated by brute-force trial-and-error: you experiment with it in your process until you get the process to work the way you want it to; i.e., your products turn out the way you want them to.
Some articles perform a measurement function. For these articles, you have the option of validating the article by having it calibrated; i.e., calibration is one technique of validating the article for its intended purpose. Calibration is not the ONLY technique of validating an article which performs a measurement function, but it is a popular one, because it is often more cost-effective than brute-force trial-and-error validation. Calibration is mainly used when you have a process, formula or recipe that you already know will work (i.e., it has already been validated elsewhere, such as in your R&D department), and you are trying to make sure that your production floor is correctly following that process, formula or recipe. For example, if your process, formula or recipe says to heat something to X degrees for Y minutes, then you want a reliable way of knowing you have heated it to X degrees for Y minutes.
This same rule applies to software-controlled articles used in testing. The first question you ask is, "What is the intended use of this article?" Probably, the intended use of the article is to distinguish good product from bad product. OK, fundamentally, you need to demonstrate that the article has the capability to do this. You have two choices. You can validate the article by brute-force trial-and-error, or you can validate the article by calibrating it.
Validating the article by trial-and-error can be done by collecting many examples of good product and many examples of bad product, running them all through your software-controlled article, and adjusting the article until it passes all the good product and rejects all the bad product.
Validating the article by calibration can be done in the usual way -- comparing it to a recognized standard.
Choose whichever validation technique works and is cheapest in your given application.
How often you revalidate the article (whether it is software-controlled or not) depends on its robustness. First you must determine how long it takes the article to drift so far out of tolerance that it no longer performs its intended use. (This information is often obtained from historical experience with the article.) Once you have this information, you establish your revalidation cycle so you are able to readjust the article before it drifts too far out of tolerance.
Thus, the key to validating software used for testing purposes is to recognize that you do not validate the software, per se; you validate the software-controlled article, and this can be done by brute-force trial-and-error or by calibration.
-- Doug Pfrang --