Background
Manufacturing, in all its simplicity, is making products by subjecting materials to certain operating conditions. A cookie is made by baking a particular mix of ingredients. The quality of a cookie is based on how much of what materials are processed under what conditions. Manufacturing companies make products and test them from time-to-time in laboratories to determine if the key product characteristics are within customer specifications.
Running Blind While Avoiding Expensive Tests
Since product quality is being evaluated only occasionally, they really don't know how good a product is between tests. They can only assume it is similar to the last test performed, that is if materials and conditions have not changed much... an assumption that is not particularly valid unless the materials and process are perfectly consistent (not likely). The manufacturer, due to expense, and sometimes because the testing is destructive, would prefer to test less often, but that would mean running "blind" longer, more of the time. They have to balance risk with cost.
Real-Time Virtual Sensors
Since data on materials and operating conditions are available from process and materials control systems, and since we have a history of lab test results, we can build and validate a mathematical model that relates materials and conditions to product characteristics. This is a virtual sensor. It is "virtual" because no physical quality sensor exists, but instead we have an estimate of product performance. Process and materials data are quite often available in "real-time" (or we can make them so), we can use that model to estimate quality in real-time, all the time not just from time to time like a lab test.
Benefits
No more running blind. We can now see, understand and act in real-time to quality. With real-time virtual sensors, operators and engineers can immediately observe the effects of making process or materials changes. If they "tweek" (adjust) the oven temperature, as the oven heats or cools, estimated cookie characteristics are seen dynamically. On the other hand, if the real-time estimated cookie characteristics are seen to be drifting off target, the operator can make adjustments to the process or materials to bring them back. An added bonus: if the virtual sensor is deemed suitably accurate, validated by further lab tests, we can reduce those lab tests to merely audit the virtual sensor's performance.
So, with a bit of software, we can enable you to see performance in real-time, make adjustments to keep your product on-target, in-spec and reduce lab testing too.
Bravo!
Later, we can talk about using the virtual sensor models to "close the loop" on quality automatically.
Blog Move...
11 years ago