Will computer vision systems and neural networks help reduce scrap and defects?
Material prepared by: Scientific Director of the AQT Center Sergey P. Grigoryev .
We can see an increase in promotional materials in the field of machine vision application in manufacturing plants for quality control.
"The computer vision camera receives an image from the surface... The resulting image is processed to identify defect areas, and is further analyzed by a neural network to classify defects. The results of the analysis are displayed on the operator's workstation, with integration into the process control system. Identified defects are classified online into more than one 30 parameters".
Of course, machine vision systems will help identify defects in the flow of products or semi-finished products, and production management and developers will be forced to pay due attention operational definitions , but without the use of knowledge in the field of statistical process control, such systems will force the operator to perform more errors of the first and second kind , only worsening the situation.
Such machine vision systems will not be able to tell the operator what he needs to do to improve the output of the process, and even more so will not take responsibility for the results of the decisions of the operator who used “defect analysis displayed on the operators’ workstation.”
"These are all attempts to cut corners on the path to quality. There are no shortcuts here."
See Donald Wheeler's article for an explanation: Correct and incorrect ways to use tolerance fields .
See an example of operator intervention in the gas flow control process at an enterprise producing biogenic methane in the article: The concept of variability in process control .