Custom algorithms were built because pattern matching was unreliable for the conditions mentioned above. For example, one algorithm was designed to detect reversal of a spring. Every time a new algorithm was created or an existing one was altered, the supplier had to validate it over hundreds of brake assembly images to ensure that it was reliable over the entire sample set. National Instruments Vision Builder, with its batch processing capabilities, accomplished this task.
To be certain that the system worked well, an adaptive technique was used where certain parameters in the algorithm, such as threshold value, were iteratively changed until the appropriate feature was found. If at the end of the iteration the feature was not found, it was concluded that it was not present. Based on the images from the sample set, the supplier was able to identify a range that could be used for these iterations, thereby reducing the processing time required. Threshold values were changed iteratively from 50 to 100 in steps of five until a single particle of a certain area remained.
Adapting to Variations in Components
A fully automated brake inspection system was built using virtual instrumentation and machine vision technology during a period of 10 weeks. Custom-built, self-adapting algorithms were used to ensure reliable inspection irrespective of the wide variations in the components. The system also gives customers complete flexibility in choosing which features are inspected and the inspection criteria, so it is very easy to change from one model to another.
Whereas defect analysis was not possible with other configurable machine vision systems, with the new system, customers can use statistical details on various defects to curb problems in raw materials and upstream processes at the source. Finally, the system makes customized reports accessible over the network.
For more information on National Instruments’ machine vision hardware and software, visit http://info.hotims.com/15132-153.