Industrial Automation branch INA
of Fraunhofer IOSB
Industrial Automation branch INA
of Fraunhofer IOSB

Evaluation of self-learning optical processes for the quality control of plastic products

Inspection detects the slightest signs of defects

Friedrichs & Rath GmbH (F&R) specializes in precision parts made from thermoplastics and manufactures a large proportion of its products for the automotive industry. In order to meet the high quality requirements of customers in the automotive industry in particular while producing high quantities, F&R would like to introduce a learning, precisely automated optical quality control system. The inspection should be able to reliably detect even the slightest signs of defects and, if possible, classify and categorize them. The aim is an inspection application that can be adapted to different types of plastic parts with little effort, even by non-specialist employees.

Procedure in the project

As part of this first evaluation study of learning image processing methods, Fraunhofer IOSB-INA initially recorded the requirements of F&R with regard to defect types and representative inspection sample types. In contrast to classic parameter-driven image processing, learning methods are based on sample images and corresponding classifications (e.g. "OK", "NOK").

 

It therefore does not require any expert knowledge in the field of image processing. For this reason, test samples of six different types of plastic parts were provided for the evaluation. These were classified in advance by F&R as either "OK" or with a defect or as "NOK". Fraunhofer IOSB-INA first photographed these test samples using a generic camera and exposure setup. As each type of plastic part has its own shape and potential defects occur at part type-specific locations, an individual frame was created for each part type using additive manufacturing. The images created were used to test the classification and anomaly detection approaches of learning processes. Classification attempts to recognize learned patterns. Anomaly detection attempts to identify deviations from a learned reference image. For the evaluation, 70% of the images were used for training, 15% of the images for automated validation/review and a further 15% for the final test.

Results of the project

OK and NOK images are taken into account for training and validation during classification. In contrast, only OK images are used for anomaly detection. The test is also performed with OK and NOK images for classification. In the case of anomaly detection, all NOK images and 15% of the OK images were tested. The results of the test were very promising. Detection rates of 96-100% were achieved with anomaly detection and 100% with classification for all part types. These high detection rates required a minimum number of 30 test samples per part type and a systematic classification of the images. Due to the good results, F&R is planning to design and create a generic testing machine that includes the evaluated procedure. This testing machine should be able to test different plastic parts and thus enable 100% testing for production lines, which was previously not economically feasible.