Hybrid Distributed Cascade Convolutional Neural Networks Model for Riveting Processes

The field of artificial intelligence is experiencing a great interest in manufacturing companies for the inspection of parts and verification of images. The current trend in the sector is to implement these novel methodologies in industrial environments so that they can benefit from their advantages over traditional systems. Quality and production managers are increasingly interested in replacing the classic inspection methods with this new approach due to its flexibility and precision. Traditional methods have some weaknesses when it comes to inspecting parts, such as sensitivity to disturbances. In an industrial environment these disturbances can be changes in lighting during the day or during the year, the appearance of external elements such as dust or dirt. The use of new convolutional neural network techniques allows training including disturbance scenarios, teaching the artificial neural network to detect non-verse defects influenced by changes in light or by the appearance of dust. In this way, it is possible to drastically reduce false positives, avoiding costly stops in production and maximizing the precision of detection and classification of each defect. This work studies the implementation of a hybrid model based on a cascade detection neural network with a classification neural network in an industrial environment.