Refine
Year of publication
Document Type
Way of publication
- Open Access (1)
Keywords
- Asset-to-Go (1)
- CAD (1)
- Machine Learning (1)
- Mensch-Roboter-Kollaboration (1)
- Mensch-Roboter-Kooperation (1)
- Multilevel switching (1)
- Programming circuit (1)
- RRAM (1)
- Sicherheitstechnik (1)
- Wandlungsfähigkeit (1)
Institute
As Convolutional Neural Network based models become reliable and efficient, two questions arise in relation to their applications for industrial purposes. The usefulness of these models in industrial environments and their implementation in these settings. This paper describes the autonomous generation of Region based CNN models trained on images from rendered CAD models and examines their applicability and performance for part handling application. The development of the automated synthetic data generation is detailed and two CNN models are trained with the aim to detect a car component and differentiate it against another similar looking part. The performance of these models is tested on real images and it was found that the proposed approach can be easily adopted for detecting a range of parts in arbitrary backgrounds. Moreover, the use of syntheic images for training CNNs automates the process of generating a detector.
Mensch-Roboter-Kollaboration im Spannungsfeld von Rekonfigurierbarkeit, Sicherheit und Effizienz
(2019)
Die heutige Automobilindustrie ist durch zunehmende Individualisierungswünsche der Kunden sowie einer Verkürzung der Konjunkturzyklen gekennzeichnet. Für Fahrzeughersteller, wie der Daimler AG äußert sich diese Marktsituation in einer hohen Variantenvielfalt und Stückzahlschwankungen. Da die Ausprägungen dieser Einflüsse weitgehend unbekannt sind, muss das Produktionssystem darauf vorbereitet sein, bedarfsgerecht der Marktsituation angepasst zu werden. Diese Anpassungen müssen schnell, mit wenig Aufwand und geringen Kosten möglich sein. Wandlungsfähige Produktionssysteme gehören deshalb zu den strategischen Erfolgsgrößen für die Automobilindustrie. Innerhalb dieses Beitrags wird ein modularer, skalierbarer Produktionsbaukasten vorgestellt, der Bestandteil einer Planungs- und Entwicklungsumgebung wandlungsfähiger roduktionskonzepte ist. Die hierfür genutzte grenzwerttheoretische Entwicklungsmethodik wird erläutert und die Skalierungsmöglichkeiten des Produktionsbaukastens mit Hilfe neuer Produktionstechnologien betrachtet. Dabei zeichnet sich die Mensch-Roboter-Kooperation als wesentlicher Bestandteil wandlungsfähiger Produktionskonzepte ab. „Smarte Produktionsassistenten“ sind intelligente und sichere Roboter-Applikationen, die sich als Möglichkeit hervortun die Mensch-Roboter-Kooperation als Schlüsseltechnologie in industriellen Anwendungen zu realisieren.
In this paper we present a novel method for performing in site real time quality inspection (QI) and consequently, digitalization of manual processes performed by human workers. It complements and improves our previous work in this area, which makes use of telemetry gathered from a smartwatch to classify manual actions as successful or unsuccessful. This new methodology provides the worker with a real time capable, robust and more accurate quality inspector. This work enhances the existing system through the elimination of input from the user by making use of a BIOX bracelet that detects gestures. The signal processing and classification methods are simplified and optimised by using assembled neural networks thus merging together the data gathered from multiple signal sources. Consequently, the overall QI system is improved with around 70%, thus furthering the necessary development needed to have a system ready to be used on a production environment.
The usage of convolutional neural networks has revolutionized data processing and its application in the industry during the last few years. Especially detection in images, a historically hard task to automate is now available on every smart phone. Nonetheless, this technology has not yet spread in the industry of car production, where lots of visual tests and quality checks are still performed manually. Even though the vision capabilities convolutional neural networks can give machines are already respectable, they still need well prepared training data that is costly and time-consuming to produce. This paper describes our effort to test and improve a system to automatically synthesize training images. This existing system renders computer aided design models into scenes and out of that produces realistic images and corresponding labels. Two new models, Single Shot Detector and RetinaNet are retrained under the use of distractors and then tested against each other. The better performing RetinaNet is then tested for performance under training with a variety of datasets from different domains in order to observe the models strength and weakness under domain shifts. These domains are real photographs, rendered models and images of objects cut and pasted into different backgrounds. The results show that the model trained with a mixture of all domains performs best.