Transparency and traceability for AI-based defect detection in PCB production. - In: Modelling and development of intelligent systems, (2023), S. 54-72
Automatic Optical Inspection (AOI) is used to detect defects in PCB production and provide the end-user with a trustworthy PCB. AOI systems are enhanced by replacing the traditional heuristic algorithms with more advanced methods such as neural networks. However, they provide the operators with little or no information regarding the reasoning behind each decision. This paper explores the research gaps in prior PCB defect detection methods and replaces these complex methods with CNN networks. Next, it investigates five different Cam-based explainer methods on eight selected CNN architectures to evaluate the performance of each explainer. In this paper, instead of synthetic datasets, two industrial datasets are utilized to have a realistic research scenario. The results evaluated by the proposed performance metric demonstrate that independent of the dataset, the CNN architectures are interpretable using the same explainer methods. Additionally, the Faster Score-Cam method performs better than other methods used in this paper.
Analyse und Entwicklung einer Softwarearchitektur für die intelligente, optische Inspektion. - Ilmenau : Universitätsverlag Ilmenau, 2023. - 1 Online-Ressource (ix, 205 Seiten)
Technische Universität Ilmenau, Dissertation 2022
Die automatische optische Inspektion ist das wichtigste Werkzeug der Qualitätskontrolle in der modernen Elektronikfertigung. Durch die automatisierte Bildaufnahme und das Ausführen vordefinierter Bildverarbeitungsschritte haben diese Systeme die manuelle optische Inspektion weitestgehend verdrängt. Trotz des großen Maßes an Automatisierung sind menschliche Experten an vielen Schritten der Prüfung unverzichtbar und damit potenzielle Fehlerquellen. In den letzten Jahren wurden zahlreiche Ansätze untersucht, welche einzelne Aspekte der optischen Qualitätssicherung durch die Anwendung von Methoden der künstlichen Intelligenz deutlich verbessern. Für den Wandel der optischen Inspektion hin zu einer verlässlichen und voll autonomen Prüfung wird in dieser Arbeit ein Modell mit fünf Phasen vorgestellt, welches die Entwicklungsschritte auf diesem Weg abbildet. Das neue Modell unterscheidet sich von bisherigen Ansätzen durch einen ganzheitlichen Blick auf die Qualitätskontrolle und die Berücksichtigung aller Prozessschritte. Um die Umsetzung dieses Modells zu tragen, zeigt diese Arbeit ein neues Architekturmuster auf, welches Lösungen auf Basis von künstlicher Intelligenz trainieren und ausführen kann. Durch seine hohe Flexibilität kann die neue Architektur über unterschiedliche Auslieferungen auf einer heterogenen Menge an Systemen angewendet werden und viele unterschiedliche Anwendungen von künstlicher Intelligenz über das Feld der optischen Inspektion hinaus umsetzen. Für die allgemeingültige Beschreibung von KI-Lösungen basiert diese Architektur auf einer Menge an Objekten, welche in dieser Arbeit definiert werden. Eine Umsetzung dieser Architektur wird diskutiert und ihre Anwendbarkeit anhand von drei Experimenten bewiesen. Die Implementierung der beschriebenen Architektur ist unter einer OpenSource-Lizenz veröffentlicht.
Latency resistant safety in distributed remote laboratories. - In: Artificial Intelligence and Online Engineering, (2023), S. 112-124
This work will focus on the problems of creating a safe distributed laboratory. We explicitly will not discuss how to make individual elements of an experiment safe, as this is highly application-dependent. Instead, the goal is to find and evaluate different methods to detect and respond to fault conditions that an individual laboratory device might detect. Specifically, the methods should differentiate between user-based faults and those introduced through network communications. We develop a mathematical model to simulate distributed laboratories. We will introduce (time-dependent) network latency and jitter between all elements. Based on the model, a discrete event simulation is created. This simulation environment simulates three different fault detecting methods: the token method, the timestamp method, and the full-state-transfer method. We will compare detection ratios, bandwidth usage, and memory usage between the three methods based on the simulation.
Hybrid Take-Home Labs for the STEM education of the future. - In: Smart Education and e-Learning - Smart Pedagogy, (2022), S. 17-26
The acceptance of digitally supported teaching has increased strongly in recent years - and not only due to Corona. In the STEM subjects, online labs are increasingly being used to ensure that the requirements for availability, usability and granularity of the offerings are met. This ensures the connection of theoretically taught fundamentals and their application and deepening in the form of practical courses in the basic subjects. However, practical experimentation and the associated haptic learning is somewhat lost as a result. The Hybrid Take-Home Labs project aims to develop and test the basis for practical support of learning processes in STEM subjects, which allows students to conduct even complex virtual and remote-controlled laboratory experiments from home using their own resources, combined as needed for student-centered teaching to meet the requirements of future-oriented competence-based learning. It is one of nine projects supported by the Thuringian Ministry of Economics, Science and Digital Society and the German Stifterverband.
Virtual environment smart house for hybrid laboratory GOLDi. - In: Mobility for Smart Cities and Regional Development - Challenges for Higher Education, (2022), S. 250-257
The necessity to integrate virtual laboratories into the study process is becoming more significant, especially in pandemic time. Virtual based e-learning is seen as a reliable and effective support of teaching and learning process in different fields of study. The hybrid laboratory GOLDi uses the possibilities of remote and virtual experiments actively. At the same time, the implementation of the new experiment for teaching students in the area of Smart House systems will expand the functionality of the laboratory. The implementation of virtual experiments in the field of home automation systems provides an interactive learning environment that allows to engage students in an active educational process and increase their motivation to study modern information technologies and processes. The paper presents the results of the development of educational virtual environment for learning basics of Smart House systems development and control.
New ways for distributed remote web experiments. - In: Learning with Technologies and Technologies in Learning, (2022), S. 257-284
Remote Laboratories are widely used in the education of stem subjects. While the first generation of remote labs was based on individually developed local experiments with an integrated web interface, the next generation combined multiple experiments in a remote laboratory management system, making it possible to share whole experiments between institutions. At the same time, with cheap hardware available more and more experiments are conducted by the students at home. The sharing of experiments is already a step towards a more prosperous learning environment. The next step is to collaboratively develop and operate experiments by offering parts of experiments that are coupled over the internet to execute the whole experiment. This form of remotely coupled experiments allows for better collaboration between different institutions and also has benefits within a single institution. By remixing the components, the curriculum can be adapted to changing teaching scenarios, especially when considering that components from other institutions might be used. Also, extensive or expensive apparatuses can be hosted by the institution while more mobile parts are given to the students creating a hybrid take-home lab which is an improvement compared to an all virtual or all remote laboratory in terms of immersion.
Adaptable digital labs - motivation and vision of the CrossLab project. - In: 2022 IEEE German Education Conference (GeCon), (2022), insges. 6 S.
The flexibility and performance of digital laboratory elements such as remote labs, VR/AR or simulations summarized under the term cross-reality labs (CrossLabs), can be seen with the development in last and has been proven under the pandemic situation. Even though the potential of cross-reality labs is obvious referring to availability and flexibility for the students, these didactic solutions remain isolated at universities as well as for individual users. The implementations are mostly so rigid that the individual didactic objectives are not interchangeable between different universities and disciplines, hence there is a lack of interoperability. The CrossLab project seeks to design didactical, technical, and organizational solutions for open digital lab objects linking student-centered teaching and a cross-university learning environment. Of importance thereby is the fact that teaching is not adaptable to the digital laboratory, but the laboratories are adaptable to the requirements of the teaching-learning setting. The four project partners are working on a cross-type and cross-element mixture of diverse types of laboratories for cross-disciplinary use in a cross-universities settings. Thereby, the project leans on existing digital laboratories in various disciplines to create an open teaching and learning environment which can be adapted to the needs of students and to provide students with the skills necessary for future working scenarios.
On the development of a unified online laboratory framework. - In: Online engineering and society 4.0, (2022), S. 10-22
This work focuses on the requirements analysis of a modular framework to simplify developing and integrating new or existing remote experiments and laboratories. Besides this technical view, this paper also gives an organizational view on developing such a framework and managing the corresponding modules, which can also be developed by a third party. On a technical side, we provide the requirements to ultimately define the interface between different modules, enabling easy integration on different abstraction levels.The work's basis is the review of individual remote laboratories and existing systems of the past two decades and the author's collective experience. In the spirit of uniting the remote laboratory community, we will follow the IEEE 1876 standard wherever possible and extend it to make our vision of an easy to use, integrable, and extendable framework possible.
KOI: an architecture and framework for industrial and academic machine learning applications. - In: Modelling and development of intelligent systems, (2021), S. 113-128
Simulating the printed circuit board assembly process for image generation. - In: 2020 IEEE 44th Annual Computers, Software, and Applications Conference, (2020), S. 245-254
The inspection of printed circuit board assemblies gradually incorporates deep-learning-based classifiers. However, such classifiers require a vast dataset. To our knowledge, such a dataset is not available. This paper proposes a method to simulate the assembly process aiming at generating such a dataset. The simulation of the solder joint shape forming during reflow and the creation of a photorealistic rendering of the assembled board have the most significant impact on the visual appearance of the results. Therefore, this paper focuses on the simulation of these steps. The calculation of the solder joint shape requires minimizing the surface tension energy. For this, the algorithm discretizes the energy equations over a heightmap. The proposed software architecture for the simulation is highly extendable and facilitate future development. Experiments with the simulation of solder joints of a chip resistor show a remarkable similarity to real images from an automatic optical inspection machine.