Refine
Year of publication
Document Type
- Article (37)
- Conference Proceeding (20)
- Part of a Book (4)
- Other (2)
- Doctoral Thesis (1)
- Report (1)
Keywords
Alternating Transfer Functions to Prevent Overfitting in Non-Linear Regression with Neural Networks
(2023)
In nonlinear regression with machine learning methods, neural networks (NNs)
are ideally suited due to their universal approximation property, which states
that arbitrary nonlinear functions can thereby be approximated arbitrarily well.
Unfortunately, this property also poses the problem that data points with
measurement errors can be approximated too well and unknown parameter
subspaces in the estimation can deviate far from the actual value (so-called
overfitting). Various developed methods aim to reduce overfitting through
modifications in several areas of the training. In this work, we pursue the
question of how an NN behaves in training with respect to overfitting when
linear and nonlinear transfer functions (TF) are alternated in different hidden
layers (HL). The presented approach is applied to a generated dataset and
contrasted to established methods from the literature, both individually and in
combination. Comparable results are obtained, whereby the common use of
purely nonlinear transfer functions proves to be not recommended generally.
Although the increasing use of automation in industry, manual assembly stations are still common and, in some situations, even inevitable. Current practice in manual assembly lines is to balance them using the takt-time of each workstation and harmonize it. However, this approach mostly does not include ergonomic aspects and thus it may lead to workforce musculoskeletal disorders, extended leaves, and demotivation. This paper presents a holistic human-centric optimization method for line balancing using a novel indicator ̶ the ErgoTakt. ErgoTakt improves the legacy takt-time and helps to find an optimum between the ergonomic evaluation of an assembly station and its balance in time. The authors used a custom version of the ErgoSentinel Software and a Microsoft Kinect depth camera to perform online and real-time ergonomic assessment. An optimization algorithm is developed to find the best-fitting solution by minimizing a function of the ergonomic RULA-value and the cycle time of each assembly workstation with respect to the worker's ability. The paper presents the concept, the system-setup and preliminary evaluation of an assembly scenario. The results demonstrate that the new approach is feasible and able to optimize an entire manual assembly process chain in terms of both, economic aspects of a well-balanced production line as well as the ergonomic issue of long term human healthy work.
Currently, many sources predict increasing use of AR technology in the industrial environment. The task of immersive productive assistance systems is to provide information contextually to the industrial user. Therefore, it is essential to explore the factors and effects that influence the visibility and the corresponding quality of this information. Caused by the technical limitations of additive display technology and application conditions, this new approach has evaluated the immersive visibility of Landolt Rings in various greyscales against ambient illuminance levels on different industrial-like surfaces, coupled with and without a white virtual background. For this purpose, an empirical study in a within-subjects-design with full factorial experimental design (n=23) was conducted on Microsoft HoloLens 2 hardware. The mean values of the main effects indicate that visibility is significantly affected by ambient illuminance (best results at lower level), greyscale (best results at middle level) and virtual background (best results with background). In contrast, the choice of surface is shown to have no statistically significant effect on visibility, however it affects the response time. Additionally, cross-interactions of variables were analyzed and lead to a design recommendation for immersive industrial applications.
Companies are confronted with increasingly demanding environments, including globalization, technologization, intergenerationality, and crises such as the coronavirus pandemic. To accept uncertainties as a challenge and to recognize opportunities for development, well-educated and resilient founders are needed who can foster innovation and sustainable development within society and the economy. The majority of today’s entrepreneurs have an academic background. Hence, institutions for higher education need to provide comprehensive educational offerings and support initiatives to train and sensitize future entrepreneurs. Therefore, since 2013, agile teaching formats have been developed in our project at a Bavarian university of applied sciences. In two stages, we founded a limited company for hands-on experimentation with entrepreneurship and also conceptualized an elective course and an annual founders’ night. Based on a theoretical model and continuous teaching evaluations, we adjusted the individual modules to suit the target group. The objective is to promote the acquisition of key competencies and exert a positive influence on the startup quotient in the region. There are six startups by students who can be traced back to our project. This indicates that a target-group-oriented educational program encourages motivation and awareness of entrepreneurial thinking and action among students.
Highly autonomous production cells are a crucial part of manufacturing systems in industry 4.0 and can contribute to a sustainable value-adding process. To realize a high degree of autonomy in production cells with an industrial robot and a machine tool, an experimental approach was carried out to deal with numerous challenges on various automation levels. One crucial aspect is the scheduling problem of tasks for each resource (machine tool, tools, robot, AGV) depending on various data needed for a job-shop scheduling algorithm. The findings show that the necessary data has to be derived from different automation levels in a company: horizontally from ERP to shop-floor, vertically from the order handling department to the maintenance department. Utilizing that data, the contribution provides a cascaded scheduling approach for machine tool jobs as well as CNC and robot tasks for highly autonomous production cells supplied by AGVs.
In the last decade many different additive manufacturing (AM) technologies for metal, plastic or ceramic processing raise from research to commercialization. As a result, AM grows into different business areas and transforms structures and processes. Hence, the contribution tends to show the change in added values though the availability of different additive manufacturing technologies based on a technology screening and market research. Regarding the named purpose, a broad market research of 83 companies and 339 printer models has been conducted to find patterns of AM technology market share and regions to structure indicators such as accuracy by processed material classes with a specified AM method. Printing materials as metal, plastic, ceramic and carbon have been considered. The categorization is done by the AM principles: power bed fusion, material extrusion, vat photopolymerization and …