Universität Stuttgart
Permanent URI for this communityhttps://elib.uni-stuttgart.de/handle/11682/1
Browse
10 results
Search Results
Item Open Access Are you sure? : prediction revision in automated decision‐making(2020) Burkart, Nadia; Robert, Sebastian; Huber, Marco F.With the rapid improvements in machine learning and deep learning, decisions made by automated decision support systems (DSS) will increase. Besides the accuracy of predictions, their explainability becomes more important. The algorithms can construct complex mathematical prediction models. This causes insecurity to the predictions. The insecurity rises the need for equipping the algorithms with explanations. To examine how users trust automated DSS, an experiment was conducted. Our research aim is to examine how participants supported by an DSS revise their initial prediction by four varying approaches (treatments) in a between‐subject design study. The four treatments differ in the degree of explainability to understand the predictions of the system. First we used an interpretable regression model, second a Random Forest (considered to be a black box [BB]), third the BB with a local explanation and last the BB with a global explanation. We noticed that all participants improved their predictions after receiving an advice whether it was a complete BB or an BB with an explanation. The major finding was that interpretable models were not incorporated more in the decision process than BB models or BB models with explanations.Item Open Access A reinforcement learning approach to view planning for automated inspection tasks(2021) Landgraf, Christian; Meese, Bernd; Pabst, Michael; Martius, Georg; Huber, Marco F.Manual inspection of workpieces in highly flexible production facilities with small lot sizes is costly and less reliable compared to automated inspection systems. Reinforcement Learning (RL) offers promising, intelligent solutions for robotic inspection and manufacturing tasks. This paper presents an RL-based approach to determine a high-quality set of sensor view poses for arbitrary workpieces based on their 3D computer-aided design (CAD). The framework extends available open-source libraries and provides an interface to the Robot Operating System (ROS) for deploying any supported robot and sensor. The integration into commonly used OpenAI Gym and Baselines leads to an expandable and comparable benchmark for RL algorithms. We give a comprehensive overview of related work in the field of view planning and RL. A comparison of different RL algorithms provides a proof of concept for the framework’s functionality in experimental scenarios. The obtained results exhibit a coverage ratio of up to 0.8 illustrating its potential impact and expandability. The project will be made publicly available along with this article.Item Open Access A survey on learning-based robotic grasping(2020) Kleeberger, Kilian; Bormann, Richard; Kraus, Werner; Huber, Marco F.This review provides a comprehensive overview of machine learning approaches for vision-based robotic grasping and manipulation. Current trends and developments as well as various criteria for categorization of approaches are provided. Model-free approaches are attractive due to their generalization capabilities to novel objects, but are mostly limited to top-down grasps and do not allow a precise object placement which can limit their applicability. In contrast, model-based methods allow a precise placement and aim for an automatic configuration without any human intervention to enable a fast and easy deployment. Both approaches to robotic grasping and manipulation with and without object-specific knowledge are discussed. Due to the large amount of data required to train AI-based approaches, simulations are an attractive choice for robot learning. This article also gives an overview of techniques and achievements in transfers from simulations to the real world.Item Open Access Smart nesting : estimating geometrical compatibility in the nesting problem using graph neural networks(2023) Abdou, Kirolos; Mohammed, Osama; Eskandar, George; Ibrahim, Amgad; Matt, Paul-Amaury; Huber, Marco F.Reducing material waste and computation time are primary objectives in cutting and packing problems (C &P). A solution to the C &P problem consists of many steps, including the grouping of items to be nested and the arrangement of the grouped items on a large object. Current algorithms use meta-heuristics to solve the arrangement problem directly without explicitly addressing the grouping problem. In this paper, we propose a new pipeline for the nesting problem that starts with grouping the items to be nested and then arranging them on large objects. To this end, we introduce and motivate a new concept, namely the Geometrical Compatibility Index (GCI). Items with higher GCI should be clustered together. Since no labels exist for GCIs, we propose to model GCIs as bidirectional weighted edges of a graph that we call geometrical relationship graph (GRG). We propose a novel reinforcement-learning-based framework, which consists of two graph neural networks trained in an actor-critic-like fashion to learn GCIs. Then, to group the items into clusters, we model the GRG as a capacitated vehicle routing problem graph and solve it using meta-heuristics. Experiments conducted on a private dataset with regularly and irregularly shaped items show that the proposed algorithm can achieve a significant reduction in computation time (30% to 48%) compared to an open-source nesting software while attaining similar trim loss on regular items and a threefold improvement in trim loss on irregular items.Item Open Access On the development of a surrogate modelling toolbox for virtual assembly(2021) Kaufmann, Manuel; Effenberger, Ira; Huber, Marco F.Virtual assembly (VA) is a method to simulate the physical assembly (PA) of scanned parts. Small local part deviations can accumulate to large assembly deviations limiting the product quality. The propagation of geometrical deviations onto the assembly is a crucial step in tolerance management to assess the assembly quality. Current approaches for VA do not sufficiently consider the physical joining process. Therefore, the propagated assembly geometry may deviate strongly from the PA. In the state of the art, only specific and complex methods for particular joining processes are known. In this paper, the concept of Surrogate Models (SMs) is introduced, representing the connection between part and assembly geometries for particular joining processes. A Surrogate Modelling Toolbox (SMT) is developed that is intended to cover the variety of joining processes by the implementation of suitable SMs. A particular SM is created by the composition of suitable Surrogate Operations (SOs). An open list of SOs is presented. The composition of a SM is studied for a laser welding process of two polymer components. The resulting VA is compared to the PA in order to validate the developed model and is quantified by the exploitation ratio R.Item Open Access Machine learning-based shear force quality prediction of ultrasonic wire bonds : utilizing process data and machine data without additional sensors(2024) Buchner, Christoph; Seidler, Christian T.; Huber, Marco F.; Eigenbrod, Hartmut; Ribbeck, Hans-Georg von; Schlicht, FranzUltrasonic wire bonding is a highly automated production process that is used billions of times a year in the electronics and electromobility industries. Due to the complexity of the process and the large number of influencing parameters, there are currently no automated methods that can be used without additional sensors to evaluate the shear force bond quality quantitatively and non-destructively with sufficiently high precision. For this reason, this paper presents a new methodology that uses machine learning to enable quantitative, non-destructive prediction of bond quality without additional sensors. For this purpose, a machine learning algorithm was developed that uses various machine data and process data from existing sensors to quantitatively predict the shear force of the bonded joint. In addition, features are extracted from process time series, such as current, power, and frequency of the ultrasonic generator as well as deformation during bonding. Of the total of 2,090 features considered, the number of features could be reduced to 26 by recursive feature elimination, while maintaining the prediction accuracy. By using optimized deep neural networks, on average, a prediction precision of the regression on the shear force of the source bond of over 89.6% R 2 -score and a mean absolute error of 241 cN can be achieved.Item Open Access Towards scalability for resource reconfiguration in robotic assembly line balancing problems using a modified genetic algorithm(2024) Albus, Marcel; Hornek, Timothée; Kraus, Werner; Huber, Marco F.Assembly lines are still one of the most used manufacturing systems in modern-day production. Most research affects the building of new lines and, less frequently, the reconfiguration of existing lines. However, the first is insufficient to meet the reconfigurable production paradigm required by volatile market demands. Consequent reconfiguration of resources by production requests affects companies’ competitiveness. This paper introduces a problem-specific genetic algorithm for optimizing the reconfiguration of a Robotic Assembly Line Balancing Problem with Task Types, including additional company constraints. First, we present the greenfield and brownfield optimization objectives, then a mathematical problem formulation and the composition of the genetic algorithm. We evaluate our model against an Integer Programming baseline on a reconfiguration dataset with multiple equipment alternatives. The results demonstrate the capabilities of the genetic algorithm for the greenfield case and showcase the possibilities in the brownfield case. With a scalability improvement through computation time decrease of up to ∼2.75 ×, reduced number of equipment and workstations, but worse objective values, the genetic algorithm holds the potential for reconfiguring assembly lines. However, the genetic algorithm has to be further optimized for the reconfiguration to leverage its full potential.Item Open Access Comparing few-shot learning methods in various application scenarios for quality prediction in ultrasonic wire bonding(2026) Buchner, Christoph; Riedle, Benjamin; Seidler, Christian T.; Huber, Marco F.; Eigenbrod, Hartmut; von Ribbeck, Hans-Georg; Schlicht, FranzFew-shot learning refers to the problem of learning underlying patterns in data from just a few training samples, which contrasts with traditional deep learning methods that usually rely on large datasets. The collection of large datasets is typically costly and time-consuming and often requires significant computational resources. In ultrasonic wire bonding production processes, it is of fundamental importance that a high quality of the joints is ensured, while the application scenario often differs depending on the process, e.g. different materials, process sequences or machine settings. In this paper, we compare various few-shot learning approaches for predicting bond qualities in ultrasonic wire bonding. This is done by quantitatively predicting the shear force of the bonding joint across different application scenarios. The prediction is based on key process variables in the form of time-series data, such as deformation, ultrasonic power, frequency and current. These time-series vary in length depending on the bonding process. The few-shot learning approaches are compared in three application scenarios: Changing the bonding program (application scenario 1), changing the transducer (application scenario 2), and changing the bonding machine (application scenario 3). For example in application scenario 1, with just 15 retraining samples-less than 2.3% of the original training data-the shear force as a quality criterion for a 380 μm aluminum wire can be predicted with high accuracy. Using an autoencoder yields a mean absolute error of 90 cN, while meta-learning improves this to just 81 cN (less than approx. 5% of the average shear force).Item Open Access Machine learning-driven multi-objective parameter optimization for sustainable, efficient, and high-quality ultrasonic wire bonding(2025) Buchner, Christoph; Riedle, Benjamin; Krauß, Jonas; Seidler, Christian T.; Huber, Marco F.; Eigenbrod, Hartmut; von Ribbeck, Hans-Georg; Schlicht, FranzUltrasonic wire bonding, a highly automated production process, finds extensive use in the electronics and electromobility sectors, with billions of applications annually. Wire bonding, a critical step in electrical manufacturing, demands high quality while rising energy costs push industries to improve efficiency. The complexity of the process and the multitude of non-linear influencing parameters in the bonding process make it difficult for engineers to quickly find the optimum parameter sets for multiple response problems simultaneously solely based on their experience. As a result, engineers commonly resort to iterative trial and error approaches to establish wire bond parameters in practice. This paper introduces a novel, machine-learning-based methodology using established optimization algorithms for automated multi-objective parameter optimization in ultrasonic copper wire bonding, considering ten key parameters that influence the normal force profile, ultrasonic power profile, and process duration. The novelty of the proposed method lies in its ability to improve process sustainability by reducing energy input and tool wear, while simultaneously maximizing bond quality (shear force) and minimizing process time, without the need for a physical model or prior process knowledge. The paper shows that the combination of Bayesian optimization with artificial neural networks is particularly effective, achieving a 3.7% reduction in energy input and a 14.4% reduction in process time, while maintaining bond quality and reducing tool wear. This approach proves to be faster, less resource-intensive, and more effective than manual optimization methods, offering a scalable solution for industrial use.Item Open Access Nest smarter, not harder : a hybrid vision-based deep reinforcement learning agent for packing 2D irregular geometries by rotational placement(2025) Abdou, Kirolos; Ibrahim, Amgad Fakhry; Binder, Kai; Huber, Marco F.Nesting is pivotal in maximizing material use and productivity within manufacturing industries and involves the ordering, rotational placement, and translation placement of 2D irregular patterns onto raw material sheets. Despite the industrial significance, few methodologies tackle the challenging rotational placement problem due to its computational complexity. Unlike traditional search-based heuristics and meta-heuristics methods, this research pioneers a Deep Reinforcement Learning (DRL)-based framework that acquires a learning-based policy for flexible rotational placement and combines it with two rule-based policies to ensure a comprehensive nesting solution. Empowered by a bespoke Deep Learning (DL)-based geometric semantics extractor module, our approach achieves a 97% improvement in computation time and a 11%enhancement in material utilization compared to an open-source nesting software on a dataset from the sheet metal industry. Additionally, it shows competitive industry-practical performance against prevailing nesting algorithms on open datasets while being at least six times faster in computation time. Furthermore, this paper introduces a novel metric for geometrical irregularity, enriching the analysis and evaluation of nesting problems.