arXiv subjects

Daniel Jung

Publications and source records attributed to Daniel Jung.

At least 19 recordsLinked to original sources

arXiv2025

Probabilistic Machine Learning for Uncertainty-Aware Diagnosis of Industrial Systems

Deep neural networks has been increasingly applied in fault diagnostics, where it uses historical data to capture systems behavior, bypassing the need for high-fidelity physical models. However, despite their competence in prediction tasks, these models often struggle with the evaluation of their confidence. This matter is particularly important in consistency-based diagnosis where decision logic is highly sensitive to false alarms. To address this challenge, this work presents a diagnostic framework that uses ensemble probabilistic machine learning to improve diagnostic characteristics of data driven consistency based diagnosis by quantifying and automating the prediction uncertainty. The proposed method is evaluated across several case studies using both ablation and comparative analyses, showing consistent improvements across a range of diagnostic metrics.

cs.LG↗

arXiv2024

Evaluating the Impact of a Specialized LLM on Physician Experience in Clinical Decision Support: A Comparison of Ask Avo and ChatGPT-4

The use of Large language models (LLMs) to augment clinical decision support systems is a topic with rapidly growing interest, but current shortcomings such as hallucinations and lack of clear source citations make them unreliable for use in the clinical environment. This study evaluates Ask Avo, an LLM-derived software by AvoMD that incorporates a proprietary Language Model Augmented Retrieval (LMAR) system, in-built visual citation cues, and prompt engineering designed for interactions with physicians, against ChatGPT-4 in end-user experience for physicians in a simulated clinical scenario environment. Eight clinical questions derived from medical guideline documents in various specialties were prompted to both models by 62 study participants, with each response rated on trustworthiness, actionability, relevancy, comprehensiveness, and friendly format from 1 to 5. Ask Avo significantly outperformed ChatGPT-4 in all criteria: trustworthiness (4.52 vs. 3.34, p<0.001), actionability (4.41 vs. 3.19, p<0.001), relevancy (4.55 vs. 3.49, p<0.001), comprehensiveness (4.50 vs. 3.37, p<0.001), and friendly format (4.52 vs. 3.60, p<0.001). Our findings suggest that specialized LLMs designed with the needs of clinicians in mind can offer substantial improvements in user experience over general-purpose LLMs. Ask Avo's evidence-based approach tailored to clinician needs shows promise in the adoption of LLM-augmented clinical decision support software.

cs.HC↗

arXiv2024

The LiU-ICE Benchmark -- An Industrial Fault Diagnosis Case Study

This paper presents the LiU-ICE fault diagnosis benchmark. The purpose of the benchmark is to support fault diagnosis research by providing data and a model of an industrially relevant system. Data has been collected from an internal combustion engine test bench operated in both nominal and faulty modes. A state-of-the-art model of the air path through an internal combustion engine with unknown parameters is provided. This benchmark has previously been used in a competition at the 12th IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes (Safe Process) 2024, Ferrara, Italy.

eess.SY↗

arXiv2023

Stability-Informed Initialization of Neural Ordinary Differential Equations

This paper addresses the training of Neural Ordinary Differential Equations (neural ODEs), and in particular explores the interplay between numerical integration techniques, stability regions, step size, and initialization techniques. It is shown how the choice of integration technique implicitly regularizes the learned model, and how the solver's corresponding stability region affects training and prediction performance. From this analysis, a stability-informed parameter initialization technique is introduced. The effectiveness of the initialization method is displayed across several learning benchmarks and industrial applications.

cs.LG↗

arXiv2023

Analysis of Numerical Integration in RNN-Based Residuals for Fault Diagnosis of Dynamic Systems

Data-driven modeling and machine learning are widely used to model the behavior of dynamic systems. One application is the residual evaluation of technical systems where model predictions are compared with measurement data to create residuals for fault diagnosis applications. While recurrent neural network models have been shown capable of modeling complex non-linear dynamic systems, they are limited to fixed steps discrete-time simulation. Modeling using neural ordinary differential equations, however, make it possible to evaluate the state variables at specific times, compute gradients when training the model and use standard numerical solvers to explicitly model the underlying dynamic of the time-series data. Here, the effect of solver selection on the performance of neural ordinary differential equation residuals during training and evaluation is investigated. The paper includes a case study of a heavy-duty truck's after-treatment system to highlight the potential of these techniques for improving fault diagnosis performance.

cs.LG↗

arXiv2022

A Unifying Approach to Efficient (Near)-Gathering of Disoriented Robots with Limited Visibility

We consider a swarm of $n$ robots in \mathbb{R}^d. The robots are oblivious, disoriented (no common coordinate system/compass), and have limited visibility (observe other robots up to a constant distance). The basic formation task gathering requires that all robots reach the same, not predefined position. In the related near-gathering task, they must reach distinct positions such that every robot sees the entire swarm. In the considered setting, gathering can be solved in $\mathcal{O}(n + Δ^2)$ synchronous rounds both in two and three dimensions, where $Δ$ denotes the initial maximal distance of two robots. In this work, we formalize a key property of efficient gathering protocols and use it to define $λ$-contracting protocols. Any such protocol gathers $n$ robots in the $d$-dimensional space in $\mathcal{O}(Δ^2)$ synchronous rounds. Moreover, we prove a corresponding lower bound stating that any protocol in which robots move to target points inside of the local convex hulls of their neighborhoods -- $λ$-contracting protocols have this property -- requires $Ω(Δ^2)$ rounds to gather all robots. Among others, we prove that the $d$-dimensional generalization of the GtC-protocol is $λ$-contracting. Remarkably, our improved and generalized runtime bound is independent of $n$ and $d$. The independence of $d$ answers an open research question. We also introduce an approach to make any $λ$-contracting protocol collisionfree to solve near-gathering. The resulting protocols maintain the runtime of $Θ(Δ^2)$ and work even in the semi-synchronous model.

cs.DC↗

arXiv2022

Data-Driven Fault Diagnosis Analysis and Open-Set Classification of Time-Series Data

Fault diagnosis of dynamic systems is done by detecting changes in time-series data, for example residuals, caused by system degradation and faulty components. The use of general-purpose multi-class classification methods for fault diagnosis is complicated by imbalanced training data and unknown fault classes. Another complicating factor is that different fault classes can result in similar residual outputs, especially for small faults, which causes classification ambiguities. In this work, a framework for data-driven analysis and open-set classification is developed for fault diagnosis applications using the Kullback-Leibler divergence. A data-driven fault classification algorithm is proposed which can handle imbalanced datasets, class overlapping, and unknown faults. In addition, an algorithm is proposed to estimate the size of the fault when training data contains information from known fault realizations. An advantage of the proposed framework is that it can also be used for quantitative analysis of fault diagnosis performance, for example, to analyze how easy it is to classify faults of different magnitudes. To evaluate the usefulness of the proposed methods, multiple datasets from different fault scenarios have been collected from an internal combustion engine test bench to illustrate the design process of a data-driven diagnosis system, including quantitative fault diagnosis analysis and evaluation of the developed open set fault classification algorithm.

stat.ML↗

arXiv2021

Gathering a Euclidean Closed Chain of Robots in Linear Time

This work focuses on the following question related to the Gathering problem of $n$ autonomous, mobile robots in the Euclidean plane: Is it possible to solve Gathering of robots that do not agree on any axis of their coordinate systems (disoriented robots) and see other robots only up to a constant distance (limited visibility) in $o(n^2)$ fully synchronous rounds? The best known algorithm that solves Gathering of disoriented robots with limited visibility assuming oblivious robots needs $Θ(n^2)$ rounds [SPAA'11]. The lower bound for this algorithm even holds in a simplified closed chain model, where each robot has exactly two neighbors and the chain connections form a cycle. The only existing algorithms achieving a linear number of rounds for disoriented robots assume robots that are located on a two dimensional grid [IPDPS'16] and [SPAA'16]. Both algorithms make use of locally visible lights (the LUMINOUS model). In this work, we show for the closed chain model, that $n$ disoriented robots with limited visibility in the Euclidean plane can be gathered in $Θ\left(n\right)$ rounds assuming the LUMINOUS model. The lights are used to initiate and perform so-called runs along the chain. For the start of such runs, locally unique robots need to be determined. In contrast to the grid [IPDPS'16], this is not possible in every configuration in the Euclidean plane. Based on the theory of isogonal polygons by Grünbaum, we identify the class of isogonal configurations in which no such locally unique robots can be identified. Our solution combines two algorithms: The first one gathers isogonal configurations; it works without any lights. The second one works for non-isogonal configurations; it identifies locally unique robots to start runs, using a constant number of lights. Interleaving these algorithms solves the Gathering problem in $\mathcal{O}(n)$ rounds.

cs.DC↗

arXiv2020

Integrated Approximate Dynamic Programming and Equivalent Consumption Minimization Strategy for Eco-Driving in a Connected and Automated Vehicle

This paper focuses on the velocity planning and energy management problems for Connected and Automated Vehicles (CAVs) with hybrid electric powertrains. The eco-driving problem is formulated in the spatial domain as a nonlinear dynamic optimization problem, in which information about the upcoming speed limits and road topography is assumed to be known a priori. To solve this problem, a novel Dynamic Programming (DP) based optimization method is proposed, in which a causal Equivalent Consumption Minimization Strategy (ECMS) is embedded. The underlying vehicle model to predict energy consumption over real-world routes is validated using experimental data. Further, a multi-layer hierarchical control architecture is proposed as a pathway to real-time implementation in a vehicle. The DP-ECMS algorithm is introduced for a long-horizon optimization problem, and then adapted for a receding horizon implementation using principles in Approximate Dynamic Programming (ADP). This computationally economical alternative to the traditional DP solution is then benchmarked and evaluated.

eess.SY↗

arXiv2020

Residual Generation Using Physically-Based Grey-Box Recurrent Neural Networks For Engine Fault Diagnosis

Data-driven fault diagnosis is complicated by unknown fault classes and limited training data from different fault realizations. In these situations, conventional multi-class classification approaches are not suitable for fault diagnosis. One solution is the use of anomaly classifiers that are trained using only nominal data. Anomaly classifiers can be used to detect when a fault occurs but give little information about its root cause. Hybrid fault diagnosis methods combining physically-based models and available training data have shown promising results to improve fault classification performance and identify unknown fault classes. Residual generation using grey-box recurrent neural networks can be used for anomaly classification where physical insights about the monitored system are incorporated into the design of the machine learning algorithm. In this work, an automated residual design is developed using a bipartite graph representation of the system model to design grey-box recurrent neural networks and evaluated using a real industrial case study. Data from an internal combustion engine test bench is used to illustrate the potentials of combining machine learning and model-based fault diagnosis techniques.

eess.SP↗

arXiv2019

Isolation and Localization of Unknown Faults Using Neural Network-Based Residuals

Localization of unknown faults in industrial systems is a difficult task for data-driven diagnosis methods. The classification performance of many machine learning methods relies on the quality of training data. Unknown faults, for example faults not represented in training data, can be detected using, for example, anomaly classifiers. However, mapping these unknown faults to an actual location in the real system is a non-trivial problem. In model-based diagnosis, physical-based models are used to create residuals that isolate faults by mapping model equations to faulty system components. Developing sufficiently accurate physical-based models can be a time-consuming process. Hybrid modeling methods combining physical-based methods and machine learning is one solution to design data-driven residuals for fault isolation. In this work, a set of neural network-based residuals are designed by incorporating physical insights about the system behavior in the residual model structure. The residuals are trained using only fault-free data and a simulation case study shows that they can be used to perform fault isolation and localization of unknown faults in the system.

eess.SP↗

arXiv2018

Competitive Routing in Hybrid Communication Networks

Routing is a challenging problem for wireless ad hoc networks, especially when the nodes are mobile and spread so widely that in most cases multiple hops are needed to route a message from one node to another. In fact, it is known that any online routing protocol has a poor performance in the worst case, in a sense that there is a distribution of nodes resulting in bad routing paths for that protocol, even if the nodes know their geographic positions and the geographic position of the destination of a message is known. The reason for that is that radio holes in the ad hoc network may require messages to take long detours in order to get to a destination, which are hard to find in an online fashion. In this paper, we assume that the wireless ad hoc network can make limited use of long-range links provided by a global communication infrastructure like a cellular infrastructure or a satellite in order to compute an abstraction of the wireless ad hoc network that allows the messages to be sent along near-shortest paths in the ad hoc network. We present distributed algorithms that compute an abstraction of the ad hoc network in $\mathcal{O}\left(\log ^2 n\right)$ time using long-range links, which results in $c$-competitive routing paths between any two nodes of the ad hoc network for some constant $c$ if the convex hulls of the radio holes do not intersect. We also show that the storage needed for the abstraction just depends on the number and size of the radio holes in the wireless ad hoc network and is independent on the total number of nodes, and this information just has to be known to a few nodes for the routing to work.

cs.DC↗

arXiv2017

Gathering Anonymous, Oblivious Robots on a Grid

We consider a swarm of $n$ autonomous mobile robots, distributed on a 2-dimensional grid. A basic task for such a swarm is the gathering process: All robots have to gather at one (not predefined) place. A common local model for extremely simple robots is the following: The robots do not have a common compass, only have a constant viewing radius, are autonomous and indistinguishable, can move at most a constant distance in each step, cannot communicate, are oblivious and do not have flags or states. The only gathering algorithm under this robot model, with known runtime bounds, needs $\mathcal{O}(n^2)$ rounds and works in the Euclidean plane. The underlying time model for the algorithm is the fully synchronous $\mathcal{FSYNC}$ model. On the other side, in the case of the 2-dimensional grid, the only known gathering algorithms for the same time and a similar local model additionally require a constant memory, states and "flags" to communicate these states to neighbors in viewing range. They gather in time $\mathcal{O}(n)$. In this paper we contribute the (to the best of our knowledge) first gathering algorithm on the grid that works under the same simple local model as the above mentioned Euclidean plane strategy, i.e., without memory (oblivious), "flags" and states. We prove its correctness and an $\mathcal{O}(n^2)$ time bound in the fully synchronous $\mathcal{FSYNC}$ time model. This time bound matches the time bound of the best known algorithm for the Euclidean plane mentioned above. We say gathering is done if all robots are located within a $2\times 2$ square, because in $\mathcal{FSYNC}$ such configurations cannot be solved.

cs.DC↗

arXiv2016

Long-range Response in AC Electricity Grids

Local changes in the topology of electricity grids can cause overloads far away from the disturbance, making the prediction of the robustness against changes in the topology - for example caused by power outages or grid extensions - a challenging task. The impact of single-line additions on the long-range response of DC electricity grids has recently been studied. By solving the real part of the static AC load flow equations, we conduct a similar investigation for AC grids. In a regular 2D grid graph with cyclic boundary conditions, we find a power law decay for the change of power flow as a function of distance to the disturbance over a wide range of distances. The power exponent increases and saturates for large system sizes. By applying the same analysis to the German transmission grid topology, we show that also in real-world topologies a long-ranged response can be found.

nlin.AO↗

arXiv2016

Cascading Failures in AC Electricity Grids

Sudden failure of a single transmission element in a power grid can induce a domino effect of cascading failures, which can lead to the isolation of a large number of consumers or even to the failure of the entire grid. Here we present results of the simulation of cascading failures in power grids, using an alternating current (AC) model. We first apply this model to a regular square grid topology. For a random placement of consumers and generators on the grid, the probability to find more than a certain number of unsupplied consumers decays as a power law and obeys a scaling law with respect to system size. Varying the transmitted power threshold above which a transmission line fails does not seem to change the power law exponent $q \approx 1.6$. Furthermore, we study the influence of the placement of generators and consumers on the number of affected consumers and demonstrate that large clusters of generators and consumers are especially vulnerable to cascading failures. As a real-world topology we consider the German high-voltage transmission grid. Applying the dynamic AC model and considering a random placement of consumers, we find that the probability to disconnect more than a certain number of consumers depends strongly on the threshold. For large thresholds the decay is clearly exponential, while for small ones the decay is slow, indicating a power law decay.

nlin.AO↗

arXiv2016

Anderson Metal-Insulator Transitions With Classical Magnetic Impurities

We study effects of classical magnetic impurities on the Anderson metal-insulator transition numerically. We find that a small concentration of Heisenberg impurities enhances the critical disorder amplitude $W_{\rm c}$ with increasing exchange coupling strength $J$. The resulting scaling with $J$ is analyzed which supports an anomalous scaling prediction by Wegner due to the combined breaking of time-reversal and spin-rotational symmetry. Moreover, we find that the presence of magnetic impurities lowers the critical correlation length exponent $ν$ and enhances the multifractality parameter $α_0$. The new value of $ν$ improves the agreement with the value measured in experiments on the metal-insulator transition (MIT) in doped semiconductors like phosphor-doped silicon, where a finite density of magnetic moments is known to exist in the vicinity of the MIT. The results are obtained by a finite-size scaling analysis of the geometric mean of the local density of states which is calculated by means of the kernel polynomial method. We establish this combination of numerical techniques as a method to obtain critical properties of disordered systems quantitatively.

cond-mat.dis-nn↗

arXiv2016

Asymptotically Optimal Gathering on a Grid

In this paper, we solve the local gathering problem of a swarm of $n$ indistinguishable, point-shaped robots on a two dimensional grid in asymptotically optimal time $\mathcal{O}(n)$ in the fully synchronous $\mathcal{FSYNC}$ time model. Given an arbitrarily distributed (yet connected) swarm of robots, the gathering problem on the grid is to locate all robots within a $2\times 2$-sized area that is not known beforehand. Two robots are connected if they are vertical or horizontal neighbors on the grid. The locality constraint means that no global control, no compass, no global communication and only local vision is available; hence, a robot can only see its grid neighbors up to a constant $L_1$-distance, which also limits its movements. A robot can move to one of its eight neighboring grid cells and if two or more robots move to the same location they are \emph{merged} to be only one robot. The locality constraint is the significant challenging issue here, since robot movements must not harm the (only globally checkable) swarm connectivity. For solving the gathering problem, we provide a synchronous algorithm -- executed by every robot -- which ensures that robots merge without breaking the swarm connectivity. In our model, robots can obtain a special state, which marks such a robot to be performing specific connectivity preserving movements in order to allow later merge operations of the swarm. Compared to the grid, for gathering in the Euclidean plane for the same robot and time model the best known upper bound is $\mathcal{O}(n^2)$.

cs.DC↗

arXiv2016

Anderson Metal-Insulator Transitions With Classical Magnetic Impurities: Supplemental material

In the supplemental materials we justify our choice of the number of Chebychev moments used within the kernel polynomial method, show some preliminary results for the large coupling behavior, discuss possible correlation effects in the local density of states, estimate the spin relaxation length and introduce the goodness of fit probability that is used to assess the quality of the fits.

cond-mat.dis-nn↗