Publications

Publications, preprints, and conference tutorials

21 publications · Publications, preprints, and conference tutorials

Closed-Form Gaussian Information Projection for Embedded Simplices and Simplicial Mixtures

Ali Darijani, Emadedin Soufian, Jürgen Beyerer
2026
Preprint
Abstract
We study Gaussian approximation of Dirichlet–Gaussian probability density functions on embedded simplices. We derive the unique information projection onto the family of nondegenerate ambient Gaussian probability density functions, obtaining closed-form expressions for the optimal Gaussian and the minimum divergence for arbitrary positive Dirichlet parameters. The minimum divergence depends only on the intrinsic simplex dimension and Dirichlet parameters, is invariant up to an affine isomorphism, independent of the ambient dimension, and independent of the normal covariance. Furthermore, by joint convexity of Kullback–Leibler divergence, we establish explicit intrinsic dimension-dependent sharp upper bounds for finite Gaussian Mixture with shared component weights. For uniform probability density functions, these bounds reduce to dimension-dependent constants. The results provide closed-form Gaussian Mixture constructions and rigorous approximation guarantees for probability density functions associated with embedded simplicial complexes. A numerical example on a circular curve illustrates the theoretical results and their connections to geometric Gaussian Mixture representations.
arXiv DOI LaTeX Source PDF URL

Closed-Form Gaussian Information Projection for Embedded Hypercubes and Beta–Gaussian Mixtures

Ali Darijani, Emadedin Soufian, Jürgen Beyerer
2026
Preprint
Abstract
We derive closed-form Gaussian information projections of Beta–Gaussian distributions whose tangential components are supported on affine embedded hypercubes and whose normal components are Gaussian. The optimal Gaussian is uniquely characterized by moment matching, and the corresponding minimum Kullback–Leibler divergence admits an explicit expression depending only on the intrinsic Beta parameters, independently of the embedding geometry, ambient dimension, and normal Gaussian parameters. In the uniform case, this divergence reduces to a linear function of the intrinsic dimension. We further extend this construction to finite, pure embedded cubical complexes by considering mixtures of the componentwise Gaussian projections. Using the joint convexity of the Kullback–Leibler divergence, we establish an explicit upper bound on the approximation error of the resulting Gaussian Mixture. For uniform tangential distributions and volume-based mixture weights, the bound is linear in the intrinsic dimension and independent of the geometry of the complex, its embedding, and the normal Gaussian parameters.
arXiv DOI LaTeX Source PDF URL

Comprehensive Description of Uncertainty in Measurement for Representation and Propagation with Scalable Precision

Ali Darijani, Jürgen Beyerer, Zahra Sadat Hajseyed Nasrollah, Luisa Hoffmann, Michael Heizmann
Preprint • 2026
Preprint
Abstract
Probability theory has become the predominant framework for quantifying uncertainty across scientific and engineering disciplines, with a particular focus on measurement and control systems. However, the widespread reliance on simple Gaussian assumptions–particularly in control theory, manufacturing, and measurement systems–can result in incomplete representations and multistage lossy approximations of complex phenomena, including inaccurate propagation of uncertainty through multi stage processes. This work proposes a comprehensive yet computationally tractable framework for representing and propagating quantitative attributes arising in measurement systems using Probability Density Functions (PDFs). Recognizing the constraints imposed by finite memory in software systems, we advocate for the use of Gaussian Mixture Models (GMMs), a principled extension of the familiar Gaussian framework, as they are universal approximators of PDFs whose complexity can be tuned to trade off approximation accuracy against memory and computation. From both mathematical and computational perspectives, GMMs enable high performance and, in many cases, closed form solutions of essential operations in control and measurement. The paper presents practical applications within manufacturing and measurement contexts especially circular factory, demonstrating how the GMMs framework supports accurate representation and propagation of measurement uncertainty and offers improved accuracy–compared to the traditional Gaussian framework–while keeping the computations tractable.
arXiv DOI PDF URL

Stakeholder-Based Deterministic and Probabilistic Quality Control Using Gaussian Mixtures for Circular Factories

Ali Darijani, Gabriel Moser, Kristian Vlajic, Zahra Sadat Hajseyed Nasrollah, Luisa Hoffmann, Mehdi Khabou, , Albert Albers
2026 Conference on Circular Production Systems and Technologies (CCPT) • 2026
Accepted
Abstract
Quality Control (QC) in Circular Factorys (CFs) is fundamentally more challenging due to the uncertainty of the incoming products to be reprocessed, compared to linear factories, which have tight tolerances on the input raw materials and parts. One must have an expressive model to manage the multimodality of circular factories. QC and the decision making process that it entails can be understood as a mapping from system requirements and variation space to discrete decisions, which can be formulated either deterministically or probabilistically. Ideally, the model must also be able to handle the deterministic and probabilistic cases seamlessly. Gaussian Mixture Models (GMMs) and their universal approximation property are considered suitable for the circular factory context. GMM can also address deterministic cases by approximating the Dirac delta as a limiting case. We present a stakeholder-oriented QC framework in which stakeholder requirements, product generations, and alternative solutions for new product generations are represented probabilistically over quality-relevant variation spaces. Two complementary decision perspectives are considered. First, stakeholder requirements, exemplified by customer requirements, are modeled as acceptance regions, and the probability that a product generation satisfies individual, joint, or alternative requirements is computed by integrating the corresponding GMM density over these regions. Second, stakeholder requirements and alternative solutions are represented as probability density functions, and distance- and divergence-based metrics are used to quantify their dissimilarity. These metrics support threshold-based acceptance and ranking of alternative solutions under uncertainty. Special attention is given to deterministic limits, threshold design, and the interpretation of distances and divergences when comparing GMMs. The proposed framework provides a basis for stakeholder-based deterministic and probabilistic QC and decision support in circular factories.
PDF URL

A Geometric Gaussian Mixture Representation of Plane Curves

Ali Darijani, Benedikt Stratmann, Jürgen Beyerer
2026 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI) • 2026
Accepted
Abstract
We introduce a user defined probabilistic polygonal representation for plane curves. Given a curve, we select vertices on the curve and connect consecutive vertices by line segments to obtain a polygonal approximation. Each segment is equipped with a user defined uncertainty parameter in the normal direction. This yields a collection of thin probabilistic geometric primitives that retain the geometrz of the underlying curve while extending it beyond the idealized deterministic one dimensional formulation. For each segment, we define a Random Variable that is uniform distributed in the tangent direction of the segment and Gaussian distributed in the normal direction of the segment. By matching the first and the second central moments, this construction induces a Gaussian component whose mean lies at the segment midpoint and whose covariance encodes both tangential and normal uncertainty. Combining the segment wise components with appropriate weights yields a Gaussian Mixture Model (GMM) representation of the user defined probabilistic polygonal representation of the plane curve. The proposed framework provides an analytically tractable probabilistic model that preserves local geometry, and uncertainty in the normal direction. It applies to smooth, closed, open, non regular, and self intersecting plane curves, allows adaptive discretization and varying uncertainty in the normal direction, and as a result supports uncertainty aware geometric modeling. Experiments on a collection of canonical plane curves show that the resulting GMM capture local tangent, local normal, and local arc length; resulting in the global shape of the underlying curves to be truthfully captured as well. The representation is particularly relevant for applications in uncertainty aware CAD and digital twins, probabilistic obstacle modeling in robotics, and probabilistic trajectory planning.
arXiv DOI LaTeX Source PDF URL

Closed-Form Gaussian Mixture Representations of d-Dimensional Triangulated Domains

Ali Darijani, Benedikt Stratmann, Ratan Bahadur Thapa, Uwe D. Hanebeck, Jürgen Beyerer, Nehal Afifi
2026 IEEE Sensor Data Fusion: Trends, Solutions, Applications (SDF) • 2026
Accepted
Abstract
We derive a Gaussian Mixture (GM) representation of a full-dimensional subset of Euclidean space with finite volume from a finite simplicial triangulation in closed form. On each simplex, we replace the uniform probability density function (PDF) with a Gaussian PDF that minimizes the Kullback–Leibler (KL) divergence, arriving at the information projection onto the family of Gaussian PDFs. By exploiting affine isomorphisms between simplices, the joint convexity and bijection invariance properties of the KL divergence, identities for Dirichlet PDFs, and a partition-of-unity argument, we establish a dimension-dependent but geometry-independent sharp upper bound on the global KL divergence between the resulting GM PDF and the uniform PDF on the original domain. Numerical examples illustrate the construction and demonstrate its potential in practical applications.
arXiv DOI LaTeX Source PDF URL

Probabilistic quality control without tears

Ali Darijani
Proceedings of the 2025 Joint Workshop of Fraunhofer IOSB and Institute for Anthropomatics, Vision and Fusion Laboratory • 2026
Published
Abstract
Quality control is often associated with strict rules, detailed calculations, and intimidating statistical language. This perception discourages many practitioners from fully engaging with probabilistic methods, even though they already rely on them implicitly. This report presents probabilistic quality control in a concept-driven, non-mathematical way. By focusing on intuition, examples, and practical decision-making, it demonstrates how probability helps manage uncertainty rather than complicate it. The aim is to make probabilistic quality control accessible, usable, and free from unnecessary anxiety.
DOI PDF URL

gmtorch: Computing with Gaussian Mixture Uncertainty in PyTorch

Ali Darijani, Benedikt Stratmann, Genís Floriach Pigem, Jürgen Beyerer, Luisa Hoffmann, Michael Heizmann
2026 IEEE Sensor Data Fusion: Trends, Solutions, Applications (SDF) • 2026
Accepted
Abstract
Gaussian mixture models (GMMs) can represent multimodal, non-Gaussian uncertainty while retaining analytical structure. Scientific workflows, however, require more than fitting a mixture to observations: uncertain quantities must remain available for subsequent transformations, arithmetic, conditioning, and information combination. We present gmtorch, a PyTorch framework for computing with GMM-represented uncertain scalar and vector quantities. Mixtures may be constructed from parameters or estimated from data; every mixture-valued result remains a complete distribution that can be evaluated, sampled, summarized, conditioned, or used in a subsequent calculation. Marginalization, conditioning, affine maps, sums and differences under independence, and normalized density products are evaluated analytically and can be composed with assumptions made explicit by the selected operations and with component counts tracked. When nonlinear maps leave the finite-GMM family, or exact operations produce impractically many components, the framework provides explicit sample–transform–refit propagation and mixture reduction. Full-covariance computations use Cholesky-based linear algebra and blocked pairwise kernels. Reference checks compare exact operations with independently assembled formulas, and a single-epoch localization example illustrates sequential likelihood-based updating. The contribution is a consistent software realization of established GMM calculus, with exact and approximate steps made explicit.
DOI GitLab PDF URL

Uncertainty Quantification and the need for better understanding of Monte Carlo and random number generation

Ali Darijani
Proceedings of the 2024 Joint Workshop of Fraunhofer IOSB and Institute for Anthropomatics, Vision and Fusion Laboratory • 2025
Published
Abstract
Uncertainty Quantification (UQ) is critical in scientific modeling and decision making, addressing the inherent uncertainty and lack of precision in computational modeling. Monte Carlo methods, one of the two prominent approaches in UQ, rely on random number generation to simulate complex systems and estimate probability density functions. However, the quality of these methods is naturally tied to the integrity of the random numbers used. Poorly designed random number generators (RNGs) can lead to biased or unreliable results, undermining the validity of UQ. Despite advancements in RNG algorithms, challenges remain in balancing computational efficiency, reproducibility, and randomness quality. A deeper understanding of these generators, their limitations, and their integration with Monte Carlo methods is essential to advancing UQ practices. This technical report tries to highlight keywords in the random number generation field and give a couple of examples showcasing the pitfalls.
DOI PDF

Gaussian mixture belief propagation for probabilistic graphical models

Luisa Hoffmann, Ali Darijani, Mehdi Khabou, Volker Schulze, Jürgen Beyerer, Michael Heizmann
2026 29th International Conference on Information Fusion (FUSION) • 2026
Published
Abstract
Inference in continuous probabilistic graphical models remains challenging due to nonlinear dependencies, multi-modal posteriors, and the absence of closed-form solutions. While Gaussian Belief Propagation (GaBP) offers efficient inference for unimodal Gaussians, it fails to capture complex, data-driven distributions. Sampling-based methods, in turn, lack analytical tractability and scalability. We present Gaussian Mixture Belief Propagation (GMBP), a novel framework that represents pairwise dependencies as joint Gaussian mixture models (GMMs) and enables closed-form message passing via analytic conditioning and marginalization. GMBP unifies the treatment of hard (Dirac) and soft (Gaussian) evidence, updating each mixture component in a Kalman-style fashion while preserving multimodality. To prevent the combinatorial explosion of components, we integrate principled Gaussian mixture reduction (GMR) techniques, ensuring computational tractability without sacrificing expressive power. Evaluated on synthetic graphs, the Abalone benchmark dataset, and real-world material science data from the Circular Factory, GMBP demonstrates superior likelihood and multimodal fidelity compared to GaBP. Our results establish GMBP as a flexible, scalable, and analytically grounded inference engine for continuous factor graphs in data- rich, nonlinear domains.
DOI URL

Instance-Specific Production Control in Circular Manufacturing Using Gaussian Mixture Models

Finn Bail, Ali Darijani, Luisa Hoffmann, Maurice Engels, Jens Becker, Martin Benfer, , Gisela Lanza
2026 Conference on Circular Production Systems and Technologies (CCPT) • 2026
Accepted
Abstract
Quality Control (QC) in Circular Factorys (CFs) is fundamentally more challenging due to the uncertainty of the incoming products to be reprocessed, compared to linear factories, which have tight tolerances on the input raw materials and parts. One must have an expressive model to manage the multimodality of circular factories. QC and the decision making process that it entails can be understood as a mapping from system requirements and variation space to discrete decisions, which can be formulated either deterministically or probabilistically. Ideally, the model must also be able to handle the deterministic and probabilistic cases seamlessly. Gaussian Mixture Models (GMMs) and their universal approximation property are considered suitable for the circular factory context. GMM can also address deterministic cases by approximating the Dirac delta as a limiting case. We present a stakeholder-oriented QC framework in which stakeholder requirements, product generations, and alternative solutions for new product generations are represented probabilistically over quality-relevant variation spaces. Two complementary decision perspectives are considered. First, stakeholder requirements, exemplified by customer requirements, are modeled as acceptance regions, and the probability that a product generation satisfies individual, joint, or alternative requirements is computed by integrating the corresponding GMM density over these regions. Second, stakeholder requirements and alternative solutions are represented as probability density functions, and distance- and divergence-based metrics are used to quantify their dissimilarity. These metrics support threshold-based acceptance and ranking of alternative solutions under uncertainty. Special attention is given to deterministic limits, threshold design, and the interpretation of distances and divergences when comparing GMMs. The proposed framework provides a basis for stakeholder-based deterministic and probabilistic QC and decision support in circular factories.
PDF URL

Automatic Gaussian Mixture Sampling for Uncertainty Management in Circular Factories

Zahra Sadat Hajseyed Nasrollah, Ali Darijani, Mehdi Khabou, Stefan Dietrich, Volker Schulze, Jürgen Beyerer
2026 Conference on Circular Production Systems and Technologies (CCPT) • 2026
Accepted
Abstract
Returned components in circular manufacturing often form heterogeneous condition populations that are poorly represented by a single Gaussian distribution. Such parts arrive with different usage, inspection, repair, and degradation histories. As a result, product condition, process duration, and quality risk are better viewed as random variables with several condition regimes rather than as one homogeneous population. In this work, we represent this structure using Gaussian mixture models (GMMs), which combine discrete regime information and continuous variability in one probabilistic description. Many quantities needed for planning and risk assessment do not admit tractable closed-form expressions once GMM uncertainties are propagated through nonlinear engineering models or decision rules. We therefore present an automatic Monte Carlo sampling layer for GMM-based uncertainty propagation. The user specifies a tolerance and confidence level, and sampling continues until the estimated confidence interval for the selected output quantity satisfies the requested precision or the maximum sampling budget is reached. The approach enables GMM-based uncertainty models to be used in simulation-based planning, routing, inspection, and quality assessment in circular factories without manually fixing the Monte Carlo sample size in advance. We demonstrate the workflow with a proof of concept fatigue strength example for returned or treated metallic components. Multimodal surface state uncertainty is propagated through a nonlinear fatigue strength relation and the output is assessed by its moments and by a low strength probability.
PDF URL

Compositional Boundaries for Density Fusion

Ratan Bahadur Thapa, Ali Darijani, Jürgen Beyerer, Steffen Staab
2026 International Conference on Scalable Uncertainty Management (SUM) • 2026
Published
Abstract
Distributed uncertainty-management systems often combine local probabilistic models along aggregation trees chosen by communication, privacy, or scheduling constraints. The final density should depend on the weighted sources, not on the particular order in which intermediate nodes combine them. We study this requirement as an algebraic compositionality problem for binary fusion of weighted probability densities. The central question is when a local fusion rule can be executed hierarchically while remaining order-invariant. We establish a compositional boundary for local segment-valued fusion rules. Within the class of continuous binary rules with additive output weights and weight-only coefficients, order-invariant hierarchical execution characterizes normalized weighted linear pooling; norm-induced segment balancing realizes the corresponding coefficient. Smooth endpoint-to-candidate f-divergence balancing has a different local geometry: its quadratic expansion induces square-root effective weights, showing why pairwise solvability alone is insufficient for schedule-independent fusion. We show that this obstruction is local to endpoint-to-candidate binary balancing, whereas global divergence barycenters retain additive-weight local limits. Finally, Gaussian mixtures show how the same issue appears in finite model classes: exact fusion is compositional, whereas stepwise compression is compositional only under a congruence condition on unnormalized component measures. These results distinguish exact schedule-independent fusion from global aggregation objectives and local approximation heuristics.
arXiv DOI LaTeX Source PDF URL

Gaussian Mixture Belief Propagation for Hybrid Inference in Factor Graphs

Luisa Hoffmann, Ali Darijani, Jürgen Beyerer, Michael Heizmann
2026 IEEE Sensor Data Fusion: Trends, Solutions, Applications (SDF) • 2026
Accepted
Abstract
Sensor data fusion in industrial systems often requires combining continuous measurements with discrete semantic information while accounting for measurement uncertainty and multimodal distributions. We propose Gaussian Mixture Belief Propagation (GMBP), a probabilistic message-passing frame-work for hybrid factor graphs that represents discrete-continuous dependencies using Gaussian mixture models (GMMs). For each discrete state, the corresponding continuous evidence is represented by a conditional GMM, enabling multimodal distributions to be propagated between continuous and discrete variables. In particular, GMBP supports both hard and uncertain continuous evidence and provides closed-form message updates in both directions. The resulting framework enables the fusion of multiple continuous evidence sources into a common discrete hypothesis while retaining the uncertainty of the individual measurements. A synthetic two-class example demonstrates the bidirectional fusion of hard and soft evidence. On a real-world fault classification benchmark, 25 repeated stratified cross-validation runs show that adaptively selecting the Gaussian mixture order per class and feature via Bayesian Information Criterion (BIC) improves accuracy by 5–10 percentage points over both a single-Gaussian baseline and a fixed-order mixture. This demonstrates the benefit of GMBP’s data-driven multimodal representation and highlights the potential of GMM-based belief propagation for probabilistic hybrid sensor data fusion with multimodal continuous evidence.

Efficient uncertainty propagation in continuous Bayesian networks using GUM and Monte Carlo methods

Luisa Hoffmann, Michael Heizmann, Ali Darijani, Jürgen Beyerer
2026 International Conference on Semantic Computing (ICSC) • 2026
Published
Abstract
In large and complex systems such as the Circular Factory, vast amounts of heterogeneous and uncertain data are generated. Classical Bayesian Networks (BNs) provide a principled framework for reasoning under uncertainty, but their reliance on discrete Conditional Probability Tables (CPTs) limits their applicability in continuous domains. In the context of semantic computing, where heterogeneous sensor data and knowledge graph information must be integrated, this limitation prevents probabilistic reasoning from being consistently applied to continuous quantities and measurement-based information. This paper introduces a framework for continuous Bayesian Networks that replaces discrete CPTs with functional dependencies and integrates the Guide to the Expression of Uncertainty in Measurement (GUM) as a standardized approach for analytical uncertainty propagation. The proposed method enables probabilistic reasoning over continuous variables while explicitly accounting for measurement and parameter uncertainty in a mathematically transparent way. To complement the analytical formulation, Monte Carlo (MC) simulations with Gaussian Mixture Models (GMMs) are employed to handle non-linear and non-Gaussian cases, allowing for numerical validation of the GUM approximation. We present illustrative examples comparing both approaches and discuss their agreement and limitations. The results demonstrate that GUM-based inference provides accurate and computationally efficient uncertainty estimates in locally linear regions, while Monte Carlo methods serve as a reliable benchmark for more complex, non-linear dependencies. This framework thus lays the foundation for robust and interpretable uncertainty reasoning in environments such as the Circular Factory. Furthermore, it provides a bridge between probabilistic reasoning and semantic representations by enabling continuous uncertainty information to be embedded within knowledge-based systems, thus advancing uncertainty-aware semantic computing.
DOI URL

Probabilistic Process-Structure-Property Modelling for Hardness and Porosity Estimation in High-Speed DED-Reprocessed Components in the context of the Circular Factory

Luisa Hoffmann, Mehdi Khabou, Helena Wexel, Ali Darijani, Jürgen Beyerer, Frederik Zanger, , Michael Heizmann
2026 Conference on Circular Production Systems and Technologies (CCPT) • 2026
Accepted
Abstract
In Circular Factory (CF) environments, used components are disassembled, characterized, and reprocessed, based on a reliable estimation of the resulting material properties and taking measurement and remanufacturing uncertainties into consideration. For reprocessing steps such as High-Speed Laser Directed Energy Deposition (HS DED-LB), uncertainties and process variability critically affect the robustness and sta- bility of process-structure-property predictions. Additionally, measurements introduce uncertainties that propagate into the estimation of the post-process material state. For remanufactured structural components, hardness and porosity serve as key quality indicators describing the ma- terial state. Both are investigated as key determinants of fatigue performance, since hardness governs local fatigue strength, while porosity acts as a stress concentrator that promotes early fatigue crack initiation and reduces fatigue strength and component lifetime. In HS DED-LB, laser power and scan speed govern the thermal input and thereby influence microstructure development and resulting material properties. The process parameters can act as physically based features representing the thermal history and its fluctuations to correlate the process to the material state. To address this issue, Gaussian Mixture Models (GMMs) are adopted as a unified probabilistic framework to represent multimodal uncertainty and enable consistent inference under process variability and measurement noise. This paper presents a probabilistic approach for modelling and fusing measurement and process uncertainties using GMMs, resulting in probability density functions of hardness and porosity after HS DED-LB reprocessing. The approach enables uncertainty-aware estimation of the material state and supports adaptive tuning of reprocessing parameters to achieve the desired target quality characteristics in circular factories.
PDF URL

A Geometric Gaussian Mixture Representation of Curves and Trajectories in Euclidean Space

Benedikt Stratmann, Emadedin Soufian, Jürgen Beyerer, Ali Darijani
2026 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI) • 2026
Accepted
Abstract
Curves and trajectories in Euclidean spaces are fundamental objects in tracking and sensor data fusion. Gaussian mixture models (GMMs) are widely used as flexible parametric probability density functions (PDFs) over the object state space. They are typically obtained by applying expectation-maximization (EM)-based schemes to point samples. However, when the underlying geometric structure is known deterministically, for example, a road centerline in map-assisted tracking, a nominal flight path, or a planned robot trajectory, such iterative fitting is unnecessary, potentially slow, and only indirectly reflects the underlying curve or trajectory. We propose a purely arithmetic, geometry-driven construction that directly maps a curve or trajectory in a Euclidean space, together with user-defined transverse uncertainty, to a GMM in a single pass, without EM or iterative optimization. Starting from a parametrized curve, we construct a uniform discretized polygonal representation and assign each segment a standard deviation in its normal subspace. This results in thin, curve-aligned Gaussian primitives that encode position, tangent, normal, and length scale, together with transverse uncertainty. By choosing component weights proportional to segment lengths, the resulting probability distribution is uniform with respect to arc length. Consequently, all GMM parameters are obtained directly from the discretized geometry and the prescribed uncertainty profile. The construction accommodates smooth, open, closed, non-regular, and self-intersecting curves and trajectories in arbitrary-dimensional Euclidean spaces. It supports adaptive discretization and spatially varying uncertainty, yielding differentiable GMMs that directly encode the underlying geometric structure. We demonstrate the approach using canonical planar curves and nominal trajectories, and conduct a simulation study in higher-dimensional Euclidean spaces to compare approximation quality and computational runtime against EM-based fitting. Potential applications include map- and road-constrained tracking and sensor data fusion, uncertainty-aware trajectory planning, probabilistic obstacle and boundary modeling in robotics, and uncertainty-aware computer-aided design (CAD) and digital twins.
arXiv DOI LaTeX Source PDF URL

Representing Measurement Uncertainty by Gaussian Mixtures

Luisa Hoffmann, M. Y. Alayoubi, Michael Heizmann, Ali Darijani, Jürgen Beyerer
Sensoren und Messsysteme (2026) • 2026
Published
Abstract
Sensor data fusion in industrial systems often requires combining continuous measurements with discrete semantic information while accounting for measurement uncertainty and multimodal distributions. We propose Gaussian Mixture Belief Propagation (GMBP), a probabilistic message-passing frame-work for hybrid factor graphs that represents discrete-continuous dependencies using Gaussian mixture models (GMMs). For each discrete state, the corresponding continuous evidence is represented by a conditional GMM, enabling multimodal distributions to be propagated between continuous and discrete variables. In particular, GMBP supports both hard and uncertain continuous evidence and provides closed-form message updates in both directions. The resulting framework enables the fusion of multiple continuous evidence sources into a common discrete hypothesis while retaining the uncertainty of the individual measurements. A synthetic two-class example demonstrates the bidirectional fusion of hard and soft evidence. On a real-world fault classification benchmark, 25 repeated stratified cross-validation runs show that adaptively selecting the Gaussian mixture order per class and feature via Bayesian Information Criterion (BIC) improves accuracy by 5–10 percentage points over both a single-Gaussian baseline and a fixed-order mixture. This demonstrates the benefit of GMBP’s data-driven multimodal representation and highlights the potential of GMM-based belief propagation for probabilistic hybrid sensor data fusion with multimodal continuous evidence.

Neural Network-Based Gaussian Mixture Reduction

Luisa Hoffmann, Linus Fischer, Alper Saritas, Ali Darijani, Jürgen Beyerer, Michael Heizmann
Accepted to 2026 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI) • 2026
Accepted
Abstract
Gaussian mixtures (GMs) are a well-established tool for the compact representation and processing of complex probability distributions and are used, among other things, in sensor data fusion. When fusing multiple GMs, the number of components grows exponentially, increasing computational and memory requirements. Reduction methods are therefore necessary to replace a large GM with a compact approximation having significantly fewer components. This paper provides an overview of existing reduction methods as a benchmark. Furthermore, a novel approach using neural network (NN) for GM reduction is derived. This includes the training structure of the NN, the per-sample fine-tuning and a hybrid approach that combines the pretrained NN with a KMeans algorithm. To overcome the issue of a collapsed NN initialization the paper presents a dead-cluster recovery rule, which improves the Integrated Squared Error (ISE). Both the benchmark methods and the novel NN methods are compared in terms of their runtime and accuracy for various dimensions and cardinalities. The runtime of the NN with KMeans matches the runtime of the cheapest classical method but surpasses these methods in terms of accuracy.

Data-Driven Decision-Making: Leveraging Digital Twins for Reprocessing in the Circular Factory

Nehal Afifi, Victor Mas, Jonas Hemmerich, Felix Leitenberger, Luisa Hoffmann, Ali Darijani, , Sven Matthiesen
Zeitschrift für wirtschaftlichen Fabrikbetrieb (ZWF) • 2025
Published
Abstract
Abstract Circular factories must ensure the functionality and reliability of used components for recombination with other components or subsystems from the same or different product generations. This paper presents a data-driven decision-making framework integrating the Functional Behavior Model and System Reliability Model within a Digital Twin. Data from physical testing is continuously incorporated, simulating recombination scenarios and guiding decision-making on component reprocessing. An angle grinder is used as a case study for demonstration. The proposed framework enhances sustainability and supports the use of reprocessed components in products designed for primary markets.
DOI URL

Bringing the Vision of Perpetual Innovative Products to Life - challenges and opportunities at the circular factory for angle grinders

Patric Grauberger, Jonas Hemmerich, Nehal Afifi, Victor Mas, Dominik Koch, Edwin Blum, , Sven Matthiesen
2026 Conference on Circular Production Systems and Technologies (CCPT) • 2026
Accepted
Abstract
In the circular economy, the contradiction between developing innovative products and reusing products from the market is a huge challenge. On the one hand, products returned from the market are, by design, unable to meet the requirements and functions of the newest product generation, hindering innovation in circularity. On the other hand, retaining the product's added value is a powerful lever for enabling sustainable circularity. Perpetual innovative products could solve this contradiction. However, the problem is that up to now it remains mostly a vision. This vision needs to address three key challenges to come to life: analyzing used products and subsystems, predicting functionality at the individual level, and reprocessing subsystems in a way that enables them to fulfill the newest functions. In this paper, we illustrate the collaborative approach to address these challenges. We structure insights and the challenges encountered in various solutions for implementing this vision, using the example of a circular factory for angle grinders. From incorporating asymmetrical degradation into product specifications to measuring strategies and reprocessing parameters, this paper shows how a vision can come to life. The uncovered insights and difficulties reveal opportunities for further development and interdisciplinary research in circular factories.
PDF URL