Jonas Nitzler

Max Diekmann, Jonas Nitzler, Jan Fischer et al.

Virtual commissioning models (VCMs) of discrete manufacturing systems are used to validate automation behavior before physical deployment, but creating and maintaining them remains labor-intensive. Relevant engineering information is distributed across programmable logic controller (PLC) engineering projects, such as Siemens TIA Portal, and kinematic simulation models, such as Siemens NX Mechatronics Concept Designer (NX MCD), where it is stored in incompatible, tool-specific data structures. In practice, IEC 61131-3-based PLC programs and variables are engineered separately from rigid-body and kinematic simulation objects such as parts, joints, sensors, and actuators. As a result, understanding system behavior, generating simulation components, and mapping PLC variables to corresponding simulation objects require cross-domain expertise and remain largely manual. This paper presents a knowledge-graph-grounded multi-agent framework for semi-automated VCM development. A deterministic setup process extracts structured data from Siemens TIA Portal and Siemens NX MCD and transforms both sources into graph-based representations within a shared graph database. The framework uses a hierarchical multi-agent architecture to support three task classes in early-stage VCM development: system understanding, simulation component generation, and cross-domain signal mapping. It provides grounded natural-language access to engineering knowledge, template-guided generation of executable NX Open journal scripts, and ranked mapping suggestions between PLC variables and NX MCD simulation objects. Evaluation on a laboratory-scale discrete manufacturing system shows that the approach reduces manual cross-domain interpretation effort and makes recurring VCM engineering tasks more actionable.

Jonas Nitzler, Maximilian Bergbauer, Stelios-Phaedon Koutsourelakis et al.

We present a unified, finite-element-native variational inference framework for very high-dimensional Bayesian spatial field reconstruction in physics-based problems governed by partial differential equations (PDEs) that are nonlinear in the inferred parameters. The framework delivers a full-covariance Gaussian variational posterior, with a probabilistic treatment of all prior and likelihood hyperparameters, on a three-dimensional curved finite-element discretization at a stochastic field dimension exceeding 400000. To our knowledge, this is the first full-covariance variational reconstruction at this scale, complementing the low-rank Hessian-Laplace approaches that dominate extreme-scale Bayesian inversion. The spatial prior is derived from the stochastic PDE (SPDE) connection and formulated natively in terms of finite-element (FE) operators. The sparse Gaussian variational distribution is parameterized via its precision Cholesky factor, with the sparsity pattern inherited from the domain's Laplacian. Unlike covariance-based sparse parameterizations, which encode only short-range correlations, the sparse precision implicitly represents dense posterior covariances through its sparse inverse, yielding smooth, physically plausible samples at O(n) memory cost and enabling direct evidence-lower-bound (ELBO) gradients via the path-derivative (sticking-the-landing) estimator. Natural gradient strategies stabilize convergence, while a variational Bayes expectation-maximization (VB-EM) loop marginalizes all hyperparameters analytically and induces an automatic coarse-to-fine continuation. The framework is demonstrated on Bayesian permeability field reconstruction for a porous-media flow problem, recovering all major spatial features with high fidelity. Algorithmic ablation and comparison with alternative inference methods quantify the improvements over state-of-the-art baselines.