Frank Liebau

Automatic anomaly detection in time series for industrial sensors, manufacturing processes, and technical systems. Ongoing measurement, sensor, and process data are continuously evaluated for deviations from expected normal operation. The result is a continuous anomaly score for trend assessment and for warning and alarm levels.

Robust statistical methods based on forecast errors, as well as deep learning models that learn patterns directly from raw signals and predict the expected temporal progression, are used. This allows anomalies in machines, plants, and processes to be detected early.

Details:

As part of a service contract with BAM, GWR GmbH developed an OpenFOAM-based simulation model for analyzing the climate in greenhouses. The model integrates the convection-diffusion equation for humidity transport and couples it with the flow field via a modified Boussinesq approximation. Furthermore, a solar model was implemented that calculates the sun's position based on geographical and temporal parameters and simulates its influence on the indoor temperature. The simulation considers transparent materials with specific transmission coefficients to realistically represent the greenhouse effect. The 'Age of Air' theory was applied to evaluate ventilation efficiency and analyze fresh air distribution within the greenhouse. The implementation was done in C++ using OpenFOAM, supplemented by Python scripts for data pre- and post-processing.

Development of numerical methods for surface-to-surface radiation models based on the radiosity equation. Focus areas include the discretization of integral equations, efficient handling of densely populated view factor matrices with hierarchical data structures and iterative solvers, and massively parallel visibility calculation on CPU/GPU using Embree and OptiX.