One Thread, Multiple LayersOne Thread, Multiple LayersOne Thread, Multiple Layers

In an automotive research project with BMW and KIT, my task was exactly that: developing onboard-capable algorithms that infer the state and remaining lifetime of high-voltage batteries directly during vehicle operation. The challenge was not only algorithmic, but fundamentally engineering-driven: a model that looks good in the lab is of little value if it cannot run reliably on the hardware.

The core challenge was not merely to build a more accurate model, but to design machine-learning and deep-learning time-series models that remain practically usable under BMS-related and edge-AI constraints. A particularly formative part was the optimization of binarized BiLSTM hybrid models under real resource constraints.

In parallel, I built a GUI-based training and evaluation framework for model training, hyperparameter tuning, real-time visualization, and data analysis. The solution was validated in cloud/edge-oriented scenarios using AWS SageMaker and Raspberry Pi.

This work deeply shaped my understanding of end-to-end ML in industrial systems: a model that runs in a notebook is not yet a solution. A model that runs reliably under resource constraints is engineering.

In the functional validation of powertrain functions, my role was not merely data analysis, but work at the intersection of HiL simulation, virtual driving cycles, functional validation, and test automation. Engine-control and powertrain functions had to go through virtual driving cycles, boundary conditions, failure scenarios, and load cases in a hardware-in-the-loop environment before they could reach the real vehicle.

The key in this type of validation is not simply to “run tests,” but to understand the relationship between real control units, simulation environments, test cases, measurement signals, and vehicle-function logic: which signals indicate normal behavior, which deviations suggest potential faults, and which test scenarios are capable of exposing boundary problems.

On top of this validation workflow, I developed Python and VBA tools that transformed manual test-data inspection into automatically generated, structured reports. These scripts were not isolated data-processing utilities; they were engineering support tools for the HiL validation process. They helped identify error patterns earlier, improve consistency in test evaluation, and make validation results more traceable.

This experience gave me a concrete understanding of the engineering value of automotive functional validation: virtual testing is not paper simulation, but a reproducible, diagnosable, decision-oriented validation system between real hardware, control logic, and simulated driving conditions. Good automation does not replace test engineers — it helps them see the real system behavior faster and more clearly.

In a scientific supply-chain-management context at FAU, I developed data-driven evaluation processes for strategic KPIs from industrial supply and value chains. In addition to classical statistical analysis, I applied structural equation modeling with R, Python, and SPSS/AMOS to examine interdependencies, moderation effects, and interaction effects in a methodologically grounded way.

The core of this work was not merely KPI reporting, but the methodological transition from processed data to reliable insights: heterogeneous data, scales, and hypotheses were translated into reproducible evidence.

The decisive point was precise, traceable, and statistically grounded validation of interdependencies — as a foundation for better decisions rather than mere data description.

At the KIT Institute of Measurement and Control Systems, I supported students in lab courses on mechatronics, automation, computer vision, and robotics. The work was not only about code or isolated experiments, but about translating theoretical models into sensor-based experiments on real prototypes.

I developed test scripts and experimental environments, maintained MATLAB/Simulink models, and automated sensor data evaluation and fault diagnosis through scripting. The work moved between digital image processing, automatic visual inspection, robotic control, mobile platforms, and real measurement technology.

The decisive aspect was the transition from theoretical model logic to experimental robustness: a model becomes truly viable only when it not only works conceptually, but also performs reproducibly under measurement noise, varying sensor conditions, practical time constraints, and real hardware.

My early roles in series production and mechanical-engineering companies shaped an important foundation of my engineering profile: I do not understand industrial systems only through theory, simulation, or data, but also through direct exposure to production environments — assembly, manufacturing, production rhythm, process flows, process control, and quality assurance. On the production floor, technology is no longer just formulas, drawings, or models. It becomes a real system of components, tools, operations, cycle times, deviations, rework, and quality decisions.

This hands-on experience taught me early that behind every data point there are real materials, machines, processes, tolerances, wear, and human operational decisions. Whether a production process is stable, why a quality issue occurs, or whether a model assumption holds often cannot be answered by paper reasoning alone. It requires going back to the site, back to the system itself.

At the same time, my exposure to heavy machinery and bearing-related engineering introduced me to FEM/BEM simulation, elastohydrodynamics, reliability analysis, and service-life analysis. These topics gave me another perspective on the physical depth of mechanical systems: load, contact, deformation, lubrication, fatigue, and failure are not abstract concepts, but the mechanisms that determine whether an engineering system can operate reliably over time.

Together, these experiences shaped my technical judgment: hands-on capability, production-floor awareness, and process understanding on one side; physical depth from numerical simulation and mechanical-system analysis on the other. For me, this was not merely an early career phase, but the starting point of my engineering mindset.