AI Strategy & Transformation: building data and AI capabilities strategically
From maturity assessment to a prioritised AI roadmap – with governance under the EU AI Act and ISO/IEC 42001.
Use Cases
Success stories from AI projects
AI-supported test automation in automotive software development: Testomat
Test case proposals, regression selection, defect classification – five testing tasks, one assistant.
Anomaly detection on sensor data
Detects anomalies in a Deep Learning Development Server cluster from sensor data.
Instance segmentation on 3D point clouds: single objects from scenes of up to ten million points
Separates single objects in point clouds of up to ten million points – up to 93% accuracy per class.
A field trial analysis pipeline: from query to dashboard
Consolidates field trial data, models it and makes it analysable in dashboards – without a line of code.
Classifying breast tumours: cost-sensitive thresholds
Internal methods study on open datasets: random forest against logistic regression, cost-sensitive threshold.
Clustering microbiome sequences: 90% cluster homogeneity
Groups microbiome sequences by nucleotide order, separates host from bacterium and names the organisms.
Drone-based crop damage assessment: measured, not estimated
Defensible figures for a 9.43-hectare maize field from drone data: damaged area, unused area, plant height and stand density.
Assessing the environmental impact of crop protection products
Assesses the environmental impact of crop protection products using recognised models – public calculator included.
Estimating crowd density instead of detecting people: 330,165 training examples
A density-based model estimates crowd counts – without identifying individuals.
Data foundation and operations for research data
FAIR-ready structures for laboratory and research data – the foundation that makes models viable.
Mapping bark beetle infestation on aerial imagery
A neural network detects dead trees on aerial imagery from the Harz and surfaces infestation hotspots as heatmaps.
CO2 emissions from satellite data: a verifiable alternative to national reporting
A space-time model quantifies CO2 emissions from satellite data – independent of national reporting.
Automated engineering documentation: auditable drafts instead of writing work
Test reports and ASPICE documents as auditable drafts – every statement with a source reference.
Engineering Knowledge Assistant: the EKA AI assistant on your development knowledge
An AI assistant on your development knowledge – answers with sources from Confluence, DOORS and Jira.
Explainable AI for field trials
A framework that makes machine learning models in field trial analysis interpretable, analytically and graphically.
Failure prediction in innovative vehicle projects
The aim of the project was to develop an algorithm that identifies fault patterns and configurations that lead to vehicle failures.
A global platform for weed control trials
A central platform that makes weed control trials from several countries comparable, and assesses cost and efficacy.
Making a graph database usable for machine learning
A Neo4j graph reconciled completely against its SQL source and made usable for dynamic model training via a BOLT interface.
Image-based analysis in screening and diagnostics
Segmentation and quantification on microscopy and histology images – as a reproducible pipeline.
Detecting concrete cracks to the pixel: 97% true positive rate
Detects concrete cracks to the pixel – 97% true positive rate – and makes width, length and change measurable.
Industrial anomaly detection in manufacturing
Approaches to anomaly detection in manufacturing benchmarked – which models run leanest.
The Kebab Challenge: spatial optimisation of sales networks
Spatial optimisation finds the two kebab shop locations in Berlin that reach a 30% market share.
ResiYou: AI-driven residue management for growers
AI platform for Bayer Crop Science: predicts residue profiles at harvest, checks spray plans.
Predictive maintenance in pneumatic systems
The goal of this project was to analyse sensor data and predict whether the pneumatic system is likely to fail.
Preclinical pipeline decisions: prioritising candidates before the expensive phases begin
Predictive models for preclinical go/no-go decisions – with an uncertainty interval, not a point estimate.
Predictive maintenance and lifetime prediction: forecasting failures before they happen
Condition monitoring for equipment, lifetime prediction for components – two separately bookable modules.
Predicting road condition: four condition variables from three survey years and traffic load
An LSTM-CNN predicts four condition variables of the A70 from three survey years and traffic density.
Product Finder: expiring patents for a generics manufacturer
Surfaces expiring drug patents, filters compound candidates and rates their market.
Mapping photovoltaic modules from orthophotos: 47,818 images, one shapefile for all of NRW
A Mask R-CNN detects PV modules on open orthophotos and produces a shapefile with location and shape for NRW.
Root cause analysis in quality management: process data analytics and agentic AI
Process data analytics plus research in past cases – substantiated root cause hypotheses per quality case.
Automated requirements engineering: reviewing requirements with the RE Assistant
Automated requirements assessment – testability, clarity, conflicts, with a justification per finding.
Surrogate models: simulation results in seconds instead of hours
ML models predict FEM and CFD results in seconds – with uncertainty and a clear validity range.
Predicting test bench times for automotive configurations
In this use case, we used machine learning to develop predictive logic for the test bench times of varying automotive configurations.
An automated forest inventory: single trees from the point cloud, whatever the scan type
LiDAR point clouds become single trees with position, height and crown extent – usable as map layers.
Detecting trucks in orthophotos and telling vehicle types apart: over 99% detection rate
A deep learning model detects trucks in orthophotos and tells vehicle types apart – over 99% detected.
From point cloud to CAD model: winner of a Geospatial World Excellence Award
CAD models emerge automatically from LiDAR highway scans – Geospatial World Excellence Award 2021.
Classifying point clouds from highway scans automatically: from weeks of manual work to hours
A 3D neural network classifies point clouds from highway scans in hours instead of weeks of manual work.
Visual quality inspection with computer vision: inspect every part, document every decision
Camera-based inspection of every part – anomaly detection for rare defects, documented and in cycle time.
A wind turbine register from satellite imagery: 3,300 turbines from 280,000 image tiles
A U-Net processes 280,000 satellite image tiles and maps 3,300 wind turbines with location and type.
Yield prediction for hybrid cereals at new locations
A model predicts how hybrid cereals perform at 20,000 new locations from genetic markers, soil and weather data.
No use cases found
No use cases are currently available for this industry.
Have a Similar Project?
Let's discuss how we can solve your data challenges together.