Edge AI prediction — on the hardware you already have.
Our software runs on the microcontrollers, PLCs, and industrial PCs, working with time-series data from any sensor — physical, electrical, chemical, biological.
Entrox Systems is a German AI company, a spin-out from the DLR German Aerospace Center. We develop edge AI prediction software based on reservoir computing, bringing technology from the research lab to real-world industrial use.
Industrial sites generate diverse time-series data from sensors and equipment. Our AI, built on reservoir computing, learns patterns from even small amounts of time-series data, enabling real-time prediction of hard-to-measure values, anomaly detection, and remaining life forecasting.
Deployable as software on your existing equipment — from industrial PCs to microcontrollers. No special hardware required.
Runs on small, low-cost devices. No cloud connection required — your data stays on-site.
Builds accurate models from small datasets — no need to wait years to accumulate training data. Adapts immediately to new equipment and new products.
Inference runs in sub-millisecond time — fast enough to keep pace with high-speed machinery and tight control loops, from slow processes to vibration above 10 kHz. Models train in seconds, not months.
From a single proven core, each product, grade, and operating condition gets its own optimized model — so accuracy holds up across high-mix, low-volume production.
Model structure is fully visible as mathematical expressions. Prediction rationale can be traced and verified.
Combine existing physics equations or empirical formulas with AI that corrects the residual error.
Addressing the full spectrum of industrial prediction challenges.
Real-time estimation of hard-to-measure values from existing sensor signals
Replaces physical sensors, lab analyses, or measurements at hard-to-reach points.
Learns "normal behavior" of equipment and detects early signs of anomalies
Detecting early signs of degradation — trained from normal data only.
Predicts remaining useful life from sensor data to enable planned maintenance
Planning maintenance based on actual condition, not fixed schedules.
Visualize which variables, at which time lags, and through which nonlinear relationships affect your process — expressed as mathematical equations
Validating expert knowledge and supporting knowledge transfer with data.
Delivered as software that runs on your existing hardware. No extra devices, no cloud connection.
The trained model is delivered as code, typically C, Python, or similar, generated to suit your target and toolchain. We provide it as source files or as a pre-built library. With no operating-system dependency at the constrained end, the same model runs on bare-metal and RTOS-based microcontrollers and scales up through PLCs, edge gateways, and industrial PCs to Linux edge servers. No GPU, no accelerator, no added hardware.
For customers who want to retrain the model themselves, we can provide a desktop tool with a graphical interface. The retrained model exports in the same format as the original delivery.
AI, Mechanical Engineering, Business Development
AI, Algorithm Development
AI, Software Development
We optimize our edge AI prediction technology for the specific needs of our customers and partners.
Our founding research group is based at DLR, the German Aerospace Center, conducting international collaborative research.
Published in IEEE, Elsevier, Springer, and other leading journals.
An integrated development pipeline from research to real-world deployment.
We conduct small-scale pilot projects using your real data to demonstrate effectiveness in your actual environment. Ultimately, you get deployable AI software optimized for your specific process, running on your existing devices. Get in touch to discuss your use case.
Have questions? See our FAQ →