Deduce Data Solutions is a scientific SME based in Santander, Spain, spun off from the University of Cantabria in 2020. Our team of PhDs in applied AI develops models for industrial processes that embed the physics and the causal structure of the process into the model, working on the data the plant already records in its SCADA, MES and ERP.

What this changes for a plant: fewer data needed to train a reliable model, predictions that process engineers can interpret and act on, and models that stay valid when operating conditions change. Typical applications: energy consumption forecasting and reduction in furnaces and thermal processes, soft sensors for variables that are hard to measure online, early fault detection and process optimisation.

Techniques: physics-informed neural networks (PINNs), causal analysis, neuro-symbolic AI, knowledge graphs and explainable AI (XAI). We integrate on top of existing systems, deliver source code and train the in-house team, with no vendor lock-in.

Sectors we have worked in: steel and forging, paper, cement, chemicals, automotive, renewables, power distribution and discrete and continuous manufacturing. We are open to industrial pilots, open innovation challenges and R&D partnerships with corporates, research groups and technology providers.

We work with industrial companies and research groups in three ways.

1. Industrial pilots. A scoped project on one process (a furnace, a kiln, a production line) using the historical data the plant already has. The pilot ends with a working model, its explanation and a measured result on the variable the plant cares about (energy, scrap, downtime, quality).

2. Open innovation challenges. We answer technology calls from corporates where the need is to predict, explain or optimise a physical process with limited or noisy data. Physics-informed and causal models are built for exactly that situation.

3. R&D partnerships. We take part in collaborative R&D projects (national and European calls) as the applied AI partner, bringing physics-informed neural networks, causal inference, neuro-symbolic AI and explainable AI, and our experience deploying them in production plants.

In all three cases we build on the client's existing SCADA, MES and ERP, hand over the source code and train the in-house team.