AI in New Energy: The Test Rigs and Analysers That Feed the Models

New energy technologies are young enough that their operating data are still being written. Electrolysers, battery recycling flowsheets, carbon storage hubs and liquid-cooled AI halls do not yet have decades of plant history to train on. Whoever generates clean, representative data now will build the models the industry relies on later.

That data comes from test rigs, pilot lines and online analysers. Here are five areas where AI is gaining ground, and where Tech Inc. equipment produces the signals each one needs.

1. Electrolyser and fuel-cell test analytics

Performance degradation limits the commercial uptake of PEM electrolysers, and because it plays out across different time and length scales, it is hard to model from first principles. A study in Energy and AI therefore used a CNN-LSTM model to predict PEM electrolyser voltage from durability-test data: 1,140 hours under constant load and 660 hours under start-stop operation. It reached R² above 0.98 on both datasets and estimated degradation of 4.5% after 1,000 hours of constant operation and 2.5% under start-stop. The model is a direct product of long, uninterrupted test records.

Where Tech Inc. fits. Our electrolyser and fuel cell system provides a precision DC supply with voltage and current control, a temperature-controlled cell housing, interchangeable membrane electrode assemblies and real-time monitoring of cell voltage, current density and hydrogen and oxygen production. Run it for durability and accelerated stress tests and you produce exactly the voltage-versus-time series these models are trained on.

2. Battery recycling process optimisation

Hydrometallurgy is a natural fit for soft sensors. A 2025 paper in Green Chemistry predicted the leaching state of black mass from online sensor data: pH, conductivity and temperature were enough to estimate how much aluminium, iron, nickel, cobalt, lithium and manganese had dissolved, with an average adjusted R² of 0.995 and an average prediction error of 1.1%. The authors worked from 46 data points, said that machine learning would depend on larger datasets, and noted that manual sampling introduced inconsistency.

Where Tech Inc. fits. Our battery recycling hydrometallurgy pilot line runs leach, filtration, purification, solvent extraction and crystallisation with PLC and SCADA, trending and batch data logging. The leach and filtration train has temperature-controlled agitated reactors with acid and reductant dosing, and the mixer-settler solvent extraction train recovers nickel, cobalt and lithium as separate streams through counter-current extraction, scrubbing and stripping. Specify pH, conductivity and temperature on each reactor with timed sample points, and the pilot campaign doubles as the training set for the commercial plant.

3. CCUS monitoring

On the storage side, the US Department of Energy's SMART initiative has evaluated machine learning for rapid CO2 and pressure plume forecasting, and for producing CO2 saturation images with uncertainty quantified in real time. Those subsurface models assume the injected stream is what the specification says it is. An IEAGHG 2025 insight paper on CO2 specifications for transport and storage networks names SOx, H2S, O2, H2O and NOx as the impurities to manage to avoid highly corrosive acid mixtures, and notes that specifications remain project-specific.

Where Tech Inc. fits. Our CO2 quality analyser system measures water, oxygen and H2S at the tie-in, with sample conditioning, validation and calibration gas, and signals to the plant DCS or SCADA. It keeps a continuous record of CO2 quality at the point of entry. Our CO2 dehydration unit runs its tower switching and regeneration from the PLC, so moisture performance and regeneration cycles sit in the same data record.

4. AI data centres and liquid cooling

The IEA's Key Questions on Energy and AI expects data-centre electricity use to roughly double from 485 TWh in 2025 to 950 TWh in 2030, reports that consumption by AI-focused data centres grew 50% in 2025, and notes that AI data centres must remove heat equivalent to 30 natural gas boilers per server rack.

HTRI cites benchmarks showing that 48% of facilities with adequate water resources have adopted direct-to-chip cooling, and explains why those loops are unforgiving: the cold plates are spaced only hundreds of microns apart with deliberately laminar flow, making them especially sensitive to deposits. Microbial films can be 10 to 100% more thermally resistant than calcium carbonate, biofilm formation accelerates below about 17% propylene glycol, and because many data centres do not routinely monitor water quality, biofilm can develop undetected. A cooling optimisation model cannot see what nobody measures.

Where Tech Inc. fits. Our coolant treatment skid runs a side stream through fine filtration with differential-pressure monitoring, doses inhibitor and glycol, and measures glycol concentration and turbidity inline alongside pressure, flow and temperature, with a PLC link to the BMS or SCADA. With our flush-and-fill rig, whether supplied as equipment or run by our crews, every loop is handed over with its flushing and fill records. Together they give the operator a coolant history from commissioning onwards.

5. SF6 and grid asset monitoring

A paper presented at the CIGRE 2025 International Symposium describes machine-learning prediction of fleetwide SF6 escape that combines historical maintenance and gas top-up records, limited online gas density monitoring and engineering knowledge, then applies what it learns to unmonitored assets. Regulation is raising the stakes. The Finnish Environment Institute's summary of the EU F-gas rules notes that switchgear holding at least 500 tonnes CO2 equivalent of F-gas needs a leak detection system, that medium-voltage switchgear up to 24 kV may not contain F-gases from 2026, and that new SF6 may not be used for servicing from 2035.

Where Tech Inc. fits. Our SF6 gas handling cart and compact SF6 recovery and filling unit recover, filter, dry, store and refill SF6 with no venting, under pressure switches and control sequences. Every recovery and top-up is a data point on how much gas a compartment lost. Specify weighed recovery and a logged service record, and each maintenance visit feeds the fleet model.

How Tech Inc. makes your new-energy data AI-ready

• Instrumented from the start. Voltage, current, pressure, flow, temperature, pH, conductivity, glycol concentration, turbidity and gas composition measured where the process happens.

• Logged and timestamped. PLC and SCADA trending, batch data logging and alarm histories on pilot lines, coolant skids and analyser systems.

• Connected. Signals to your DCS, SCADA or BMS, ready for the historian your data team uses.

• Calibrated and validated. Validation and calibration gas on analyser systems, and sample points for laboratory checks.

• Scale-up ready. Pilot data structured so it carries through to the design of a commercial plant.

Talk to us about your data

Whether you are testing membrane electrode assemblies, piloting a recycling flowsheet or protecting a liquid-cooled hall, tell us what your model needs to see. We will build the rig, pilot line or analyser system that records it. Email mail@techincresearch.com or call +1 (289) 622-1032.