Engineering notes from inside plant operations.

Operational guide/6 min read

Green energy operations

Operating renewable and power assets with earlier causal visibility across the plant.

Operational guide/7 min read

Green hydrogen operations

Operational guidance for stacks, utilities, gas handling, and plant response.

Technical explainer/7 min read

Electrolyzer diagnostics

Earlier fault detection and root-cause signals before threshold alarms.

Technical explainer/7 min read

Physics-driven AI vs generic industrial analytics

Why models that respect plant physics behave differently from dashboard analytics.

Operational guide/8 min read

Electrolyser performance guarantees

How a performance guarantee is measured, where the argument usually starts, and what evidence settles it.

Operational guide/10 min read

Reading plant data safely

The acquisition paths, what each one asks of the control system, and how to answer the warranty question before you sign.

Operational guide/6 min read

Reporting to lenders after COD

What changes at commercial operation, what the lender's technical advisor actually looks at afterwards, and which of it your plant can evidence.

Selection guide/7 min read

Alkaline vs PEM vs AEM

Datasheet comparisons assume steady state. If the plant follows a solar or wind profile, four other properties decide the outcome.

Diagnostic guide/7 min read

Higher kWh/kg than the datasheet

Five causes, in the order worth checking. Only one of them is the stack.

Technology guide/7 min read

Digital twins for power plants

Three different things share the name. Only one of them answers an operating question, and most failed projects are a mismatch between which one was bought and which one was needed.

Operational guide/7 min read

Electrolyser predictive maintenance

The stack has no rotating parts, so most of what predictive maintenance means elsewhere does not transfer. What replaces it is electrochemical.

Technology guide/7 min read

Reducing false positives

Five sources of false alarms in industrial anomaly detection, and why precision matters more than recall for anything an operator has to read.

Operational guide/9 min read

Monitoring evidence a BESS insurer can price

Battery storage is underwritten largely on class assumptions. What continuous condition data changes in that conversation, and what it does not.

Operational guide/8 min read

Evidencing an availability guarantee

Availability disputes are almost never arithmetic. They are definitions applied to events after the fact, and the side with the record decides how the definition reads.

Operational guide/8 min read

RUL reporting for lenders

Remaining useful life has no meaning without a stated end-of-life criterion. How a measured trend becomes a replacement date, what that date does to the cost of production, and what a technical advisor will accept.

Operational guide/8 min read

Surviving a chain of custody audit

Certification audits fail on the measurement layer far more often than on the process. What an auditor samples, and which gaps are unrecoverable once the period has closed.

Technology guide/9 min read

Digital twin for an electrolyser

What a twin of an electrolyser actually models, why the geometric and statistical kinds do not help here, and what a worked architecture looks like.

Operational guide/7 min read

Start-stop cycling degradation

A plant following solar stops every night. What that costs in stack life, through which mechanisms, and how a cycle gets defined in the contract.

Diagnostic guide/7 min read

Membrane thinning in PEM stacks

Which signals move first when a membrane thins, why the order of arrival carries information, and why one signal never attributes the cause on its own.

Operational guide/7 min read

Does connecting to a DCS void the warranty

What supply-contract exclusions typically cover, which connection paths sit inside them, and why the site management-of-change procedure is a separate gate.

Technical explainer/7 min read

On-premise and air-gapped AI for OT

What air-gapped means for process data specifically, what an on-premise host removes, and the four things it does not.

Technical explainer/7 min read

Physics-informed prognostics

What physics-informed prognostics means in an operating plant, where the failure-time estimate comes from, and why the answer is a distribution rather than a date.

Technical explainer/7 min read

What a hybrid digital twin is

What a hybrid twin is, how residual learning splits the work between the physics and the data, and what it costs to build and keep calibrated.

Selection guide/8 min read

Build or buy industrial analytics

The four layers of an analytics stack, which of them a plant team can genuinely build, and why ownership in year three decides the answer.

Diagnostic guide/7 min read

Recoverable and non-recoverable degradation

Separating fouling you can wash off from deterioration you cannot, using corrected data the plant already has.

Operational guide/7 min read

HRSG performance degradation monitoring

What to monitor on a heat recovery steam generator in cycling duty, and what each signal separates once it has been normalised for load and ambient.

Diagnostic guide/8 min read

Hydrogen compressor failure under variable duty

What fails first when a compressor follows an electrolyser following the sun, and why an hours-based maintenance plan mistimes it.

Technical explainer/7 min read

Ammonia loop under variable hydrogen supply

What constrains a synthesis loop when its hydrogen follows a solar profile: the catalyst temperature window, inert accumulation and the buffer storage available.

Technical explainer/8 min read

AI and safety instrumented systems

Where machine learning can and cannot sit relative to a safety function, written from the standards rather than around them.

Technical explainer/7 min read

Operator decision support systems

What an operator needs at the moment of a decision, why alarm management is the precedent that matters, and how to tell whether anything is being done differently.

Operational guide/8 min read

HAZOP and LOPA for battery storage

How HAZOP and LOPA transfer to an asset whose natural nodes are electrical and thermal, and why each protection layer has to be credited scenario by scenario.

Technical explainer/8 min read

State of health from operational data

How state of health is estimated when the asset never stops to run a full characterisation cycle, and why the method has to travel with the number.

Operational guide/8 min read

BESS warranty compliance evidence

The owner-side view of a battery warranty: what the operating envelope conditions, how far cycle-counting definitions differ, and what evidence a claim needs.

Technical explainer/7 min read

Thermal runaway: prediction against detection

What is genuinely predictable, what is only detectable, and why blurring the two turns a marketing claim into a safety claim.

Operational guide/9 min read

CEA 2027 rules and BESS monitoring

What the notified 2026 amendment means for monitoring and instrumentation, which evidence a site has to keep, and which of it is cheap at construction.

Technical explainer/8 min read

Performance ratio against CUF

Two metrics quoted side by side that answer different questions, how curtailment distorts them differently, and which one a lender is asking for.

Reference guide/14 min read

Green hydrogen certification: RFNBO, 45V and GHCI

What the EU, US and Indian regimes each test, why a renewable electricity certificate does not carry a hydrogen claim, and which evidence has to exist before the first kilogram.

Operational guide/9 min read

What a control system does not show on an electrolyser plant

Five gaps between what a distributed control system reports and what an operating decision needs, what each looks like on a running plant, and what closing it requires.

Operational guide/11 min read

Pump condition monitoring from existing signals

A vibration threshold tells you a limit was crossed. It does not tell you whether the machine was faulty or answering a hydraulic load, and the start-up window where the answer usually sits is the part the alarm layer is configured to ignore.