What to Measure Before You Electrify the Boiler

Behind a steel door most of the plant walks past, a gas train hisses. Inside sits a fire-tube boiler the size of a delivery van, turning natural gas into steam at ten bar, and one custody-transfer meter at the property line counts the fuel going in. That meter is the number the boiler house reports with real confidence. It's also the wrong number to start from.

Ask around and someone will quote the plant's annual gas spend to the euro, and the header pressure to the decimal. Ask where that heat goes once it leaves the header, though, which use, at what temperature, at which hour of which shift, and the room goes quiet. Steam fans out through a manifold into a dozen destinations, and almost nobody has metered them apart. The gap is harmless while gas is cheap and the boiler just runs. It stops being harmless the moment someone asks whether the boiler should be electric.

On 17 July 2026 the European Commission published its Electrification Action Plan, with an indicative target of 46% of final energy from electricity by 2040 and an estimate that reaching it would trim the EU's fossil-fuel import bill by €260 billion a year. For anyone who runs a boiler house, all of that policy compresses into one practical question: when does the gas boiler become a heat pump? The honest answer isn't a date on a capital plan. It's a measurement campaign. The case for electrifying any given duty stands or falls on two quantities the boiler house has never bothered to log: the temperature the process genuinely needs, and the shape of the load across the day.

Neither is written on the boiler nameplate. What follows is the data you gather before you spec a single kilowatt of heat pump, why the instrumentation on a typical boiler house can't see it yet, and how those measurements sort your heat into the duties that electrify now and the ones that wait.

Where the heat actually goes

A heat pump doesn't make heat so much as move it, and the cost of moving it rises with the lift: the gap between the temperature it draws from and the temperature it delivers. Pull warmth from a 30°C effluent stream and deliver it at 70°C, and a good industrial unit returns three to four units of heat for every unit of electricity it draws. Ask that same machine to reach past boiling and the coefficient of performance sags toward the point where the sums stop working. So the single most useful thing you can know before electrifying isn't how much heat you use. It's the temperature you actually need it at, duty by duty.

The boiler makes one grade of steam, and that single high setpoint is a disguise. It runs hot because it has to satisfy the hottest thing on site, and everything downstream throttles it down. The header feeds a pasteuriser that needs a fraction of that grade. It heats wash water. It keeps a tank farm warm. It runs space heating through the winter. Each of those is a distinct duty at a distinct temperature, and the boiler's one setpoint hides every one of them. Read at the header, the plant looks like it needs 180°C steam. Read at the point of use, most of the load sits well below boiling.

Disaggregating that spread is the first field exercise, and it's mostly a temperature-logging job: a resistance thermometer on the supply and return of each header branch, sampled often enough to catch the duty cycling on and off, trended long enough to cover a full production week. What comes back is a histogram of demand against temperature. On a food or beverage site it usually leans hard toward the low end, and that lean is the whole opportunity.

Read the histogram and the strategy half-writes itself. The duties clustered under boiling are the ones a heat pump serves at a coefficient of performance that pays. The high-temperature tail, the reheat furnace on a rolling line or a calciner, is a different machine's problem, and for now often a different fuel's. The Commission's own plan puts the reach of industrial heat pumps and electric boilers at 400-500°C, which drags a good deal of mid-grade drying and concentration duty into range that wasn't there a few years back. The economics still favour the bottom of the spread, where the lift is gentle and the COP is high. But the reachable envelope keeps climbing, and a duty that's marginal this year may not be next.

One caution on reading temperatures off the header. The setpoint isn't the delivered temperature; distribution losses, flash from condensate, and a control margin all sit between the boiler and the process. A duty logged as needing steam may in truth need eighty-something degrees at the jacket wall, and that difference decides whether a heat pump clears it. Measure at the wall, not at the boiler, or the histogram lies to you in the direction that kills good projects.

The peak the boiler never charged you for

Swap the fuel and you swap the billing physics with it. Natural gas is priced on volume: burn what you burn, pay for the therms, and the boiler couldn't care less whether you drew them in a smooth line or a burst of spikes. Electricity is priced on two axes at once. There's the energy you consume, in kilowatt-hours, and there's the demand you impose, the peak power in kilowatts the network has to stand ready to deliver. And that second axis is the one that ambushes electrification projects. A gas boiler serving a spiky load simply pays for the gas. An electric system serving the same spiky load pays for the energy and then pays again, month after month, for the highest fifteen minutes it ever hit.

Which is why the second dataset isn't a total, it's a profile. You need the load resolved at the interval the utility settles on, which across most of the EU is the quarter-hour. Averages bury the peaks that cost you. A plant that burns a steady baseline all day and then fires every CIP set and every batch heat-up into the same afternoon window has a demand problem no annual figure will ever surface. The ratio you're hunting is the load factor: average demand over peak demand. A high load factor, a flat and well-spread profile, makes electric heat cheap. A low one, spiky and all at once, makes the effective price per delivered kilowatt-hour brutal, because the demand charge smears across too little energy.

How long do you log? Longer than feels necessary. A production week captures the batch rhythm but misses the season. A dairy's heat load in July looks nothing like its load in January, and a plant that sizes on a summer month will find the same unit gasping on the first hard frost, when it has to pull heat from colder ambient air and lift it further for less return. Twelve months of quarter-hour data is the honest baseline. If that history doesn't exist, and on most sites it doesn't, the campaign starts now and the decision waits for it, because sizing a seven-figure asset on a fortnight of readings is how you end up with a machine that's wrong in both directions: too small for the winter peak, too big for the summer base, badly loaded all year.

There's a second reading buried in the same interval data, and it's the one that turns a demand problem into a design lever. Look at when the peaks land, not just how tall they are. If the CIP surge and the morning batch stack onto the same half-hour by habit rather than necessity, some of that peak is schedule, not physics. Spread the draws, or buffer them, and the peak you have to build electric capacity for shrinks before you've bought a single machine.

Sizing for the sweet spot, not the peak

With the temperature histogram and the load profile side by side, a sizing strategy appears that isn't "match the boiler." Rank every quarter-hour of the year by how much heat the site drew, highest to lowest, and you get a load-duration curve: a line that starts at the rare screaming peak on the left and falls to the long, low base load running almost every hour. The shape of that curve is the whole argument against electrifying everything at once.

But the peak is expensive and brief. The last slice of capacity (the coldest morning, the simultaneous CIP surge) may show up only a few dozen hours a year, yet sizing a heat pump to cover it means buying a much larger machine and running it at part-load the rest of the time. The base is the opposite: unglamorous, nearly constant, cheap to serve at a high COP. So the money case usually points to a hybrid. A heat pump sized to the base and the shoulder, carrying the bulk of the annual kilowatt-hours, paired with a booster (an electric boiler, or for a while yet a retained gas boiler) that only wakes for the peak.

Where you put that split is the design decision, and only the two datasets together can place it. A unit rated for roughly half the site's peak heat duty will, on a well-spread load, still carry the large majority of the year's kilowatt-hours, because the top of the duration curve is so thinly populated with hours. Undersize it and you lean on the expensive booster too often. Oversize it and you've spent capital on capacity that idles, and saddled the plant with a machine that spends its life cycling at part-load, where both efficiency and reliability suffer. The curve settles that trade-off. A rule of thumb won't.

Thermal storage is the piece that quietly rewrites the sizing sum. A hot-water buffer tank, even a modest one, decouples the heat pump's steady output from the process's spiky demand: the pump charges the store through the quiet hours and the store rides out the surge. That does two things at once. It lets a smaller, steadier heat pump serve a peakier load, and it flattens the electrical draw, lifting the load factor that sets your effective price. Storage is cheaper per unit of capacity than heat-pump compressor, so on a peaky site the buffer tank often earns its place before the last increment of heat-pump does. The interval data tells you how big a store you'd need. Without the profile, you're guessing at a tank size, and a guessed tank is either useless or a monument.

The instruments the boiler house never had

All of this assumes data the boiler house doesn't produce, and that's the uncomfortable part. The gas meter at the boundary is revenue-grade and trustworthy; almost nothing downstream of it is. Steam flow, where it's metered at all, is often read off an orifice plate that's been drifting for a decade. Condensate return goes unmeasured. The header branches carry no flow instrumentation. Temperatures show on local dial gauges no historian ever sees. You can't profile what you don't sample, so before any analysis there's an instrumentation campaign, and it's more than clipping on a single meter.

What it takes, in practice:

• Clamp-on ultrasonic flow meters on the header branches and, where you can reach it, the condensate return. They're non-invasive, so no shutdown to fit them.
• Resistance thermometers or thermocouples on the supply and return of each duty, because a flow reading without its temperature difference tells you nothing about the heat actually delivered.
• A revenue-grade power analyser on the main incomer, logging real power finely enough to reconstruct the quarter-hour peaks the tariff bills against.
• Gas sub-metering per header, not just the one custody meter at the fence, so each duty's fuel input ties back to its heat output.

Sampling rates matter as much as sensor placement, and the instinct to log everything once a second forever is the wrong one. Temperature and flow on a batch duty want a sample every few seconds to a minute, fast enough to catch a CIP cycle heating and dumping, slow enough that you're not drowning in a year of sub-second noise. Electrical demand is the exception. To see the peak the way the meter does, log real power at least once a second and aggregate up to the quarter-hour the tariff settles on. Energy totals reconcile daily against the utility bill. Get the rates wrong in either direction and you either miss the transient that matters or bury the historian in data nobody reads.

Then it all has to land somewhere aligned. A pile of CSV files from four loggers with four different clocks is worse than useless. A flow reading and its temperature reading have to share a timestamp to the second, or the heat calculation built on them is fiction. This is the unglamorous spine of the whole exercise: an edge layer that timestamps every stream on one clock, does the flow-and-delta-T heat maths at the source, and streams tidy, reconciled energy-by-duty into the historian instead of raw pulses. Get that right and the temperature histogram and the load-duration curve fall out of a query. Get it wrong and you've built a warehouse of numbers no one trusts.

Data quality is not a footnote here; it's the deliverable. An ultrasonic meter on a badly chosen straight run, an RTD reading the pipe clamp instead of the fluid, a power channel scaled to the wrong CT ratio: any one of them poisons the analysis quietly, and you won't notice until a heat balance refuses to close. Cross-check the streams against each other. The sub-metered gas should sum to the custody meter; the heat delivered should reconcile against the fuel burned and the boiler's efficiency. When those don't tie out, trust the physics and go find the bad sensor.

The spark gap, honestly

Strip the project to its frame and the go/no-go on any single duty is one comparison. What does a delivered unit of heat cost from the boiler, and what does it cost from the heat pump? The fuel side is the gas price over the boiler's efficiency: you burn a bit more than a unit of gas to land a unit of heat in the process. The electric side is the electricity price over the COP: you draw well under a unit of electricity to deliver that same unit of heat, because the pump multiplies it. Electrify the duty when the second figure drops below the first. That's the spark gap, and it's the sum that finally decides.

Two quantities move that comparison, and you just measured both. The COP is set by the temperature lift, so the low-grade duties clear the bar first: a wash-water loop at a gentle lift might run a COP of four, and four units of heat for one of electricity is enough to beat gas even though electricity is the pricier fuel per unit of energy, as it remains across most of Europe. The effective electricity price is set by the load factor, so the same duty on a flat, well-behaved load clears more easily than on a spiky one, where demand charges inflate every kilowatt-hour. That's the reason to run the sum per duty, not per plant.

A plant doesn't electrify. Its heat duties do, one at a time, each when its own coefficient of performance and its own load factor say the number works.

One more correction the data has to carry: normalise it against what the plant actually made. Raw energy totals rise and fall with production, so a slow month flatters a duty and a flat-out month damns it, and neither tells you anything about the heat itself. Divide the heat by the tonnes of product and you get specific energy consumption, the same per-tonne yardstick we used in an earlier piece on benchmarking energy use. Cast the electrification case in those terms and seasonal swings in throughput stop masquerading as swings in efficiency. If the site runs an energy management system to ISO 50001, this is the energy review and the baseline it already asks for, pointed at one capital decision instead of a yearly audit.

None of this makes the decision for you, and it shouldn't. What it does is replace a hunch with a curve. The plant that measures first knows which duties clear the spark gap today, which sit a tariff round away, and which stay on fuel for the foreseeable future. The plant that guesses buys one big heat pump, bolts it onto the highest-temperature duty because that's the loudest steam line, watches the COP disappoint, and concludes electrification doesn't work here. Both spent money. Only one spent it on the right duty.

Notes

This is a duty-by-duty method, not a universal verdict. A genuinely high-temperature process (metals reheat, cement, glass) is a different electrification problem, closer to induction, resistance, or plasma than to a heat pump, and nothing here settles it. The electricity-to-gas price ratio that sets the spark gap varies by country, contract, and year, so a duty that doesn't clear today may clear after the next tariff round or a carbon-price move; re-run the sum, don't file it once. The 46% figure is the Commission's indicative target, not an obligation on any single site. And the COP values used above illustrate the temperature-lift relationship rather than specifying any machine; take real numbers from the manufacturer's performance curve at your actual source and sink temperatures, not from a blog.