Every water treatment program is designed for a single moment in time. From day one, everything else changes — seasonality, process conditions, equipment aging, instrument drift, and the slow divergence of operating reality from commissioning assumptions. Here is what that costs, and where the industry is heading.
When I started designing programs for refineries, chemical plants, and power plants, the robust technology didn’t exist to manage a water treatment system in near real-time and respond to changing conditions proactively — let alone predictively. The goal was to “design for the expected extremes” and then discuss what risk the customer was willing to accept to reduce cost.
That approach shifted over the past 15 years to something more troubling: “justify the lowest cost approach based on the least-stressed operating conditions — just to win the bid.”
Either way, you’re still designing for a single operating condition — a Design Basis — and not the changing, everyday conditions the system actually experiences.
I’ve sat in enough post-mortems to know how this story usually ends. A scaled exchanger. A biofilm event that took a cell out of service. A corrosion coupon that came back looking nothing like the model predicted it should. And in almost every case, when we went back and pulled the original design basis, we found the same thing: a document that was accurate on the day it was written, and increasingly wrong every day after that.
A design basis is, by definition, a single point in time. In a cooling system, it captures makeup water chemistry, cooling load, cycles of concentration, flow rates, and metallurgy as they existed — or as they were projected to exist — at commissioning. It is the best available estimate of one moment. That’s not a flaw in the engineering. That’s just what a design basis is.
The problem is what happens next, which is that nothing stays still. The system that gets commissioned on day one is not the system running on day 400, and it is certainly not the system running on day 4,000. Water chemistry drifts with the seasons. Process conditions shift with production demands. Heat exchangers foul. Instruments drift out of calibration. Operators make reasonable, defensible decisions in the moment that nonetheless move the system further from where the original model assumed it would sit.
None of this is a surprise to anyone who has run a plant. What’s surprising is how many treatment programs are still built — and left — as if the design basis were a permanent operating condition rather than a starting point. We build a beautiful, defensible model to win the job or pass the commissioning review, and then we treat it as though it were self-maintaining. It isn’t. A design basis doesn’t age gracefully. It just ages, quietly, in the background, while everyone assumes the program is still doing what it was designed to do.
Systems are not snapshots. They’re movies. And most treatment programs are still working from the first frame.
Start with seasonality, because it’s the most predictable driver of change and still routinely underestimated. A surface water source in January is a different raw water source than the same intake in July. Temperature swings change scaling and corrosion kinetics directly. Alkalinity, silica, calcium hardness, and total dissolved solids all move with seasonal runoff, snowmelt, drought conditions, and reservoir turnover. A treatment program tuned to a winter makeup water profile can be meaningfully undertreating — or overtreating — by mid-summer. The chemistry didn’t fail. The assumption that the chemistry would hold steady did.
Then there are process changes, which tend to be larger and less predictable than seasonal drift. In a refinery, a crude slate shift changes sulfur content, heat loads, and the contaminant profile that ultimately finds its way into the cooling water system. A heavier, higher-sulfur crude changes exchanger duty and corrosion risk in ways that ripple straight through to the cooling tower basin. In chemical plants, a product mix change can alter cooling demand and the thermal and chemical load on the water system just as dramatically. Nobody redesigns the water treatment program every time the plant changes what it’s running. But the water system doesn’t know that. It just responds to whatever is actually flowing through it.
Turnaround intervals compound the problem further. Where five-year turnarounds used to be common, many operators are now pushing seven to ten years between major outages — a defensible economic decision on its own terms, but one with direct consequences for water systems. Equipment accumulates fouling, corrosion, and mechanical degradation across that entire span, none of which was built into the original design model. The exchangers, tower fill, and piping at year seven are not the exchangers, fill, and piping at commissioning. The system has changed. The model, in most cases, has not.
Equipment failures and process excursions create their own step-changes. A heat exchanger leak introduces process contamination into cooling water and resets the baseline for corrosion and biological activity, sometimes permanently. A biocide overdose or underdose event can shift the microbiological population in ways that persist long after the event is corrected. A makeup water upset — a turbidity spike, a treatment plant excursion upstream — can introduce a chemistry the original model never contemplated. Each of these events creates a new normal. The static model wasn’t built to recognize that the normal changed, let alone recalibrate to it.
Instrumentation drift is the quiet one, and it’s the one I think gets the least attention relative to the damage it causes. Conductivity probes, pH sensors, and ORP electrodes all drift over time, some faster than others depending on fouling and maintenance quality. A program running on a conductivity-based cycles-of-concentration setpoint can be operating at a materially different CoC than the model assumes, and nobody notices because the instrument is still reporting a number — just not the right one. It just quietly moves the actual operating point further from the design basis every week until something fails and everyone is surprised. Anyone who has run a CoC analysis side by side with a “confirmed” conductivity setpoint after a probe has been in service for a year or two knows exactly how far apart those numbers can get.
And finally, there’s the slow drift of sub-optimal operation that has nothing to do with a discrete event at all. Systems simply tend to run hotter, dirtier, and harder than their design basis as they age. Approach temperatures climb as fill fouls. Fan efficiency degrades. Evaporation rates shift with ambient conditions and load. Cycles of concentration creep in directions nobody intended. None of this is dramatic in any single week. Over a few years, it adds up to a system that bears only a passing resemblance to the one the original model described.
None of this is theoretical. Scale, corrosion, and biological fouling all have real, measurable dollar costs — in energy penalties from fouled heat transfer surfaces, in exchanger and tower replacement costs pulled forward years ahead of schedule, in chemical costs that are either wasted on overtreatment or insufficient to protect assets running outside the assumed operating window. Or the biggest cost…lost or reduced production.
The costs that get the most attention, though, are the ones tied to unplanned shutdowns. A planned turnaround is expensive but manageable — it’s scheduled, resourced, and built into the operating budget. An unplanned outage caused by a tube failure, a fouled condenser, or a corrosion event that finally crossed a threshold is a different category of expense entirely, in lost production as much as in repair cost. In almost every case I’ve reviewed, the gap between what the design model assumed and what the system was actually experiencing is exactly where that failure was incubating. Nobody was lying to themselves on purpose. They were just running the plant against a model that had quietly stopped describing reality.
That’s the real risk of the static model: not that it was wrong on day one, but that everyone keeps trusting it long after it stops being right. Running a treatment program against an outdated design basis is running blind, even though the instrumentation is telling you otherwise. You’re not managing the system in front of you. You’re managing the system as it existed at commissioning, and hoping the two haven’t diverged too far.
To the industry’s credit, this problem has not gone unaddressed. It’s just been addressed in stages, each one better than the last but still incomplete on its own.
The first wave was better monitoring — more sensors, more frequent manual sampling, and more disciplined data logging. Running alongside this, for more than 30 years now, has been tracer chemistry: the use of inert fluorescent tracers co-dosed with treatment chemicals to measure actual chemical concentration in the system independent of conductivity or ion-specific measurements. Tracer technology was a genuine step forward — it gave operators a more reliable way to verify that the chemistry was actually present at the intended concentration, decoupled from the drift and interference that plagues conductivity-based dosing control. But it’s worth being honest about what tracer chemistry is and isn’t. It confirms that a chemical is present. It does not tell you whether that chemical is performing, whether the system’s saturation state has drifted into a scaling or corrosive regime, or whether the operating conditions the chemistry was designed for still exist. Tracer chemistry is a 30-year-old technology — a genuine and reliable tool for chemical inventory management — but it is not a predictive technology. It tells you what’s in the water. It does not tell you what the water is about to do.
The second wave connected that monitoring together — remote data access, automated alarming, trend analysis across sites and time periods that used to live in disconnected logbooks. This closed some of the gap between when a deviation occurred and when someone found out about it. But trend analysis still tells you what already happened. It doesn’t tell you what’s about to happen, or why.
The third wave is where things get genuinely different: predictive technologies that model saturation states and corrosion risk in real time, flagging conditions that are trending toward a failure mode before that failure manifests physically. This is the shift from “here’s what happened” to “here’s what’s about to happen if nothing changes.” It’s a fundamentally different relationship with the system — one where the model is doing continuous work instead of sitting on a shelf. Saturation modeling, done properly, is a much more honest description of scaling risk than the older index-based shortcuts most of us grew up using.
The frontier — and where I think this is all heading — is the digital twin: a live model of the system that updates continuously with actual operating data rather than sitting frozen at the commissioning date. Not a snapshot. A continuously calibrated simulation that reflects the makeup water chemistry today, the process conditions today, the fouling and instrument health today — and that recalculates risk in near real-time as those inputs change.
Rudimentary digital twins for industrial water systems do exist today. A handful of platforms are on the market, and some larger companies have invested in proprietary versions. But it’s worth being clear-eyed about where they stand. Most carry price tags in the hundreds of thousands to millions of dollars per year — a threshold that makes economic sense for only a small handful of the largest industrial facilities, and that places the technology out of reach for the vast majority of plants where it would provide real value. And cost is only part of the problem. Many of these early platforms are still missing significant pieces of the chemistry and operational context that determine whether a digital model actually reflects what a system is doing — knowledge that only comes from hard-learned field experience across crude slate shifts, seasonal source water swings, failure mode time-scales of different asset types, realistic field-level chemistry interactions and inhibition, and the slow degradation of fill and instrumentation that doesn’t show up cleanly in a sensor feed. Building a digital twin that is useful rather than just impressive requires that knowledge to be embedded in the model architecture from the start, not discovered after deployment. Without it, only limited value will ever be derived from operating digital twins.
A digital twin done right doesn’t replace the water treatment program. It makes the program responsive to reality instead of anchored to commissioning assumptions. It turns the design basis from a static document you file away after startup into a living reference point the program continuously checks itself against. The technology is coming, and the cost will drop dramatically in the next 6–12 months. The question is whether it arrives with the field knowledge built in — or whether the industry spends another decade learning at scale what we already learned one plant at a time.
Here’s the question I’d ask if I were still running these programs full-time, and the one I now ask every client who tells me their treatment program is “working fine.”
The question isn’t whether your system has drifted from its design basis. It has. Every system does, from the day it’s commissioned. Seasonality, process changes, turnaround intervals, equipment events, instrument drift, and the ordinary wear of daily operation guarantee it.
The real question is how much it has drifted — and whether your treatment program actually knows it, or is just quietly assuming the commissioning-day model still holds.
If you’re not sure how to answer that, that’s exactly the conversation worth having. Or, are you interested in Digital Twins, whether you’re building one or implementing one, and want an independent expert review? Industrial Water Advisory offers a free 30-minute discovery call to walk through where your program stands against its original design basis and what a living, continuously calibrated model could look like for your system. Schedule a discovery call with Industrial Water Advisory to find out how far your system has really moved — and what to do about it, or to talk to an expert in Digital Twin technology.
Industrial Water Advisory offers a free 30-minute discovery call to walk through where your program stands against its original design basis — and what a living, continuously calibrated model could look like for your system.