AI Integrated Water Treatment System

Why Cooling-Water Chemistry Is Still a Manual Process — and Why That Has to Change

Most plant rooms still dose biocides, scale inhibitors and corrosion inhibitors on a timer or a thumb-rule. The dosing pump runs at the same stroke length whether the cooling tower is at peak load or barely evaporating; whether the recirculating water has just been topped up with fresh RO permeate or has been concentrated by the sun all week; whether the biofilm count is creeping up because of a change in the incoming make-up water or holding steady through a quiet month. The result is over-dosing in some conditions and under-dosing in others, with consequences ranging from wasted chemical spend to accelerated equipment damage.

The chemistry itself is well understood. Cyclotetrasiloxane-based scale inhibitors work, oxidising biocides work, dispersants work. The gap is not in the chemistry — it is in the feedback loop. The dosing pump does not know what just happened in the tower. It runs a programme that was decided by an operator who last looked at a water test three weeks ago.

Closing that gap is the object of an AI-integrated control layer. The necessary inputs already exist on most plant floors: flow meters on make-up and bleed, conductivity probes, ORP sensors, sometimes an inline corrosion probe or a particle counter. They are individually accurate; they are collectively ignored because no one has time to watch six trend lines at once. Adding a control layer that does watch them, correlates them with the chiller load profile, and adjusts the dosing pumps on a continuous basis is the incremental change that moves the whole system from “good chemistry on a good day” to “good chemistry every day”.

This is not hypothetical AI; it is supervisory control with model-predictive logic. The same regime is already deployed in power-plant feed-water chemistry under the umbrella of Industry 4.0 / IIoT. Adapting it to commercial and district-cooling loops is mainly a question of economics — the loop must be large enough that chemical and water savings justify the instrumentation and control cost. That threshold is typically reached around 1,500 TR of cooling capacity; below it, conventional timer-based dosing remains the right answer.

Inputs the AI Layer Reads and Outputs It Adjusts

Most plant floors already have 2–4 of the six instruments installed; the AI layer integrates what you have rather than insisting on a parallel sensor stack. That is also why the economic break-even sits at ~1,500 TR and not at 500 TR.

Water-Treatment Control, Answered

Is this an “AI” product or a real control system?

It is a real industrial controller running model-predictive logic on top of standard cooling-water chemistry. The “AI” label refers to the feedback correlation engine; the chemistry itself is conventional inhibitor / biocide chemistry that has been in use for decades. Nothing exotic is added to the water.

Does it require internet connectivity?

No — edge control is the default architecture. The controller runs locally and stores trend data on board; remote access (Modbus / 4G / Ethernet) is available for operators who want it, but is optional. This keeps the cybersecurity surface to whatever the plant BMS already exposes.

Will it work with our existing dosing pumps?

In most cases yes. The controller outputs 4–20 mA or Modbus setpoints to the existing chemical-feed skid; the pump hardware stays unchanged. A site survey takes about 90 minutes and confirms the integration points before any equipment is delivered.

What does early access involve?

A site survey (loop volume, sensor inventory, BMS integration requirements), a phased rollout plan against your operating calendar, and a reference agreement covering data we can and cannot publish. Early-access projects are scheduled in two slots per quarter.