There are countless variables affecting chiller plant efficiency.
They move every second.
- Outdoor wet-bulb
- Building load
- Water quality
- Fouling
- Flow regime
- Lift
Fortunately:
AI does not need to track all of them to optimize plant efficiency.
What Is Really Needed?
There are only four main controllable variables required to optimize a chiller plant:
1. Leaving Chilled Water Temperature
2. Chilled Water Flow (DP Setpoint)
3. Entering Condenser Water Temperature
4. Condenser Water Flow
That’s it.
Every reasonably programmed BAS already has minimum and maximum setpoints for these variables.
However:
• Manufacturer limits are designed primarily to protect equipment.
• They are not necessarily selected to match loading profile.
• They are not necessarily selected to maximize efficiency.
In many plants:
• Minimum and maximum setpoints are simply manufacturer limits plus a safety buffer.
• Nominal setpoints sit somewhere near the midpoint.
The Core Concept
By controlling these four variables — and continuously measuring total kW/ton — AI can optimize a chiller plant to the limit permitted by mechanical condition.
That limit is real And it matters.
The Physical Limit of Optimization
Optimization potential is not infinite.
It is constrained by mechanical maintenance condition. To illustrate lets consider the following two scenarios.
Scenario 1 – Clean Equipment
A cooling tower producing X tons
At outside wet-bulb temperature Y°F
• Nozzles are clean
• Spray pattern is uniform
• Thin film formation is optimal
• Fan power is minimal for required tonnage
AI identifies:
• CWF_Sp1
• ECWT_Sp1
That produce minimum plant kW/ton.
⸻
Scenario 2 – Two Years Later
Same wet-bulb.
Same required tonnage.
But:
• Nozzles are partially clogged
• Spray pattern is deteriorated
• Fill media has dry spots
• Fan must work harder to achieve same evaporation
Now AI selects:
• CWF_Sp2 > CWF_Sp1
To improve water distribution and compensate for degradation.
Plant efficiency is lower than Scenario 1.
But AI still finds the best possible efficiency within mechanical limits.
⸻
This Extends Everywhere
The same logic applies to:
• Fouled condenser tubes
• Oil accumulation
• Tower fill degradation
• Pump wear
• Lift shifts
• Heat transfer deterioration
AI does not eliminate physics.
It operates within it.
Proven Field Performance
This is not theoretical.
In real installations, we have observed:
Average chiller plant efficiency gains of 48%.
If chillers represent 30–60% of building electrical consumption (typical for 24/7 facilities), total electric bill reductions can range from:
15% to 30%.
And that is only on the water side.
Perceived Risks
1. “AI will run my plant at the limits.”
No.
AI operates within your existing minimum and maximum setpoints.
In practice, we observe:
• Setpoints hover near mid-range
• Plants operate mostly at part load (≈98% of the year)
• Equipment is not forced into extreme conditions
AI resets within boundaries you already define.
⸻
2. “I don’t know how AI will operate my plant.”
If your BAS already has min/max setpoints:
AI simply takes control of the reset strategy.
It modulates slowly.
It measures impact on kW/ton.
It learns which combinations work best.
No new mechanical risks are introduced.
If limits are not defined:
They should be established by an engineer in coordination with the AI provider, based on:
• Manufacturer envelope
• Maintenance status
• Operating requirements
Modern plant equipment is designed for setpoint modulation.
AI simply does it intelligently.
The True AI Edge
You may think:
“I can do this with traditional programming.”
Engineers have tried for 30 years.
The problem is not intelligence.
The problem is dimensionality.
To replicate AI manually, you would need:
• A mechanical engineer
• Sitting at the workstation
• 24/7/365
• Modulating four variables
• Recording wet-bulb
• Recording load
• Recording kW/ton
• Identifying optimal combinations
For every weather condition.
For every load condition.
For every maintenance state.
That is not programming.
That is pattern recognition.
That is learning.
That is AI.
Where Do We Go From Here?
This series focused on the water side.
But that is only 50% of HVAC.
Air-side optimization is needed to unlock the full HVAC/Energy savings potential. (However these are not the scope of this series.)
When water-side and air-side optimization are combined:
Buildings commonly see:
25%–40% total electric savings from optimization alone. However optimization is not the only lever available on an HVAC system. Often times load reduction opportunities are present which can yield additional savings.
That is what we have observed in Puerto Rico.
Final Caution
Not every AI is equal.
Not every provider is equal.
Owners should evaluate:
• Technology maturity
• Field results
• Engineering understanding
• Implementation capability
Because optimization unlocks long-term asset value.
And when done correctly:
It is not risky.
It is inevitable.
Written by SIMA Intelligent Buildings
Articles in this Series:
- Chiller Plant Efficiency Series – Introduction
- Pump Efficiency Is About Operating Point, Not Speed
- Chiller Plant Efficiency Series – Cooling Tower Efficiency Part 1
- Chiller Plant Efficiency Series – Cooling Tower Efficiency Part 2
- Chiller Plant Efficiency Series – Chiller Efficiency Part 1
- Chiller Plant Efficiency Series – Chiller Efficiency Part 2
- AI – The Limit of Optimization Potential
