Chiller-plant automation fails for unglamorous reasons. The controller is usually capable enough; the problem is a point schedule that omits the flow switch, a staging sequence with no minimum run time, or a Modbus map that is one register out. When the plant behaves unpredictably, the operators switch the system to manual and the automation becomes decoration.
This article covers the four parts that determine whether a chiller PLC works: the point schedule, the control sequences, communications, and commissioning. The example is a plant with three 500 kW chillers, three chilled-water pumps, three condenser-water pumps and two cooling towers.
1. Point schedule: list everything before writing logic
The point schedule is both the design input and the acceptance checklist. It should cover analog inputs (AI), analog outputs (AO), digital inputs (DI) and digital outputs (DO) for each item of equipment.
| Object | AI | AO | DI | DO |
|---|---|---|---|---|
| Chiller | Entering/leaving water temperature, evaporator and condenser pressure, current | — | Run, fault, remote ready, flow | Start/stop, alarm reset |
| Chilled-water pump | Current, frequency | Speed reference | Run, fault, local/remote | Start/stop |
| Condenser-water pump | Current, frequency | Speed reference | Run, fault | Start/stop |
| Cooling tower | Leaving water temperature | Fan speed | Run, fault | Start/stop |
| Header | Supply/return temperature, flow, differential pressure | Bypass valve position | — | — |
| Outdoor | Dry-bulb, humidity | — | — | — |
A plant of this size typically has 80–100 points. At an installed cost of 800–1,500 CNY per point, including controller, cable, containment and programming, the automation package is about 80,000–150,000 CNY.
Three points are most often missing: the flow switch, the chiller remote-ready signal and the bypass valve. Without a flow switch, a chiller can start with no evaporator flow and freeze; without remote-ready, the controller does not know whether the chiller will accept a remote command; without a bypass valve, low-load operation can fall below the minimum flow and trip the chiller.
2. Six sequences that must be explicit
Start/stop interlock. Start the cooling-tower fan, condenser-water pump, chilled-water pump and then the chiller, with 30–60 seconds between steps. Stop in reverse, with the pumps running for 5 minutes after the chiller stops. The chiller must not start until the flow switch is proven.
Staging. Use return-water temperature deviation plus time. Add a chiller when the return is 1.5 K above setpoint for 15 minutes; remove one when the return is 1 K below setpoint for 30 minutes. A load-rate criterion is more accurate but needs reliable flow and temperature measurement.
Minimum run and minimum stop. Once started, a chiller should run for at least 20 minutes. After stopping, it should remain off for at least 15 minutes. Without these timers, small load changes cause repeated starts and shorten compressor life.
Runtime rotation. Start the chiller with the lowest accumulated running hours. Rotate weekly or monthly, preferably during a stable-load period.
Fault changeover. If a running chiller faults, start the standby unit and close the failed unit’s isolation valves. The supply temperature may rise briefly, but the pumps must keep running.
Freeze and limit protection. Alarm at 4 °C leaving-water temperature, stop the chiller at 2 °C while keeping the pump running, and trip on high condenser pressure or low evaporator pressure. Record the time and value of every protection action.
| Logic item | Setting | Reason |
|---|---|---|
| Add delay | 15 min | Avoid response to short load swings |
| Remove delay | 30 min | Removal should be more conservative than addition |
| Minimum run | 20 min | Protect the compressor |
| Minimum stop | 15 min | Allow pressure equalization |
| Pump run-on | 5 min | Carry residual cooling, prevent evaporator over-temperature |
| Freeze alarm | 4 °C | Leave time to react |
3. Communications and register maps
Three combinations are common: a chiller with a Modbus RTU port connected through a serial server; a chiller with Modbus TCP connected directly to the plant switch; or a gateway to BACnet for the building automation system.
| Protocol | Physical layer | Typical use | Watch point |
|---|---|---|---|
| Modbus RTU | RS485 | Chillers within one mechanical room | Common baud rate; 120 Ω termination |
| Modbus TCP | Ethernet | Plant LAN with many points | IP planning; avoid address conflicts |
| BACnet IP | Ethernet | Interface to BAS | Device instance planning |
| BACnet MS/TP | RS485 | Existing building systems | Lower speed; limited point count |
Five failures account for most commissioning problems:
| Symptom | Cause | Action |
|---|---|---|
| No data | Register address offset by one | Confirm whether the manufacturer uses zero- or one-based addressing |
| Nonsense value | Wrong data type or byte order | Confirm high/low word order for 32-bit floats |
| Intermittent data | Baud rate or parity mismatch | Agree 9600, 8, N, 1 on both sides |
| Some points missing | Wrong slave address or function code | Check the map against the manual |
| Write has no effect | Writing a holding register instead of a coil | Confirm read/write area and permissions |
Address offset is the most common. A register described as 40001 in one manual may be 0-based in another. During commissioning, read each point with a test tool and align the actual address with the schedule before programming. Byte order is second: a 32-bit float can be arranged four ways, and a value such as 3.6 × 10⁻³⁸ almost always means the word order is wrong.
4. Interface to the building automation system
The technical protocol is rarely the real dispute. The dispute is who supplies the point list and which naming convention applies.
A workable division is: the HVAC designer supplies the process point list (name, type, range, alarm value); the controls contractor supplies the communication point list (address, data type, read/write permission); both sides perform a point-by-point check and sign it before programming. That single step removes most rework.
Agree equipment identifiers early. If the HVAC drawing calls a chiller “CH-01”, the PLC schedule calls it “Chiller 1” and the BAS calls it “Chiller 1”, maintenance will need three documents to find one fault. Use one identifier from design drawing through to the HMI.
Agree the data granularity too. The BAS usually needs status, alarm, supply and return temperature, not every internal chiller register. Uploading dozens of internal parameters makes the graphic unreadable and adds no operational value.
5. Three tests before acceptance
Interlock test. Open the flow switch and confirm the chiller stops. Inject a fault signal and confirm the standby starts. Do this under load; a no-load test will not reveal a sequencing error.
Staging test. Raise the return temperature by closing terminal valves or reducing flow, then record whether the next chiller starts after the set delay. Reduce load and confirm the stop sequence and delay.
Communication-loss test. Disconnect the communication cable and observe whether the controller alarms and what the chiller does. The default should be defined in the control narrative: hold, stop, or run to a safe condition. Communication failure is more common than equipment failure, yet this test is often skipped.
| Acceptance item | Criterion | Record |
|---|---|---|
| Point accuracy | Every point matches the schedule | Point check sheet |
| Interlock | Chiller stops within 5 s of flow loss | Time-stamped action record |
| Staging | Delay and threshold match settings | Time/temperature trend |
| Communication loss | Alarm and defined safe state | Alarm log |
| Data retention | At least 12 months | Historian export |
Key parameters and acceptance criteria for the point schedule
The point schedule is the contract between the mechanical design and the control program. If it is incomplete, the program is incomplete. The reference case has three 500 kW chillers, paired hydronic pumps and two tower cells.
| Item | Reference | Acceptance check |
|---|---|---|
| Points per chiller | 12–15, including AI, DI and DO | Every point in the schedule is readable and writable |
| Total points | About 80–100 for the example plant | No point is left as a spare without a name |
| Control cycle | 1–5 s; interlock loop 1 s | Flow-switch trip stops the chiller within 5 s |
| Communications | 9,600 or 19,200 baud, 8N1; 120 Ω termination | Point-by-point read matches the register map |
| Data retention | At least 12 months | Trend export covers a full cooling season |
| Mandatory points | Flow switch, remote-ready, bypass valve | Any one missing fails the schedule |
| Staging test | Add and remove according to the written delay | Time and temperature recorded against the set point |
| Communication loss | Alarm plus defined safe state | Chiller holds or stops as the narrative says |
The three mandatory points are not optional extras. Without a flow switch, the chiller can start dry and freeze the evaporator. Without a remote-ready signal, the controller does not know whether the machine will accept a remote command. Without a bypass valve, low-load flow falls below the chiller’s minimum and the machine trips on protection.
At acceptance, prove each point against the schedule, not against the drawing. A drawing can be revised; the schedule is the as-built list. Record the point name, address, data type and engineering unit, and sign it before the graphics are built. That single signed list removes most of the rework at the end of the project. Longcom’s water chiller page gives the communication options and the factory point list for each model.
Common questions
Does a chiller with its own controller still need a PLC?
It depends on scale. One or two chillers can usually use the manufacturer’s group function. Three or more, or a plant that must interface to BAS and meter energy, benefits from an independent PLC or plant controller. The proprietary route is cheaper initially but limits later expansion.
What should be checked first when Modbus reads nothing?
Check the slave address, baud rate and parity, register offset and data type/byte order, in that order. These four items solve most no-data cases. Test one device with a service tool before troubleshooting the network.
Should staging use return temperature or load rate?
Return temperature is simple, reliable and does not depend on a flow meter. Load rate is more precise but depends on accurate flow and temperature difference. If the flow meter is unreliable, use return temperature plus a time delay.
Should chillers stop or keep running when communications fail?
Define it by risk. An unattended plant should stop or move to a safe state; a staffed plant may hold status and alarm. The decision belongs in the control narrative and should be tested at commissioning.
How much energy does better sequencing save?
Typical plant-level savings from improved staging and pump control are 5–15%, depending on the original operating practice. A plant that was switched manually has more upside than one already using load-based control. Accurate sensors are a precondition.
How can an old chiller without a communication port be connected?
Add power meters, temperature sensors and relays for status and fault, and use digital outputs for start/stop. Accuracy is lower than native communication, but it supports basic sequencing and energy metering at modest cost.
How many I/O points?
A three-chiller plant needs roughly 80–100 points, including status, alarms, valves, flows and temperatures. Count the spares and the gateway tags.
What must be hardwired?
Emergency stop, fire alarm, high pressure and loss of flow. Do not rely on the PLC alone for safety.
Related reading
- Chiller Plant Group Control
- Chiller Plant and BAS Integration
- Chiller Remote Monitoring Platform
- Water Chiller Commissioning
LONGCOM can prepare the I/O schedule, staging logic, communication map and acceptance tests for a chiller plant. For a project review, contact 400-878-0280 or 13002001280.
Related reading
Chiller plant sensors · Chiller plant group control · Chiller plant to BAS