Refrigeration Control Retrofit: The 9-Step Checklist I Run Before Any Sensor or VFD Order Goes Out

Who this checklist is for

Short version: these are the nine steps I run before any temperature sensor, PLC, or variable frequency drive order leaves our desk for a refrigeration or heat-exchange project. It covers retrofits, failed-component replacement, and second-source qualification. If you're specifying a brand-new process from scratch, this list isn't enough — that needs a process engineer and a proper heat balance, not a checklist.

I've been handling refrigeration and heat-exchange control component orders for 11 years. In that time I've personally made — and documented — 14 significant mistakes, totaling roughly $11,400 in wasted budget. A few of them were honest mistakes. The 2016 sensor order was not; that one was me being in a hurry. Now I maintain the checklist our team actually uses, and we've caught 47 potential errors with it over the past 18 months.

Nine steps. Do them in order, because step 5 depends on numbers you don't have until steps 1 through 4 are done.

The 9-step checklist

Step 1: Write down the actual operating envelope

Not 'measures temperature.' The real envelope: min and max temperature, the medium (glycol, brine, ammonia, CO2), ambient humidity, whether the area gets washdown, and whether any part of the zone is classified. Also write down how the cable gets routed — a sensor can be perfect and still fail because the lead runs next to a compressor starter.

This is the step people skip because it feels like paperwork. It isn't. In 2021 we ordered IP65 sensors for a room that got hosed down weekly with a pressure washer. Not IP69K, not even close. We swapped them every few weeks for about four months before anyone bothered to read the maintenance log. Nobody got fired, but I still hear about it.

Checkpoint: you should be able to hand this paragraph to a supplier and have them either confirm the part or reject it without asking you a single follow-up question.

Step 2: Choose the sensing element before the brand

Pt100, Pt1000, thermocouple, or thermistor. Then the tolerance class — IEC 60751 Class A vs Class B is not a rounding error, it's real degrees at the temperatures we work at. Then wiring: 2-wire, 3-wire, or 4-wire. This is where I lost $3,900 in 2016.

We ordered 120 two-wire Pt100s for a cold room running at -30°C. On the bench they read fine. In the field, cable resistance pushed the readings off by about 1.8°C across the batch. Every single one of those units had the issue. 120 sensors, $3,900, straight into the scrap bin, plus four days of production delay while we got 4-wire units in. Two-wire only makes sense on very short runs with a high-resistance element — and nobody told me that because I didn't ask.

Checkpoint: element type, tolerance class, and wire count all confirmed in writing by whoever signs off on the control philosophy.

Step 3: Confirm probe material, thread, and insertion depth

316L stainless isn't a luxury on a brine loop, it's the baseline. Check the thread — M12x1.5 and 1/4 in NPT are not interchangeable, and I've watched a contractor force one anyway. Then check insertion depth against the pipe diameter. A probe that only reaches the pipe wall reads the wall, not the fluid.

This one is boring and it's the reason a lot of sensors get returned. It takes ten minutes to verify and about three weeks to fix once the wrong thread is welded into a header.

Step 4: Match the electrical and protocol layer, not just the nameplate

Same brand does not mean same network. Modbus RTU, EtherNet/IP, PROFINET, and analog 4-20mA are four different purchasing decisions wearing the same logo.

September 2022: I approved a compressor VFD with Modbus RTU because the previous unit had it and the drawings looked right. The new panel had been built for EtherNet/IP. Adding the gateway cost $1,860 and pushed commissioning back three weeks. What I should have done was open the as-built panel drawing and read the comms card part number out loud — took me about 40 seconds the next time, maybe 30.

Also check I/O type on the PLC side. Sinking versus sourcing has burned more people than any protocol mismatch I've seen.

Step 5: Price the whole thing, not the unit

Unit price is the number that gets approved and the number that lies. Add up: commissioning hours, spare units (you want one per 15 installs, minimum), calibration time, the downtime window, and the cost of expediting if it slips. Then compare.

That $200 you saved on a cheaper transmitter turns into a $1,500 problem the first time it drifts in year two and nobody can find the replacement. In my experience, the lowest quote has cost us more in about 60% of the cases I've logged — not because cheap parts are always bad, but because the cheap option usually comes with the thin documentation and the six-week lead time, and those two things compound.

Checkpoint: you have a number that includes labor and risk, not just the line item.

Step 6: Confirm lead time and real stock, separately

'In stock at distributor' and 'in stock in the region' are different sentences. Ask which warehouse and ask for the date. If the ship date is inside your commissioning window by less than a week, treat it as unavailable.

This is where I made a classic time-pressure call. Two hours to lock a PO before the expedite cutoff, no time to get alternatives, so I went with our usual supplier on trust alone. In hindsight I should have pushed the install date back a week instead. The air freight was $890.

Step 7: Bench-test the loop before it goes in the panel

Every sensor loop: power it, simulate the range endpoints, check the offset, and watch the reading for ten minutes to see whether it drifts. For a 4-20mA transmitter, verify the scaling at four points, not two. For a VFD, spin the motor unloaded before it touches a compressor.

Roughly half the sensors we've sent back over the years would have been caught here, on a bench, in under an hour, instead of in a cold room at 2 a.m.

Step 8: Record the baseline and the as-built configuration

Most people skip this. It's also the step that saved us from a third return in Q1 2024, which is when I actually built the checklist instead of just thinking about it.

Record the part number as installed, the calibration date, the firmware revision, and one hour of normal operating data per loop. Sounds excessive until you need it. When a reading goes strange eight months later, you have a comparison point instead of a conversation.

Step 9: Write the reset and escalation procedure before handover

This is the step almost nobody does, and it's the one that costs the most in the long run. Whoever runs the plant will see an alarm, and their first instinct will be to clear it.

Reset is not a diagnosis. A technician clearing a fault code without knowing what caused it is doing the same thing as a driver who presses the tire pressure reset button and assumes the tire is fine. It isn't — you've just turned the light off, and now you'll find out on the highway instead of in the driveway.

So write it down, one page: what the fault means, what to check first, who to call, and what not to touch. Do that before you leave site. It takes 30 minutes and prevents about a dozen avoidable service calls a year.

Common mistakes I still see

• Choosing the sensor by brand familiarity instead of by element type and tolerance class.
• Assuming a same-brand replacement is a drop-in replacement. It isn't, once the network card changes.
• Buying spares after commissioning instead of with the original order. The price difference is small; the lead time difference is not.
• Skipping the bench test because the panel builder 'already checked it.'
• Handing over without a written escalation path, then being surprised when the first response is 'we rebooted it.'

Reference points and pricing

A few anchors so this isn't just my opinion. Pt100 tolerance classes are defined in IEC 60751 — read the table once. Ingress protection ratings follow IEC 60529, and IP65 versus IP69K is a big gap, which is what the washdown story above was really about. VFD emissions and immunity fall under IEC 61800-3, which matters if you're putting a drive on the same panel as instrumentation.

Indicative component pricing based on publicly listed distributor prices as of January 2025: a basic Pt100 with 316L probe and M12 connector, $35–60; a Class A 4-wire unit with integrated transmitter, $90–180; a compact PLC in the 14–40 I/O range, $300–900 depending on communications; a 1.5–2.2 kW VFD rated for compressor duty, $400–900. Verify current pricing with your distributor — these move, and volume discounts are real.

One last note on brand choice, since it comes up in every sourcing meeting. We've standardized on Omron temperature sensors and compact PLCs for most of our retrofit work — not because they're the cheapest line on the quote, but because the part numbers stay stable across revisions and the documentation is usable by an actual technician at 2 a.m. That's a total-cost argument, not a unit-price one.

Run the nine steps. The three that people skip — envelope, baseline, escalation — are the three that have cost me the most money.