IoT and Predictive Maintenance for Industrial Refrigeration Packages: A Practical Specification Guide

Ask any rotating-equipment engineer what the worst outcome of a refrigeration package failure is and the answer is rarely the compressor itself. It is the downstream process — a reformer train, an LPG recovery unit, a chilled water loop feeding half a plant — that stops with it. That asymmetry is why refrigeration packages sit near the top of most criticality rankings, and why the shift from calendar-based overhaul to condition-based maintenance keeps coming up in technical bid clarifications.

The intent of this article is narrow and practical. It is not a survey of Industry 4.0 platforms. It is a guide to the decisions a project team actually has to make: which parameters to measure on an oil-injected screw compressor refrigeration package, how the monitoring architecture is layered, what to write into the specification so the data can leave the skid, and where the equipment supplier's responsibility begins and ends.

1. From calendar-based to condition-based: the business case

Three maintenance philosophies compete for the same budget line, and it is worth being blunt about what each one costs.

Run-to-failure is defensible for non-critical, redundant, cheap-to-replace items. A refrigeration package feeding a continuous process is none of those things.

Calendar or running-hour based overhaul is the default in most plants and it fails in both directions. Set the interval short and you strip a compressor whose bearings were perfectly serviceable, paying for labour, spares and an outage you did not need. Set it long and the failure arrives before the window does. Neither error is visible in the maintenance KPI — the first shows up as cost, the second as an unplanned shutdown attributed to "bad luck".

Condition-based and predictive maintenance schedules intervention against measured degradation. Published industry estimates for the shift to condition-based regimes cluster around a 25–30% reduction in maintenance spend, with the larger and more reliable saving coming from avoided unplanned downtime rather than from the maintenance budget itself. Treat those figures as directional. The number that matters on your site is the cost of one unplanned trip of the refrigeration package multiplied by its historical frequency — and for a continuously operating machine, that calculation usually settles the argument on its own.

The framework for making this rigorous already exists and is worth citing in your own internal business case: ISO 17359 sets out the general procedure for condition monitoring and diagnostics of machines — criticality assessment, failure mode identification, parameter selection, alarm criteria, then diagnosis and prognosis. Working through it in that order stops teams from buying sensors before they have decided what failure they are trying to catch.

2. What to monitor on a screw compressor refrigeration package

For an oil-injected screw compressor refrigeration unit, the measurement set that earns its keep falls into four groups.

Compressor and driver. Discharge temperature, oil temperature, oil pressure and differential across the oil filter, and vibration on the compressor and motor bearing housings. This group catches bearing degradation, oil circuit restriction, rotor rub and liquid carry-over risk. Vibration is the parameter that buys time: spectral and envelope trends on a rolling-element bearing typically deviate weeks before any temperature signature appears. Where casing-mounted accelerometers are not sufficient — large, high-value machines with sleeve bearings — the industry reference for the instrument scope, transducer mounting and alarm/shutdown philosophy is API 670, and it is entirely reasonable to cite it in a specification for critical duty even on packaged plant.

Refrigeration cycle. Evaporating and condensing pressure, suction superheat, subcooling and liquid level. Drifting condensing pressure at constant load points to fouling or non-condensables; falling superheat points to expansion device or charge problems; a slow divergence in the approach temperature across the evaporator or condenser is the earliest indication of heat exchanger fouling.

Motor. Current, winding and bearing temperature, and start counts. Start frequency in particular is a maintenance-relevant variable that is almost never trended and almost always exceeded.

Cooling and auxiliary systems. Cooling water inlet/outlet temperature difference and flow, oil cooler duty, and where relevant, oil quality. Online oil analysis is a genuinely useful but genuinely expensive option; for most packages, scheduled offline sampling against a trend baseline delivers the same decision quality at a fraction of the cost.

A pragmatic tiering rule keeps the instrument list honest. Pressures, temperatures and levels are standard scope — they are needed for control and protection regardless of any maintenance strategy. Permanent online vibration monitoring is worth specifying on packages above roughly 500 kW absorbed power, or on any package whose failure stops production, with route-based portable measurement acceptable below that. Online oil analysis and advanced analytics are for the small number of machines where the consequence of failure justifies them.

3. The three-layer architecture

Monitoring architecture on packaged plant is not complicated, but the layers must be built in order.

Layer 1 — sensing. Transmitters, RTDs, pressure transducers, accelerometers and current transformers physically installed on the package. This layer decides data quality and nothing downstream can repair a bad decision made here. An accelerometer mounted on a guard rather than on the bearing housing produces numbers, not information. Thermowell placement, transmitter accuracy class, hazardous-area certification and cable routing all belong to the equipment scope and all belong in the datasheet.

Layer 2 — control and protection. The package PLC and HMI: real-time display, trending, high/low alarms, capacity control, and interlocked shutdown. This layer is not "digitalisation" — it is the safe-operation baseline every package should already have. What varies between suppliers is how much history the controller retains, whether that history can be exported, and whether the tag list is documented and open.

Layer 3 — supervisory and remote monitoring. Data leaves the skid into the plant DCS, a historian, an asset performance management platform, or a supplier-side remote diagnostic connection. Trend analysis, cross-machine comparison and any form of prediction live here, and none of it is possible until layers 1 and 2 are complete and connected.

The failure pattern is consistent across sites and industries: the sensing layer is adequate, the control layer works, and nothing crosses from layer 2 to layer 3. Surveys of industrial asset connectivity repeatedly show that the proportion of connected assets is far higher than the proportion whose data is actually integrated across systems. Instrumented is not the same as connected; connected is not the same as used.

4. Protocols, tag lists and CMMS integration

The gap between layer 2 and layer 3 is closed with three deliverables, all of which are cheap at enquiry stage and expensive afterwards.

A defined protocol. Modbus TCP remains the pragmatic default for packaged equipment — universally supported, trivially mapped, adequate for periodic process values. OPC UA (IEC 62541) is the better choice where semantics, security and self-description matter: it carries engineering units, timestamps and asset structure rather than raw register numbers, and it authenticates and encrypts by design. If the plant is building a historian or asset performance management layer with any ambition, specify OPC UA server capability at the package controller and accept Modbus as a fallback.

A signal exchange list. Every argument about integration on site traces back to a missing tag list. The exchange list should name every point, its engineering unit, range, update rate, alarm limits and direction, and it should be issued and agreed at document review — not derived by reverse engineering during commissioning. Where remote access is involved, the direction column matters most: the common and defensible arrangement for a packaged machine is data out only, with all control commands local, a gateway or data diode enforcing the separation, and the whole scheme reviewed by the operator's own OT security function before order.

A route into the maintenance system. Condition data that does not generate a work order changes nothing. The practical integration is modest: threshold and trend exceptions from the historian or monitoring layer raise notifications in the CMMS or EAM system against the correct functional location, so that a rising bearing vibration trend becomes a planned job in the next window rather than a note in a shift log. This is where the maintenance cost saving is actually realised, and it is almost always an operator-side integration rather than an equipment supplier scope — which is precisely why it needs to be assigned to someone explicitly.

5. Writing it into the equipment specification

Five clauses cover most of what is needed, and all five are inexpensive when they are written before order placement.

• Instrument schedule. Every measured point with tag number, service, type, range, accuracy class, hazardous-area rating and physical location on the package. Ask for it as a deliverable, not as a promise.
• Vibration scope. Whether permanent accelerometers are required, on which bearings, and whether alarm and shutdown functionality is required or monitoring only. State the applicable reference — API 670 for critical duty, ISO 20816 evaluation zones for acceptance.
• Communication interface. Protocol, physical layer, redundancy, and the signal exchange list as a review-stage document.
• Data retention and ownership. How long the controller retains history, in what resolution, in what export format, and — if any supplier-hosted platform is involved — an explicit statement that the operator owns the data and can export it. This clause prevents platform lock-in and costs nothing to insert.
• Factory acceptance test scope. Verify the communication interface during FAT with the actual tag list, not just the mechanical and performance run. An interface tested in the workshop is an interface that works on site.

6. Retrofitting packages already in service

For existing units, the cost of retrofit divides sharply by control system generation. A package with a modern PLC usually needs only a communication module, a documented tag list and a small number of additional transducers — a low-cost, low-disruption exercise that can often be done without an outage. A relay-logic or obsolete-controller package needs a control system replacement first, which is a different order of investment and should be planned into a scheduled turnaround rather than justified on monitoring benefits alone.

In both cases the sequence is the same: connect what already exists and start accumulating history before purchasing any analytical layer. Predictive models need a baseline. A platform installed on a machine with no stored history produces dashboards, not predictions.

7. Five procurement pitfalls

Mistaking the HMI for digitalisation. A touchscreen displays; it does not necessarily store, transmit or analyse. The two questions that separate a display from a data source: how long is history retained, and in what format can it be exported?

Buying analytics before data. Sequence matters. Sensing, then connectivity, then storage, then analytics. Reversing it wastes the software budget while the underlying gap remains.

Evaluating the platform instead of the instrument list. Demonstrations are easy to make impressive. During technical bid evaluation, compare instrument schedules and transducer locations across bidders — that comparison is objective and predicts real capability far better than a user interface does.

Leaving the interface to site. Protocol and tag list agreed at document review cost days of engineering. The same work done during commissioning costs weeks and usually a change order.

Ignoring statutory inspection. Condition-based maintenance replaces unnecessary opening of machinery. It does not replace mandatory pressure equipment inspection, relief device testing or classification survey requirements. Those obligations continue on their own schedule and should be stated as such in the maintenance strategy document, so nobody later assumes monitoring bought an exemption it did not.

8. Where the package supplier fits

It is worth being precise about scope, because the market is noisy on this point. Suzhou Lmart Energy Equipment is an equipment manufacturer and package integrator — pressure vessels, shell-and-tube heat exchangers, and the integration and skid-mounting of screw compressor refrigeration units. We are not a software vendor and we do not sell an analytics platform. What we are responsible for is the part of the chain that cannot be added later: instrument selection and tapping point design, transducer mounting provisions, PLC control and protection logic, HMI configuration, and open communication interfaces prepared to the specified protocol and tag list so that the operator's DCS, historian or CMMS can take the data.

Two delivered projects illustrate how differently that scope is configured by application. On a continuous catalytic reforming unit at a large domestic refining and petrochemical complex, two direct-cooling packages of 12,650 kW refrigerating capacity each, with 4,500 kW drivers, were supplied with full temperature, pressure and vibration instrumentation and integrated into the plant-wide DCS — at that unit criticality, the monitoring scope is a rounding error against a single unplanned trip. On a propane refrigeration system for an oilfield gas processing station (5,045 kW at −38 °C), the same equipment class was configured differently: skid-mounted, local PLC control with remote data transmission, because the station has no resident equipment engineer and the problem being solved is distance rather than criticality.

The common thread is that the monitoring scope was fixed during design review against the actual operating context, rather than bought as a generic package and adapted afterwards. That is the whole argument of this article in one sentence: predictive maintenance is decided at specification stage, and everything after that is execution.

Conclusion

The sequence that works is unglamorous and reliable. For new packages, define measurement points, protocol and data ownership in the specification and verify the interface at FAT. For units in service, close the layer 2 to layer 3 gap and accumulate history before spending on analytics. For every asset, decide which failure modes you are trying to catch before selecting the sensors that catch them — the ISO 17359 discipline — and route the resulting exceptions into the maintenance system where they become work orders.

Data first. Prediction after. If you are preparing a refrigeration package specification and want the instrumentation and interface scope reviewed against your operating context, our engineering team is available to work through it.