Ammonia Refrigeration Automation: What Changes Between One Plant and Twenty

Installing automation at twenty ammonia plants is not twenty times harder than installing it at one. Keeping it working is. Automated control degrades quietly after commissioning, and the mechanisms that cause the degradation, setpoints adjusted during a difficult shift and never restored, undocumented local modifications, and faults nobody is looking for, operate independently at each site. After three years, twenty installations that started identical are running twenty different configurations, and no one can say precisely how they differ.

This is the part of multi-site automation that rarely appears in a project plan. Commissioning has a completion date. Configuration drift does not have a start date, which is why it gets no owner and no budget line.

Two plants receiving the same automation package in the same year can end up in very different places depending on how thoroughly each was commissioned. Commissioning is not the act of switching a system on. It is the verification that what was installed does what the owner specified, under real operating conditions, with the results documented.

The formal discipline is well defined. ASHRAE's training on the principles of building commissioning under Guideline 0 and Standard 202 places emphasis on initiating the commissioning process during pre-design, so that the owner's project requirements document can guide the verification of success across the design, construction, and operation phases of a project. Standard 202 was reissued in 2024.

The phrase carrying the most weight there is across the operation phase. Commissioning that ends at handover verifies a snapshot. It confirms the plant behaved correctly on the day the contractor was present, and says nothing about whether it still behaves correctly eighteen months later.

Where an owner's project requirements document does not exist, and at many facilities it does not, there is no written record of what the automation was supposed to achieve. Subsequent changes cannot be evaluated against intent, because intent was never written down. Each site then develops its own understanding of what correct looks like, and those understandings diverge.

Drift is not a failure of discipline so much as an accumulation of reasonable decisions. A room runs warm during a heavy receiving week, so a setpoint moves. A defrost schedule gets stretched because the crew needs the dock clear. A high-pressure alarm limit gets widened because it was nuisance-tripping. Each adjustment solves a real problem on the day it is made. Almost none get reverted, and almost none get recorded.

At a single plant this is manageable, because the person who made the changes is generally still there and remembers most of them. Across twenty plants it is not, and the loss compounds with turnover. The configuration becomes the only record of the decisions, and reading a configuration tells you what a system does without telling you why.

Enterprise treatments of refrigeration automation controls draw the line at attribution: whether every change is traceable to a person, a time, and a reason. At one facility that record is a convenience for the next engineer. Across a portfolio it is the only mechanism by which leadership can know which sites are still running the strategy that was commissioned and which have quietly moved somewhere else.

Version sprawl runs alongside this. Controllers, interface software, and communication modules all get patched at different times by different contractors, and the resulting matrix of what is installed where is rarely maintained anywhere. The question of whether a control strategy proven at one site will behave identically at another cannot be answered without it.

Why faults persist: A 2022 paper presented at the International Refrigeration and Air Conditioning Conference at Purdue observes that HVAC equipment frequently suffers from inadequate maintenance, improper installation, and unrepaired faults, which are responsible for an increase in energy consumption and utility costs.

Degraded performance is difficult to notice because it does not announce itself. Research on automated fault detection and diagnosis presented at the Purdue refrigeration conference notes that such systems have the potential to provide early detection of these faults, so that energy waste and excessive equipment wear can be avoided. The same work identifies a practical obstacle, observing that a major challenge in field implementation is the cost and availability of sensors.

The operational version of this problem is familiar to anyone who has run a refrigeration plant. A temperature alarm fires when product is at risk, which is late. Nothing fires when the plant is holding temperature correctly while consuming noticeably more energy than it did last season, because no threshold was ever set for that condition and no operator is asked to watch for it.

Across a portfolio, the absence of that signal is what makes performance comparison so difficult. Twenty sites all reporting in-range temperatures look equivalent on a status report. Some of them are running well and some are compensating for a fault, and nothing in the routine reporting distinguishes the two.

Every argument about automation eventually returns to who is available to operate it. Industrial refrigeration has a defined competency ladder, and the entry rung is genuinely entry-level.

The Refrigerating Engineers and Technicians Association describes its Certified Assistant Refrigeration Operator credential as an entry-level examination designed to identify whether an operator has the basic knowledge to function safely in an engine room under the supervision of a more experienced operator, and states that there are no minimum experience requirements to sit it.

Under the supervision of a more experienced operator is the load-bearing phrase. The credential is explicitly designed around a supervised model, which assumes an experienced operator is present to supervise. Portfolios that have thinned their senior bench through retirement are running an arrangement the certification structure did not anticipate.

This is where consistency stops being an efficiency argument and becomes an operational one. When a less experienced operator covers an unfamiliar site, every difference in screen layout, alarm naming, and setpoint convention between that site and the one they trained on is a chance to misread a situation. Standardized configurations do not replace expertise. They reduce how much site-specific expertise a person needs before they can be useful.

Temperature control in these facilities is not only a product quality matter. FDA guidance on the Food Safety Modernization Act final rule for preventive controls for human food states that process controls include procedures ensuring control parameters are met, that such controls can include operations such as refrigerating foods, and that they must include parameters and values such as critical limits appropriate to the nature of the control and its role in the facility's food safety system.

The guidance is equally specific about evidence, noting that monitoring is conducted as appropriate to the preventive control, giving the recording of temperature values as its example, and stating that monitoring must be documented. A control system that holds temperature but cannot produce a clean record of having done so satisfies half the requirement.

That documentation burden scales badly when each site handles it differently. Twenty facilities producing twenty formats of temperature record is workable until somebody has to assemble a portfolio-level answer under time pressure, at which point the differences become the whole problem.

The refrigeration engineering does not change. Ammonia behaves the same way in a plant that is one of twenty as it does in a plant that is one of one, and the control strategies that work at a single site work everywhere.

What changes is that every ungoverned decision now happens twenty times, in twenty places, without anyone in a position to see the pattern. Drift that a single competent engineer would catch and correct at one plant goes unnoticed across a portfolio, not because the people are less capable, but because nobody is looking at all twenty configurations side by side. Automation at scale is less an installation problem than a question of whether the organization can still describe, accurately, what each of its plants is doing.