Publication date: October 2026
A food processing plant has combustible dust in its mill and six metric tons of ammonia in its machine room. Four hundred rows of risk analysis, eleven years without an incident, and two hazards that were never assessed together. This is how using the wrong tool leaves a plant with something worse than no analysis at all: documented certainty that the job is already done.
The folder is always heavy.
It is the first thing I notice when a plant manager slides it across the conference table, with the mix of pride and urgency that comes with turning in a completed assignment. Four hundred rows. Color-coded ratings. Signatures at the bottom. Today’s binder belongs to a food ingredients plant: drying towers, mills, bagging lines, and five cold rooms served by an ammonia refrigeration system.
Row 93 lists a caught-in hazard at the point of operation. Row 120 lists skin exposure to a solvent. Row 207 lists a forced upper-limb posture. All three have the same rating—medium risk—the same pair of criteria, and the same one-to-five scale.
These are three phenomena that have absolutely nothing in common, except for the rows they occupy.
People defending the universal risk matrix often invoke ISO 31000. It is worth reading: the standard is a framework that emphasizes tailoring risk management to the organization’s context. It does not prescribe techniques. It does not validate matrices. Using it to justify a single tool for every hazard means making it say the opposite of what it actually says.
But the deeper objection is not regulatory; it is arithmetic. An ordinal scale—rare, occasional, frequent—assumes the events being compared are alike in nature. They are not. A caught-in hazard depends on a machine’s geometry and the sequence in which its interlocks fail. Dermatitis depends on the relationship between concentration and exposure time. Putting them in the same cell does not produce an imprecise comparison; it produces an impossible comparison that only looks as if one has been made.
The first question, then, is not how risky something is. It is what kind of question we are asking—and which tool was designed to answer it.
There is a set of established techniques, and they are standardized. IEC 31010:2019 organizes them by function, not popularity: each technique answers a different question, and only that question.
Read in sequence, they form a process that almost no plant completes. HAZID opens the inventory when everything is still conceptual. HAZOP examines the process once the diagrams are available. FMEA drills down to the component, where the hazard exists in the part rather than the process variable. LOPA takes the high-consequence scenarios and asks the one question HAZOP cannot ask on its own: Are the safeguards in place sufficient? Bow-tie analysis turns the findings into a diagram a director can understand in 30 seconds.
One important clarification: HAZID does not appear under that name in IEC 31010. It is an established industry practice, not a standardized method, and should be presented as such. The other techniques are included, and where they fit matters. HAZOP belongs to the risk identification family; LOPA belongs to risk control analysis. Anyone who completes a HAZOP, binds the report, and files it away has answered a question that was never asked.
There is a scenario where this confusion stops being methodological and becomes physical.
Let’s return to the binder. Between the hammer mill and the packaging machine, the hazard “explosive atmosphere” appears, rated using the same two criteria as everything else. The controls column lists three items: fire protection system verification, grounding inspections, and cleaning routines. A responsible person is assigned, and evidence is available. The row is closed.
All three controls are necessary. None answers the question.
The question is whether an explosive atmosphere exists; how far it extends; how often and for how long it is present; and which ignition sources could ignite it within that volume. That is called hazardous area classification. It is not an inspection; it is an engineering study with its own deliverables: a zoning drawing, material characterization, and equipment specifications derived from both.
In Mexico, this study is governed by the current edition of NOM-001-SEDE, which adopted the National Electrical Code’s Class and Division system. The class describes what is in the air: Class I for gases and vapors, Class II for combustible dusts, and Class III for fibers. The division describes when it is present: Division 1 if it may be present during normal operations, and Division 2 if it appears only when something fails.
There is a second map. NFPA 70 allows it in Articles 505 and 506—a detail often omitted in discussions in Mexico—and it is used by the IEC 60079 series and the European directives known as ATEX. Its logic is graduated rather than binary.
Look at the column on the right. Division 1 combines Zone 0 and Zone 1 in a single category, even though the two zones require different equipment categories. The conversion therefore cannot be reversed. Converting zones to divisions erases a distinction; converting divisions to zones requires recovering information—the frequency and duration of the atmosphere’s presence—that a Class and Division study never recorded. Anyone who converts the classification on the drawing without returning to the original study either over-specifies the equipment or leaves the most hazardous volume unprotected. I have seen both, and the second is not discovered until something ignites.
Two additional attributes are assigned to the zone to determine which equipment can be installed. The first is the gas group, and it presents a costly trap: the naming systems run in opposite directions. NEC Group A, the most hazardous, corresponds to international Group IIC; Group D, the least hazardous, corresponds to Group IIA.
The second attribute is the temperature class, which sets how hot the equipment’s surface can become before it could serve as an ignition source: T1 is 450°C; T2, 300°C; T3, 200°C; T4, 135°C; T5, 100°C; and T6, 85°C. These measure different things—the gas group addresses flame propagation through gaps, while the temperature class addresses ignition from a hot surface—and both must be verified separately.
When the material is combustible dust, characterization requires parameters no visual inspection can provide: minimum ignition energy, minimum ignition temperature of a dust cloud and a dust layer, minimum explosible concentration, Kst, and maximum pressure. NFPA 660 documents how these are determined in its 2025 edition.
ATEX refers to two European directives: 2014/34/EU, covering equipment, and 1999/92/EC, covering the protection of workers exposed to explosive atmospheres. ATEX is not Mexican regulation and is not independently enforceable in Mexico.
It enters through a contract. A European parent company or an insurer requires ATEX marking, and the plant discovers that its installation—classified legitimately under the Class and Division system—must now be reconciled with a system that does not recognize it. The instinctive response is to reclassify everything, and that is the most expensive option. The right approach is to retain the original study and use documented engineering judgment to derive the zone and equipment category required by the specification.
Let’s return to the three controls. A checklist verifies that a control is present. It does not establish that the control is sufficient. There is a gap between being present and being sufficient—and incidents happen in that gap.
That is the deepest harm caused by using a tool that is not suited to the task, and it is not technical. A completed checklist creates a record. The record creates confidence. And confidence closes the question. A false sense of control is not an individual oversight; it is the predictable result—documented, signed, and filed—of using a basic method for a complex operation.
A plant with no analysis knows it does not know. A plant with a generic risk assessment believes it knows. The second is harder to help because all its evidence points the other way.
One might argue that everything above is good international practice, not a Mexican requirement. That argument does not survive a careful reading of the standard.
Section 7.1(e) of NOM-028-STPS-2012 lists what the selected risk analysis techniques must provide: identify potential hazards, analyze causes, analyze consequences, recommend measures, identify issues for further study, identify the frequency of occurrence, and determine the required level of protection.
Focus on the last two. That is, word for word, the function of a layers of protection analysis. Since 2012, the Mexican standard has required the output of a LOPA without calling it by name.
A generic matrix does not provide frequency. An ordinal scale does not become a frequency just because more columns are added. Nor does a generic matrix determine the required level of protection, because that requires assessing whether the existing safeguards are sufficient—a question the tool never asks. There are also two requirements that often go unnoticed until an inspector points them out. Section 7.1(b) calls for 11 input documents, including piping and instrumentation diagrams, electrical drawings, and alarm lists. Section 7.2 requires the study to be updated every five years, before any modification, and after an investigation of a major accident.
One blind spot remains, and it is the most costly. Section 8.1(c) lists activities that require a procedure and authorization: lockout of lines and energy sources, confined spaces, welding and cutting, hazardous reactions, handling flammable and toxic substances, monitoring explosive atmospheres, and maintaining vessels that have contained hazardous materials. These are all nonroutine activities.
And with almost comic precision, these are exactly the activities the matrix overlooks. It assesses the plant under steady-state conditions and ignores a scheduled shutdown, a changeover, a restart after maintenance, and contractor entry. In short, it ignores the moments when the most severe risks are concentrated.
It is worth returning to the plant in the binder, because it has both scenarios within 40 meters of each other. In the process area there is combustible dust: a drying tower, a mill, pneumatic conveying, a silo, and a bagging machine. In the machine room there is anhydrous ammonia: three compressors, a liquid separator, rooftop condensers, and the network supplying five cold rooms.
In the binder, each hazard occupies one row. “Dust explosion,” medium risk, controls: cleaning, fire extinguishers, grounding. “Ammonia release,” high risk, controls: detectors, respiratory protection, response team, annual drill. Both have a responsible person and supporting evidence. Both are marked closed.
And yet.
No one checked the inventory. The system contains about 6,200 kilograms of anhydrous ammonia. The threshold quantity in Appendix A of NOM-028-STPS-2012 is 4,600 kilograms. The plant exceeded the threshold when it expanded the fourth cold room six years ago, and no one noticed because no one compared the inventory with the appendix. The standard applies in full, yet the plant has spent six years operating under the belief that NOM-028 “is for petrochemical plants.”
The machine room was treated as an unclassified location, and that is the most revealing gap. The plant operates it as an ordinary location because “it always has been one.” But that designation is not a fact of nature; it is a condition. IIAR 2 allows the designation only when emergency ventilation and ammonia detection meet its requirements. In other words, the ventilation and detectors are not extra controls installed in an ordinary room. They are the reason the room qualifies as ordinary. When the detector has gone 14 months without documented calibration, the basis for the classification has disappeared. No one filed a report because no one knew there was a report to file.
And no one assessed the interaction between the two hazards, which is where the major scenario lies. A primary deflagration in the mill is not the feared event. The feared event is the secondary explosion: the pressure wave lifts dust accumulated on beams, light fixtures, and cable trays, triggering a second explosion of incomparably greater magnitude. If this happens in a building that houses ammonia piping, the fire stops being a fire and becomes an emergency the response team cannot approach.
A row-based matrix cannot represent this scenario. Not because the person who filled it out lacked rigor, but because of a structural limitation in the tool: the rows do not communicate with one another. Each hazard is assessed in isolation, and the interaction—which is what turns an incident into a catastrophe—has no row in which to be recorded.
What should be done—and what went wrong
The sixth point deserves its own line, because it is the one I see most often. The ammonia detector and the ventilation start-up share a controller. The response team is the same small crew working the night shift. And the cleaning, counted as protection against a dust explosion, depends on the availability of workers during the shift with the fewest people. None of the three is independent of the initiating cause, so none is, strictly speaking, an independent layer of protection. An administrative control is not a weak layer; it is a different kind of layer. Adding it to the same column as a safety instrumented system is not an error of degree. It is an error of category.
The right conclusion from a review like this is not that the plant did too little. It did a lot: 400 rows, 11 years of discipline, impeccable supporting evidence. It did a lot of what did not protect it. The study that would have provided protection was not more expensive than the matrix. It was different.
The binder will still be heavy. That part does not change. What changes is what we ask.
The question is not whether the plant has a risk analysis—it almost always does. The question is whether the tool answered what the operation required it to answer. Were nonroutine activities included in the scope? Was hazardous area classification based on an engineering study or a walkthrough? Would each safeguard described as a layer of protection still be effective at 3 a.m. on a Tuesday, with half the usual crew and a new operator?
That is why I present reports in a meeting instead of sending them by email; include training for operational personnel as part of the deliverable instead of quoting it separately; and present each control according to the hierarchy of controls, including how much it reduces risk and the specific scenario in which that reduction occurs. A recommendation without a scenario and a measure of risk reduction cannot be prioritized, budgeted, or defended before an auditor. It survives in the meeting minutes and disappears from the budget.
When the answer to all four questions is yes, the analysis provides protection. When it is not, the analysis simply documents. A file that documents without protecting is the most expensive form of compliance: it pays the full cost of the paperwork while leaving the risk untouched.
At HF Safety & Environmental Services, we help manufacturing, food processing, pharmaceutical, and chemical plants review the methodology behind their risk analyses. This includes selecting the right technique for each scenario, classifying hazardous areas, verifying the independence of protection layers, and aligning the analysis with NOM-028-STPS-2012. If you are reviewing your file with these questions in mind, let’s talk.
My plant processes food, not petrochemicals. Does NOM-028-STPS-2012 apply to me?
It depends on the inventory, not the industry. If the ammonia refrigeration system exceeds the threshold quantity in Appendix A—4,600 kilograms of anhydrous ammonia—the standard applies in full. Many food processing plants exceed this threshold when they expand their cold rooms without realizing it.
When is a LOPA needed instead of only a HAZOP?
A LOPA is needed when a HAZOP identifies high-consequence scenarios that depend on multiple safeguards for mitigation, and the plant needs to determine whether those safeguards are independent and sufficient. HAZOP identifies the scenario; LOPA determines whether it is adequately protected.
Does an ammonia machine room need to be classified as a hazardous area?
IIAR 2 allows it to be designated as an ordinary location, but only under specific conditions: it must have emergency ventilation and ammonia detection that meet the standard’s requirements. If either system is no longer operational or verified, the basis for that designation no longer exists.
Can a Division classification be converted to a Zone classification?
Not directly. Division 1 combines Zone 0 and Zone 1, which require different equipment categories. Converting the classification requires going back to the data on the frequency and duration of the atmosphere’s presence—information that a Class and Division study does not record.
Todos los Derechos Reservados 2026