1Introduction
Air handling unit selection is usually described as a sizing problem, and that description is what makes it go wrong. Sizing implies a lookup: state the duty, find the model. But an AHU is built as an ordered assembly of sections, and the properties a customer eventually measures — the pressure the fan develops, the state of the air leaving the coil, the sound at the neck — belong to the assembly rather than to any component chosen from a table.
The practical consequence is a specific and repeatable error. An engineer sizes the coil from a catalogue at a reference air state, sizes the fan against an estimated external static pressure, and adds an allowance for the filters. The unit is quoted. On site the filters load, the actual entering air differs from the reference, and the fan runs at a duty nobody computed. The unit does not fail; it simply does not do what the submittal said.
This paper describes a tool built on the opposite premise: that the section is the unit of selection, and the unit's behaviour is derived from the sections placed in it. The design consequence is that pressure drop accumulates as the engineer builds, the coil is rated at the state it will actually see, and the price is emitted by the engine that did the physics.
1.1What a spreadsheet cannot hold
The incumbent tool is a spreadsheet, and its failure is not arithmetic. A typical AHU selection touches five inter-dependent calculations: coil performance in sensible and latent terms, fan curve fit, motor sizing, filter pressure drop, and acoustic level at the neck and casing. Each is tractable alone. What a spreadsheet cannot hold is the dependency between them — change the coil rows and the pressure drop moves, which moves the fan duty, which moves the motor and the sound.
Nor can it hold provenance. A selection sheet that lives on a sales engineer's laptop, using a coil curve from 2019 that no one has audited in eighteen months, will produce an internally consistent answer that no longer corresponds to the components the factory will build with. The question an auditor asks is not whether the arithmetic is right; it is whether the curve can be traced. The blunt test is whether that trace takes under a minute.
Version drift across regional offices is the same failure in a different costume: several answers, each defensible in isolation, none reconcilable with the others.
1.2Contributions
- A section-level configuration model in which pressure drop accumulates along the air path, so fan duty is derived rather than assumed.
- A costing method that emits the bill of materials from the same engine that produced the performance, per decker and per tag, with SAP part codes.
- A drawing and submittal path in which the schematic the engineer builds is the artefact exported to 2D CAD, removing the redraw step.
- An eight-dimension capability reference framework for evaluating AHU selection tooling by demonstration.
2Background and Related Work
Three regimes govern what an AHU selection must be able to state: the casing and component standards, the filter classification standards, and the fan rating standards. A fourth — ecodesign — converts several of them into a compliance obligation.
2.1Casing and component rating
EN 1886 specifies laboratory test methods, requirements and classifications for the casings of non-residential air handling units — mechanical strength, air leakage, thermal transmittance and thermal bridging. It is a casing standard: it says nothing about what the unit does thermally, only about the box.
EN 13053 covers the other half: requirements, classification and testing for the rating and performance of the unit as a whole and of its components and sections. It applies both in the laboratory and in situ, and to mass-produced and tailor-made units alike — which matters here, because an engineered-to-order AHU is by definition not a catalogue model.
Read together, the two standards describe exactly the architecture this tool implements: a classified casing enclosing rated sections. A selection tool that models the unit as an indivisible model number cannot express that structure, and therefore cannot state a section-level claim an auditor can check.
2.2Filter classification and the pressure band
ISO 16890 has been the international classification basis for general-ventilation air filters since December 2016, and the mandatory basis since 30 June 2018, superseding EN 779. Its innovation is to express efficiency against the ambient particulate fractions — PM1, PM2.5 and PM10 — rather than against a single synthetic test particle, with classes ISO ePM1, ISO ePM2.5, ISO ePM10 and ISO Coarse assigned where a filter separates at least half of the corresponding range across a 0.3 to 10 micrometre spectrum.
Where a project is specified in North American terms, ANSI/ASHRAE 52.2 supplies the MERV reporting system over the same particle-size range. A selection tool serving both markets must hold both, because a filter section specified as ISO ePM1 70% and one specified as MERV 14 are not interchangeable statements.
For unit design the classification matters less than what accompanies it: each filter carries an initial and a final pressure drop. The section is not a single number but a band, and the fan must be selected against the loaded end of it. Sizing to the clean value is the most common way a unit that passed at the factory fails to hold airflow in service.
2.3Fan rating and ecodesign
ISO 5801 specifies procedures for determining fan performance using standardized airways, with rules for converting results across changes in speed, gas and — for model tests — size. ANSI/AMCA 210, published jointly as ANSI/ASHRAE 51, establishes uniform laboratory methods for certified aerodynamic performance rating in terms of airflow rate, pressure developed, power, density, rotational speed and efficiency.
Commission Regulation (EU) No 1253/2014 turns these ratings into an obligation for units placed on the European market, setting ecodesign requirements for ventilation units and distinguishing residential from non-residential units. The regulation is why specific fan power is a design constraint rather than a reporting nicety, and why a tool that cannot compute it at selection time cannot demonstrate compliance.
2.4Failure modes of catalogue-driven selection
- Reference-state coil rating. The coil is rated at a catalogue air state rather than the state the preceding sections actually deliver, so the sensible and latent split are both wrong.
- Clean-filter fan sizing. The fan is selected against initial filter pressure drop, so the unit loses airflow as the filters load through their service interval.
- Assumed external static. External static pressure is entered as a number rather than derived from the sections placed, so the internal contribution is double-counted or omitted.
- Detached costing. The bill of materials is built in a separate sheet from the selection, so a late coil change updates the price or the performance but rarely both.
- Redrawn general arrangement. The schematic is rebuilt by hand in a CAD package, introducing a second description of the unit that can disagree with the first.
All five follow from treating the unit as a catalogue entry. None is fixed by a better catalogue.
3System Overview
The tool is organised as a guided five-step workflow over a governed section catalogue. The steps are project information, unit configuration, section design, bill of materials and costing, and reports and submission. Each step narrows what the next may do, so an infeasible unit cannot be reached by accident.
3.1Unit configuration
Configuration fixes the properties that belong to the unit rather than to any section: certification regime, unit category, air balance across fresh, return and supply, panel build-up, damper face velocities and unit orientation. These are the parameters that constrain what sections may subsequently be placed and in what order.
Fixing the air balance here rather than per section is a deliberate ordering choice. Fresh, return and supply flows are conservation constraints on the whole unit; allowing them to be edited inside a section dialogue would permit an assembly whose mass flows do not close.
3.2Section design
Section design is the core of the tool. Sections are dragged onto the air-flow path and reordered directly, and a live schematic redraws as the assembly changes. The governed catalogue includes inlet and mixing boxes, pre-filters, bag filters, fine filters, HEPA filters, coils, fan sections, empty plena and empty heater sections, among others.
Each section carries its own engineering dialogue rather than a shared property sheet. A mixing-box dialogue configures fresh- and return-air dampers; a filter dialogue captures the initial and final pressure drops in pascals; a coil dialogue takes rows, circuiting and fin spacing. Section ordering and component feasibility are rule-enforced, so a HEPA placed upstream of a pre-filter, or a coil that will not fit the configured casing depth, is refused at the point of placement.
3.3Governance and the section master
Reusable section templates are maintained centrally in a section master, which is what makes a library update propagate rather than diffuse. When research and development releases a new coil family, it is added once and becomes available to every engineer on the next selection; the alternative — emailing a revised spreadsheet — is the mechanism by which a 2019 curve survives to 2026.
Inquiries are governed by branch and financial year through an explicit allocation state machine, with per-engineer attribution. This is what allows an organisation with several regional offices to hold one answer rather than one answer per office.
4Computational Methods
Notation is collected in Appendix A; worked numerical examples in Appendix B.
4.1Pressure accumulation along the air path
The defining computation is the simplest one, and it is the one catalogue tools omit. Sections placed in series each impose a pressure drop at the design airflow, and the fan must develop their sum plus the external system resistance.
For filter sections the term is a band rather than a value, bounded by the initial and final pressure drops the filter dialogue captures. Selecting the fan against the initial value guarantees the unit loses airflow before the filters are changed, so the design duty is taken at the final condition.
4.2Coil duty at the actual entering state
The coil is rated at the air state the preceding sections deliver, not at a catalogue reference point. For an air-side mass flow, the total and sensible duties follow from the enthalpy and dry-bulb differences across the coil, and the latent duty is their difference.
The reason this matters for section-level design is that the entering state is itself an output of the assembly. A mixing box upstream sets it from the fresh and return proportions; a pre-heater shifts it. Rating the coil against a fixed reference discards exactly the information the configuration produced.
4.3Specific fan power
Once total pressure is derived rather than assumed, fan power follows, and specific fan power becomes computable at selection time rather than estimable afterwards.
This is the compliance-relevant consequence of section-level design. Under Regulation (EU) No 1253/2014 the specific fan power of a non-residential ventilation unit is regulated, and it can only be demonstrated if total pressure was derived from the sections actually placed.
4.4Cost roll-up from section to project
The bill of materials is derived from the same configuration object that produced the performance. Each section contributes its components; each component resolves to a part code, a quantity, a unit of measure and a rate.
The per-decker granularity is what makes the output usable downstream. A project total is a commercial number; a per-decker bill of materials with part codes is a manufacturing instruction, and it exports to SAP, Oracle or Microsoft Dynamics over standard formats and REST interfaces at the enterprise boundary IEC 62264 describes.
4.5Schematic and submittal generation
The live schematic is the design artefact, so exporting it is a serialisation rather than a re-drawing. It is written as SVG directly, or through a single plugin as a 2D CAD general-arrangement drawing in DXF or DWG.
The engineering value is not the time saved drawing. It is that there is one description of the unit. Where a general arrangement is redrawn by hand, the drawing and the selection are two artefacts that can disagree, and the disagreement is typically discovered on the shop floor.
5Reported Outcomes and Field Evidence
This section separates three kinds of number, because they carry different weight. Figures published for this module; a deployment result that belongs to a four-module programme; and a modelled scenario. Each is labelled, and each appears in the provenance table with its source.
5.1Figures published for this module
| Measure | Reported | Where stated |
|---|---|---|
| Selection-cycle time | 60-70% reduction, typically within the first quarter | AHU blog post |
| Audit findings after adoption | Zero in the first quarter | AHU blog post |
| Manufacturing lead time | 40% reduction | AHU product page |
| Workflow speed, inquiry to submittal | 10x | AHU product page |
| Adoption | 50+ HVAC manufacturers | AHU product page |
| Time to live deployment | 4-8 weeks | AHU FAQ |
The implementation figure is worth reading alongside the architecture. Four to eight weeks is not integration time; the critical path is loading the customer's own coil, fan and accessory libraries into the section master. That is a direct consequence of the design: a tool whose selections are bound to governed component data cannot be useful until that data is present.
5.2A deployment in which this module was one of four
MileSoft's published HVAC case study describes a deployment at an HVAC OEM in which this module was installed alongside FCU Selection, Chiller Selection and Lifecycle Cost Analysis. The programme reported a 70% reduction in selection-cycle time, zero audit findings since rollout, three times as many proposals per engineer per month, and a 12 percentage-point uplift in proposal-to-order conversion, across four regional offices.
Those four figures belong to the programme, not to this module. Four tools were deployed together with process change; no attempt was made to isolate the contribution of AHU Selection alone, and this paper does not claim it.
What can be said more narrowly is mechanistic rather than quantitative. The reported operator experience — that work which took a week now takes an afternoon, and that submittals are audit-ready without preparation — is consistent with removing the redraw step and emitting the bill of materials from the selection engine, which are properties of this module specifically.
5.3Modelled engineer-hour recovery
MileSoft publishes an open return-on-investment model for this module whose assumptions are stated rather than embedded. It is a modelled scenario, not a measurement, and is labelled as such throughout.
| Parameter | Default |
|---|---|
| Manual selection time | 120 minutes |
| Selection time in MileSoft | 15 minutes |
| Selections per month | 40 |
| Engineer-hours released per year | 840 |
The model's structure is more informative than its output. It attributes the saving entirely to selection time and claims nothing for rework avoided, drawing time, or costing effort — all of which the architecture also affects. It is therefore a conservative floor rather than a projection.
6Discussion
6.1Derived duty is the load-bearing decision
Of everything described here, the decision that produces the rest is that total pressure is derived from the sections placed rather than entered beside them. Once duty is derived, the coil is rated at a real entering state, specific fan power becomes computable, ecodesign compliance becomes demonstrable, and a change anywhere in the assembly propagates rather than requiring a manual sweep.
The corollary is that a catalogue tool cannot be improved into this. Adding more models to a lookup does not make the lookup account for the section upstream of the coil. The change is structural, which is why the section rather than the unit has to be the object the tool manipulates.
6.2Why costing must live in the selection engine
It is tempting to treat the bill of materials as a downstream concern — select first, cost afterwards in the enterprise system. The argument against it is not effort but consistency. Selection and costing describe the same physical object, and when they are produced by different tools they drift at exactly the moments that matter: late design changes, value engineering, and tender revisions.
Emitting both from one engine also changes what a discount means. A discount applied to a per-component build-up is inspectable line by line, so a margin decision can be examined rather than merely accepted. A discount applied to a lump sum conceals which component absorbed it.
6.3A capability reference framework for AHU selection tooling
| Dimension | Question the tool must answer by demonstration |
|---|---|
| D1 Section granularity | Is the object being configured a section, or a model number with options? |
| D2 Derived duty | Add a filter section. Does the fan duty change without anyone retyping it? |
| D3 Filter band | Is the fan selected against the filter's final pressure drop or its initial one? |
| D4 Real entering state | Is the coil rated at the state the upstream sections deliver, or at a catalogue reference? |
| D5 Rule enforcement | Attempt an infeasible section order. Is it refused at placement or discovered at review? |
| D6 One engine, one price | Change the coil. Do capacity and price both move in the same action? |
| D7 Single drawing | Is the exported general arrangement the same artefact as the design, or a redraw? |
| D8 Library governance | Publish a new coil family. How long until every engineer is selecting from it? |
D2 is the single most diagnostic question. A tool that cannot make fan duty move when a section is added is a catalogue with a configurator on top.
6.4Generalisability
The architecture generalises to any engineered-to-order assembly whose performance is a property of its ordering, which in HVAC includes air handling units and, with a different section vocabulary, packaged and rooftop equipment. It generalises poorly to genuinely catalogue products, where the assembly is fixed and a lookup is the correct model. The published figures are specific to AHU manufacturers and MEP consultants and should not be read across to those cases.
7Threats to Validity and Limitations
- Vendor-reported figures. Every quantitative claim in Section 5 is published by the vendor from its own record. None is independently audited, and no controlled comparison against an alternative tool was performed.
- Confounded deployment result. The 70% selection-cycle figure belongs to a four-module programme with concurrent process change. This module's individual contribution was not isolated.
- Range without dispersion. The 60-70% selection-cycle figure is a range across customers with no sample size, distribution, or definition of the baseline it improves on.
- Modelled ROI is not measurement. Appendix B's engineer-hour recovery follows from published assumptions about manual and software selection times. Those times are vendor estimates, not observed distributions.
- Baseline heterogeneity. A 60-70% reduction against a mature spreadsheet estate and against an undisciplined one are different claims; the published record does not distinguish them.
- No cost data. Licence, implementation and library-loading costs are not reported, so no return-on-investment conclusion is drawn.
The architecture in Sections 3 and 4 is reproducible from this text. Section 5 should be read as the vendor's reported experience, and the deployment figure as a programme result.
8Future Work
- Isolating the module's contribution. Instrumenting selection time per module in a multi-module deployment would let the programme figure be decomposed, replacing a shared attribution with a measured one.
- Acoustic prediction at section level. Sound at the neck and casing is currently a design check; deriving it from the placed sections the way pressure already is would close the last assumption in the assembly.
- Filter loading over the service interval. Selecting at the final pressure drop is conservative but static; modelling the loading trajectory would let energy over the change interval be reported rather than bounded.
- Ecodesign declaration output. Specific fan power is computed at selection time but is not yet emitted as a conformity declaration in the form Regulation (EU) No 1253/2014 anticipates.
- Measured versus predicted follow-up. No published data compares the submittal's stated performance against commissioning measurements. That comparison is the only test that would settle whether derived duty closes the gap it is designed to close.
9Conclusion
An air handling unit behaves as an assembly, and a selection tool that models it as a catalogue entry will mis-state the quantity that gets measured on site. This paper has described a tool built on the opposite premise — the section as the unit of selection — and traced the consequences: pressure accumulates along the air path so fan duty is derived, the coil is rated at the state the assembly actually delivers, specific fan power becomes computable at selection time, and the bill of materials is emitted by the same engine that produced the performance.
The published record for this module reports a 60-70% reduction in selection-cycle time with zero audit findings in the first quarter, a 40% reduction in manufacturing lead time, and adoption by more than fifty HVAC manufacturers. A four-module deployment reported larger programme figures, which this paper deliberately does not attribute to this module alone.
The capability reference framework of Section 6.3 is offered as the durable contribution, and its second question is the one that separates architectures rather than vendors: add a filter section, and see whether the fan duty moves on its own.
Appendix ANomenclature
| Symbol / term | Meaning |
|---|---|
| V(flow) | Design volume flow rate through the unit |
| dP(total) | Total pressure the fan must develop |
| dP(external) | External system resistance, outside the unit |
| dP(i) | Pressure drop across section i at the design airflow |
| dP(filter, initial/final) | Clean and loaded pressure drop bounding a filter section's service band |
| m(air) | Air-side mass flow rate |
| h(in), h(out) | Specific enthalpy of the air entering and leaving the coil |
| T(in), T(out) | Dry-bulb temperature of the air entering and leaving the coil |
| cp(air) | Specific heat capacity of moist air at constant pressure |
| Q(total), Q(sensible), Q(latent) | Total, sensible and latent coil duty |
| SHR | Sensible heat ratio, Q(sensible) divided by Q(total) |
| SFP | Specific fan power, fan power per unit volume flow |
| eta(total) | Combined fan, drive and motor efficiency at the operating point |
| q(c), r(c) | Quantity and unit rate of component c in the bill of materials |
| C(unit) | Unit price after component roll-up, discount and margin |
| AHU | Air handling unit |
| BOM | Bill of materials |
| MERV | Minimum Efficiency Reporting Value (ANSI/ASHRAE 52.2) |
| ISO ePM1 / ePM2.5 / ePM10 | ISO 16890 filter classes by particulate fraction |
Appendix BWorked Numerical Examples
Appendix B.1Why the filter band changes the fan selection
A unit is designed for 8,000 m3/h against 250 Pa of external resistance. The assembly is a mixing box (40 Pa), a pre-filter rated ISO Coarse (initial 55 Pa, final 250 Pa), a fine filter rated ISO ePM1 (initial 105 Pa, final 450 Pa), a cooling coil (160 Pa) and an attenuator (35 Pa).
Applying Equation (dp) at the clean condition: dP(total) = 250 + 40 + 55 + 105 + 160 + 35 = 645 Pa. At the loaded condition, applying Equation (dpfilter): dP(total) = 250 + 40 + 250 + 450 + 160 + 35 = 1,185 Pa.
The difference is 540 Pa, or 84% above the clean figure. A fan selected at 645 Pa does not merely lose a little airflow as the filters load — it is selected for a duty the unit spends almost none of its service life at. The filters alone move from 160 Pa to 700 Pa, which is why the band, not the value, is the design input.
Applying Equation (sfp) with a combined efficiency of 0.62 at the operating point: SFP at the loaded condition is 1,185 / 0.62 = 1,911 W per m3/s, against 645 / 0.62 = 1,040 W per m3/s clean. A compliance statement made on the clean number understates the regulated quantity by nearly half.
Appendix B.2How the mixing box moves the coil duty
The same unit handles 8,000 m3/h at an air density of 1.2 kg/m3, so the air-side mass flow is 8,000 x 1.2 / 3,600 = 2.667 kg/s. Take a specific heat of 1.02 kJ/kg-K.
Case A, rated at a catalogue reference of 27 degrees Celsius dry-bulb and 55.5 kJ/kg enthalpy, leaving at 13 degrees and 36.0 kJ/kg. Applying Equation (coil): Q(total) = 2.667 x (55.5 — 36.0) = 52.0 kW; Q(sensible) = 2.667 x 1.02 x (27 — 13) = 38.1 kW; Q(latent) = 13.9 kW; SHR = 0.73.
Case B, rated at the state the mixing box actually delivers — 30% fresh air at 38 degrees and 95.0 kJ/kg mixed with 70% return at 24 degrees and 47.5 kJ/kg, giving 28.2 degrees and 61.7 kJ/kg — leaving at the same 13 degrees and 36.0 kJ/kg. Q(total) = 2.667 x (61.7 — 36.0) = 68.5 kW; Q(sensible) = 2.667 x 1.02 x (28.2 — 13) = 41.4 kW; Q(latent) = 27.1 kW; SHR = 0.60.
Total duty is 32% higher and the sensible heat ratio falls from 0.73 to 0.60. The second number is the operationally dangerous one: a coil selected for SHR 0.73 and asked to deliver 0.60 will not dehumidify to the design condition, and the complaint arrives as humidity rather than as temperature. Nothing in Case A is arithmetically wrong; it is simply rated at a state the unit will never see.
Appendix B.3The published engineer-hour model
MileSoft's published model assumes a manual AHU selection takes about 120 minutes against about 15 minutes in the tool, at 40 selections per month.
Time saved per selection is (120 — 15) / 60 = 1.75 h. Annual selections are 40 x 12 = 480. Engineer-hours released are 1.75 x 480 = 840 h per year, which is roughly half a full-time engineer.
Two caveats belong with the figure. The 120 and 15 minute times are vendor estimates rather than observed distributions, and the model claims nothing for rework avoided, drawing time, or costing effort — so it is a floor rather than a projection. It is reported here because its assumptions are published and adjustable, which is the property that makes a vendor ROI model examinable at all.