Fintube HX Community

[ English | 简体中文 ]

An open, reproducible Python core for baseline fin-tube heat-exchanger thermal-hydraulic calculations (MIT Licensed).
Supports forward rating, bounded inverse mass-flow solving, parametric scans, and CSV/Excel exports for water coils, evaporators, and condensers.
Repositories: GitHub · Gitee Mirror | Video Overview: YouTube (EN) · Bilibili (中文)

Ecosystem & Scope Overview

Fintube HX follows a three-tier architecture where Community and Pro share the same thermophysical and baseline solver core (regression-tested and reviewed in three external rounds):

Implemented models

Water geometry uses Wang-Chi plain-fin j/f, Hausen/Gnielinski tube-side convection, Churchill friction and an equal-area annular-fin efficiency approximation. Wet geometry recomputes a wet-fin efficiency proxy from the enhanced air-side coefficient. It is not a resolved partially wet coil model. Version 0.1.0.dev8 adds an optional geometry-mode alternative, model.wet_surface_model: "threlkeld_braun": Braun–Klein–Mitchell (1989) enthalpy-potential ε-NTU with the Threlkeld wet-fin efficiency (Lewis number 1, dry j applied to the wet surface, condensate-film resistance neglected); the coil is taken as the larger of the dry and fully wet heat rates. Omitting the field keeps the legacy contact approximation.

Version 0.1.0.dev9 adds two fin types to the water geometry: geometry.fin_type: "louver" (Wang, Lee, Chang & Lin 1999, IJHMT 42:1945; requires louver_pitch_m, louver_height_m) and "wavy_herringbone" (Wang, Hwang & Lin 2002, IJR 25:673; requires wave_height_m, wave_projected_length_m; fin faces use the developed area sec θ). Both are restricted to the published sample ranges and default to the Schmidt plate-fin efficiency used in the papers' data reduction (geometry.fin_efficiency_method, also selectable for plain fins). Both published j correlations are split at Re_Dc = 1000; for multi-row coils the two branches differ by up to tens of percent at that point (low-Re "level-off" in the data); the equations are used as published. Wet surfaces use the dry j/f. Version 0.1.0.dev10 adds geometry.correlation_range_tolerance (at most 0.01): an explicit, declared extension of the geometric sample bounds of the louver/wavy correlations (never Re or rows); every extended parameter is listed in the result and a "boundary extrapolation" warning is issued.

Version 0.1.0.dev7 adds two optional geometry items: geometry.fin_count (actual counted fins; changes area only, the Wang hydraulic diameter stays periodic) and model.exposed_tubes (unfinned straight tube ends and U-bends standing in the air stream; Churchill–Bernstein single-cylinder crossflow at a declared approach velocity, dry only, heat reported separately to a bypass stream).

Features: JSON input validation, isolated input/result snapshots, CLI single-case and batch execution, parameter scans, CSV/XLSX export, and bounded mass-flow targets for water outlet temperature, refrigerant superheat or subcooling. An outlet-air-temperature target is not implemented.

Quick start

Python 3.12, 3.13 or 3.14 (>=3.12); the pinned dependencies below are tested on all three. From PyPI: pip install fintube-hx. For development, select the interpreter explicitly rather than using the system default:

uv venv --python 3.12 .venv
uv pip install --python .venv --index-url https://pypi.tuna.tsinghua.edu.cn/simple -r requirements-development-lock.txt -e .
# Activate: source .venv/bin/activate (POSIX), .venv\Scripts\activate (Windows)

Alternatively, use python3.12 -m venv .venv (Windows: py -3.12 -m venv .venv), activate it and install with pip. If the mirror lacks a package, use official PyPI with TLS verification enabled.

fintube-hx capabilities
fintube-hx run examples/water_geometry.json --output output/result.json --xlsx output/result.xlsx
fintube-hx scan examples/water_dry.json --parameter air.volume_flow_m3_h --values "[800,1000,1200]" --output output/scan.json
pytest -q

from fintube_hx.engine import run_case
from fintube_hx.storage import load_case
result = run_case(load_case("examples/water_geometry.json"))
print(result["heat_transfer_w"], result["fluid_out"]["temperature_c"])

Input and output conventions

• Temperatures: °C; pressures: absolute Pa; dimensions: m; mass flow: kg/s; air volume flow: actual inlet m³/h; RH: 0–1.
• Provide either fluid.inlet_t_c or fluid.inlet_h_j_kg, not both. Two-phase inputs use P-h; the JSON contract has no inlet-quality field.
• model.segments controls refrigerant axial cells only. Water is lumped: the field may be omitted; legacy supplied values do not subdivide the coil and are reported as not applied in numerical_controls.
• Inverse flow bounds must bracket the desired root. Same-sign endpoints do not prove that an internal root is absent; the solver does not certify monotonicity or global uniqueness.
• Water pressure outputs are scoped straight-tube/core friction, not total installed-coil or system losses; return bends, headers and inlet/outlet fittings are not included. Under the default legacy scope the air side is dry core friction only; the optional model.pressure_scope: "v1" (0.1.0.dev16) adds the acceleration term and the wet-surface increment (see below). Every result lists what is included, excluded, extrapolated or degraded in model_coverage.

Current engineering scope

By default the water wet/dry policy uses inlet water temperature and a contact approximation, not local wall temperature or partially wet area; the optional wet_surface_model: "threlkeld_braun" (dev8) evaluates the enthalpy-potential wet surface with wet-fin efficiency instead. Air properties are inlet-based; water properties use an iterated mean temperature. Wang uses the fin-collar diameter, not the bare-tube OD. For the equal-thickness collar convention this is Dc = OD + 2×fin thickness; actual formed collar dimensions are not independently specified in this baseline. The default fin count is L/fin pitch; fin_count supplies the actual count. The outer-edge geometry is idealized.

Louver and herringbone-wavy fins (dev9), micro-fin tube correlations (dev11, for extensions), optional contact conductance and inner fouling (dev13) and the clean-surface onset-of-frost limit (dev14, tube networks only) are available. There are no actual tube-route networks, row-by-row crossflow, intermediate split/merge junctions, brine properties, transient frost growth or tube-to-tube conduction through shared fins in this core. Evaporation at/below 0°C is rejected unless declared (subzero_evaporation). Correlation range checks reject extrapolation rather than silently clipping geometry.

Known issue (measured): condenser outlet subcooling tends to be over-predicted — by 2.3–3.0 K on the in-range CETIAT points and by 2.5–5 K on the out-of-range Joppolo 2015 condensers (capacity +1 … +9 %). The cause is not yet identified; nothing has been tuned to it (see validation/measured-v6/).

Measured validation

Current set: validation/measured-v6/. It records sources with page/table locators, the D0–D4 grade of every input, the tolerances declared before running, and every result. Nothing is fitted and no multipliers are used.

Known issue: condenser subcooling is over-predicted (by 2.3–3.0 K on CETIAT and 2.5–5 K on Joppolo).

The refrigerant points need a tube-circuit model, which is part of Fintube HX Pro and not of this repository; their Pro results are published for scrutiny.

Why the lumped model and the tube network give different numbers. The lumped model in this repository treats the whole coil as ideal counterflow and cannot represent blind (water-free) tubes. The Pro network follows the actual circuits row by row. On the 48 Tan units computed both ways the means are +5.3 % (lumped) and +4.6 % (network). Almost all of that gap comes from coil pre-03#, which has two blind tubes: the two models differ by 5.3 points there and by 0.04 points on average on the other coils. The headline means (+5.7 % and +3.9 %) also differ because the network additionally covers 14 units (post-03#/04#) that the lumped run does not include.

Reproduce the lumped results with python tools/run_measured_v6_tan.py --check. It is bit-for-bit on the recording platform; across CPUs and maths libraries the capacity deviation agrees within 1e-7.

The earlier v5 set (validation/measured/, six points) is superseded. Its water-coil boundaries were incomplete because of a duct bypass, and its CETIAT points used a plain-fin substitute for the louver fin. Its table is kept there for traceability.

Reproducible comparisons

references/python-hvac/ contains a pinned source reproduction and a mapped water case. The mapped case produces 4363.49 W, versus 4084.46 W in the reproduced reference (+6.83%); this is a software comparison with differing geometry/model definitions, not a measured accuracy result or a fitted calibration.

Version 0.1.0.dev19 adds an optional, declared extrapolation of the wet-friction correlation's fin pitch (plain bare-aluminium fins, 3.2–4.0 mm; flagged as extrapolated, never silent); results without the new key are unchanged.

Version 0.1.0.dev18 widens the supported Python range to 3.12–3.14 with the same pinned dependencies; no calculation changes (all 230 tests, including the bit-for-bit baselines, pass on 3.12, 3.13 and 3.14).

Version 0.1.0.dev16 (core API 2) adds, without changing any legacy result (bit-for-bit checked on 72 Community and 289 Professional cases):

• model.pressure_scope: absent or "legacy_v0" keeps the previous outputs; "v1" reports air_pressure_drop_breakdown_pa (core friction, entrance/exit, acceleration, wet increment) and air_pressure_drop_total_pa. Checked against the originals, the Wang 1999/2000/2002 friction factors were reduced with the entrance/exit losses included, so these are reported as included_in_f and not added again; the acceleration term (1+σ²)(ρ1/ρ2−1) that those reductions subtract is added. v1 requires geometry.fin_surface ("bare_aluminum" or "hydrophilic", no "unknown").
• Wet-surface air friction for non-hydrophilic plain fins: Wang, Lin & Lee 2000 (IJHMT 43:1869, Eqs. 11–15; 31 uncoated samples, Re 300–5000, Fp 1.19–3.2 mm). Hydrophilic coatings, louver and herringbone fins, or values outside the sample range report the wet increment as not_computed with the reason; capacity is unaffected.
• Declared extrapolations: geometry.fin_pitch_extrapolation: true for plain fins 8.7–12 mm (only the pitch and the Dh it implies are extended), and geometry.row_extrapolation: "saturate_at_row_6" with the per-row API air_side_factors_by_row for 7–12-row tube networks (rows 1–6 use the 6-row average, rows 7+ the row-6 local value 6·j(6)−5·j(5)). The lumped models here still reject more than 6 rows. All extrapolated results are flagged and are not validation evidence.
• Structured Chinese errors fintube_hx.errors.ModelRangeError (a ValueError) with field, allowed range, source and suggested fallbacks; planned-but-unsupported items (slit fins, Dc < 6.9 mm, inline banks, flat tubes, mixtures that CoolProp does not provide such as R454B) say so explicitly.

Version 0.1.0.dev17 adds exact (bit-identical) shortcuts for the humid-air states used inside the root solvers, about 1.2–1.9× faster tube networks in the Professional edition with unchanged results, an optional solver.time_limit_s, and a lower bound j6 ≥ 0.5·j(6) for declared row extrapolation (the difference formula is ill-conditioned at low Re and is refused rather than silently distorted).

Version 0.1.0.dev14 adds subzero_evaporation: "initial_frost" support in the wet-surface element (clean-surface onset of frosting: deposit as ice with its fusion enthalpy, piecewise saturation curve at 0 °C; no frost thickness, frost resistance or blockage). The lumped contact-factor refrigerant model rejects it; it is intended for strict wet-surface tube networks. Version 0.1.0.dev13 adds optional geometry.contact_conductance_w_m2_k (fin-collar/tube contact conductance on the tube outer area π·Do·L) and geometry.inner_fouling_resistance_m2_k_w, both constant series resistances with no default value; nothing changes when they are absent. Version 0.1.0.dev12 adds the optional model.subzero_evaporation: "dry_only" for evaporators: surfaces at or below 0 °C are accepted only while they stay at or above the local air frost point (CoolProp humid-air saturation is over ice below 0 °C), i.e. without deposition; otherwise the case is rejected because a frost model would be required. The refrigerant temperature represents the surface in this check, which is conservative. The default (reject) is unchanged. Version 0.1.0.dev11 adds geometry.tube_type: "microfin" with module fintube_hx.microfin: TN 2224 micro-fin geometry (Kedzierski & Lin 2023, NIST), their convective-boiling Nusselt correlation with the mixture glide factor, and the NISTIR 6333 two-phase pressure drop (Choi, Kedzierski & Domanski 1999). Geometric parameters outside the TN 2224 database raise. In this core the lumped water model rejects micro-fin tubes explicitly; the correlations are available to extensions (the Professional evaporator network uses them). They are implementation-verified against the published example and geometry, not yet validated against measured coils. Version 0.1.0.dev10 adds the declared correlation_range_tolerance; without it nothing changes. Version 0.1.0.dev9 adds louver and herringbone-wavy fins and the Schmidt fin efficiency (module fintube_hx.airside); plain-fin cases without fin_efficiency_method are bit-for-bit unchanged. Version 0.1.0.dev8 adds the optional wet_surface_model (module fintube_hx.wet_surface); without it all existing results are bit-for-bit unchanged. Version 0.1.0.dev7 adds fin_count, exposed_tubes and the measured-validation set; without these fields all existing results are bit-for-bit unchanged.

For lumped refrigerant models, an omitted solver.max_iterations defaults to max(200, 2*segments + 20). An explicit limit is never increased; a low budget is disclosed and an unconverged run still exits with code 2. Version 0.1.0.dev6 adds this budget correction; it does not change heat-transfer equations.

The status of every review item (fixed / clarified / open) is tracked in docs/EXTERNAL_REVIEW_STATUS.md. The v3 follow-up records the controlled wet-branch diagnostic and corrected current-version status. The external review response records confirmed issues, corrections and reasons for disputed claims. Version 0.1.0.dev5 corrects fin blockage in the diagonal minimum-flow gap and the misleading inverse-bracket error message; existing transverse-gap reference capacities are unchanged.

Author and citation

Project lead: Zhichao Wang (ORCID 0000-0001-6662-835X). If you use this software in a paper or an engineering report, please cite it as described in CITATION.cff (the Cite this repository button on GitHub exports BibTeX/APA). If it is useful to you, a star on GitHub or Gitee helps other engineers find it. Questions, bug reports and public data: Issues.

Cooperation and enquiries: wangzhichao@chinaibee.com

License

Original code: MIT. Dependencies and adapted python-hvac expressions have separate licences and attribution; see THIRD_PARTY_NOTICES.md, CORE_API.md and CONTRIBUTING.md.