Validation

PyFlange is verified against the reference implementation of the polynomial model by Siemens Gamesa (SGRE): a set of Excel workbooks that compute the model of IEC 61400-6 AMD1 step by step. The comparison is automated and runs as part of the test suite, so every change to the code is checked against the reference values.

Validation cases

Case	Flange	Diameter	Gap	Tilt	Shell stiffness
L1	L, 120 × M80	7.5 m	sinusoidal	0°	interpolated / simplified
L2	L, 120 × M80	7.5 m	sinusoidal	0.25°	interpolated / simplified
L3	L, 120 × M80	7.5 m	shape factor 1.2	0°	interpolated / simplified
L4	L	4.6 m	sinusoidal	0°	interpolated / simplified
T5	T	7.5 m	sinusoidal	0°	interpolated / simplified
T6	T	7.5 m	shape factor 1.5	0°	interpolated / simplified
T7	T	7.5 m	shape factor 1.5	1°	interpolated / simplified
T8	T	7.5 m	sinusoidal	1°	interpolated / simplified

Each case is computed for the gap angles 30°, 60°, 90° and 120°, with the AMD1 design gap heights. The 16 case files contain 3264 comparisons in total: the bolt and flange stiffnesses, the gap height distribution, the points and coefficients of the bolt force and moment polynomials, the gap closing forces and the fatigue damage.

A quantity passes when it differs from the reference value by less than 1%. For some quantities, each case file sets a larger tolerance, listed in its tolerances object:

• the shell stiffness interpolated from the FEA database, and the gap stiffness and closing forces derived from it: up to 3%, mostly for the 30° gap, because the workbooks interpolate the database differently;
• the fatigue damage: up to 5.5%, because it amplifies small differences of the bolt stress ranges;
• the bolt moment curve of T-flanges (initial slope c_ini_M and closed-gap moment Ms_min): up to 7%, because the reference workbooks fit a quadratic where PyFlange uses the exact cubic (see below);
• the slope of the bolt force polynomial at P1 (c_ini_true), a diagnostic quantity that is very sensitive to the polynomial coefficients: up to 30%, and more in two cases where the reference value is close to zero.

Run the validation with --failed and a lower --rtol to see these differences.

Running the validation

The validation cases are in the validations folder of the source repository. After cloning it:

py validations/validate.py validations/references/L1-interp.json

validations/references/L1-interp.json
    PASS  gap  30 deg  bolt.fy   [MPa]  expected      900   obtained      900   mismatch +0.000%   (allowable 1.00%)
    PASS  gap  30 deg  bolt.As   [mm²]  expected  4344.07   obtained  4344.05   mismatch -0.000%   (allowable 1.00%)
    ...
    PASS  gap  30 deg  Z2        [kN]   expected  1909.43   obtained  1920.17   mismatch +0.562%   (allowable 1.00%)
    ...

Useful options:

Option	Effect
--failed	Show only the comparisons that fail.
--rtol 0.005	Use a different allowable mismatch.
--excel	Also write the results to copies of the SGRE workbooks in validations/results/ (requires Excel and xlwings).

To run all the cases: py validations/validate.py validations/references/*.json.

Known differences

• The reference workbooks fit a quadratic polynomial through three points for the tensile bolt moment of T-flanges, while PyFlange evaluates the exact cubic of ref. [2], eq. G.61. The validation compares the quadratic coefficients of the workbooks with the quadratic through the PyFlange points.
• The ED2 classes are not covered by the reference workbooks. They are tested against the formulas of ref. [5] in the unit tests.

Unit tests

Besides the validation, the test suite (py -m pytest) contains unit tests of every module, and runs the example scripts of the tutorials.