Run life cycle¶

Start-up checks¶

Before the flow exists, at start-up, GALE checks:

the fixed step is a whole multiple of each OpenFAST DT;
CompInflow = 2;
Wake_Mod matches the sampling mode, and TwrShadow = 0 with a forced tower.

Each distinct model is read on one rank (erf_openfast::gather_model_facts) and the results are shared. The solver requirements (anelastic, fixed step, traps off with OpenFAST) are listed in Limitations.

Start-up audit¶

At the first step, once the node layout, the flow at the nodes and ERF’s density at the hubs are known, every OpenFAST turbine is audited. Findings are printed together under OpenFAST input audit for <name>. A fatal finding aborts after the whole list is printed.

Fatal	Warning
Model `AirDens` differs from `air_density`	Model gravity differs from ERF’s
ERF’s hub density differs from the model’s by more than `density_tolerance` (an ABL base state thins with height, hence the tolerance)	Forced tower with `TwrAero` off
`CompAero = 0` while loads go into the flow	Fewer than 8 cells across the rotor
Rotor below the ground, above the top or across a non-periodic side	A kernel in use narrower than a cell, named by its key
Base outside the domain	Line points farther apart than `epsilon`; tower points farther apart than `epsilon_tower`
Two rotors whose swept discs intersect	`num_force_points_tower > 0` with a model that has no tower nodes
`nacelle_cd × nacelle_area` above \(2\pi\epsilon_{nacelle}^2\)	Rotor kernel cut by a non-periodic boundary; rotor facing away from the wind or yawed more than 30°

For a disk the audit also notes rings whose points lie more than two kernel widths apart, where the ring’s force breaks into separate blobs. The stub’s deck is not an OpenFAST primary file, so its model-file checks are skipped with a note. The tip-travel limit and the hub-axis check follow at later steps.

Diagnostics¶

The files are listed in Outputs. Some definitions:

Thrust is the sum of the hub and blade node forces as OpenFAST reports them: the force of the fluid on the structure, so along the inflow. Power is torque times rotor speed.
Hub axis is the shaft direction. It follows yaw, tilt and the tower’s deflection.
Total load is thrust plus tower plus nacelle, all on the structure.
momentum_source.csv equals the sum of the forces on the fluid: minus the turbines’ total loads (for an actuator disk, minus the thrust’s shaft component), plus the prescribed-Ct disks’ \(-T\mathbf{n}\).
_erf and _flow rows carry the time at the end of their step; their first row is the start-up state. Files that describe a step’s forcing carry the time at its start.

Wake lines¶

wake.lines_xD puts sampling lines behind every rotor (OwnedWakeLines, the owner-written copy of the actuator core’s WakeLines).

Each distance is along the horizontal projection of the rotor axis through the hub. A wake follows the wind at hub height, and a shaft tilt of a few degrees would otherwise carry far lines into the ground.
At each distance: one horizontal line normal to that direction and one vertical line, each of num_points points spanning half_width diameters either side. The vertical line is clipped at the ground; each point’s offset s is in the file.
Lines are built at the first sample from the rotor’s diameter. Every point must lie in the domain.
Every wake.int steps the velocity is written to <output_root>_wake.csv (time, xD, line, s, x, y, z, u, v, w) and, from avg_start, accumulated into <output_root>_wake_avg.csv.
The running sums and the line layout are checkpointed. A restart with another layout stops.
Nothing is sampled when prescribed_velocity replaces the flow.

Statistics¶

From avg_start, every step adds each body’s state to its running statistics. <output_root>_stats.csv holds the sample count, first and last times and, per quantity, the mean, RMS, minimum and maximum.

Turbine: thrust along the shaft and along x, torque, power, rotor speed, hub velocity and blade-mean streamwise velocity.
Prescribed-Ct disk: upstream and disk speeds, thrust and power.

The sums are checkpointed, so statistics continue across a restart. These are the numbers a turbulent-inflow LES is compared on.

Checkpoint and restart¶

An ERF checkpoint carries the bodies’ state under <chk>/moving_bodies:

File	Contents
`state`	Step count, start time, OpenFAST’s stop time, anchor level and step, every body’s placing and forcing inputs, each turbine’s OpenFAST time index.
`<name>.chkp`	OpenFAST’s own checkpoint (`FAST_CreateCheckpoint`), written by the owner.
`<name>_nodes.dat`, `<name>_fllc.dat`	Node loads and the FLLC state.
wake and statistics files	Running sums.

With an anchor level above 0, ERF writes the source the coarser levels need to <chk>/StructureCoarseSource, as for every structure model.

On restart:

Owners are chosen again from the new boxes; a restart may use another number of ranks.
Turbines are restored with FAST_ExtInfw_Restart. ERF checks that OpenFAST’s time index is the one its own checkpoint expects.
The anchor level’s sources are rebuilt at the next step from the restored loads. Coarser levels take theirs from the stored source.
With fllc = automatic the kernel width is set again from the restored rotor.
Logs keep their rows up to the checkpoint and append from there. In a clean directory they start afresh.
A checkpoint written without bodies (a precursor) can be restarted with bodies; they start afresh at its time.

The constraints on a restart are in Limitations.

Timing¶

Every phase is a BL_PROFILE region. A -DERF_ENABLE_TINY_PROFILE=ON build lists their times at the end of the run (min, mean and max over ranks); in other builds they cost nothing.

timing_int = K prints, in any build, every K anchor steps (counted across restarts), the wall-clock seconds per phase since the last summary, as min, mean and max over ranks, and the fastest, mean and slowest turbine’s FAST_CFD_Step time.