Repository navigation
Hydraulic network - #2
Closed
lamdangelo wants to merge 42 commits into
Closed
lamdangelo wants to merge 42 commits into
lamdangelo wants to merge 42 commits into
Conversation
Stage 0 of the hydraulic field-circuit coupling: a modified-nodal-analysis solver for a lumped pressure/mass-flow network (nodes, pipe/valve/pump branches), sharing the CoolProp property interface and friction-factor models with the 1D channel solver and mirroring its theta/BDF2 time integration. This is a standalone module with no wiring into the conductor solver yet; that bordered (Schur-complement) port coupling is stage 1. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Stage 1 of the field-circuit coupling: channel ends declared as ports in the hydraulic_network section are coupled to network nodes through a monolithic bordered (Schur complement) solve around the existing banded system. Ports are pressure-driven overlays on the standard boundary conditions: the port-end pressure row becomes the continuity constraint p_port = p_node with the network node column as a border, and the port mass flow rho_old*A*v enters the node mass balance implicitly as the Lagrange multiplier of that constraint. The network is solved in increment form to avoid the cancellation of a zero-pressure particular solution, inherits the conductor's time integration method (coupled node rows forced backward Euler under Crank-Nicolson/Galerkin), and is initialized by a steady solve with the conductor's initial port flows. The degenerate-limit test runs CASE_1 with the CHAN_1 outlet discharging through an almost ideal valve into a reservoir at the imposed outlet pressure and matches the fixed-boundary baseline to solver roundoff; the committed CASE_1/2/3 regressions are unchanged (the uncoupled path is untouched). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Stage 2 of the field-circuit coupling: both ends of a channel (and several channels sharing manifold nodes) coupled to a pump + bypass loop. The stage-1 bordered solve already handled multiple ports; the changes fix the two issues the pump loop exposed: - Network initialization now honors the declared initial node pressures: branch flows are initialized from each branch's own characteristic at those pressures (initialize_branch_flows_from_pressures) instead of relocating the nodes to the network's steady operating point, which applied a pressure step to the coupled channel ends on the first time step. Residual node imbalance is resolved by the transient itself, through manifold compliance volumes or the algebraic junction balances. - resolve-time warning when a ported channel has hydraulic-parallel partners (open interface) not ported to the same node at the same end: a differential plenum pressure between openly connected channels is strongly amplified by the transverse exchange terms (measured on CASE_1: a common-mode 10 Pa boundary shift responds with ~1e-5 K, the same shift on one channel only with ~1 K scale), so parallel groups should be manifolded together. Two new integration tests: a droop-free pump with a near-ideal drain reproduces the INTIAL = 1 imposed-pressure-drop baseline to solver roundoff, and the boosted CASE_1 heat pulse redistributes ~70% of the channel intake into the bypass while the pump rides up its droop curve -- the loop physics no fixed-boundary run can represent. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The node pressures and branch mass flow rates of the coupled hydraulic network are recorded every time step and written to hydraulic_network_te.tsv in the coupled conductor's Time_evolution directory, following the conventions of the channel inlet/outlet record (header at initialization, rows appended in CHUNCK_SIZE batches or at the end of the transient). Node temperatures are prescribed inputs until the network-side energy transport stage and are not recorded. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Stage 3 of the field-circuit coupling: conductors coupled to the network keep their own banded systems and meet only in the shared dense Schur complement -- each contributes C_k A_k^-1 B_k from its stacked-right-hand- side banded solve, one small dense solve closes all conductors and the network together, and the network state advances exactly once per step. To support this, the transient step is split into its natural phases: assemble_thermal_hydraulic_system (through boundary conditions and row scaling), the banded solve, and finalize_step (error estimate, history, checks, solution reorganization). step() composes them for uncoupled conductors and rejects ported ones; the simulation loop routes coupled conductors to solve_coupled_conductors_step after preparing every conductor, and forces the smallest proposed time step on all coupled conductors (they share the network unknowns). The network build validates that coupled conductors share the time integration method, and the network state file is written once per step by the first coupled conductor. The two-conductor test clones the CASE_1 conductor and discharges one channel of each into a shared manifold: in the degenerate limit both conductors match the fixed-boundary baseline and the recovery line carries the sum of the two outlet flows. The committed regression suites are unchanged, confirming the step() split is behavior-preserving. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Stage 4 of the hydraulic field-circuit coupling: internal node temperatures become dynamic. After every pressure/flow solve a segregated, fully implicit step of the node enthalpy balances advances them -- inflows carry the upstream temperature (upwind by the just-computed flows, branch cp frozen at the upstream state), outflows carry the node's own well-mixed temperature and cancel, zero-volume nodes reduce to the algebraic flow-weighted inflow mixture. The resulting temperature rate feeds the thermal-expansion part of the node mass storage, d(rho V)/dt = V (rho kappa_T dp/dt - rho beta dT/dt), into the next step's mass balance as a lagged source. Conductor ports report their new-state mass flow and temperature to the balances, so back-flow and recirculation now carry physically consistent enthalpy into the manifolds and back into the channels through the existing inflow temperature boundary condition. Node temperatures join the pressure and flow columns of the network state output. The energy transport changes the coupled pulse scenario qualitatively: the strong heat pulse makes the channels expel hot fluid into the manifolds, whose near-incompressible helium (thermal pressure coefficient beta/kappa_T of order 1e6 Pa/K) pressurizes the rigid loop by tens of kPa, pushing the pump backward through its droop curve and venting the expansion surplus through the drain -- enthalpy the static node temperatures used to discard silently. The pump-loop test is renamed and rewritten around that complete physics; the degenerate-limit regressions are untouched and new unit tests pin the algebraic inflow mixing and the volume-node relaxation time constant to closed-form references. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…state Valve branches accept a reverse_linear_resistance (check-valve semantics, switched on the previous-step flow sign); reservoirs can be re-stated at runtime (set_reservoir_state) and internal nodes accept external mass/enthalpy sources, so drivers can impose measured boundary schedules and tie-ins. Node enthalpy balances carry the transported energy through junctions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Auxiliary current/field tables can be a single csv/tsv keyed by the component identifier in the first column, equivalent to the legacy one-sheet-per-component xlsx layout (ragged blocks padded and trimmed). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…grity Simulation accepts a step_callback invoked at the top of every transient iteration (drivers use it for measured boundary schedules, quench detection and source switching); a truthy return stops the run. Runs that stop before TEND now flush the partially filled time-evolution buffers and reorganize the spatial files by the saves actually performed, so no recorded data is lost and plots label the true save times. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ents File-driven patch exchange between winding-geometry-adjacent positions (turn-to-turn, layer-to-layer, and across conductors) that the 1D metric cannot see: each CSV row maps a local x-interval onto a partner component interval and exchanges g_eff*(T_partner - T_local) per metre, with an optional (min(T_mean, T_sat)/T_ref)^n law for the insulation conductivity. Patch pairs listed on both partners are exactly antisymmetric (previous-step fields on both sides). A reserved BOUNDARY partner couples patches to a fixed-temperature reservoir (casing sink). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Opt-in strand/jacket heat sources for the field-transient losses, all
default-off and bit-identical to the prior model when disabled:
* coupling loss (n*tau/mu0)*(dB/dt)^2*A via coupling_loss_time_constant
(refactored to the shared per-step _field_rate helper);
* eddy-current loss sigma(T[,B])*(dB/dt)^2*C via
eddy_loss_geometry_constant (jacket-material sigma(T) on jackets,
Cu sigma(T,B,RRR) on strand matrices; material dispatcher in
properties_of_materials.electrical_conductivity);
* NbTi hysteresis loss (2/3pi)*Jc(B,T)*d_f*|dB/dt|*A_sc via
filament_diameter (linear in |dB/dt|: energy set by the field swing).
A FROM_FILE field profile can scale with I(t)/I0
(magnetic_field_scales_with_current), which drives the dB/dt of these
sources during a discharge. All sources (incl. the transverse coupling)
are wired into the nodal/Gauss heat-source assembly, the yaml schema and
the input loaders.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
gitignore treats '#' as a comment only at the start of a line, so patterns like 'TDD_examples/**/ # input files ...' were literal (unmatchable) paths and their targets showed up as untracked. Move every trailing comment onto its own line. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A pressure-triggered by-pass with hysteresis (quench protection): kind: relief_valve with trip_pressure / reseat_pressure / monitored_node (default: the from-node) in the network input. Blocked until the monitored node exceeds the trip, then a one-directional valve with its open-state resistances until the pressure falls below the reseat. State switches once per step on the previous-step pressures, mirroring the check-valve convention. Replaces the fragile driver-managed pattern (a plain closed valve mutated by a step callback keyed on the branch NAME), which left model files non-self-describing and silently unprotected under drivers without the callback. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two new YAML-only simulation settings: autosave_interval stores a checkpoint of the complete evolving state every N accepted time steps (atomic npz under Checkpoints/, two most recent kept, time-evolution buffers flushed first so output files are always complete up to a checkpoint); restart resumes an interrupted run from the newest checkpoint after the normal deterministic initialization, truncating the output files to the checkpoint time. The checkpoint state surface covers the time-integration history (theta/BDF2/AM4 levels), per-conductor counters and diagnostic accumulators, coolant primary fields, solid multi-level heat-source arrays and the dB/dt cache, the electric solution and sub-step counters, and the full hydraulic-network state including the relief-valve hysteresis latches. A restarted run continues bit-identically to an uninterrupted one (regression-tested per-file by sha256 for a plain and a network-coupled CASE_1 scenario). Restart refuses finalized runs (post-processing consumes the raw spatial saves), missing checkpoints, and manifests written by a different integration method or checkpoint schema. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… conjugation
New per-conductor input HYDRAULIC_FORMULATION (auto | velocity |
mass_flow, YAML-only). The mass-flow formulation makes the channel mass
flow rate the first fluid unknown - the native variable of the hydraulic
network coupling - by an in-place similarity transform M^-1 X M of the
assembled Gauss-point flux/upwind/source blocks with the analytic
change-of-variables Jacobian frozen at the linearization state
(hydraulics/formulation.py). The fluid transient block is the identity
and the fluid load-vector rows are zero, so both are invariant; boundary
rows are overwritten after assembly and carry the native Dirichlet value
(an imposed mass flow becomes exact instead of lagged-density).
auto resolves at setup: network coupling declared -> mass_flow, else
velocity; explicit velocity on a coupled conductor warns; conductors
sharing a network must resolve identically. In mass-flow mode the port
coupling coefficient is the exact +-sigma (no frozen-density Picard
linearization), the port flow report reads the native unknown, Reynolds
numbers use the density-free mass-flow expression, and the adaptive-step
flow-slot floor is the dimensional image rho*A*(1 m/s) of the velocity
floor.
The velocity path is untouched (guard hook only) and verified
byte-identical on the full CASE_1 decks, yaml and excel. Eigenvalue
invariance {v, v+-c} and kernel-vs-dense-reference agreement are unit
tested; the coupled degenerate limit matches the fixed-pressure baseline
to the 1e-6 roundoff floor under BOTH formulations.
Known, measured limitation (documented in the tests and dev notes): the
frozen-M route omits the coefficient-derivative terms
M^-1 (dM/dt + A dM/dx) W', so during density transients the two
formulations differ at O(v*drho/rho) in the flow fields (~12-14% on the
100 W/m CASE_1 pulse), invariant under mesh AND time-step refinement;
steady/near-steady agreement is 1e-6..5e-3. Mass-conservation drift on a
flow-reversal scenario (BE-consistent quadrature): velocity 6.19e-2,
mass_flow 3.24e-2 of inlet throughput.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every run - regardless of the active hydraulic formulation - now writes the channel mass flow rate alongside the velocity: a mass_flow_rate column in the spatial-distribution files (and their reorganized per-property form) and a per-channel mass_flow_rate time-evolution file. The inlet/outlet records and the Solution files already carried both. Existing columns are unchanged (verified column-wise byte-identical against the pre-change CASE_1 baselines, yaml and excel). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Key reference: the new conductor input (auto resolution rule) and the two simulation-level checkpoint settings. Developer notes record the M / M^-1 definitions, the conjugation route and its verified invariants, the measured frozen-M transient limitation (refinement-invariant O(v*drho/rho) flow difference, with the numbers), the flow-slot semantics table, the checkpoint/restart design, and the restart formulation policy (fail on mismatch, no state conversion). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…rmulation The pump-loop degenerate limit, the shared-node two-conductor test and the heat-pulse redistribution test compare against uncoupled baselines (or validate loop physics) that predate the formulation selection; with auto now resolving coupled decks to mass_flow they measured the known cross-formulation transient difference (1.2e-1 / 5.3e-4) instead of the coupling mechanics, and the boosted 500 W/m pulse drives the mass-flow coupled loop to an invalid negative pressure at t = 0.7 s (the mass-flow counterpart of the velocity path's own blow-up threshold on this pulse, documented in the dev notes). Pinning both runs of each test to velocity restores their original meaning; the mass_flow coupled path stays covered by the parametrized single-port degenerate test and the formulation suite. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ot the Jacobian The Gauss-point source block multiplies the state, not its derivative, so the chain-rule similarity M^-1 S M evaluates the source at a parasitic state (1 + beta*T - kappa_T*p ~ 2.5 for supercritical helium at 6 K), inflating the effective channel friction and producing an 18-27 % steady flow deficit on the AAB51 deck. The right factor must be the Picard secant P = diag(1/(rho*A), 1, 1), which reproduces Phi(W') exactly at the frozen state. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A hard Dirichlet on the inflow temperature injects a many-kelvin step in one time step when a port reverses against a quench-heated header; the consistent-mass Galerkin scheme answers with spatial ringing that can drive the neighbouring node below 0 K. The imposed value now relaxes from the channel end temperature toward the node temperature at the physical boundary-cell renewal rate v*dt/dz, reducing to the hard Dirichlet whenever v*dt >= dz. Built from current fields only, so checkpoint identity is preserved. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
FluidType.CONSTANT serves user-given constant properties (beta = 0, so kappa_s = kappa_T and cp = cv) from the coolprop_interface entry points, letting the solvers run against closed-form solutions unchanged. The suite covers Kirchhoff splits, RLC discharge, water hammer, transmission line and thermal lag, with runner scripts under scripts/verification. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ewton New opt-in ELECTRIC_CURRENT_CONSISTENCY conductor flag: the quasi-static electric solve iterates F(x) = K(|I|)x - b = 0 with a damped Newton method so the power-law strand resistances are exactly consistent with the solved network currents. The Jacobian replaces only the diagonal resistance block with per-regime differential resistances (SC n*R, sharing via implicit differentiation of the current divider, normal R_stab, chained jacket-parallel combination), keeping K's sparsity and the cached RCM permutation; an Armijo backtracking line search globalizes the iteration and blockwise volt/ampere tolerances with an E0*L voltage floor define convergence. Components store their electric regime partition and strand-only resistance for the derivative builders; get_current binds current_for_resistance aliases so the loop can rebind solved currents without touching op_current semantics. Equipotential groups containing a fixed-potential node are now grounded joints (all members take the fixed value) instead of merged free DOFs. Also fixes the ind_array.any() guards in strand_mixed get_electric_resistance (an index set containing only element 0 was silently skipped) and exempts mode-none strands from imposed-current bookkeeping. Replaces the Picard fixed-point loop, which oscillates without converging (100-cap, ~4 min/step) once a real quench creates an extended dissipative zone at power-law exponent n = 20. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The quasi-static path never advanced conductor.electric_time (only the transient sub-stepping loop does) and never rebuilt the nodal current-source vector after initialization, so CURRENT_IS_FROM_FILE waveforms were interpolated at t = 0 forever: time-varying transport currents - dump/discharge templates in particular - were silently frozen at their initial value for every steady-solver run. Set the electric clock to the thermal time before the steady solve and rebuild the current source via eval_total_operating_current after the operating-condition update, mirroring the per-sub-step sequence of the transient loop. Constant-current decks are bit-identical (the rebuild rewrites the same two boundary entries). Found by a dump-time-constant scan whose tau = 0.15 s and tau = 0.65 s transients were bit-identical. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…search A wildly scaled Newton direction near a singular Jacobian (~zero transport current during a dump tail) crashed the run twice over: first scipy.bisect probes the current-divider residual with Python floats, where i**n raises OverflowError instead of saturating; with that saturated, the stalled line search then adopted a non-finite iterate whose Joule power crashed the thermal banded solve. The divider residual and its derivatives now compute through np.float64 under errstate(over=ignore) so overflow yields a sign-correct inf, the line search rejects non-finite trial norms, and if every backtracked trial is non-finite the solver keeps the best iterate (currents rebound for consistent resistances) with a warning; a non-finite Newton step bails out the same way. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New SimulationSummary writes simulation.log at run end (runtime timers, feature flags, per-solid extrema) plus a per-step hotspot_log.tsv with per-strand max temperatures and normal-zone lengths (from the electric regime partition). A Normal zone section reports onset, maximum length and the normal-zone propagation velocity, fitted as half the growth rate over the 25-75 % window and restricted to t past heater-off with growth beyond the disturbance-made baseline - heater ignition of the heated zone otherwise reads as an absurd 60-90 m/s front. Restart re-seeds the extrema from the restored state. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The final stop-time save has no scheduled num_step_save slot, so the spatial reorganization indexed past the performed saves and aliased the step-0 file (unwritten slots are 0). _performed_saves now returns the per-save step numbers, appending the final step counter for the stop-time save. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
All integration methods now share the simulation_results/<name>/... folder instead of per-method BE/CN/AM4/GAL/BDF2 subtrees (the checkpoint manifest already guards restarts against a method change), and the legacy GUI results root is no longer created eagerly on construction - headless runs override it with the model directory anyway. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
get_registry's suffix heuristic sent existing directories like "..._return1.8L" to their parent, resolving the registry against the wrong folder. An existing directory is now always taken as-is; the heuristic only serves (possibly non-existent) workbook paths. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Figures and tables under analysis/sensitivity/results* are exempt from the global *.png / *.csv rules; the simulation cache and per-sample run clones stay untracked. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ts fit range The NIST copper magnetoresistance correction is a log10 polynomial fitted for xxx >~ 1. Extrapolated to vanishing field (a field map scaled by an exponentially dumped current) it diverges, 10**aaa overflows to inf and poisons the copper conductivity and the thermal system matrix. The correction at xxx = 1 is only ~2e-3, so it is skipped below that, in copper.py and in the duplicate inside magnesium_diboride.py. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ty dispatch electrical_resistivity_of accepts 'ss' / 'stainless_steel' and returns the existing stainless-steel correlation, so jackets can be used in the eddy-current loss and the resistance floor like copper and Al6063. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ernel assemble_system_matrices looped over (local row, local column) pairs with fancy indexing per pair. The scatter is now one strided-view addition per local row in pure numpy, and a compiled numba kernel (about 2.5x faster at 10^4 elements) when numba is importable. Both reproduce the original accumulation bit for bit (each band entry receives at most two contributions), which tests/test_band_assembly.py checks against the reference loop. numba is added to the dependencies; the numpy scatter is the fallback without it. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…e -1) ConductorMesh accepts a mesh_file (grid key MESH_FILE, resolved against the deck directory) for ITYMSH = -1: a one-column text file with the node z coordinates, validated (finite, strictly increasing, 0 to the conductor length, MAXNOD when given), with NELEMS derived from the file. The refinement keys may be omitted on such a deck. Straight and helicoidal component coordinates follow the file nodes; the electric topology is built once since the mesh is static. The legacy per-component external_grid workbook path is kept as the fallback when no mesh_file is given. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…mulation A second route to the mass-flow unknowns, selected by the boolean conductor input explicit_mass_flow_formulation (EXPLICIT_MASS_FLOW_ FORMULATION, default false): the Gauss-point flux-Jacobian, upwind and fluid-row source blocks are assembled from the written-out (mdot, p, T) coefficients in hydraulics/mass_flow_equations.py instead of conjugating the velocity-form blocks at runtime. The resolved formulation is the new internal HydraulicFormulation.MASS_FLOW_EXPLICIT (not an input value); MASS_FLOW_FORMULATIONS groups both mass-flow routes wherever the first fluid slot is the mass flow rate (boundary conditions, initial seeding, network ports, error controller floors, solution reorganization). OPENSC2_DEBUG_CHECKS=1 asserts the heat-source consistency q_p = phi rho c_v q_T on the assembled block. Checkpoints now store the native mass flow rate (restored, or rebuilt from v and rho(p, T) for older archives): a restart under an evolving flow otherwise refreshed the velocity and the Reynolds number from the stale initialization value on its first step. With the flag off every deck is bit-identical. Unit tests check the eigenstructure and the similarity to the transform route, the integration tests the steady channel, the pressure pulse and CASE_1; docs/mass_flow_formulation_notes.md records the measured differences (stabilisation block only, open TODO). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
… under network coupling tests/verification/formulation_suite/ compares the velocity (v, p, T) and the explicit mass-flow (mdot, p, T) formulations on a W7-X NbTi channel, three conductors in parallel between two volumes fed through linear valves: steady flow split, low-Mach acoustic step, symmetric and asymmetric heating with Richardson extrapolation, and a manufactured solution. The latter uses the new opt-in Conductor.fluid_source_callback hook, which adds nodal sources to the fluid rows of the Gauss-point source vector and is None on every production path. Solver-free references (steady compressible network ODE, linear transmission line with characteristic closure, MMS residual) live in tests/reference/ with their own unit tests. scripts/run_formulation_matrix.sh runs the full matrix, scripts/make_plots.py draws figures/ from results/ (both ignored). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…solve New conductor operation electric_resistance_floor (ELECTRIC_RESISTANCE_ FLOOR, default 0 = off). Below Ic the power law gives R -> 0, so the split of the transport current among parallel superconducting strands or stacks has zero residual for any loop current and the steady Newton solve drifts from step to step until a loop current exceeds Ic (cold-DL drift seen on the HELIAS-VIPER 2D decks). With the floor the Gauss-element resistance of the superconducting regime is bounded below by floor * E0 * L_e / Ic, which pins the split to the resistive share and leaves the sharing and normal regimes untouched; the differential resistance is R_floor on floored elements. Applied by the mixed-strand and the stack resistance builders; tests/test_resistance_floor.py checks the floor, the derivative and the no-op without the key. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…-stack hysteresis Coupling loss: coupling_loss_time_constant (n*tau) may now be a list, one entry per loop family, with the new key coupling_loss_relaxation_time (tau, scalar or list). With tau > 0 the coupling magnetisation follows tau dM/dt + M = -(n*tau/mu0) dB/dt (backward Euler over the field-rate sampling interval) and q = mu0 M^2/(n*tau) A, so an exponential dump deposits the fraction n*tau/(tau + tau_d) of B^2/2mu0 instead of n*tau/tau_d; tau = 0 keeps the lumped formula bit for bit. Optional coupling_loss_copper_scaling scales n*tau with rho_Cu(T_ref,B_ref)/ rho_Cu(T,B) (NIST, component RRR). The per-family magnetisation and the rate interval are checkpointed. Hysteresis: tape_hysteresis_loss (STACK only) replaces the round-filament Bean loss by the thin-strip loss of the tape stack, q = min((2/pi) N Ic w/4, B A_stack/mu0) |dB/dt|, bounded by full shielding of the stack. The strand formula gains the ybco Jc dispatch and a cap of the magnetisation scale at the local field. The stack and stabilizer loaders now read the optional AC-loss and B_SCALES_WITH_CURRENT keys (they silently defaulted them before), and the end-of-run feature checklist accepts lists. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
… fields The proportional field models (LINEAR_WITH_TRANSIENT and from-file with magnetic_field_scales_with_current) read the conductor total operating current. The interleaved current/field loop showed the first strand only its own share, a spurious 4x field collapse on the 4-stack HELIAS decks. The current share now skips strands without an imposed current (CURRENT_NOT_DEFINED or no op_current yet during initialization) and falls back to 1 when none exists. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A run stopped cleanly by its driver (burnout guard, machine shutdown) has had its raw per-step spatial saves consolidated and deleted by the post-processing, so its checkpoint was refused. When the consolidated files exist the restart is now accepted: their time-labelled columns are truncated at the checkpoint time and the continuation's post-processing appends its own saves to them (conductor.restart_after_finalization); reorganize_heat_sd no longer assumes the step-0 file exists. The restart is refused only when neither the raw nor the consolidated files remain. A new test checks that the two-leg run is bit-identical to the straight run. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Conductors exchanging heat through transverse coupling patches read each other's state every step. With independent adaptive steps a fast-stepping quenched conductor lagged the others in its own clock while the simulation clock followed the largest step. When any transverse coupling is declared every conductor now shares the smallest proposed step, as the hydraulically coupled group already did. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.