CONVERTER_NETWORK.md

Quasi-steady torque-converter network

English · 简体中文 · Français · Русский · 日本語 · 한국어 · Deutsch · Español · Italiano · Português

torque_converter participates in the same shaft, motor, cylinder, clutch and ideal-gear solve. Pump and turbine are distinct rotational nodes with explicit inertia. The stator is stationary ground; its reaction is observable but it performs no work. Fluid loss feeds an optional thermal node or the external heat-rejection ledger. A separate parallel clutch supplies lockup. All parameters in the laboratory are synthetic and unverified.

Explicit maps and equations

Four maps are required: pump_positive, pump_negative, turbine_positive and turbine_negative. The member with larger absolute speed is the reference driver; the pump wins an exact tie. The sign of that member selects its positive/negative map. This is a mathematical reference-member convention, including during counterrotation. It does not infer an unavailable reverse or coast characteristic.

For driver speed wD and follower speed wF, s = wF/wD lies in [-1,1]. Each map contains 2–32 explicit points with dimensionless speed_ratio, torque_ratio R and capacity_coefficient C in nm_s2_rad2. C multiplies squared speed; it is not an inverse K-factor. R and C interpolate linearly and are never extrapolated. Both speeds zero produce zero reactions and mode zero.

Tdriver   = -C(s) * wD * abs(wD)
Tfollower =  R(s) * C(s) * wD * abs(wD)
Tstator   = -(Tpump + Tturbine)
Qdot      = C(s) * abs(wD)^3 * (1 - s*R(s))

Compilation requires strictly increasing knots spanning [-1,1], nonnegative C/R, and s*R(s) <= 1 throughout every segment. Checking only knots is insufficient: R linear in s makes efficiency quadratic; any interior maximum is checked too. Slopes must be finite. At s=1, C must be zero and R one, giving zero fluid torque at equal same-direction speed. At s=-1, pump-positive/turbine-negative and pump-negative/turbine-positive maps must give matching physical reactions. This prevents a jump when the reference member changes during counterrotation. The endpoint comparison allows only 64*epsilon*(abs(a)+abs(b)) rounding tolerance. Map arrays are owned and immutable; normalized coefficients enter the model fingerprint.

These restrictions define Power!'s current passive signed-map model. They do not claim to cover arbitrary measured converter curves. General map-based driving/coasting conventions are documented by MathWorks Torque Converter and its two-mode example. The four signed maps, interpolation validation and solver below are Power! design; no source code or measured parameter set was copied from those references.

Coupled integration and limits

Converter reactions use interval midpoint speeds. When cylinders are present, a joint Newton system solves their discrete crank-angle work and both converter-port speeds. Every unit-torque response includes the electromechanical system and permanent gear projection. Clutch constraint iterations call this same nonlinear solve; capture and reversal subdivision regenerate interval responses. There is no lagged converter torque applied after cylinder or clutch integration.

The nonlinear solve has 24 iterations and 12 halving line-search attempts per iteration. The cylinder angle residual tolerance is 2e-14 rad. A converter speed residual uses 2e-12 + 64*epsilon*(abs(value)+abs(free_prediction)) rad/s. Analytic piecewise map Jacobians and finite-difference cylinder work derivatives build the joint system. Existing 0.25-rad cylinder travel, valve/burn resolution, clutch iteration/event and gear constraint limits still apply. At most eight converters are supported within the existing 32-node, 64-component and 64-state budgets. Each converter adds four logical observable history states: two mean torques, mean heat power and cumulative heat. Compensated heat summation also participates in copy/hash/rollback.

Accepted interval heat is -h*(Tp*wp_mid + Tt*wt_mid), so the mechanical work removed is the heat recorded. Negative heat beyond the solved-speed rounding tolerance rejects the interval; only rounding-sized negative residue is clamped to zero. Torques and heat power are duration-weighted across accepted internal intervals, then divided by the full tick. Speculative event trials never commit their heat or reactions. All converter, clutch, gear, gas, burn-history, input and global-ledger state rolls back on a failed or cancelled batch. Forks own their workspaces. Tests exercise allocation-free capture.

Midpoint integration is second order for smooth standalone converter motion. The coupled fired model retains explicit wall-temperature coupling and interval-average clutch breakaway, so it does not claim uniform second order through all transitions. Refine ticks for a numerical failure or accuracy study; inspect map slopes, inertia/speed scales and clutch constraints before recreating a session. A valid map is not a guarantee that every chosen timestep is solvable.

Shared model and observable contract

JSON uses required node_a (pump), node_b (turbine), four arrays under parameters, and optional heat_node. It accepts no converter input channel, carrier port or implicit map defaults. ComponentDefinition.TorqueConverter exposes the same Core model. Map errors carry the component ID and an actionable field such as converter.pump_positive or converter.counter_rotation.

Field	Meaning	Unit
torque	Last complete tick's mean pump torque	Nm
torque_at_b	Last complete tick's mean turbine torque	Nm
torque_at_c	Last complete tick's mean stationary-stator reaction	Nm
heat_flow	Last complete tick's mean fluid heat power	W
fluid_heat	Cumulative accepted fluid heat	J
speed_ratio	Current follower/driver signed speed ratio; zero when stopped	fraction
converter_drive	0 stopped, 1 pump positive, 2 pump negative, 3 turbine positive, 4 turbine negative	state code

Mean torques/power start at zero. Boundary input updates do not rewrite prior-tick mean outputs. torque_at_c names the stator reaction here; this component has no third rotor. Converter-containing models advertise quasisteady_converter_powertrain, and add fingerprint tag 10. Converter-free model fingerprints and trajectories remain unchanged. Portable asset v10 retains all four maps; v1–v9 readers and authentic fixtures remain.

Laboratory and evidence

Request MCP example fired-converter, or run:

dotnet src/Power.Cli/bin/Release/net10.0/Power.Cli.dll assets/labs/fired-converter.power.json --output artifacts/reports/fired-converter.json

The 0.8-s laboratory starts with a 600-rpm pump and 300-rpm turbine, feeding the sun of a planetary and a 3:1 final drive. Ring braking selects reduction; a sun/ring clutch selects direct drive. Independent scheduled lockup, release and recapture exercise fluid and friction paths. These are prescribed events, not an automatic transmission controller.

At 50,000-ns ticks, all 87 report boundaries replay exactly through CLI, portable assets and MCP. The model fingerprint is 839d03901973668d and final state hash is 834a679376b7a6fd. Final pump/turbine speed is approximately 73.37748 rad/s, load speed 6.988331 rad/s, fluid heat 24.27663 J and lockup heat 22.84709 J. The shift clutch and brake add 157.18199 J and 83.42289 J. The shared heat node reaches 301.438643 K; total energy residual is about 3.30e-11 J. These numbers describe a synthetic transient.

Independent tests cover the analytic C(s)=k(1-s), R=1 two-inertia solution, analytic stall decay, second-order refinement, thermal and external heat routing, stator balance, reverse/coast/counterrotating states, shared converter ports, gear reflection, lockup, full-batch rollback, cancellation, forks and zero allocation. Fired-model refinement checks a dimensionless combined distance in final speed, fluid heat and lockup heat to a 3,125-ns run, using five tick sizes. It also bounds absolute differences below 0.0002 rad/s or J respectively. Individual heat errors need not decrease at every halving near events. This check is separate from standalone analytic order. Signed-map ownership/validation and re-signed malformed portable records have dedicated checks.

The hydraulic network now supplies pressure-derived lockup and shift capacity. Fluid angular momentum, converter fill/pressure dynamics, rotating/freewheeling stator mechanics, temperature-dependent properties, pump/regulator and piston dynamics, complete DCT/AT topology and ECU/TCU coordination remain open. Complete engine behavior, OEM measurements and vehicle calibration also remain open. Studio has schematic fluid ports and prepared tests; actual Editor/Play/IL2CPP evidence remains separate and unavailable in this environment.