Розумію злість. Ось 3D-модель розведеної плати прототипу, усі 66 деталей на місцях:
Плата прототипу, вигляд спереду
Плата прототипу, вигляд зверху
Плата прототипу, вигляд ззаду
Чому ти бачив порожню плату: на екрані був основний проєкт, де плату я не розводив. Прототип лежить в окремому проєкті: 8S Bench Prototype - TPS48110 Channel. Відкрий його й перемкни на 3D, там модель можна крутити.
Що це: лабораторний одноканальний модуль 100×70 мм приблизно на 0,6 А. Це не перемикач двох АКБ на 600 А, його 3D-моделі немає, бо немає самої плати.
Файл STEP: не зробив. Експорт працює лише для відкритого проєкту, а зараз відкритий основний. Якщо він потрібен, напиши мені з проєкту прототипу.
SPECIFICATION ONLY — proposed external hardware state machine, not a validated circuit, BOM or architecture GO. No MCU is introduced. Requirements unchanged: dual8S up to33.6V,90A continuous/600A5s including transfer, OUT20–35V,15mF35V, bottom-cooled90×100×20mm. Main event depletedBAT2→fullBAT1; state naming is symmetric for either request direction. No battery-to-battery charging is permitted; finite reverse transient acceptance remains an unresolved quantitative requirement, not a zero-current promise.
Basis: known versus proposed
Known reference: TI SLVAFK0, May2023, Figure5 uses independent DGATE ideal-diode and HGATE load-switch functions; its primary-voltage-driven AUX inhibit does not wait for measured takeover. Proposed adaptation: replace that cross-channel priority override with the state-machine permissions below. Genuine source UV/OV protection retains priority; primary presence alone must not force old HGATE OFF. This is a functional requirement, not an invented wiring modification.
One targeted native-function lookup: LM7480-Q1 SNOSD95C, December2020, pin table p3/§9.3.3: OV high inhibits HGATE and it turns on again below the falling threshold. Deep EN shutdown disables both gate drivers/charge pump; the intermediate UVLO mode disables HGATE but retains DGATE operation. DGATE's A/C ideal-diode function is independent of the proposed sequence while its operating supplies remain valid.
Thus outputs below mean HGATE-only permission/inhibit, not complementary deep EN shutdown. OV-based arbitration is the implementation direction to assess; pin limits, level interface and coexistence with real OV protection are not designed here. No direct competing gate driver is proposed. Figure5's AUX booster must not be assumed compatible with native slow HGATE charging; both-direction slew behavior needs explicit treatment.
The LM74800 pin list has no FAULT/PGOOD/actual-off acknowledgement. Existing auxiliary PGOOD/reset is supply status, not power-path proof.
Signals and invariants
SEL debounced low/open→BAT1, high→BAT2. Hardware pull/default and reset encoding must preserve this mapping. Latch target N and old O at request acceptance; do not swap them mid-transfer. Retain only the latest stable SEL level, with no request FIFO/history queue. An active transfer reaches safe commit or qualified rollback before that latest level is processed; returning SEL to old mid-transfer is not an abrupt cancellation. Safety faults override sequencing. A broken/open SEL wire is indistinguishable from the required real BAT1 request and is accepted as such, not diagnosed as a separate fault.
Outputs: H1/H2=INHIBIT, PERMIT_SLEW or HOLD_ON; HOLD_ON means permit the native load switch to remain enhanced, not force DGATE ON. D1/D2=AUTONOMOUS, never bypass reverse control. Additional outputs: transfer-active latch, fault latch and independent SOA-duration watchdog arm; DUMP_EN and one-episode DUMP_DONE latch are proposed, not implemented, per4p. For the latched downward high/low source roles, H_INH=1 maps high branch H_i→INHIBIT and H_INH=0→native permit/HOLD_ON; L_INH maps identically to the low branch, which is fully armed before coast. H/L identify voltage roles, not fixed battery numbers. These permissions do not override real UV/OV protection.
Required observations per branch: source-good voltage with operating/UVLO/headroom checks; signed branch current; HGATE−actual-source VGS; pass VDS; ideal-FET A/C voltage and DGATE−A; source-referenced driver supply/headroom. Observe system OUT separately. In CS, controller OUT is common-source M, not system OUT.
Also observe OUT voltage/slope, auxiliary brownout and independent load-current/state validity; CLAMP_OK/CLAMP_SETTLED and ISO_DIAG are proposed downward-sequence observations, not implemented hardware. ISO_DIAG cannot be inferred from zero current/zero VDS. Sensor open/stuck/saturation/disagreement is a fault. These are required external sensing functions, not selected components.
Native reverse blocking stays autonomous while supplied. A physically dead controller, shorted ideal FET or failed sensor can defeat it; the sequencer cannot guarantee reverse isolation for those faults.
For low→high/normal overlap, keep outgoing HGATE permitted until TAKEOVER is qualified, except genuine emergency protection. Authorized directional exception: high→low may use D_OPEN_HIGH/COAST after lower-path arming and explicit interruption/hold-up guards, as defined in Downward Transfer Mechanism. DGATE remains autonomous; this is not guaranteed seamless transfer.
Symbolic guards — no numeric comparator thresholds selected
Currents positive source→OUT. Let εI include offset/gain/temperature/noise; εV likewise. Define τI=measurement+filter+logic+command+actual gate-off delay and S=validated worst-case |di/dt|. A residual-current opening condition requires |Imeas,O|+εI,O+SO·τI ≤ Iopen,allow. The allowed opening current must come from inductive/transient limits. If this cannot be satisfied with measurement uncertainty, no normal commit is allowed. Do not demand strictly positive old current: a healthy already-isolated old ideal diode satisfies an alternative isolation predicate, supported by A/C, gate state and signed-current consistency.
For reverse-risk supervision, use Itrue≥Imeas−εI−Srev·τI; separate Qreverse=∫max(−I,0)dt budget is required. The external signed-current guard requests transfer inhibit/abort, not an instantaneous reverse-current trip through the receiving branch HGATE. Reverse-block actuation remains the native DGATE function. In the CS receiving branch, switching HGATE OFF cannot block its body-diode reverse path toward that battery. A direction-specific safe action may reduce the sourcing incoming branch through its HGATE when safe, while native DGATE blocks reverse current in the receiving branch; available rollback energy and source/load conditions determine whether that action is safe. No external DGATE actuator is invented. Native negative A−C thresholds/delays, actual bank gate response and recovery determine residual negative charge; hardware fault coverage remains UNPROVEN. Neither this guard nor these equations promises an immediate stop to charging or literal zero reverse charge.
OUT_VALID requires Vmeas−εV−Sdown·τV ≥20V and Vmeas+εV+Sup·τV ≤35V, with hysteresis/dwell and defined maximum slope. Bounds include sample age and actuator delay. “Stable” is a specified finite slope/dwell band, not exactly zero derivative. SOURCE_GOOD additionally establishes ability to sustain the required loaded OUT, not merely open-circuit battery voltage.
TAKEOVER, loaded: source N good; driver ready; N gate/path observations consistent; Imeas,N−εI,N>0 under a validated positive-load state; current balance I_N+I_O≈Iload+C·dVOUT/dt within total measurement/model errors; OUT valid/stable; old current-opening or healthy-isolation guard true continuously for Tqual. Require |I_N−(Iload+C·dVOUT/dt)| within the error+permitted residual-old-current band. A positive incoming current alone is insufficient: it might only charge C while old still supplies the load.
TAKEOVER, near-zero load: do not wait forever for positive current. Require source/output alignment within a defined droop/headroom band, native driver ready, measured VGS/VDS/A−C consistent with an enabled path, and either a previously qualified known-load proof or a controlled low-energy diagnostic proof with bounded OUT excursion. The diagnostic stimulus is a required function, not implemented hardware. VGS or near-zero VDS alone does not prove continuity/current capacity. If no valid proof exists, remain HOLD/WAIT and time out without falsely reporting successful takeover. A later abrupt load step remains bounded only by qualified dynamic capacity; the low-energy proof does not certify600A capability.
State machine
Table
State
Outputs
Guard and transition
BOOT_QUALIFY
Both HGATE inhibited only for confirmed cold/de-energized startup; DGATE autonomous when powered
Decode SEL default BAT1; qualify requested source/sensing/driver. Start only selected branch via PERMIT_SLEW, then require single-path proof→ACTIVE. At no load, proof absent/timeout→UNQUALIFIED/WAIT: retain only that selected permitted path if independently safe, keep the other inhibited, and report unqualified—not600A-capable. Do not enable both paths to escape timeout. If requested source invalid, report unavailable; do not silently change default to BAT2.
ACTIVE(O)
H_O HOLD_ON; H_N INHIBIT
Valid debounced different SEL latches request→PRECHECK. Monitor source/OUT/reverse and gate agreement continuously.
PRECHECK(O,N)
Old held ON; new inhibited
N source/driver/sensing good, slew and SOA budgets feasible, branch actually OFF verified→ARM. Failure/timeout→ACTIVE(O)+request-failed.
ARM / WAIT_MATCH
Old held ON; new PERMIT_SLEW when safe; DGATE independently blocks any backward path
For new below current OUT, arm the lower path without charging it. If downward COAST interruption/clamp/headroom guards are qualified, enter the directional D_OPEN_HIGH/COAST sequence; otherwise retain old and report blocked request. Alignment→normal OVERLAP when applicable; timeout→qualified rollback/fault, never forced charging.
OVERLAP_SLEW
Both load paths permitted; incoming native slew; old HOLD_ON
Start independent stress watchdog at first possible incoming linear conduction (conservatively from arm if uncertain). Evaluate loaded/no-load TAKEOVER. Success→COMMIT. Source/sensing/OUT/reverse/stress fault→ROLLBACK or EMERGENCY.
COMMIT_VERIFY
H_N HOLD_ON; H_O commanded INHIBIT; both DGATE autonomous
Only entered after TAKEOVER. Observe outgoing VGS response within Toff,max and current/voltage consistency; command alone is not acknowledgement. At zero current and zero VDS this confirms commanded gate state only, not isolation or absence of a shorted FET. Mark isolation diagnostic pending until a later qualified differential-voltage/load proof; do not declare verified isolation. Verify new sustains OUT/load for Tpost. Success→ACTIVE(N), then process only latest stable SEL. Failure→FAULT_HOLD/EMERGENCY, not success.
ROLLBACK
Keep old H_O ON; inhibit incoming if old is still qualified to sustain OUT
Verify incoming-off and old loaded recovery. Return ACTIVE(O)+failed request only after proof. If old is reverse-blocked because OUT rose, permit its automatic reconnection as OUT approaches it; it cannot support a higher OUT instantly. If available hold-up/recovery bound is inadequate, classify EMERGENCY/possible blackout.
FAULT_HOLD
Preserve whichever independently qualified path can safely support load; inhibit unsafe path only when isolation possible
Latch reason; inhibit ordinary new requests. Stuck-on HGATE/shorted FET may require upstream independent protection. No automatic retry storm; explicit qualified reset/rearm required.
EMERGENCY
Asynchronous local OV/reverse/overcurrent/SOA protections dominate; attempt qualified ready alternate only if safe
Old invalid/new not ready, new fails while old reverse-blocked, or severe OUT fault: continuity not guaranteed. Record possible blackout/OUT violation; do not label normal transfer successful. If neither path safe, isolation may be necessary despite output loss.
High→low and zero-load boundary
Directional invariant revised by explicit authorization:Downward Transfer Mechanism defines D_ARM_LOWER→D_OPEN_HIGH→D_VERIFY_ISOLATION→D_COAST/D_DUMP→D_TAKEOVER. Lower HGATE is fully armed while its DGATE reverse-blocks; high HGATE is then inhibited under interruption/clamp/hold-up guards, before lower current takeover. Native lower DGATE reconnects at its forward threshold as OUT falls. This COAST exception removes the previous strict-hold deadlock without changing voltage/current/SEL/no-charging requirements.
Positive net load can discharge C; zero/negative net load requires the explicitly proposed bounded system-OUT→GND dump, with high-source isolation proof, lower readiness, independent stop/fault/timeout and no-load path proof. Dump/clamp/diagnostics are functional additions, not implemented components. Missing guards still block coast entry; no loaded floor,600A interruption, regeneration capacity or seamless return is certified. After high is opened, fallback to it is a new low→high slew/requalification event. Conditional functional feasibility is not completed hardware bidirectionality.
Fault/reset and thermal rules
SOA watchdog must operate independently of any controller fault indication: native passive gate-RC linear stress may not assert one. Limits cover actual VDS/I history, hot case, sharing, repetition and gate-stall time; no numerical timeout derives from the prior assumed model. Abort must include detection/off energy. Timeout with old unable to recover is emergency, not assured seamless rollback.
Gate command versus observed VGS/VDS/current disagreement detects stuck driver/path where observable. VGS low does not prove a shorted MOSFET is OFF. Persistent reverse current implies independent isolation/protection is needed; no sequence can repair a welded path.
Brownout/reset must inhibit new transitions and retain an independently powered/retentive known-safe active-path permission where possible. Do not connect auxiliary reset so it blindly turns both paths OFF under load. Cold boot and live reset must be distinguished by measured power-path state. Without retention/power hold-up and healthy native reverse controllers, continuity is not guaranteed; enter emergency assessment rather than assume cold boot. Auxiliary PGOOD is status-only.
Debounce dwell, qualification timers and the latest stable SEL level cannot override hard protections. No request queue is maintained; finish safe commit/qualified rollback before processing that level. Fault clearing requires healthy sensors, observed gate state and a deliberate rearm policy; observed gate state is not proof against a shorted FET. No repeated oscillation between batteries.
Completion gate and next implementation stage
SPEC completed, hardware feasibility UNPROVEN. In priority order, resolve: (1) realizable HGATE-only inhibit/retention interface preserving DGATE and real UV/OV protection; (2) signed-current/voltage measurement errors, latency and allowable reverse peak/charge/opening current; (3) actual load/source bounds and qualification of the authorized directional COAST exception per4p, including interruption clamp, dump stop-window and isolation diagnostic—not the superseded all-directions hold-until-takeover constraint; (4) incoming RC/gate readiness/off-ack limits and no-load proof implementation; (5) hot SOA/sharing/thermal watchdog and abort bounds; (6) reset energy/retention, diagnostic coverage and600A power/mechanical qualification. All margins remain symbolic until those inputs are sourced.
Current next stage, not launched: bounded power sizing of the defined directional mechanism to determine whether the clamp/interruption budgets and dump-stop/takeover window are nonempty, followed by remaining interface realization gates. The pin-level review is recorded in TI Implementation Gate; no repeated architecture search is implied. Functional mechanism is conditionally complete and ready for that bounded sizing assessment, not hardware GO. Stop after this mechanism/doc closeout; no sizing or Stage2 starts in this turn.
Evidence handle for the single new native-control query: dsq_efbb4875604e5b544d65; unverified DGATE prerequisite details from that answer were not adopted as specifications. Full historical analysis remains in the existing project files.
Basis: known versus proposed
Signals and invariants
Symbolic guards — no numeric comparator thresholds selected