BD × AI Lab

DGTE

Double-Gravity Thermoelectric Engine — Low-Grade Heat to DC
DGTE uses gravity and buoyancy as the two motion-conversion strokes, while the real energy input must come from a heat or resource gradient. Hot capsules become lighter than the surrounding fluid and rise; cold capsules become heavier and descend. A co-axial recuperator couples the columns with target effectiveness ε ≥ 0.90. No turbomachinery. No refrigerants. No moving seals in the thermal path.
4–8%
Net Efficiency
≥0.90
Recuperator ε
≥0.85
Duty Factor φ
≥99%
30-Day Availability
DC
Direct Output

ABSTRACT

DGTE is a double-gravity heat engine. The name says what it does: gravity acts twice per cycle. When a sealed capsule is heated, its density drops below the surrounding fluid — gravity pulls the fluid down, displacing the capsule upward. When the capsule cools, its density rises above the fluid — gravity pulls the capsule down directly. Same force, two directions, two power strokes. Segmented linear coils harvest electromagnetic energy from both strokes — no turbines, no pistons, no refrigerants. A full-height annular counter-current recuperator couples the HOT and COLD columns with target effectiveness ε ≥ 0.90, recovering heat that would otherwise be lost.

Two crossover variants are formalised: V1 (rotary airlocks, baseline — proven mechanics, 9/10 confidence) and V2 (magnetic coupling, R&D — higher long-term upside, 5/10 confidence). The architecture accepts both as swappable cartridges. Pilot acceptance targets: net efficiency η = 4–8% at 90–120/20–30 °C, duty factor φ ≥ 0.85, availability ≥ 99% over 30 days, parasitics ≤ 100W, zero crossover-attributable jams.

Best-fit sources: industrial waste heat (compressor oil coolers, CIP/pasteurisation), low-enthalpy geothermal, solar-thermal, district heat. FOAK economics: €40–80k/kWₑ with 12–18 year payback at €0.20/kWh; volume targets €10–15k/kWₑ with 3–5 year payback at high tariffs. The system trades peak thermodynamic efficiency for simplicity, safety, availability, and direct DC output.

Conceived, designed, and authored by Bojan Dobrečevič with AI teamwork (ChatGPT, Claude, Gemini, DeepSeek). The recuperation architecture was seeded by a human insight — "capsules don't traverse the exchanger; fluid does" — and elaborated through multi-LLM iteration into the co-axial annular design adopted as baseline.

Current Arena Status · DGTE v0.7a

Pure DGTE remains alive but hard. The current bench planner says not to build the full machine first. The useful next move is to attack transfer, crossover, pressure balance, and thermal closure as separate bench tests.

#1 Pure-DGTE testRotary/starwheel crossover dye and carry-over rig.
#2 Pure-DGTE testHydraulic balanced-lock rig for equalization energy, leakage, valve loss, and J/transfer.
#3 Pure-DGTE testIntegrated thermal slice with measured hot/cold temperatures, thermal input, transfer losses, auxiliaries, and net electrical proxy.
Experiment 0 helperWater-sealed bell / flooded-cup intake as a passive priming and anti-air helper only, not an energy source.

P_net = min(P_thermal_gross, P_PTO_path) − transfer losses − auxiliaries

Open DGTE Bench Arena

1How It Works

DGTE uses gravity twice as a motion-conversion mechanism, not as a free energy source. A capsule heated in the HOT column becomes less dense than the bath fluid — the fluid sinks under gravity, and the capsule is displaced upward. At the top, it crosses to the COLD column, cools, becomes denser than the fluid — and gravity pulls it straight down. Both motions pass through segmented coils that generate electricity, but net output must be closed against real thermal input, transfer losses, and auxiliary loads.

The 12-Step Capsule Loop

  1. Preheat loops to setpoints; baseline ε
  2. Prime and degas; verify Δp
  3. Start crossovers — bottom airlock feeds HOT column
  4. HOT ascent → power generation (stroke 1)
  5. HOT→COLD crossover at top
  6. COLD descent → power generation (stroke 2)
  7. COLD→HOT crossover at bottom
  8. Spacing control via back-EMF timing
  9. Throughput ramp to achieve φ_duty ≥ 0.85
  10. Steady run — hold C_r ∈ [0.6, 1.0]
  11. De-rate on ε/Δp drift
  12. Orderly stop — cool; log ε, Δp

Co-Axial Vessel Architecture

Central HOT riser and annular COLD downcomer separated by a full-height annular recuperator wall. Crossovers at top and bottom complete the capsule loop. The recuperator is not a separate unit — it IS the wall between the columns, transferring heat continuously via counter-current flow.

Key Physics

F_net = (ρ_bath − ρ_capsule) · V · g → gravity on the density difference drives both strokes
F_drag = ½ · C_D · ρ_bath · A · v² → design v ≤ 0.4 m/s in coil zones
ε_recup = [1 − exp(−NTU(1−C_r))] / [1 − C_r · exp(−NTU(1−C_r))]
P_net = η_gen · P_gross − P_parasitic → direct DC output
2Variants — V1 Rotary vs V2 Magnetic vs V3 Dry-Porch
AspectV1 Rotary (baseline)V2 Magnetic (R&D)V3 Dry-Porch
Confidence9/105/10contextual
CrossoverFully-wetted rotary airlocksWet-tolerant magnetic couplingGas-phase dry tunnels + N₂ puffs
Moving partsRotating seals/bearingsNone through pressure boundaryValves + pneumatics
Parasitics5–15W per airlockPotentially lowerN₂ + ventilation
Recuperator ε≥ 0.90 (counter-current)≥ 0.90 (counter-current)≈ 0.70 (regenerative)
Dev riskLow (COTS mechanics)High (novel physics)Medium
Long-term upside7/109/10Limited
Decision: Proceed to pilot with V1 as baseline. Develop V2 under gates. Design crossover interface as swappable cartridges to enable future V2 field trials without re-shelling the vessel.

V2 R&D Gates (must pass before integration)

  1. Torque & slip ≥ 2× worst-case with zero slip over 10⁶ cycles in hot oil
  2. Fouling tolerance + CIP recovery
  3. Thermal/EMI compatibility
  4. Δp transient integrity
  5. Deterministic transfer latency compatible with spacing and φ_duty
3Physics & Models

Thermodynamic Model

Ideal ceiling: η_C = 1 − T_c/T_h. Net efficiency accounts for recuperator effectiveness, duty factor, generation efficiency, geometry, and parasitics:

η_net ≈ (α · η_C) · f_hx(ε, C_r) · φ_duty · η_gen · χ − P_par/Q̇_in

Where α ≈ 0.30 captures non-HX irreversibilities, η_gen ≈ 0.8–0.9, P_par/Q̇_in ≈ 0.2–0.35%.

Worked Example — V1 at 110/25 °C

ParameterValue
Carnot efficiency η_C22.17%
Usable fraction f_hx (ε=0.90)0.96
Duty factor φ0.86
Generation efficiency η_gen0.85
Irreversibility factor α0.30
Gross η~4.9%
Parasitic deduction−0.25%
Net η~4.6%

With ε → 0.92 and geometry factor χ = 1.02, net approaches 5.5–6.0%.

Sensitivity (Tornado)

Most sensitive: ε → f_hx (strongest lever), then φ_duty, then η_gen, then P_par. A 2%-point improvement in ε yields +0.2–0.3 %-point η_net.

Capsule Inversion Dwell

τ_th = C_cap / (h · A_s) → t_inv ≈ τ_th · ln[(T₀ − T_b)/(T* − T_b)]

Rule: t_inv ≤ 0.10 · t_s and t_× ≤ 0.05 · t_s → φ_duty ≥ 0.85

4Controls & Instrumentation

Primary KPIs (Real-Time)

  • ε_recup ≥ 0.90 — from RTD ladders + flowmeters
  • φ_duty ≥ 0.85 — from back-EMF timestamps
  • t_inv ≤ 0.10 · t_s — inversion dwell budget
  • Carry-over < 0.5% — dye-trace verification
  • Availability ≥ 99% — 30-day continuous

State Machine

COLD_START → WARM_THROUGH → RAMP → STEADY ↔ PART_LOAD → CIP/SERVICE. Fault path: FAULT_LATCHED ↔ JAM_CLEAR → RAMP. Transitions tied to ε, φ, Δp, STO, carry-over.

Loop Hierarchy

LevelLoopActuatorRate
InnerSpacing PI (sensorless back-EMF)Throughput & crossover rate≤5% /s
MiddleΔp PID with feedforwardPump VFD~3 t_h
Outerε guard + ΔT ladder monotonicityMaldistribution routineMinutes

Sensorless Position Tracking

Back-EMF zero-crossing and peak timing from each coil segment provides capsule position and velocity without physical sensors. Controller maintains target spacing t_s and φ_duty. This is the same principle as DCC — the system reads its own electrical signature.

5Test Protocols & Decision Gates
TestGoalAcceptance
D2 — ε-LadderVerify recuperator performanceε ≥ 0.90 at design; ≥ 0.88 at −20% throughput; ladder monotone
D3 — Duty & Dwellφ_duty ≥ 0.854h continuous; t_inv ≤ 0.10·t_s; jitter σ < 0.2·t_×
D4 — Carry-OverBound cross-contamination< 0.5% per transfer (dye-trace)
D5 — ParasiticsConfirm P_par/Q̇_in0.2–0.35% (V1/V2); DC cross-check ±1%
D6 — Availability30-day continuous≥ 99%; faults recover within 10 min to ≥95% baseline
D8 — V1 RotaryRotor performanceTorque map; jam-clear ≤ 3 cycles; carry-over across ±20% throughput
D9 — V2 MagneticCoupling & EMISlip ≤ 20°; spur < −60 dBV; equivalence to V1 on all KPIs
D10 — CIP RecoveryFouling recoveryPost-CIP ε within −1 %-pt; Δp within +5%
Executive rule: D2–D6 + D8 pass → V1 baseline qualified. Additionally D9 passes with no regressions → V2 pilot authorised. V3 remains non-baseline unless D7 shows compelling site-specific advantages.
6Safety & Regulatory

Governing Standards

Pressure: ASME BPVC VIII / EN 13445. Piping: B31.3 / EN 13480. Relief: API 520/521. Machinery: ISO 12100, EN 60204-1. Electrical: IEC 60204-1, IEC 61439. EMC: IEC 61000. Functional safety: IEC 61508/61511 or ISO 13849.

Key Safeguards

  • Overpressure: PSV(s) sized to controlling scenario + optional rupture disk. Hydrotest ≥ 1.3× MAWP.
  • Hardwired STO/E-stop to crossovers, coil stage, pump VFD
  • Trips: vessel over-pressure, Δp spike >2×, bath T out-of-window, DC bus OVP, door open, rotor stall >1s
  • Soft derates: ε drop >5 %-pt, φ < 0.80, coil T high, EMI mask violation
  • CIP: isolate → drain → flush → recirculate → neutralise → rinse → sample → release

Risk Summary

HazardMitigation
Over-temperatureHi-temp trip; redundant RTDs; relief path
Over-pressureRelief + rupture; expansion volume; proof 1.5×
Oil leak/spillContainment pan; leak detect; spill plan
Crossover jam (V1)Screens; torque limit; STO; quick-swap cartridge
Recuperator foulingFiltration; CIP manifolds; ε-ladder monitoring
Electrical faultOCPD; SPD; thermal sensors per segment
7Economics & Deployment

CAPEX Breakdown (FOAK 2–3 kWₑ Skid)

BlockShareLearning
Vessel + annular HX28–36%↓ with volume
Recuperator inserts18–24%↓ with tooling
Crossovers10–16% (V1)V2 falls if standardised
Coils + power electronics10–15%↓ PCB panelisation
Pumps/piping6–10%Stable
I&C + enclosure16–24%↓ after FOAK

Economics by Scale

ScaleCAPEX/kWₑPayback @ €0.20Payback @ €0.40
FOAK pilot€40–80k12–18 yr6–9 yr
NOAK volume€10–15k5–7 yr3–5 yr

Worked Example (Site B: 110/25 °C)

V1, η_net = 5.5%, Q̇_in = 60 kW_th → P_net ≈ 3.3 kWₑ. CF = 0.9 → E_yr ≈ 26 MWh. FOAK CAPEX €85k → LCOE ≈ €475/MWh. With replication (CAPEX → €55k, η → 6.2%) → LCOE ≈ €295/MWh.

Best-Fit Sources

  • Industrial waste heat: compressor oil coolers, jacket water, food/beverage CIP/pasteurisation
  • Low-enthalpy geothermal (60–120 °C)
  • Solar-thermal / district heat return
  • Combined with PV panels (DGTE cools panel, extracts waste heat → HTH hybrid)

Learning Curves

Wright's law on replicable blocks (coils, rectifiers, rotor cartridges, HX inserts): C(N) = C₁(N/N₁)^(−β), β ∈ [0.15, 0.25]. Engineering/commissioning overheads drop ≥50% after FOAK.

8Comparators — DGTE vs. Alternatives
AttributeDGTEMicro-ORCStirlingTEG
Working fluidThermal oilRefrigerants/HFOsGasSolid-state
TurbomachineryNoneTurbine + pumpPistons + sealsNone
η @ 90–120/20–30°C4–8% netOften off-designNeeds higher ΔT<3%
MaintenanceLow (few wear points)Turbine/pump/valvesSeals/clearancesVery low
Direct DCYesInverter pathVariableYes
RefrigerantsNoneYes (HFOs)NoNo
DGTE trades peak thermodynamic efficiency for simplicity, safety, availability, and direct DC output. It fills the niche where ORC is overkill, Stirling needs higher ΔT, and TEGs are too inefficient.
9Prototype Roadmap (C → A → B)
PrototypeGoalOutputCAPEX
C — Physics DemoDemonstrate dual gravity, crossovers, φ_duty, DC on both strokes0.2–0.6 kWₑ (no recuperator)€20–50k
A — Mini w/ RecuperatorProve ε ≥ 0.90, low Δp, stable φ, crossover reliability0.3–0.8 kWₑ€35–60k
B — Pilot (near-product)Full performance validation, 30-day run2–3 kWₑ€100–200k

Upgrade Path

Design C as a platform: flanged ports for future recuperator cartridges; pre-route manifolds and CIP tees; install RTD ladders, Δp ports, and flowmeters from day one; modular coils/rectifiers; same capsule geometry throughout.

UpgradeIncremental CAPEXWhat Changes
C → A€15–30kAdd recuperator cartridge + instrumentation
C → B€70–120kFull recuperator + production crossovers + enclosure

Decision Gates

  • C-gate: Stable bidirectional EMF, no jams, repeatable φ_duty histogram
  • A-gate: ε ≥ 0.90; Δp within design limits; parasitics ≤ 100W
  • B-gate: 30-day run with A ≥ 0.99; zero crossover-attributable jams; validated energy bookkeeping
10Contributions & Provenance

Authorship

ContributorRoleShare
Bojan DobrečevičConception, regeneration lineage, decisive recuperation seed & selection, requirements, final authority55%
ChatGPT 5 ThinkingSynthesis, models, controls, testing, drafting, integration18%
GPT-o5Extended recuperation from human seed, crossovers, targets12%
ClaudeSafety/regulatory drafting, clarity edits, glossary8%
GeminiRe/Pr/Nu references, sensitivity grid, annular corrections5%
DeepSeekSlip detection nuances, control-aware perturbations2%

Key Design Decision

The recuperation architecture was seeded by a human insight: "capsules don't traverse the exchanger; fluid does." This reframing — from regenerative (capsules carry heat) to recuperative (fluid carries heat through a wall) — was the decisive architectural pivot. GPT-o5 elaborated the seed into the co-axial annular counter-current design; multi-LLM review confirmed it as baseline.

Design lineage: RC1 → RC2 (Dry-Porch) → RC3 (transitional) → RC4 (Recuperation v1, co-axial annular, fully-wetted crossovers) — adopted as FOAK baseline.