MOTOR-23 — Thermal Design and Magnet Temperature

FACT — A traction motor cannot run continuously at full power — it may deliver 2-3x rated power for a few seconds off the line, but sustaining that peak for minutes would melt winding insulation and permanently weaken its magnets. The answer to “how much power, for how long” lives not in the electromagnetic design, but in the thermal design. This lesson models heat’s journey through the motor using the language of electrical circuits, and examines how we protect the two most critical hotspots — the winding end-turns and the magnets.

In this lesson we cover:

1. Heat Sources: Copper, Iron, Magnet, Mechanical
2. Lumped Parameter Thermal Network (LPTN): Modeling the Heat Path as an RC Circuit
3. Winding-Head Hotspot Management
4. NdFeB Magnets: Curie Temperature and Knee-Point Temperature ($T_k$)
5. Cooling Method Comparison: Water-Glycol Jacket vs. In-Shaft Oil Spray
6. Thermal Time Constants and S1/S2 Duty Cycles
7. Simple LPTN Worked Example

---

1. Heat Sources: Copper, Iron, Magnet, Mechanical

Four primary loss mechanisms convert to heat in a traction motor:

Loss Type	Formula/Dependency	Dominant Regime
Copper loss ($P_{Cu}$)	$P_{Cu} = 3 I_{rms}^2 R_s(T)$	Low speed, high torque (launch)
Iron loss ($P_{Fe}$)	$P_{Fe} \propto f \cdot B^2 + f^2 \cdot B^2$ (MOTOR-22 Section 6)	High speed, field weakening
Magnet/rotor eddy loss	Driven by high-frequency MMF harmonics	High speed, FSCW windings
Mechanical loss (friction + windage)	$\propto n^{2-3}$	Very high speed

FACT — $R_s(T)$ rises with temperature (copper’s temperature coefficient of resistance is $\approx 0.393\%/^\circ C$), creating a positive feedback loop where copper loss self-amplifies: temperature ↑ → resistance ↑ → loss ↑ → temperature ↑. Thermal design balances this loop against cooling capacity.

---

2. LPTN: Modeling the Heat Path as an RC Circuit

A full 3D thermal Finite-Element Analysis (thermal FEA) can take minutes to hours to run — impractical for control algorithm design or drive-cycle simulation. Instead, engineers use a Lumped Parameter Thermal Network (LPTN): each physical region of the motor (winding, slot liner, tooth, yoke, housing, coolant) is collapsed into a single node, and heat flow between nodes is represented via an electrical circuit analogy.

Electrical-Thermal Analogy:
  Electrical Circuit         Thermal Circuit
  -------------------         ---------------
  Current (I)          <->   Heat flow rate (Q, Watts)
  Voltage (V)           <->   Temperature (T, °C)
  Resistance (R)         <->   Thermal resistance (R_th, °C/W)
  Capacitance (C)         <->   Thermal capacitance (C_th, J/°C)
  Current source          <->   Loss/heat-generating source (P_loss)

Thermal resistance $R_{th}$ opposes heat flow through a material layer:

$R_{th} = \frac{\Delta T}{Q} = \frac{L}{k \cdot A} \quad \text{(conduction, } k\text{: thermal conductivity)}$

Thermal capacitance $C_{th}$ is a node’s heat storage capacity (mass × specific heat):

$C_{th} = m \cdot c_p$

A node’s temperature over time behaves exactly like a capacitor voltage in an RC circuit:

$C_{th} \frac{dT}{dt} = P_{loss} - \frac{T - T_{ref}}{R_{th}}$

The motor’s complete heat path — winding → slot liner → stator core (tooth+yoke) → housing → coolant — is modeled as a chain of series-connected $R_{th}$ nodes:

Heat Path (Winding -> Coolant), LPTN Chain:

 P_Cu (copper loss)     P_Fe (iron loss)
    |                        |
    v                        v
[T_wdg]--R_liner--[T_core]--R_core--[T_housing]--R_conv--[T_coolant]
    |                  |                 |
   C_wdg              C_core            C_housing
    |                  |                 |
   (ground/ref)      (ground/ref)      (ground/ref)

CRITICAL — LPTN’s greatest value is speed: with node counts typically kept to 10-50, an LPTN model can compute transient temperature response in seconds — a fraction of the compute time of full FEA. This makes it practical to run drive-cycle simulations (WLTP, real driving data) spanning thousands of seconds, and even to run as a real-time temperature observer inside the control software — full FEA is impractical for either. However, the LPTN parameters ($R_{th}$, $C_{th}$ values) are typically calibrated using a one-time detailed FEA run or bench test data.

---

3. Winding-Head Hotspot Management

Intuitively, one might expect the hottest point to be the conductor embedded in the slot, since that’s where current flows. In real traction motors, however, the hottest point is usually the winding end-turns (winding heads) — the parts of the coils that protrude beyond the two ends of the stator stack and are not embedded in the slot.

The reason: heat from the slot-embedded conductor can escape via two conduction paths (radially through the slot liner into the core, and from there to the housing/coolant), while the end-turn lacks this low-resistance conduction path — it can only reject heat via convection into the surrounding air or oil, which generally represents a much higher $R_{th}$ than conduction through the core.

$R_{th,endturn} \gg R_{th,slot-to-core}$

WARNING — Winding insulation class (e.g., Class H, $180^\circ C$ continuous limit) is typically uniform across the whole winding, but the actual temperature distribution is not uniform — the end-turn can run $20$–$40^\circ C$ hotter than the in-slot section. If thermal protection monitors only the in-slot temperature (or a single average-temperature sensor), the end-turn can exceed the real insulation limit and this goes undetected.

This is why high-power-density traction motors favor direct winding-head oil spray: oil nozzles spray directly onto the end-turns, sharply lowering that region’s $R_{th}$ and substantially raising the convective heat transfer coefficient — cooling capacity is targeted precisely where the temperature is highest.

---

4. NdFeB Magnets: Curie Temperature and Knee-Point Temperature ($T_k$)

Neodymium-Iron-Boron (NdFeB) magnets lose remanent magnetic flux as temperature rises. Two distinct temperature thresholds must not be conflated:

• Curie Temperature ($T_C$): The theoretical temperature, typically $310$–$400^\circ C$, at which the material completely loses its ferromagnetic property. At this point the magnet is fully demagnetized.
• Knee-Point Temperature ($T_k$): The much lower — typically $80$–$150^\circ C$ for common NdFeB grades — practical temperature at which the “knee” of the B-H demagnetization curve enters the second quadrant.

B-H Demagnetization Curve (Second Quadrant), Temperature-Dependent:

  B
  ^
  |  T = 20°C (cold)           T > Tk (hot)
  |  \                         \
  |   \                         \___
  |    \                             \
  |     \_____ (flat, reversible)     \  <- "knee" bends
  |                                     \___
  |                                          \
  +------------------------------------------+---> H (demagnetizing field)
  0                                         Hc(T)

CRITICAL — Reversible vs. Irreversible Demagnetization — If the magnet stays below $T_k$ while exposed to a demagnetizing field (large negative $I_d$, MOTOR-19), the B-H curve remains flat, and the magnet largely recovers its original flux once the field is removed — this is reversible demagnetization. But once temperature exceeds $T_k$, the curve bends at the “knee,” and the same demagnetizing field now pushes the magnet into the curve’s nonlinear region — even after the field is removed, magnet flux does not recover, and the loss is permanent (irreversible demagnetization). This is why the practically relevant limit for engineering purposes is not the deceptively high Curie temperature, but the much lower $T_k$.

ASSUMPTION — $T_k$ varies by magnet grade (N42SH, N38UH, etc. — the “SH/UH/EH” suffixes denote increasing temperature resistance) and by the strength of the demagnetizing field encountered (operating point load); the $80$–$150^\circ C$ range given here is a general order-of-magnitude indicator, not a verified datasheet value for any specific grade.

Because of this, traction motor design estimates magnet temperature either via direct sensing (rare, since instrumenting a rotating rotor is mechanically difficult) or, far more commonly, via a thermal observer model that infers it from control data, applying derating to keep it below $T_k$ (Section 6).

---

5. Cooling Method Comparison: Water-Glycol Jacket vs. In-Shaft ATF Oil

Attribute	Water-Glycol Jacket Cooling	In-Shaft ATF Oil Cooling
Heat transfer coefficient	$\approx 40$–$60\text{ kW/m}^2\text{K}$	Can be higher at the local spray point, but varies with fluid volume/flow rate
Targeting	Outer housing surface (indirect — reaches stator via conduction)	Directly onto winding head / rotor shaft (point targeting)
Electrical isolation	Galvanic separation between housing and winding (insulated housing)	ATF is electrically insulating → can contact winding directly
System complexity	Moderate (pump, jacket channels, separate loop)	High (oil pump, filter, nozzle design, often shares a loop with gear oil)
Rotor cooling	Weak (rotor thermally isolated from housing)	Strong (direct rotor cooling possible via in-shaft channel)
Magnet hotspot effect	Indirect, delayed	Direct, fast response

INTERPRETATION — Water-glycol jacket cooling is simpler and more mature technology, effective at cooling the stator housing broadly, but it is indirect for localized hotspots like the rotor and end-turns, since heat must first cross the air gap or shaft bearings. In-shaft ATF oil cooling, able to reach the rotor directly and — depending on nozzle design — the end-turns, manages magnet and end-turn hotspots far more effectively, but adds an oil loop (pump, filter, heat exchanger) plus electrical-isolation/sealing engineering, raising system complexity and cost. In modern high-performance traction motors (e.g., Porsche Taycan, Lucid Air motor families), hybrid cooling architectures combining both methods are common — housing cooled by a water-glycol jacket, rotor/end-turns cooled by oil.

ASSUMPTION — The exact cooling architecture of specific OEMs (Porsche, Lucid, etc.) is not a confirmed public specification; the “hybrid water-glycol + oil” design referenced here reflects a widely known industry architectural trend, not a verified detail of a specific model year or trim.

---

6. Thermal Time Constants and S1/S2 Duty Cycles

Every $R_{th}$-$C_{th}$ node pair carries a thermal time constant, exactly like a classical first-order RC circuit:

$\tau = R_{th} \cdot C_{th}$

This time constant sets how fast or slow the motor’s temperature responds to a power step — a large-mass, low-thermal-resistance housing node may carry a $\tau$ on the order of hours, while a small-mass winding node may carry a $\tau$ on the order of seconds to minutes.

The duty type classification in IEC 60034-1 rests directly on this physics:

• S1 (Continuous duty): The motor runs at constant load until temperature reaches steady state ($dT/dt = 0$). S1 power is limited by the thermal equilibrium of the slowest (largest-$\tau$, typically housing/coolant) node.
• S2 (Short-time duty): The motor runs at a power far exceeding S1, for a defined duration (e.g., $30\text{ s}$, $60\text{ s}$), without allowing temperature to reach steady state, then is allowed to cool.

$T(t) = T_{final} - (T_{final} - T_0)e^{-t/\tau}$

FACT — Peak (S2) power is bounded by a duration that must not exceed the limit of the motor’s smallest-$\tau$, fastest-heating node (typically the winding, given its small mass and high loss). This is why an EV traction motor’s “peak power for 30 s” or “10 s launch torque” specifications derive directly from the winding’s small thermal time constant and the $\tau$ value the LPTN model assigns to that node. MOTOR-26 covers how the S1/S2 classification shapes control software and driver-facing behavior (e.g., “boost” mode, peak-torque timeout) in more detail.

---

7. Simple LPTN Worked Example

Let’s build a simplified three-node model: Winding (W) → Core (C) → Coolant (Amb, treated as a fixed reference).

Node	$R_{th}$ (to its output, °C/W)	$C_{th}$ (J/°C)	Loss Input (W)
Winding (W) → Core	$0.08$	$450$	$P_{Cu} = 600\text{ W}$
Core (C) → Coolant	$0.03$	$1200$	$P_{Fe} = 250\text{ W}$

EXAMPLE — At steady state (S1, $t \to \infty$), the temperature difference between winding and core is set purely by $R_{th}$ and the total power flowing through that node (capacitance has no effect at steady state): $\Delta T_{W-C} = R_{th,W-C} \cdot P_{Cu} = 0.08 \times 600 = 48^\circ C$ $\Delta T_{C-Amb} = R_{th,C-Amb} \cdot (P_{Cu}+P_{Fe}) = 0.03 \times 850 = 25.5^\circ C$ If coolant temperature is $65^\circ C$: $T_{core} = 65 + 25.5 = 90.5^\circ C$, $T_{winding} = 90.5 + 48 = 138.5^\circ C$ — safely below the Class H insulation limit ($180^\circ C$).

The winding node’s time constant: $\tau_W = R_{th,W-C} \cdot C_{th,W} = 0.08 \times 450 = 36\text{ s}$. This means the winding temperature reaches roughly 95% of its new steady-state value in about $3\tau \approx 108\text{ s}$ after a power step — so a $30$–$60$-second S2 peak-power window operates in a regime where winding temperature has not yet reached steady state, still climbing along its rise curve. This is the physical reason S2 permits substantially higher instantaneous power than S1.

---

8. Frequently Asked Questions (FAQ)

Why use LPTN instead of full FEA?

FACT — LPTN can compute transient temperature over a full drive cycle in seconds, while full thermal FEA can take hours for the same simulation. LPTN is the only practical option for a real-time temperature observer running inside control software; FEA is instead used once, to calibrate LPTN parameters.

Why isn’t magnet temperature measured directly?

INTERPRETATION — Mounting a wired temperature sensor (thermocouple, RTD) on a rotating rotor is mechanically difficult (requires slip rings or telemetry) and costly; most manufacturers instead use an observer model that infers magnet temperature from fixed stator sensors and motor current/speed data, validating accuracy against worst-case test data.

Does exceeding the knee-point temperature ($T_k$) instantly damage the motor?

FACT — No, the effect is gradual — just above $T_k$, a small, partial, permanent flux loss occurs; the loss rate grows with both temperature and the strength of the demagnetizing field encountered. But every permanent loss is irreversible and gradually degrades torque/efficiency performance, which is why derating strategies are designed to leave a meaningful safety margin below $T_k$.

---

9. Summary

• The primary heat sources are copper ($I^2R$, self-amplifying with temperature), iron (frequency/flux-dependent), magnet/rotor eddy, and mechanical losses.
• LPTN models the heat path as an RC electrical circuit of $R_{th}$/$C_{th}$ nodes; it is far faster than full FEA and the only practical route to real-time temperature estimation and derating.
• Winding end-turns lack a low-resistance conduction path and are usually hotter than the in-slot conductor; direct oil spray specifically targets this hotspot.
• The practical demagnetization limit for NdFeB magnets is the much lower knee-point temperature ($T_k$), not the deceptively high Curie temperature; demagnetization is reversible below $T_k$ and irreversible above it.
• Water-glycol jacket cooling is mature and indirect for the stator housing, while in-shaft ATF oil cooling directly and rapidly targets rotor/end-turn hotspots; high-performance motors often combine both.
• Thermal time constants ($\tau = R_{th}C_{th}$) are short for small-mass nodes like the winding, which is why S2 short-time peak power can exceed S1 continuous power substantially — a direct consequence of thermal physics.

10. Sources and Verification Note

• Pyrhönen, Jokinen & Hrabovcová — Design of Rotating Electrical Machines (thermal modeling fundamentals).
• Lipo — Introduction to AC Machine Design (loss mechanisms).
• Staton, Boglietti & Cavagnino — Solving the More Difficult Aspects of Electric Motor Thermal Analysis (LPTN methodology, end-turn modeling).
• IEC 60034-1 (S1/S2 duty type definitions).

ASSUMPTION — The numerical LPTN example ($R_{th}$, $C_{th}$ values) and $T_k$ range given in this lesson are order-of-magnitude illustrations, not verified datasheet values for any specific motor model. For vehicle-level thermal loops, see EV Academy → EV-14.

Next Lesson

• MOTOR-24 — Mechanical Design, NVH, and Rotor Limits: We turn to the mechanical strength of the motor we just managed thermally — rotor critical speed, magnet retention (banding/sleeve) design, sources of acoustic noise and vibration (NVH), and the mechanical stress analysis that sets maximum speed limits.