🎓 Lesson 21
D5
Thermal Imaging & SoC Correlation Validation
Thermal imaging helps engineers see how hot a battery gets while it’s working, and SoC (State of Charge) tells them how full the battery is—this lesson shows how to check if those two things match up reliably.
🎯 Learning Objectives
- ✓ Analyze thermal image sequences to extract mean cell-surface temperature trends across SoC intervals
- ✓ Design a validation test protocol that isolates SoC–temperature correlation from confounding variables (e.g., current, ambient, aging)
- ✓ Apply linear and polynomial regression to quantify correlation strength (R²) and residual error between thermal features and reference SoC
- ✓ Explain the physical origins of non-monotonic SoC–temperature behavior in NMC and LFP chemistries
- ✓ Calculate the maximum allowable thermal deviation (°C) for a given SoC bin based on IEEE 1679.2 SoC accuracy requirements
📖 Why This Matters
In grid-scale battery energy storage systems (BESS), undetected thermal gradients or inaccurate SoC estimates can lead to overcharging, thermal runaway, or premature capacity loss—costing millions in downtime and safety remediation. Thermal imaging is now deployed for remote, non-contact monitoring, but its value hinges on proven correlation with true SoC. Without rigorous validation, thermal data is just pretty pictures—not actionable intelligence. This lesson bridges infrared sensing and electrochemistry to build trust in thermal-based BMS decisions.
📘 Core Principles
The SoC–temperature relationship arises from three interdependent phenomena: (1) Joule heating (∝ I²R), which dominates during charge/discharge; (2) reversible entropic heating/cooling (ΔT/ΔSoC ≈ ∂T/∂SoC|ₚ), intrinsic to electrode material thermodynamics; and (3) irreversible parasitic reactions (e.g., SEI growth), which accelerate at high SoC and temperature. Crucially, this relationship is *not* globally linear: NMC cells exhibit peak surface temperature near 80–90% SoC due to entropic maxima, while LFP shows near-flat thermal response until >95% SoC. Validation must therefore account for chemistry-specific, rate-dependent, and hysteresis-affected behavior—and distinguish surface temperature from core temperature using calibration curves derived from embedded thermocouples.
📐 Correlation Confidence Metric (CCM)
CCM quantifies the statistical reliability of thermal–SoC correlation by normalizing residual error against both measurement uncertainty and SoC bin width. It is used to accept/reject thermal imaging as a proxy for SoC within a defined operational envelope.
Correlation Confidence Metric (CCM)
CCM = (ΔSoC_bin / 100) × (u_combined / |T_obs − T_pred|)Quantifies reliability of thermal imaging as a proxy for SoC within a specified SoC interval.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| ΔSoC_bin | SoC bin width | % | Width of the SoC interval over which correlation is evaluated (e.g., 10 for 10% bins) |
| u_combined | Combined measurement uncertainty | °C | Root-sum-square of IR camera accuracy and reference SoC uncertainty |
| T_obs | Observed mean surface temperature | °C | Mean temperature measured by calibrated thermal camera over cell surface |
| T_pred | Predicted temperature | °C | Temperature estimated by correlation model (e.g., polynomial fit) for the same SoC |
Typical Ranges:
LFP, C/5, 25°C ambient: 0.15 – 0.35
NMC811, 1C, 35°C ambient: 0.08 – 0.22
💡 Worked Example
Problem: A 280 Ah LFP prismatic cell is cycled at C/5 from 10% to 100% SoC in 10% bins. IR camera (±0.8°C accuracy) measures mean surface T. Reference SoC from coulomb counting + OCV lookup has ±0.5% uncertainty. At 70% SoC, observed T = 32.4°C; model-predicted T = 31.9°C. Bin width = 10% SoC.
1.
Step 1: Compute absolute residual error = |32.4 − 31.9| = 0.5°C
2.
Step 2: Compute combined uncertainty = √(0.8² + 0.5²) = √(0.64 + 0.25) = √0.89 ≈ 0.94°C
3.
Step 3: Apply CCM = (bin_width / 100) × (combined_uncertainty / residual_error) = (10/100) × (0.94 / 0.5) = 0.1 × 1.88 = 0.188
4.
Step 4: Compare to threshold: CCM ≥ 0.20 required per UL 9540A Annex D for thermal-proxy SoC use → 0.188 < 0.20 → correlation fails at 70% SoC under these conditions.
Answer:
The CCM is 0.188, below the minimum acceptable threshold of 0.20, indicating insufficient confidence to use thermal imaging alone for SoC estimation at 70% SoC under C/5 cycling.
🏗️ Real-World Application
At the Notrees BESS (Texas, 36 MW / 24 MWh, AES Energy Storage), thermal imaging was integrated into the BMS after field validation revealed strong SoC–T correlation only above 85% SoC for their LFP modules. Below that, entropy effects were negligible and noise dominated. Engineers therefore disabled thermal-based SoC correction below 85%, reverting to coulomb counting + OCV fusion—reducing SoC estimation error from ±3.1% to ±0.8% across full cycle life (per 2022 EPRI Report 3002011157). This decision emerged directly from structured thermal–SoC correlation validation per IEEE 1679.2 Section 6.4.