Operando monitoring of thermal runaway in commercial lithium-ion cells via advanced lab-on-fiber technologies

Operando monitoring of complex physical and chemical activities inside rechargeable lithium-ion batteries during thermal runaway is critical to understanding thermal runaway mechanisms and giving early warning of safety-related failure. However, most existing sensors cannot survive during such extremely hazardous thermal runaway processes (temperature up to 500 °C accompanied by fire and explosion). To address this, we develop a compact and multifunctional optical fiber sensor (12 mm in length and 125 µm in diameter) capable of insertion into commercial 18650 cells to continuously monitor internal temperature and pressure effects during cell thermal runaway. We observe a stable and reproducible correlation between the cell thermal runaway and the optical response. The sensor’s signal shows two internal pressure peaks corresponding to safety venting and initiation of thermal runaway. Further analysis reveals that a scalable solution for predicting imminent thermal runaway is the detection of the abrupt turning range of the differential curves of cell temperature and pressure, which corresponds to an internal transformation between the cell reversible and irreversible reactions. By raising an alert even before safety venting, this new operando measurement tool can provide crucial capabilities in cell safety assessment and warning of thermal runaway.

The response spectra of FBG to temperature and pressure Wavelength response data (symbol) and linear fitting curve (solid line).Considering the pressure limit of the safety valve of the 18650 batteries, 2 MPa was selected here as the maximum pressure.The temperature sensitivities of an FBG and an FPI are both determined by the following Eq.(S1) 1 : where δ = 8.3×10 -6 °C -1 and α = 0.55×10 -6 °C -1 are the thermo-optic coefficient and linear thermal expansion coefficient of silica, respectively;  is free space wavelength; Δλ is the wavelength shift caused by external perturbations; Δ is external temperature change; Δ/L is strain.In practice, strain is eliminated by fixing the FBG-PPI assembly at only one end in a 0.5 mm diameter drilled central hole into the cell, where it remains suspended.For the aircavity FPI, its temperature sensitivity is mainly attributed to the thermal expansion effect of the silica wall.Therefore, its temperature sensitivity is much lower than that of the FBG.In our experiment, the measured temperature sensitivities of the FPI and the FBG are 0.5 pm °C -  4/21 and 10.3 pm °C -1 , respectively.The "temperature insensitivity" is a relative concept.In many publications, they claim that their sensor is temperature insensitive when the sensitivity is less than 1 pm °C -1 .What's more, the pressure sensitivity of our FPI sensor is very high.The temperature-pressure cross-sensitivity coefficient is calculated to be (0.5 pm °C -1 )/(4185.3pm MPa -1 ) ≈ 1.210 -4 MPa °C -1 .This means that a 100 °C temperature fluctuation only leads to a neglectable pressure measurement error of 0.01 MPa.So, we believe we could claim our FPI is "temperature insensitive".

Error assessment of the FBG temperature measurement
In order to achieve the error assessment of the FBG temperature measurement in both normal cycling conditions and thermal runaway conditions, the FBG sensor and thermocouple are both implanted into the 18650 cells simultaneously to character their temperature response.The positions of the FBG and TC are displayed in Fig. S3a.The Pearson correlation analysis 2,3 is used to quantify their relationship.The Pearson correlation coefficient (PCC) is calculated as 4 : where   and   represent the temperature measured by FBG and TC, respectively.PCC value is between -1 and 1, where 1 or -1 represents a 100% linear relevance, and 0 means 0% relevance.
During a consecutive charge-discharge cycling at 0.5 C, 1 C, 1.5 C and 2 C with 30-minute relaxation setting between charge and discharge for temperature recovery, the internal temperature evolution of the FBG and TC is nearly identical as shown in Fig. S3b.The PCC is 99.86%, the maximum error between the temperature measured by FBG and TC during normal cycling is 0.12 °C, the relative error is merely 0.31%.And when the cell with 100% SOC is triggered thermal runaway by overheating of a 100 W heater, the internal temperature monitored by FBG and TC is also highly overlapped as shown in Fig. S3c.The PCC measured is 99.89%, the absolute error between the maximum temperature measured by FBG and TC during normal cycling is 2.43 °C, the relative error is merely 0.46% as shown in Fig. S3d.
Therefore, for both normal cycling conditions and thermal runaway conditions, the relative temperature errors between thermocouple and FBG are less than 0.5% and the PCCs are higher than 99.85%.

Heating mode and temperature
The thermal runaway is triggered by a heater attached to the cell, and the external heater is turned off near the beginning of Stage III since the internal processes drive the heat generation inside the cell at that time.Note: the subscripts "s" and "in" indicate "surface" and "internal", which represent the surface and internal temperature monitored by thermocouple and FBG, respectively."/" indicates this parameter cam be ignored or does not exist.

14/21
Table S2 summarizes the characteristic parameters of the cell with 0% SOC, 50% SOC and 100% SOC during thermal runaway, it can be seen that the time and temperature at which safety venting occurs show little correlation with the SOC 6 as the electrolyte property is the same for different SOCs, which is attributed to that the safety venting relies heavily on the electrolyte evaporation but not on the generation of gases during chemical reactions 7 .With regards to the internal short circuit, it depends mostly of the separator melting and shrinkage and it displays weak correlations with SOC.

Characterizations of cell over thermal runaway using DSC, SEM, EDS, XRD methods
In-depth thermal runaway mechanism is revealed by post-mortem approaches involving DSC (Differential Scanning Calorimetry), SEM (Scanning Electron Microscope), EDS (Energy Dispersive Spectrometer) and XRD (X-rays Diffraction) in characterizing thermal stability, surface morphology, element composition and structure evolution experiencing thermal runaway, represented by the cell with 100% SOC.

• DSC characterizations
The heat flow of separator, the sample involving lithiated graphite anode, delithiated cathode and electrolyte are measured by the DSC 214 (NETZSCH) where the samples with electrolyte are encapsulated in high pressure gold-plated crucible with argon-filled glove box.
All of the samples are performed with a ramp rate of 10 °C min −1 , where the separator is heated from ambient temperature to 200 °C and the sample of "cathode+anode+electrolyte" is heated from ambient temperature to 450 °C.
The heat flow of the separator (Fig. S12a) indicates that the separator melting starts at 158.0 °C and is followed by the further shrinkage at about 164.0 °C, and the collapse ends at 171.0 °C, which explains exactly that the internal short circuit temperature at the cell with different SOCs all locates within this range as summarized in Table S2.The first exothermic peak of heat flow of the combination of fully lithiated graphite, fully delithiated LFP and electrolyte (Fig. S12b) is associated with SEI decomposition 8 appearing from 84.4 °C, which is the beginning of parasitic side reaction to generate gases for further internal pressure accumulation.Subsequently, the second exothermic peak characterizes the reactions between active materials and electrolyte with the massive reactions starting at 168.4 °C and peaking at 242.4 °C, which is correlated to the trigger temperature of thermal runaway ( TR,s =194.1 °C and  TR,s =171.4 °C), implying that the reactions between active materials and electrolyte occupy the dominant position in triggering thermal runaway.Finally, the third slight 15/21 exothermic peak represents the reaction between graphite electrode with binder 9 from 289.4 °C, resulting in sustained gases generation and accelerated temperature rise rate 10 .•

SEM-EDS characterizations
The SEM-EDS patterns of cathode and anode before/after thermal runaway are acquired by Gemini SEM 500 manufactured by ZEISS.The as-prepared samples before thermal runaway are pasted on a self-made container by conductive glue in glove box, and then quickly put on the SEM-EDS equipment within 30-min after taking container out of the glove box.The asprepared samples after thermal runaway are directly pasted on the sample table in the air.
SEM images, EDS images and EDS spectrum are all obtained.
SEM patterns of anode before thermal runaway with 100%SOC are smooth and flat with the distinct lithiated graphite particles (Fig. S13a and b).Yet there seems to be plenty of impurities covering the graphite surface to lead to a rough pattern showing in Fig. S13e and f, which is estimated to be decomposition products after thermal runaway.The EDS mapping and spectrum in Fig. S13c, d, g, h further confirm this conjecture, where the additional fluorine (F) element with intensified signal is detected on the anode surface after thermal runaway, demonstrating that F-containing substances decomposed from electrolyte are accumulated on anode surface after thermal runaway.It is not coincidence but appeared at repeated samples provided in Fig. S14.The analogous rough SEM images are also discovered at cathode after thermal runaway as shown in Fig. S13i, j, m, n, while difference emerges at merely sporadic impurities appearing on cathode surface rather than covering the whole cathode surface like anode.Although the EDS patterns imply no variation in element distribution and composition for cathode.Remarkably, compared with the SEM images of cathode before thermal runaway, the non-uniform particle size distribution together with indistinct particle boundaries are discovered after thermal runaway, indicating the changed structure that will be disclosed by XRD displayed in Fig. S15.

• XRD characterizations
The XRD pattern of cathode and anode before/after thermal runaway are acquired by Smartlab (Cu Kα) manufactured by Rigaku to reveal the structure evolution during thermal runaway.The anode materials before thermal runaway are scraped off from current collector and ground into powder in glove box, while anode after thermal runaway and cathode are operated with the same procedures in the air.
In Fig. S15a all diffraction peaks before thermal runaway manifest excellent consistency with the standard card of the delithiated LFP, while divergence appears at cathode after thermal runaway that the predominant crystalline phase detected by XRD is Fe2P2O7 and Fe7(PO4)6, demonstrating that LFP has been decomposed with the existence of electrolyte 11,12 .
This observation matches well with the predicted three-step decomposition of FePO4 using DFT calculations 13 as expressed in Eq.( 1).
FePO 4 → 0.1(Fe 7 (PO 4 ) 6 + Fe 3 (P 2 O 7 ) 2 ) → 0.167(Fe 3 (PO 4 ) 2 + Fe 3 (P 2 O 7 ) 2 ) → 0.5Fe 2 P 2 O 7 (1)   As for anode, the existed XRD peaks of LiC6 and LiC12 provide evidence for the lithiated graphite 14 before thermal runaway, the coexistence peak of graphite-2H and graphite-3R is attributed to the unutilized graphite ascribed as the excessive anode capacity in design to avoid Li plating.Similarly, in Fig. S15b and c XRD pattern of anode after thermal runaway exhibits the highly typical graphite crystallized structure with the sharp coexistence peak of graphite-2H and graphite-3R, implying that the intercalated Li can be removed from lithiated graphite at elevated temperature to reserve the graphite structure.This can be rationalized by considering that Lewis acid PF5 decomposed by LiPF6 facilitates Li diffusion by removing electrons from graphite 15,16 .

Fig. S15
The XRD evidences of cathode and anode before/after thermal runaway revealing structure evolution.
(a) XRD patterns of cathode.XRD patterns of anode and zoomed section (c).

Fig. S1
Fig. S1 Response spectra of FBG to temperature and pressure.(a) The full spectral response of FBG from 25 to 600°C, the arrows indicate the direction of wavelength change upon increase of temperature.(b) With the temperature sensitivity of 10.3 pm °C -1 .(c) The full spectral response of FBG sensor in air at 0-2 MPa in 0.2 MPa increment, there exhibits superior linear relationship (d) with the pressure sensitivity of -3.2 pm MPa -1 .(c and d) Fig. S2 Response spectra of FPI to temperature and pressure.(a) The full spectral response of FPI sensor from 25 to 600°C.The three resonance peaks of FPI at ~1520 nm, ~1550 nm and ~1580 nm as a function of T. (b and d) Raw data (symbol) and linear fitting curve (solid line).The temperature response sensitivity of FPI is 0.5 pm °C -1 .(c) The full spectral response of FPI sensor in air at 0-2 MPa in 0.2 MPa increment, the arrows indicate the direction of wavelength change upon increase of pressure.(d) The pressure response sensitivity of FPI is 4185.3pm MPa -1 .

Fig. S3
Fig. S3 Error assessment of the FBG temperature measurement during charge-discharge cycling and thermal runaway.(a) The position of the implanted thermocouple and FBG.(b) The voltage and internal temperature evolution monitored by thermocouple and FBG during normal cycling at 0.5 C, 1 C, 1.5 C and 2 C, where a 30minute relaxation is set between charge and discharge.(c) The voltage and internal temperature evolution monitored by thermocouple and FBG during thermal runaway of the cell with 100% SOC.(d) The Pearson correlation coefficient (PCC), absolute and relative error of maximum temperature during normal cycling and thermal runaway.

Fig. S11
Fig. S11 The heating mode and temperature evolution of the cell with 100% SOC, 50% SOC and 0% SOC.(a) The position of thermocouples and FBG.(b,c,d) Temperature of heater and cell at three cases.

Fig. S12
Fig. S12 The DSC results and the derived side reaction sequence to reveal thermal runaway mechanism.(a, b) The heat flow curves of separator (a) and graphite in fully lithiated state, LFP in fully delithiated state with electrolyte (b) monitored by DSC.(c) The side reaction sequence with the corresponding temperature.(d, e) The repeated DSC curves of separator (d) and an+ca+ele (e).

Fig. S13
Fig. S13 The SEM-EDS patterns exhibiting surface morphology and elemental composition before and after thermal runaway.(a-d) Anode before thermal runaway.(e-h) Anode after thermal runaway.(i-l) Cathode before thermal runaway.(m-p) Cathode after thermal runaway.a, b, e, f, I, j, m, n, SEM pattern.c, g, k, o, EDS mapping.d, h, l, p, EDS spectrum.
Fig. S14 The repeated SEM-EDS patterns demonstrating rough surface morphology with impurities and additional fluorine (F) element after thermal runaway.(a) The anode surface morphology after thermal runaway.(b) The cathode surface morphology after thermal runaway.(c) The EDS mapping and spectrum of the anode after thermal runaway.

Table S2
Characteristic parameters of the cell with 0% SOC, 50% SOC and 100% SOC during thermal runaway