Coaxially nested component with asymmetric fiber resonant cavity and separation membrane for gaseous and dissolved gases detection

Abstract

Dynamically, simultaneously and sensitively embedded or external detection of multi-component gaseous or dissolved phase gases in millilitre volumes is critical for many frontier applications. Raman spectroscopy based on a single-wavelength laser can achieve simultaneous detection of various gases (except noble gases), but it lacks sensitivity. Therefore, this paper firstly develops a theoretical framework for Raman signal generation within the fiber resonant cavity (FRC), and based on which, an asymmetric FRC is proposed to balance the laser power enhancement and the intracavity loss caused by a high-reflectivity input cavity mirror, enhancing the Raman signal intensity by 170 times when compared to only hollow-core fiber and 36 times to only geometry resonant cavity, achieving sensitive gas detection with a limit of detection (LOD) as low as 0.01 ppm*bar. Meanwhile, a coaxially nested component with FRC and separation membrane is proposed, which can enable approximate synchronization of liquid-gas separation and detection. This work provides an effective solution of measurement instruments to realize d high-precision detection of multi component gases in various states with gas consumption ranging from 0.3 mL to 2 mL, enabling further refinement of the associated generation and decomposition pathways of different gases, and the coaxially nested component also can be used in other scattering-based spectroscopic techniques (e.g., Brillouin spectroscopy) and absorption-based spectroscopic techniques.

Simultaneous and sensitive Raman detection of gaseous or dissolved phase gases is critical. Here, authors report for the first time a coaxially nested component consisting of a side-drilled-fiber resonant cavity and a membrane for enabling near simultaneous gas generation, diffusion, separation, and sensitive detection.

Introduction

Disease diagnosis based on breath analysis, status assessment of battery and large-scale equipment such as power equipment and space rocket, etc. urgently require dynamic, simultaneous and highly sensitive (0.1 ppm level) detection of multi-component gaseous gases^1–3 or dissolved gases^4–7 in low-volume (a few ml) whether it is embedded detection by diffusion or external detection by pump. Conventional absorption spectroscopy^8–10, photoacoustic spectroscopy, and photothermal spectroscopy^11–13 based on absorption effect can typically employ wavelength-specific lasers preconfigured for known gases to perform time-sharing detection of multi-component gases. Notably, Ye et al.¹⁴ proposed a modulated ringdown comb interferometry method for simultaneous measurement of multiple gases. By sweeping the comb lines across the cavity resonance with sufficient modulation depth, all comb lines can be coupled through the cavity in sequence, and then the complex cavity ring-down mode containing gas information in the cavity can be extracted with the help of Michelson interferometer and non-uniform Fourier transform. However, this method faces challenges in detecting homonuclear diatomic molecules (e.g., H₂ and O₂) and suffers from relatively complex and expensive system. Raman spectroscopy using a single-wavelength laser can simultaneously and cost-effectively detect highly complex gas mixtures even with undefined composition (excluding noble gases) while performing real-time calibration during detection^15–17. However, its sensitivity remains limited. At present, multi-pass cavity^18,19, resonant cavity^20,21, and hollow-core fiber^22–25 are often used to improve the interaction distance, laser power build-up within the cavity, or the Raman signal collection efficiency, improving the gas detection sensitivity. However, the Raman signal enhancement in multi-pass cavity is limited by the finite number of light reflections, with detection limit typically at 10 ppm*bar levels. High-finesse resonant cavity, by establishing intracavity laser resonance, can achieve a detection limit of 0.015 ppm*bar²¹. However, this approach requires gas sample volumes exceeding 1 L. The hollow-core fiber effectively enhances Raman signal collection efficiency (spontaneous Raman scattering is an 4Pi process) by confining both the excitation laser and the gas within the hollow-core area, enabling Raman spectroscopy to achieve ppm-level detection limits^26–30, with a gas sample volume of a few mL, but it is incapable of detecting dissolved gases.

In this paper, we propose a coaxially nested component with asymmetric fiber resonant cavity and separation membrane for gaseous and dissolved gases detection by Raman spectroscopy. At first, synergistic enhancement model for Raman signal is built-up. With that, an asymmetric-reflectivity configuration is proposed to reduce the transmission losses of both the laser and Raman signals within the cavity, with the gain effect of the asymmetric reflectivity FRC is 2.5 to 22 times higher than that of the symmetric reflectivity case at different wavenumbers, achieving a Raman detection sensitivity as low as 0.01 ppm*bar. In addition, a coaxially nested component with asymmetric FRC and separation membrane for integration of liquid-gas separation and detection is proposed, which enables simultaneous liquid–gas separation and detection. To verify the effectiveness of this technology, Raman spectroscopy detection experiments of human breath gas, dissolved gases in battery electrolytes, dissolved gases in mineral insulating oil, SF₆ and C₅F₁₀O as well as their respective decomposition products by partial discharge fault are carried out with gas sample consumption ranging from 0.3 mL to 2 mL.

Results

Theory

To obtain the optimal structural parameters for maximum signal enhancement, the transmission characteristics of laser light and Raman scattered light in the FRC are studied, as shown in Fig. 1.

Fig. 1. The transmission characteristics within fiber resonant cavity.

a laser field build-up process of within the FRC. b The output process of the backward Raman scattering signal after multiple reflections within the FRC.

The build-up process of the laser light field within the cavity is shown in Fig. 1a. At this stage, the transmission characteristic of the built-up laser light field at point z, denoted as l(z), can be expressed as:

$$l\left(z\right)=\frac{t_{1L}\left(e^{\frac{-\alpha z}{2}}{e}^{-i\Delta \theta z}+r_{2L}{e}^{\frac{-\alpha \left(2d-z\right)}{2}}{e}^{-i\Delta \theta \left(2d-z\right)}\right)}{1-r_{1L}{r}_{2L}{e}^{-i2\Delta \theta d}{e}^{-\alpha d}}$$ 1

where t_1L represents the transmission coefficient of the input mirror to the laser light field, r_1L and r_2L represent the reflection coefficient of the input and output mirrors to the laser light field respectively, $\Delta \theta =2\pi n_f/\lambda$ is the phase shift per unit length of the laser light field propagating in the hollow-core fiber, α represents the transmission media’s loss of the laser light, and d represents cavity length of the FRC.

Therefore, the transmission characteristic of laser power at point z, T_L (z), can be expressed as:

$$T_L\left(z\right)=l\left(z\right)l^{*}\left(z\right)$$ 2

where l^*(z) represents the conjugate of l(z). On the other hand, the Raman signal undergoes multiple reflections within the FRC and exits through the input mirror and out mirror, the exiting process through the input mirror is as shown in Fig. 1b. The transmission characteristic of the backward Raman scattering signal T_R (z, λ) can be expressed as:

$$T_R\left(z,\lambda \right)=\frac{\left(e^{-\beta \left(\lambda \right)z}+r_{2R}^2\left(\lambda \right)e^{-\beta \left(\lambda \right)\left[2d-z\right]}\right)t_{1R}^2\left(\lambda \right)}{1-r_{1R}^2\left(\lambda \right)r_{2R}^2\left(\lambda \right)e^{-2\beta \left(\lambda \right)d}}\frac{{\Omega }_{fiber}}{4\pi }{\eta }_{\mathrm{mod}e}$$ 3

where $t_{1R}^2\left(\lambda \right)$ represents the transmittance of the input mirror for Raman signal varies with wavelength λ, $r_{1R}^2\left(\lambda \right)$ and $r_{2R}^2\left(\lambda \right)$ represent reflectivity of the input and output mirrors for the Raman signal, β(λ) represents the transmission loss of the Raman signal inside the FRC as function of wavelength λ, and Ω_fiber represents the acceptance angle of the hollow-core fiber. The ratio Ω_fiber /4π denotes the collection efficiency of the hollow-core fiber for the 4π Raman signal generated within the FRC. ${\eta }_{\mathrm{mod}\mathrm{e}}$ is modes collection efficiency of Raman signals^26–28. Therefore, the backward Raman scattered signal from the FRC can be expressed as:

$$I_R\left(d,\lambda \right)\propto m\sigma \pi {\left(\frac{D}{2}\right)}^2{\eta }_{\mathrm{coupling}}{I}_0{\int }_0^d{T}_L\left(z\right)T_R\left(z,\lambda \right)dz$$ 4

where m represent the concentration of gas sample; σ is scattering cross-section of gas, respectively, which is proportional to the inverse fourth power of laser wavelength; I₀ is the intensity of the incident laser; η_coupling represents the collection efficiency of the spectrometer for Raman signals, and T is exposure time of the spectrometer.

Equation (4) tell us during the resonant enhancement of laser light undergoing round trips within the resonant cavity, Raman signals can be excited at arbitrary positions and achieve a synergistic enhancement. To analyze the influence of different parameters on the Raman intensity, we first consider the transmission loss of the hollow-core fiber. Decreasing the transmission loss of the hollow-core fiber will lead to an exponential growth in the intracavity built-up laser power. Meanwhile, the enhancement of the Raman signal is also influenced by the reflectivity of both the output and input mirrors. A higher reflectivity of the output mirror allows more laser light to return into the cavity, enhancing the backward Raman signal. Since excessively high reflectivity is prone to contamination and current hollow-core fibers in the visible range are still far from achieving the typical several dB/km loss level of single-mode fibers in visible band (the hollow-core fiber used in the experiment, both α and β are assigned 0.06 dB/m^31,32), $r_{2L}^2$ is taken as 0.999. The reflectivity of the input mirror simultaneously affects both the build-up of intracavity laser power and the transmission loss of Raman signals from the FRC. Simply selecting an input mirror with higher-reflectivity to enhance the intracavity built-up laser power does not necessarily ensure a higher Raman signal enhancement. Therefore, the influence of different values of $r_{1L}^2$ on the FRC Raman intensity is analyzed, while assuming $r_{2R}^2$ and $r_{1R}^2$ are both 0.97.

The Raman intensity under resonance in the FRC at different values of $r_{1L}^2$ can be calculated using Eq. (4), as shown in Fig. 1. The figure shows that the Raman signal reaches maximum intensity when $r_{1L}^2$ is 0.992(we should emphasize here the reflectivity of the input cavity mirror is not fixed; its optimal reflectivity decreases as the intracavity loss increases). Therefore, we propose an FRC with an asymmetric-reflectivity mirror to enhance the Raman intensity. However, Raman enhancement in the FRC depends on both the fiber length and wavelength-dependent reflectivity of the cavity mirrors. To analyze the Raman intensity variation across different wavenumbers, the measured mirror reflectivity (as shown in Supplementary Fig. 2) were used in Eq. (4) to calculate the Raman intensity for different fiber lengths, as shown in Fig. 2.

Fig. 2. Wavelength-dependent Raman enhancement influenced by mirror reflectivity.

a Cavity length optimization across the spectral range. b Performance comparison of symmetric and asymmetric reflectivity FRC at the optimal length.

Figure 2a shows that the hollow-core fiber with a length of 32 cm provides stronger Raman enhancement than other lengths for most wavenumbers. Figure 2b compares the Raman enhancement of the optimized asymmetric reflectivity FRC and the symmetric reflectivity FRC. Since the transmission loss of hollow-core anti-resonant fiber (HC-ARF) in the visible band cannot be neglected, using an $r_{1L}^2$ of 0.999 causes laser light and the low-wavenumber Raman signals to circulate more times within the HC-ARF. This leads to higher intracavity loss and overall lower Raman intensity than when $r_{1L}^2$ is 0.992, as illustrated by the black squares and circles in Fig. 2b. The red dots in Fig. 2b are the Raman enhancement ratios of the two configurations. It can be seen that the gain effect of the asymmetric reflectivity FRC is 2.5 to 22 times higher than that of the symmetric reflectivity case, which demonstrates the remarkable performance of the asymmetric reflectivity FRC. However, the theoretical enhancement shown in Fig. 2b is limited by the intracavity round-trip losses of both the laser and Raman signal, as well as by the laser linewidth. Unlike the single-pass transmission loss measured with a broadband source, the round-trip loss accumulates over multiple resonance cycles. Minor alignment imperfections can lead to a small fraction of light coupling into higher-order modes, resulting in a slightly increased effective round-trip loss compared to the single-pass loss of the fundamental mode. In addition, the laser linewidth should be much narrower than cavity resonance width to maintain stable resonance coupling and achieve enhancement close to theoretical value.

Experimental setup

The schematic diagram of the coaxially nested component for integration of liquid-gas separation and detection is shown in Fig. 3a, where the FRC is formed by the tight coupling of large mode area fibers (LMAFs) with mirrored ends and a drilled hollow-core anti-resonant fiber^33,34. The composition of the liquid-gas separation membrane is divided into two cases. For gaseous or water-dissolved gas detection, a hydrophobic Polytetrafluoroethylene (PTFE) membrane with an internal porous structure (as shown in Supplementary Fig. 3a–b) is used as the separation membrane, which can not only filter impurities in the sample gas but also achieve water-gas separation. For oil-dissolved gas detection, a Teflon AF2400 membrane is deposited on the surface of PTFE membrane via dip-coating to achieve oil-gas separation (as shown in Supplementary Fig. 3c–d).

Fig. 3. Gaseous and dissolved characteristic gases detection system based on FRCERS.

a The schematic diagram of the coaxially nested component. b The mechanical structure of the coaxially nested component. c The schematic diagram of the gaseous and dissolved characteristic gases detection system.

Figure 3b shows the mechanical structure of the coaxially nested component (more details can be found in Supplementary Note 4), and its fabrication process and characteristics can be found in Methods section. Among them, both gas and liquid samples can enter through port 1. For gas detection, port 2 should be closed, and the sample gas is filtered through the separation membrane into the FRC for detection. For dissolved gas detection, a peristaltic pump circulates the sample through ports 1 and 2, and then the separation membrane separates the characteristic gas dissolved in the sample into the hollow-core area of the drilled HC-ARF for detection.

Figure 3c is the schematic diagram of the gaseous and dissolved characteristic gases detection system based on FRCERS. The laser (Precilasers, 532 nm) is modulated at a high frequency (20 MHz) by an electro-optic modulator (EOM) and then aligned collinearly by adjusting the tilt and rotation of a mirror and a dichroic mirror. The aligned laser is coupled into the FRC using an adjustable lens and a five-axis nano-positioning stage. Under optimal coupling, the Pound-Drever-Hall (PDH) error signal exhibits a monotonic linear region with a single zero-crossing, enabling stable frequency locking to the FRC resonance. However, if fiber bending or alignment errors occur, the error signal becomes non-monotonic and introduces multiple spurious zero-crossings, which severely degrade the locking stability. To address this issue, a loose PDH locking technique has been proposed to achieve stable locking (can be maintained over many tens of hours, further details are provided in Supplementary Note 6). The theoretical finesse of asymmetric-reflectivity FRC can be derived as:

$$F_{\mathrm{theory}}=\frac{\pi \sqrt[2]{r_{1L}{r}_{2L}{e}^{-\alpha d-\epsilon }}}{1-r_{1L}{r}_{2L}{e}^{-\alpha d-\epsilon }}$$ 5

where ε is total loss from mirror scattering and other non-ideal factors apart from intrinsic fiber loss.

The backward Raman signal generated by FRC is coupled into the spectrometer (HRS-300-S, Princeton Instruments) via M2. The edge filter (Semrock, LP03-532RE-25) in front of the spectrometer is used to suppress Rayleigh scattering at laser wavelength. Additionally, an aperture is placed before the edge filter, in combination with the spectrometer’s slit and CCD, can effectively suppress the background signal^35,36 generated by the FRC (further details are provided in Supplementary Note 7).

The enhancement performance of asymmetric-reflectivity FRC

To evaluate the influence of input mirror reflectivity on Raman signal enhancement performance, FRCs consisting of a 32 cm HC-ARF (transmission loss: 0.06 dB/m) and mirrors with different reflectivity are constructed. Under a laser power of 500 mW and a measurement time of 200 s, the Raman spectra of a standard-concentration gas mixture were measured using symmetric- and asymmetric-reflectivity FRCs, as shown in Supplementary Fig. 16. Compared to Supplementary Fig. 16a and b, it is evident that for the symmetric-reflectivity FRC, the Raman signal is significantly suppressed at wavenumbers below 2500 cm⁻¹, and this suppression trend is consistent with the simulation results (red line in Supplementary Fig. 16a). Supplementary Fig. 16c presents a comparison between theoretical and measured enhancement factors, showing a similar trend for both. The measured enhancement factors agree with the simulated values at wavenumbers above 2000 cm⁻¹, whereas below 2000 cm⁻¹, the measured enhancement deviates more noticeably from the theoretical predictions. This discrepancy may be attributed to the relatively asymmetric intracavity light field in the asymmetric-reflectivity FRC, which can lead to low-wavenumber Raman signals more easily exciting higher-order transverse modes in the hollow-core fiber during propagation, thereby increasing transmission loss at low wavenumbers and reducing the corresponding Raman enhancement. To demonstrate the Raman signal enhancement effects of the asymmetric-reflectivity FRC, Raman spectra of ambient air are collected using asymmetric-reflectivity FRC(${\mathrm{r}}_{1\mathrm{L}}^2$ of 0.992, ${\mathrm{r}}_{\mathrm{2L}}^2$ of 0.999), symmetric-reflectivity FRC(${\mathrm{r}}_{1\mathrm{L}}^2$ of 0.999, ${\mathrm{r}}_{\mathrm{2L}}^2$ of 0.999), single hollow-core fiber(α of 0.06 dB/m and d of 32 cm), and resonant cavity (${\mathrm{r}}_{1\mathrm{L}}^2$ of 0.999, ${\mathrm{r}}_{\mathrm{2L}}^2$ of 0.999) at a power of 18 mW and a measurement time of 30 s, as shown in Supplementary Fig. 17. After noise standardization, the corresponding N₂ Raman signal intensities of the latter three were 6.33 × 10⁵, 2.24 × 10⁴ and 1.06 × 10⁵, which are 6 (the difference to theoretical value of 1.6 times in Fig. 2b is from the difference between true and simulation value of input mirror reflectivity for the N₂ Raman signal, as shown Supplementary Fig. 2a), 170 times and 36 times lower than 3.81 × 10⁶ of the proposed asymmetric-reflectivity FRC. Meanwhile, asymmetric-reflectivity FRC also exhibits good operation stability, and the possibility of laser-induced heating inside the cavity has been discussed in Supplementary Note 10.

To evaluate the performance of the coaxially nested component for gaseous and dissolved characteristic gases in real samples, embedded and external detection experiments were conducted at a laser power of 500 mW on various samples including human breath gases, dissolved gases in battery electrolyte, and dissolved gases in mineral insulating oil.

Human breath gas detection

The detection of CO₂ isotopes, CH₄, and H₂ in breath gases can be used for diagnosing gastrointestinal diseases. Before gas detection, the coaxially nested component is first evacuated. The sample gas from a 28-year-old volunteer then flows into the component under negative pressure, with a response time of 20 s, as shown in Supplementary Fig. 20. To compare the changes in gas composition during human respiration, the coaxially nested component is used to analyse ambient air and breath samples, as shown in Fig. 4.

Fig. 4. Comparison of detection results between ambient air and human breath gas.

Ambient air is drawn into the coaxially nested assembly by a vacuum pump for detection, human breath samples are collected using aluminium foil sampling bags and then introduced into the coaxially nested component for detection.

Due to the high CO₂ concentration in breath gas, its rotational-vibrational peak can be clearly obtained within 0.2 s of spectral acquisition. Using the CO₂ peak at 1388 cm⁻¹ for quantitative analysis, the CO₂ concentration in breath is determined to be 3.9%. Moreover, a distinct characteristic peak at 1370 cm⁻¹, attributed to ¹³CO₂ is observed, which can be utilized for diagnosing gastric and duodenal diseases. When the measurement times are 60 s and 200 s, the CH₄ (v₂ vibrational mode, 2915 cm⁻¹) and H₂ (Q1(1) at 4157 cm⁻¹), which can reflect the state of gut microbiota, can be observed in breath gas. The corresponding detected concentrations of CH₄ and H₂ are 2.1 ppm and 2.4 ppm, respectively. The signal-to-noise ratios (SNR) of Raman peaks are calculated as 437.2 and 105.9, respectively (calibration curve is shown in Supplementary Fig. 21). Corresponding limits of detection (LOD) for CH₄ and H₂ at SNR of 3σ are 0.014 ppm*bar and 0.068 ppm*bar. In addition, reproducibility of LOD is shown in Supplementary Fig. 22.

Detection of dissolved gases in battery electrolytes

When a lithium-ion battery (LIB) experiences internal faults such as local internal short circuits, overcharging, thermal runaway, or mechanical damage, the electrolyte and electrode materials of the LIB decompose and generate gases, and then dissolve into the electrolyte. Detecting dissolved gases (CH₄, C₂H₄, H₂, CO₂, CO and HF, etc.) in the electrolyte can help determine the type of fault and the decomposition pathways of LIB materials^35,36. In this experiment, a LiNi₁Co₁Mn₁O₂ (NCM111)-graphite LIB (the maximum charging voltage is 4.2 V) is used as a sample and the coaxially nested component is implanted into the battery, as shown in Supplementary Fig. 23. The implantation procedure is shown in the Methods section. During electrolyte detection, the measurement time was set to 200 s and the Raman spectra (with a spectral resolution of 1.59 cm⁻¹) of dissolved gases in the electrolyte, before and after multiple overcharge cycles, are shown in Fig. 5.

Fig. 5. Raman spectra of dissolved gases in the battery electrolyte before and after overcharging.

Before charging, the Raman spectrum of dissolved gases in the electrolyte of the NCM111-graphite LIB was collected with a measurement time of 200 s. Subsequently, the battery was cycled for 2 h at 4.3 V, and then the Raman spectrum of dissolved gases in the electrolyte after overcharging was collected with the same measurement time.

Before overcharging (black line), slight decomposition of the electrolyte occurs, leading to the appearance of CO₂ Raman peaks at 1285 cm⁻¹ and 1388 cm⁻¹, C₂H₄ Raman peaks at 1346 cm⁻¹, 1627 cm⁻¹, and 3018 cm⁻¹, as shown by the black line in Fig. 5. Peaks 1388 cm⁻¹ and 3018 cm⁻¹ are used for quantitative analysis of CO₂ and C₂H₄, and their concentrations are 40.3 ppm and 4.5 ppm, respectively (the calibration curve is shown in Supplementary Fig. 24). After overcharging (red line), characteristic gas peaks of H₂, CO₂, CH₄, C₂H₄ and CO can be clearly observed. The lithium dendrite growth caused by overcharging can penetrate the solid electrolyte interphase (SEI) layer, leading to further decomposition of the electrolyte and generating large amounts of CO₂ and C₂H₄. The S₀(1), S₀(2) and S₀(3) vibration peaks of H₂ can be observed at 582 cm⁻¹, 812 cm⁻¹ and 1034 cm⁻¹, which are generated by the reaction between the graphite electrode and the binder. The v₀ vibration mode produced by CO can be found at 2146 cm⁻¹, which is generated by the decomposition of the battery electrolyte and the SEI layer. The v₁ vibration mode of CH₄ can be found at 2915 cm⁻¹, which also resulting from SEI layer decomposition. In addition, by comparing the spectra before and after overcharging, a weak Raman peak near 3958 cm⁻¹ is observed, which corresponds to the v₀ vibrational mode of hydrofluoric acid. Using the Raman peaks at 1388 cm⁻¹, 3018 cm⁻¹, 582 cm⁻¹, 2145 cm⁻¹ and 2915 cm⁻¹ for quantitative analysis, the concentrations of dissolved CO₂, C₂H₄, H₂, CO and CH₄ in the electrolyte are determined to be 2135.9 ppm, 1673.4 ppm, 3657.3 ppm, 146.8 ppm and 133.6ppm, respectively. Benefiting from the capability of simultaneously and dynamically detecting multiple gaseous species, the proposed technique enables further refinement of the battery material decomposition pathways during LIB failure (the reaction pathways provided in the Supplementary Note 14).

Detection of dissolved gases in mineral insulating oil

When internal faults (such as overheating of oil and paper, partial discharge, spark discharge, arc discharge, etc.) occur oil-immersed power equipment, the internal insulating materials (insulating oil (mineral insulating oil, vegetable insulating oil, synthetic ester insulating oil, etc.) and insulating paper of the equipment will decompose and generate gases, and then dissolve in the insulating oil. Detecting the gases dissolved in the insulating oil can be used to determine the fault type of equipment and to better understand the decomposition pathways of the insulating materials. To detect dissolved gases in insulating oil, the gases are generally separated from the oil and then pumped into the gas chamber for detection. The whole process takes up to 3 h, resulting in delayed fault detection. To evaluate the application performance of coaxially nested component in this field, a needle-plate electrode is used to perform breakdown discharge in a liquid pool of mineral insulating oil (the oil temperature was stabilized at 50 °C using a temperature control system, as shown in Supplementary Fig. 25) for 10 min. Then, mineral insulating oil containing dissolved gases was circulated into coaxially nested component at flow rate of 50 mL/min for gas detection (measurement time of 60 s). The variation trends of each gas concentrations with separation time are shown by black dots in Fig. 6 (characteristic peaks for gas are selected as follows: H₂: 852 cm⁻¹, C₂H₂: 1976 cm⁻¹, CO: 2145 cm⁻¹, CH₄: 2915 cm⁻¹, C₂H₄: 3018 cm⁻¹, and C₂H₆: 2952 cm⁻¹).

Fig. 6. Comparison of the measured and estimated gas-liquid separation trends of dissolved gases in mineral insulating oil.

a H₂, b C₂H₂, c CO, d CH₄, e C₂H₆, f C₂H₄.

The equilibrium concentrations of dissolved gases are determined as follows: H₂: 14.75 ppm C₂H₂: 54.04 ppm, CO: 26.63 ppm, CH₄: 4.86 ppm, C₂H₆: 9.36 ppm, and C₂H₄: 6.16 ppm (the calibration curve is shown in Supplementary Fig. 26). Meanwhile, the physical equilibrium time of each gas can be effectively reduced to less than 50 min (further details on the effects of membrane parameters on diffusion kinetics can be found in Supplementary Note 19). Furthermore, to enable rapid evaluation of gas concentration in mineral insulating oil, a dynamic estimation algorithm based on the dissolution-diffusion process of separation membrane is proposed (the flowchart is shown in Supplementary Fig. 27). The measurement results from the first 10 min were used as fitting data to estimate the gas concentration trends during the gas-oil separation process beyond 10 min. The estimated results are shown by the red line in Fig. 6. It can be seen that the separation times required to reach 90% of stable concentration for each gas are as follows: H₂: 49 min, C₂H₂: 35 min, CO: 48 min, CH₄: 37 min, C₂H₆: 35 min, and C₂H₄: 37 min. Meanwhile, the estimated equilibrium concentrations of each gas are: H₂: 15.05 ppm C₂H₂: 55.13 ppm, CO: 26.83 ppm, CH₄: 4.99 ppm, C₂H₆: 9.50 ppm, and C₂H₄: 6.29 ppm, with corresponding absolute errors of 2.10%, 2.02%, 0.75%, 2.67%, 1.50% and 2.11%, respectively. All absolute errors are within 3%, demonstrating the effectiveness of the dynamic estimation algorithm. The accurate and fast detection of dissolved gases in insulating oil can serve for refining decomposition pathways of insulation oil and insulation paper under different faults, which can be summarized in Supplementary Note 20.

Discussion

This study demonstrates that the coaxially nested component consisting of an asymmetric-reflectivity FRC and separation membrane can achieve a LOD of gas Raman spectroscopy as low as 0.01 ppm*bar, with a gas consumption ranging from 0.3 mL to 2 mL.

Firstly, we proposed the Raman signal intensity model of the fiber resonant cavity to optimize the cavity structural parameters. By analysing the influence of the FRC construction parameters on Raman signal intensity, it is found that the high transmission loss of HC-ARF in the visible band (about 0.06 dB/m, 30 times higher than in the infrared band^33,37) leads to increased Raman signal attenuation when higher-reflectivity mirrors are used. Based on the Raman signal intensity model of FRC, it is found that an asymmetric-reflectivity FRC (with the input mirror reflectivity slightly lower than that of the output mirror) can reduce the transmission loss of the laser and Raman signal in FRC, the gain effect for N₂ at 2328 cm⁻¹ of the asymmetric reflectivity FRC is 6 times higher than that of the symmetric reflectivity case. We should emphasize here the reflectivity of the input cavity mirror is not fixed; its optimal reflectivity decreases as the intracavity loss increases.

At a laser power of 500 mW and a measurement time of 60 s, the LOD of CH₄ can reach 0.014 ppm*bar. Compared with the normalized LOD⁴¹ of CH₄ (0.40 ppm *bar*W * $\sqrt{s}$) in Reference²¹ (using a Fabry–Perot cavity with symmetric reflectivity of the cavity mirrors, both at 99.996%), this work achieves a value of 0.054 ppm *bar*W * $\sqrt{s}$, which is about 7.4 times lower (gas consumption is at least 200 times lower) and represents the best Raman spectroscopy LOD reported to date. In addition, our technique offers a more favourable combination of sensitivity and sensor compactness when compared with existing Raman-based gas sensing approaches, as shown in Table 1.

Table 1.

Sensitivity, length and gas consumption comparison of different Raman gas sensor

Ref	Enhancement method	Sensitivity (ppm*bar W* $\sqrt{\mathrm{s}}$)	Length	Gas consumption
18	Multi-pass cavity	CH4: 2.35	3 m	>3 L
19	Multi-pass cavity	CO2: 339.41	0.3 m	>2 L
20	Resonant cavity	CO2: 60.74	0.3 m	>200 mL
21	Resonant cavity	CH4: 0.40	0.7 m	>425 mL
28	HC-ARF	CH4:1.59	12 m	/
Our work	FRC	CH4:0.054	0.32 m	0.3 mL to 2 mL

Finally, a coaxially nested component consisting of side-drilled FRC and separation membrane is proposed, which can enable approximate synchronization of liquid-gas separation and detection. Based on which, high-sensitivity and dynamic Raman detection of gaseous and dissolved characteristic gases in the fields of medical diagnostics, energy storage, and energy power systems are achieved. For gaseous gas detection, the response time is less than 20 s; for oil-dissolved gas detection, when combining with the proposed dynamic estimation algorithm, the response time is less than 10 min. Additionally, we demonstrate the applicability of this technique for dynamic, highly sensitive and simultaneous detection of SF₆ and C₅F₁₀O as well as their respective decomposition products. Based on which, the corresponding refined decomposition pathways of SF₆ and C₅F₁₀O are provided in Supplementary Note 21.

In the future, the sensitivity of this component can be improved by further reducing the transmission loss of HC-ARF in the visible band. If the loss can be further reduced to 0.001 dB/m, the sensitivity of Raman spectroscopy in this study can reach ppb level or even ppt level. Sensitivity may also be enhanced by increasing the cavity finesse through higher output mirror reflectivity (one should know that higher finesse makes frequency stabilization more difficult and may limit applicability in harsh environments) or by combining the coaxially nested architecture with absorption based methods such as absorption, photothermal, or photoacoustic spectroscopy, etc.

Methods

Fabrication and characterization of coaxially nested component

The coaxially nested component is composed of an asymmetric hollow-core FRC and a separation membrane, in which the FRC is composed of a 32 cm long HC-ARF and LMAFs with high-reflection mirrored ends. The physical picture is shown in Supplementary Fig. 4, and the fabrication process is as follows: (1) the LMAFs are first fixed on the ceramic ferrule with epoxy resin, and then high-reflectivity TiO₂/SiO₂ dielectric coatings with reflectivity of 99.2% and 99.9% (corresponding to the 532 nm band) are deposited onto the end faces of the two LMAFs via ion beam deposition; (2) the 27.5 cm long separation membrane is passed through the drilled HC-ARF and fixed to the end of the ceramic ferrule by epoxy resin; (3) the end faces of the HC-ARF are tightly attached to the mirrored end faces of the LMAFs and fixed with a ceramic sleeve to ensure that the resonant light does not leak from the cavity (the coupling efficiency from large-mode-area fiber to HC-ARF reaches 90%).

At the same time, a gas cell composed of two tees and a Teflon tube is designed to enable the coaxially nested component to realize the detection of gaseous and dissolved characteristic gases, as shown in Supplementary Fig. 4c. Meanwhile, the ceramic ferrule of the HC-ARF is sealed and fixed in the tee with epoxy resin, ensuring the airtightness of the FRC.

Subject definition and test conditions in breath analysis

A 28-year-old healthy male volunteer who had fasted overnight participated in the morning measurement. Human breath samples were collected in aluminum foil sampling bags and subsequently introduced into the coaxially nested component by evacuating the component to introduce the gas sample.

The process of coaxially nested component implanted into NCM111-graphite LIB

The coaxially nested component is inserted through the interior of the NCM111-graphite LIB, allowing the dissolved gases in the LIB electrolyte to pass through the separation membrane of the component and enter the drilled HC-ARF for the detection of dissolved gases, as shown in Supplementary Fig. 23. The implantation procedure is as follows: (1) inside an argon-filled glovebox, 1.5 mm-diameter holes are drilled at both ends of the aluminium-plastic film of the NCM111-graphite LIB using a bench drill. Then, the drilled HC-ARF equipped with a separation membrane is inserted through the LIB and sealed in place with epoxy resin; (2) LMAFs with high-reflection mirrored ends are connected to the HC-ARF using a ceramic sleeve to form a coaxially nested component; (3) the interfaces between the ceramic sleeve and the ceramic ferrule are sealed with epoxy resin to prevent oxygen from entering the NCM111-graphite LIB.

Data acquisition and processing

In our experiments, Raman spectra were recorded using an HRS-300-S spectrometer (Princeton Instruments) equipped with a PIXIS™ 400B CCD detector and a 1200 g/mm grating. Baseline correction was performed using the asymmetric least squares smoothing method with a smoothing factor of 2, a threshold of 0.001, and 10 iterations.

Author contributions

The manuscript was written through the contributions of all authors. F.W. and W.K. conceived this idea, designed and performed most of the experiments and analyzed the data. H.S. performed the simulation and the analytical modelling, T.R. built the separation membrane. F.W. and W.C. supervised the project.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by State Key Laboratory of Power Transmission Equipment Technology (Project No. SKLPET-01-2024-03), National Natural Science Foundation of China (Project No. 62275035) and Fundamental Research Funds for the Central Universities (Project No. 2024IAIS-QN014).

Data availability

The data supporting the findings of this article are provided in Zenodo under the accession code DOI link of 10.5281/zenodo.20797299.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75228-w.