Anti-interference and noise suppression system and method for multi-parameter gas detection in coal mine underground full-optical cooperation

Abstract

The present application relates to the technical field of gas detection, and discloses a multi-parameter gas detection anti-interference and noise suppression system and method based on full-optical cooperation in a coal mine, which collects original optical characteristic signals in a detection area in the coal mine, constructs an interference characteristic parameter set and an environmental compensation vector; calculates real-time estimated noise variance and real-time reliability weight of each detection channel based on the interference characteristic parameters; inverses the original optical characteristic signals to obtain original concentration, carries out regression compensation by using the environmental compensation vector, and carries out weighted fusion on the compensated concentration based on the real-time reliability weight to output final fused concentration; and finally outputs a composite decision in combination with the interference characteristic parameters; the present application realizes active perception and adaptive weighted fusion of the health state of the detection channel by establishing quantitative mapping of the interference parameters and the noise variance, effectively solves the measurement distortion problem caused by performance degradation of the sensor, and significantly improves the robustness and accuracy of the system.

Description

Technical Field

[0001] This invention relates to the field of gas detection technology, and more specifically, to a multi-parameter gas detection anti-interference and noise suppression system and method with all-optical coordination in underground coal mines. Background Technology

[0002] Underground gas monitoring in coal mines is a core component of ensuring safe production. Due to the extremely harsh underground environment, characterized by high dust levels, high humidity and heat, strong vibrations, and complex electromagnetic interference, traditional electrochemical sensors often suffer from poisoning failure, large zero-point drift, and short maintenance cycles. In recent years, optical gas detection technology based on spectral absorption principles has gradually become the mainstream development direction for precision underground detection due to its advantages such as high selectivity, long lifespan, and calibration-free operation. To improve the accuracy and reliability of detection, the industry has begun to explore the use of multi-sensor fusion technology to address interference from complex environments.

[0003] In the prior art, for example, Chinese Patent No. CN120470236B discloses a method for multi-parameter environmental monitoring in underground mines. This method uses multi-source sensors to collect equipment pose, vibration spectrum, and environmental parameters in real time, performs spatiotemporal alignment and filtering noise reduction on the raw data, and intelligently compensates for sensor reading errors based on the coupling model of vibration spectrum and pose drift, thereby eliminating the interference of mechanical vibration on the accuracy of gas monitoring. Another Chinese patent application, CN109459403A, discloses a multi-gas detection device and method in coal mines. This device detects the concentration of multiple gases through an integrated gas acquisition unit and multi-parameter gas sensors, uses a monitoring substation to control the operation of the gas pump, and uploads the data to the ground central station, realizing continuous monitoring and remote transmission of gases in the goaf.

[0004] However, while existing monitoring solutions have made some progress in terms of vibration resistance and system integration transmission, significant technical blind spots remain when facing the core environmental stress of continuous deposition of micron-sized coal dust underground, particularly in the field of high-sensitivity photoacoustic spectroscopy, where the problem of "implicit sensitivity attenuation induced by mechanical resonance loss of lock" is prevalent. Specifically, quartz tuning fork enhanced photoacoustic spectroscopy (QEPAS) technology heavily relies on the extremely high quality factor (Q value) and precise resonant frequency of the quartz tuning fork to amplify weak acoustic signals. After operating for a period of time in the high-dust environment of coal mines, micron-sized coal dust particles inevitably adhere to the vibrating arm of the quartz tuning fork. According to the physical model of the harmonic oscillator, the resonant frequency of the tuning fork is inversely proportional to the square root of the effective mass. This minute amount of dust adsorption will produce a mass loading effect, causing an irreversible drift in the physical resonant frequency of the tuning fork. Existing laser modulation systems are typically locked to a fixed center frequency set at the factory or perform scanning within an extremely narrow range. When dust load causes the physical resonance peak of the tuning fork to shift beyond the system's locked range, the system is actually exciting the tuning fork at the "resonance shoulder" or even in the non-resonance region of its amplitude-frequency response curve, resulting in an exponential decrease in acoustic gain. This sensitivity attenuation caused by the dynamic mismatch between the excitation frequency and the physical response frequency is easily misinterpreted as a decrease in gas concentration, causing the system to output a low concentration reading when the actual gas concentration exceeds the limit, resulting in a serious missed detection accident of "gas present but undetectable." Moreover, this physical mismatch cannot be repaired by simple circuit gain compensation. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a multi-parameter gas detection anti-interference and noise suppression system and method with full optical coordination in coal mines. By constructing a set of interference characteristic parameters including indicators such as frequency drift and optical window attenuation rate, an error sensitivity coefficient matrix is ​​established to quantify the noise variance and reliability weight of each detection channel in real time. Combined with frequency sweep tracking and multi-dimensional regression compensation techniques, proactive sensing and adaptive weighted fusion of the sensor's physical health status are achieved. This invention effectively solves the measurement distortion problem caused by physical damage such as dust mass load and optical path contamination, significantly improves the system's robustness under extreme operating conditions, and ensures the logical consistency and accuracy of gas monitoring data.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An all-optical detection array is deployed in the area to be detected in the underground coal mine. The array collects the original optical characteristic signals under the multi-dimensional physical field disturbance in the underground coal mine, constructs a set of interference characteristic parameters, and simultaneously constructs an environmental compensation vector. The all-optical detection array includes an NDIR infrared absorption detection module, a QEPAS photoacoustic resonance detection module, and an LDO fluorescence quenching detection module.

[0008] Based on the all-optical detection array, the detection channels are defined, the error sensitivity coefficient matrix is ​​constructed, and based on the error sensitivity coefficient matrix and the set of interference feature parameters, the real-time estimated noise variance of each detection channel is calculated. Based on the real-time estimated noise variance, the real-time confidence weight of each detection channel is calculated.

[0009] The original optical feature signals are inverted to output the original inverted concentrations of each target gas;

[0010] The original inversion concentrations of each target gas are processed based on the environmental compensation vector and real-time confidence weight, and the final fused concentrations of each target gas are output.

[0011] Based on the final fusion concentration of each target gas and the set of interference characteristic parameters, a composite decision is output.

[0012] The original optical characteristic signals include the intensity of pure infrared absorption signal, acoustic resonance signal, and fluorescence lifetime signal;

[0013] The method for acquiring the intensity of the pure infrared absorption signal includes: controlling the NDIR infrared absorption detection module to acquire the blackbody background radiation reference value when the infrared light source is off, acquiring the dual-wavelength light intensity signal when the infrared light source is on, separating the light intensity of the signal channel and the light intensity of the reference channel from the dual-wavelength light intensity signal, and subtracting the blackbody background radiation reference value from the light intensity of the signal channel to obtain the intensity of the pure infrared absorption signal.

[0014] The method for acquiring the acoustic resonance signal includes:

[0015] The frequency sweep range is set based on the factory-set center frequency of the QEPAS photoacoustic resonance detection module. Within the frequency sweep range, a rapid frequency sweep excitation is performed and the amplitude-frequency response curve of the quartz tuning fork is detected. The frequency corresponding to the maximum amplitude point is extracted from the amplitude-frequency response curve as the real-time physical resonance frequency. The excitation frequency is locked at the real-time physical resonance frequency to collect the acoustic resonance signal and the amplitude of the acoustic resonance signal.

[0016] The method for acquiring the fluorescence lifetime signal includes: controlling the LED light source of the LDO fluorescence quenching detection module to sinusoidally modulate and excite the oxygen-sensitive fluorescent membrane at a preset modulation frequency to generate an excitation light signal; the oxygen-sensitive fluorescent membrane emits a fluorescence signal after being excited by the excitation light signal; measuring the phase delay of the fluorescence signal relative to the excitation light signal to obtain the fluorescence phase difference; simultaneously measuring the reference fluorescence lifetime; and combining the fluorescence phase difference and the reference fluorescence lifetime to form a fluorescence lifetime signal.

[0017] The method for constructing the set of interference feature parameters includes:

[0018] The blackbody background offset is obtained by comparing the blackbody background radiation reference value with the factory-calibrated reference value.

[0019] The frequency drift is calculated by subtracting the real-time physical resonance frequency from the factory-set center frequency.

[0020] The current transmitted light intensity of the reference beam passing through the air chamber light window is measured, and the light window attenuation rate is obtained by calculating the ratio of the current transmitted light intensity to the reference value of the transmitted light intensity of the clean light window.

[0021] The frequency drift, optical window attenuation rate, and blackbody background offset are combined to form a set of interference characteristic parameters.

[0022] The method for defining the detection channel is as follows:

[0023] The NDIR infrared absorption detection module is the first detection channel, the QEPAS photoacoustic resonance detection module is the second detection channel, and the LDO fluorescence quenching detection module is the third detection channel.

[0024] The environmental compensation vector includes ambient temperature, relative humidity, and atmospheric pressure.

[0025] The method for constructing the error sensitivity coefficient matrix includes:

[0026] Frequency drift influence factor, optical decay influence factor and thermal radiation influence factor are respectively configured for the frequency drift, optical window attenuation rate and blackbody background offset in the interference feature parameter set to form an error sensitivity coefficient matrix;

[0027] The calculation method for the real-time estimated noise variance of each detection channel includes:

[0028] The frequency drift, optical window attenuation rate, and blackbody background shift are weighted and summed with their respective influence factors to calculate the noise variance increment.

[0029] The historical noise variance of each detection channel at the previous sampling time is obtained. The historical noise variance at the previous time is iteratively updated using the noise variance increment to obtain the real-time estimated noise variance of each detection channel at the current time.

[0030] The target gas includes methane, carbon monoxide, carbon dioxide, and oxygen;

[0031] The original inversion concentrations of each target gas include the infrared absorption inversion concentrations of methane and carbon dioxide, the photoacoustic inversion concentration of carbon monoxide, and the fluorescence inversion concentration of oxygen.

[0032] The method for obtaining the infrared absorption inversion concentrations of methane and carbon dioxide includes: substituting the light intensity of the signal channel and the light intensity of the reference channel into the Beer-Lambert law to perform absorbance inversion calculations to obtain the infrared absorption inversion concentrations of methane and carbon dioxide.

[0033] The method for obtaining the photoacoustic inversion concentration of carbon monoxide includes:

[0034] The acoustic gain function under the current state is calculated based on the real-time physical resonance frequency. The amplitude of the acoustic resonance signal is divided by the product of the acoustic gain function and the excitation light power to obtain the photoacoustic inversion concentration of carbon monoxide. The excitation light power is the emission power of the LED light source in the QEPAS photoacoustic resonance detection module.

[0035] The method for obtaining the fluorescence inversion concentration of oxygen includes:

[0036] The fluorescence lifetime is calculated and measured based on the fluorescence phase difference and the preset modulation frequency. The temperature-compensated reference fluorescence lifetime is obtained by using the ambient temperature in the environmental compensation vector to compensate for the temperature drift.

[0037] The fluorescence inversion concentration of oxygen was calculated based on the measured fluorescence lifetime and the temperature-compensated reference fluorescence lifetime.

[0038] The method for generating the composite decision includes:

[0039] Multiple safety thresholds are set for each target gas. The final fusion concentration of each target gas is compared with the corresponding multiple safety thresholds. The gas concentration alarm level is determined based on the comparison results.

[0040] A health threshold is set for each interference feature parameter in the interference feature parameter set. Each interference feature parameter is compared with its corresponding health threshold, and the health status level of each detection channel is determined based on the comparison results.

[0041] The gas concentration alarm level is logically combined with the health status level of each detection channel to generate a composite decision.

[0042] The method for processing the original inversion concentration of each target gas based on environmental compensation vector and real-time confidence weight includes:

[0043] The environmental compensation vector is used to perform environmental factor regression compensation on the original inversion concentration of each target gas to obtain the environmental compensation concentration. The environmental compensation concentration is then weighted and fused using real-time reliability weights to output the final fused concentration of each target gas.

[0044] A multi-parameter gas detection anti-interference and noise suppression system with full optical coordination in underground coal mines, used to implement the aforementioned multi-parameter gas detection anti-interference and noise suppression method with full optical coordination in underground coal mines, the system comprising:

[0045] Signal acquisition module: used to deploy a full optical detection array in the area to be detected in the underground coal mine, acquire the original optical feature signals under multi-dimensional physical field disturbance in the underground coal mine based on the full optical detection array, construct a set of interference feature parameters, and simultaneously construct an environmental compensation vector;

[0046] Credibility weight module: Define detection channels based on the all-optical detection array, construct an error sensitivity coefficient matrix, calculate the real-time estimated noise variance of each detection channel based on the error sensitivity coefficient matrix and the set of interference feature parameters, and calculate the real-time credibility weight of each detection channel based on the real-time estimated noise variance.

[0047] Concentration Inversion Module: Used to invert the original optical feature signals and output the original inverted concentrations of each target gas;

[0048] Environmental compensation fusion module: Based on the environmental compensation vector and real-time confidence weight, the original inversion concentration of each target gas is processed, and the final fused concentration of each target gas is output.

[0049] Composite decision module: Outputs composite decision based on the final fusion concentration of each target gas and the set of interference characteristic parameters.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] This invention establishes a dynamic mapping mechanism between the degree of physical interference and the data fusion weights by constructing a set of interference feature parameters containing multidimensional physical field disturbance information and calculating the real-time estimated noise variance and real-time reliability weights of each detection channel using an error sensitivity coefficient matrix. This mechanism can perceive and quantify the physical damage to sensor performance caused by the complex environment in coal mines (such as resonance loss due to dust mass load and signal-to-noise ratio reduction due to optical path contamination). At the data fusion level, it adaptively reduces the weights of observations from damaged or failed channels and uses environmental compensation vectors to perform multidimensional regression correction on the original inversion concentration. This effectively eliminates systematic measurement deviations caused by fluctuations in environmental temperature, humidity, and other factors, achieving high-reliability collaborative detection under heterogeneous detection arrays. It ensures that the system can still output logically consistent and accurate final fused concentrations and composite decisions when facing sensor performance degradation or instantaneous strong interference, thereby fundamentally avoiding safety hazards caused by the inability to distinguish between gas concentration changes and sensor performance degradation. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A flowchart illustrating the principle of the multi-parameter gas detection anti-interference and noise suppression method with all-optical coordination in coal mines provided in this embodiment of the invention;

[0054] Figure 2 This is a schematic diagram of the deployment of the all-optical detection array provided in an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the dual-wavelength infrared detection principle provided in an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the mass loading effect of a quartz tuning fork provided in an embodiment of the present invention;

[0057] Figure 5 This is a functional block diagram of the multi-parameter gas detection anti-interference and noise suppression system with all-optical coordination in underground coal mines provided in an embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1

[0060] Please see Figure 1 As shown, this embodiment provides a multi-parameter gas detection anti-interference and noise suppression method with all-optical coordination in underground coal mines, including:

[0061] Step S10: Deploy a full-optical detection array in the area to be detected in the coal mine, and construct a set of interference characteristic parameters based on the original optical characteristic signals collected by the full-optical detection array under the multi-dimensional physical field disturbance in the coal mine; simultaneously collect ambient temperature, relative humidity and atmospheric pressure to form an environmental compensation vector; the original optical characteristic signals include pure infrared absorption signal intensity, acoustic resonance signal and fluorescence lifetime signal;

[0062] Further, step S10 includes:

[0063] Step S11, see Figure 2 An all-optical detection array, comprising an NDIR infrared absorption detection module, a QEPAS photoacoustic resonance detection module, and an LDO fluorescence quenching detection module, is deployed in the underground coal mine area to be tested. Simultaneously, ambient temperature, relative humidity, and atmospheric pressure are collected to form an environmental compensation vector. The NDIR infrared absorption detection module is the first detection channel, the QEPAS photoacoustic resonance detection module is the second detection channel, and the LDO fluorescence quenching detection module is the third detection channel.

[0064] The underground environment of coal mines is characterized by complex interference factors such as high temperature, high humidity, coal dust deposition, airflow noise, mechanical vibration, and absorption of infrared blackbody background radiation in the tunnel. These factors prevent sensors based on a single detection principle from distinguishing between changes in gas concentration and the aliasing of environmental interference. In traditional laser scanning gas detection solutions, the laser and temperature control module are costly and consume a lot of power, the intrinsically safe isolation circuit is complex, and the life-cycle maintenance cost is high. Furthermore, they are insufficient in terms of multi-principle collaborative anti-interference. Step S10 deploys a fully optical detection array based on three different physical principles to achieve the separation and acquisition of original optical feature signals and interference feature parameters, providing a data foundation for subsequent health modeling and reliability-weighted fusion.

[0065] An all-optical detection array refers to a collection of detection systems that do not rely on a laser scanning temperature control system as the main sensing component. It includes an NDIR infrared absorption detection module, a QEPAS photoacoustic resonance detection module, and an LDO fluorescence quenching detection module, which detect gas concentration based on infrared absorption spectroscopy, photoacoustic resonance effect, and fluorescence quenching principle, respectively. NDIR is the abbreviation for non-dispersive infrared, and its detection principle is based on Beer-Lambert's law, which states that the degree to which gas molecules absorb infrared light of a specific wavelength is directly proportional to the gas concentration. Figure 2 As shown, this module contains an infrared light source, which, together with a detector, forms a detection path based on the principle of infrared absorption. QEPAS is an abbreviation for Quartz Tuning Fork Enhanced Photoacoustic Spectroscopy. It utilizes the piezoelectric effect of a quartz tuning fork to convert the acoustic signal generated by the absorption and modulation of light energy by gas molecules into an electrical signal. Through high-quality factor resonant amplification, it achieves high-sensitivity detection of trace gases, such as... Figure 2 As shown, this module uses a quartz tuning fork as the core sensing element to detect sound waves generated by the photoacoustic effect. LDO is an abbreviation for fluorescence lifetime detection, which is based on the Stern-Walmer fluorescence quenching model and inverses the oxygen concentration by measuring the fluorescence lifetime change of an oxygen-sensitive fluorescent membrane, such as... Figure 2 As shown, the module is equipped with an oxygen-sensitive fluorescent membrane for excitation and generation of fluorescence signals. The detection array is constructed using three detection modules based on different physical principles, resulting in varying response characteristics of each detection channel to environmental interference. When one detection channel fails due to specific interference, other detection channels can still provide valid data. This complementarity creates conditions for subsequent reliability-weighted fusion.

[0066] Multidimensional physical field disturbances refer to the superimposed effects of multiple physical interference factors simultaneously existing in the underground coal mine environment on sensor measurement results. These disturbance factors include: coal dust particles adhering to... Figure 2The mass loading effect on the quartz tuning fork arm of the second detection channel causes resonant frequency drift; coal dust deposition on the optical window surface leads to decreased light transmittance; infrared radiation generated by the heated coal wall and coal dust creates blackbody background interference on the detector; and changes in ambient temperature and humidity alter the fluorescence quenching efficiency. The original optical characteristic signal refers to the signal directly characterizing gas concentration acquired by the all-optical detection array, including the intensity of pure infrared absorption signal, acoustic resonance signal, and fluorescence lifetime signal. The interference characteristic parameter set refers to the set of quantitative indicators characterizing the degree of physical damage to the sensor, including frequency drift, optical window attenuation rate, and blackbody background shift. Separating the acquisition of the original optical characteristic signal and the interference characteristic parameters allows subsequent processing to distinguish between "gas concentration change" and "sensor performance degradation," avoiding misjudging sensor failure as abnormal gas concentration.

[0067] The NDIR infrared absorption detection module is set as the first detection channel, the QEPAS photoacoustic resonance detection module as the second detection channel, and the LDO fluorescence quenching detection module as the third detection channel. This channel numbering mechanism provides a unified index for the noise variance matrix calculation in step S20 and the weighted fusion operation in step S40. The environmental compensation vector consists of ambient temperature, relative humidity, and atmospheric pressure. Ambient temperature affects the infrared absorption coefficient, fluorescence quenching efficiency, and resonance characteristics of the quartz tuning fork; relative humidity affects the infrared absorption cross-section of gas molecules; and atmospheric pressure affects gas density and sound wave propagation characteristics. Environmental parameters and original optical feature signals are acquired synchronously to ensure that the environmental compensation vector and the signal to be compensated strictly correspond in the time dimension, avoiding compensation distortion due to acquisition time differences.

[0068] Step S12: Control the NDIR infrared absorption detection module to collect the blackbody background radiation reference value when the infrared light source is off, and to collect the dual-wavelength light intensity signal when the infrared light source is on. Separate the signal channel light intensity and the reference channel light intensity from the dual-wavelength light intensity signal. Subtract the blackbody background radiation reference value from the signal channel light intensity to obtain the pure infrared absorption signal intensity. Calculate the ratio between the blackbody background radiation reference value and the factory-calibrated reference reference value to obtain the blackbody background offset.

[0069] In step S12, the NDIR infrared absorption detection module uses time-division multiplexing logic to separate and acquire infrared spectral absorption signals from blackbody background radiation. Time-division multiplexing refers to acquiring different types of signals at different time periods, avoiding signal aliasing through temporal separation. The infrared light source is off when the system controls the infrared LED or filament to be completely powered off; in this state, the detector only receives thermal radiation from the external environment. Blackbody background radiation refers to the infrared electromagnetic radiation emitted by coal walls, coal dust particles, and other heat sources in coal mine tunnels due to their temperature being above absolute zero. According to Planck's law of blackbody radiation, any object with a temperature above absolute zero will emit infrared radiation. In underground coal mines, the temperature of the tunnel walls can reach tens of degrees Celsius, and the resulting blackbody radiation falls precisely within the response band of the infrared detector, creating an interference background for gas detection based on the infrared absorption principle. After the system controls the infrared light source to enter the off state, it acquires the ambient thermal radiation signal through a pyroelectric detector or thermopile detector, defines this signal as the blackbody background radiation reference value, and records it as V. dark A pyroelectric detector is a sensor sensitive to infrared radiation. Its working principle is based on the pyroelectric effect, where certain crystalline materials generate surface charges when their temperature changes. When ambient thermal radiation is incident on the sensitive element of the pyroelectric detector, the element's temperature changes, generating an electrical signal proportional to the incident radiation power. Acquiring a blackbody background radiation baseline is to obtain a signal baseline purely generated by ambient thermal radiation. This baseline does not contain any useful signal components generated by infrared light sources and can be used as a subtraction factor in subsequent signal purification processing.

[0070] See Figure 3 After the system controls the infrared light source to turn on, it uses a dual-wavelength detection architecture to acquire signals. Figure 3 The diagram illustrates that the infrared light source emits a beam containing a signal wavelength λ1 and a reference wavelength λ2. This beam enters the gas chamber and contacts the target gas within. After transmission through the gas chamber, the light intensities of the signal channel and the reference channel are output separately. The dual-wavelength detection architecture refers to the system simultaneously using two optical channels with different center wavelengths for detection. One wavelength channel is located at the characteristic absorption peak of the target gas, while the other wavelength channel is located at a reference position where the target gas has no or weak absorption. The signal wavelength is denoted as λ1, and its center wavelength strictly corresponds to the infrared absorption peak of the target gas. For example, for methane detection, the signal wavelength λ1 corresponds to the strong absorption peak of methane molecules in the mid-infrared band; for carbon dioxide detection, the signal wavelength λ1 corresponds to the asymmetric stretching vibration absorption peak of carbon dioxide molecules. The reference wavelength is denoted as λ2, and its selection principle is to avoid the absorption peaks of all analyte gases and common interfering gases, so that changes in the light intensity of the reference channel only reflect the influence of non-gaseous factors such as light source aging and window contamination.

[0071] The light intensity of the signal channel is denoted as I. sig(λ1), which physically represents the infrared light intensity at the signal wavelength after passing through the gas chamber, is affected by both gas absorption and environmental interference. The reference channel light intensity is denoted as I. ref (λ2), its physical meaning is the infrared light intensity at the reference wavelength after passing through the gas chamber. Since there is no absorption in the gas at the reference wavelength, this light intensity only reflects the changes in the light source intensity and the transmittance of the light path. The pure infrared absorption signal intensity (SIR) is obtained by subtracting the blackbody background radiation reference value from the signal channel light intensity. Its calculation formula is: SIR = I sig (λ1)-V dark The physical meaning of this formula is that by subtracting the ambient thermal radiation background from the signal acquired in the on state, the effective signal component generated solely by the infrared light source is obtained. The subtraction operation in the formula is based on the principle of linear superposition of signals, meaning that the detector's output signal equals the sum of the signals generated by all incident radiation sources. When the infrared light source is on, the total radiation received by the detector is the superposition of the light source radiation and the ambient thermal radiation; when the infrared light source is off, the detector only receives ambient thermal radiation. Subtracting the two allows the separation of the signal component generated by the light source radiation. This combination of time-division acquisition and differential processing can effectively suppress the interference of blackbody background radiation on infrared absorption detection without increasing hardware complexity, making the intensity of the pure infrared absorption signal more accurately reflect the absorption characteristics of the gas.

[0072] Black body background offset Tbb bias The calculation uses a ratio between the blackbody background radiation reference value and the factory-calibrated reference value. The calculation formula is: Tbb bias =(V dark / V ref,baseline )×100%, where V ref,baseline This is a reference value collected under standard environmental conditions during factory calibration. This value reflects the sensor's background thermal radiation level in a clean, normal temperature environment. V dark With V ref,baseline The ratio represents the degree of deviation of the current environmental thermal radiation from the standard environment. Multiplying it by 100% converts the ratio to a percentage for easier subsequent threshold determination. Blackbody background offset Tbb biasThe physical meaning lies in quantifying the increment of the current environmental thermal radiation interference relative to the factory baseline. A larger increment indicates more severe environmental thermal radiation interference and lower reliability of the sensor measurement results. Including the blackbody background offset as one of the interference characteristic parameters in the interference characteristic parameter set provides input parameters for the noise variance update in step S20, enabling the system to dynamically adjust the reliability weight of the detection channel according to changes in environmental thermal radiation. Using time-division multiplexing logic for signal acquisition separates the infrared absorption signal from the blackbody background radiation in the time dimension, avoiding the complex filtering operations that might be introduced by signal separation in the frequency or spatial domain. The adoption of a dual-wavelength detection architecture allows common-mode interference such as light source aging and window contamination to be compensated through the reference channel, improving the long-term stability of infrared absorption concentration inversion. The introduction of the blackbody background offset transforms the previously difficult-to-quantify environmental thermal radiation interference into a measurable numerical indicator, providing a quantitative basis for subsequent error budgeting and health diagnosis.

[0073] Step S13: Set the sweep frequency range based on the factory-set center frequency of the QEPAS photoacoustic resonance detection module, perform rapid sweep frequency excitation within the sweep frequency range, and detect the amplitude-frequency response curve of the quartz tuning fork using second-order harmonic phase-locked loop detection technology. Extract the frequency corresponding to the maximum amplitude point from the amplitude-frequency response curve as the real-time physical resonance frequency.

[0074] Step S14: Calculate the frequency drift by performing a difference calculation between the real-time physical resonance frequency and the factory-set center frequency, and lock the excitation frequency at the real-time physical resonance frequency to collect the acoustic resonance signal and the amplitude of the acoustic resonance signal.

[0075] Steps S13 and S14 implement the physical resonance frequency tracking and mass load characterization of the quartz tuning fork based on frequency sweep excitation in the QEPAS photoacoustic resonance detection module. A quartz tuning fork is a high-precision resonant device made using the piezoelectric effect of quartz crystal. Its shape resembles a tuning fork structure, with two symmetrical vibrating arms. When external sound pressure is applied to the vibrating arms, they vibrate and generate an electrical signal proportional to the vibration amplitude under the piezoelectric effect. The core performance indicators of a quartz tuning fork include the resonant frequency and the quality factor. The resonant frequency refers to the vibration frequency at which the tuning fork resonates, at which the vibration amplitude reaches its peak. The quality factor is a dimensionless parameter characterizing the energy loss characteristics of the resonant system. A higher quality factor indicates lower system energy loss, a sharper resonance peak, and stronger amplification capability for weak signals.

[0076] See Figure 4 The mass loading effect refers to the phenomenon that when external material is attached to the surface of a resonator, the effective mass of the resonator increases, resulting in a change in the resonant frequency. Figure 4 This illustrates the process of a quartz tuning fork changing from a clean state to a dust-laden state. For example... Figure 4As shown on the left, in a clean state, the quartz tuning fork has no foreign matter on its surface, and its vibration frequency remains at the factory-set center frequency. According to the physical model of the harmonic oscillator, the resonant frequency of the tuning fork is inversely proportional to the square root of its effective mass, i.e.: f∝1 / √m, where f is the resonant frequency and m is the effective mass. Figure 4 As shown on the right, when coal dust particles adhere to the vibrating arm of a quartz tuning fork, the system enters a dust-attached state, increasing the effective mass of the vibrating arm. According to the aforementioned inverse relationship, the resonant frequency will decrease. This causes the resonant frequency of the quartz tuning fork to drift from the factory-set center frequency to the real-time physical resonant frequency. This frequency change is directly related to the amount of dust attached; the more dust attached, the more significant the frequency decrease. After operating in a coal mine environment for several months, micron-sized coal dust particles (such as...) Figure 4 The continuous deposition of dust (as shown by the black dot) can cause the resonant frequency to drift by several hertz or even more. Since the resonant peak width (half-maximum width at half maximum) of a quartz tuning fork is only on the order of several hertz, this means that even a small frequency drift can cause the excitation frequency to deviate from the center of the resonant peak, resulting in a significant gain reduction. Traditional laser modulation systems are typically locked to a fixed center frequency set at the factory or perform only a small-range scan. When dust causes the physical resonant peak of the tuning fork to shift beyond the locked range of the laser modulation, the system is actually exciting the tuning fork at the resonant shoulder or even in the non-resonant region. The resonant shoulder refers to the region where the gain drops rapidly on both sides of the resonant peak. When excited in this region, the amplification factor of the acoustic signal decreases exponentially compared to the center of the resonant peak, resulting in a significant reduction in the sensor's sensitivity to the gas. This sensitivity reduction cannot be completely recovered by simply adjusting the gain coefficient, because the essence of the problem lies in the dynamic mismatch between the excitation frequency and the physical response frequency, rather than insufficient gain in the signal amplification circuit.

[0077] Step S13 uses the factory-set center frequency as a reference to set the sweep frequency range, thus resolving the dynamic mismatch between the excitation frequency and the physical response frequency. The factory-set center frequency is denoted as f. center This value is calibrated using a precision frequency measuring instrument before the product leaves the factory, representing the frequency of the quartz tuning fork. Figure 4 The physical resonant frequency under clean conditions is shown. The sweep range is denoted as Δf, and its setting is based on the following: the sweep range should be greater than the maximum frequency drift that the quartz tuning fork may experience throughout its entire service life, and should be less than the frequency interval between adjacent resonant modes to avoid mode confusion. For example, the sweep range can be set to plus or minus tens of hertz of the factory-set center frequency. The system at f... center -Δf to f center The system performs a fast frequency sweep excitation within the frequency range of +Δf. The frequency sweep excitation refers to the system changing the excitation frequency point by point according to the preset frequency step and collecting the response signal of the quartz tuning fork at each frequency point.

[0078] Second-harmonic phase-locked loop (HLL) detection is a highly sensitive method for detecting weak signals, based on modulation and demodulation techniques. In a QEPAS system, the intensity of the excitation light source is modulated at a certain frequency. The photoacoustic signal generated by gas molecules absorbing the modulated light also carries modulation information of the same frequency. The lock-in amplifier correlates the piezoelectric signal output from the quartz tuning fork with the modulation signal, extracting the signal component in phase with the modulation frequency, effectively suppressing noise unrelated to the modulation frequency. The second harmonic refers to the signal component at twice the modulation frequency. Second-harmonic detection has a higher signal-to-noise ratio than fundamental frequency detection because the contribution of noise sources such as light source intensity fluctuations and optical path interference is significantly reduced at the second harmonic frequency. The system uses HLL to detect the amplitude-frequency response curve of the quartz tuning fork, which is the relationship between the amplitude of the quartz tuning fork output signal and the excitation frequency. Due to the high quality factor of the quartz tuning fork, its amplitude-frequency response curve exhibits a sharp peak, and the frequency corresponding to the peak is the real-time physical resonant frequency of the quartz tuning fork. The system extracts the frequency corresponding to the maximum amplitude point from the amplitude-frequency response curve as the real-time physical resonant frequency, denoted as f. peak The algorithm for extracting the point with the maximum amplitude can be the peak search algorithm. This algorithm traverses the amplitude of all sampling points within the frequency sweep range and finds the sampling point with the maximum amplitude and its corresponding excitation frequency.

[0079] Step S14 will set the real-time physical resonance frequency f peak With factory-set center frequency f center The frequency drift f is obtained by performing interpolation. drift Its calculation formula is: f drift =|f peak -f center The absolute value operation in the formula ensures that the frequency drift is non-negative because the impact on sensor performance is the same regardless of whether the frequency drifts towards higher or lower frequencies. Frequency drift f drift The influence of dust mass load on the resonant characteristics of the quartz tuning fork is directly quantified. A larger value indicates more severe dust adhesion and a more significant attenuation of the sensor's resonant gain. The frequency drift is included as one of the interference characteristic parameters in the interference characteristic parameter set, providing input parameters for the noise variance update in step S20, enabling the system to dynamically adjust the confidence weight of the QEPAS detection channel according to the degree of dust adhesion.

[0080] In obtaining the real-time physical resonance frequency f peakThen, the excitation frequency is locked at this frequency for signal acquisition. Excitation frequency locking means that the system precisely sets the modulation frequency of the light source to half of the real-time physical resonant frequency, so that the frequency of the generated photoacoustic signal falls exactly at the center of the resonant peak of the quartz tuning fork. In the excitation frequency locked state, the system acquires the piezoelectric signal output by the quartz tuning fork, and the amplitude of this signal is defined as the acoustic resonant signal amplitude A. meas This signal is defined as the acoustic resonant signal SPA. The amplitude A of the acoustic resonant signal is... meas The carbon monoxide concentration is directly proportional to the gas concentration and serves as the raw data for carbon monoxide concentration inversion in subsequent step S30. A combination of dynamic frequency sweep tracking and frequency locking is employed to ensure the system always operates at the center of the resonance peak of the current physical state, avoiding gain attenuation due to frequency mismatch. Frequency sweep excitation is performed in each sampling cycle, ensuring the system can track the gradual frequency changes caused by continuous dust deposition in real time. The calculation of frequency drift transforms the abstract degree of dust adhesion into a quantifiable numerical indicator, providing a basis for health diagnosis and enabling the system to distinguish between "low gas concentration" and "resonance lockout," avoiding the risk of missed detections due to resonance lockout.

[0081] Step S15: Control the LED light source of the LDO fluorescence quenching detection module to sinusoidally modulate and excite the oxygen-sensitive fluorescent film at a preset modulation frequency to generate an excitation light signal. After the oxygen-sensitive fluorescent film is excited by the excitation light signal, it emits a fluorescence signal. The phase delay of the fluorescence signal relative to the excitation light signal is measured to obtain the fluorescence phase difference. The reference fluorescence lifetime is measured simultaneously. The fluorescence phase difference and the reference fluorescence lifetime are combined to form a fluorescence lifetime signal.

[0082] Step S15 involves acquiring the oxygen detection signal for the LDO fluorescence quenching detection module. Fluorescence quenching refers to the phenomenon where fluorescent molecules, in their excited state, undergo energy or charge transfer with quencher molecules, leading to a decrease in fluorescence intensity or a shortened fluorescence lifetime. In oxygen detection applications, oxygen molecules act as quenchers, interacting with excited-state fluorescent molecules through a collisional quenching mechanism, causing the fluorescence lifetime to shorten with increasing oxygen concentration. The oxygen-sensitive fluorescent membrane is a thin film made by immobilizing fluorescent dye molecules in an oxygen-permeable polymer matrix. When oxygen molecules diffuse into the polymer matrix and come into contact with the fluorescent dye molecules, a collisional quenching reaction occurs. Fluorescence lifetime detection has higher stability than fluorescence intensity detection because fluorescence lifetime is an intrinsic property of fluorescent molecules and is not affected by factors such as fluctuations in light source intensity, optical path loss, and changes in detector sensitivity.

[0083] The system controls the LED light source of the LDO fluorescence quenching detection module to operate at a preset modulation frequency f. mod The oxygen-sensitive fluorescent film was excited by sinusoidal modulation. Sinusoidal modulation refers to the periodic variation of the luminous intensity of the LED light source with time in the form of a sine wave, with a modulation frequency f. modThe period of the sine wave is determined. The selection of the preset modulation frequency must meet the following conditions: the modulation frequency should be on the same order of magnitude as the reciprocal of the fluorescence lifetime to ensure sufficient resolution for phase delay measurement; the modulation frequency should avoid interference frequencies that may exist in the environment. For example, the modulation frequency can be set to a fixed value within the range of 1~50 kHz. After being excited by an excitation light signal, the oxygen-sensitive fluorescent membrane emits a fluorescence signal. Due to the time delay in the fluorescence emission process, the fluorescence signal has a phase lag relative to the excitation light signal. The magnitude of the phase lag is directly related to the fluorescence lifetime; the longer the fluorescence lifetime, the greater the phase lag. The system measures the phase lag of the fluorescence signal relative to the excitation light signal through a phase detection circuit, defining this phase lag as the fluorescence phase difference ϕ. meas The phase detection circuit works on the principle of quadrature demodulation technology, which performs correlation operations on the in-phase and quadrature components of the fluorescence signal and the excitation signal respectively, and obtains the phase difference through arctangent operation.

[0084] Reference fluorescence lifetime τ ref Measurements are performed through an independent reference channel. The reference channel employs a sealed, standard oxygen concentration environment, ensuring the internal fluorescent membrane remains unaffected by the analyte gas and its fluorescence lifetime remains constant. The purpose of the reference channel is to provide a lifetime baseline unaffected by changes in the analyte gas concentration, used in subsequent ratio calculations to eliminate common-mode interference such as fluorescent dye aging and variations in light source intensity. The reference fluorescence lifetime τ... ref Phase difference ϕ with fluorescence meas The combination constitutes the fluorescence lifetime signal SF: SF=(ϕ meas ,τ ref The fluorescence lifetime signal SF is a binary tuple containing all the raw data required for oxygen concentration inversion. Phase-based lifetime measurement offers higher robustness to interference compared to time-domain decay measurement because phase measurement is independent of the absolute amplitude of the signal and insensitive to fluctuations in light source intensity and optical path loss. The introduction of a reference channel enables dual lifetime ratio measurement, further eliminating interference from long-term, slowly changing factors such as fluorescent dye aging and temperature drift.

[0085] Step S16: Measure the current transmitted light intensity of the reference beam passing through the air chamber light window using an independently set light window attenuation sensor; calculate the ratio between the current transmitted light intensity and the baseline value of the transmitted light intensity of the clean light window to obtain the light window attenuation rate; combine the frequency drift, light window attenuation rate, and blackbody background offset to form a set of interference characteristic parameters.

[0086] Step S16 synchronizes the acquisition of optical window attenuation and environmental parameters. The optical window is a transparent insulating element between the gas chamber and the external environment, allowing infrared or visible light to penetrate into the gas chamber for gas detection. A clean optical window refers to the initial ideal state when the gas chamber optical window is uncontaminated (i.e., free from coal dust deposition or obstruction). In coal mine environments, coal dust particles deposit on the optical window surface, forming an obstruction layer and reducing the window's transmittance. The impact mechanism of optical window contamination on each detection channel differs: for NDIR detection, optical window contamination reduces the intensity of infrared light incident on the gas chamber; for QEPAS detection, optical window contamination reduces the excitation light power, indirectly affecting the photoacoustic signal intensity; for LDO detection, optical window contamination reduces the excitation light intensity. Traditional detection systems treat optical window contamination as a purely interfering factor, unable to quantify or compensate for it.

[0087] The system measures the degree of contamination in the optical window using an independently configured optical window attenuation sensor. The sensor consists of an independent light source and detector pair; its emitted reference beam passes through the gas chamber optical window but does not participate in gas detection. The wavelength of the reference beam is selected to avoid the absorption peaks of all the gases being measured, ensuring that changes in beam intensity reflect only changes in the transmittance of the optical window, not gas absorption. The system measures the current transmitted light intensity I after the reference beam passes through the optical window. trans It is then compared with the pre-calibrated cleanroom light window transmitted light intensity reference value I0 to calculate the light window attenuation rate I. atten :I atten =(1-I trans / I0)×100%, I in the formula trans / I0 represents the ratio of the current light window transmittance to the clean light window transmittance. The closer this ratio is to 1, the cleaner the light window; the closer it is to 0, the more severe the light window contamination. Subtracting this ratio from 1 yields the light window attenuation rate, therefore the light window attenuation rate I0. atten The closer to 0, the cleaner the light window; the closer to 100%, the more severe the light window contamination. atten A value of zero indicates that the light window is completely clean. atten A value of 100% indicates that the light window is completely blocked. The light window attenuation rate is included as one of the interference feature parameters in the interference feature parameter set, providing input parameters for the noise variance update in step S20, enabling the system to dynamically adjust the confidence weight of each detection channel according to the degree of light window contamination.

[0088] Ambient temperature T, relative humidity RH, and atmospheric pressure P are synchronously collected by a temperature and humidity sensor module, forming an environmental compensation vector [T,RH,P]. Ambient temperature affects the infrared absorption coefficient, fluorescence quenching constant, and resonant frequency of the quartz tuning fork; relative humidity affects the infrared absorption cross-section of gas molecules and the oxygen permeability of the fluorescent membrane; atmospheric pressure affects gas density and the acoustic wave propagation characteristics of photoacoustic signals. The environmental compensation vector provides input parameters for concentration inversion in step S30 and environmental regression compensation in step S40. The frequency drift f is... drift , Light window attenuation rate I atten and bold background offset Tbb bias The interference feature parameters are combined to form a set. The construction of this set integrates the interference information, originally scattered across various detection modules, into a unified data structure, facilitating centralized processing in step S20. The three interference feature parameters characterize the impact of three main types of interference factors in the coal mine environment on the sensor: frequency drift characterizes the impact of dust mass load effect on the QEPAS detection channel, window attenuation rate characterizes the impact of dust deposition on the optical detection channel, and blackbody background offset characterizes the impact of environmental thermal radiation on the infrared absorption detection channel. This parallel acquisition mechanism of multi-dimensional interference parameters enables the system to evaluate the sensor's operating status from multiple perspectives, providing comprehensive input data for subsequent health modeling and confidence weighting.

[0089] Step S10 transforms interference factors (dust mass load, window contamination, and environmental thermal radiation) that are often overlooked in traditional detection systems into measurable and quantifiable numerical parameters. This transformation turns interference factors from purely negative influences into positive information that can be used for health diagnosis. By incorporating interference characteristic parameters into the system's feedback loop, step S10 achieves a shift from passively responding to interference to actively sensing it, creating conditions for subsequent closed-loop compensation and health diagnosis. By deploying all-optical detection arrays based on three different physical principles, parallel detection capabilities for four target gases—methane, carbon monoxide, carbon dioxide, and oxygen—in coal mines are achieved. Through signal acquisition techniques such as time-division multiplexing logic, frequency sweep excitation tracking, and phase-based lifetime measurement, effective separation of the original optical characteristic signals from the interference background is achieved. By constructing a set of interference characteristic parameters including frequency drift, window attenuation rate, and blackbody background offset, a multi-dimensional quantitative characterization of the sensor's health status is realized. The parallel deployment of multiple detection channels based on different physical principles allows each channel to complement the response characteristics to various interference factors. When a detection channel experiences measurement deviation due to specific interference, other detection channels can still provide valid data. This redundancy design improves the system's reliability under extreme conditions. Effective separation of the original optical characteristic signal from the interference background enables the concentration inversion in step S30 to be calculated based on a clean signal, reducing the systematic bias of the inversion results and improving the accuracy of gas concentration measurement. Quantitative characterization of interference characteristic parameters allows step S20 to perform health assessment based on specific numerical values ​​rather than vague qualitative judgments, providing a mathematical basis for the accurate calculation of confidence weights. The dynamic frequency sweep tracking mechanism implemented in step S10 addresses the resonance lock-up problem faced by quartz tuning fork enhanced photoacoustic spectroscopy in high-dust environments. By tracking the physical resonance frequency in real time and locking the excitation frequency at the center of the resonance peak, the system always operates in the resonance gain region, avoiding sensitivity attenuation caused by the mismatch between the excitation frequency and the physical response frequency. Real-time acquisition of frequency drift allows the system to distinguish whether the decrease in sensitivity stems from a decrease in gas concentration or resonance lock-up, eliminating the risk of missed detections due to resonance lock-up.

[0090] Step S20: Construct the error sensitivity coefficient matrix. Based on the error sensitivity coefficient matrix and the set of interference feature parameters, calculate the real-time estimated noise variance of each detection channel. Calculate the real-time confidence weight of each detection channel based on the real-time estimated noise variance.

[0091] Specifically, step S20 transforms three types of physical interference parameters—frequency drift, optical window attenuation rate, and blackbody background shift—into mathematical quantitative indicators characterizing the measurement reliability of each detection channel. Traditional gas detection systems cannot self-assess the reliability of their output data when the environment deteriorates. When sensor performance degrades due to dust adhesion or optical window contamination, the system still outputs measurement results with fixed weights, failing to distinguish whether the measurement deviation stems from changes in gas concentration or sensor performance degradation, leading to missed detections or false alarms. Step S20 establishes a mathematical mapping relationship between interference parameters and noise variance, enabling dynamic assessment of the health status of each detection channel and providing a weighting basis for the subsequent credibility-weighted fusion in step S40. The three interference feature parameters in the interference feature parameter set characterize the physical damage state of the complex underground coal mine environment to the all-optical detection array from three dimensions: mass load, optical path obstruction, and thermal radiation interference, constituting the input data for sensor health assessment.

[0092] Further, step S20 includes:

[0093] Step S21: Configure frequency drift influence factor, optical attenuation influence factor and thermal radiation influence factor for frequency drift, optical window attenuation rate and blackbody background offset in the interference feature parameter set, respectively, to form an error sensitivity coefficient matrix.

[0094] Step S21 configures the corresponding influence factors for the three types of interference characteristic parameters and establishes the error sensitivity coefficient matrix. The influence factor is a dimensionless coefficient characterizing the contribution of a specific interference parameter to the noise variance of the detection channel; its value reflects the coupling strength between the interference parameter and the measurement error. The frequency drift influence factor, denoted as β, is used to quantify the frequency drift f. drift The influence weight of noise variance; the light decay influence factor, denoted as γ, is used to quantify the optical window attenuation rate I. atten The influence weight of noise variance; the thermal radiation influence factor, denoted as δ, is used to quantify the blackbody background offset Tbb. bias The weights of the influence on noise variance. The three influencing factors together constitute the error sensitivity coefficient matrix, which describes the mapping relationship from the interference feature parameter space to the noise variance space.

[0095] The determination of each influencing factor in the error sensitivity coefficient matrix requires a combination of physical mechanism analysis and calibration experiments. The frequency drift influencing factor β is determined based on the following: According to the harmonic oscillator physical model, the acoustic gain of a quartz tuning fork is related to the degree to which the excitation frequency deviates from the resonant frequency via a Lorentz function. The greater the deviation, the more significant the gain attenuation, the more severe the signal amplitude fluctuation, and the greater the increase in noise variance. During the factory calibration phase, different degrees of frequency offset are artificially applied to the system in a clean environment, and the corresponding changes in signal noise variance are recorded. The slope coefficient between the frequency offset and the increase in noise variance is obtained through linear regression fitting; this slope is the frequency drift influencing factor β. For example, if every 1 Hz increase in frequency drift leads to a 0.02 unit increase in noise variance, then β is set to 0.02. The optical attenuation influencing factor γ is determined based on the following: A decrease in the transmittance of the optical window leads to a decrease in the incident light power to the detector. According to the signal-to-noise ratio (SNR) theory of photodetectors, the SNR is proportional to the square root of the incident light power. A decrease in light power will lead to a decrease in the SNR and an increase in noise variance. During the calibration phase, the system used standard attenuators with different transmittances to simulate optical window contamination, recording the corresponding changes in noise variance, and obtaining the γ value through regression analysis. The determination of the thermal radiation influence factor δ is based on the following: an increase in blackbody background offset indicates intensified environmental thermal radiation interference, leading to an increase in background noise components in the detector output signal and thus increasing the uncertainty of the signal net value. During the calibration phase, the system measured the correspondence between blackbody background offset and noise variance under different temperature environments, and obtained the δ value through regression analysis.

[0096] The construction of the error sensitivity coefficient matrix enables interference parameters with different physical properties to be uniformly mapped to the mathematical space of noise variance increment, achieving normalization of heterogeneous interference information. The independent setting of the frequency drift influence factor β, optical attenuation influence factor γ, and thermal radiation influence factor δ allows the system to differentiate weighting based on the actual impact of various interferences, avoiding evaluation biases that might result from treating all interferences equally. The influence factors, obtained through factory calibration, are stored in the system's non-volatile memory and remain stable throughout the entire usage cycle, requiring no on-site recalibration. Simultaneously, the influence factors can also be adaptively updated through an online learning mechanism: the system continuously collects corresponding data of interference parameters and measurement errors, using recursive least squares or Kalman filtering algorithms to correct the influence factors online, enabling the error sensitivity coefficient matrix to adapt to the slow drift of sensor performance parameters during long-term use.

[0097] Step S22: The frequency drift, optical window attenuation rate and blackbody background offset are weighted and summed with the corresponding influencing factors to calculate the noise variance increment; the historical noise variance of each detection channel at the previous sampling time is obtained, and the historical noise variance at the previous time is iteratively updated using the noise variance increment to obtain the real-time estimated noise variance of each detection channel at the current time.

[0098] Step S22 calculates the noise variance increment using interference characteristic parameters and influencing factors, and iteratively updates the historical noise variance. Noise variance is a statistical indicator characterizing the dispersion of a random variable. In the field of gas detection, noise variance reflects the volatility and instability of the output signal of the detection channel. A larger noise variance indicates a greater random error in the measurement results and lower reliability. Historical noise variance refers to the estimated noise variance of the i-th detection channel at the previous sampling time, denoted as σ. i old Its initial value is determined by factory calibration and characterizes the background noise level of the sensor in a clean environment. The subscript i ranges from 1 to 3, where i = 1 corresponds to the NDIR infrared absorption detection module, i = 2 corresponds to the QEPAS photoacoustic resonance detection module, and i = 3 corresponds to the LDO fluorescence quenching detection module.

[0099] The real-time noise variance estimation update employs an iterative formula based on the Kalman filter concept, combining historical noise variance with the interference characteristic parameters at the current moment to calculate the real-time estimated noise variance σ for each detection channel at the current moment. i new The updated formula is: σ i new =σ i old +β×|f drift,i |+γ×|I atten,i |+δ×|Tbb bias,i |, the first term on the right side of the formula σ i old The historical noise variance reflects the error level of the detection channel at the previous moment and serves as the basis for estimating the noise variance at the current moment. The second term β×|f drift,i | represents the noise variance increment caused by frequency drift, where f drift,i Let f be the frequency drift corresponding to the i-th detection channel. For the QEPAS photoacoustic resonance detection module, the frequency drift directly characterizes the mass load of the quartz tuning fork; for the NDIR infrared absorption detection module and the LDO fluorescence quenching detection module, the frequency drift does not directly affect their working principle, and the corresponding f can be used as a reference. drift,i Set to zero or use a smaller weight. Absolute value calculation ensures that the frequency drift, whether shifted towards higher or lower frequencies, produces a positive increase in noise variance, consistent with physical reality. The third term γ×|I atten,i | represents the noise variance increment caused by the attenuation of the optical window, where I atten,i Let be the optical window attenuation rate corresponding to the i-th detection channel. Optical window attenuation affects all three detection channels because all three detection modules rely on optical signals for gas detection. The fourth term δ×|Tbb bias,i | represents the noise variance increment caused by the blackbody background shift, where Tbb bias,iThis represents the blackbody background offset corresponding to the i-th detection channel. This mainly affects the NDIR infrared absorption detection module because the band of blackbody radiation overlaps with the operating band of infrared absorption detection.

[0100] The formula uses an additive approach to sum the historical noise variance with the increments caused by the three types of interference. This is because the effects of each type of interference on measurement error are cumulative and independent. Frequency drift caused by dust adhesion, light attenuation caused by window contamination, and background interference caused by environmental heat radiation coexist and are independent of each other; their contributions to the noise variance can be linearly added together. The iterative form of the formula gives the noise variance estimation a temporal continuity; as environmental interference continues to intensify, σ... i new It will gradually increase; when environmental interference decreases or the sensor is maintained and cleaned, by increasing σ i old Resetting to the initial calibration value allows the system to recover to a normal noise variance level. This state update mechanism based on the Kalman filter concept ensures that the noise variance estimation reflects both the current disturbance state and maintains continuity with historical states, avoiding the drastic fluctuations that might result from estimating solely based on current sampled data.

[0101] Step S23: Perform inverse normalization on the real-time estimated noise variance of each detection channel to obtain the real-time confidence weight of each detection channel.

[0102] Specifically, step S23 converts the real-time estimated noise variance output in step S22 into weighting coefficients for subsequent data fusion. The real-time reliability weight is a numerical indicator representing the proportion of the detection channel's output data in the fusion calculation; a larger weight indicates higher reliability of the channel's data and a greater contribution to the fusion result. The physical meaning of the reciprocal normalization operation is that detection channels with larger noise variances have higher uncertainty in their measurement results and should be assigned lower weights in the fusion; detection channels with smaller noise variances have more reliable measurement results and should be assigned higher weights in the fusion. The real-time reliability weight W for each detection channel is... i The calculation formula is: Where k is the summation index, representing all detection channels participating in the fusion. In this embodiment, k ranges from 1 to 3, corresponding to the NDIR infrared absorption detection module, the QEPAS photoacoustic resonance detection module, and the LDO fluorescence quenching detection module, respectively. The formula uses the reciprocal of the noise variance as the weight base, with the denominator being the sum of the weight bases of all detection channels participating in the fusion, thus achieving normalization. Normalization ensures that the sum of the weights of all detection channels equals 1, satisfying the mathematical requirements of weighted fusion. The statistical basis for using the reciprocal of the variance as the weight is that, in multi-sensor data fusion theory, when the measurement errors of each sensor follow a zero-mean Gaussian distribution, using the reciprocal of the variance as the weight for weighted averaging can obtain the optimal estimate in the sense of minimum mean square error. This calculation method ensures that the system can automatically allocate the optimal weights according to the real-time health status of each channel (characterized by the noise variance).

[0103] Step S20 constructs a complete mathematical link from physical interference perception to fusion weight allocation in the all-optical collaborative detection scheme, realizing unified measurement and closed-loop feedback of multi-dimensional interference. By constructing an error sensitivity coefficient matrix, heterogeneous interference parameters such as frequency drift, optical window attenuation rate, and blackbody background offset are uniformly mapped to noise variance increments, realizing normalized quantification and differentiated weighting of interference in different physical dimensions, avoiding evaluation distortion caused by single-dimensional or equal-weighted processing. An iterative update mechanism based on Kalman filtering is adopted, so that noise variance estimation takes into account both historical accumulation and current disturbance, which can smooth the random fluctuations of instantaneous sampling and sensitively capture the gradual trend of sensor performance, providing early warning support for preventive maintenance. The real-time reliability weight generated by the reciprocal normalization operation breaks the limitations of traditional fixed weights and provides a unified health comparison benchmark for three different detection channels based on infrared absorption, photoacoustic resonance, and fluorescence quenching principles. This dynamic weight allocation mechanism ensures that in the complex environment of underground coal mines, when a channel fails due to specific interference (such as dust load or light window contamination), the system can automatically reduce its weight and tilt towards healthy channels, achieving smooth degraded operation of the measurement system and effectively avoiding the risk of missed detections or false alarms caused by blindly relying on data from failed channels.

[0104] Step S30: Combine the ambient temperature in the environmental compensation vector to invert the original optical feature signal and output the original inverted concentration of each target gas.

[0105] Further, step S30 includes:

[0106] Step S31: Substitute the light intensity of the signal channel and the light intensity of the reference channel into the Beer-Lambert law to perform absorbance inversion calculation and obtain the infrared absorption inversion concentrations of methane and carbon dioxide.

[0107] Specifically, the three types of raw signals—pure infrared absorption signal intensity, acoustic resonance signal, and fluorescence lifetime signal—are converted into numerical values ​​characterizing gas concentration. Concentration inversion refers to the process of converting the raw signal output by the sensor into the concentration value of the analyte using a mathematical model established based on the physical detection principle. Traditional concentration inversion methods use fixed conversion coefficients or lookup tables, which cannot adapt to the drift of sensor performance parameters during long-term operation. When environmental factors cause changes in the sensor's response characteristics, the inversion results will exhibit systematic deviations. Step S30 establishes dedicated inversion models for the three detection modules based on different physical principles, and introduces frequency drift gain reconstruction and temperature drift compensation mechanisms during the inversion process, enabling the inversion results to adapt to changes in sensor performance caused by the complex environment underground in coal mines.

[0108] Step S31 utilizes the signal channel light intensity I obtained in step S12 sig (λ1) and reference channel light intensity I ref (λ2) Infrared absorption concentration inversion is performed. The Beer-Lambert law is a fundamental law describing the absorption and attenuation of light as it propagates through a medium. According to this law, when monochromatic light passes through a homogeneous medium, the natural logarithm of the ratio of transmitted light intensity to incident light intensity is proportional to the product of the medium concentration and the optical path length. In NDIR infrared absorption detection, infrared light at the signal wavelength λ1 is absorbed by the target gas molecules, while infrared light at the reference wavelength λ2 is not absorbed by the target gas. The intensity ratio of the two wavelengths can be used to eliminate the effects of variations in light source intensity and common-mode interference in the optical path.

[0109] Infrared absorption inversion concentration C abs The calculation formula is: In the formula, α(λ1) is the absorption coefficient of the target gas at the signal wavelength λ1, characterizing the absorption capacity of gas molecules at that wavelength of infrared light per unit concentration and per optical path length. This coefficient is determined by the vibrational energy level structure of the target gas molecules and is obtained through measurement using standard concentration gas during factory calibration. L is the optical path length, representing the effective propagation path length of infrared light within the gas chamber. This length is determined by the geometry of the gas chamber. For gas chambers employing multi-reflection folded optical paths, the effective optical path length is the product of the physical path length and the number of reflections. sig (λ1) is the light intensity of the signal channel, acquired in step S12, reflecting the intensity of infrared light at the signal wavelength after passing through the gas chamber containing the target gas. ref (λ2) is the light intensity of the reference channel, which is acquired in step S12. It reflects the intensity of infrared light at the reference wavelength after passing through the gas chamber. Since there is no absorption in the gas at the reference wavelength, the light intensity is only affected by the changes in the light source intensity and the light path transmittance.

[0110] The formula uses the ratio of the light intensity of the signal channel to the light intensity of the reference channel for calculation. When the light source intensity decreases due to aging, the light intensities of both wavelengths decrease synchronously, and the ratio remains unchanged. When the transmittance of the optical window decreases due to contamination, the light intensities of both wavelengths decrease proportionally, and the ratio still remains unchanged. Using this ratio calculation eliminates the common-mode influence of light source intensity changes and optical window contamination on the inversion results. During long-term operation, even if the light source gradually ages or the optical window gradually becomes contaminated, as long as the attenuation ratio of the two wavelength channels is consistent, the inversion results can still remain accurate, extending the effective working cycle of the sensor and reducing calibration frequency and maintenance costs. The formula uses the natural logarithm for the comparison value because the original form of the Beer-Lambert law is an exponential decay relationship, where transmittance equals the exponent of negative absorbance. Taking the logarithm of both sides transforms the exponential relationship into a linear relationship, making absorbance proportional to concentration, which facilitates subsequent linear regression compensation and weighted fusion calculations. A negative sign is added before the logarithmic result because when the gas concentration increases, the light intensity I of the signal channel decreases. sig (λ1) decreases due to enhanced absorption, and the ratio I sig ( λ1) / I ref (λ2) decreases, the logarithm is negative, adding a negative sign makes the inversion concentration C... abs It is a positive value, which is consistent with the physical meaning.

[0111] For methane detection, the signal wavelength λ1 corresponds to the mid-infrared absorption peak of the carbon-hydrogen bond stretching vibration of the methane molecule. Substituting the light intensity of the signal channel and the light intensity of the reference channel obtained in step S12 into the above formula, the infrared absorption inversion concentration of methane is calculated and denoted as CCH. 4,IR For carbon dioxide detection, the signal wavelength λ1 corresponds to the mid-infrared absorption peak of the asymmetric stretching vibration of the carbon dioxide molecule. The infrared absorption inversion concentration of carbon dioxide is obtained using the same calculation formula, denoted as CCO. 2,IR When two gases are detected using the same NDIR infrared absorption detection module, the filter can be switched at different time periods using time-division multiplexing, or two independent optical channels can be used for simultaneous detection using spatial separation. Step S31 uses the Beer-Lambert law to perform concentration inversion, converting the infrared light intensity signal into a gas concentration value with clear physical meaning, and establishing a quantitative relationship between the sensor output signal and the concentration of the gas to be measured. The inversion model has a rigorous physical theoretical basis, and the model parameters (absorption coefficient and optical path length) can be obtained through physical measurement or factory calibration, without relying on a large number of samples for machine learning training, reducing the complexity of model establishment and the requirement for standard gases. The introduction of the reference channel makes the inversion results insensitive to light source aging and window contamination, enabling the sensor to operate for a long time in coal mines without frequent calibration, meeting the requirement of a six-month calibration cycle in mining safety regulations.

[0112] Step S32: Calculate the acoustic gain function under the current state based on the real-time physical resonance frequency and the factory-calibrated standard resonance frequency. Divide the acoustic resonance signal amplitude by the product of the acoustic gain function and the excitation light power to obtain the photoacoustic inversion concentration of carbon monoxide. The excitation light power is the emission power of the LED light source in the QEPAS photoacoustic resonance detection module.

[0113] Step S32 performs carbon monoxide concentration inversion for the QEPAS photoacoustic resonance detection module, introducing an acoustic gain function to compensate for gain changes caused by frequency drift. After gas molecules absorb modulated light energy, they generate pressure fluctuations due to periodic heating; these pressure fluctuations constitute the acoustic signal. In the QEPAS system, the acoustic signal generated after gas molecules absorb modulated light acts on the quartz tuning fork arm, exciting the tuning fork to vibrate and outputting an electrical signal through the piezoelectric effect. The resonant amplification characteristics of the quartz tuning fork enable it to amplify weak photoacoustic signals hundreds to thousands of times, achieving highly sensitive detection of trace amounts of gas. The resonant amplification factor of the quartz tuning fork is closely related to the degree of matching between the excitation frequency and the physical resonance frequency. When the excitation frequency is exactly equal to the physical resonance frequency, the amplification factor reaches its peak; when the excitation frequency deviates from the physical resonance frequency, the amplification factor drops sharply. Traditional QEPAS systems use a fixed excitation frequency for signal acquisition. When the physical resonant frequency of the quartz tuning fork drifts due to dust mass loading, a mismatch occurs between the excitation frequency and the physical resonant frequency, resulting in a decrease in the resonant amplification and a reduction in the sensor's sensitivity to gases. Because traditional systems cannot detect changes in the physical resonant frequency, a decrease in the response signal amplitude is misinterpreted as a decrease in gas concentration, leading to a risk of missed detections. Step S32 utilizes the real-time physical resonant frequency f... peak and frequency drift f drift An acoustic gain function is established to model the gain change caused by frequency drift, thereby achieving gain compensation during the concentration inversion process.

[0114] Acoustic gain function G(f QTF The formula for calculating ) is: In the formula, G0 is the standard resonance state reference gain, representing the resonant amplification factor of the quartz tuning fork when operating at its physical resonance frequency in a clean state. This value is obtained through factory calibration and reflects the intrinsic amplification capability of the quartz tuning fork. f0 is the factory-calibrated standard resonance frequency, numerically equal to the factory-set center frequency f defined in step S13. center , representing the physical resonant frequency of a quartz tuning fork in a clean state. QTF f is the actual operating frequency of the quartz tuning fork. QTF Equal to the real-time physical resonance frequency f obtained in step S13 peak k fThis is the frequency offset gain correction coefficient, characterizing the impact of resonant frequency offset on acoustic gain. This coefficient is obtained by measuring gain changes under different frequency offset conditions during factory calibration, reflecting the slope characteristics of the quartz tuning fork's amplitude-frequency response curve. The formula uses relative frequency offset. The independent variable used for gain correction is because the amplitude-frequency response curve of a quartz tuning fork can be approximated as a Lorentz function near the resonance peak. Within a small offset range, it can be linearized using a first-order Taylor expansion, resulting in a linear relationship between the relative frequency offset and the gain change. The reason for using a relative offset instead of an absolute offset is that different batches of quartz tuning forks have different physical resonance frequencies; using a relative offset allows for consistent correction coefficient k. f This method is applicable to different batches of products, enhancing the versatility of the parameters. The formula uses a multiplicative form to multiply the reference gain G0 by the correction factor because the change in gain is a proportional change relative to the reference state. Using a multiplicative form ensures that the gain function equals the reference gain G0 when the frequency offset is zero.

[0115] Photoacoustic inversion concentration of carbon monoxide C CO The calculation formula is: In the formula, A meas The amplitude of the acoustic resonance signal is obtained by step S14 and represents the piezoelectric signal amplitude output by the quartz tuning fork when the excitation frequency is locked at the real-time physical resonance frequency. The response coefficient for carbon monoxide concentration was calibrated and obtained through factory calibration, characterizing the system's sensitivity to carbon monoxide concentration response under unit optical power and unit resonant gain; G(f QTF P is the acoustic gain function under the current state, reflecting the resonant amplification capability of the quartz tuning fork under the current frequency offset state. opt The excitation power represents the emission power of the LED light source in the QEPAS photoacoustic resonance detection module. This power can be measured in real time through the built-in optical power monitoring circuit or using the nominal value set at the factory calibration. The formula is in the form of dividing the acoustic resonance signal amplitude by the product of the carbon monoxide concentration calibration response coefficient, the acoustic gain function, and the excitation optical power. Its physical meaning is: the acoustic resonance signal amplitude is proportional to the product of the gas concentration, the carbon monoxide concentration calibration response coefficient, the acoustic gain, and the excitation optical power. Dividing the signal amplitude by the calibration response coefficient, gain, and optical power yields the value proportional to the gas concentration. The acoustic gain function G(f) is introduced into the denominator. QTFWhen dust mass load causes resonant frequency drift, the acoustic gain decreases, and the signal amplitude generated by the same gas concentration decreases. Without gain compensation, the retrieved concentration will be lower than the true value. Introducing a gain function reduces the denominator as the gain decreases, keeping the fractional value constant, thus the retrieved concentration is unaffected by frequency drift. Step S32 introduces an acoustic gain function for gain compensation, converting the frequency drift into a gain correction factor to compensate for the sensitivity attenuation caused by dust mass load during concentration retrieval. Even if the resonant frequency of the quartz tuning fork drifts due to continuous dust deposition during long-term operation in a coal mine environment, the concentration retrieval results remain accurate, avoiding sensitivity attenuation and missed detection risks caused by resonance lock-up.

[0116] Step S33: Calculate and measure the fluorescence lifetime based on the fluorescence phase difference and the preset modulation frequency, and use the ambient temperature in the environmental compensation vector to perform temperature drift compensation on the reference fluorescence lifetime to obtain the temperature-compensated reference fluorescence lifetime.

[0117] Step S34: Calculate the fluorescence inversion concentration of oxygen based on the measured fluorescence lifetime and the reference fluorescence lifetime after temperature compensation;

[0118] Steps S33 and S34 involve oxygen concentration inversion for the LDO fluorescence quenching detection module. The core of these steps is the introduction of a temperature drift compensation model to correct the reference fluorescence lifetime. Fluorescence lifetime refers to the average time constant of fluorescence emission during the transition of a fluorescent molecule from an excited state back to its ground state. This lifetime is determined by the intrinsic electronic structure of the fluorescent molecule and is independent of excitation light intensity and detector sensitivity. Therefore, fluorescence lifetime-based detection has higher long-term stability compared to fluorescence intensity-based detection. In oxygen detection, oxygen molecules interact with excited-state fluorescent molecules through a collisional quenching mechanism, dissipating the excitation energy of the fluorescent molecules in a non-radiative manner, thus shortening the fluorescence lifetime. According to the Stern-Wolmer fluorescence quenching model, fluorescence lifetime is inversely proportional to oxygen concentration; therefore, oxygen concentration can be inverted by measuring changes in fluorescence lifetime.

[0119] Step S33 is based on fluorescence phase difference meas and preset modulation frequency f mod Calculate and measure fluorescence lifetime τ meas Phase-domain fluorescence lifetime measurement is a lifetime measurement method based on frequency domain analysis. Its principle is as follows: when a fluorescent molecule is excited by sinusoidally modulated excitation light, the emitted fluorescence signal also exhibits a sinusoidal change. However, due to the time delay in fluorescence emission, the fluorescence signal lags behind the excitation light signal in phase. This phase lag has a definite mathematical relationship with the fluorescence lifetime. The measurement of fluorescence lifetime τ... meas The calculation formula is: , in the formula, tan( measThe excitation light signal is represented by the tangent of the fluorescence phase difference, reflecting the ratio of the quadrature to the in-phase components of the fluorescence signal. tan(t) is the tangent trigonometric function. 2π is twice pi and is used to convert angular frequency to linear frequency. The physical derivation of the formula is as follows: Let the excitation light signal be a sinusoidal function E(t) = E0sin(2πf). mod ×t), where E(t) is the instantaneous intensity of the excitation light signal, representing the light intensity emitted by the LED light source in the LDO fluorescence quenching detection module at time t, E0 is the amplitude of the excitation light signal, characterizing the maximum value (peak value) of the modulated light intensity of the LED light source, t is the time variable, sin() is the sine function, and the fluorescence emission process can be described by a first-order linear differential equation, the steady-state solution of which is the fluorescence signal F(t) = F0sin(2πf mod ×t- ), where F(t) is the instantaneous intensity of the fluorescence signal, characterizing the intensity of the fluorescence emitted by the oxygen-sensitive fluorescent film after excitation at time t, and F0 is the amplitude of the fluorescence signal, characterizing the maximum value of the emitted fluorescence intensity. Usually, F0 is less than E0 because the fluorescence quantum efficiency is usually less than 1 and there is loss in the optical path. This is due to phase lag. Because there is a time delay (lifetime) between the absorption and emission of fluorescent molecules, the waveform of the fluorescence signal lags behind the excitation light signal on the time axis. Satisfy tan( )=2πf mod ×τ, after transformation, we get fluorescence lifetime τ=tan( ) / (2πf mod The formula shows that fluorescence lifetime is directly proportional to the tangent of phase lag and inversely proportional to the modulation frequency, which is consistent with physical intuition: the longer the fluorescence lifetime, the greater the time delay of fluorescence emission and the more obvious the phase lag.

[0120] Step S33 utilizes the ambient temperature T in the environmental compensation vector to adjust the reference fluorescence lifetime τ. ref Temperature drift compensation was performed to obtain the temperature-compensated reference fluorescence lifetime τ. ref,compFluorescence lifetime is affected by temperature, and the mechanism of change is as follows: increased temperature leads to increased molecular thermal motion, increased nonradiative transition rates, and shortened fluorescence lifetime. In underground coal mines, ambient temperature may fluctuate significantly. Without temperature compensation, temperature changes will be misinterpreted as changes in oxygen concentration, leading to distorted inversion results. The fluorescence lifetime temperature drift compensation model is: τ(T) = τ0 × (1 + λ × (T - T0)), where τ0 is the standard fluorescence lifetime at the reference temperature, representing the intrinsic fluorescence lifetime of the fluorescent molecule at the reference temperature T0, obtained through factory calibration. T0 is the reference temperature, usually set as the standard temperature under room temperature conditions, serving as the benchmark for temperature compensation. λ is the temperature coefficient, characterizing the sensitivity of fluorescence lifetime to temperature changes. This coefficient is obtained by measuring the change in fluorescence lifetime under different temperature conditions during factory calibration, reflecting the temperature characteristics of the fluorescent dye. The formula uses a linear model to describe the relationship between fluorescence lifetime and temperature because, within the temperature range of industrial applications, the temperature dependence of fluorescence lifetime can be approximated as a first-order linear relationship, and the influence of higher-order nonlinear terms can be ignored. The formula uses the relative temperature difference (T-T0) as the independent variable, making the compensation factor equal to 1 at the reference temperature, and the model output equal to the standard fluorescence lifetime τ0. A negative temperature coefficient λ indicates that increasing temperature shortens the fluorescence lifetime, while a positive value indicates that increasing temperature prolongs the fluorescence lifetime. The specific value is determined by the molecular structure of the fluorescent dye.

[0121] Temperature-compensated reference fluorescence lifetime τ ref,comp The calculation formula is: In the formula, T ref To measure the ambient temperature during reference fluorescence lifetime measurement, it can be recorded using a temperature sensor or assumed to be the same as the current ambient temperature. The physical meaning of the formula is: normalize the reference fluorescence lifetime from the temperature conditions at the time of measurement to the current ambient temperature conditions, eliminating lifetime deviations caused by temperature differences between the two measurement times. The temperature compensation factor in the numerator projects the reference lifetime onto the current temperature state, while the temperature compensation factor in the denominator is used for normalization, ensuring that τ is the same when the measurement temperature is the same as the current temperature. ref,comp Equal to τ ref .

[0122] Step S34 is based on the measurement of fluorescence lifetime τ meas and temperature-compensated reference fluorescence lifetime τ ref,comp Calculate the fluorescence inversion concentration of oxygen The Stern-Wolmer equation describes the quantitative relationship between fluorescence lifetime and quencher concentration during fluorescence quenching. This equation allows changes in fluorescence lifetime to be converted into oxygen concentration values.

[0123] Fluorescence inversion concentration of oxygen The calculation formula is: , in the formula The oxygen quenching constant characterizes the quenching efficiency of oxygen molecules on fluorescence lifetime. This constant is determined by the interaction strength between the fluorescent dye and oxygen molecules and is obtained through factory calibration. The derivation of the formula is as follows: based on the Stern-Wolmer equation, After transformation, we get The formula uses the temperature-compensated reference fluorescence lifetime τ. ref,comp Replacement of the original reference fluorescence lifetime τ ref This eliminates the effect of temperature drift on the lifetime ratio. When the oxygen concentration is zero, τ meas Equal to τ ref,comp The ratio is equal to 1, and the concentration is equal to zero; when the oxygen concentration increases, τ meas The shortening due to the quenching effect results in a ratio greater than 1, with a positive concentration. Oxygen quenching constant. The reciprocal of the lifetime ratio is used as a proportionality coefficient to convert the change in lifetime ratio into a numerical value of oxygen concentration. Steps S33 and S34 introduce a temperature drift compensation model, eliminating the influence of ambient temperature changes on fluorescence lifetime and ensuring that the oxygen concentration inversion results are not affected by temperature fluctuations. In the environment of coal mines where temperatures may fluctuate significantly, the oxygen detection results remain accurate, avoiding false alarms caused by misinterpreting temperature changes as abnormal oxygen concentrations. The introduction of a reference channel enables the system to have dual lifetime ratio measurement capabilities. Through ratio calculation, the influence of common-mode interference such as fluorescent dye aging and LED light source intensity changes on the inversion results is eliminated, improving the long-term stability of the sensor.

[0124] Step S35: Combine the infrared absorption inversion concentrations of methane and carbon dioxide, the photoacoustic inversion concentration of carbon monoxide, and the fluorescence inversion concentration of oxygen to form the original inversion concentrations of each detection channel.

[0125] Step S35 involves inverting the concentrations (CCH) of methane and carbon dioxide using infrared absorption. 4,IR and CCO 2,IR The photoacoustic inversion concentration of carbon monoxide, C CO Fluorescence inversion concentration of oxygen The original inversion concentrations of each target gas are combined. The original inversion concentration refers to the concentration value that has only been inverted through the physical model but has not yet undergone environmental factor regression compensation. This provides input data for the environmental regression compensation and confidence-weighted fusion in subsequent step S40. The original inversion concentration of each target gas can be represented as a concentration vector, with each component of the vector corresponding to the inversion result of different gases. This data organization method facilitates subsequent matrix operations and batch processing. The frequency drift gain reconstruction mechanism embedded in step S30 accurately compensates for the resonance frequency drift caused by dust mass load in the QEPAS photoacoustic resonance detection module, fundamentally solving the sensitivity attenuation problem caused by the mismatch between the excitation frequency and the physical response frequency. Simultaneously, the temperature drift compensation model effectively corrects the measurement deviation of the LDO fluorescence quenching detection module in the large temperature variation environment of underground coal mines. By combining common-mode interference cancellation achieved through the reference channel, step S30 not only achieves accurate mapping from physical signals to concentration data, but also overcomes the effects of light source aging, dye decay, and environmental disturbances through intrinsic adaptive compensation. While reducing calibration complexity, it significantly improves the robustness and measurement accuracy of the system in long-term operation.

[0126] Step S40: Use the environmental compensation vector to perform environmental factor regression compensation on the original inversion concentration of each target gas to obtain the environmentally compensated concentration. Use the real-time reliability weight of each detection channel to perform weighted fusion calculation on the environmentally compensated concentration and output the final fused concentration of each target gas.

[0127] Further, step S40 includes:

[0128] Step S41: Extract ambient temperature and relative humidity from the environmental compensation vector, perform multidimensional regression compensation calculation on the original inversion concentration of each target gas, and obtain the environmentally compensated concentration of each target gas.

[0129] Step S42: Establish a mapping mechanism between target gases and detection channels, construct a time sliding window, use the mapping mechanism to match the real-time reliability weights corresponding to each target gas, perform time-domain weighted fusion calculation on the environmentally compensated concentrations of each target gas, and output the final fused concentrations of methane, carbon monoxide, carbon dioxide and oxygen.

[0130] Specifically, step S40 eliminates the residual influence of environmental factors such as temperature and humidity on concentration inversion through environmental regression compensation, and then integrates the concentration data from multiple detection channels into a single, highly reliable output value through reliability-weighted fusion. Environmental regression compensation refers to the process of establishing a mathematical relationship between environmental parameters and concentration measurement deviations using statistical regression methods, and correcting the original inverted concentration based on current environmental conditions. Reliability-weighted fusion refers to a data processing method that assigns different weight coefficients according to the measurement reliability of each detection channel, and synthesizes multiple concentration values ​​into a single output value through weighted averaging. Traditional multi-sensor gas detection systems typically use fixed weights or simple averaging to fuse data from each channel. When the measurement deviation of a certain detection channel increases due to environmental interference, its erroneous data will still participate in the fusion calculation at a fixed proportion, contaminating the final output result. Step S40 introduces a dynamic weighting mechanism, enabling the fusion algorithm to automatically adjust the weight allocation according to the real-time health status of each detection channel, placing the data of healthy channels in a dominant position and effectively suppressing the data of damaged channels, achieving a leap from passive detection to active sensing capabilities.

[0131] Step S41 extracts the ambient temperature T and relative humidity RH from the environmental compensation vector, and performs multidimensional regression compensation calculations on the original inversion concentrations of each detection channel output in step S30. The environmental compensation vector contains three components: ambient temperature T, relative humidity RH, and atmospheric pressure P. Ambient temperature T is the thermodynamic temperature of the air in the coal mine roadway, which affects the temperature dependence of the infrared absorption coefficient, the temperature drift of the fluorescence quenching constant, and the resonance characteristics of the quartz tuning fork. Relative humidity RH is the ratio of the actual partial pressure of water vapor in the air to the saturated partial pressure of water vapor, which affects the infrared absorption cross section of gas molecules and the oxygen permeability of the fluorescent membrane.

[0132] The multidimensional regression compensation model establishes a linear relationship between the original inverted concentration and environmental parameters, quantifying the impact of environmental factors on concentration measurement into a calculable correction term. The environmental regression compensation concentration C for the j-th target gas is... j comp The calculation formula is: C j comp =α×S j +β T ×T+γ RH ×RH+δ0, where S in the formula j Let be the original inverted concentration value of the j-th target gas, where j ranges from 1 to 4, corresponding to the four target gases: methane, carbon monoxide, carbon dioxide, and oxygen. α is the concentration signal scaling factor, characterizing the direct contribution of the original inverted concentration to the environmentally compensated concentration. A coefficient close to 1 indicates that the original inverted concentration is basically accurate and requires no significant correction; a coefficient deviating from 1 indicates a systematic bias in the original inverted concentration, requiring scaling adjustment through a regression model. βT β is the temperature correction coefficient, characterizing the concentration deviation caused by a unit temperature change. Its physical meaning lies in the fact that as the ambient temperature increases, physical parameters such as the infrared absorption coefficient and fluorescence quenching efficiency change, causing a drift in the sensor output signal under the same gas concentration conditions. T This is used to compensate for this temperature-dependent bias. γ RH The humidity correction factor characterizes the concentration deviation caused by a unit change in relative humidity. Its physical significance lies in the fact that water vapor molecules have absorption lines in the mid-infrared band, which may cross-interfere with the absorption peaks of methane and carbon dioxide. Simultaneously, water vapor alters the oxygen permeability of the fluorescent membrane, γ... RH This is used to compensate for this humidity-dependent bias. δ0 is a constant bias term that characterizes the zero-point drift of the sensor background. When both ambient temperature and humidity are at reference levels, δ0 reflects a constant deviation between the sensor output and the true concentration.

[0133] Multidimensional regression coefficients α, β T γ RH The acquisition of δ0 requires combining factory calibration data with an online learning mechanism. During factory calibration, the system changes temperature and humidity conditions under a standard gas environment, records the deviation data between the sensor output and the standard concentration, and obtains the initial values ​​of the regression coefficients through least squares fitting. Least squares is a well-known regression analysis method that finds the coefficient values ​​that minimize the fitting error by minimizing the sum of squared residuals. For example, the temperature correction coefficient β... T The determination process is as follows: Under fixed gas concentration conditions, the ambient temperature is adjusted in multiple gradients within the typical working temperature range of underground coal mines. The sensor output deviation corresponding to each temperature point is recorded. A linear regression is performed with temperature as the independent variable and output deviation as the dependent variable. The slope of the regression line is β. T Humidity correction factor γ RH A similar method was used to determine the relative humidity under fixed temperature conditions, recording the output deviation and performing regression analysis. The formula uses a linear combination to superimpose the original concentration, temperature correction term, humidity correction term, and constant bias because within the temperature and humidity range of industrial applications, the influence of environmental factors on the sensor output can be approximated as a first-order linear relationship, and the contribution of higher-order nonlinear terms is negligible. The reason for using a multiplicative superposition form instead of a multiplicative form is that the influence mechanisms of temperature and humidity on the sensor output are independent, and the deviations caused by temperature changes and humidity changes can be linearly superimposed. Each correction term in the formula is related to S... j Decoupling makes the environmental compensation process and the concentration inversion process independent of each other, avoiding additional nonlinear distortion introduced by the compensation calculation.

[0134] The multidimensional regression coefficients can also be adaptively updated through an online learning mechanism, enabling the compensation model to adapt to the slow drift of performance parameters during long-term sensor use. The online learning mechanism is implemented as follows: the system continuously collects raw inversion concentration, environmental parameters, and reference concentration data, and incrementally updates the regression coefficients using a recursive least squares algorithm. The recursive least squares algorithm is a well-known adaptive filtering algorithm that introduces a forgetting factor to give higher weight to recent data, allowing the regression coefficients to track the slow changes in system parameters. The reference concentration can be obtained by using a standard gas to obtain the true concentration value during periodic calibration, or by using the most consistent concentration value as a quasi-reference value when multiple detection channels output the same value. The online learning mechanism enables the multidimensional regression model to have self-calibration capabilities. When the sensor's response characteristics drift due to aging or contamination, the regression coefficients can automatically adjust to compensate for this change, reducing the frequency of reliance on on-site calibration and allowing the system to meet the six-month calibration cycle requirement of mining safety regulations.

[0135] Step S41 performs multidimensional regression compensation calculations on the four target gases respectively, obtaining four environmentally compensated concentration values: environmental regression compensated concentration of methane CCH4. comp Environmental regression compensation concentration of carbon monoxide (CCO) comp Environmental regression compensation concentration of carbon dioxide (CCO2) comp Environmental regression compensation concentration of oxygen CO2 comp The regression coefficients differ for different gases, reflecting the varying sensitivities of each gas detection principle to environmental factors. For example, methane detection is based on the NDIR infrared absorption principle, and its temperature correction coefficient β... T The primary function is to compensate for the temperature dependence of the infrared absorption cross section of methane molecules; carbon monoxide detection is based on the QEPAS photoacoustic resonance principle, with a temperature correction coefficient β. T The primary function is to compensate for the temperature dependence of photoacoustic signal generation efficiency; oxygen detection is based on the LDO fluorescence quenching principle, with a temperature correction coefficient β. TThe primary function is to compensate for the temperature dependence of the fluorescence quenching constant. Step S41 employs a multidimensional regression compensation method to separate and eliminate the influence of ambient temperature and relative humidity on concentration measurement from the original inverted concentration, making the environmentally compensated concentration more accurately reflect the true gas concentration. In the environment of large temperature and humidity fluctuations in underground coal mines, the concentration output of each target gas no longer experiences systematic deviations due to changes in environmental conditions, avoiding false alarms or missed detections caused by misinterpreting environmental changes as changes in gas concentration. The linear structure of the multidimensional regression model provides clear physical interpretation for the compensation calculation, and each correction coefficient can be traced back to a specific physical mechanism, facilitating parameter verification during system debugging and fault diagnosis. The online learning mechanism enables the regression coefficients to track the slow drift of sensor performance, maintaining the effectiveness of the compensation model during long-term operation and extending the effective working cycle of the sensor.

[0136] Step S42 aims to address the weight allocation and data fusion problem of "multiple gases, few channels" in heterogeneous detection arrays. Since different detection modules in the all-optical detection array perform different detection tasks, step S42 first establishes a deterministic mapping relationship between the environmentally compensated concentration and the real-time reliability weight. In this embodiment, the target gas index j ranges from 1 to 4, corresponding to methane (j=1), carbon monoxide (j=2), carbon dioxide (j=3), and oxygen (j=4), respectively; the detection channel index i ranges from 1 to 3, corresponding to the NDIR infrared absorption detection module (i=1), the QEPAS photoacoustic resonance detection module (i=2), and the LDO fluorescence quenching detection module (i=3), respectively. When j=1 (methane) or j=3 (carbon dioxide), the corresponding detection channel index i=1 is mapped. This is because both methane and carbon dioxide are detected by the NDIR infrared absorption detection module, and the physical health status of this module (such as window attenuation and blackbody interference) determines the reliability of the measurement data for these two gases; therefore, they share the same real-time reliability weight W1. When j=2 (carbon monoxide), the corresponding detection channel index i=2. This gas is detected independently by the QEPAS photoacoustic resonance detection module, and its data reliability is affected by factors such as quartz tuning fork frequency drift, determined by the real-time reliability weight W2. When j=4 (oxygen), the corresponding detection channel index i=3. This gas is detected independently by the LDO fluorescence quenching detection module, and its data reliability is determined by the real-time reliability weight W3. Based on the above mapping relationship, the system performs time-domain weighted fusion calculations for each type of target gas. This calculation introduces a sliding window mechanism on the time axis, using the real-time health status of the detection channels (represented by weights) to adaptively filter the concentration data.

[0137] Final fusion concentration of each target gas The calculation formula is as follows: ,in, Indexing for the target gas, ; This is a mapping function used to map data based on gas indices. Return the corresponding detection channel index ,Right now . The size of the time sliding window represents the number of historical sampling points participating in the fusion (e.g., taking the most recent one). (Sampling period). This serves as an index for the sampling time within the time sliding window. This refers to the current moment. For the first The target gas in Concentration after environmental compensation at any given time. In order to be with the first The detection channels corresponding to the gases are in The formula assigns real-time reliability weights based on the physical source channel. The mapping function Map(j) ensures strict alignment of concentration data with the health status of the physical source channel. For the NDIR channel (i=1), if W1 decreases at a certain moment due to window contamination, this weight change will simultaneously affect the fusion calculation of methane (j=1) and carbon dioxide (j=3), synchronously suppressing the abnormal data contribution of these two gases at that moment. The numerator uses weighted summation, and the denominator uses weighted normalization, ensuring that the final output physical quantity is in concentration units. This time-domain fusion mechanism based on physical attribution not only solves the data fluctuation problem of single-channel detection under instantaneous interference but also rigorously handles the weight reuse logic under the "one-to-two" detection architecture of the NDIR module, ensuring the logical consistency and measurement accuracy of the output data of the all-optical collaborative detection system.

[0138] Step S40 employs a multidimensional regression compensation method for environmental factors combined with time-domain reliability weighted fusion. This method uses a regression model to eliminate nonlinear interference from changes in ambient temperature and relative humidity on the original inverted concentration, thus normalizing the concentration data to standard environmental conditions. A time-sliding window mechanism is used to dynamically reconstruct data based on the health status of each detection channel at different sampling times (determined by historical and current noise variance). This automatically suppresses the weight of data during periods of high interference (such as instantaneous dust impact or light window obstruction) and enhances the contribution of data during stable periods. This successfully solves the measurement distortion problem caused by the variable climate and frequent physical interference in underground coal mine environments, ensuring the accuracy of gas concentration readings under significant temperature and humidity fluctuations. It also provides robustness to the single-channel detection architecture in the face of strong instantaneous interference. This mechanism allows the system to achieve "soft redundancy" through time-dimensional information complementarity without hardware redundancy, ensuring the logical consistency and long-term stability of monitoring data for key gases such as methane and carbon monoxide under complex operating conditions, and avoiding false alarms or missed detections caused by sudden environmental changes.

[0139] Step S50: The final fusion concentration of each target gas is compared with a preset safety threshold to obtain the gas concentration alarm level. At the same time, each interference feature parameter in the interference feature parameter set is compared with a preset health threshold to determine the health status level of each detection channel. A composite decision is generated based on the gas concentration alarm level and the health status level of each detection channel.

[0140] Further, step S50 includes:

[0141] Step S51: Set multi-level safety thresholds for each target gas, compare the final fusion concentrations of methane, carbon monoxide, carbon dioxide and oxygen with the corresponding multi-level safety thresholds, and determine the gas concentration alarm level based on the comparison results.

[0142] Step S52: Set health thresholds for frequency drift, optical window attenuation rate and blackbody background offset respectively; compare each interference feature parameter in the interference feature parameter set with the corresponding health threshold; and determine the health status level of each detection channel based on the comparison results.

[0143] Step S53: Logically combine the gas concentration alarm level with the health status level of each detection channel to generate a composite decision.

[0144] Specifically, step S50 transforms the final fused concentration of each target gas and the set of interference characteristic parameters into executable safety decisions and equipment maintenance instructions. Traditional coal mine gas detection systems only issue a single alarm for concentration exceeding limits. When sensor performance degrades due to dust adhesion or window contamination, the system cannot recognize the decreased reliability of the measurement results. It may output normal concentration readings even when the sensor is severely faulty, or misjudge sensor performance degradation as abnormal gas concentration. Step S50 establishes a dual-channel decision-making mechanism of graded gas concentration alarm and sensor health diagnosis, enabling the system to simultaneously possess the capabilities of environmental safety monitoring and equipment status monitoring. The output decision signal not only reflects the safety status of the underground gas environment but also includes a quantitative assessment of the sensor's own operating status, providing more comprehensive information support for mine safety management.

[0145] Step S51 sets multi-level safety thresholds for each target gas and executes graded alarms. Multi-level safety thresholds refer to multiple concentration thresholds set for the same target gas, with different thresholds corresponding to different levels of danger and alarm levels. The reason for using multi-level thresholds instead of a single threshold is that the danger of gas concentrations in underground coal mines is continuously changing; a single threshold can only distinguish between normal and exceeding limits, failing to reflect the gradual change in danger. Setting multi-level thresholds allows the system to issue alerts when the gas concentration is still within the warning range, providing a time window for personnel evacuation and emergency response. The multi-level safety thresholds for each target gas adopt a three-level structure, with the three thresholds increasing sequentially and defined as the warning threshold, alarm threshold, and emergency threshold, respectively. The warning threshold corresponds to the critical concentration where the gas concentration begins to deviate from the normal range but has not yet constituted an immediate danger; the alarm threshold corresponds to the critical concentration where the gas concentration has entered the danger zone and emergency measures are required; and the emergency threshold corresponds to the limit concentration where the gas concentration has reached the point where an explosion or severe poisoning may occur. The values ​​of the multi-level safety thresholds are determined according to coal mine safety regulations and industry standards. The threshold setting for each target gas needs to consider the gas's toxicological characteristics, explosion limits, and underground ventilation conditions. For example, the three-level safety thresholds for each target gas can be set as shown in Table 1:

[0146] Table 1. Examples of Level 3 Safety Thresholds for Various Target Gases

[0147]

[0148] The execution process of the graded alarm is as follows: the final fusion concentration of methane C is sequentially increased. CH4,final The final fusion concentration of carbon monoxide, C CO,final The final fusion concentration of carbon dioxide C CO2,final The final fusion concentration of oxygen C O2,final The values ​​are compared with their respective multi-level safety thresholds. Taking methane as an example, let the warning threshold for methane be Th. 1,CH4 The alarm threshold is Th 2,CH4 The emergency threshold is Th 3,CH4 The alarm level determination logic is as follows: when C CH4,final Less than Th 1,CH4 When the alarm level is determined to be normal, the system does not trigger any alarm action; when C CH4,final Greater than or equal to Th 1,CH4 And less than Th 2,CH4 When the alarm level is determined to be a warning, the system triggers a yellow audible and visual alert and sends a warning message to the monitoring center; when C CH4,final Greater than or equal to Th 2,CH4 And less than Th 3,CH4 When the alarm level is determined to be an alarm, the system triggers a red audible and visual alarm and sends a personnel evacuation suggestion to the monitoring center; when C CH4,final Greater than or equal to Th3,CH4 When the alarm level is determined to be emergency, the system triggers a continuous alarm and sends a power-off command to the linked control equipment. The alarm logic for oxygen is the opposite of that for methane, carbon monoxide, and carbon dioxide, employing a lower-limit alarm mechanism: an early warning is triggered when the oxygen concentration falls below the warning threshold, an alarm is triggered when it falls below the alarm threshold, and an emergency alarm is triggered when it falls below the emergency threshold, as insufficient oxygen in coal mines also threatens personnel safety. This tiered alarm mechanism allows different levels of gas concentration anomalies to trigger differentiated response levels. A warning level can trigger audible and visual alerts and enhanced ventilation measures; an alarm level can trigger personnel evacuation commands; and an emergency level can trigger equipment power-off protection. This tiered response avoids the potential for over- or under-response that might result from a single threshold alarm.

[0149] Step S52 sets a health threshold for the interference characteristic parameter and performs sensor health diagnosis. The health threshold is a critical value used to determine whether the sensor's operating state is within the normal range. When the interference characteristic parameter exceeds the health threshold, it indicates that the corresponding detection channel has experienced significant performance degradation, and the reliability of its measurement data decreases. The health threshold Th for frequency drift is... f The setting is based on the full width at half maximum (FWHM) of the quartz tuning fork's amplitude-frequency response curve. FWHM refers to the frequency interval between two frequency points where the amplitude drops to half of the peak value on the amplitude-frequency response curve; this parameter characterizes the resonant bandwidth of the quartz tuning fork. When the frequency drift exceeds FWHM, the excitation frequency has deviated from the center of the resonant peak to the position where the amplitude drops to half, and the acoustic gain drops to less than half of the standard state. The attenuation of the signal amplitude will cause a significant deviation in the concentration inversion results. For example, if the factory-calibrated FWHM of the quartz tuning fork is 4 Hz, then the health threshold Th for the frequency drift... f It can be set to 4 Hz. The health threshold Th for the optical window attenuation rate. I The setting is based on the lower limit of the dynamic range of the photodetector. Dynamic range refers to the range of light intensity within which the detector can respond linearly. When the incident light intensity is below the lower limit of the dynamic range, the linear relationship between the detector's output signal and the incident light intensity fails, increasing the risk of signal distortion. For example, if the detector enters the nonlinear region when the light intensity decays to less than 30% of its initial value, then the health threshold Th for the light window decay rate... I It can be set to 70%. The health threshold for the black body background offset (Th) bb The setting is based on the signal-to-noise ratio (SNR) requirement of the infrared absorption signal. When the increase in blackbody background radiation causes the background noise power to exceed a certain proportion of the effective signal power, the accuracy of concentration inversion will significantly decrease. For example, if the system requires an SNR of no less than ten times, then the health threshold Th for the blackbody background offset... bb It can be set to 10%.

[0150] The sensor health diagnosis process is as follows: the frequency drift f collected in step S10 is processed... drift , Light window attenuation rate I atten and bold background offset Tbb bias The parameters are compared with the corresponding health thresholds, and the health status level is determined based on the ratio of the interference feature parameter to the health threshold. The health status level is divided into three levels: Good, Attention, and Deterioration. The determination rules for each level are as follows: When the interference feature parameter is less than the health threshold multiplied by a preset attention coefficient, the health status level is determined to be Good. The attention coefficient is a positive number less than 1, used to provide a warning buffer before the interference feature parameter approaches the health threshold. For example, the attention coefficient can be set to 0.5. When the interference feature parameter is greater than or equal to the health threshold multiplied by the attention coefficient and less than the health threshold, the health status level is determined to be Attention, indicating that the sensor performance has begun to degrade but has not yet reached a critical state. When the interference feature parameter is greater than or equal to the health threshold, the health status level is determined to be Deterioration, indicating that the sensor performance has degraded to a critical state, and the reliability of the measurement data has significantly decreased.

[0151] Taking the QEPAS photoacoustic resonance detection module as an example, the specific execution of health diagnosis is explained: Let Th be the health threshold for frequency drift. f If the frequency is 4 Hz and the attention coefficient is 0.5, then the attention threshold is 2 Hz. When the frequency drift is f... drift When the frequency is less than 2 Hz, the health status of the QEPAS photoacoustic resonance detection module is determined to be good, and the system operates normally without the need for maintenance; when f drift When the frequency is greater than or equal to 2 Hz and less than 4 Hz, the health status level is determined to be "Caution". The system displays maintenance prompts on the output interface and records them in the equipment log, prompting the user to schedule quartz tuning fork cleaning in the next maintenance cycle; when f drift When the frequency is greater than or equal to 4 Hz, the health status is determined to be degraded, and the system immediately issues a device fault alarm. For the NDIR infrared absorption detection module and the LDO fluorescence quenching detection module, the same three-level health status classification logic is used, with health diagnoses based on the optical window attenuation rate and blackbody background offset, respectively. The sensor health diagnosis transforms the interference characteristic parameters collected in step S10 into a quantitative assessment of the device status, making the sensor performance degradation process observable and predictable. Maintenance prompts are issued before complete performance failure, avoiding the risk of missed detections due to sensor failure.

[0152] Step S53 logically combines the gas concentration alarm level and the detection channel health status level to generate a composite decision output. Composite decision refers to a decision result formed after comprehensively considering both environmental safety conditions and equipment operating status, offering higher reliability compared to single-dimensional decisions. The logical combination is implemented using a decision matrix approach. The decision matrix is ​​a two-dimensional lookup table, with row indices representing gas concentration alarm levels, column indices representing detection channel health status levels, and matrix elements representing corresponding composite decision type codes. The construction rules for the decision matrix are: along the row direction, higher alarm levels indicate higher decision urgency; along the column direction, lower health status levels result in more confidence level annotations for the decision. For example, the specific content of the decision matrix is ​​shown in Table 2:

[0153] Table 2 Two-dimensional table of decision matrix

[0154]

[0155] The process of generating composite decisions is as follows: Based on the alarm levels of each target gas output in step S51 and the health status levels of each detection channel output in step S52, the decision matrix is ​​queried using the alarm level as the row index and the health status level as the column index to obtain the corresponding composite decision type code. When the same detection channel corresponds to multiple target gases, the highest alarm level is used as the comprehensive alarm level for that channel in the decision matrix query, following the principle of taking the highest alarm level. For example, if the NDIR infrared absorption detection module detects methane and carbon dioxide simultaneously, and the methane alarm level is warning while the carbon dioxide alarm level is normal, then the comprehensive alarm level for that channel is warning. When the gas concentration alarm level is normal but the health status level is deteriorated, the composite decision is to maintain the alarm and mark the credibility downgrade. The system adds a credibility downgrade identifier to the output concentration data, prompting the downstream system that the data comes from a detection channel with deteriorated performance and should be used with caution. When the gas concentration alarm level is alarm but the health status level is deteriorated, the composite decision is to issue an alarm and suggest manual verification. This is because it is impossible to determine whether the high concentration reading is due to an actual increase in gas concentration or a measurement deviation caused by sensor failure. The system sends a manual verification request to the monitoring center at the same time as triggering the alarm, requiring on-site personnel to carry a portable calibration instrument for secondary confirmation.

[0156] Step S50 forms a complete closed-loop collaboration with steps S10, S20, and S40. The interference feature parameter set collected in step S10 includes three parameters: frequency drift, optical window attenuation rate, and blackbody background offset. This parameter set serves both as input variables in the noise variance update formula of step S20 for calculating real-time reliability weights and as a basis for health diagnosis in step S52 for sensor state assessment, achieving dual utilization of interference information at both the data and decision levels. The real-time estimated noise variance calculated in step S20 based on the interference feature parameters is used in the reliability weighted fusion operation of step S40 to determine the fusion weights of each detection channel, and also serves as an auxiliary reference indicator for health diagnosis in step S52: when the real-time reliability weight of a detection channel continuously falls below a preset lower weight threshold, even if the corresponding interference feature parameter has not yet exceeded the health threshold, step S52 still marks the channel as requiring attention and generates an auxiliary maintenance prompt. The lower weight threshold is set based on the statistical lower bound of the weight distribution of each detection channel under normal operating conditions. The final fusion concentration of each target gas output in step S40 is converted into a concentration alarm level after the hierarchical alarm judgment in step S51. Together with the health status level output in step S52, it serves as the index for the decision matrix query in step S53, completing the full-link information flow from the original signal acquisition to the safety decision output.

[0157] The execution of the system degradation operation instruction in step S50 and the noise variance update in step S20 form a feedback loop. When step S52 determines that the health status level of a detection channel is degraded, the system degradation operation instruction generated in step S53 includes an operation code that forces the noise variance of the channel to be set to the degradation penalty value. This operation code is passed to the noise variance update module in step S20 for execution, so that the real-time estimated noise variance of the channel in the next sampling period is forcibly overwritten with the degradation penalty value instead of being calculated according to the conventional formula. The degradation penalty value is much larger than the normal state noise variance, causing the real-time reliability weight of the channel calculated in step S23 to approach zero, and the concentration data contribution of the channel in the weighted fusion operation in step S40 is effectively suppressed. This feedback mechanism enables the health diagnosis result to affect the data fusion process in real time. After the sensor performance degradation is detected at the decision level, the weight of the damaged channel is immediately reduced at the data level, avoiding the pollution of the final fused concentration output by erroneous data from the degraded channel.

[0158] Step S50 establishes a dual-channel decision-making architecture for graded gas concentration alarms and sensor health diagnosis, unifying environmental safety monitoring and equipment status monitoring into the system output. The graded alarm mechanism sets three threshold levels—early warning, alarm, and emergency—based on the toxicological characteristics and explosion limits of each target gas. This allows the system to issue alerts when the gas concentration is still within the early warning range. Different levels of gas concentration anomalies trigger differentiated response levels, enabling ventilation enhancement measures to be initiated during the early warning stage, providing ample time for personnel evacuation and emergency response. The health diagnosis mechanism sets health thresholds for frequency drift, optical window attenuation rate, and blackbody background shift based on the full width at half maximum (FWHM) of the quartz tuning fork, the detector's dynamic range, and signal-to-noise ratio requirements. Combined with the attention coefficient, it classifies health states into three levels: good, attenuated, and deteriorated. This makes the performance degradation process of sensors due to dust mass load, optical window contamination, or environmental thermal radiation observable and predictable, issuing maintenance prompts before complete performance failure, thus transforming equipment management from passively responding to faults to proactively preventing maintenance. The composite decision-making mechanism combines concentration alarm levels and health status levels through a decision matrix. When sensor performance deteriorates, it automatically labels data with a reliability degradation flag and suggests manual verification. This avoids misjudging abnormal readings caused by sensor malfunctions as actual gas concentration anomalies, or missing actual gas concentration anomalies due to sensor failure. It eliminates safety hazards caused by the inability to distinguish between gas concentration changes and sensor performance degradation. The feedback loop of system degradation operation commands and noise variance updates allows the detection system to maintain effective operation even when some channels fail by dynamically adjusting the weights of damaged channels, ensuring the continuity and reliability of multi-parameter gas detection in coal mines.

[0159] Step S50 achieves deep integration of environmental safety monitoring and equipment status monitoring by constructing a dual-channel decision architecture that combines gas concentration-level alarm and sensor health diagnosis. This step not only implements differentiated graded responses based on multi-level safety thresholds, initiating intervention measures during the early warning stage to reserve an emergency window, but also uses interference characteristic parameters such as frequency drift and optical window attenuation rate to quantify and grade sensor performance degradation, transforming equipment management from passive maintenance to proactive preventative maintenance. Through composite decision generation using the decision matrix, the system can automatically identify and mark data with reduced reliability due to sensor deterioration, effectively avoiding the risk of false alarms or missed detections caused by sensor failure. Furthermore, once a detection channel is determined to be degraded, its noise variance is immediately forcibly increased to suppress its weight in the fusion calculation in step S40, ensuring that the system can maintain high reliability operation through dynamic weight adjustment even under complex operating conditions with partial channel failures, completely eliminating safety hazards caused by the inability to distinguish between gas concentration changes and sensor performance degradation.

[0160] Example 2

[0161] This embodiment, based on Embodiment 1, provides a multi-parameter gas detection anti-interference and noise suppression system with full optical coordination in underground coal mines, such as... Figure 5 As shown, it includes:

[0162] Signal acquisition module: used to deploy a full optical detection array in the area to be detected in the underground coal mine, acquire the original optical feature signals under multi-dimensional physical field disturbance in the underground coal mine based on the full optical detection array, construct a set of interference feature parameters, and simultaneously construct an environmental compensation vector;

[0163] Credibility weight module: Define detection channels based on the all-optical detection array, construct an error sensitivity coefficient matrix, calculate the real-time estimated noise variance of each detection channel based on the error sensitivity coefficient matrix and the set of interference feature parameters, and calculate the real-time credibility weight of each detection channel based on the real-time estimated noise variance.

[0164] Concentration Inversion Module: Used to invert the original optical feature signals and output the original inverted concentrations of each target gas;

[0165] Environmental compensation fusion module: Based on the environmental compensation vector and real-time confidence weight, the original inversion concentration of each target gas is processed, and the final fused concentration of each target gas is output.

[0166] Composite decision module: Outputs composite decision based on the final fusion concentration of each target gas and the set of interference characteristic parameters.

[0167] Furthermore, in the signal acquisition module, the original optical characteristic signals include the intensity of pure infrared absorption signals, acoustic resonance signals, and fluorescence lifetime signals;

[0168] The method for acquiring the intensity of the pure infrared absorption signal includes: controlling the NDIR infrared absorption detection module to acquire the blackbody background radiation reference value when the infrared light source is off, acquiring the dual-wavelength light intensity signal when the infrared light source is on, separating the light intensity of the signal channel and the light intensity of the reference channel from the dual-wavelength light intensity signal, and subtracting the blackbody background radiation reference value from the light intensity of the signal channel to obtain the intensity of the pure infrared absorption signal.

[0169] The method for acquiring the acoustic resonance signal includes:

[0170] The frequency sweep range is set based on the factory-set center frequency of the QEPAS photoacoustic resonance detection module. Within the frequency sweep range, a rapid frequency sweep excitation is performed and the amplitude-frequency response curve of the quartz tuning fork is detected. The frequency corresponding to the maximum amplitude point is extracted from the amplitude-frequency response curve as the real-time physical resonance frequency. The excitation frequency is locked at the real-time physical resonance frequency to collect the acoustic resonance signal and the amplitude of the acoustic resonance signal.

[0171] The method for acquiring the fluorescence lifetime signal includes: controlling the LED light source of the LDO fluorescence quenching detection module to sinusoidally modulate and excite the oxygen-sensitive fluorescent membrane at a preset modulation frequency to generate an excitation light signal; the oxygen-sensitive fluorescent membrane emits a fluorescence signal after being excited by the excitation light signal; measuring the phase delay of the fluorescence signal relative to the excitation light signal to obtain the fluorescence phase difference; simultaneously measuring the reference fluorescence lifetime; and combining the fluorescence phase difference and the reference fluorescence lifetime to form a fluorescence lifetime signal.

[0172] Furthermore, in the credibility weight module, the method for constructing the error sensitivity coefficient matrix includes: configuring frequency drift influence factor, light decay influence factor and thermal radiation influence factor for the frequency drift, optical window attenuation rate and blackbody background offset in the interference feature parameter set, respectively, to form an error sensitivity coefficient matrix;

[0173] The calculation method for the real-time estimated noise variance of each detection channel includes:

[0174] The frequency drift, optical window attenuation rate, and blackbody background shift are weighted and summed with their respective influence factors to calculate the noise variance increment.

[0175] The historical noise variance of each detection channel at the previous sampling time is obtained. The historical noise variance at the previous time is iteratively updated using the noise variance increment to obtain the real-time estimated noise variance of each detection channel at the current time.

[0176] Furthermore, in the concentration inversion module, the original inversion concentrations of each target gas include the infrared absorption inversion concentrations of methane and carbon dioxide, the photoacoustic inversion concentration of carbon monoxide, and the fluorescence inversion concentration of oxygen.

[0177] The method for obtaining the infrared absorption inversion concentrations of methane and carbon dioxide includes: substituting the light intensity of the signal channel and the light intensity of the reference channel into the Beer-Lambert law to perform absorbance inversion calculations to obtain the infrared absorption inversion concentrations of methane and carbon dioxide.

[0178] The method for obtaining the photoacoustic inversion concentration of carbon monoxide includes:

[0179] The acoustic gain function under the current state is calculated based on the real-time physical resonance frequency. The amplitude of the acoustic resonance signal is divided by the product of the acoustic gain function and the excitation light power to obtain the photoacoustic inversion concentration of carbon monoxide. The excitation light power is the emission power of the LED light source in the QEPAS photoacoustic resonance detection module.

[0180] The method for obtaining the fluorescence inversion concentration of oxygen includes:

[0181] The fluorescence lifetime is calculated based on the fluorescence phase difference and the preset modulation frequency. The reference fluorescence lifetime is then compensated for temperature drift using the ambient temperature in the environmental compensation vector. Based on the measured fluorescence lifetime and the temperature-compensated reference fluorescence lifetime, the fluorescence inversion concentration of oxygen is calculated.

[0182] The methods and systems of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated.

[0183] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0184] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-parameter gas detection anti-interference and noise suppression method with all-optical coordination in underground coal mines, characterized in that, The method includes: A fully optical detection array is deployed in the area to be detected underground in a coal mine. The array collects the original optical characteristic signals under multidimensional physical field disturbances in the coal mine, including the intensity of pure infrared absorption signals, acoustic resonance signals, and fluorescence lifetime signals. The fully optical detection array includes an NDIR infrared absorption detection module, a QEPAS photoacoustic resonance detection module, and an LDO fluorescence quenching detection module. The method for acquiring the intensity of the pure infrared absorption signal includes: controlling the NDIR infrared absorption detection module to acquire the blackbody background radiation reference value when the infrared light source is off, acquiring the dual-wavelength light intensity signal when the infrared light source is on, separating the light intensity of the signal channel and the light intensity of the reference channel from the dual-wavelength light intensity signal, and subtracting the blackbody background radiation reference value from the light intensity of the signal channel to obtain the intensity of the pure infrared absorption signal. The method for acquiring the acoustic resonance signal includes: setting a sweep frequency range based on the factory-set center frequency of the QEPAS photoacoustic resonance detection module; performing rapid sweep frequency excitation within the sweep frequency range and detecting the amplitude-frequency response curve of the quartz tuning fork; extracting the frequency corresponding to the maximum amplitude point from the amplitude-frequency response curve as the real-time physical resonance frequency; and locking the excitation frequency at the real-time physical resonance frequency to acquire the acoustic resonance signal and the amplitude of the acoustic resonance signal. The blackbody background radiation reference value is compared with the factory-calibrated reference value to obtain the blackbody background offset; the real-time physical resonance frequency is compared with the factory-set center frequency to obtain the frequency drift; the current transmitted light intensity of the reference beam passing through the air chamber window is measured, and the current transmitted light intensity is compared with the clean light window transmitted light intensity reference value to obtain the light window attenuation rate; the frequency drift, light window attenuation rate and blackbody background offset are combined to form a set of interference characteristic parameters, and an environmental compensation vector is constructed simultaneously. Based on the all-optical detection array, the detection channels are defined, the error sensitivity coefficient matrix is ​​constructed, and based on the error sensitivity coefficient matrix and the set of interference feature parameters, the real-time estimated noise variance of each detection channel is calculated. Based on the real-time estimated noise variance, the real-time confidence weight of each detection channel is calculated. The original optical feature signals are inverted to output the original inversion concentrations of each target gas; the target gases include methane, carbon monoxide, carbon dioxide, and oxygen; the original inversion concentrations of each target gas include the infrared absorption inversion concentrations of methane and carbon dioxide, the photoacoustic inversion concentration of carbon monoxide, and the fluorescence inversion concentration of oxygen; the method for obtaining the photoacoustic inversion concentration of carbon monoxide includes: calculating the acoustic gain function under the current state based on the real-time physical resonance frequency, dividing the amplitude of the acoustic resonance signal by the product of the acoustic gain function and the excitation light power to obtain the photoacoustic inversion concentration of carbon monoxide; The original inversion concentrations of each target gas are processed based on the environmental compensation vector and real-time confidence weight, and the final fused concentrations of each target gas are output. Based on the final fusion concentration of each target gas and the set of interference characteristic parameters, a composite decision is output.

2. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 1, characterized in that, The method for acquiring the fluorescence lifetime signal includes: The LED light source of the LDO fluorescence quenching detection module is controlled to sinusoidally modulate and excite the oxygen-sensitive fluorescent membrane at a preset modulation frequency to generate an excitation light signal. After being excited by the excitation light signal, the oxygen-sensitive fluorescent membrane emits a fluorescence signal. The phase delay of the fluorescence signal relative to the excitation light signal is measured to obtain the fluorescence phase difference. The reference fluorescence lifetime is measured simultaneously. The fluorescence phase difference and the reference fluorescence lifetime are combined to form a fluorescence lifetime signal.

3. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 2, characterized in that, The method for defining the detection channel is as follows: The NDIR infrared absorption detection module is the first detection channel, the QEPAS photoacoustic resonance detection module is the second detection channel, and the LDO fluorescence quenching detection module is the third detection channel. The environmental compensation vector includes ambient temperature, relative humidity, and atmospheric pressure.

4. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 3, characterized in that, The method for constructing the error sensitivity coefficient matrix includes: Frequency drift influence factor, optical decay influence factor and thermal radiation influence factor are respectively configured for the frequency drift, optical window attenuation rate and blackbody background offset in the interference feature parameter set to form an error sensitivity coefficient matrix; The calculation method for the real-time estimated noise variance of each detection channel includes: The frequency drift, optical window attenuation rate, and blackbody background shift are weighted and summed with their respective influence factors to calculate the noise variance increment. The historical noise variance of each detection channel at the previous sampling time is obtained. The historical noise variance at the previous time is iteratively updated using the noise variance increment to obtain the real-time estimated noise variance of each detection channel at the current time.

5. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 4, characterized in that, The method for obtaining the infrared absorption inversion concentrations of methane and carbon dioxide includes: substituting the light intensity of the signal channel and the light intensity of the reference channel into the Beer-Lambert law to perform absorbance inversion calculations to obtain the infrared absorption inversion concentrations of methane and carbon dioxide.

6. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 5, characterized in that, The excitation light power is the emission power of the LED light source in the QEPAS photoacoustic resonance detection module.

7. The method for anti-interference and noise suppression of multi-parameter gas detection with full optical coordination in coal mines according to claim 6, characterized in that, The method for obtaining the fluorescence inversion concentration of oxygen includes: The fluorescence lifetime is calculated and measured based on the fluorescence phase difference and the preset modulation frequency. The temperature-compensated reference fluorescence lifetime is obtained by using the ambient temperature in the environmental compensation vector to compensate for the temperature drift. The fluorescence inversion concentration of oxygen was calculated based on the measured fluorescence lifetime and the temperature-compensated reference fluorescence lifetime.

8. The method for multi-parameter gas detection with full optical coordination in coal mines to resist interference and suppress noise, as described in claim 7, is characterized in that... The method for generating the composite decision includes: Multiple safety thresholds are set for each target gas. The final fusion concentration of each target gas is compared with the corresponding multiple safety thresholds. The gas concentration alarm level is determined based on the comparison results. A health threshold is set for each interference feature parameter in the interference feature parameter set. Each interference feature parameter is compared with its corresponding health threshold, and the health status level of each detection channel is determined based on the comparison results. The gas concentration alarm level is logically combined with the health status level of each detection channel to generate a composite decision.

9. The method for multi-parameter gas detection with full optical coordination in coal mines to resist interference and suppress noise, as described in claim 8, is characterized in that... The method for processing the original inversion concentration of each target gas based on environmental compensation vector and real-time confidence weight includes: The environmental compensation vector is used to perform environmental factor regression compensation on the original inversion concentration of each target gas to obtain the environmental compensation concentration. The environmental compensation concentration is then weighted and fused using real-time reliability weights to output the final fused concentration of each target gas.

10. A multi-parameter gas detection anti-interference and noise suppression system with full optical coordination in coal mines, used to implement the multi-parameter gas detection anti-interference and noise suppression method with full optical coordination in coal mines as described in any one of claims 1-9, characterized in that, The system includes: Signal acquisition module: used to deploy a full optical detection array in the area to be detected in the underground coal mine, acquire the original optical feature signals under multi-dimensional physical field disturbance in the underground coal mine based on the full optical detection array, construct a set of interference feature parameters, and simultaneously construct an environmental compensation vector; Credibility weight module: Define detection channels based on the all-optical detection array, construct an error sensitivity coefficient matrix, calculate the real-time estimated noise variance of each detection channel based on the error sensitivity coefficient matrix and the set of interference feature parameters, and calculate the real-time credibility weight of each detection channel based on the real-time estimated noise variance. Concentration Inversion Module: Used to invert the original optical feature signals and output the original inverted concentrations of each target gas; Environmental compensation fusion module: Based on the environmental compensation vector and real-time confidence weight, the original inversion concentration of each target gas is processed, and the final fused concentration of each target gas is output. Composite decision module: Outputs composite decision based on the final fusion concentration of each target gas and the set of interference characteristic parameters.

Images