Nondestructive testing method and system for flame detector based on multi-physical field coupling and impedance spectrum analysis

Abstract

The application discloses a multi-physical field coupling and impedance spectrum analysis flame detector nondestructive testing method and system, and relates to the technical field of helicopter flame detector nondestructive testing.The temperature field and the electromagnetic field are coupled to replace light, so that a simulation environment of fire radiation effect is provided for the flame detector, and the photochemical reaction of the photoelectric material caused by light is avoided; the error is eliminated through wide frequency scanning and the four-electrode method, the impedance spectrum is obtained through inversion calculation, and the physical parameters are obtained through deconstruction, so that the problem that the continuous relaxation process information contained in the wide frequency data is lost in traditional impedance analysis is avoided; the classical physical parameters of the sensor fault are mapped into the Hamiltonian of the quantum system, the fault is sensed by solving the ground state, and the defects that the classical machine learning is easy to fall into local optimization when processing high-dimensional, nonlinear and strongly correlated fault features, and the model is poor in interpretability.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for helicopter flame detectors, specifically to a non-destructive testing method and system for flame detectors based on multi-physics coupling and impedance spectral analysis. Background Technology

[0002] Helicopter fire detectors are key components of aviation safety systems, used to monitor the fire status in critical areas such as the engine compartment and transmission compartment in real time. Among them, photosensitive flame detectors work based on the photoelectric effect principle. Their core sensing element (photodiode / transistor) receives infrared or ultraviolet radiation of a specific wavelength and converts the light signal into an electrical signal to identify and alarm on the flame.

[0003] Currently, the factory generally uses the mercury lamp illumination method for functional testing of this type of helicopter fire detector: specific wavelengths of light generated by a mercury lamp are used to simulate the real flame spectrum, stimulating the detector with light, and the detector's output signal is observed to determine whether its function is up to standard. While this method can visually verify the detector's photoelectric response characteristics, it has significant inherent limitations.

[0004] The spectral energy distribution of mercury lamp light sources differs from that of real flames, and the light intensity is difficult to control precisely. During repeated testing, photoelectric sensing elements are exposed to non-natural light radiation for extended periods, accelerating the light decay process of photodiodes / transistors and causing irreversible "photo-fatigue" effects on the sensor's photoelectric materials—that is, a continuous decrease in the quantum efficiency of the photosensitive materials, an increase in dark current, and a drift in spectral response characteristics. This photo-induced aging damage is cumulative and irreversible, directly shortening the service life of the detector.

[0005] A more prominent problem is the severe gap in the functional coverage of existing detection methods. Conventional methods such as resistance testing can only verify the continuity of the detector's internal circuitry, but cannot effectively assess the health status of the core photoelectric conversion function. In other words, even if the detector circuitry is intact, its photosensitive element may be on the verge of performance degradation or failure due to accumulated damage from historical testing, yet this is difficult to identify using traditional methods. This risk of "hidden failure" leads to detectors with degraded performance being mistakenly judged as qualified products for installation, or detectors that have not yet completely failed being prematurely scrapped due to the inability to accurately assess their performance.

[0006] The aforementioned technical contradictions directly led to a sharp increase in the scrap rate of photosensitive flame detectors, resulting not only in a large waste of spare parts and economic losses, but also in an increase in the frequency of helicopter maintenance and maintenance costs, which adversely affected the safety and economy of aviation operations. Summary of the Invention

[0007] Based on the problems mentioned above, the purpose of this invention is to provide a non-destructive testing method and system for flame detectors using multi-physics coupling and impedance spectral analysis. This solves the problem that repeated illumination testing with mercury lamps induces photo-fatigue effects in luminescent materials, leading to irreversible performance degradation, detection damage, and a vicious cycle of misjudgment and scrapping, resulting in a high detector replacement rate and significantly increasing the life-cycle maintenance cost of helicopters.

[0008] This invention is achieved through the following technical solution: The first aspect of this invention provides a non-destructive testing method for flame detectors based on multi-physics coupling and impedance spectral analysis, comprising the following steps: Step S1: Apply an electromagnetic field and a temperature field to the flame detector respectively, and couple the electromagnetic field and the temperature field to simulate the thermal radiation effect of a fire. Step S2: Under the fire thermal radiation effect, measure the impedance data of the flame detector, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis; Step S3: Extract quantum features from the state matrix to generate a fault feature vector; Step S4: Based on the hybrid adjudication architecture of convolutional quantum neural network and decision tree, perform fault diagnosis on the fault feature vector to obtain the fault status of the flame detector.

[0009] In the above technical solution, the temperature field and electromagnetic field are physically coupled to replace the light, thereby providing a simulated environment for the fire detector to simulate the thermal radiation effect of the fire. This avoids the irreversible photochemical reaction (photo-fatigue) that light (especially ultraviolet light) can cause to the photoelectric materials, and the difficulty in accurately controlling the light intensity and wavelength distribution inside the chip.

[0010] In a simulated fire thermal radiation effect environment, raw data from flame detectors are collected, and errors are eliminated using broadband scanning and the four-electrode method. Impedance spectra are obtained through inversion calculations, and the physical parameters deconstructed from them are used to construct a state matrix. The state matrix is ​​then used as input data for the next step of quantum feature extraction. Compared with traditional impedance analysis methods that only focus on impedance magnitude or phase at specific frequencies, this method preserves the continuous relaxation process information contained in broadband data.

[0011] The state matrix contains the physical parameters of the sensors in the flame detector. This method maps the state matrix to the Hamiltonian of the quantum system and encodes the physical parameters into quantum observations through quantum feature extraction.

[0012] Finally, the fault feature vector is input into a hybrid adjudication architecture of convolutional quantum neural network and decision tree for fault diagnosis. This solves the black box problem of quantum machine learning, achieves high-precision and physically interpretable final diagnosis, and solves the problems of pure neural network diagnosis results being difficult to explain with physical principles, engineering credibility, and the inability of pure rule systems to handle complex, concurrent, and unknown faults.

[0013] In one optional embodiment, an electromagnetic field and a temperature field are applied to the flame detector, and the electromagnetic field and the temperature field are coupled, including: An axial alternating magnetic field is applied using an axial alternating magnetic field generator, and an instantaneous temperature field is applied using a Peltier temperature control platform. The axial alternating magnetic field and the instantaneous temperature field are coupled to generate a multi-physics coupling environment. Calculate the carrier generation rate in the multiphysics coupling environment. Based on the carrier generation rate, the non-equilibrium carrier density distribution can be solved. The total carrier density distribution is determined by combining the carrier concentration and the non-equilibrium carrier density distribution. The conductivity and complex permittivity of the sensor material in the flame detector are determined by the total carrier density distribution.

[0014] In one optional embodiment, measuring the impedance data of the flame detector includes: The current terminals of the four-electrode impedance analyzer are connected to the current electrodes of the flame detector sensor via wires to form a turbulent current loop; and an orthogonal alternating current is injected into the turbulent current loop. The voltage terminals of the four-electrode impedance analyzer are connected to the voltage electrodes of the flame detector sensor via wires to construct a sensing voltage loop and measure the voltage drop. The impedance data is calculated using the orthogonal alternating current and the voltage drop.

[0015] In one optional embodiment, impedance analysis is performed on the impedance data, including: The complex capacitance is calculated using the impedance data, the complex capacitance is normalized to the complex permittivity, the absolute admittance is calculated based on the complex permittivity, and the real part of the admittance is extracted from the absolute admittance. The impedance data is decomposed into resistive and capacitive components. A Nyquist plot is drawn with the resistive component as the horizontal axis and the capacitive component as the vertical axis. The Nyquist plot is then fitted with a least-squares circle to obtain the Cole-Cole plot radius of curvature. Based on the impedance data, curves showing the changes in impedance amplitude and phase difference with frequency are plotted. The curves and the Nyquist plot together construct a broadband impedance spectrum. The broadband impedance spectrum is then inverted into a dielectric relaxation time distribution function. The dielectric relaxation time is determined by the dielectric relaxation time distribution plot.

[0016] In one optional embodiment, quantum feature extraction is performed on the state matrix, including: The parameters in the state matrix are normalized and scaled, and the normalized and scaled parameters are encoded into the energy operator to obtain the parameterized Hamiltonian. Construct a parameterized quantum circuit, and adjust the quantum gate parameters in the parameterized quantum circuit based on the parameterized Hamiltonian to obtain a trial quantum state; The expected energy value of the trial quantum state is measured, and the expected energy value is iteratively optimized using an optimizer to obtain a fault feature vector.

[0017] In one optional embodiment, fault diagnosis is performed on the fault feature vector based on a hybrid adjudication architecture of convolutional quantum neural networks and decision trees, including: The fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed to obtain a discriminative feature vector. The discriminative feature vector is input into the quantum pooling layer for quantum variable layering processing to obtain quantum features; The quantum features are measured to obtain the original fault probability distribution, and the original fault probability distribution is processed through a fully connected neural network layer to obtain a fault occurrence probability vector. The failure probability vector is verified using a decision tree to obtain the failure status of the flame detector.

[0018] In one optional embodiment, the fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed, including: A local feature map is obtained by sliding computation on the fault feature vector using a set of one-dimensional convolution kernels. The local feature map is nonlinearly activated using a nonlinear activation function, and then the nonlinearly activated local feature map is fed into a pooling layer for downsampling to obtain a discriminative feature vector.

[0019] In an optional embodiment, the discriminative feature vector is input into a quantum pooling layer for quantum variable layering processing, including: The discriminative feature vector is encoded onto a quantum state to obtain a quantum state feature vector; Quantum convolutional layers are used to perform convolution operations on adjacent qubits in the quantum state feature vector to extract quantum state correlation features; The quantum state correlation features are quantum pooled by a quantum pooling layer to generate quantum features.

[0020] In one alternative embodiment, measuring the quantum feature includes: Each qubit in the quantum feature is sampled multiple times under the computational basis to generate sampling results; The probability of a set of "0" and "1" bit strings appearing in the sampling result is measured.

[0021] A second aspect of the present invention provides a non-destructive testing system for flame detectors based on multi-physics coupling and impedance spectral analysis, comprising: A fire thermal radiation effect simulation module is used to apply electromagnetic fields and temperature fields to a flame detector, and couple the electromagnetic fields and temperature fields to simulate the fire thermal radiation effect. The impedance analysis module is used to measure the impedance data of the flame detector under the fire thermal radiation effect, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis. The fault feature extraction module is used to extract quantum features from the state matrix and generate a fault feature vector. The fault diagnosis module is used to diagnose the fault feature vector based on a hybrid decision architecture of convolutional quantum neural network and decision tree to obtain the fault status of the flame detector.

[0022] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. Physical fields through thermo-magnetic field coupling replace light, achieving non-destructive stimulation and avoiding irreversible photochemical reactions in optoelectronic materials caused by traditional light exposure methods. Furthermore, the intensity and wavelength distribution of light are difficult to control precisely within the chip. 2. By using wideband scanning and the four-electrode method, errors are eliminated, and the impedance spectrum is obtained through inversion calculation. The physical parameters deconstructed from it avoid the problem that traditional impedance analysis only focuses on the impedance magnitude or phase at a specific frequency, losing the information of the continuous relaxation process contained in the wideband data. 3. The classical physical parameters of sensor faults are mapped to the Hamiltonian of the quantum system. Faults are perceived by solving its ground state. This overcomes the shortcomings of classical machine learning, which is prone to getting trapped in local optima and has poor model interpretability when dealing with high-dimensional, nonlinear, and strongly correlated fault features. 4. By using the quantum-classical hybrid structure of convolutional quantum neural networks and physical rule decision tree verification, the "black box" problem of quantum machine learning is solved, and high-precision and physically interpretable final diagnosis is achieved. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating the non-destructive testing method for flame detectors based on multi-physics coupling and impedance spectrum analysis provided in Embodiment 1 of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0025] Embodiment 1 of this invention provides a non-destructive testing method for flame detectors based on multi-physics coupling and impedance spectrum analysis. This method relates to related technical fields such as smart sensors and magnetic sensors. Figure 1 As shown, the non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectroscopy analysis includes the following steps: Step S1: Apply an electromagnetic field and a temperature field to the flame detector respectively, and couple the electromagnetic field and the temperature field to simulate the thermal radiation effect of a fire. Step S2: Under the fire thermal radiation effect, measure the impedance data of the flame detector, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis; Step S3: Extract quantum features from the state matrix to generate a fault feature vector; Step S4: Based on the hybrid adjudication architecture of convolutional quantum neural network and decision tree, perform fault diagnosis on the fault feature vector to obtain the fault status of the flame detector.

[0026] It should be noted that, compared with the traditional mercury lamp light source irradiation method, this method uses the physical coupling of temperature field and electromagnetic field to replace light irradiation, thereby providing a simulated environment of fire thermal radiation effect for flame detectors. This avoids the irreversible photochemical reaction (photo-fatigue) that light irradiation (especially ultraviolet light) can cause to photoelectric materials, and the difficulty in precisely controlling the light intensity and wavelength distribution inside the chip.

[0027] In a simulated fire thermal radiation effect environment, raw data from flame detectors are collected, and errors are eliminated using broadband scanning and the four-electrode method. Impedance spectra are obtained through inversion calculations, and the physical parameters deconstructed from them are used to construct a state matrix. The state matrix is ​​then used as input data for the next step of quantum feature extraction. Compared with traditional impedance analysis methods that only focus on impedance magnitude or phase at specific frequencies, this method preserves the continuous relaxation process information contained in broadband data.

[0028] The state matrix contains the physical parameters of the sensors in the flame detector. This method maps the state matrix to the Hamiltonian of the quantum system and encodes the physical parameters into quantum observations through quantum feature extraction.

[0029] Finally, the fault feature vector is input into a hybrid adjudication architecture of convolutional quantum neural network and decision tree for fault diagnosis. This solves the black box problem of quantum machine learning, achieves high-precision and physically interpretable final diagnosis, and solves the problems of pure neural network diagnosis results being difficult to explain with physical principles, engineering credibility, and the inability of pure rule systems to handle complex, concurrent, and unknown faults.

[0030] In one optional embodiment, an electromagnetic field and a temperature field are applied to the flame detector, and the electromagnetic field and the temperature field are coupled, including: An axial alternating magnetic field is applied using an axial alternating magnetic field generator, and an instantaneous temperature field is applied using a Peltier temperature control platform. The axial alternating magnetic field and the instantaneous temperature field are coupled to generate a multi-physics coupling environment. Calculate the carrier generation rate in the multiphysics coupling environment. Based on the carrier generation rate, the non-equilibrium carrier density distribution can be solved. The total carrier density distribution is determined by combining the carrier concentration and the non-equilibrium carrier density distribution. The conductivity and complex permittivity of the sensor material in the flame detector are determined by the total carrier density distribution.

[0031] In this step, a multi-physics coupling environment is first constructed, simultaneously establishing the coupling effect of electromagnetic and temperature fields within the photoelectric sensing element of the flame detector. The electromagnetic field is excited by an alternating electrical signal of a specific frequency to simulate the photon energy coupling effect during photoelectric conversion; the temperature field is established through controllable thermal excitation, reflecting the thermal load characteristics of the detector's operating environment. The two physical fields work together to drive the carrier excitation process within the semiconductor material, providing a physical excitation basis for subsequent electrical characterization.

[0032] For flame detectors, an axial alternating magnetic field generator (also known as a multi-frequency eddy current generator) is used to apply an axial alternating magnetic field of 0.1-5 mT and 0.1-10 MHz. The axial alternating magnetic field is shown below: in, : A magnetic field vector that varies with time; Peak amplitude of magnetic field, with an axial alternating magnetic field generator applying a magnetic field of 0.1mT-5mT; Angular frequency, determined by the frequency of the axial alternating magnetic field generator. Decision, that is (frequency f) (Range from 0.1Hz to 10MHz). :time; The z-axis is a unit vector, and the magnetic field lines are parallel to the normal direction of the flame detector sensor.

[0033] Simultaneously, using a Peltier temperature control platform, a ±0.1℃ perturbation was applied to obtain the transient temperature field as T(r, t). ), r: position vector.

[0034] The changing magnetic field induces a vortex electric field inside the sensor of the flame detector. That is, Faraday's electromagnetic induction: ; Where r: position vector; : Vorticity; : Vortex electric field; Alternating magnetic field.

[0035] Temperature perturbations in the transient temperature field can alter the intrinsic carrier concentration of a material. and conductivity The intrinsic carrier concentration formula is as follows: ; in, The density of quantum states near the bottom of the conduction band that can be occupied by electrons and holes; The density of quantum states available for electrons and holes near the top of the valence band; The bandgap width of the flame detector sensor; Boltzmann constant; The Arrhenius relation reflects the need for charge carriers to cross the bandgap. Only then can it be thermally activated.

[0036] It's important to note that the photosensitive sensor in a flame detector is essentially a semiconductor device, and its working principle can be visualized using an "electronic parking lot" model. This model abstracts the semiconductor material as a multi-layered parking lot specifically designed to house the microscopic particle "electron." In this parking lot, the valence band top corresponds to the highest level filled with electrons, with all spaces below this level (i.e., the valence band) completely occupied by electrons. The conduction band bottom corresponds to the lowest level completely empty, with no electrons distributed in this level and above (i.e., the conduction band), providing space for electrons to move freely. Between the valence band top and the conduction band bottom lies an energy gap that prevents electrons from residing; this gap is called the band gap, and its energy scale is the "band gap width."

[0037] In actual physics, due to the continuous presence of thermal disturbances (similar to the impact of high summer temperatures on a parking lot), some electrons at the top of the valence band have the opportunity to gain enough energy to cross the band gap and transition from the top of the valence band to the bottom of the conduction band. This transition process produces two important results: on the one hand, a new freely moving electron is added at the bottom of the conduction band; on the other hand, a freely moving vacancy is left at the top of the valence band, which is called a "hole". Electrons and holes are generated in pairs, together constituting the "intrinsic carriers" of a semiconductor. The concentration of intrinsic carriers is closely related to temperature, following the thermal excitation law—the higher the temperature, the greater the average thermal energy gained by electrons, the more electrons successfully cross the band gap, and the higher the concentration of intrinsic carriers. This physical mechanism constitutes the microscopic basis of the photoelectric conversion function of photosensors and also provides a theoretical basis for subsequent non-destructive testing through temperature field control.

[0038] The formula for conductivity is as follows: ; in, Elementary charge; Carrier mobility; : Carrier concentration.

[0039] The temperature field first rapidly modulates the conductivity of the flame detector sensor. Subsequently, the alternating magnetic field induces an eddy current electric field within this modulated conductivity. Therefore, the eddy current effect is dynamically modulated by temperature, and the conductivity σ is a function of temperature T. This means that the temperature field, by changing the conductivity, directly alters the diffusion and distribution of the electromagnetic field within the flame detector sensor. The two processes are intertwined in real time and dynamically.

[0040] Specifically, the electromagnetic field governing equations mainly describe how an alternating magnetic field induces an eddy current electric field, that is, they describe the electric field generated at an angular frequency of ω. external current density The distribution of the magnetic vector potential A (whose curl is the magnetic field) under excitation is given by the following formula: ; The equivalent equation is: ; Its coupling term is conductivity. .

[0041] The temperature field governing equations primarily describe the diffusion of temperature in space and the sources of heat, as well as the material density. Specific heat capacity and thermal conductivity Given a constant temperature T, the formula for the change in temperature T with time t is as follows: ; Among them, heat source item (Time derivative) (Frequency domain representation).

[0042] Its coupling term is the heat source term. The Joule heat generated by the electromagnetic field is directly determined by the solution of the electromagnetic field.

[0043] The key link between electromagnetic fields and temperature fields is conductivity. ,Will Substituting the expression into the temperature field control equation, and... Substituting the relationships into the electromagnetic field equations, we obtain a set of interdependent equations: ; Magnetic potential A and temperature T are variables in each other's equations, through Connect with |A|².

[0044] The differences between this coupling method and the traditional coupling method are shown in the table below:

[0045] The table above compares the core differences between this method and traditional techniques in three application areas: induction heat treatment, thermo-electromagnetic nondestructive testing, and semiconductor parameter testing. It reveals the essential innovation of this method in terms of physical mechanism and application goals. In the field of induction heat treatment, traditional methods use the macroscopic high-temperature effect of eddy currents to achieve overall heating of the specimen, with thermal processing as the core objective. The electromagnetic field is merely a means of heat generation, while the temperature field is the final output. This method, however, induces micro-carriers through a magnetic field, using the temperature field as a perturbation modulation parameter with an accuracy of ±0.1℃. Using physical field excitation as a means, it achieves precise control of the energy scale and mechanism of action. The core is that the temperature field is transformed from a "processing tool" to a "modulation tool." In the field of thermo-electromagnetic nondestructive testing, traditional methods use the thermal and electromagnetic fields independently or sequentially, with loose coupling. The thermal field is the passive detection object, while the electromagnetic field is the active detection means. This method achieves real-time synchronous deep fusion of the two fields, using them together as a joint excitation source to generate a composite electrical response for diagnosis. The core is that the temperature field is transformed from a detection object to a detection means. In the field of semiconductor parameter testing, traditional methods treat temperature as an independent variable... To obtain the temperature dependence of parameters, the electrical characteristics of devices are statically calibrated under different isothermal conditions. This method uses temperature as a dynamic modulation variable, working in real-time with an alternating magnetic field to dynamically change the electromagnetic parameters of the material over a wide frequency range, thus realizing a dynamic parameter probe function. The core is to overcome the limitations of isothermal testing and achieve real-time interactive modulation of temperature and electromagnetic parameters. These innovations in physical mechanisms and application goals provide unique technical advantages and solutions for non-destructive testing of flame detectors: This method draws on the energy control concept in the field of induction heat treatment, limiting the electromagnetic field energy to the level of carrier excitation in a micro-region, and modulating the temperature field with a precision of ±0.1℃, avoiding direct photon irradiation, fundamentally solving the photo-fatigue effect caused by traditional mercury lamp testing, and achieving truly non-destructive testing; it introduces a real-time synchronous coupling mechanism of heat and electromagnetic, turning the temperature field into an excitation means, inducing carriers through the electromagnetic field and modulating the carrier concentration through the temperature field, reproducing the photo-electric conversion physical process, and realizing the equivalent evaluation of the core photoelectric conversion function, breaking through the limitation of traditional resistance testing that can only verify the continuity of the circuit; it uses temperature as a dynamic modulation variable, working in real-time with an alternating magnetic field, and dynamically changing the electromagnetic parameters of the material over a wide frequency range, thus realizing a dynamic parameter probe function. The method utilizes a real-time coordinated MHz broadband alternating magnetic field to extract microscopic physical parameters through impedance spectroscopy analysis, constructing a dynamic-broadband diagnostic system to achieve quantitative assessment of the detector's health status and provide data support for predictive maintenance and lifespan assessment. Combining these three aspects, this method forms a novel detection paradigm, replacing traditional optical excitation with precise electromagnetic-temperature field coupling, static electrical testing with dynamic impedance spectroscopy analysis, and macroscopic functional judgment with quantitative characterization of microscopic charge carriers. This breakthrough overcomes existing technological bottlenecks and provides an effective technical approach to reduce the scrap rate and maintenance costs of helicopter fire alarm detectors.

[0046] vortex electric field When work is done on charge carriers, their power density is converted into the non-equilibrium charge carrier generation rate: ; The generated charge carriers will diffuse and recombine, as described by the following equations: ; in, : Carrier generation rate generated by eddy current electric field; Carrier generation rate is generated by temperature perturbation; Non-equilibrium carrier concentration; D: Carrier lifetime, a key parameter (performance indicator) for determining sensor aging; D: Carrier diffusion coefficient. The degree of non-uniformity in carrier concentration The increment generated by the eddy current effect; The increment caused by thermal disturbance.

[0047] The above equations are used to solve for the two non-equilibrium increments generated by the excitation, which are spatially non-uniform. and Among them, two independent increments and This is determined by the fundamental difference between the two physical mechanisms. The core reason is that the physical sources and mathematical descriptions of the two excitation mechanisms are completely different. That is, using a "magnetic-thermal field" to replace "light" actually involves simultaneously activating two completely different physical processes to simulate them in a coordinated manner.

[0048] The carrier density distribution can instantly change the conductivity and complex permittivity of the flame detector sensor material. The formula for calculating the complex permittivity is as follows: ; in, High-frequency dielectric constant; Imaginary unit; represents a 90° phase lag in the complex plane; Electrical conductivity; Angular frequency; Vacuum permittivity; : Term contributed by free carriers.

[0049] An axial alternating magnetic field and a perturbation temperature field are simultaneously applied to the flame detector. The electromagnetic-thermal multiphysics coupling simulates the thermal radiation effect of a fire, generating a carrier density distribution inside the sensor that is equivalent to that of illumination, thereby avoiding optical fatigue damage caused by traditional illumination. This carrier distribution state instantly changes the conductivity and complex permittivity of the sensor material, thereby modulating the electrical response signal collected by the four-electrode impedance analyzer to achieve non-destructive testing.

[0050] In one optional embodiment, measuring the impedance data of the flame detector includes: The current terminals of the four-electrode impedance analyzer are connected to the current electrodes of the flame detector sensor via wires to form a turbulent current loop; and an orthogonal alternating current is injected into the turbulent current loop. The voltage terminals of the four-electrode impedance analyzer are connected to the voltage electrodes of the flame detector sensor via wires to construct a sensing voltage loop and measure the voltage drop. The impedance data is calculated using the orthogonal alternating current and the voltage drop.

[0051] It is important to note that conductivity and dielectric constant are two core physical parameters describing the response characteristics of materials (sensor chips) to an applied electric field. This response manifests macroscopically as the generation of current. When a material is placed in an electric field, two types of microscopic current mechanisms are simultaneously excited within it: one is conduction current, originating from the directional movement of free charges under the influence of the electric field, following Ohm's law in a microscopic form; the other is displacement current, originating from the polarization response of bound charges within the dielectric caused by the change of the electric field over time, determined by the time rate of change of the electric displacement vector. Under the influence of an alternating electric field, the above two types of currents superimpose to form the total current density, which can be further combined into the equivalent conductivity in complex form. Thus, under alternating field conditions, the conductivity and dielectric properties of a material are coupled together through a complex form to determine the total current response, where the real part represents energy dissipation and the imaginary part represents energy storage. The ratio of their proportions determines the phase relationship between the current and the voltage—when the conduction mechanism dominates, the current and voltage are in phase; when the polarization mechanism dominates, the current leads the voltage; the actual response lies somewhere in between.

[0052] Based on the above principles, in step S1, under the simulated fire radiation effect, this method uses a four-electrode impedance analyzer to collect impedance data. First, the I+ and I- terminals of the four-electrode impedance analyzer are connected to the two current electrodes (I+, I-) of the flame detector sensor via wires to form an excitation current loop. Then, a known, constant-amplitude small sinusoidal current is injected into this excitation current loop through the current electrodes. .

[0053] Simultaneously, the V+ and V- terminals of the four-electrode impedance analyzer are connected to the other two independent voltage electrodes (V+, V-) of the flame detector sensor via wires, forming a sensing voltage loop with a small sinusoidal current. A complex electric field distribution is generated inside the device. Its distribution is entirely determined by the geometry of the device and the different internal structures. The decision (i.e., the boundary value problem combining Maxwell's equations with the material distribution) is made. At this point, the voltage electrode measures the potential difference (voltage drop) along a specific path. : ; The impedance data is calculated using orthogonal alternating current and voltage drop: ; Impedance data is the complex ratio of these two.

[0054] When the magnetic-thermal field alters the carrier density at various points within the sensor, it directly changes the conductivity and dielectric constant at those points. Therefore, with the same injection current... This will generate an electric field distribution inside the sensor that differs from that in a healthy state, thus affecting the measured voltage. A change has occurred. We analyze the impedance of this change. This leads to the deduction of changes within the material.

[0055] In one optional embodiment, impedance analysis is performed on the impedance data, including: The complex capacitance is calculated using the impedance data, the complex capacitance is normalized to the complex permittivity, the absolute admittance is calculated based on the complex permittivity, and the real part of the admittance is extracted from the absolute admittance. The impedance data is decomposed into resistive and capacitive components. A Nyquist plot is drawn with the resistive component as the horizontal axis and the capacitive component as the vertical axis. The Nyquist plot is then fitted with a least-squares circle to obtain the Cole-Cole plot radius of curvature. Based on the impedance data, curves showing the changes in impedance amplitude and phase difference with frequency are plotted. The curves and the Nyquist plot together construct a broadband impedance spectrum. The broadband impedance spectrum is then inverted into a dielectric relaxation time distribution function. The dielectric relaxation time is determined by the dielectric relaxation time distribution plot.

[0056] It should be noted that at each scanning angular frequency At the point, the four-electrode impedance analyzer directly obtains a pair of impedance data. : Impedance magnitude: ; Phase difference: ; Equivalent representation of complex impedance: ; in, : Real part (resistive component); Imaginary part (capacitive component); Voltage phase; Current phase, : Imaginary number.

[0057] Calculate the complex capacitance using impedance data. : ; in, .

[0058] Normalize the complex capacitance to the complex permittivity. : ; Among them, vacuum capacitor : Vacuum dielectric constant; A: Electrode area; d: Material thickness.

[0059] Calculation of absolute admittance based on the permittivity : ; Real Admittance .

[0060] Obtaining impedance data Then, it can be drawn into a... The horizontal axis is... A curve plot with the vertical axis (i.e., a Nyquist plot).

[0061] Using least squares circle fitting, we find the center coordinates (a, b) and radius R for each data point. The squared algebraic distance of the fitted circle : ; Least squares circular fitting (i.e., minimizing the cost function S): ; in, : Refers to the total number of discrete data points used for fitting.

[0062] By solving The optimal solution can be obtained, yielding the center coordinates (a, b) and radius R. According to the definition of differential geometry, the radius of curvature at a point on an arc is numerically equal to the radius of the best-fit circle at that point. For a capacitive arc on a Cole-Cole diagram, its maximum radius of curvature... This is the R obtained from the fitting. That is: .

[0063] Plot impedance magnitude based on impedance data Phase difference The frequency-varying curve and the Nyquist plot together constitute a broadband impedance spectrum. The broadband impedance spectrum is then inverted into a dielectric relaxation time distribution function. ; in, At angular frequency The complex impedance measured below; Resistance at infinite frequency. The intensity of the relaxation process; : Dielectric relaxation time distribution function; : Imaginary number.

[0064] Dielectric relaxation time distribution function Main peak position As flame detector sensors age, defects increase and carrier lifetime decreases. It will shorten significantly, leading to Moving in the direction of shorter time.

[0065] Obtaining the temperature coefficient of impedance Together with the real part of the admittance, the radius of curvature of the Cole-Cole diagram, and the dielectric relaxation time, a state matrix is ​​constructed: .

[0066] In one optional embodiment, quantum feature extraction is performed on the state matrix, including: The parameters in the state matrix are normalized and scaled, and the normalized and scaled parameters are encoded into the energy operator to obtain the parameterized Hamiltonian. Construct a parameterized quantum circuit, and adjust the quantum gate parameters in the parameterized quantum circuit based on the parameterized Hamiltonian to obtain a trial quantum state; The expected energy value of the trial quantum state is measured, and the expected energy value is iteratively optimized using an optimizer to obtain a fault feature vector.

[0067] It should be noted that the state matrix can be constructed as a classical field vector. ; in, The logarithm of carrier lifetime indicates that during performance degradation, defects trap carriers. Decrease; p2: The radius of curvature in the Cole-Cole diagram reflects the microscopic uniformity of the material; p3: Temperature coefficient of impedance. Sensitive to thermal defects; p4: Relaxation strength (real part amplitude of admittance); p5: The relaxation symmetry factor normalizes and scales these parameters so that their range is in the interval [-1, 1].

[0068] The normalized and scaled parameters are encoded into the energy operator of the quantum system, thus constructing the parameterized Hamiltonian H(P): ; in: This is a Pauli operator term.

[0069] The interpretation of Pauli operator terms from the perspective of quantum measurement is shown in the table below:

[0070] Each Pauli operator term (e.g.) This represents a specific quantum "measurement perspective," whose weight pn is determined by the magnitude of the corresponding classical physical parameters. For example, the larger p1 (lifetime) is, the greater the value of the Hamiltonian. The greater the weight of a term, the more the quantum system tends to perceive faults from the dimension of "lifetime decay".

[0071] Construct a parameterized quantum circuit consisting of a series of tunable parameters. It is composed of quantum gates (such as rotation gates and controlled NOT gates). Through continuous adjustment... , to test quantum state The goal is to get as close as possible to the ground state (lowest energy state) described by the Hamiltonian H(P). This optimization process is the process by which a quantum system learns and understands the fault characteristics in a flame detector.

[0072] The prepared test quantum state Prepared multiple times (e.g., 8192 times) on a physical quantum processor. Measure each term of the Hamiltonian H(P). For example, for The term can be obtained by simply measuring the first qubit under the computational basis and statistically analyzing the probability difference between 0 and 1. The estimated value is obtained by weighted summation of the measurements of all terms, yielding the expected energy value in the current trial quantum state. : ; in, : Variables used for iterating over formulas.

[0073] Quantum measurement This is passed to a classic optimizer (such as the gradient descent variant COBYLA). The optimizer then... Based on previous records, a new set of potentially better parameters is calculated. .Will Send it back to the quantum processor to begin the next iteration.

[0074] The above process is repeated until... It stops decreasing and reaches convergence. At this point, we have: ; in, The ground state energy of H(P); : The ground state of H(P).

[0075] Finally, the fault feature vector is obtained. : ; Each component of this vector is a quantum observable, directly representing the probability of a specific, potential microscopic failure mode, namely: the ground state energy. Overall fault severity. The lower the value (the more negative), the further the system deviates from a healthy state; f1 can be used to represent this. Bandgap / level defects. A value close to -1 indicates severe bandgap distortion or deep level defect traps, primarily associated with p1 (lifetime). (f2 can be used to refer to it because deep level traps drastically shorten carrier lifetime); : Lattice disorder / interface state. A value deviating from 0 indicates that the lattice periodicity is disrupted or the interface state density is increased. It is mainly associated with p2 (homogeneity Γ) and p5 (symmetry), reflecting structural disorder, and can be referred to by f3. —Carrier mobility / scattering center. An abnormal value indicates that carrier transport is hindered and the mobility decreases. It is mainly related to p3 (temperature coefficient) and p4 (relaxation intensity). It is sensitive to impurity scattering and can be referred to as f4.

[0076] In one optional embodiment, fault diagnosis is performed on the fault feature vector based on a hybrid adjudication architecture of convolutional quantum neural networks and decision trees, including: The fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed to obtain a discriminative feature vector. The discriminative feature vector is input into the quantum pooling layer for quantum variable layering processing to obtain quantum features; The quantum features are measured to obtain the original fault probability distribution, and the original fault probability distribution is processed through a fully connected neural network layer to obtain a fault occurrence probability vector. The failure probability vector is verified using a decision tree to obtain the failure status of the flame detector.

[0077] It should be noted that the present invention uses a quantum computing coprocessor to process the fault characteristics.

[0078] In one optional embodiment, the fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed, including: A local feature map is obtained by sliding computation on the fault feature vector using a set of one-dimensional convolution kernels. The local feature map is nonlinearly activated using a nonlinear activation function, and then the nonlinearly activated local feature map is fed into a pooling layer for downsampling to obtain a discriminative feature vector.

[0079] It should be noted that, although These are already abstract features, but classic one-dimensional convolutional layers can further explore the potential local correlations and sequence patterns among these four components. For example, the simultaneous anomalies of f2 (bandgap defect) and f4 (decrease in mobility) may point to a specific composite failure mode.

[0080] Therefore, in this method, a set of small one-dimensional convolutional kernels is used on the fault feature vector. The algorithm performs sliding computation, passes through a non-linear activation function (such as ReLU), and then downsamples the data using a pooling layer (such as MaxPooling) to output a new, more discriminative feature vector. .

[0081] In an optional embodiment, the discriminative feature vector is input into a quantum pooling layer for quantum variable layering processing, including: The discriminative feature vector is encoded onto a quantum state to obtain a quantum state feature vector; Quantum convolutional layers are used to perform convolution operations on adjacent qubits in the quantum state feature vector to extract quantum state correlation features; The quantum state correlation features are quantum pooled by a quantum pooling layer to generate quantum features.

[0082] It should be noted that the preprocessed features Sent into the quantum circuit, Encode it into a quantum state. Specifically, encode each eigenvalue. This is mapped to the rotation angle of a qubit. After encoding, an n-qubit system can simultaneously represent... The superposition of these states maps fault information onto an exponentially large quantum feature space.

[0083] Then, quantum convolutional layers are used to convolve adjacent qubits in the quantum state feature vector. Each quantum convolutional layer consists of a set of parameterized quantum gates acting on adjacent qubits. The quantum convolution operation establishes coherence and entanglement between qubits through controlled rotation gates and controlled NOT gates, encoding classical fault features at the quantum state level for mixing, thereby extracting deep, nonlinear intrinsic correlations between feature components. This correlation, leveraging quantum parallelism and entanglement nonlocality, can capture high-order complex patterns that are difficult to model using classical computation, providing more accurate feature representation capabilities for flame detector fault diagnosis. For the impedance spectrum and temperature coefficient characteristics of flame detectors, the deep correlation extraction capability established by quantum convolution through quantum entanglement holds promise for discovering hidden high-order fault fingerprints from these seemingly independent feature components, achieving earlier and more accurate fault warnings.

[0084] Quantum pooling achieves dimensionality reduction by measuring a subset of qubits and utilizing measurement collapse properties: first, a subset of qubits is measured, causing their quantum states to collapse into classical states; then, based on this measurement result, controlled unitary operations are applied to the remaining unmeasured qubits, conditionally manipulating their quantum states. This process is similar to the downsampling function of classical pooling, but by leveraging the intrinsic randomness and collapse effect of quantum measurement, it can reduce the number of qubits while retaining the most significant quantum characteristic information, thereby effectively reducing the complexity and measurement overhead of subsequent quantum circuits.

[0085] In one alternative embodiment, measuring the quantum feature includes: Each qubit in the quantum feature is sampled multiple times under the computational basis to generate sampling results; The probability of a set of "0" and "1" bit strings appearing in the sampling result is measured.

[0086] The final quantum state output by the quantum pooling layer is measured. Typically, each qubit is sampled multiple times under the Z-basis (computation basis). The probability of the measurement result (a string of bits with "0" and "1") constitutes the original fault probability distribution. For example, the probability of a 4-qubit system outputting 1010 represents the likelihood of a certain combination of faults.

[0087] The probability distribution obtained from quantum measurements is input into a classic fully connected neural network layer and normalized using the Softmax function, ultimately outputting a 23-dimensional probability vector P=[p1,p2,...,p23]. Each pi corresponds to the probability of one of the 23 defined fault types occurring.

[0088] To avoid misjudgments caused by quantum noise or model uncertainty, the final diagnostic result is not simply selected based on the highest probability value. Instead, a three-dimensional fault decision tree based on physical rules is introduced for verification. (1) Threshold filtering: Only fault categories with a probability pi exceeding a preset high threshold (e.g., 85%) will be initially confirmed.

[0089] (2) Rule verification: Check whether the physical characteristics of high-probability faults match the fault feature vector. The observations must be consistent. For example, if the system has a high probability of indicating an "interface state density anomaly" (fault class #2), then it is necessary to check... Does f3 (lattice disorder observation) also exhibit significant anomalies?

[0090] (3) Conflict resolution: If multiple fault probabilities are similar, then rules based on fault tree analysis and historical data are activated to select the most likely cause.

[0091] Embodiment 2 of the present invention provides a non-destructive testing system for flame detectors based on multi-physics coupling and impedance spectrum analysis, comprising: A fire thermal radiation effect simulation module is used to apply electromagnetic fields and temperature fields to a flame detector, and couple the electromagnetic fields and temperature fields to simulate the fire thermal radiation effect. The impedance analysis module is used to measure the impedance data of the flame detector under the fire thermal radiation effect, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis. The fault feature extraction module is used to extract quantum features from the state matrix and generate a fault feature vector. The fault diagnosis module is used to diagnose the fault feature vector based on a hybrid decision architecture of convolutional quantum neural network and decision tree to obtain the fault status of the flame detector.

[0092] 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 description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., 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 non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectroscopy analysis, characterized in that, Includes the following steps: Step S1: Apply an electromagnetic field and a temperature field to the flame detector respectively, and couple the electromagnetic field and the temperature field to simulate the thermal radiation effect of a fire. Step S2: Under the fire thermal radiation effect, measure the impedance data of the flame detector, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis; Step S3: Extract quantum features from the state matrix to generate a fault feature vector; Step S4: Based on the hybrid adjudication architecture of convolutional quantum neural network and decision tree, perform fault diagnosis on the fault feature vector to obtain the fault status of the flame detector.

2. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 1, characterized in that, Applying an electromagnetic field and a temperature field to the flame detector, and coupling the electromagnetic field and the temperature field, includes: An axial alternating magnetic field is applied using an axial alternating magnetic field generator, and an instantaneous temperature field is applied using a Peltier temperature control platform. The axial alternating magnetic field and the instantaneous temperature field are coupled to generate a multi-physics coupling environment. Calculate the carrier generation rate in the multiphysics coupling environment. Based on the carrier generation rate, the non-equilibrium carrier density distribution can be solved. The total carrier density distribution is determined by combining the carrier concentration and the non-equilibrium carrier density distribution. The conductivity and complex permittivity of the sensor material in the flame detector are determined by the total carrier density distribution.

3. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 1, characterized in that, The impedance data of the flame detector is measured, including: The impedance data of the flame detector is measured, including: The current terminals of the four-electrode impedance analyzer are connected to the current electrodes of the flame detector sensor via wires to form a turbulent current loop; and an orthogonal alternating current is injected into the turbulent current loop. The voltage terminals of the four-electrode impedance analyzer are connected to the voltage electrodes of the flame detector sensor via wires to construct a sensing voltage loop and measure the voltage drop. The impedance data is calculated using the orthogonal alternating current and the voltage drop.

4. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 3, characterized in that, Impedance analysis is performed on the impedance data, including: The complex capacitance is calculated using the impedance data, the complex capacitance is normalized to the complex permittivity, the absolute admittance is calculated based on the complex permittivity, and the real part of the admittance is extracted from the absolute admittance. The impedance data is decomposed into resistive and capacitive components. A Nyquist plot is drawn with the resistive component as the horizontal axis and the capacitive component as the vertical axis. The Nyquist plot is then fitted with a least-squares circle to obtain the Cole-Cole plot radius of curvature. Based on the impedance data, curves showing the changes in impedance amplitude and phase difference with frequency are plotted. The curves and the Nyquist plot together construct a broadband impedance spectrum. The broadband impedance spectrum is then inverted into a dielectric relaxation time distribution function. The dielectric relaxation time is determined by the dielectric relaxation time distribution plot.

5. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 1, characterized in that, Quantum feature extraction of the state matrix includes: The parameters in the state matrix are normalized and scaled, and the normalized and scaled parameters are encoded into the energy operator to obtain the parameterized Hamiltonian. Construct a parameterized quantum circuit, and adjust the quantum gate parameters in the parameterized quantum circuit based on the parameterized Hamiltonian to obtain a trial quantum state; The expected energy value of the trial quantum state is measured, and the expected energy value is iteratively optimized using an optimizer to obtain a fault feature vector.

6. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 1, characterized in that, Based on a hybrid adjudication architecture combining convolutional quantum neural networks and decision trees, fault diagnosis is performed on the fault feature vector, including: The fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed to obtain a discriminative feature vector. The discriminative feature vector is input into the quantum pooling layer for quantum variable layering processing to obtain quantum features; The quantum features are measured to obtain the original fault probability distribution, and the original fault probability distribution is processed through a fully connected neural network layer to obtain a fault occurrence probability vector. The failure probability vector is verified using a decision tree to obtain the failure status of the flame detector.

7. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 6, characterized in that, The fault feature vector is input into a convolutional quantum neural network, and the fault feature vector is preprocessed, including: A local feature map is obtained by sliding computation on the fault feature vector using a set of one-dimensional convolution kernels. The local feature map is nonlinearly activated using a nonlinear activation function, and then the nonlinearly activated local feature map is fed into a pooling layer for downsampling to obtain a discriminative feature vector.

8. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 6, characterized in that, The discriminative feature vector is input into the quantum pooling layer for quantum variable layering processing, including: The discriminative feature vector is encoded onto a quantum state to obtain a quantum state feature vector; Quantum convolutional layers are used to perform convolution operations on adjacent qubits in the quantum state feature vector to extract quantum state correlation features; The quantum state correlation features are quantum pooled by a quantum pooling layer to generate quantum features.

9. The non-destructive testing method for flame detectors based on multiphysics coupling and impedance spectral analysis according to claim 6, characterized in that, include: Measuring the quantum characteristics includes: Each qubit in the quantum feature is sampled multiple times under the computational basis to generate sampling results; Measure the probability of a set of "0" and "1" bit strings appearing in the sampling result.

10. A non-destructive testing system for flame detectors based on multi-physics coupling and impedance spectral analysis, used to implement the non-destructive testing method for flame detectors based on multi-physics coupling and impedance spectral analysis as described in any one of claims 1 to 9, characterized in that, The flame detector non-destructive testing system includes: A fire thermal radiation effect simulation module is used to apply electromagnetic fields and temperature fields to a flame detector, and couple the electromagnetic fields and temperature fields to simulate the fire thermal radiation effect. The impedance analysis module is used to measure the impedance data of the flame detector under the fire thermal radiation effect, perform impedance analysis on the impedance data, and construct a state matrix based on the results of the impedance analysis. The fault feature extraction module is used to extract quantum features from the state matrix and generate a fault feature vector. The fault diagnosis module is used to diagnose the fault feature vector based on a hybrid decision architecture of convolutional quantum neural network and decision tree to obtain the fault status of the flame detector.

Images