A method and apparatus for detecting magnetic fields and magnetic gradients

Abstract

This application belongs to the field of magnetic anomaly detection technology, specifically a method and apparatus for detecting magnetic fields and magnetic gradients. It includes three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor housed within a spherical shell. The current in the spherical feedback coil is adjusted to bring the values ​​of the three components of the reference time-difference fluxgate sensor close to zero. The output signals of the three-component and reference time-difference fluxgate sensors are acquired. The time difference is calculated from the output signals. The measured magnetic field is calculated based on the time difference. A magnetic gradient tensor is constructed based on the magnetic induction intensity signals along different axes from the three-component and reference time-difference fluxgate sensors. This method completely solves the problems of signal drift and weak anti-interference capability inherent in traditional amplitude-type fluxgate sensors, greatly improving the accuracy of magnetic field measurement.

Description

Technical Field

[0001] This application belongs to the field of magnetic anomaly detection technology, specifically a method and device for detecting magnetic fields and magnetic gradients. Background Technology

[0002] Magnetic gradient tensor measurement devices (MTTMs), as high-precision magnetic field measurement instruments, are used to capture changes in magnetic fields in different directions in space. Their working principle involves the coordinated combination of multiple fluxgate magnetometers or magnetic sensors to simultaneously and accurately measure the three-dimensional rate of change of the magnetic field in a Cartesian coordinate system (typically along the X, Y, and Z axes), thereby obtaining the magnetic field gradient tensor and providing crucial data for in-depth analysis of magnetic field characteristics. Given the high precision requirements of these sensors in sensing magnetic field direction, current magnetic gradient tensor probes primarily employ a superconducting quantum interference device (SQUID) combination scheme. SQUID sensors, manufactured based on standard Nb / AlOX / Nb technology, are waterproof and resistant to thermal cycling. The system has undergone numerous experiments and flight tests, successfully measuring the magnetic field characteristics of helicopters and conducting flight tests on single-gradient meters and multiple generations of systems. However, superconducting materials face many challenges in practical applications. They are extremely sensitive to external magnetic field environments; even minute magnetic field fluctuations can interfere with the stability of the superconducting system, thus affecting measurement accuracy. Therefore, in high-precision measurement scenarios, comprehensive and efficient magnetic field shielding measures are essential to isolate external interference. Moreover, the superconducting properties of superconducting materials are predicated on extremely low-temperature environments, typically requiring cryogenic cooling media such as liquid nitrogen or liquid helium to maintain the cryogenic state. This undoubtedly significantly increases the operation and maintenance costs of superconducting systems, adds complexity to the system architecture, and introduces high energy consumption, limiting the expansion of their application scenarios. While existing spherical feedback three-component fluxgate magnetic gradient full tensor probes achieve the measurement of the magnetic gradient tensor, the limitations of inductive fluxgates mean that inductive magnetic sensors can only measure alternating magnetic fields and cannot measure DC magnetic fields. Their adaptive capability is also weak. In many practical applications such as geological exploration, magnetic anomaly detection, and military reconnaissance, the lack of DC magnetic field information greatly limits the effectiveness of such probes, making it difficult to meet the needs for comprehensive and accurate magnetic field measurement in complex environments. Summary of the Invention

[0003] The technical problem to be solved by this application is to provide a method for detecting magnetic fields and magnetic gradients, which solves the problem that inductive magnetic sensors can only measure alternating magnetic fields and cannot measure DC magnetic fields, and have poor adaptive capabilities.

[0004] Another aspect of this application provides a magnetic field and magnetic gradient detection device.

[0005] A magnetic field and magnetic gradient detection method according to a first aspect of this application includes:

[0006] Three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor are placed inside a spherical shell, forming a cross-shaped structure centered on the origin in space. Two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis. A spherical feedback coil is wound in three enameled wire winding grooves that intersect perpendicularly along the X, Y, and Z directions, respectively, on the surface of the spherical shell.

[0007] Adjust the current in the spherical feedback coil to make the values ​​of the three components of the reference time difference type fluxgate sensor approach zero;

[0008] Excitation signals are passed into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0009] Acquire the output signals of the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor respectively;

[0010] The time difference is calculated using the output signal;

[0011] Calculate the measured magnetic field based on the time difference;

[0012] The magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors are obtained by superimposing the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

[0013] Furthermore, the excitation signal is a triangular wave, and the resulting excitation magnetic field is expressed as follows:

[0014] ,

[0015] In the formula For the slope of the triangular wave excitation magnetic field, For time, For the number of cycles, For the incentive cycle, To excite the magnetic field.

[0016] Furthermore, the formula for calculating the measured magnetic field based on the time difference is as follows: In the formula For the measured magnetic field, The magnetic field frequency, For time difference.

[0017] Furthermore, based on the superposition of the measured magnetic field and the excitation magnetic field, the calculation formula for the magnetic induction intensity signal of different three-component time-difference fluxgate sensors is as follows:

[0018] B is the magnetic flux density signal. For vacuum permeability, The relative permeability of the core material. To excite the amplitude of the magnetic field.

[0019] Furthermore, a magnetic gradient tensor is constructed based on the magnetic induction intensity signals along different axes on the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor, including:

[0020] The gradients of the magnetic field components in the X, Y, and Z directions are obtained by dividing the difference in magnetic field induction intensity between two three-component time-difference fluxgate sensors on the X-axis by the distance between the two three-component time-difference fluxgate sensors in the X direction.

[0021] The gradients of the magnetic field components in the X, Y, and Z directions are obtained by dividing the differences in magnetic induction intensity between the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor on the Y-axis by the distance between the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor in the Y-axis.

[0022] The gradient of the magnetic field component in the Z direction is obtained by taking the negative sum of the gradients of the magnetic field component in the X direction and the gradients of the magnetic field component in the Y direction.

[0023] Furthermore, the current in the spherical feedback coil is adjusted to bring the values ​​of the three components of the reference time difference type fluxgate sensor close to zero, including:

[0024] Acquire the output data of the reference time difference type fluxgate sensor;

[0025] A preliminary statistical characteristic model of the noise is constructed, and an objective function for minimizing the noise is set. The feedback parameters are iteratively adjusted with a fixed step size.

[0026] In each iteration, the degree of realization of stochastic resonance under the current feedback parameter settings is evaluated by comparing the output data with the expected output.

[0027] If the ideal state is not achieved, the feedback parameters are adjusted based on the error signal.

[0028] Furthermore, wavelet transform decomposition is performed on the output signals of the three three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensor. High-frequency noise components are filtered out by a threshold function while retaining key features to obtain the denoising coefficients.

[0029] The denoising coefficients are reconstructed to generate a low-noise signal, which is then input into the untrained neural network.

[0030] The improved particle swarm optimization algorithm is started, the particle velocity, position and dynamic parameters are initialized, and the particle state is adjusted in real time during the iteration.

[0031] When the global fitness accuracy is met or the maximum number of iterations is reached, the optimal particle position sequence is output as the initial weights and threshold of the neural network.

[0032] Drive the neural network to train until the output error meets the set conditions;

[0033] The output of a neural network is the denoised output signal.

[0034] Furthermore, after calculating the time difference from the output signal, the obtained time difference is denoised, including:

[0035] Anomaly detection in isolated forests is performed on time differences. An anomaly detection model is constructed by dynamic window segmentation and adaptive parameter adjustment. For each detected anomaly, the location and degree of anomaly are recorded. For each anomaly, surrounding normal points are selected as reference points and the weighted average is used to replace the anomaly to obtain a preprocessed signal.

[0036] The preprocessed signal is input into the VMD decomposition module, and the mode number K and penalty factor are adaptively selected based on the characteristics of the preprocessed signal. The preprocessed signal is decomposed into K modal components and the energy entropy of each component is calculated.

[0037] By normalizing the entropy value, weights are assigned according to the energy entropy of each modal component and then weighted and fused to generate an intermediate signal with enhanced main features.

[0038] The intermediate signal with enhanced main features is subjected to hierarchical adaptive threshold calculation. Soft threshold denoising is applied layer by layer and the processing progress is iteratively judged. The time difference is reconstructed after all levels are completed.

[0039] A magnetic field and magnetic gradient detection device according to a second aspect of this application includes:

[0040] Three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor are placed inside a spherical shell, forming a cross-shaped structure centered on the origin of the coordinate system. Two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis.

[0041] The surface of the spherical shell is engraved with three enameled wire winding grooves that intersect perpendicularly in the X, Y, and Z directions, respectively, for winding the spherical feedback coil;

[0042] The controller is used to adjust the current in the spherical feedback coil so that the values ​​of the three components of the reference time difference fluxgate sensor approach zero;

[0043] The detection circuit is used to pass excitation signals into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0044] Used to acquire the output signals of three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0045] Used to calculate the time difference from the output signal;

[0046] Used to calculate the measured magnetic field based on the time difference;

[0047] This is used to obtain the magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors based on the superposition of the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

[0048] Compared with existing technologies, the advantages of this application are as follows: This application avoids the defects of amplitude-type fluxgate sensors and uses a time-difference fluxgate sensor for detection, obtaining accurate time difference measurements. This completely solves the problems of signal drift and weak anti-interference ability inherent in traditional amplitude-type fluxgate sensors, greatly improving the accuracy of magnetic field measurements. It is suitable for applications requiring high accuracy in magnetic field measurements, particularly in complex electromagnetic environments. Attached Figure Description

[0049] Figure 1 A flowchart illustrating a magnetic field and magnetic gradient detection method provided in this application embodiment;

[0050] Figure 2 A relative position diagram of a time-difference fluxgate sensor provided in an embodiment of this application;

[0051] Figure 3 A block diagram of the controller, detection circuit, and acquisition module provided in the embodiments of this application;

[0052] Figure 4 The circuit schematic diagram of the acquisition module provided in the embodiment of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] See Figure 1 The flowchart illustrates a method for detecting magnetic fields and magnetic gradients. An embodiment of this application provides a method for detecting magnetic fields and magnetic gradients, comprising: placing three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor within a spherical shell, forming a cross-shaped structure centered on the origin in space; wherein two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis; a spherical feedback coil is wound within three enameled wire winding grooves perpendicularly intersecting along the X, Y, and Z directions, respectively, etched on the surface of the spherical shell.

[0055] Adjust the current in the spherical feedback coil to make the values ​​of the three components of the reference time difference type fluxgate sensor approach zero;

[0056] Excitation signals are passed into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0057] Acquire the output signals of three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0058] The time difference is calculated using the output signal;

[0059] Calculate the measured magnetic field based on the time difference;

[0060] The magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors are obtained by superimposing the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

[0061] The three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor here belong to the same structure. Based on the principle of electromagnetic induction and the nonlinear characteristics of soft magnetic core materials, they can measure the external magnetic field. When the soft magnetic core is subjected to a periodically alternating excitation magnetic field and reaches a bidirectional oversaturation state, when the external magnetic field is 0, the time difference between the positive and negative saturation states during the magnetization of the soft magnetic core does not change. However, when the external magnetic field is not 0, the time difference between the positive and negative saturation states during the magnetization of the core changes. This time difference has a linear relationship with the measured magnetic field, thus obtaining the measured magnetic field in a single direction.

[0062] In this embodiment, the current in the spherical feedback coil is adjusted according to the output signal of the reference time difference fluxgate sensor, so that the values ​​of the three components of the reference time difference fluxgate sensor approach zero. The current in the spherical feedback coil is adjusted to generate a feedback magnetic field, which cancels out the measured magnetic field. When the measured magnetic field of the reference time difference fluxgate sensor approaches zero, the magnetic field environment in which the four time difference fluxgate sensors work is a near-zero magnetic field. The induced signal obtained in this way is the residual magnetic field generated by the measured magnetic field and the spherical feedback coil. This residual magnetic field will not be zero, but it approaches zero. The smaller the value of the residual magnetic field, the more accurate the measurement result.

[0063] The output signals of the magnetic field, i.e. the induced signals, are obtained by collecting the output signals of the magnetic field from three other three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor, and then the time difference and magnetic gradient tensor are calculated.

[0064] Three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors measure the output signal of the magnetic field. The time difference can be obtained by means of peak detection method, which locates the peak time of the sensing signal pulse through peak detection circuit, and calculates the time difference based on the time of three adjacent peak points, so that the influence of the amplitude change of the sensing signal on the reading of the peak time of the pulse is not considered.

[0065] In one embodiment, the excitation signal is a triangular wave, and the resulting excitation magnetic field is represented as follows:

[0066] In the formula For the slope of the triangular wave excitation magnetic field, For time, For the number of cycles, For the incentive cycle, To excite the magnetic field. When the excitation period is excitation magnetic field When used as an excitation signal, the detection principle is based on the hysteresis saturation bistable characteristic of the magnetic core, and the time difference of bidirectional saturation is used to measure the magnetic field being tested. The relationship between the two is ultimately determined by the time difference to obtain the measured magnetic field. In one embodiment, a triangular wave is selected as the excitation signal, which acts on the excitation coils of the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor to measure the magnetic field. This excitation signal has low requirements for the coercivity of the magnetic core, and the sensitivity depends only on the amplitude of the excitation magnetic field. and magnetic field frequency Decide.

[0067] In one embodiment, the formula for calculating the measured magnetic field based on the time difference is as follows: In the formula For the measured magnetic field, The magnetic field frequency, The time difference is used to calculate the magnetic flux density signal of different three-component time-difference fluxgate sensors, based on the superposition of the measured magnetic field and the excitation magnetic field.

[0068] B is the magnetic flux density signal. For vacuum permeability, The relative permeability of the core material. To excite the amplitude of the magnetic field.

[0069] In one embodiment, a magnetic gradient tensor is constructed based on magnetic flux density signals along different axes of the magnetic field measured by a three-component time-difference fluxgate sensor and a reference time-difference fluxgate sensor, including:

[0070] The gradients of the magnetic field components in the X direction, Y direction, and Z direction are obtained by dividing the difference in magnetic induction intensity between two three-component time-difference fluxgate sensors in the X direction by the distance between the two three-component time-difference fluxgate sensors in the X direction.

[0071] The gradients of the magnetic field components in the X, Y, and Z directions are obtained by dividing the difference in magnetic induction intensity between the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor in the Y direction by the distance between them.

[0072] The gradient of the magnetic field component in the Z direction is obtained by taking the negative sum of the gradients of the magnetic field component in the X direction and the gradients of the magnetic field component in the Y direction.

[0073] Two three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and one three-component time-difference fluxgate sensor and a reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis. The magnetic gradient tensor of the full-tensor probe, composed of three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor, is... The form can be expressed as:

[0074] ,

[0075] In the formula , , representing the elements in the magnetic gradient tensor, which contains nine components. However, in a source-free space (referring to a specific region without external sources of magnetic or electric fields. In such a space, all electromagnetic fields are generated by boundary conditions or distant sources, rather than directly by sources within the region), in mathematical and physical descriptions, a source-free space typically means that the charge density and current density are both zero within that region. This leads to the validity of two important electromagnetic field equations:

[0076] Divergence theorem. In source-free space, the divergence of the electric field is zero, i.e., ∇ B=0 indicates that the electric field lines are closed loops, with no beginning and no end.

[0077] Ampère's circuital law. In a source-free space, the curl of a magnetic field is zero, i.e., ∇². B=0, which means that the magnetic field lines are also closed and there is no magnetic monopole.

[0078] These conditions lead to the symmetry of the magnetic gradient tensor and the linear dependence of some components. Since the curl of the magnetic field is zero, the off-diagonal elements of the magnetic gradient tensor satisfy... = , = , = This means there are only three independent off-diagonal elements; furthermore, the curl of the magnetic field is zero. The trace of the magnetic gradient tensor (the sum of all diagonal elements) must also be zero, i.e. + + =0, which means that only two of the three diagonal elements are independent; therefore, only five quantities are independent in the source-free space.

[0079] The computed elements are:

[0080] (1) , This is achieved by taking the difference in magnetic induction intensity in the X-direction of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis, and dividing it by the distance between them. It is used for calculation, representing the gradient of the magnetic field component in the X direction as it changes with the X direction.

[0081] This refers to the magnetic flux density component in the X-direction of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis. In one of two three-component time-difference fluxgate sensors symmetrically distributed along the X-axis, the magnetic flux density component in the X-direction of the other three-component time-difference fluxgate sensor is transmitted through the time difference. To obtain.

[0082] (2) , This is achieved by taking the difference in magnetic induction intensity in the Y direction of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis, and dividing it by the distance between them. The gradient of the magnetic field component in the Y direction as a function of the X direction is calculated, and the magnetic induction intensity changes with time difference. To obtain.

[0083] This refers to the Y-axis component of the magnetic flux density of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis. The magnetic flux density of one of the two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis is the component of the magnetic flux density in the Y direction.

[0084] (3) , This is achieved by taking the difference in magnetic induction intensity in the Z direction of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis, and dividing it by the distance between them. The gradient of the magnetic field component in the Z direction as a function of the X direction is calculated, and the magnetic induction intensity changes with time difference. To obtain.

[0085] This refers to the Z-direction component of the magnetic flux density of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis. The component of the magnetic flux density in the Z direction of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis.

[0086] (4) , It is the difference in magnetic induction intensity in the X direction between the reference time difference type fluxgate sensor and the three-component time difference type fluxgate sensor on the Y-axis, divided by the distance between them. , which represents the gradient of the magnetic field component in the X direction as it changes with the Y direction.

[0087] This represents the X-axis component of the magnetic flux density of a reference time-difference fluxgate sensor on the Y-axis. The magnetic flux density of the three-component time-difference fluxgate sensor on the Y-axis is the component in the X-direction.

[0088] (5) , It is the difference in magnetic induction intensity in the Y-axis direction between a reference time difference type fluxgate sensor and a three-component time difference type fluxgate sensor, divided by the distance between them. It is used for calculation, representing the gradient of the magnetic field component in the Y direction as it changes with the Y direction.

[0089] This represents the component of the magnetic flux density in the Y-direction of a reference time-difference fluxgate sensor. The magnetic flux density of the three-component time-difference fluxgate sensor on the Y-axis is the component in the Y-direction.

[0090] (6) , It is the difference in magnetic flux density in the Z direction between the reference time difference type fluxgate sensor and the three-component time difference type fluxgate sensor on the Y-axis, divided by the distance between them. It is used for calculation, representing the gradient of the magnetic field component in the Z direction as it changes with the Y direction.

[0091] This represents the Z-axis component of the magnetic flux density of a reference time-difference fluxgate sensor on the Y-axis. The component of the magnetic flux density of the three-component time-difference fluxgate sensor on the Y-axis is in the Z-direction.

[0092] (7) , This is achieved by taking the difference in magnetic induction intensity in the Z direction of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis, and dividing it by the distance between them. It is used for calculation, representing the gradient of the magnetic field component in the Z direction as it changes with the X direction.

[0093] This refers to the Z-direction component of the magnetic flux density of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis. The component of the magnetic flux density in the Z direction of one of two three-component time-difference fluxgate sensors that are symmetrically distributed along the X-axis.

[0094] (8) , It is the difference in magnetic flux density in the Z direction between the reference time difference type fluxgate sensor and the three-component time difference type fluxgate sensor on the Y-axis, divided by the distance between them. The calculation represents the gradient of the magnetic field component in the Z direction as it changes with the Y direction, but the output of the reference time difference fluxgate sensor is zero.

[0095] This represents the Z-axis component of the magnetic flux density of a reference time-difference fluxgate sensor on the Y-axis. The component of the magnetic flux density of the three-component time-difference fluxgate sensor on the Y-axis is in the Z-direction.

[0096] (9) , Essentially, it involves calculating the gradient of the magnetic field in the Z direction, which is affected by changes in the magnetic field in the X and Y directions. This is the result of the combined effect of changes in the magnetic field in both the X and Y directions. Furthermore, since the curl of the magnetic field is zero, the trace of the magnetic gradient tensor (the sum of all diagonal elements) must also be zero. ,so A negative sign is required.

[0097] In one embodiment, adjusting the current in the spherical feedback coil to make the three components of the reference time difference fluxgate sensor zero includes:

[0098] Acquire the output data of the reference time difference type fluxgate sensor;

[0099] A preliminary statistical characteristic model of the noise is constructed, and an objective function for minimizing the noise is set. The feedback parameters are iteratively adjusted with a fixed step size.

[0100] In each iteration, the degree of realization of stochastic resonance under the current feedback parameter settings is evaluated by comparing the output data with the expected output.

[0101] If the ideal state is not achieved, the feedback parameters are adjusted based on the error signal.

[0102] The magnetic field being measured can be determined by the output signal of the reference time difference fluxgate sensor. The current in the spherical feedback coil is adjusted based on the output signal to generate a canceling magnetic field. The ultimate goal is to make the output signal of the reference time difference fluxgate sensor approach zero. Understandably, the other three three-component time difference fluxgate sensors are in the same environment as the reference time difference fluxgate sensor, and their measured values ​​also approach zero. However, experiments have shown that the values ​​cannot be zero; there will be residual magnetic fields.

[0103] The statistical features in constructing a preliminary statistical feature model of noise can include: Probability distribution: Noise signals typically possess certain statistical characteristics, and their amplitude and phase variations can be described by probability distributions. Common probability distributions include Gaussian, uniform, and Poisson distributions. Mean: The mean of a noise signal reflects its average level; typically, the mean is close to zero. Variance and standard deviation: Variance and standard deviation measure the fluctuation range of the noise signal's amplitude and are important parameters for measuring noise intensity. The standard deviation is the positive square root of the variance. Correlation function and power spectral density: The correlation function describes the degree of correlation of the noise signal at different time points, while the power spectral density describes the power distribution of the noise signal in the frequency domain. Choosing a suitable preliminary statistical feature model, such as a Gaussian model or a Poisson model, is also important.

[0104] The objective function can be either mean squared error or absolute error. The feedback parameters are adjusted iteratively to gradually optimize the objective function. The basic idea of ​​iterative adjustment is: in each iteration, the gradient (or derivative) of the objective function is calculated based on the current parameter values, and then the feedback parameters are updated in the opposite direction of the gradient, so that the value of the objective function gradually decreases.

[0105] The fixed step size (also known as the learning rate) is an important parameter in the iterative adjustment process. It determines the magnitude of the feedback parameter update in each iteration.

[0106] In another embodiment, the following scheme can also be used to adjust the current in the spherical feedback coil so that the values ​​of the three components of the reference time difference type fluxgate sensor are zero:

[0107] A feedback structure is selected, referring to the odd-order, cubic, and higher-order odd-order feedback structures described below. Feedback parameters are set, the excitation signal is input, and the output is processed through Runge-Kutta iteration. The particle swarm optimization algorithm is introduced to dynamically solve for the odd-order polynomial coefficients. Simultaneously, cross-correlation analysis is performed on the excitation signal. By analyzing multiple sets of data, when the cross-correlation value between the excitation signal and the output signal is closer to 1, the feedback system is considered to have resonated. This data is recorded and stored. After finding the optimal feedback parameters, they are reset to achieve resonance in the feedback system, including the reference time-difference fluxgate sensor. The process then terminates, achieving the goal of active adaptation of the three three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensor to external noise. The above process achieves automatic adjustment of the feedback parameters through software calculation.

[0108] This application studies the conditions for stochastic resonance in a bistable system. Based on the equivalent mathematical model of the bistable system, an adaptive control strategy is designed to achieve stochastic resonance. Since the bistable characteristics of a time-difference fluxgate sensor can be described by the Langevin equation: In the formula, The input signal is, and , For input signal, It is a Gaussian white noise signal. , , Since the input is a positive number, the Langevin equation exhibits obvious odd-order characteristics after differentiation. Therefore, utilizing the odd-order function characteristics of bistable systems, and considering that the choice of feedback form directly affects the sensitivity of the time-difference fluxgate sensor, odd-order, cubic, and higher-order odd-order feedback structures were compared. Experiments show that the shape of the noise response characteristic curve is basically consistent under different noise intensities. When using a higher-order odd-order feedback structure, the time-difference fluxgate sensor can achieve maximum sensitivity within a certain noise range; cubic feedback can adjust the sensitivity peak position, allowing it to actively adapt to external noise. Therefore, this application combines linear and higher-order odd-order feedback structures, which can effectively adjust the coercivity and sensitivity peak position, and significantly improve the sensitivity of the time-difference fluxgate sensor and reduce the influence of noise under uncertain noise characteristics.

[0109] Considering the unknown characteristics of environmental noise, it is difficult to achieve stochastic resonance by individually changing system parameters (coefficients of odd-degree functions). Therefore, an adaptive adjustment algorithm for feedback parameters needs to be designed to dynamically solve for the odd-degree polynomial coefficients. Combining this with the real-time magnetic field measurement requirements of time-difference fluxgate sensors, a particle swarm optimization (PSO) algorithm is introduced to dynamically solve for the odd-degree polynomial coefficients. Here, the particles represent the feedback coefficient values ​​to be solved. By initializing the feedback coefficient set, the impact of using the optimal coefficient feedback on improving sensor sensitivity is determined based on the fitness function value. The fitness function is used to evaluate a potential feasible solution, i.e., the optimal value of the feedback coefficients. Therefore, this application uses mean squared error as the fitness function, which measures the difference between the model's predicted value and the actual value. When the fitness value reaches a certain value, the time difference is maximized, and the time-difference fluxgate sensor reaches its optimal resonance state. At this point, the PSO algorithm output value is the optimal combination of coefficients for the feedback function. By calculating the fitness value of the new position of each feedback coefficient particle, updating the individual and global optimal values ​​based on the fitness value of each feedback coefficient, updating the velocity and position of each feedback coefficient particle, and checking whether the termination condition of the optimal fitness is met, if not, the loop returns to the step of calculating the fitness value of the new position of each feedback coefficient particle; if it is met, the optimal solution of the feedback coefficient is output. By reaching the resonance state, the sensitivity of the time difference fluxgate sensor is indirectly improved.

[0110] In one embodiment, wavelet transform decomposition is performed on the output signals of the three three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensor, and the high-frequency noise components are filtered out by a threshold function while retaining key features to obtain the denoising coefficient.

[0111] The denoising coefficients are reconstructed to generate a low-noise signal, which is then input into the untrained neural network.

[0112] The improved particle swarm optimization algorithm is started, the particle velocity, position and dynamic parameters are initialized, and the particle state is adjusted in real time during the iteration.

[0113] When the global fitness accuracy is met or the maximum number of iterations is reached, the optimal particle position sequence is output as the initial weights and threshold of the neural network.

[0114] Drive the neural network to train until the output error meets the set conditions;

[0115] The output of a neural network is the denoised output signal.

[0116] In this process, the low-noise signal generated by reconstructing the denoised coefficients is input into the untrained neural network using inverse wavelet transform. This process avoids the inertial weights found in traditional particle swarm optimization algorithms. Typically, these values ​​are fixed or simply linearly decreasing, making dynamic adaptation impossible based on the search progress. This application's embodiments employ a linear decreasing strategy:

[0117] ,

[0118] in, and These are the initial and final values ​​of the inertia weight, respectively. This is the current iteration number; This is the maximum number of iterations. (Setting) For high weighting, this is the early stage weighting, and the value is set relatively large to enhance global exploration. This is a low-weight setting, indicating a later-stage weight, with a relatively small value to focus on localized development.

[0119] Traditional particle swarm optimization (PSO) algorithms lack compression factor constraints for velocity updates, which can lead to excessive velocities, particle oscillations, or divergence. The improved velocity update formula is as follows:

[0120] ,

[0121] in, It is a particle In dimensions The speed of the (t+1)th iteration (the speed to be updated). It is a particle In dimensions The speed of the t-th iteration, It is a particle In dimensions The best historical position of the particle swarm in its t-th iteration, where the entire particle swarm is in dimension d. It is the globally optimal position in the t-th iteration. It is the compression factor, and its calculation formula is usually based on the learning factor. and The sum and the dimension of the particle swarm Dimension It may not appear directly in the formula for calculating the compression factor, but it will affect the performance of the algorithm and the choice of parameters.

[0122] The difference between the traditional and improved speed updates lies in the addition of a compression factor. Limit the velocity amplitude to prevent particles from going out of control.

[0123] Among them, compression factor The calculation formula. When At that time, the compression factor Forced compression speed ensures convergence stability:

[0124] ,

[0125] in, yes and A function of the sum and their product, generally set >4 is used to ensure that the square root in the formula is meaningful and to optimize the compression factor. It is a positive value.

[0126] In one embodiment, the time difference is calculated from the output signal and then denoised, including:

[0127] Anomaly detection in isolated forests is performed on time differences. An anomaly detection model is constructed by dynamic window segmentation and adaptive parameter adjustment. For each detected anomaly, the location and degree of anomaly are recorded. For each anomaly, surrounding normal points are selected as reference points and the weighted average is used to replace the anomaly to obtain a preprocessed signal.

[0128] The preprocessed signal is input into the VMD decomposition module (variational mode decomposition), and the mode number K and penalty factor are adaptively selected based on the characteristics of the preprocessed signal. The preprocessed signal is decomposed into K modal components and the energy entropy of each component is calculated.

[0129] By normalizing the entropy value, weights are assigned according to the energy entropy of each modal component and then weighted and fused to generate an intermediate signal with enhanced main features.

[0130] The intermediate signal with enhanced main features is subjected to hierarchical adaptive threshold calculation. Soft threshold denoising is applied layer by layer and the processing progress is iteratively judged. The time difference is reconstructed after all levels are completed.

[0131] Since the high-frequency random noise (bandwidth 10kHz~100kHz) generated by the irreversible jumping of domain walls during the magnetization process of soft magnetic core is distributed near the peak of induced voltage pulse, it is easy to cause time difference fluctuations. Therefore, by using isolated forest anomaly detection and VMD decomposition module to denoise the time difference signal, signal distortion is effectively suppressed and denoising accuracy is enhanced.

[0132] Traditional VMD decomposition requires the mode number K to be manually preset or calculated using fixed empirical formulas, making it susceptible to noise interference, leading to insufficient or excessive decomposition. This application's embodiment employs an improved VMD decomposition based on signal length. The joint rule with energy entropy avoids over-decomposition caused by high-frequency noise interference. The mode number K is as follows:

[0133] ,

[0134] in The signal length is used to avoid over-decomposition under high-frequency noise interference.

[0135] Traditional VMD decomposition cannot adapt to different noise scenarios, potentially leading to loss of signal details or noise residue. This application's embodiments improve VMD decomposition: based on signal variance... Adaptive adjustment of penalty factor High noise Increase the value, approaching 90, to enhance bandwidth constraints and suppress noise; penalty factor at low noise levels. Decrease the value, approaching 80, to preserve signal details. The formula is as follows:

[0136] ,

[0137] Signal variance in high-noise scenarios Increase, penalty factor Approaching 90, enhance modal bandwidth constraints. In low-noise scenarios. Approaching 80 preserves signal details while maintaining modal orthogonality.

[0138] Traditional VMD decomposition uses the Alternating Direction Multiplier Method (ADMM), but it does not dynamically adjust the iterative parameters based on signal characteristics, resulting in weak convergence and adaptability. This application's embodiment improves VMD decomposition by iteratively updating using the Alternating Direction Multiplier Method, including: mode update, center frequency update, and Lagrange multiplier update. Wherein:

[0139] Modal update: Frequency domain update :

[0140] ,

[0141] It indicates the first During the nth iteration, the 1st Each mode in the frequency domain Update value at that location, For the first Each mode in the frequency domain The value at that location, The frequency domain representation of the original signal (input signal). Indicates except the first Besides the one mode, other modes in the frequency domain The components at the point and, These represent Lagrange multipliers, used to constrain decomposition conditions. Representing the The center frequency of each mode.

[0142] Center frequency update:

[0143] ,

[0144] | | indicates the first During the nth iteration, the 1st The energy density (squared of the modulus) in the frequency domain after each mode update is obtained by weighted integration (using...). The center frequency is calculated by taking the ratio of the weighted average (weighted average) to the energy integral, thus concentrating the modal energy around the center frequency. Indicates the first The next iteration yields the updated center frequency.

[0145] Lagrange multipliers update:

[0146] ,in, It indicates the first The updated value of the Lagrange multipliers in the next iteration. Indicates the first Lagrange multipliers in the next iteration. To update the step size (learning rate), control the magnitude of the Lagrange multiplier update.

[0147] After decomposition, weights are assigned based on modal energy entropy, and wavelet denoising is fused: high-energy-entropy modes (containing more noise) are assigned low weights, and low-energy-entropy modes (containing more effective signals) are assigned high weights.

[0148] ,

[0149] ,

[0150] In the formula It indicates the first Modal signals in the time domain Energy. Indicates the first The entropy of the modal signals in each time domain is calculated. High-entropy modes (containing complex noise) are assigned low weights to suppress noise reconstruction contributions. Low-entropy modes (containing valid signals) are assigned high weights. For multi-scale threshold denoising, the entropy of the modal signals in each time domain is calculated. Adaptive wavelet thresholding is applied: high-frequency modes are treated with a low threshold (to preserve details), while low-frequency modes are treated with a high threshold (to suppress baseline drift). Indicates weight, Indicates the first The entropy of a modal signal in the time domain.

[0151] In one embodiment, the wavelet transform is improved by applying an adaptive soft thresholding process to the wavelet coefficients at each scale. This improved method is more adaptive than traditional soft thresholding, effectively removing noise while preserving signal features as much as possible. The formula is as follows: The adaptive soft thresholding process applies a soft threshold to the wavelet coefficients at each scale:

[0152] ,

[0153] This represents the wavelet coefficients updated after thresholding. This represents the threshold for the corresponding scale j, with different thresholds used for different scales j. , For wavelet coefficients, the threshold is increased for high-noise scales (to suppress noise) and decreased for low-noise scales (to preserve details).

[0154] See Figure 2 The relative position diagram of the time difference fluxgate sensor shown, and see also... Figure 3 The illustrated embodiment of this application provides a structural block diagram of a magnetic field and magnetic gradient detection device. The magnetic field and magnetic gradient detection device includes:

[0155] Three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor are placed inside a spherical shell, forming a cross-shaped structure centered on the origin of the coordinate system. Two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis.

[0156] Three enameled wire winding grooves are engraved on the surface of the spherical shell, which intersect perpendicularly along the X, Y, and Z directions respectively, for winding the spherical feedback coil;

[0157] The controller is used to adjust the current in the spherical feedback coil so that the values ​​of the three components of the reference time difference fluxgate sensor approach zero;

[0158] The detection circuit is used to pass excitation signals into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0159] Used to acquire the output signals of three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor;

[0160] Used to calculate the time difference from the output signal;

[0161] Used to calculate the measured magnetic field based on the time difference;

[0162] This is used to obtain the magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors based on the superposition of the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

[0163] The spherical housing is made of non-conductive materials, which include various types such as polycarbonate plastic, fiberglass, polyurethane foam, phenolic plastic, or mica. The enameled wire winding grooves are characterized by equal width and spacing but unequal depth. Inside the spherical housing, a three-component time-difference fluxgate sensor and a reference time-difference fluxgate sensor are fixed using a cross-shaped bracket and bolts.

[0164] The three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor have the same structure, consisting of a sensing unit and a detection circuit. The sensing unit consists of a soft magnetic core and a coil, which completes the sensing of the magnetic field to be measured.

[0165] Three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors are based on the principle of electromagnetic induction and the nonlinear characteristics of soft magnetic core materials to measure external magnetic fields. (Periodic variation diagram) When a periodically alternating excitation magnetic field acts on the soft magnetic core to reach a bidirectional oversaturation state, when the external magnetic field is 0, the time difference between the positive and negative saturation states during core magnetization does not change. However, when the external magnetic field is not 0, the time difference between the positive and negative saturation states during core magnetization changes. This time difference is linearly related to the measured magnetic field, thus obtaining the measured magnetic field in a single direction.

[0166] The detection circuit includes an excitation signal generation module, an induction signal processing module, and a time difference reading and processing module. The excitation signal generation module is connected to the soft magnetic core and is used to generate a triangular wave. The induction signal processing module is used to preprocess the induction signal and remove noise. The noise removal process is to perform wavelet transform decomposition, filter out high-frequency noise components through a threshold function, and retain key features to obtain the noise reduction coefficient.

[0167] The denoising coefficients are reconstructed to generate a low-noise signal, which is then input into the untrained neural network.

[0168] The improved particle swarm optimization algorithm is started, the particle velocity, position and dynamic parameters are initialized, and the particle state is adjusted in real time during the iteration.

[0169] When the global fitness accuracy is met or the maximum number of iterations is reached, the optimal particle position sequence is output as the initial weights and threshold of the neural network.

[0170] Drive the neural network to train until the output error meets the set conditions;

[0171] The output of a neural network is the denoised output signal.

[0172] The time difference of the denoised output signal is calculated and read by the time difference reading and processing module. The measured magnetic field is calculated based on the time difference, and the magnetic induction signals of different three-component time difference type fluxgate sensors and reference time difference type fluxgate sensors are obtained by superimposing the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time difference type fluxgate sensors and reference time difference type fluxgate sensors.

[0173] The time difference reading and processing module is also used to denoise the time difference obtained after calculating the time difference from the output signal, including:

[0174] Anomaly detection in isolated forests is performed on time differences. An anomaly detection model is constructed by dynamic window segmentation and adaptive parameter adjustment. For each detected anomaly, the location and degree of anomaly are recorded. For each anomaly, surrounding normal points are selected as reference points and the weighted average is used to replace the anomaly to obtain a preprocessed signal.

[0175] The preprocessed signal is input into the VMD decomposition module, and the mode number K and penalty factor are adaptively selected based on the characteristics of the preprocessed signal. The preprocessed signal is decomposed into K modal components and the energy entropy of each component is calculated.

[0176] By normalizing the entropy value, weights are assigned according to the energy entropy of each modal component and then weighted and fused to generate an intermediate signal with enhanced main features.

[0177] The intermediate signal with enhanced main features is subjected to hierarchical adaptive threshold calculation. Soft threshold denoising is applied layer by layer and the processing progress is iteratively judged. The time difference is reconstructed after all levels are completed.

[0178] See Figure 4 As shown, in one embodiment, the controller further includes a data acquisition module connected to the reference time difference type fluxgate sensor. The data acquisition module is connected to the input terminal of the DSP module through an operational amplifier U1 in series with a resistor R1. After processing by the DSP module, the signal is converted into an analog signal by the DAC module (digital-to-analog converter). Finally, the analog signal outputs current to the spherical feedback coil through the current output module. Through feedback adjustment, the values ​​of the three components of the reference time difference type fluxgate sensor are made to approach zero. The processing procedure of the digital signal processing module includes:

[0179] Acquire the output data of the reference time difference type fluxgate sensor;

[0180] A preliminary statistical characteristic model of the noise is constructed, and an objective function for minimizing the noise is set. The feedback parameters are iteratively adjusted with a fixed step size.

[0181] In each iteration, the degree of realization of stochastic resonance under the current feedback parameter settings is evaluated by comparing the output data with the expected output.

[0182] If the ideal state is not achieved, the feedback parameters are adjusted based on the error signal.

[0183] The acquisition module may also include a voltage follower to isolate the reference time difference fluxgate sensor from subsequent circuits, while improving the output drive capability of the reference time difference fluxgate sensor. A DSP module runs a digital integration algorithm, filters, and denoises, then outputs the processed digital signal to the DAC module. However, the output digital signal power is low and cannot directly drive the spherical feedback coil. Therefore, the output of the DAC module is connected to a current output module. The current output module uses an operational amplifier U2. Since the current required by the spherical feedback coil may reach over 30mA, and ordinary operational amplifiers cannot provide sufficient output current, a push-pull buffer Q composed of two PNP transistors needs to be added after the current output module to enhance the forward and reverse current capability of the output. The output terminals of the two PNP transistors are connected to the inverting input terminal of the current output module. When using a power amplifier to drive the spherical feedback coil, excessive current may occur. The output terminals of the two PNP transistors are connected to a resistor R2 to limit the current magnitude, protecting the current output module and spherical feedback coil from overload damage before outputting to the spherical feedback coil.

[0184] In a magnetic field and magnetic gradient detection device according to an embodiment of this application, to achieve effective compensation for the environmental magnetic field, taking a maximum magnetic field strength of 50000 nT in a certain direction as a design example, the radius R of the spherical shell is set to 16 cm, and the output current I of the current output module is required to be greater than 30 mA. When the number of slots for the spherical feedback coil is 12, the slot spacing is set to 0.8 cm, the slot width is 0.2 cm, and the distance from the enameled wire winding slot to the center of the spherical shell is... Meanwhile, enameled wire with a diameter of 0.1mm is used, and 25 turns are wound around each groove.

[0185] according to In the formula B is the magnetic flux density. Where is the dielectric constant. Feedback coil current, The radius of the feedback coil is the distance between the location to be calculated and the spherical feedback coil. When =0, the number of turns can be derived. ,at this time Turns / m, then calculate according to the formula for the number of turns. M is the number of revolutions, L is the length, and N can be derived from this. lock up.

[0186] The maximum magnetic field generated by the spherical feedback coil is calculated as follows: B = It can be known that B 50000nT.

[0187] In this embodiment, the magnetic gradient tensor measurement mainly relies on the difference between two three-component time-difference fluxgate sensors. Therefore, the error caused by magnetic field inhomogeneity has a negligible impact on the tensor output accuracy. Based on this characteristic, the baseline distance between the two three-component time-difference fluxgate sensors is set to 12cm. For the other two directions, the same design concept and method are used. Finally, the three time-difference fluxgate sensors and one reference time-difference fluxgate sensor are arranged in a cross-shaped structure with a spacing of 12cm.

[0188] The embodiments of this application can effectively complete the measurement of the magnetic gradient tensor. In this case, the time-difference fluxgate sensor operates stably near a zero magnetic field, reducing nonlinear errors and improving measurement accuracy. It is important to note that this does not mean the magnetic field is completely canceled out to an absolute zero. In practical applications, there will always be a certain residual magnetic field, which will cause the time-difference fluxgate sensor to produce a corresponding output. This not only ensures high stability and accuracy but also avoids the effects of crosstalk.

[0189] Compared to traditional fluxgate sensors, this application's embodiment employs a time-difference fluxgate sensor based on a novel time-difference measurement principle. This completely solves the problems of signal drift and weak anti-interference capability inherent in traditional amplitude-type fluxgate sensors, significantly improving the accuracy of magnetic field measurement and demonstrating a major advantage in high-end detection fields with extremely high magnetic field precision requirements. Furthermore, the spherical feedback coil exhibits superior performance through its magnetic field feedback mechanism. Compared to other types of feedback coils, the spherical feedback coil can generate a more uniform and wider-ranging magnetic field. While maintaining the same magnetic field uniformity, the spherical feedback coil is smaller and has higher space utilization. By integrating four time-difference fluxgate sensors into a single triaxial spherical feedback coil, a stable zero-magnetic-field operating environment is effectively created. This fundamentally avoids mutual interference caused by the presence of feedback coils when sensors provide individual feedback, greatly simplifies the system calibration process, and significantly shortens the baseline distance.

[0190] Compared to existing technologies, this application's embodiments avoid the shortcomings of amplitude-type fluxgate sensors from the detection principle perspective, employing a time-difference fluxgate sensor for detection. Through method optimization, precise time difference measurements are obtained, thus providing higher measurement accuracy. High-resolution computational parameters can be set according to requirements, effectively reducing errors and enabling more accurate processing of fluxgate sensor signals. This is particularly suitable for applications requiring high precision in magnetic field measurement. For measurements in complex electromagnetic environments, adaptive adjustment of feedback parameters ensures proactive adaptation to different environments, improving adaptability and robustness. Furthermore, a complete signal denoising method for three-component time-difference fluxgate sensors is employed, utilizing wavelet transform decomposition and particle swarm optimization (PSO) algorithm-based neural network optimization. This method deeply integrates wavelet domain feature extraction, dynamic parameter PSO optimization, and neural network parameter initialization: on one hand, wavelet transform reduces signal complexity; on the other hand, improved PSO compression factor constraints and weight reduction strategies achieve automated and precise adaptation of network parameters. This accurately captures the high-frequency component changes corresponding to the three-axis time difference that the magnetic gradient tensor depends on in the three-component time-difference fluxgate sensor measurement, significantly improving denoising performance and making the measured magnetic gradient tensor values ​​more accurate. Combining the high permeability and low coercivity of the soft magnetic core in a time-difference fluxgate sensor, the induced signal exhibits "narrow pulse and high amplitude" characteristics. This application matches the wavelet transform with the nonlinear characteristics of the core magnetization process, not only preserving the spike pulses of the saturation stage but also utilizing its frequency domain localization characteristics to accurately filter out Barkhausen noise during the magnetization of the soft magnetic core in the time-difference fluxgate sensor. Compared to traditional single-method denoising, this enhances the fitting ability of the induced signal, effectively suppresses signal distortion, and improves denoising accuracy. Addressing the abrupt changes in the full tensor magnetic gradient caused by sudden time-difference anomalies due to external boundaries in magnetic gradient measurement, a dynamic window-based isolated forest anomaly detection method is designed. This solves the problems of dynamic parameter coupling, adaptive adjustment of VMD modal parameters and wavelet thresholds, and cross-domain collaborative denoising (time-domain anomaly processing → frequency-domain modal separation → multi-scale wavelet refinement). Considering the complexity of the measured magnetic field, the measurement of the full magnetic gradient tensor depends on the multi-scale characteristics of time difference. An entropy-weighted adaptive VMD decomposition module is constructed to address the entropy-weighted decision fusion mechanism (quantifying the value of modal information with energy entropy to avoid feature dilution caused by equal weighting). Furthermore, the uniform magnetic field generated by the spherical feedback coil is stronger than other types of feedback coils, and under the same uniformity conditions, the spherical feedback coil occupies the least space. By arranging multiple time-difference fluxgate sensors in a triaxial spherical feedback coil, stable operation of these time-difference fluxgate sensors in a near-zero magnetic field environment can be ensured, avoiding potential mutual interference that may occur when the individual time-difference fluxgate sensors provide independent feedback.This application not only facilitates system calibration but also significantly shortens the distance between baselines, thereby improving measurement accuracy. It is suitable for scenarios with stringent requirements for magnetic field precision, such as scientific research experiments, high-end industrial testing, and geological exploration.

[0191] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for detecting magnetic fields and magnetic gradients, characterized in that, include: Three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor are placed inside a spherical shell, forming a cross-shaped structure centered on the origin in space. Two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis. A spherical feedback coil is wound in three enameled wire winding grooves that intersect perpendicularly along the X, Y, and Z directions, respectively, on the surface of the spherical shell. Adjust the current in the spherical feedback coil to make the values ​​of the three components of the reference time difference type fluxgate sensor approach zero; Excitation signals are passed into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor; Acquire the output signals of the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor respectively; The time difference is calculated using the output signal; Calculate the measured magnetic field based on the time difference; The magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors are obtained by superimposing the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

2. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, The excitation signal is a triangular wave, and the resulting excitation magnetic field is expressed as follows: In the formula For the slope of the triangular wave excitation magnetic field, For time, For the number of cycles, For the incentive cycle, To excite the magnetic field.

3. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, The formula for calculating the measured magnetic field based on the time difference is: In the formula For the measured magnetic field, The magnetic field frequency, For time difference.

4. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, The formula for calculating the magnetic flux density signal of different three-component time-difference fluxgate sensors, based on the superposition of the measured magnetic field and the excitation magnetic field, is as follows: B is the magnetic flux density signal. For vacuum permeability, The relative permeability of the core material. To excite the amplitude of the magnetic field.

5. The magnetic field and magnetic gradient detection method according to claim 3, characterized in that, A magnetic gradient tensor is constructed based on the magnetic induction intensity signals along different axes on a three-component time-difference fluxgate sensor and a reference time-difference fluxgate sensor, including: The gradients of the magnetic field components in the X, Y, and Z directions are obtained by dividing the difference in magnetic field induction intensity between two three-component time-difference fluxgate sensors on the X-axis by the distance between the two three-component time-difference fluxgate sensors in the X direction. The gradients of the magnetic field components in the X, Y, and Z directions are obtained by dividing the differences in magnetic induction intensity between the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor on the Y-axis by the distance between the three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor in the Y-axis. The gradient of the magnetic field component in the Z direction is obtained by taking the negative sum of the gradients of the magnetic field component in the X direction and the gradients of the magnetic field component in the Y direction.

6. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, Adjusting the current in the spherical feedback coil to bring the values ​​of the three components of the reference time difference fluxgate sensor close to zero, including: Acquire the output data of the reference time difference type fluxgate sensor; A preliminary statistical characteristic model of the noise is constructed, and an objective function for minimizing the noise is set. The feedback parameters are iteratively adjusted with a fixed step size. In each iteration, the degree of realization of stochastic resonance under the current feedback parameter settings is evaluated by comparing the output data with the expected output. If the ideal state is not achieved, the feedback parameters are adjusted based on the error signal.

7. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, Wavelet transform decomposition is performed on the output signals of the three three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensor. High-frequency noise components are filtered out by a threshold function while retaining key features to obtain the denoising coefficient. The denoising coefficients are reconstructed to generate a low-noise signal, which is then input into the untrained neural network. The improved particle swarm optimization algorithm is started, the particle velocity, position and dynamic parameters are initialized, and the particle state is adjusted in real time during the iteration. When the global fitness accuracy is met or the maximum number of iterations is reached, the optimal particle position sequence is output as the initial weights and threshold of the neural network. Drive the neural network to train until the output error meets the set conditions; The output of a neural network is the denoised output signal.

8. The magnetic field and magnetic gradient detection method according to claim 1, characterized in that, After calculating the time difference from the output signal, the obtained time difference is denoised, including: Anomaly detection in isolated forests is performed on time differences. An anomaly detection model is constructed by dynamic window segmentation and adaptive parameter adjustment. For each detected anomaly, the location and degree of anomaly are recorded. For each anomaly, surrounding normal points are selected as reference points and the weighted average is used to replace the anomaly to obtain a preprocessed signal. The preprocessed signal is input into the VMD decomposition module, and the mode number K and penalty factor are adaptively selected based on the characteristics of the preprocessed signal. The preprocessed signal is decomposed into K modal components and the energy entropy of each component is calculated. By normalizing the entropy value, weights are assigned according to the energy entropy of each modal component and then weighted and fused to generate an intermediate signal with enhanced main features. The intermediate signal with enhanced main features is subjected to hierarchical adaptive threshold calculation. Soft threshold denoising is applied layer by layer and the processing progress is iteratively judged. The time difference is reconstructed after all levels are completed.

9. A magnetic field and magnetic gradient detection device, characterized in that, include: Three three-component time-difference fluxgate sensors and one reference time-difference fluxgate sensor are placed inside a spherical shell, forming a cross-shaped structure centered on the origin of the coordinate system. Two of the three-component time-difference fluxgate sensors are symmetrically distributed along the X-axis, and the other three-component time-difference fluxgate sensor and the reference time-difference fluxgate sensor are symmetrically distributed along the Y-axis. The surface of the spherical shell is engraved with three enameled wire winding grooves that intersect perpendicularly in the X, Y, and Z directions, respectively, for winding the spherical feedback coil; The controller is used to adjust the current in the spherical feedback coil so that the values ​​of the three components of the reference time difference fluxgate sensor approach zero; The detection circuit is used to pass excitation signals into three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor; Used to acquire the output signals of three three-component time-difference fluxgate sensors and a reference time-difference fluxgate sensor; Used to calculate the time difference from the output signal; Used to calculate the measured magnetic field based on the time difference; This is used to obtain the magnetic induction signals of different three-component time-difference fluxgate sensors and reference time-difference fluxgate sensors based on the superposition of the measured magnetic field and the excitation magnetic field. The magnetic gradient tensor is constructed based on the magnetic induction intensity signals of different axes on the three-component time-difference fluxgate sensors and the reference time-difference fluxgate sensors.

Images