Sensor-based inductance winding loss measurement method and system

Abstract

The invention relates to the technical field of inductance loss measurement, and discloses a sensor-based inductance winding loss measurement method and system. The method comprises the following steps: synchronously acquiring a real-time current signal, a voltage signal and a surface temperature distribution signal during operation of the hollow winding inductor through an array type electrical sensor; and calculating a skin depth threshold value under the current working condition based on the skin effect dynamic model in combination with the current signal frequency and the conductor material parameters, and performing multi-channel adaptive filtering on the current signal to extract a matched effective current component. A stray capacitance compensation coefficient matrix is generated by combining the spatial distribution parameters of the segmented winding structure, and the instantaneous power loss value of each segment is calculated according to the effective current component, the voltage signal and the compensation coefficient matrix; thermal coupling errors are corrected through the mapping relation between the temperature distribution signals and the power loss values, the corrected loss values of all the segments are finally integrated, the comprehensive loss index of the hollow winding inductor is output, and the method is suitable for the loss evaluation scene of the hollow winding inductor.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of inductance loss measurement, in particular to a sensor-based inductance winding loss measurement method and system. BACKGROUND

[0002] In the fields of power electronics, new energy power generation, industrial control, etc., as a core component for energy storage, filtering, and signal coupling, the power loss of the hollow winding inductor during operation directly affects the efficiency, stability, and service life of the entire circuit system. Accurate measurement of the winding loss of the hollow winding inductor is an important prerequisite for optimizing inductor design and improving system performance.

[0003] Currently, the winding loss of the hollow winding inductor is mainly measured by traditional electrical measurement methods, such as direct measurement by power meter, voltage and current integration method, etc. These methods rely on single-channel sensors to collect current and voltage signals, making it difficult to achieve synchronous acquisition of multiple parameters. In actual operation, the current signal of the hollow winding inductor is often accompanied by harmonic interference, and is affected by changes in working conditions, so the current frequency will fluctuate dynamically. The traditional method lacks targeted filtering processing, which easily leads to the inclusion of invalid interference components in the collected current signal, thereby affecting the accuracy of the loss calculation.

[0004] The conductor of the hollow winding inductor will produce a skin effect under the action of alternating current, and the skin depth changes with the current frequency and the material of the conductor, directly affecting the distribution of current on the cross section of the conductor, and further affecting the winding loss value. Existing measurement methods use fixed skin depth parameters for calculation, without considering the influence of dynamic changes in working conditions on the skin effect, which cannot accurately match the current distribution characteristics under the current operating state, resulting in deviations between the calculated loss and the actual loss.

[0005] The hollow winding inductor is often wound in sections to optimize heat dissipation and structural stability, and parasitic capacitance is generated during the sectional winding process. These parasitic capacitances will form additional leakage currents under high-frequency working conditions, which will interfere with the measurement accuracy of voltage and current signals. Existing methods do not construct a parasitic capacitance compensation mechanism for the spatial distribution characteristics of the sectional winding structure, making it difficult to eliminate the influence of parasitic capacitance on loss calculation, further reducing the reliability of the measurement results.

[0006] The hollow winding inductor generates power loss during operation, and the heat generated by the loss causes uneven temperature distribution on the winding surface, while temperature changes affect the resistance characteristics of the conductor, forming a thermal coupling effect. Existing measurement methods do not establish a correlation between temperature distribution and power loss, and cannot correct the error of the sectional loss calculation caused by the thermal coupling effect, making the final output comprehensive loss index difficult to accurately reflect the actual loss state of the inductor.

[0007] With the development of power electronic systems towards high frequency, miniaturization, high efficiency, the precision requirement of air-core winding inductance winding loss measurement is continuously improved, and the shortcomings of traditional measurement methods in multi-parameter synchronous acquisition, dynamic skin effect adaptation, parasitic capacitance compensation, thermal coupling error correction and other aspects are increasingly prominent, which cannot meet the fine measurement demand in actual application scene, and a kind of inductance winding loss measurement method considering the above influencing factors is needed. SUMMARY

[0008] The purpose of the present application is to provide a sensor-based inductance winding loss measurement method and system to solve the problems raised in the background art.

[0009] To achieve the above purpose, the present application provides a sensor-based inductance winding loss measurement method, which comprises:

[0010] Synchronously collecting real-time current signal, voltage signal and surface temperature distribution signal of air-core winding inductance in running state by array type electrical sensor;

[0011] Based on the skin effect dynamic model, the skin depth threshold value under the current working condition is calculated according to the current signal frequency and the conductor material parameters;

[0012] The collected current signal is subjected to multi-channel adaptive filtering processing, and the effective current component matched with the skin depth threshold value is extracted;

[0013] Combining the spatial distribution parameters of segmented winding structure, a parasitic capacitance compensation coefficient matrix is generated;

[0014] According to the effective current component, the voltage signal and the compensation coefficient matrix, the instantaneous power loss value of each winding segment is calculated;

[0015] Through the mapping relationship between the temperature distribution signal and the power loss value, the thermal coupling error of the segmented winding structure is corrected;

[0016] The power loss values of each segment after correction are integrated, and the comprehensive loss index of air-core winding inductance is output.

[0017] Preferably, the calculation process of the skin effect dynamic model is specifically:

[0018] The resistivity, magnetic permeability of the conductor material and the fundamental frequency of the current signal are obtained;

[0019] The product of the fundamental frequency and the magnetic permeability is used to generate a frequency-dependent factor;

[0020] The ratio of the resistivity and the frequency-dependent factor is squared to obtain the dynamically updated skin depth threshold value.

[0021] Preferably, the multi-channel adaptive filtering processing comprises:

[0022] A signal channel corresponding to the number of winding segments is established, and each channel is configured with an independent band-pass filter;

[0023] The center frequency and bandwidth of each channel filter are adjusted according to the skin depth threshold value;

[0024] The current signal is subjected to time-frequency decomposition by using a sliding time window algorithm to suppress high-frequency harmonic interference.

[0025] Preferably, the generation process of the parasitic capacitance compensation coefficient matrix is specifically as follows:

[0026] The dielectric constant, interlayer spacing and wire geometry topology of the insulation layer in the segmented winding structure are obtained;

[0027] A basic compensation unit is generated according to the ratio of the dielectric constant to the interlayer spacing;

[0028] The electric field distribution weight of the wire geometry topology is superimposed to construct a multi-dimensional compensation coefficient matrix.

[0029] Preferably, the calculation process of the instantaneous power loss value is specifically as follows:

[0030] The effective current component is subjected to phase alignment processing with the voltage signal;

[0031] The aligned signal is subjected to matrix point multiplication operation according to the segment number;

[0032] The point multiplication result is weighted and corrected by introducing the compensation coefficient matrix.

[0033] Preferably, the thermal coupling error correction includes:

[0034] A linear regression model of temperature gradient and power loss is established;

[0035] The thermal conduction offset between adjacent segments is calculated according to the surface temperature distribution signal;

[0036] The distribution weight of the power loss value is iteratively adjusted by using a back propagation algorithm.

[0037] Preferably, the method further includes:

[0038] The frequency drift of the current signal is monitored in real time, and the dynamic model is recalculated when the drift exceeds a preset threshold value;

[0039] The filter channel parameters and the compensation coefficient matrix are updated according to the recalculated result.

[0040] Preferably, the output process of the comprehensive loss index is specifically as follows:

[0041] The power loss values of the segments after correction are subjected to time integration operation;

[0042] The superposition integral result and a dynamic compensation term of the frequency drift amount;

[0043] After normalization processing, a standardized loss index sequence is generated.

[0044] Preferably, the method further comprises:

[0045] By the redundant check mechanism of the electrical sensor array, abnormal temperature or current data points are eliminated;

[0046] When the continuous abnormal data exceeds the fault tolerance upper limit, a hardware self-check signal is triggered.

[0047] Preferably, the application further comprises a sensor-based inductance winding loss measurement system, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-mentioned sensor-based inductance winding loss measurement method.

[0048] Compared with the prior art, the application has the following beneficial effects:

[0049] Through the arrayed electrical sensor, real-time current signal, voltage signal and surface temperature distribution signal under the operation state of the hollow winding inductance are synchronously collected, compared with the traditional single-channel sensor collection mode, electrical parameters and thermal parameters of the inductance operation can be simultaneously obtained, the information lag problem caused by the time difference in the collection process of different parameters is avoided, the correlation and timeliness between multi-dimensional parameters are ensured, more comprehensive and synchronous basic data are provided for subsequent loss calculation, and the input parameters of the loss calculation are more suitable for the actual operation state of the inductance.

[0050] In the skin effect processing aspect, the method calculates the skin depth threshold under the current working condition based on the skin effect dynamic model, in combination with the current signal frequency and the conductor material parameter, instead of using a fixed skin depth parameter. This dynamic calculation method can adapt to the fluctuation of the current frequency and the inherent characteristics of the conductor material in the inductance operation process in real time, and accurately reflect the distribution rule of the current on the conductor cross section under the current working condition. Meanwhile, through the multi-channel adaptive filtering processing of the collected current signal, the effective current component matched with the skin depth threshold can be extracted, and invalid signals such as harmonic interference can be filtered out, so as to reduce the current parameter error caused by the dynamic change of the skin effect and signal interference, and provide more accurate current basic data for subsequent segmented power loss calculation.

[0051] For the problem of parasitic capacitance generated by the segmented winding structure of the air-core inductor, the method generates a parasitic capacitance compensation coefficient matrix in combination with the spatial distribution parameters of the segmented winding structure. The spatial distribution parameters of the segmented winding structure are directly related to the size and distribution characteristics of the parasitic capacitance. The compensation coefficient matrix constructed based on the parameters can accurately match the parasitic capacitance characteristics of different winding segments. During the loss calculation process, the voltage signal and the current signal are compensated specifically to eliminate the interference of the leakage current formed by the parasitic capacitance under high-frequency conditions, ensure the accuracy of the instantaneous power loss value calculation of each winding segment, and avoid the loss calculation deviation caused by the uncompensated parasitic capacitance.

[0052] In the thermal coupling error correction link, the method corrects the thermal coupling error of the segmented winding structure through the mapping relationship between the temperature distribution signal and the power loss value. During the operation of the air-core inductor, the heat generated by the power loss changes the temperature distribution of the winding surface, and the temperature change affects the conductor resistance, forming a thermal coupling effect. By establishing the mapping relationship between the temperature distribution and the power loss, the influence of the temperature factor on the resistance and the loss calculation can be included in the correction category, and the loss calculation results of each segmented winding structure are adjusted, so that the segmented loss value is more consistent with the actual loss under the heating state, and the interference of the thermal coupling effect on the final comprehensive loss index is reduced.

[0053] By integrating the corrected power loss values of each segment, the comprehensive loss index can be output, which can realize accurate aggregation from segmented loss to overall loss, avoid the overall loss evaluation deviation caused by the accumulation of segmented loss calculation errors. Compared with the traditional method of directly calculating the overall loss, this segmented calculation and correction integration mode can more accurately capture the loss differences of different winding segments, clearly reflect the loss distribution characteristics of each part of the inductor, not only output accurate comprehensive loss index, but also provide more specific loss distribution information for inductor winding structure optimization, heat dissipation design improvement, etc., and help the subsequent inductor performance improvement work. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The working principle diagram of the inductor winding loss measurement method based on sensors described in the present application;

[0055] Figure 2 The flowchart of multi-channel adaptive filtering processing;

[0056] Figure 3 The flowchart of parasitic capacitance compensation coefficient matrix generation. DETAILED DESCRIPTION

[0057] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0058] Please refer to Figure 1 The present application provides a sensor-based inductance winding loss measurement method, which comprises: an array-type electrical sensor is arranged on the surface and connection terminal of a hollow-wound inductance to capture current signal, voltage signal and surface temperature distribution signal in real time. The acquisition process ensures signal synchronization and avoids measurement errors caused by time offset. A skin effect dynamic model dynamically calculates the skin depth threshold according to the current signal frequency and the conductor material parameters, including resistivity and magnetic permeability, which are obtained from the inductance design specifications. Multi-channel adaptive filtering processing is performed on the acquired current signal, the number of filtering channels matches the segmented winding structure of the hollow-wound inductance, each channel is independently configured with a band-pass filter, and the filter parameters are adjusted in real time according to the skin depth threshold. After the effective current component is extracted, the parasitic capacitance compensation coefficient matrix is generated in combination with the spatial distribution parameters of the segmented winding structure, which involves the dielectric constant of the insulating layer, the interlayer spacing and the winding geometric topology. The instantaneous power loss value calculation is based on the effective current component, the voltage signal and the compensation coefficient matrix, and the power evaluation of each winding segment is realized through matrix operation. There is a mapping relationship between the temperature distribution signal and the power loss value, which is used to correct the thermal coupling error of the segmented winding structure, and the correction process considers the influence of temperature gradient. The integrated power loss values of each segment after correction are output as a comprehensive loss index, which includes numerical sequence or graphical display.

[0059] Embodiment 1: Please refer to Figure 2, the skin effect dynamic model requires the resistivity and permeability of the conductor material, which are obtained from the material technical specification of the air-core inductor. The technical specification is stored in the system memory in the form of an electronic table. The fundamental frequency of the current signal is obtained by real-time monitoring of the current waveform. The frequency spectrum characteristics of the current signal are analyzed using the fast Fourier transform algorithm. The number of sampling points for frequency spectrum analysis is set to 1024 points. The Hanning window function is applied to the sampled data to reduce spectral leakage. The frequency-dependent factor is calculated by multiplying the fundamental frequency and the permeability. The multiplication operation is performed using a 32-bit floating-point processing unit. The operation result is temporarily stored in the cache register. The dynamically updated skin depth threshold is obtained by mathematical operation of the ratio of the resistivity and the frequency-dependent factor. The square root operation uses the Newton-Raphson iteration algorithm. The iteration accuracy is set to six decimal places. The convergence condition is that the difference between the results of the adjacent two iterations is less than one ten-thousandth. The multi-channel adaptive filtering process requires the establishment of a signal channel consistent with the number of winding segments. The number of signal channels is determined according to the physical structure of the air-core inductor. Each signal channel is configured with an independent digital bandpass filter. The design of the digital bandpass filter uses an infinite impulse response filter structure. An eight-order elliptic filter is selected to achieve steep transition band characteristics. The skin depth threshold is used to dynamically adjust the center frequency and bandwidth of each channel filter. The adjustment mechanism is based on the pre-established frequency-bandwidth mapping table. The mapping table contains the filter parameter combinations corresponding to different skin depth thresholds. The sliding time window algorithm is used for time-frequency decomposition of the current signal. The Blackman-Harris window is selected to optimize the frequency resolution. The window length is set to an integer multiple of the fundamental period to avoid spectral leakage. The suppression of high-frequency harmonic interference is realized through multi-stage filtering. The first-stage filter eliminates frequency components above the Nyquist frequency. The second-stage filter sets the notch characteristics for specific harmonic frequencies.

[0060] The resistivity data of the conductor material is temperature compensated when read from the material database, and the compensation coefficient is calculated according to the temperature coefficient curve of the conductor material, which is stored in the read-only memory in the form of a polynomial. The value of magnetic permeability considers the influence of frequency change, and the magnetic permeability attenuation at high frequency is compensated by a correction factor, which is determined based on the frequency response characteristic curve of the magnetic core material. The detection of the fundamental frequency uses the autocorrelation algorithm to enhance the anti-interference ability, and the frequency corresponding to the peak value of the autocorrelation function is the fundamental frequency, and the frequency resolution reaches zero point one hertz. The calculation of the frequency dependent factor introduces temperature drift compensation, and the compensation amount is adjusted in real time according to the reading of the environmental temperature sensor, which is integrated in the electrical sensor array. The parameter adjustment of the digital band-pass filter uses the gradient descent method for optimization, and the optimization goal is to minimize the passband ripple and stopband attenuation, and the optimization process is completed in the initialization stage of the filter. The filter parameters of each signal channel are independently stored in the multi-port random access memory, and the memory stores the coefficient matrix according to the channel number. The adjustment of the center frequency follows the principle of linear interpolation, and linear interpolation calculation is performed between two adjacent preset frequency points, and the interpolation step is set to one tenth of hertz. The adjustment of the bandwidth maintains a constant percentage relationship with the center frequency, and the percentage value is dynamically adjusted according to the signal-to-noise ratio, which is calculated by analyzing the signal power spectral density. The implementation of the sliding time window algorithm uses a ring buffer structure, and the buffer size matches the window length, and the data update uses a pointer circular moving method. The time-frequency decomposition uses the short-time Fourier transform algorithm, and the transform length is set to two hundred and fifty-six sampling points, and the overlap rate is set to fifty percent to ensure time continuity. The suppression of high-frequency harmonic interference increases the adaptive noise cancellation link, and the noise reference signal is extracted from the frequency band far from the fundamental wave, and the cancellation algorithm uses the least mean square algorithm to update the filter weight. The application of Hanning window function cooperates with windowed fast Fourier transform, and the window function coefficient is calculated and stored in the lookup table in advance, which reduces the real-time calculation amount.

[0061] The signal synchronization of the multi-channel adaptive filtering process is realized by a global timer, and the sampling time deviation of each channel is controlled at the nanosecond level. The phase consistency between channels is maintained by a digital delay-locked loop, which compares the zero-crossing time of signals of each channel and generates a phase correction signal. The filtered signal is subjected to amplitude normalization processing, and the normalization coefficient is dynamically adjusted according to the maximum amplitude of the signal to avoid data overflow. The signal quality monitoring module evaluates the signal-to-noise ratio of the output signal of each channel in real time, and triggers an alarm signal when the signal-to-noise ratio is lower than the threshold. The update mechanism of the skin depth threshold is synchronized with the change of the current signal frequency, and the skin depth threshold is immediately recalculated when the frequency change exceeds the set threshold. The recalculating process is realized by an interrupt service program, and the interrupt priority is set to the highest level to ensure real-time performance. After the threshold is updated, the filter parameter adjustment adopts a smooth transition strategy, and the new parameter gradually replaces the old parameter within ten sampling periods to avoid signal mutation. The signal continuity during parameter adjustment is guaranteed by an interpolation algorithm, and the interpolation algorithm uses cubic spline interpolation to maintain waveform smoothness. The multi-channel processing of the current signal adopts a parallel computing architecture, and each channel is allocated an independent digital signal processor core. The cores communicate through a crossbar network. Direct memory access is used for data transmission to reduce the intervention of the central processing unit and improve processing efficiency. The data consistency between channels is maintained by a hardware synchronization signal, which is generated by dividing the main clock, and the division coefficient can be programmed. The processing delay is compensated by a timestamp mechanism, and each sampling point is attached with a high-precision timestamp for delay alignment in subsequent processing steps.

[0062] The stability monitoring of the filter is achieved by calculating the pole position, and the filter coefficients are automatically reset when the pole position exceeds the unit circle. The coefficient reset uses a backup parameter set, which is stored in non-volatile memory to prevent data loss. The filter performance evaluation is achieved by calculating the passband ripple and stopband attenuation, and the evaluation results are recorded in the system log for later analysis. The step size parameter of the adaptive filtering algorithm is dynamically adjusted according to the signal characteristics, and the step size is reduced to improve stability when the signal changes dramatically, and the step size is increased to speed up convergence when the signal is smooth. The calibration of the signal channel is completed by injecting a standard test signal, and the test signal frequency covers the entire working frequency band, and the calibration coefficients are stored in non-volatile memory. The gain difference between channels is compensated by a digital multiplier, and the multiplier coefficients are determined during calibration to ensure the amplitude consistency of each channel. The phase calibration adjusts the group delay using an all-pass filter, and the filter coefficients are calculated based on the phase difference between channels, and the phase difference is measured by the cross-correlation algorithm. The influence of temperature on the characteristics of the filter is eliminated by the temperature compensation coefficient, and the compensation coefficient is calibrated by temperature cycle test, and the calibration data is stored in the lookup table. The real-time performance of multi-channel adaptive filtering processing is guaranteed by the pipeline architecture, and the data acquisition, filtering processing and signal output form a three-stage pipeline. The data buffer between pipeline stages uses a double buffering mechanism to avoid data access conflicts. The processing timing is controlled by a programmable logic device, and the timing parameters are dynamically configured according to the sampling rate. The system resource allocation uses a static priority scheduling algorithm to ensure that high-priority tasks are responded to in a timely manner. The memory management uses a fixed block size allocation strategy to reduce memory fragmentation and improve access efficiency.

[0063] The anti-aliasing processing of the current signal is completed before analog-to-digital conversion, and the anti-aliasing filter uses a Butterworth low-pass filter with a cutoff frequency of half the sampling frequency. The resolution of the analog-to-digital converter is set to sixteen bits, and the sampling rate is set according to ten times the highest frequency component of the signal. The suppression of quantization noise is achieved by oversampling technology, and the oversampling multiple is set to four, which is combined with digital filtering to improve the signal-to-noise ratio. The signal conditioning circuit includes a programmable gain amplifier, and the gain is automatically adjusted according to the signal amplitude to fully utilize the dynamic range of the analog-to-digital converter. The fault detection of multi-channel adaptive filtering processing is achieved by comparing the output signals of each channel, and the abnormal channel is automatically isolated and the backup processing path is enabled. The fault record includes fault type, occurrence time and channel number, and the record data is used for system maintenance and fault analysis. The system self-checking program is run periodically, and the self-checking content includes memory test, processor load test and communication interface test. The self-checking result forms a health status report, which is uploaded to the monitoring system through the communication interface.

[0064] Example 2: see Figure 3The dielectric constant of the insulation layer is obtained from a technical document of a material supplier, the technical document is stored in a system database in a standardized electronic format, and the numerical value of the dielectric constant has a corresponding correction coefficient under different temperature conditions. The interlayer spacing is obtained by measuring the physical structure of the air-core inductor with a high-precision laser range finder, the measurement point is selected at the center position of each winding segment, and the measurement data is averaged by three repeated measurements to eliminate accidental errors. The winding geometric topology is collected by a three-dimensional scanner to obtain point cloud data of the air-core inductor, and the point cloud data is processed by triangular meshing to generate a digital model, which contains spatial coordinate information of the wire path. The calculation of the basic compensation unit is based on the ratio of the dielectric constant to the interlayer spacing, and the calculation process is performed by a double-precision floating-point operation unit, and the operation result is temporarily stored in a register file. The temperature correction of the dielectric constant is realized by querying a temperature-dielectric constant relationship table, the relationship table contains data points from minus forty degrees Celsius to one hundred and fifty degrees Celsius, and the data points are spaced five degrees Celsius apart. The measurement data of the interlayer spacing needs to be processed by surface fitting, and the fitting algorithm uses the least squares method to fit a quadratic surface to eliminate local distortion caused by measurement noise. The numerical range of the basic compensation unit is normalized, and the normalized reference value is the ratio of the typical dielectric constant of the insulation material to the standard interlayer spacing.

[0065] The electric field distribution weight of the winding geometric topology is calculated by a finite element analysis software, the material properties of the finite element analysis software are set to the actual used insulation material and conductor material, and the boundary condition is set to open circuit condition. The calculation of the electric field distribution weight considers the influence of the working frequency, the frequency value is taken from the fundamental frequency of the real-time monitored current signal, and the electric field distribution weight is recalculated when the frequency changes. The division of the finite element grid adopts an adaptive grid refinement technique, which automatically refines the grid in areas with large electric field gradients, and the grid quality coefficient is controlled to be above zero point eight. The output format of the electric field distribution weight is a three-dimensional tensor, and the three dimensions of the tensor correspond to the winding segment number, the spatial coordinate direction and the frequency point respectively. The construction of the multi-dimensional compensation coefficient matrix is realized by tensor multiplication, the basic compensation unit is multiplied with the electric field distribution weight tensor as a scalar factor, and the multiplication operation is performed in parallel on the graphics processing unit. The dimensions of the matrix are consistent with the number of winding segments of the winding structure, and each matrix element contains two parts of real and imaginary parts, which represent the amplitude and phase compensation of the capacitive effect respectively. The arrangement order of the matrix elements is according to the physical position number of the winding segments, and the numbering rule is arranged clockwise from the inner layer to the outer layer of the inductor. The symmetry check of the matrix is completed by comparing the difference between the transposed matrix and the original matrix, the difference threshold is set to five ten-thousandths, and the electric field distribution weight is recalculated when the threshold is exceeded.

[0066] The storage of the parasitic capacitance compensation coefficient matrix adopts a sparse matrix compression format, and the non-zero elements are stored in a coordinate list format to reduce the memory occupation space. The matrix access interface is designed in a direct memory access mode, supporting multi-thread concurrent reading operation, and the reading delay is controlled within ten clock cycles. The matrix update mechanism is triggered by the frequency monitoring module, and the matrix update process is started when the frequency change exceeds one percent. The update process is completed in the background thread without affecting real-time calculation. The matrix version management adopts a timestamp mechanism, and each matrix is attached with a generated timestamp to ensure that the used matrix matches the current frequency condition. The frequency characteristics of the dielectric constant of the insulating layer are measured by a wideband impedance analyzer, and the measurement frequency range covers ten hertz to ten megahertz. The measurement points are uniformly distributed in the logarithmic coordinates. The dielectric constant frequency variation curve is fitted into a Debye model, and the model parameters are stored in a parameter database. The dynamic change of the interlayer spacing considers the thermal expansion effect, and the thermal expansion coefficient is read from the material thermodynamic parameter table. The temperature data come from the integrated temperature sensor. The deformation analysis of the winding geometry topology introduces the influence of mechanical stress, and the stress distribution is calculated by a structural mechanics simulation software. The simulation conditions include installation stress and electromagnetic force. The calculation of the electric field distribution weight adds a correction of the proximity effect, and the correction factor is obtained by measuring the coupling capacitance between adjacent segments. The measurement uses an LCR meter at a specific frequency point. The frequency interpolation of the weight uses a cubic spline interpolation algorithm, and the interpolation nodes are selected as ten characteristic frequency points. The characteristic frequency points are logarithmically uniformly distributed according to the working frequency range. The normalization processing of the multi-dimensional compensation coefficient matrix adopts the maximum modulus normalization, and the normalization coefficient is the maximum value of the modulus of all elements of the matrix. The modulus of the normalized matrix elements ranges from zero to one. The condition number evaluation of the matrix is completed by calculating the singular value of the matrix. When the condition number is too large, the matrix reconstruction process is automatically triggered.

[0067] The boundary effect processing of the segmented winding structure is realized by adding virtual segments, which are set at both ends of the actual segments. The parameters of the virtual segments are determined by extrapolation. The electric field distribution weight at the boundary is calculated by the mirror method, and the mirror surface of the mirror method is set on the symmetry plane of the segmented structure. The verification of the compensation coefficient matrix is performed by comparing the calculated value with the measured value. The measured value is obtained by measuring the scattering parameters between the segments with a vector network analyzer and converting the capacitance parameters. The matrix correction factor is obtained by least square fitting the error between the calculated value and the measured value, and the correction factor is applied to all elements of the matrix. The real-time calling of the parasitic capacitance compensation coefficient matrix is realized by the memory mapping file, and the matrix data is mapped to the virtual address space of the process to realize fast access. The matrix operation is accelerated by using the single instruction multiple data stream instruction set, which supports simultaneous processing of four floating point operations. The block optimization of matrix multiplication divides the large matrix into small block matrices, and the block size is optimized according to the cache size of the processor to reduce the cache miss times. The matrix inversion operation adopts the LU decomposition algorithm, and the partial pivot selection method is used in the decomposition process to improve the numerical stability. The backup mechanism of the compensation coefficient matrix adopts the double memory bank design, which automatically switches to the backup memory bank when the main memory bank is damaged. The matrix integrity check is realized by the cyclic redundancy check code, which is stored in the header of the matrix file and checked every time it is read. The matrix compression algorithm uses dictionary encoding to compress the repeated mode, and the compression ratio reaches more than 50%. The matrix version rollback function retains the last five versions of the matrix data, and the rollback trigger conditions are matrix check failure or calculation abnormality.

[0068] The real-time update of the winding geometry topology is achieved through a visual recognition system, which uses a high-resolution industrial camera to capture inductance images, and an image processing algorithm to recognize changes in wire position. Topology change detection is triggered when the difference between the current topology and the reference topology exceeds a threshold, triggering matrix recalculation. The lighting conditions of the visual recognition system use uniform backlighting to eliminate the effects of shadows on image recognition. The image acquisition frequency is synchronized with the temperature sampling frequency, with each temperature sampling point corresponding to a topology image. Sensitivity analysis is used for uncertainty propagation analysis of matrix elements, calculating the partial derivative of each input parameter on the matrix element. Range checking of matrix application limits matrix elements to a physically reasonable range, with values outside the range automatically truncated to the boundary value. Exception handling for matrix operations includes division by zero checking, overflow checking, and illegal value checking, with exceptions recorded in the system log and triggering an alarm. The integration of the compensation coefficient matrix and the signal processing flow uses a pipeline architecture, with the matrix data preparation stage and the power calculation stage executed in parallel. The matrix cache management uses the least recently used algorithm to keep the most commonly used matrices in the cache. The matrix data format is standardized using IEEE754 floating-point numbers to ensure data compatibility between different platforms. The matrix access interface is encapsulated to provide thread-safe read-write lock mechanisms to prevent multi-thread access conflicts. Matrix life cycle management uses reference counting to automatically release memory resources when they are not in use.

[0069] In embodiment 3, the effective current component is phase-aligned with the voltage signal. The phase alignment is calculated using a cross-correlation function to determine the time delay between the two signals. The peak value of the cross-correlation function corresponds to the optimal alignment point. The digital signal processor has a built-in delay line buffer that stores historical sampling data. The buffer depth meets the storage requirements of the maximum expected delay. The aligned signals are reorganized into a three-dimensional data array according to the segment number. The first dimension represents the time sequence index, the second dimension corresponds to the winding segment number, and the third dimension saves the signal amplitude. Matrix point multiplication is performed on the floating-point operation unit, following the rule of element-by-element multiplication and summation. The point multiplication result represents the instantaneous power loss value without compensation. The parasitic capacitance compensation coefficient matrix serves as a weight factor in the calculation process. The combination of the weight factor and the instantaneous power value uses Hadamard product operation, which multiplies corresponding elements of matrices of the same dimension. The compensated power value needs to be normalized in dimension. The normalization coefficient is taken from the system's calibration parameter library, which is calibrated by standard instruments when the device is shipped. The thermal coupling error correction is based on the mapping relationship between the temperature distribution signal and the power loss value. The mapping relationship is described by a multiple linear regression model, and the model parameters are trained by historical operation data. The temperature distribution signal comes from the infrared temperature measurement unit of the array-type electrical sensor. The spatial resolution of the temperature measurement unit matches the winding segment size, with at least three temperature measurement points corresponding to each segment.

[0070] The processing of the surface temperature distribution signal includes noise filtering and outlier rejection. The noise filtering uses a median filter to protect the edge features of the temperature signal. The outlier rejection is based on a statistical outlier detection algorithm. The thermal conduction offset between adjacent segments is obtained by solving the thermal conduction differential equation. The equation is discretized using the finite difference method, and the boundary condition is set as an adiabatic boundary. The back propagation algorithm is used in error correction to adjust the distribution weight of the power value. The weight update amount is determined based on the gradient direction of the loss function, and the learning rate is set to an adaptive adjustment mode. The iteration convergence condition is set according to the relative change of the power value, and the change threshold is set to one thousandth. The maximum number of iterations is limited to two hundred times. The calculation of the instantaneous power loss value introduces a frequency compensation term based on the harmonic analysis result of the current signal, and the harmonic analysis uses the fast Fourier transform algorithm. The numerical integration in the power calculation process uses the composite Simpson rule, and the integration interval is adaptively divided according to the signal period. The integration step size and the sampling interval maintain an integer multiple relationship. The signal resampling process is performed before phase alignment, and the resampling algorithm uses a polynomial interpolation method. The interpolation order is selected as a cubic polynomial to balance accuracy and computational load. The signal amplitude calibration is realized by referring to the standard signal source, and the calibration curve is stored as a lookup table. Real-time calibration is completed through linear interpolation.

[0071] The thermal coupling error correction model considers the temperature dependence of the material thermal conductivity, and the thermal conductivity variation curve is read from the material property database, which contains measured data at different temperatures. The non-linear correction of the temperature sensor is realized by polynomial fitting, and the fitting coefficients are stored in the sensor's calibration certificate. The numerical stability of the heat conduction equation is guaranteed by controlling the time step, which meets the requirements of the Courant condition to avoid numerical divergence. The verification of the error correction result is completed by comparing the measurement data of the infrared thermal imager, and the spatial temperature distribution of the thermal imager is used as the reference benchmark. The visualization of the power loss value's spatial and temporal distribution is realized by the color mapping algorithm, which converts the power value to the hue component of the HSV color space with the saturation fixed at the maximum value. The real-time display refresh rate is synchronized with the data acquisition rate to avoid image flickering. The data storage adopts a hierarchical structure, with the original sampling data stored in the cache and the processing results written to the non-volatile memory. The storage format is selected as the standard HDF5 format, which supports metadata embedding and fast retrieval.

[0072] Thermal coupling error correction introduces a temperature compensation factor which is determined by the following relationship:

[0073]

[0074] where: represents the temperature compensation factor, represents the number of temperature sensors, represents the weight coefficient of temperature sensor, temperature rise value measured by the temperature sensor, maximum temperature difference allowed by the system. Temperature compensation factor is used to correct the thermal coupling error in power loss calculation, and the corrected power value is closer to the true loss. Both sides of the formula are dimensionless values, and the dimension is consistent. Weight coefficient According to the distance between the temperature sensor and the winding segment, the closer the distance, the greater the weight. Temperature rise value is the difference between the current temperature and the reference temperature, and the reference temperature is the initial temperature when the system starts. Maximum temperature difference read from the system safety operation parameters.

[0075] The timing synchronization of the signal processing link is controlled by the global clock signal generated by the FPGA, and the clock signal is distributed to each processing unit with a deviation of less than one nanosecond. The compensation of processing delay adopts the look-ahead buffer mechanism, and the buffer size is sufficient to accommodate the number of sampling points corresponding to the maximum processing delay. The system resource allocation adopts a static priority scheduling strategy, and the power calculation task is given the highest priority to ensure real-time requirements. Memory access optimization is achieved through data prefetching technology, and the prefetching algorithm is based on access pattern prediction of the next required data block. The temperature field reconstruction algorithm adopts the Kriging interpolation method, and the interpolation parameters are determined by the variogram model, and the model type is selected as the Gaussian model. The resolution enhancement of the reconstruction result is realized through super-resolution technology, and the technical basis is the convolutional neural network model, and the network weight is trained offline. The temperature gradient calculation adopts the central difference formula, and the difference step is automatically set according to the sensor spacing. The heat flow density distribution is calculated through the Fourier heat conduction law, and the temperature gradient in the law comes from the reconstructed temperature field.

[0076] The convergence of the error correction process is monitored by calculating the residual norm, and the norm type is selected as the two-norm, and the convergence threshold is set according to the measurement accuracy requirement. The numerical stability in the iteration process is maintained by the regularization method, and the regularization parameter is determined by the L curve method. The matrix inversion operation adopts the singular value decomposition algorithm, and the small singular value truncation threshold in the decomposition process is set to ten times the machine precision. The uncertainty analysis of the calculation result adopts the Monte Carlo method, randomly samples the input parameters within the measurement error range, and statistically analyzes the distribution characteristics of the output value. The real-time data processing pipeline includes five stages: data acquisition, signal conditioning, feature extraction, model calculation and result output. The data transfer between pipeline stages adopts the double buffer mechanism to avoid read-write conflict. The affinity of the processing thread is set to bind the calculation-intensive task to a specific CPU core, reducing the context switching overhead. Memory management uses the pooling technology to pre-allocate memory blocks of a fixed size, improving allocation efficiency.

[0077] The calibration procedure of the system includes two parts: direct current calibration and alternating current calibration. The direct current calibration uses a standard resistance load, and the alternating current calibration uses an impedance analyzer. The least squares fitting of the calibration data uses Legendre orthogonal polynomials to reduce the correlation between the coefficients. The temperature cycle test is performed in an environmental chamber, with a temperature range of minus twenty-five degrees Celsius to plus eighty-five degrees Celsius, and a test point interval of ten degrees Celsius. The long-term stability evaluation is performed by running continuously for one thousand hours, and recording the change trend of the key parameters every hour. The fault detection and diagnosis system monitors the signal quality indicators, including signal-to-noise ratio, harmonic distortion, and direct current offset. The abnormal situation triggers a hierarchical response mechanism, and the slight abnormality records the log, and the serious abnormality starts the protection process. The system self-checking program runs at startup, and the self-checking range covers the sensor interface, memory, and communication module. The maintenance interface supports parameter configuration and diagnostic data export, and the configuration protocol uses encrypted transmission to ensure security.

[0078] In embodiment 4, the frequency drift is monitored by short-time Fourier transform of the current signal, and the transform window length is dynamically adjusted according to the signal stability. The window length setting follows the principle of integer multiple of signal period. The preset threshold is set according to the rated working frequency of the air-core inductor, and the threshold exceeding triggers the skin effect dynamic model recalculation. The recalculation process updates the filter channel parameters and compensation coefficient matrix, and the update operation is performed in the interrupt service routine to avoid blocking the main process. The output of the comprehensive loss index performs time integration operation on the power loss value of each segment after correction, and the integration interval is from the measurement start time to the current time. The integral method uses the trapezoidal rule. The dynamic compensation term of the frequency drift is added to the integral result as an additional term, and the compensation term coefficient is determined by the correlation between the frequency drift and the power loss. The integral result is scaled to the standard range by normalization, and the normalization reference is taken from the rated power value of the air-core inductor. The standardized loss index sequence is output in array form, and the sequence length matches the sampling period. The output interface supports serial communication or file storage.

[0079] The frequency monitoring module processes the current signal using a sliding window mechanism, with a window width set to sixteen times the fundamental period and a window sliding step of one quarter of the window width. The spectrum analysis uses the Blackman-Harris window function to reduce spectral leakage, with a frequency resolution of 0.5 Hz. The frequency drift calculation is achieved by tracking the change in the fundamental frequency peak position, with a parabolic interpolation method used for peak detection to improve positioning accuracy. The threshold comparator uses a hysteresis comparison method to prevent frequent triggering, with a hysteresis width set to plus or minus 2% of the rated frequency. The response time of the interrupt service routine is controlled within 10 microseconds, with an interrupt priority set to the highest level of the system. The skin effect dynamic model recalculation process calls the material parameter database, with direct memory mapping used to reduce delays. The filter channel parameter update includes center frequency, bandwidth, and gain coefficient, with floating-point operation units used for parallel processing. The update of the compensation coefficient matrix uses an incremental update strategy, updating only the matrix elements affected by frequency changes. Memory management uses a double buffering mechanism, with smooth transition of new and old parameters to avoid signal jumps. Parameter verification is achieved through a boundary check algorithm, with default parameters used to replace parameters that exceed the reasonable range.

[0080] The time integration operation uses an accumulator structure, with an accumulator word length set to 64 bits to prevent overflow. The integration step size is adaptively adjusted, with a smaller step size used when the signal changes dramatically to improve accuracy, and a larger step size used when the signal is stable to improve efficiency. The calculation of the dynamic compensation term is based on frequency drift history data, with a history data length of the last 100 sampling points. The compensation coefficient is determined through linear regression analysis, with the regression model recalibrated every five minutes. Temperature compensation factors are introduced in the normalization process, with temperature data from an integrated temperature sensor. The output data format uses the standard IEEE floating-point number format, with a data packet structure including a timestamp, loss value, and status word. The communication protocol supports two industry standards, MODBUS and PROFIBUS, with protocol conversion achieved through a programmable logic device. Data storage uses a circular buffer method, with a buffer size that can store 24 hours of continuous data. The storage data compression uses the LZ77 algorithm, with a compression ratio of 3:1. Remote data transmission encryption uses the AES256 algorithm, with the key updated every 24 hours.

[0081] The system self-diagnosis function monitors the running state of each processing module in real time, with state information output through LED indicator lights and digital interfaces. Fault codes are recorded in non-volatile memory, with fault history traceable to the last 100 events. The watchdog timer monitors the program running state, triggering a system restart when the timeout is not reset. The power management module monitors the supply voltage and current, starting the backup power switch when there is an anomaly, as shown in Table 1.

[0082] Table 1: Frequency drift monitoring parameter configuration table

[0083]

[0084] The signal processing pipeline adopts a four-stage architecture with data check points between each stage. The check algorithm uses cyclic redundancy check, and data retransmission is triggered when the check fails. Pipeline throughput optimization is achieved through instruction rearrangement, and the compiler optimization level is set to three. The cache prefetch strategy is based on data access pattern prediction, with a prediction accuracy of over ninety percent. Memory access conflicts are avoided through cross-addressing, and the addressing scheme uses a low cross-mode. The calculation of the temperature compensation factor takes into account the rate of change of the ambient temperature, which is obtained through differential operation. The temperature sensor data is denoised using median filtering with a window size of five sampling points. The dew point temperature calculation is introduced for humidity compensation, and the dew point temperature is solved by the Magnus formula. The atmospheric pressure compensation uses standard atmospheric pressure conversion, and the conversion coefficient is obtained from the meteorological database. The comprehensive influence of environmental parameters is corrected by a multiple regression model, and the model parameters are updated every twenty-four hours. The system calibration process includes automatic calibration and manual calibration modes, and automatic calibration is performed once a week. The calibration signal source uses a high-precision function generator, and the output signal amplitude accuracy reaches one ten-thousandth. The phase calibration uses a phase-locked loop technology, and the phase error is less than zero one degree. The amplitude calibration uses piecewise linear interpolation, and the interpolation point interval is one-tenth of the range. The calibration data is stored in the calibration file, and the file version management uses incremental numbering.

[0085] The electromagnetic compatibility design includes a shielding layer and a filter circuit, and the shielding layer has a grounding resistance of less than one ohm. The power filter uses a π-type filter with a cutoff frequency set to ten kilohertz. The signal isolation uses an optoelectronic coupler with an isolation voltage of two thousand five hundred volts. The grounding system uses a single-point grounding method, and the grounding wire cross-sectional area is not less than four square millimeters. The radiation emission test meets the CISPR22 standard, and the test distance is ten meters. The mechanical structure design considers the heat dissipation and shockproof requirements, and the surface area of the heat sink is calculated according to the maximum power consumption. The mounting bracket uses shock-absorbing materials, and the shock-absorbing frequency avoids the system resonance point. The terminal uses a spring crimping method, and the contact resistance is less than one milliohm. The shell protection level reaches the IP54 standard, and the dust and water resistance performance meets the requirements of industrial environment. The material selection considers temperature adaptability, and the working temperature range covers minus twenty-five degrees to seventy degrees. The user interface design includes a touch screen and physical buttons, and the touch screen resolution reaches eight hundred by four hundred pixels. The display content supports multi-language switching, and the character encoding uses UTF8 format. The alarm information is displayed in three levels, and the color coding conforms to the industrial standard. Data export supports USB and Ethernet interfaces with a transmission rate of one hundred megabits per second. Remote monitoring supports Web access, and security authentication uses two-factor verification. The log record includes operation records and system events, and the record retention time is not less than three years.

[0086] Example 5: Redundancy verification mechanism of electrical sensor array achieves data cross-verification by configuring multiple sensor nodes. Each sensor node includes an independent signal conditioning circuit and analog-to-digital converter. Node data acquisition adopts synchronous sampling technology. The synchronization signal is obtained by frequency division of the master clock generated by the FPGA, and the clock deviation is controlled within five nanoseconds. The data verification algorithm adopts a three-modulus redundancy voting mechanism. Three sensor nodes form a voting group, and the output result is the data value that is consistent between two or more nodes. Abnormal data point identification is based on the statistical outlier detection principle. The detection window includes the most recent one hundred sampling points, and the threshold is set to three times the standard deviation. Continuous abnormal data counting is implemented by a hardware counter with a counter bit width of sixteen bits. When the count value reaches the fault tolerance limit, a hardware self-test signal is triggered. The hardware self-test signal initiates the sensor diagnostic process, which includes impedance testing, noise level detection, and communication link checking. Impedance testing is performed by injecting a 1000 Hz sine wave signal to measure the amplitude and phase angle of the sensor output impedance. Noise level detection is performed with the input short-circuited, and the root mean square value is calculated by collecting ten minutes of background noise data. The communication link check sends test data packets to verify the transmission error rate. The test data packet length is set to 128 bytes. The self-test result is compared with a preset benchmark value, which is stored in the calibration area of ​​non-volatile memory and updated every three months. If the self-test fails, the system enters a safe mode. In safe mode, the power output is turned off and the alarm indicator light is activated. The alarm signal is transmitted to the control center through optical isolation.

[0087] The redundant data replacement strategy employs a spatiotemporal correlation interpolation method, repairing abnormal data points based on data from adjacent sensors and historical data trends. Time-dimensional interpolation uses the Lagrange interpolation algorithm with a cubic polynomial order. Spatial-dimensional interpolation uses an inverse distance weighted method, with weighting coefficients inversely proportional to the sensor spacing. Historical data trend analysis utilizes an autoregressive model, with model parameters fitted to the most recent 1000 normal data points using the least squares method. Interpolation results undergo amplitude limitation checks, with the limitation range set at ±20% of the normal data fluctuation range. Data repair records are marked with a special flag, stored in reserved bytes within the data packet. The verification log recording system uses a circular buffer structure, with a buffer capacity storing 100,000 event records. Log entries include timestamps, event types, sensor numbers, original data, and repaired data. Timestamp accuracy reaches millisecond levels, with a base time provided by a real-time clock chip. Event categories include data anomalies, sensor failures, communication interruptions, and system restarts. Log retrieval supports combined queries by time range and event type, and query results can be exported as CSV files. The log file automatic archiving function generates a new log file every month, and old files are compressed and saved for twelve months before being automatically deleted.

[0088] The sensor node health state monitoring is achieved by periodically collecting node operating parameters, including power supply voltage, chip temperature and signal quality indicators. The power supply voltage monitoring uses a sixteen-bit analog-to-digital converter, with a measurement accuracy of plus or minus five millivolts. The chip temperature is detected by an integrated temperature sensor, and the temperature data is recorded every ten minutes. The signal quality indicator calculates the signal-to-noise ratio and total harmonic distortion, based on a one-minute data collection window. The health state scoring model weighs the various parameters, with the weight coefficients dynamically adjusted according to the importance of the parameters. When the score is below the threshold, a warning signal is triggered, and the warning signal is identified by a system status register bit. The fault prediction model uses a machine learning algorithm to analyze the sensor data features, including time domain statistics, frequency domain components and nonlinear features. The time domain statistics calculate the mean, variance, skewness and kurtosis, with a time window length of ten minutes. The frequency domain components are obtained by fast Fourier transform, and the first ten harmonic amplitude values are extracted. The nonlinear features include approximate entropy and sample entropy, with an embedding dimension of three. The machine learning model selects a support vector machine classifier, and the training data comes from a historical fault case library. The model update period is set to three months, and the model parameters are updated incrementally with new data.

[0089] The electromagnetic compatibility design includes multi-layer circuit board layout and shielding shell structure, and the signal wiring uses differential transmission mode. The power filter network uses LCπ type filter, the inductance value is selected as ten micro henry, and the capacitance value is selected as one hundred nanofarad. The shielding shell material uses galvanized steel plate, the thickness reaches zero point eight millimeter, and the grounding point is designed at the geometric center of the shell. The radiation emission suppression is realized by adding magnetic beads and common mode choke coil, and the sensitive frequency point sets a special filter. The conducted emission test meets the EN55022 standard limit value, and the test frequency range is from one hundred fifty kilohertz to thirty megahertz. The system maintenance interface provides remote diagnosis and parameter configuration functions, and the communication protocol uses encrypted transmission mode. The identity authentication uses digital certificate mechanism, and the certificate validity period is set to one year. Parameter modification requires double authorization, and the operator account and administrator account are verified at the same time. The configuration change record details the operation log, which includes the pre-modification value and the post-modification value. Remote diagnosis supports real-time data monitoring and fault code reading, and the data refresh rate can be set to one second to one minute. The firmware upgrade adopts block check mechanism, and each data block adds CRC32 check code. The environmental adaptability design considers the influence of temperature cycle and mechanical vibration, and the temperature cycle test range is from minus forty degrees Celsius to eighty-five degrees Celsius. The thermal design uses a combination of heat dissipation fins and heat conductive silicone, and the chip junction temperature is controlled below one hundred twenty-five degrees Celsius. The mechanical vibration protection uses shock absorbing rubber pads and wire clamps, and the resonance frequency avoids the fifty hertz power frequency point. The protective coating uses three-proof paint treatment, and the moisture-proof, mildew-proof and salt mist-proof performance meets the industrial standards. The wiring terminal adopts anti-misplug design, and the terminal material selects phosphor bronze gold plating treatment.

[0090] Reliability verification is completed by accelerated life testing, including high temperature and humidity and voltage fluctuation. High temperature and humidity testing is conducted at eighty-five degrees Celsius, eighty-five percent humidity for one thousand hours. Voltage fluctuation testing simulates twenty percent positive and negative voltage variation, with a five percent positive and negative frequency variation. Fault injection testing intentionally introduces sensor short circuits, open circuits, and signal interference to verify system fault tolerance capabilities. Mean time between failures metrics are calculated by a Weibull distribution model, with a ninety-five percent confidence interval. Reliability reports include failure rate curves and failure mode analysis, with reports updated every six months. Data security is protected by encryption storage and access control mechanisms, with storage encryption using AES256 algorithm to encrypt data files. Access control is based on role-based permission management, with permissions divided into three levels: read-only, operation, and maintenance. Audit logs record all data access operations, with log files using digital signatures to prevent tampering. Data transmission channels are encrypted using TLS1.3 protocol, with key exchange using elliptic curve cryptography. Secure boot mechanisms verify firmware integrity, with hash value comparison failure denying system startup. Security chips store key material independently, with physical protection reaching EAL4+ security level. System integration testing includes unit testing and system integration, with unit testing covering all software modules and hardware functions. Test cases are designed using equivalence class partitioning and boundary value analysis methods, with exception scenario testing including power failure and signal interference. System integration simulates actual operating conditions, with load variation ranging from zero to one hundred and twenty percent of rated load. Performance metrics testing includes measurement accuracy, response time, and stability, with accuracy verification using a standard power source as a reference. Compatibility testing verifies adaptation with different types of inductors, with interface protocols supporting custom extensions. Test reports detail test conditions and result data, with reports archived as technical documentation.

[0091] It should be noted that the relational terms herein, such as first and second, and the like, are used solely to distinguish one from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0092] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, combinations, and variations of the embodiments can be made by those skilled in the art without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

Claims

1. A sensor-based method for measuring inductor winding losses, characterized in that, Includes the following steps: The real-time current signal, voltage signal and surface temperature distribution signal of the hollow wire-wound inductor are synchronously acquired by an array of electrical sensors during operation. Based on the skin effect dynamic model, the skin depth threshold under the current working condition is calculated according to the current signal frequency and conductor material parameters. The acquired current signal is subjected to multi-channel adaptive filtering to extract the effective current component that matches the skin depth threshold. By combining the spatial distribution parameters of the segmented winding structure, a parasitic capacitance compensation coefficient matrix is ​​generated. The instantaneous power loss value of each winding segment is calculated based on the effective current component, voltage signal, and compensation coefficient matrix. By mapping the temperature distribution signal to the power loss value, the thermal coupling error of the segmented winding structure is corrected; By integrating the corrected power loss values ​​of each segment, the overall loss index of the air-wound inductor is output.

2. The sensor-based inductor winding loss measurement method according to claim 1, characterized in that, The calculation process of the skin effect dynamic model is as follows: Obtain the resistivity, permeability, and fundamental frequency of the current signal of the conductor material; A frequency dependence factor is generated based on the product of the fundamental frequency and the permeability. The square root of the ratio of resistivity to frequency dependence factor is used to obtain the dynamically updated skin depth threshold.

3. The sensor-based inductor winding loss measurement method according to claim 2, characterized in that, The multi-channel adaptive filtering process includes: Establish signal channels corresponding to the number of winding segments, and configure an independent bandpass filter for each channel; Adjust the center frequency and bandwidth of each channel filter according to the skin depth threshold; A sliding time window algorithm is used to perform time-frequency decomposition on the current signal to suppress high-frequency harmonic interference.

4. The sensor-based inductor winding loss measurement method according to claim 3, characterized in that, The specific process for generating the parasitic capacitance compensation coefficient matrix is ​​as follows: Obtain the dielectric constant, interlayer spacing, and winding geometry of the insulating layers in the segmented winding structure; The basic compensation unit is generated based on the ratio of dielectric constant to interlayer spacing; By superimposing the electric field distribution weights of the winding geometry, a multidimensional compensation coefficient matrix is ​​constructed.

5. The sensor-based inductor winding loss measurement method according to claim 4, characterized in that, The calculation process for the instantaneous power loss value is as follows: Phase alignment processing is performed between the effective current component and the voltage signal; Perform matrix dot product on the aligned signal according to the segment number; A compensation coefficient matrix is ​​introduced to perform weighted correction on the dot product result.

6. The sensor-based inductor winding loss measurement method according to claim 5, characterized in that, The thermal coupling error correction includes: Establish a linear regression model between temperature gradient and power loss; Calculate the thermal conduction offset between adjacent segments based on the surface temperature distribution signal; The distribution weights of power loss values ​​are iteratively adjusted using the backpropagation algorithm.

7. The sensor-based inductor winding loss measurement method according to claim 6, characterized in that, Also includes: The frequency drift of the current signal is monitored in real time, and the dynamic model is recalculated when the drift exceeds a preset threshold. Update the filter channel parameters and compensation coefficient matrix based on the recalculation results.

8. The sensor-based inductor winding loss measurement method according to claim 7, characterized in that, The specific process for outputting the comprehensive loss index is as follows: Perform time integration on the corrected power loss values ​​for each segment; The dynamic compensation term for the superimposed integral result and the frequency drift; After normalization, a standardized loss index sequence is generated.

9. The sensor-based inductor winding loss measurement method according to claim 8, characterized in that, Also includes: Abnormal temperature or current data points are eliminated through the redundancy verification mechanism of the electrical sensor array. When consecutive abnormal data exceeds the fault tolerance limit, a hardware self-test signal is triggered.

10. A sensor-based inductor winding loss measurement system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the sensor-based inductor winding loss measurement method as described in any one of claims 1 to 9.

Images