An inductor online detection system with multi-parameter synchronous acquisition

Abstract

The application discloses an inductor online detection system with multi-parameter synchronous acquisition, and particularly relates to the technical field of inductor online monitoring, and comprises: a distortion detection module which triggers synchronous acquisition of voltage, current and temperature based on a saturation distortion capture threshold; a flux linkage analysis module which performs nonlinear flux linkage analysis to generate instantaneous differential inductance of each sampling point; a window average module which averages the instantaneous differential inductance of each sub-window in a switching cycle to obtain a window differential inductance value sequence and a change rate thereof; and a risk assessment module which assesses a risk level and gives an early warning according to the temperature, the window differential inductance and the change rate thereof. The application generates instantaneous differential inductance at each sampling point through nonlinear flux linkage analysis, thereby breaking the linear inductance hypothesis. Based on the window differential inductance sequence and the change rate thereof, the time resolution is improved, and early warning is realized. The distortion capture threshold and the integral time window length are adjusted in a closed loop through feedback parameters, thereby realizing adaptive optimization detection.

Description

Technical Field

[0001] This invention relates to the field of inductor online monitoring technology, and more specifically, to an inductor online detection system with multi-parameter synchronous acquisition capabilities. Background Technology

[0002] Inductors are fundamental components in power electronic systems such as switching power supplies, motor drives, and grid-connected converters. Their inductance value is not constant but changes dynamically with variations in core saturation, temperature, and bias current. Especially in high-power-density applications, inductors often operate close to the critical state of core saturation. At this point, even small fluctuations in inductance can cause a significant increase in current ripple, loop gain instability, and may even trigger overcurrent protection or damage power devices.

[0003] Several online inductance detection methods have been proposed, among which the most widely used is the online calculation method based on high-frequency ripple slope. This method uses the rising or falling slope of the inductor current within a switching cycle to infer the inductance value, offering advantages such as simplicity and no need for external excitation signals. However, this method implicitly assumes that the inductance remains constant within a switching cycle and the current change follows a linear triangular wave. When the inductor core rapidly enters the deep saturation region, the inductance value decreases nonlinearly with the instantaneous current, resulting in significant distortion of the current waveform. Under this condition, calculations based on linear slopes will produce significant lag and numerical deviations, failing to accurately reflect the instantaneous dynamic inductance. Furthermore, this algorithm requires at least half a switching cycle of data to complete a single calculation, resulting in insufficient time resolution and difficulty in timely capturing sub-microsecond saturation events, potentially leading to overcurrent protection delays, current loop oscillations, or equipment damage. Therefore, this invention proposes an online inductor detection system with multi-parameter synchronous acquisition to address these issues. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: An online inductor testing system with multi-parameter synchronous acquisition includes: The distortion detection module is based on a saturation distortion capture threshold. When the distortion characteristics in the inductor current waveform reach the saturation distortion capture threshold, it triggers the synchronous acquisition of the voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core at the same time reference to obtain voltage data, current data, and temperature data. The flux linkage analysis module performs nonlinear flux linkage analysis based on voltage and current data, generating the instantaneous differential inductance at each sampling point; The window averaging module divides a switching cycle into multiple sub-cycle sampling windows. Within each sub-cycle sampling window, the instantaneous differential inductance corresponding to all sampling points within the window is averaged to obtain a window differential inductance. The window differential inductance value sequence is obtained, and the rate of change of the window differential inductance value over time is calculated. The risk assessment module evaluates the risk level of the current state based on temperature data, the instantaneous differential inductance value sequence, and the rate of change of the instantaneous differential inductance value over time, and issues corresponding graded early warning signals according to the risk level. The fingerprint feedback module compares the feature data generated in the above process, including voltage data, current data, temperature data, instantaneous differential inductance value sequence, and risk level, with a pre-stored multimodal failure fingerprint database to identify the current failure mode. Based on the identification result, feedback parameters are generated, and the saturation distortion capture threshold and the length of the integration time window used when generating the instantaneous differential inductance are adjusted using the feedback parameters.

[0005] In a preferred embodiment, the saturation distortion capture threshold is an initial threshold that is set to five to fifteen percent of the peak rated current of the inductor. Furthermore, in the subsequent closed-loop feedback, the feedback parameters are proportionally adjusted to the initial threshold, and the adjusted threshold is used to replace the initial threshold in the next monitoring trigger.

[0006] In a preferred embodiment, the synchronous acquisition of the voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core under the same time reference is achieved by simultaneously triggering all sensor channels with a common analog-to-digital converter start signal.

[0007] In a preferred embodiment, nonlinear flux linkage analysis is performed based on voltage and current data to generate the instantaneous differential inductance at each sampling point, specifically including the following calculation steps: The first step is to denote the voltage data sequence as V, the current data sequence as I, and the inductor parasitic resistance as R, and then construct a dynamic error function: ; The second step involves using the response system equations of a chaotic synchronization algorithm to calculate the flux linkage sequence over time from the measured values ​​V and I in real time by bringing the dynamic error function close to zero. The response system equation is: ; among which symbols Represents the real magnetic flux linkage, which is a function of time t, with the symbol... Indicates the estimated flux linkage, symbol This represents the preset coupling coefficient, and the symbol tanh represents the hyperbolic tangent function; The third step is to solve for the magnetic flux sequence. For current sequences Find the numerical derivative The instantaneous differential inductance at each sampling point is obtained.

[0008] In a preferred embodiment, a switching cycle is divided into multiple sub-cycle sampling windows. Within each sub-cycle sampling window, the average of the instantaneous differential inductances corresponding to all sampling points within the window is taken to obtain a window differential inductance. A sequence of window differential inductance values ​​is obtained, and the rate of change of the window differential inductance values ​​over time is calculated. Specifically, a complete pulse width modulation cycle is uniformly divided into sub-cycle sampling windows of equal length according to a preset number of windows, with the preset number of windows being no less than four. The arithmetic mean of the instantaneous differential inductances of all sampling points within each window is taken as the window differential inductance of that window. The rate of change is the difference in window differential inductance between adjacent windows divided by the window time length.

[0009] In a preferred embodiment, the risk level of the current state is assessed based on temperature data, the window differential inductance value sequence, and the rate of change of the window differential inductance value over time. This assessment includes the following calculation process: The temperature data is converted into a thermodynamic entropy correction factor, which is calculated using the following formula: The thermodynamic entropy correction factor is equal to the natural logarithm of the temperature value divided by the reference temperature value, and then multiplied by the Boltzmann constant, where the reference temperature value is 300 Kelvin and the Boltzmann constant is 1.38 times 10 to the power of negative 23 joules per Kelvin. Construct a geodesic distance based on differential geometry The geodesic distance is used to measure the curvature difference between the current window differential inductance value sequence and the window differential inductance value sequence under historical normal operating conditions. The calculation formula is as follows: ; Where the symbol T represents a switching cycle, the symbol... Represents the window differential inductance value, symbol This represents the average window differential inductance value under normal historical operating conditions, with the symbol α representing the preset curvature weighting coefficient; The thermodynamic entropy correction factor and geodesic distance Multiply the results and then normalize the index to the base of the natural constant e to obtain a risk index between zero and one. Based on this risk index, four consecutive risk level intervals are divided.

[0010] In a preferred embodiment, the average window differential inductance value representing historical normal operating conditions is compared with a pre-stored multimodal failure fingerprint database to identify the current failure mode. Specifically, this includes: extracting three feature vectors—time-domain peak value, frequency-domain dominant frequency energy, and rate of change slope—from voltage, current, and temperature data, respectively; extracting three feature vectors—mean, variance, and maximum curvature—from the window differential inductance value sequence; directly using the risk level as a discrete feature value; concatenating all the above feature vectors into a comprehensive feature vector; calculating the Euclidean distance between this comprehensive feature vector and the standard feature vector corresponding to each type of failure mode in the fingerprint database; and determining the failure mode corresponding to the minimum distance as the current failure mode.

[0011] In a preferred embodiment, the feedback parameters include a scaling factor for the saturation distortion capture threshold and a correction value for the length of the integration time window used to generate the instantaneous differential inductance; the scaling factor and the correction value are generated in the following manner: Calculate the Euclidean distance between the current integrated feature vector and the standard feature vector corresponding to the normal operation mode in the pre-stored multimodal failure fingerprint database, and record the value of the Euclidean distance; The mean of the window differential inductance value sequence in the current comprehensive feature vector is compared with the mean of the window differential inductance value sequence in the standard feature vector corresponding to the normal operation mode: if the current mean is greater than the standard mean, a positive sign is generated; if the current mean is less than the standard mean, a negative sign is generated; if the current mean is equal to the standard mean, a zero sign is generated. Divide the Euclidean distance value by a preset reference distance value to obtain a dimensionless adjustment range coefficient. Then multiply the adjustment range coefficient by the generated positive, negative, or zero sign to obtain the signed adjustment amount. Add one to the signed adjustment amount to obtain the proportional adjustment coefficient; multiply the signed adjustment amount by the currently used integral time window length to obtain the updated integral time window length, which is the correction value of the integral time window length. Multiply the scaling factor by the currently used saturation distortion capture threshold, and use the product as the updated saturation distortion capture threshold. The updated saturation distortion capture threshold is used for the next monitoring trigger, and the updated integral time window length is used for the flux linkage integral calculation when generating the instantaneous differential inductance next time.

[0012] The technical effects and advantages of this invention are as follows: This invention overcomes the inherent flaw of traditional methods that rely on the linear inductance assumption by synchronously acquiring voltage, current, and temperature data under the same time reference and directly generating the instantaneous differential inductance at each sampling point based on nonlinear flux linkage analysis. When the inductor core rapidly enters the deep saturation region, the inductance value decreases nonlinearly with the current. Traditional algorithms based on high-frequency ripple slope, which assume a constant inductance, exhibit significant lag and numerical deviation. This invention directly calculates the instantaneous differential inductance using the derivative of flux linkage with respect to current, without assuming a constant inductance over a switching cycle. Therefore, it accurately reflects the transient changes in inductance during sub-microsecond saturation, providing a true parameter basis for loop compensation and overcurrent protection. It fundamentally solves the detection failure problem caused by linear model mismatch in existing technologies and significantly improves adaptability to dynamic nonlinear operating conditions.

[0013] This invention achieves a balance between time resolution and noise suppression by dividing a switching cycle into multiple sub-cycle sampling windows and averaging the instantaneous differential inductance within each window to obtain a sequence of window differential inductance values ​​and their rate of change. This invention independently generates a window differential inductance within each sub-cycle window, increasing the inductance value update frequency several times over, while reducing the impact of sampling noise through averaging within the window. The calculated rate of change of the window differential inductance value over time directly reflects the inductance decay rate, allowing for subsequent risk level assessment and early warning. It can prompt the system to take derating or protection actions when saturation is still in its early stages, preventing power transistor damage due to protection delays.

[0014] This invention identifies the current failure mode by comparing it with a pre-stored multimodal failure fingerprint database and generates feedback parameters to dynamically adjust the saturation distortion capture threshold and the integration time window length used when generating the instantaneous differential inductance, forming a closed-loop optimization mechanism. Based on the real-time identified failure mode (e.g., uniform saturation, local saturation, or aging-induced brittle fracture), this invention uses a comparison of Euclidean distance and the inductance mean to generate a signed adjustment amount, automatically lowering or raising the capture threshold while shortening or extending the integration time window. This ensures the system always operates within the optimal sensitivity range. This adaptive closed-loop significantly improves the robustness and generalization ability of the detection system, avoiding performance degradation of fixed parameters over long-term operation. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of an online inductor detection system with multi-parameter synchronous acquisition, as described in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 The following examples were obtained: Example 1: An online inductor testing system with multi-parameter synchronous acquisition, comprising: The distortion detection module is based on a saturation distortion capture threshold. When the distortion characteristics in the inductor current waveform reach the saturation distortion capture threshold, it triggers the synchronous acquisition of the voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core at the same time reference to obtain voltage data, current data, and temperature data. The flux linkage analysis module performs nonlinear flux linkage analysis based on voltage and current data, generating the instantaneous differential inductance at each sampling point; The window averaging module divides a switching cycle into multiple sub-cycle sampling windows. Within each sub-cycle sampling window, the instantaneous differential inductance corresponding to all sampling points within the window is averaged to obtain a window differential inductance. The window differential inductance value sequence is obtained, and the rate of change of the window differential inductance value over time is calculated. The risk assessment module evaluates the risk level of the current state based on temperature data, the instantaneous differential inductance value sequence, and the rate of change of the instantaneous differential inductance value over time, and issues corresponding graded early warning signals according to the risk level. The fingerprint feedback module compares the feature data generated in the above process, including voltage data, current data, temperature data, instantaneous differential inductance value sequence, and risk level, with a pre-stored multimodal failure fingerprint database to identify the current failure mode. Based on the identification result, feedback parameters are generated, and the saturation distortion capture threshold and the length of the integration time window used when generating the instantaneous differential inductance are adjusted using the feedback parameters.

[0018] The saturation distortion capture threshold is an initial threshold used to determine whether the inductor current waveform exhibits distortion characteristics before entering saturation. This initial threshold is obtained by multiplying the inductor's rated peak current by a scaling factor, which is between 5% and 15%. This range is based on statistical analysis of experimental data from multiple inductor models at the saturation critical point. For example, if an inductor's rated peak current is 10 amps, the initial threshold is set to a fixed value between 0.5 amps and 1.5 amps, typically 0.8 amps.

[0019] When feedback parameters are generated in subsequent closed-loop feedback, these parameters are proportional adjustment coefficients, the values ​​of which are dynamically determined by the current failure mode and the Euclidean distance calculation results. This proportional adjustment coefficient is multiplied by the currently used saturation distortion capture threshold, and the product becomes the updated saturation distortion capture threshold. The updated threshold replaces the original threshold for the next monitoring trigger. For example, if the current threshold is 0.8 amps and the feedback parameter is 1.2, the new threshold is updated to 0.96 amps.

[0020] The system synchronously acquires the voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core under the same time reference. This is achieved by simultaneously triggering all sensor channels via a shared analog-to-digital converter (ADC) start signal. This shared start signal is generated by a timer in the controller, and its trigger period is equal to an integer multiple of the inductor's operating frequency to ensure that the sampling time deviation of each channel is less than one sampling clock cycle. For example, if the inductor's operating frequency is 100 kHz, the timer generates a start signal with a period of 10 microseconds, simultaneously triggering the ADCs of the voltage, current, and temperature sensors. Each channel begins sampling on the rising edge of the start signal, and the sampling and holding time is the same, thus ensuring that the time error between the data does not exceed 0.1 microseconds. This synchronous sampling method is based on the fact that transient analysis of multiple physical quantities requires strict time alignment; otherwise, the flux linkage integral calculation will produce accumulated errors. Experimental data shows that when the time deviation exceeds one microsecond, the error in the differential inductance calculation increases by 5%.

[0021] Nonlinear flux linkage analysis is performed based on voltage and current data to generate the instantaneous differential inductance at each sampling point. The specific calculation steps include the following: First, denote the voltage data sequence as V, the current data sequence as I, and the inductor parasitic resistance as R. The value of the parasitic resistance R is based on the inductor's resistance measurement at DC or low frequencies, typically ranging from milliohms to ohms. For example, for power inductors, R is usually taken as 0.05 ohms. A dynamic error function is constructed, defined as the difference between V and the product of I and R, minus the time derivative of the flux linkage. That is, the error function equals (VI·R) minus dλ / dt. The physical meaning of this dynamic error function is to measure the degree to which the law of electromagnetic induction is satisfied; when the estimate of the flux linkage λ is accurate, the error function approaches zero.

[0022] The response system equation employing a chaotic synchronization algorithm is used to calculate the flux linkage sequence over time from measured values ​​V and I in real time by bringing the dynamic error function close to zero. The response system equation is: dλ / dt = VI·R + k·tanh(λ - λ_est). Here, λ represents the actual flux linkage as a function of time t, λ_est represents the estimated flux linkage, k represents the preset coupling coefficient, and tanh represents the hyperbolic tangent function. The coupling coefficient k is set experimentally based on the system's convergence speed and stability, typically ranging from 0.1 to 10; for example, for an inductor with a 100 kHz switching frequency, k is set to 2.5. Through iterative solving, the estimated flux linkage λ_est gradually approximates the actual flux linkage λ, thus obtaining the flux linkage sequence λ(t).

[0023] Finally, the numerical derivative of the obtained flux linkage sequence λ(t) with respect to the current sequence I(t) is calculated to obtain the instantaneous differential inductance at each sampling point. The numerical derivative is calculated by dividing the flux linkage difference between adjacent sampling points by the corresponding current difference, i.e., dλ / dI≈(λ_{i+1}-λ_i) / (I_{i+1}-I_i). This calculation is repeated for all sampling points to obtain the sequence of instantaneous differential inductances. For example, at a certain sampling moment, if the flux linkage difference is 0.001 Weber and the current difference is 0.1 Ampere, then the instantaneous differential inductance is 0.01 H. This calculation method does not rely on the linear inductance assumption, and therefore can accurately reflect the nonlinear characteristics when the magnetic core is saturated.

[0024] In the response system equation, the symbol t represents time, with units corresponding to the time interval between voltage and current sampling sequences. For example, the time difference between adjacent sampling points is ten microseconds. The symbol i represents the index of the sampling point, for example, i takes values ​​from 1, 2, 3 up to the total number of sampling points, used to identify the position of each discrete data in the sequence. Therefore, λ(t) represents a continuous function of flux linkage changing with time, and λ_i represents the flux linkage value corresponding to the i-th sampling point.

[0025] When dividing a switching cycle into multiple sub-cycle sampling windows, a complete pulse width modulation cycle is uniformly divided into sub-cycle sampling windows of equal length according to a preset number of windows. The preset number of windows is set based on the response time of the inductor core material and the processing capability of the sampling system. Experimental data shows that a minimum of four windows can capture transient changes in the saturation process; typical values ​​are eight or sixteen. For example, if the switching cycle is 20 microseconds and the number of windows is eight, then the length of each sub-cycle sampling window is 2.5 microseconds.

[0026] Within each sub-period sampling window, the arithmetic mean of the instantaneous differential inductances corresponding to all sampling points within the window is taken to obtain a window differential inductance. The purpose of averaging is to suppress random noise at individual sampling points and improve the stability of the window-level inductance. For example, if a window contains 10 sampling points, and the instantaneous differential inductance values ​​of each sampling point are 10.1 μH, 9.9 μH, 10.0 μH, 9.8 μH, 10.2 μH, 9.7 μH, 10.0 μH, 9.9 μH, 10.1 μH, and 10.0 μH, then the arithmetic mean is 9.97 μH, which is the window differential inductance of that window.

[0027] Repeat the above averaging operation for each window to obtain a sequence of window differential inductance values. The length of this sequence is equal to the number of windows in one switching cycle. For example, when there are 8 windows, the sequence contains 8 values, which correspond to the inductance values ​​of the 8 windows respectively.

[0028] The rate of change of the window differential inductance value over time is calculated. This rate of change is equal to the difference in window differential inductance between two adjacent windows divided by the window duration. The window duration is the length of each sub-cycle sampling window. For example, if the window duration is 2.5 microseconds, the differential inductance value of the first window is 10.0 microhenries, and the differential inductance value of the second window is 9.5 microhenries, then the rate of change is negative 0.2 microhenries per microsecond. This rate of change reflects the rate at which the inductance value decreases within a window interval. In practical applications, it can also indicate that the magnetic core is rapidly saturating when the rate of change exceeds an empirical threshold (e.g., negative 0.5 microhenries per microsecond), requiring an early warning.

[0029] When assessing the risk level of the current state based on temperature data, the window differential inductance value sequence, and the rate of change of the window differential inductance value over time, the temperature data is first converted into a thermodynamic entropy correction factor. The formula for calculating this thermodynamic entropy correction factor is as follows: The thermodynamic entropy correction factor is equal to the natural logarithm of the temperature value divided by the reference temperature value, multiplied by the Boltzmann constant. The reference temperature is 300 Kelvin, and the Boltzmann constant is 1.38 × 10⁻⁶. -23 Joules per Kelvin (i.e., 1.38 multiplied by 10 to the power of -23 joules per Kelvin). For example, if the measured temperature of an inductor core is 350 Kelvin, then the temperature value divided by the reference temperature value is 1.1667, the natural logarithm is 0.154, and then multiplied by 1.38 × 10⁻⁶. -23 The thermodynamic entropy correction factor is approximately 2.13 × 10⁻⁶. -24 Joules per Kelvin.

[0030] Construct a geodesic distance based on differential geometry This is used to measure the curvature difference between the current window differential inductance value sequence and the window differential inductance value sequence under historical normal operating conditions. The calculation formula is: ; Where the symbol T represents a switching cycle, typically 20 microseconds; Represents the window differential inductance value, in microhenries; symbol This represents the average window differential inductance value under historical normal operating conditions, for example, 10.0 microhenries based on the statistical results of 100 normal cycles; the symbol α represents the preset curvature weight coefficient, with a value range of 0.1 to 10, and a typical value of 2.0. The setting is based on making the curvature term and the inductance deviation term of the same order of magnitude.

[0031] The thermodynamic entropy correction factor and geodesic distance Multiplying these values ​​and then performing exponential normalization with the natural constant e as the base, yields a risk index between zero and one. This risk index is then used to divide the risk into four consecutive risk level intervals. For example, the thermodynamic entropy correction factor is 2.13 × 10⁻⁶. -24 The geodesic distance is 0.35, and the product is approximately 7.46 × 10⁻⁶. -25 If the negative power of e is approximately 1, then the risk index approaches 0; if the product is large, then the risk index approaches 1.

[0032] The risk index is divided into four consecutive risk level ranges: 0 to 0.2 is level 1, 0.2 to 0.5 is level 2, 0.5 to 0.8 is level 3, and 0.8 to 1.0 is level 4. The level is set based on the experimental statistics that the normal fluctuation range corresponds to level 1, slight saturation corresponds to level 2, moderate saturation corresponds to level 3, and deep saturation corresponds to level 4.

[0033] Based on the risk level, corresponding graded warning signals are issued. When the risk index is in the range of 0 to 0.2, a green indicator light is constantly on, indicating that the inductor is in normal operating condition. This range is set based on the statistical results of the risk index from 100 measurements of the inductor under rated operating conditions, with a typical upper limit of 0.15. When the risk index is in the range of 0.2 to 0.5, a blue flashing signal is output at a frequency of 1 Hz, indicating that the inductor is showing slight signs of saturation. This range corresponds to an experimentally observed inductance value decrease of less than 5%, and is set based on empirical data showing that a risk index exceeding 0.2 requires attention. When the risk index is in the range of 0.5 to 0.8, a yellow audible and visual alarm signal is output, and a buzzer is triggered to emit intermittent beeps, indicating moderate saturation requiring load reduction. The typical threshold for this range is 0.65, corresponding to an inductance value decrease of 5% to 15%. When the risk index is in the range of 0.8 to 1.0, a red flashing signal is output and an emergency stop command is generated simultaneously, with a flashing frequency of 5 Hz. The emergency stop command is sent directly to the upper-level controller to disconnect the power supply. The setting is based on experimental data showing that if the inductance value drops by more than 15% or the rate of change is less than -0.5 microhenries per microsecond, continued operation will damage the power transistor.

[0034] The process of identifying the current failure mode is as follows: Three feature vectors are extracted from voltage, current, and temperature data: time-domain peak value, frequency-domain dominant frequency energy, and rate of change slope. The time-domain peak value is the maximum value of the physical quantity within one switching cycle. For example, if the highest value in the voltage data is 48 volts, then the time-domain peak value is 48 volts, set based on the fact that the peak value reflects the degree of instantaneous impact. The frequency-domain dominant frequency energy refers to the energy value corresponding to the frequency component with the largest amplitude after performing a Fast Fourier Transform on the physical quantity sequence. For example, the dominant frequency of the current data is 100 kHz, and its energy is 0.5 ampere squared; this energy value is used to identify harmonic anomalies. The rate of change slope refers to the slope of the least-squares fitted line of the physical quantity within one switching cycle. For example, if the temperature data rises from 25 degrees Celsius to 30 degrees Celsius in 20 milliseconds, the slope is 0.25 degrees Celsius per millisecond; the magnitude of the slope reflects the rate of change.

[0035] The mean, variance, and maximum curvature are extracted from the window differential inductance value sequence. The mean is the average of all values ​​in the sequence. For example, in a window differential inductance value sequence of [10.0, 9.5, 9.0, 8.5, 8.0] microhenries, the mean is 9.0 microhenries. The variance is the average of the sum of the squares of the differences between each value and the mean, reflecting the degree of inductance fluctuation. For example, the variance of the above sequence is approximately 0.5. The maximum curvature is obtained by calculating the second difference of the sequence. Specifically, the first difference of the sequence is calculated again, and the value with the largest absolute value is taken as the maximum curvature. For example, if the first difference is [-0.5, -0.5, -0.5, -0.5] and the second difference is [0, 0, 0], then the maximum curvature is 0. If the sequence has an inflection point, the second difference will be non-zero. The maximum curvature is used to identify abrupt changes in the inductance descent curve.

[0036] The risk level is directly treated as a discrete feature value, which is assigned the values ​​0, 1, 2, and 3 in order of level, corresponding to the four risk level intervals mentioned above. For example, if the current risk level is level 2, then the discrete feature value is 1.

[0037] All the above feature vectors are concatenated into a single composite feature vector. For example, the three feature values ​​of voltage data are [48 volts, 0.3 joules, 0.1 volts per millisecond], the three feature values ​​of current data are [5.2 amperes, 0.5 amperes squared, -0.2 amperes per millisecond], the three feature values ​​of temperature data are [30 degrees Celsius, 0.05 degrees Celsius squared per hertz, 0.2 degrees Celsius per millisecond], the three feature values ​​of the window differential inductance value sequence are [9.0 microhenries, 0.5, 0.05], and the discrete feature value of risk level is 1. Then the composite feature vector is [48, 0.3, 0.1, 5.2, 0.5, -0.2, 30, 0.05, 0.2, 9.0, 0.5, 0.05, 1].

[0038] Calculate the Euclidean distance between the integrated feature vector and the standard feature vector corresponding to each failure mode in the fingerprint database. The Euclidean distance is calculated as the square root of the sum of the squares of the differences between the corresponding components. For example, the standard feature vector for the normal mode in the fingerprint database is [47, 0.28, 0.09, 5.0, 0.48, -0.18, 29, 0.04, 0.19, 9.2, 0.45, 0.03, 0]. The square root of the sum of the squares of the differences between this vector and the current vector yields a distance of 2.1. Similar calculations are performed for the distances corresponding to saturation modes, aging modes, etc. The failure mode corresponding to the minimum distance is determined as the current failure mode. For example, if the minimum distance of 2.1 corresponds to the normal mode, it is determined as normal; if the minimum distance corresponds to the saturation mode, it is determined as a saturation failure.

[0039] The feedback parameters include a scaling factor for the saturation distortion capture threshold and a correction value for the length of the integration time window used to generate the instantaneous differential inductance. This scaling factor and correction value are generated as follows: Calculate the Euclidean distance between the current integrated feature vector and the standard feature vector corresponding to the normal operating mode in the pre-stored multimodal failure fingerprint database, and record the value of this Euclidean distance. For example, if the current integrated feature vector is [48,0.3,0.1,5.2,0.5,-0.2,30,0.05,0.2,9.0,0.5,0.05,1], and the standard feature vector corresponding to the normal operating mode is [47,0.28,0.09,5.0,0.48,-0.18,29,0.04,0.19,9.2,0.45,0.03,0], calculate the square root of the sum of the squares of the differences of each component, and obtain an Euclidean distance of 2.1. This preset baseline distance value is determined based on the normal fluctuation range of experimental statistics, and a typical value is 5.0. For example, the Euclidean distance is usually less than 3.0 under normal conditions.

[0040] The mean of the window differential inductance value sequence in the current comprehensive feature vector is compared with the mean of the window differential inductance value sequence in the standard feature vector corresponding to the normal operating mode. If the current mean is greater than the standard mean, a positive sign is generated; if the current mean is less than the standard mean, a negative sign is generated; and if the current mean is equal to the standard mean, a zero sign is generated. For example, if the mean of the current window differential inductance value sequence is 9.0 microhenries and the standard mean is 9.2 microhenries, then the current mean is less than the standard mean, and a negative sign is generated.

[0041] Dividing the Euclidean distance by a preset baseline distance value yields a dimensionless adjustment factor. For example, if the Euclidean distance is 2.1 and the baseline distance is 5.0, the adjustment factor is 0.42. This adjustment factor is then multiplied by a positive, negative, or zero sign to obtain a signed adjustment amount. A positive sign represents multiplying by +1, a negative sign represents multiplying by -1, and a zero sign represents multiplying by 0. For example, multiplying the adjustment factor 0.42 by -1 yields a signed adjustment amount of -0.42.

[0042] Adding the signed adjustment amount to the value of 1 yields the proportional adjustment coefficient. For example, -0.42 plus 1 equals 0.58, so the proportional adjustment coefficient is 0.58. Simultaneously, adding the signed adjustment amount to the value of 1 and then multiplying it by the currently used integration time window length yields the updated integration time window length. This updated integration time window length is the correction value for the integration time window length. For example, if the currently used integration time window length is 10 microseconds, multiplying it by (1-0.42)=0.58 gives an updated integration time window length of 5.8 microseconds.

[0043] Finally, the proportional adjustment coefficient is multiplied by the currently used saturation distortion capture threshold, and the product is used as the updated saturation distortion capture threshold. For example, if the currently used saturation distortion capture threshold is 0.8 amps, multiplying it by 0.58 yields 0.464 amps, which is the updated saturation distortion capture threshold. The updated saturation distortion capture threshold is used for the next monitoring trigger, and the updated integral time window length is used for the flux linkage integration calculation when generating the instantaneous differential inductance next time. The setting is based on the following: when the average value of the window differential inductance is low (i.e., the inductance decreases), a negative sign is generated, and the threshold and window length are reduced to make the system more sensitive to capturing early saturation characteristics; conversely, a positive sign is generated to relax the threshold and avoid false triggering. This closed-loop adjustment enables the system to adaptively track inductor aging or changes in operating conditions.

[0044] It should be noted that in practical applications, the final updated saturation distortion capture threshold should be limited to a preset threshold range. The minimum value of this range is 1% of the inductor's rated peak current, and the maximum value is 30% of the inductor's rated peak current. The setting is based on the following: the lower limit of 1% ensures the system will not be frequently falsely triggered by noise, and the upper limit of 30% ensures that saturation events will not be missed due to an excessively high threshold. This range is derived from experimental statistics on ten different inductor models, with a typical minimum of 3% and a typical maximum of 20%. For example, if an inductor's rated peak current is 10 amps, the adjustment range for the saturation distortion capture threshold is 0.1 amps to 3.0 amps. If the calculated updated threshold is lower than 0.1 amps, it should be forcibly set to 0.1 amps; if it is higher than 3.0 amps, it should be forcibly set to 3.0 amps.

[0045] The final updated integration time window length should also be limited to a preset window length range, with a minimum of one sampling period and a maximum of one switching period. The setting is based on the following: a lower limit of one sampling period to ensure at least one data point is used for integration, and an upper limit of one switching period to avoid introducing errors across integration periods. For example, if the sampling period is one microsecond and the switching period is twenty microseconds, then the adjustment range for the integration time window length is one to twenty microseconds. If the calculated update window length is less than one microsecond, it is forcibly set to one microsecond; if it is greater than twenty microseconds, it is forcibly set to twenty microseconds. This limiting operation is performed after each closed-loop feedback update to ensure that the system parameters are always within a safe and effective operating range.

[0046] The design logic for the positive and negative signs is based on the physical characteristics of the inductor and the requirements of closed-loop control. The specific reasons are as follows: When the current average value of the window's differential inductance is less than the standard average value under normal operating conditions, it indicates that the inductance has begun to decrease, which is usually a precursor to the core entering the saturation region. At this point, the system needs to improve its detection sensitivity, i.e., lower the saturation distortion capture threshold (making it easier to trigger) and shorten the integration time window length (improving the response speed of flux linkage calculation), in order to capture subsequent saturation distortion characteristics earlier. Therefore, a negative sign is used to make the adjustment amount negative, thus the proportional adjustment coefficient is less than 1 (threshold reduction), and the integration time window length is also shortened accordingly.

[0047] When the current average value of the window's differential inductance is greater than the standard average, it may be caused by measurement noise, transient load fluctuations, or abnormal inductor material, rather than true saturation. In this case, blindly increasing the sensitivity will lead to frequent false triggers. Therefore, the system needs to appropriately reduce the sensitivity, i.e., increase the saturation distortion capture threshold and extend the integration time window length to filter out false distortions. Therefore, a positive sign is used to make the adjustment amount positive, the proportional adjustment coefficient is greater than 1 (threshold increased), and the integration time window length is extended accordingly.

[0048] When the current mean equals the standard mean, the system considers the current state to be without deviation from normal operation and requires no adjustment. Therefore, a zero sign is used, the adjustment amount is zero, and the original parameters are maintained. The positive and negative signs reflect the causal relationship between the direction of physical deviation and the direction of adjustment action: negative deviation (inductance decrease) requires negative adjustment (reducing the threshold and window), and positive deviation (inductance increase) requires positive adjustment (increasing the threshold and window), thus forming a negative feedback closed loop to ensure that the system adaptively tracks the health state of the inductor.

[0049] The above-mentioned models or function formulas are all dimensionless and numerical calculations. The models or function formulas are obtained by software simulation based on a large amount of collected data to obtain the most recent real situation. The preset parameters in the models or function formulas are set by those skilled in the art according to the actual situation.

[0050] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0051] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0052] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0053] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An online inductor testing system with multi-parameter synchronous acquisition, characterized in that, include: The distortion detection module is based on a saturation distortion capture threshold. When the distortion characteristics in the inductor current waveform reach the saturation distortion capture threshold, it triggers the synchronous acquisition of the voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core at the same time reference to obtain voltage data, current data, and temperature data. The flux linkage analysis module performs nonlinear flux linkage analysis based on voltage and current data, generating the instantaneous differential inductance at each sampling point; The window averaging module divides a switching cycle into multiple sub-cycle sampling windows. Within each sub-cycle sampling window, the instantaneous differential inductance corresponding to all sampling points within the window is averaged to obtain a window differential inductance. The window differential inductance value sequence is obtained, and the rate of change of the window differential inductance value over time is calculated. The risk assessment module evaluates the risk level of the current state based on temperature data, the instantaneous differential inductance value sequence, and the rate of change of the instantaneous differential inductance value over time, and issues corresponding graded early warning signals according to the risk level. The fingerprint feedback module compares the feature data generated in the above process, including voltage data, current data, temperature data, instantaneous differential inductance value sequence, and risk level, with a pre-stored multimodal failure fingerprint database to identify the current failure mode. Based on the identification result, feedback parameters are generated, and the saturation distortion capture threshold and the length of the integration time window used when generating the instantaneous differential inductance are adjusted using the feedback parameters.

2. The inductor online detection system with multi-parameter synchronous acquisition according to claim 1, characterized in that, The saturation distortion capture threshold is an initial threshold, and in the subsequent closed-loop feedback, the feedback parameters are proportionally adjusted to this initial threshold. The adjusted threshold is used to replace the initial threshold in the next monitoring trigger.

3. The inductor online detection system with multi-parameter synchronous acquisition according to claim 2, characterized in that, The voltage across the inductor, the current flowing through the inductor, and the temperature of the inductor core are synchronously acquired under the same time reference. All sensor channels are triggered simultaneously by a common analog-to-digital converter start signal.

4. The inductor online detection system with multi-parameter synchronous acquisition according to claim 3, characterized in that, Nonlinear flux linkage analysis is performed based on voltage and current data to generate the instantaneous differential inductance at each sampling point. The specific calculation steps include the following: The first step is to denote the voltage data sequence as V, the current data sequence as I, and the inductor parasitic resistance as R, and then construct a dynamic error function: ； The second step involves using the response system equations of a chaotic synchronization algorithm to calculate the flux linkage sequence over time from the measured values ​​V and I in real time by bringing the dynamic error function close to zero. The response system equation is: ; among which symbols Represents the real magnetic flux linkage, which is a function of time t, with the symbol... Indicates the estimated flux linkage, symbol This represents the preset coupling coefficient, and the symbol tanh represents the hyperbolic tangent function; The third step is to solve for the magnetic flux sequence. For current sequences Find the numerical derivative The instantaneous differential inductance at each sampling point is obtained.

5. The online inductor testing system with multi-parameter synchronous acquisition according to claim 4, characterized in that, A switching cycle is divided into multiple sub-cycle sampling windows. Within each sub-cycle sampling window, the instantaneous differential inductance corresponding to all sampling points within the window is averaged to obtain a window differential inductance. The window differential inductance value sequence is obtained, and the rate of change of the window differential inductance value with time is calculated.

6. The online inductor testing system with multi-parameter synchronous acquisition according to claim 5, characterized in that, Based on temperature data, the window differential inductance value sequence, and the rate of change of the window differential inductance value over time, the risk level of the current state is assessed, specifically... Includes the following processes: The temperature data is converted into a thermodynamic entropy correction factor, which is calculated using the following formula: The thermodynamic entropy correction factor is equal to the natural logarithm of the temperature value divided by the reference temperature value, and then multiplied by the Boltzmann constant. Construct a geodesic distance based on differential geometry The geodesic distance is used to measure the curvature difference between the current window differential inductance value sequence and the window differential inductance value sequence under historical normal operating conditions. The calculation formula is as follows: ； Where the symbol T represents a switching cycle, the symbol... Represents the window differential inductance value, symbol This represents the average window differential inductance value under normal historical operating conditions, with the symbol α representing the preset curvature weighting coefficient; The thermodynamic entropy correction factor and geodesic distance Multiply the results and then normalize the index to the base of the natural constant e to obtain a risk index between zero and one. Based on this risk index, four consecutive risk level intervals are divided.

7. The inductor online detection system with multi-parameter synchronous acquisition according to claim 6, characterized in that, Identifying the current failure mode specifically includes: Three feature vectors are extracted from voltage, current, and temperature data: peak value in the time domain, dominant frequency energy in the frequency domain, and slope of the rate of change. Three feature vectors are extracted from the window differential inductance value sequence: mean, variance, and maximum curvature. The risk level is directly used as a discrete feature value. All the above feature vectors are concatenated into a comprehensive feature vector. The Euclidean distance between this comprehensive feature vector and the standard feature vector corresponding to each type of failure mode in the fingerprint database is calculated. The failure mode corresponding to the minimum distance is determined as the current failure mode.

8. The online inductor testing system with multi-parameter synchronous acquisition according to claim 7, characterized in that, The feedback parameters include a scaling factor for the saturation distortion capture threshold and a correction value for the length of the integration time window used to generate the instantaneous differential inductance; the scaling factor and the correction value are generated in the following manner: Calculate the Euclidean distance between the current integrated feature vector and the standard feature vector corresponding to the normal operation mode in the pre-stored multimodal failure fingerprint database, and record the value of the Euclidean distance; The mean of the window differential inductance value sequence in the current comprehensive feature vector is compared with the mean of the window differential inductance value sequence in the standard feature vector corresponding to the normal operation mode: if the current mean is greater than the standard mean, a positive sign is generated; if the current mean is less than the standard mean, a negative sign is generated; if the current mean is equal to the standard mean, a zero sign is generated. Divide the Euclidean distance value by a preset reference distance value to obtain a dimensionless adjustment range coefficient. Then multiply the adjustment range coefficient by the generated positive, negative, or zero sign to obtain the signed adjustment amount. Add the signed adjustment amount to the value of one to obtain the proportional adjustment coefficient. Multiply the proportional adjustment coefficient by the currently used saturation distortion capture threshold, and use the product as the updated saturation distortion capture threshold. Add the signed adjustment amount to the value of one, and then multiply it by the currently used integration time window length to obtain the updated integration time window length. This updated integration time window length is the correction value of the integration time window length. The updated saturation distortion capture threshold is used for the next monitoring trigger, and the updated integration time window length is used for the flux linkage integration calculation when generating the instantaneous differential inductance next time.

Images