A non-invasive current inversion and monitoring method for parallel chips of a compression type IGBT based on modal decoupling

Abstract

This disclosure provides a non-invasive current inversion and monitoring method for press-fit IGBT parallel chips based on modal decoupling. The inversion method includes: acquiring non-invasive external magnetic field data of the device under test (DUT); based on the non-invasive external magnetic field data, inverting the modal coefficients corresponding to the external magnetic field data using a pre-constructed linear mapping model between the external magnetic field data and modal coefficients, where the modal coefficients are the expansion coefficient vectors of the parallel chip current distribution under a plane polynomial basis function; and based on the modal coefficients, reconstructing the internal parallel chip current distribution of the DUT using a pre-constructed modal basis matrix. This method, without damaging the device package, obtains modal coefficients from external magnetic field data and then reconstructs the internal parallel chip distribution using the modal coefficients, enabling accurate inversion of the internal current distribution of press-fit IGBTs with multiple chips connected in parallel.

Description

Technical Field

[0001] This disclosure relates to the field of power electronic device testing technology, specifically to a non-invasive current inversion and monitoring method for press-fit IGBT parallel chips based on modal decoupling. Background Technology

[0002] With the development of high-voltage direct current transmission systems, press-fit IGBTs are widely used due to their high reliability and short-circuit failure mode. However, the module usually contains dozens of parallel chips. Due to uneven pressure distribution and differences in stray parameters, uneven current distribution is prone to occur, leading to local overheating and premature failure.

[0003] Currently, the main method for monitoring the internal current distribution of press-fit IGBTs is invasive monitoring. This method requires damaging the device package, thereby compromising its hermeticity and accelerating device aging. Therefore, non-invasive measurement methods have received widespread attention due to their advantage of no physical contact. However, most existing non-invasive measurement methods can only detect the offset of the device's internal components relative to the current sharing condition, and cannot obtain more detailed "chip-level" current distribution information. Summary of the Invention

[0004] This disclosure addresses the problems existing in the prior art by providing a non-invasive current inversion and monitoring method for press-fit IGBT parallel chips based on modal decoupling. This method can achieve accurate inversion of the internal current distribution of multiple press-fit IGBTs connected in parallel, solving the problem of poor accuracy in obtaining the current distribution of parallel chips under non-invasive conditions in the prior art.

[0005] To achieve the above objectives, the technical solution adopted in this disclosure is as follows: The first aspect of this disclosure provides a non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling, comprising: acquiring non-invasive external magnetic field data of the device under test; based on the non-invasive external magnetic field data, inverting the modal coefficients corresponding to the external magnetic field data through a pre-constructed linear mapping model between the external magnetic field data and the modal coefficients, wherein the modal coefficients are the expansion coefficient vectors of the parallel chip current distribution under a plane polynomial basis function; wherein the linear mapping model is constructed based on multiple sets of offline calibration data, each set of offline calibration data including: calibration modal coefficients determined by invasive parallel chip current calibration data, and corresponding non-invasive external magnetic field calibration data; based on the modal coefficients, reconstructing the internal parallel chip current distribution of the device under test through a pre-constructed modal basis matrix; wherein the modal basis matrix is ​​constructed based on the planar coordinates of the center positions of each internal parallel chip of the device under test and a pre-selected plane polynomial basis function.

[0006] In some embodiments, the linear mapping model includes: In the formula, For modal coefficients, External magnetic field data, This is an L×M dimensional modal inversion matrix, where L is the modal order and M is the number of external magnetic field sensors. The L-dimensional constant bias is used to compensate for the systematic errors of zero drift in the measurement channel and the integrator bias. The T in the upper right corner indicates matrix transpose.

[0007] In some embodiments, the modal basis matrix is ​​constructed in the following manner: determining the planar coordinates of the center positions of each parallel chip inside the device under test and normalizing them to obtain the normalized planar coordinates of each parallel chip inside the device under test; selecting planar polynomial basis functions and constructing an initial modal basis matrix based on the normalized planar coordinates; performing zero-sum processing on the column vectors other than the first column in the initial modal basis matrix to decouple the total current from the current distribution shape, thereby obtaining the decoupled modal basis matrix.

[0008] In some embodiments, multiple sets of offline calibration data are obtained as follows: a plane polynomial basis function is selected, and the modal order of the plane polynomial basis function is determined; based on the relationship between the number of sensors and the modal order, the minimum number of sensors corresponding to the modal order is determined; any device of the same type as the device under test is selected, and corresponding current sensors are arranged on the current paths of each parallel chip inside the same type device, and magnetic field sensors with a number no less than the minimum number of sensors are arranged circumferentially outside the same type device; multiple sets of calibration experiments are performed, and in each set of calibration experiments, the invasive parallel chip current calibration data and the non-invasive external magnetic field calibration data of the same type device are collected simultaneously by the current sensors and the magnetic field sensors; based on the invasive parallel chip current calibration data and the pre-constructed modal basis matrix of the same type device, the corresponding calibration modal coefficients are obtained by solving the linear least squares method; the calibration modal coefficients and the non-invasive external magnetic field calibration data corresponding to each set of calibration experiments are taken as a set of offline calibration data to obtain multiple sets of offline calibration data.

[0009] In some embodiments, the normal equation for the linear least squares method is: In the formula, The modal coefficients of sample s are... Here is the modal basis matrix. The current data of the invasive parallel chip for sample s is shown. The upper right corner -1 indicates the inverse matrix, and the upper right corner T indicates the matrix transpose.

[0010] In some embodiments, a linear mapping model is constructed as follows: non-invasive external magnetic field calibration data from multiple sets of offline calibration data are stacked row-wise to form an external magnetic field sample matrix, and the corresponding calibration mode coefficients are stacked row-wise to form a mode coefficient sample matrix; based on the external magnetic field sample matrix and the mode coefficient sample matrix, the mode inversion matrix and constant bias are obtained by solving the linear least squares method; and a linear mapping model is constructed based on the mode inversion matrix and the constant bias.

[0011] In some embodiments, the internal parallel chip current distribution of the device under test is reconstructed using the following formula: In the formula, For the current distribution of internal parallel chips, For modal coefficients, Here is the modal basis matrix. In some embodiments, the modal order of the plane polynomial basis functions is selected as second order, corresponding to a minimum number of sensors of six.

[0012] A second aspect of this disclosure provides a monitoring method for press-fit IGBT parallel chips, comprising: reconstructing the internal parallel chip current distribution of the device under test according to the non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling provided in the first aspect or any embodiment of the first aspect; calculating the corresponding state monitoring index based on the internal parallel chip current distribution; wherein the state monitoring index includes the current variation coefficient and the maximum relative deviation; and monitoring the parallel chips of the device under test based on the current variation coefficient and the maximum relative deviation.

[0013] In some embodiments, monitoring the current distribution of the device under test includes: if the current variation coefficient or the maximum relative deviation exceeds a preset safety threshold, issuing an aging or fault warning for the device under test, and mapping the current distribution of the internal parallel chips into a planar cloud map for visualization, thereby completing the monitoring.

[0014] Compared with the prior art, this disclosure has the following beneficial effects: The method provided in this disclosure acquires non-invasive external magnetic field data and combines it with a pre-constructed linear mapping model to invert mode coefficients. Then, it reconstructs the internal parallel chip current distribution based on the mode basis matrix constructed using chip plane coordinates and plane polynomial basis functions. The linear mapping model is constructed from calibration mode coefficients determined by invasive current calibration data and corresponding non-invasive magnetic field calibration data to form offline calibration data. This method enables accurate reconstruction of the internal parallel chip current distribution of a press-fit IGBT using only external non-invasive magnetic field data acquisition, without damaging the device's packaging. This avoids the impact of invasive monitoring on device airtightness and lifespan. The high-dimensional chip current distribution is transformed into low-dimensional mode coefficients for inversion calculation. Combined with the offline calibrated linear mapping relationship, the underdetermined inverse problem of directly inverting a large number of internal chip currents from a small number of external sensors is effectively solved, reducing computational complexity and improving the stability and accuracy of the inversion. In addition, a mode basis matrix is ​​constructed based on the plane polynomial basis function and the actual plane coordinates of the chip, so that the mode coefficients are highly matched with the spatial characteristics of the chip current distribution. Combined with the offline calibrated mapping model, the chip-level accuracy of the current distribution reconstruction is guaranteed, realizing fine and non-invasive sensing of the current distribution of parallel chips inside the press-fit IGBT. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a non-intrusive current inversion method for press-fit IGBT parallel chips based on modal decoupling provided in this disclosure embodiment; Figure 2 This is a schematic diagram illustrating the correspondence between modal order and the number of sensors provided in an embodiment of this disclosure; Figure 3 This is a schematic flowchart of a monitoring method for a press-fit type IGBT parallel chip provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram illustrating the modal superposition and inversion principle of current distribution in a press-fit IGBT parallel chip provided in this embodiment of the disclosure; Figure 5 This is a schematic diagram of a press-fit IGBT with internal and external sensors installed, provided in an embodiment of this disclosure; Figure 6 This is a visualization cloud map of the non-intrusive inversion results of the press-fit IGBT parallel chip provided in this embodiment under normal operating conditions; Figure 7 This is a visualization cloud map of the non-intrusive inversion results of the press-fit IGBT parallel chip under abnormal operating conditions provided in this embodiment of the disclosure. Detailed Implementation

[0016] The present disclosure will now be further described with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present disclosure and should not be construed as limiting the scope of protection of the present disclosure. It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application.

[0017] The acquisition, transmission, storage, use, and processing of data in this disclosed technical solution comply with relevant national laws and regulations. In the embodiments of this disclosure, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this disclosure, and do not imply that the applicant has already used or necessarily used such solutions.

[0018] All terms used in this disclosure have the same meaning as understood by one of ordinary skill in the art to which this disclosure pertains, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries should be interpreted as having meanings consistent with their meanings in the context of the relevant art, and not as idealized or highly formalized, unless expressly defined herein.

[0019] Example 1; Figure 1 This is a flowchart illustrating a non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling, as provided in this disclosure embodiment. Figure 1 As shown, the specific steps include S11 to S13.

[0020] Step S11: Obtain non-invasive external magnetic field data of the device under test.

[0021] It should be noted that in the embodiments of this disclosure, the device under test refers to a press-fit IGBT device containing multiple parallel chips, which is the object of current distribution inversion in this disclosure; non-invasive external magnetic field data refers to the magnetic field related measurement data of the external space of the device under test collected without damaging the packaging structure of the device under test or making physical contact with the internal circuit of the device.

[0022] In one possible implementation, acquiring non-invasive external magnetic field data of a device under test (DUT) includes: arranging multiple segmented PCB Rogowski coils circumferentially on a two-dimensional plane outside the DUT; the output voltage of the segmented PCB Rogowski coils is processed by an integrator to obtain non-invasive external magnetic field data. Here, the two-dimensional plane refers to a spatial plane parallel to the chip layout plane of the DUT and located outside the DUT; circumferential arrangement means arranging sensors in a ring around the outer contour of the DUT; segmented PCB Rogowski coils refer to Rogowski coils fabricated using PCB (Printed Circuit Board) technology and divided into multiple segments, which are non-invasive sensors used to measure magnetic fields or currents; other non-invasive sensors can also be used, and this disclosure does not specifically limit this; the integrator refers to an electronic device capable of integrating electrical signals, used to convert the induced voltage signal output by the segmented PCB Rogowski coils into a measurement signal related to the magnetic field.

[0023] For example, in one specific implementation, a press-fit IGBT containing five parallel chips is used as the device under test. On a two-dimensional plane parallel to the chip arrangement plane of the device, six segmented PCB Rogowski coils are arranged circumferentially around the periphery of the device at equal 60° intervals. All coils are arranged close to the device casing. The induced voltage signal output by the coils is fed into an integrator with a bandwidth of 0-10MHz for integration processing. The integrator transmits the processed electrical signal to a data acquisition system, which samples and quantizes the signal to finally obtain the non-invasive external magnetic field data of the device under test.

[0024] The core purpose of acquiring non-invasive external magnetic field data of the device under test is to provide raw external sensing data for subsequent mode coefficient inversion. This step avoids the damage to the device's encapsulation and sealing caused by invasive measurements through a non-invasive acquisition method.

[0025] Step S12: Based on non-invasive external magnetic field data, the modal coefficients corresponding to the external magnetic field data are obtained by inversion through a pre-constructed linear mapping model between the external magnetic field data and the modal coefficients.

[0026] It should be noted that, in the embodiments of this disclosure, the linear mapping model refers to a mathematical model that establishes a linear correspondence between non-invasive external magnetic field data and modal coefficients, which is used to realize the inversion calculation from external magnetic field data to modal coefficients; the modal coefficients refer to the expansion coefficient vector of the parallel chip current distribution under the plane polynomial basis function, which can characterize the high-dimensional chip current spatial distribution characteristics in a low-dimensional form.

[0027] In some embodiments, the linear mapping model includes: , In the formula, For modal coefficients, External magnetic field data, This is an L×M dimensional modal inversion matrix, where L is the modal order and M is the number of external magnetic field sensors. The L-dimensional constant bias is used to compensate for the systematic errors of zero drift in the measurement channel and the integrator bias. The T in the upper right corner indicates matrix transpose.

[0028] The linear mapping model is constructed based on multiple sets of offline calibration data. Each set of offline calibration data includes: calibration mode coefficients determined by invasive parallel chip current calibration data, and corresponding non-invasive external magnetic field calibration data.

[0029] It should be noted that offline calibration data refers to sample data collected in an offline experimental environment for constructing and calibrating linear mapping models, including calibration modal coefficients and corresponding non-invasive external magnetic field calibration data; invasive parallel chip current calibration data refers to the real current data of each parallel chip inside the calibration device collected by invasive sensors, serving as the reference data for calibration; calibration modal coefficients refer to the true values ​​of modal coefficients used to calibrate linear mapping models, obtained by combining invasive parallel chip current calibration data with the modal basis matrix. In some embodiments, multiple sets of offline calibration data are obtained as follows: a plane polynomial basis function is selected, and the modal order of the plane polynomial basis function is determined; based on the relationship between the number of sensors and the modal order, the minimum number of sensors corresponding to the modal order is determined; any device of the same type as the device under test is selected, and corresponding current sensors are arranged on the current paths of each parallel chip inside the same type device, and magnetic field sensors with a number no less than the minimum number of sensors are arranged circumferentially outside the same type device; multiple sets of calibration experiments are performed, and in each set of calibration experiments, the invasive parallel chip current calibration data and the non-invasive external magnetic field calibration data of the same type device are collected simultaneously by the current sensors and the magnetic field sensors; based on the invasive parallel chip current calibration data and the pre-constructed modal basis matrix of the same type device, the corresponding calibration modal coefficients are obtained by solving the linear least squares method; the calibration modal coefficients and the non-invasive external magnetic field calibration data corresponding to each set of calibration experiments are taken as a set of offline calibration data to obtain multiple sets of offline calibration data.

[0030] It should be noted that planar polynomial basis functions refer to basis functions in polynomial form constructed based on planar coordinates, which are the foundation for realizing current distribution mode decomposition; identical devices refer to press-fit IGBT devices with the same model, specifications, and internal chip arrangement as the device under test, serving as experimental objects for offline calibration; current sensors refer to sensors used to measure the actual current of the chip, which in this embodiment is an embedded PCB Rogowski coil, where an embedded PCB Rogowski coil refers to a device using a PCB (Printed Circuit Board)... A Rogowski coil, fabricated using PCB (Printed Circuit Board) technology and embedded in the chip's current path, can also be used; no specific limitation is made. Linear least squares is a mathematical optimization method that finds the best function match for data by minimizing the sum of squared errors. It is used to solve for calibration mode coefficients from invasive parallel chip current calibration data. The modal basis matrix is ​​a matrix constructed based on chip planar coordinates and plane polynomial basis functions, mapping the current distribution to the modal coefficients. The modal order refers to the highest order of the plane polynomial basis functions, determining the dimension of the modal coefficients and the accuracy of the inversion. The minimum number of sensors refers to the minimum number of external magnetic field sensors required to achieve the current distribution inversion for the corresponding modal order.

[0031] In one possible implementation, such as Figure 2 The correspondence shown determines the minimum number of sensors corresponding to a given modal order. (See also...) Figure 2 When the modal order of the plane polynomial basis functions is selected as second order, the minimum number of sensors is 6.

[0032] In some embodiments, the normal equation for the linear least squares method is: , In the formula, The modal coefficients of sample s are... Here is the modal basis matrix. The current data of the invasive parallel chip for sample s is shown. The upper right corner -1 indicates the inverse matrix, and the upper right corner T indicates the matrix transpose.

[0033] In some embodiments, a linear mapping model is constructed as follows: non-invasive external magnetic field calibration data from multiple sets of offline calibration data are stacked row-wise to form an external magnetic field sample matrix, and the corresponding calibration mode coefficients are stacked row-wise to form a mode coefficient sample matrix; based on the external magnetic field sample matrix and the mode coefficient sample matrix, the mode inversion matrix and constant bias are obtained by solving the linear least squares method; and a linear mapping model is constructed based on the mode inversion matrix and the constant bias.

[0034] The system is based on non-invasive external magnetic field data and uses a pre-built linear mapping model to invert modal coefficients. The core idea is to use the linear correspondence between the magnetic field data obtained from offline calibration and the modal coefficients to convert the online-acquired external magnetic field data into low-dimensional modal coefficients that can characterize the internal current distribution. This step transforms the underdetermined inverse problem of directly inverting a large number of internal chip currents from a small number of external sensors into an easily solvable problem of inverting low-dimensional modal coefficients from magnetic field data through modal dimensionality reduction, which greatly reduces the computational complexity. At the same time, the linear mapping model is built based on multiple sets of offline calibration data, which ensures the accuracy and stability of the inversion results.

[0035] Step S13: Based on the modal coefficients, the internal parallel chip current distribution of the device under test is reconstructed using the pre-constructed modal basis matrix.

[0036] It should be noted that, in the embodiments of this disclosure, the internal parallel chip current distribution refers to the current value and spatial distribution characteristics of each parallel chip inside the device under test, which is the final inversion target of this disclosure.

[0037] In some embodiments, the internal parallel chip current distribution of the device under test is reconstructed using the following formula: , In the formula, For the current distribution of internal parallel chips, For modal coefficients, is the modal basis matrix.

[0038] The modal basis matrix is ​​constructed based on the planar coordinates of the center positions of the parallel chips inside the device under test and pre-selected planar polynomial basis functions.

[0039] It should be noted that the planar coordinates of the center position refer to the coordinate values ​​of the geometric center of each parallel chip inside the device under test in a preset two-dimensional coordinate system; normalization processing refers to the mathematical processing method of converting the planar coordinates of the chip into values ​​in the range of 0-1, which is used to eliminate the influence of the absolute value of the coordinates on the construction of the modal basis matrix.

[0040] Additionally, it should be noted that the construction of the modal basis matrix is ​​closely related to the chip arrangement inside the device under test. Different numbers of chips and different arrangement methods of press-fit IGBTs result in different dimensions and element values ​​of the modal basis matrix. The system will construct a dedicated modal basis matrix based on the actual chip coordinates of the device under test to ensure the accuracy of current distribution reconstruction.

[0041] In some embodiments, the modal basis matrix is ​​constructed in the following manner: determining the planar coordinates of the center positions of each parallel chip inside the device under test and normalizing them to obtain the normalized planar coordinates of each parallel chip inside the device under test; selecting planar polynomial basis functions and constructing an initial modal basis matrix based on the normalized planar coordinates; performing zero-sum processing on the column vectors other than the first column in the initial modal basis matrix to decouple the total current from the current distribution shape, thereby obtaining the decoupled modal basis matrix.

[0042] The system reconstructs the current distribution of internal parallel chips based on modal coefficients and a pre-constructed modal basis matrix. The core idea is to use the modal basis matrix to achieve the inverse mapping from low-dimensional modal coefficients to high-dimensional chip current distribution. The low-dimensional modal coefficients that can characterize the current distribution are restored to the specific current values ​​of each parallel chip. This step is the inverse process of modal dimensionality reduction inversion. Through the pre-constructed modal basis matrix, the mapping from modal coefficients to current distribution can be accurately realized. At the same time, the modal basis matrix is ​​decoupled from the shape of the total current and the current distribution through zero-sum processing, so that the reconstructed current distribution can clearly distinguish the overall current level and local distribution characteristics, providing accurate chip-level current data for subsequent device state assessment.

[0043] Example 2; Figure 3 This is a flowchart illustrating a monitoring method for a press-fit type IGBT parallel chip provided in an embodiment of this disclosure, as shown below. Figure 3 As shown, the specific steps include S21 to S23.

[0044] Step S21: Obtain the internal parallel chip current distribution of the device under test.

[0045] Based on the non-invasive current inversion method for press-fit IGBT parallel chips provided in the above embodiments, the internal parallel chip current distribution of the device under test is reconstructed. Specific steps will not be elaborated here.

[0046] Step S22: Calculate the corresponding status monitoring indicators based on the current distribution of the internal parallel chips; wherein, the status monitoring indicators include the current variation coefficient and the maximum relative deviation.

[0047] Step S23: Monitor the parallel chips of the device under test based on the current variation coefficient and the maximum relative deviation.

[0048] In some embodiments, monitoring the parallel chips of the device under test includes: If the current variation coefficient or the maximum relative deviation exceeds the preset safety threshold, an aging or fault warning for the device under test will be issued, and the current distribution of the internal parallel chips will be mapped as a planar cloud map for visualization and characterization, thus completing the monitoring.

[0049] Example 3; This embodiment provides a non-invasive current inversion and monitoring method for press-pack IGBT parallel chips based on modal decoupling. Its technical route mainly includes: constructing a modal-based current distribution model, abstracting the inverse problem in current modes and modal space, state monitoring indicators based on modal currents, solving current modal coefficients, calibrating and inverting modal coefficients of a segmented Rogowski coil sensor. The detailed explanations are as follows.

[0050] Constructing a modal-based current distribution model: Denote the internal chip current of a press-pack IGBT device with N parallel chips at any moment as Equation 1: ; Suppose M sensors are arranged outside the device. Usually M < N. The outputs after signal processing at the same moment can be denoted as Equation 2: ; According to the Biot-Savart law, the magnetic field result y measured by the external sensors is driven by the current i of the press-pack IGBT parallel chips. The relationship between it and the internal chip current i can be written as Equation 3: ; Where S is the sensitivity matrix determined by the geometric structure and the coupling relationship of the field source. Since M < N, directly reconstructing N chip currents i from M external sensors, the corresponding inverse problem is highly underdetermined and ill-posed, extremely sensitive to noise and modeling errors, and is not feasible in engineering.

[0051] Therefore, this disclosure does not directly solve for N current values, but by constructing appropriate spatial modes, maps the high-dimensional parallel chip current distribution to low-dimensional modal coefficients, and then constructs state indicators directly related to operation and maintenance requirements based on the modal coefficients to achieve identifiable low-dimensional state perception.

[0052] To characterize the spatial distribution of the parallel chip current inside the module, first introduce the planar geometric coordinates of each parallel chip. Denote the center coordinates of the nth parallel chip as ( x n , y n ), n = 1, …, N. On the normalized planar coordinates (x, y), consider a set of planar polynomial basis functions, that is, Equation 4: ; Where d is the highest polynomial order, and the modal order is uniquely determined by the highest order d. Typically, in ascending order, the planar polynomial basis functions can be arranged as Equation 5: ; Where L is the modal order. Consider the current distribution I(x, y) inside the press-pack power module as a discrete current field on the plane. Under polynomial modes, expand I(x, y) as Equation 6: ; in a p,q These are the modal expansion coefficients.

[0053] For the nth parallel chip, its current can be approximately expressed as Equation 7: ; The elements of the modal basis matrix T are defined as shown in Equation 9: ; Where L is the number of basis functions.

[0054] Then all chip currents can be written in matrix form, i.e., Equation 10: ; When the modal order d is selected and L is much smaller than the number of parallel chips N, a It can provide a low-dimensional, physically meaningful characterization of the current distribution, facilitating inversion from a limited number of sensor results, such as... Figure 4 As shown.

[0055] exist Figure 4 The top six subplots illustrate the spatial distribution of the six modes (modes 1 to 6) corresponding to the second-order planar polynomial basis functions on the chip plane. Mode 1 represents the total current mode, corresponding to a basis function uniformly distributed across the entire plane, characterizing the total current level of the device. Its value is the same across all chips and is directly related to the total current. Modes 2 to 6 represent the current distribution shape modes, corresponding to basis functions with spatial variation characteristics, characterizing the relative distribution shape of the current among the chips. After zero-sum processing, the sum of the currents of each shape mode across all chips is zero, completely decoupled from the total current. In the figure, the two-dimensional coordinates (x, y) of each subplot are consistent with the chip plane layout coordinates. The color gradient represents the basis function value of that mode at the corresponding position, and the color scale on the right provides a quantization reference. The bottom left subplot illustrates the core process of mode superposition: the coefficients of each mode are weighted and superimposed with the corresponding mode basis functions, and then all the weighted mode results are summed to restore a high-dimensional chip-level current distribution cloud map. The right-hand sub-figure below maps the continuous current distribution cloud map obtained by modal superposition onto the actual chip array. Each colored dot corresponds to a parallel chip, and the color of the dot corresponds one-to-one with the current value, realizing a precise mapping from continuous distribution to discrete chip current.

[0056] Abstraction of current modes and inverse problems in mode space: In condition monitoring of press-fit devices, total current and current distribution shape typically play different roles: total current is mainly determined by system control and load, while non-uniform parallel current more directly reflects device condition and degradation level. Therefore, it is necessary to separate the "overall current level" and "spatial non-uniform shape" at the modal level.

[0057] In polynomial mode construct 5, the first mode is a constant term, corresponding to the mode basis matrix T. n,1 =1. At this point, the coefficient of this mode is... a 1. Average chip current I ave Total current of press-fit IGBTs I tot Equation 11 is satisfied between them: .

[0058] To further determine the total current I determined by the system and load tot To decouple the current distribution shape mode from the press-fit IGBT device, the mode basis of the 2nd to Lth columns of the mode basis matrix T is zero-sum processed, i.e., Equation 12 is executed: ; After processing, the corresponding columns of T are replaced to obtain a new modal basis matrix, still denoted as T. At this point, for each mode with l≥2, its sum across all chips is zero; they only change the relative distribution between chips, without changing the total current.

[0059] Under this modal configuration, the total current can be approximated by Equation 13: ; This indicates the total current mode coefficient a 1. Only the total current of the press-fit IGBT is described, while the shape modal coefficients are not. a 2,…, a L This characterizes the spatial non-uniform shape of the current distribution, completely decoupling it from the total current. This makes it convenient to subsequently evaluate the total current mode and the current distribution shape mode.

[0060] After obtaining Equation 10 above, the original inverse problem Equation 3 can be rewritten in modal space. The relationship between external sensor measurements and chip current can be expressed as Equation 14: ; Here, H is an M-row L-column matrix, representing the equivalent sensitivity matrix from each mode to the sensor output. Thus, the original inverse problem, "inverting the chip current i from external measurement y," is transformed into "first inverting the mode coefficients a from external measurement y, then reconstructing i from the modes." Given that the number of modes L is much smaller than the number of chips N and does not exceed the effective information dimension provided by the external sensors, this inverse mapping can reflect the main characteristics of the current distribution under a limited sensor configuration, possessing an engineering-feasible implementation path.

[0061] Condition monitoring indicators based on modal current: This disclosure uses the current coefficient of variation (CV) and the maximum relative deviation D. max The two indicators serve as state variables for press-fit IGBTs, measuring their overall non-uniformity and extreme non-uniformity, respectively.

[0062] For the chip current i at a certain moment, its standard deviation is given by Equation 15: ; When the current of all chips is close to the average value, the CV is small, which means that the current sharing of parallel chips is good; when there is a significant high-low current split, the CV increases, indicating that the overall current unevenness of the press-fit IGBT increases.

[0063] To reflect the deviation of the "worst chip" from the average current, the maximum relative deviation formula 16 is defined: .

[0064] CV characterizes the overall non-uniformity of current distribution in press-fit IGBTs, D max This characterizes localized extreme non-uniformity, and the combination of these two factors can effectively describe the current distribution state of the module. In online monitoring, thresholds based on the safe operating area can be set for these characteristics to provide early warning of worsening trends in current non-uniformity.

[0065] Solution of current mode coefficients: like Figure 5 As shown, an embedded PCB Rogowski coil is installed in the current path of each parallel chip, and the actual current i of the chip is measured. n Meanwhile, M segmented PCB Rogowski coils are arranged on the plane where the components are located on the outside of the module. Each segmented coil covers a portion of the magnetic flux around the module, and its output voltage is integrated to obtain the external magnetic flux measurement result y.

[0066] In each calibration experiment, the transient current waveform of the internal parallel chip and the transient magnetic flux waveform of the external segmented coil are simultaneously acquired, and the waveforms are sampled sequentially according to the sampling step size to obtain several pairs of samples (i s,y s After obtaining the chip coordinates and modal basis matrix T, for each sample (i) s ,y s The chip current i, measured invasively, will be... s The modal coefficients are obtained by performing least-squares fitting under the modal basis T. a s According to the normal equation of linear least squares, 17: ; Repeat the above process for all calibration samples to obtain a set. a s The "true value modal coefficients" of the calibration samples serve as the target quantities for subsequent "external coil → modal coefficient" mapping calibration.

[0067] Modal coefficient calibration and inversion of segmented Rogowski coil sensor: This disclosure does not directly solve for H, but instead uses a data-driven approach to directly calibrate from y in the sample space. s to a s The linear mapping from the outer coil to the modal coefficients for the s-th sample can be described by Equation 18: ; Where G is the L×M mode inversion matrix, and b is an L-dimensional constant bias used to compensate for systematic errors such as zero drift in the measurement channel and integrator bias.

[0068] All calibration samples are stacked row by row to form the outer coil sample matrix Y and the modal coefficient matrix A, as shown in Equation 19: ; Based on the acquired calibration data, G and b are solved using the standard linear least squares method. After calibration, invasive sensors are no longer required during online operation. For the external segment Rogowski coil measurement y(t) at any given time, the estimated modal coefficients are obtained using the aforementioned linear model Equation 18. a Pred Then, combined with Equation 10, the chip current distribution estimation can be reconstructed. i Pred Furthermore, the current variation coefficient CV and the maximum relative deviation D are calculated based on equations 15 and 16. max This allows for online monitoring and condition assessment of the current non-uniformity of press-fit IGBTs, relying only on a small number of external segmented Rogowski coil sensors.

[0069] Example 4; This embodiment provides a non-intrusive current inversion and monitoring method for press-fit IGBT parallel chips based on modal decoupling, which is specifically divided into the following calibration stage and testing stage.

[0070] Calibration phase: Step 1: Construct the monitoring system hardware platform. Taking the press-fit power module as the object under test, establish a coordinate system and determine the center coordinates of the N parallel chips based on the geometric layout of the parallel chips inside the module. x n , y n On the two-dimensional plane outside the module, M segmented PCB Rogowski coils are arranged circumferentially to non-invasively acquire the external magnetic flux signal y of the module; embedded PCB Rogowski coils are arranged around each parallel chip inside the module to obtain the actual current i of the parallel chip as a calibration reference.

[0071] Step 2: Construct the decoupled modal basis matrix. Select a set of plane polynomials as basis functions to construct the modal basis matrix T. To achieve state decoupling, perform zero-sum processing on the higher-order mode column vectors in matrix T except for the first column (constant term), so that the higher-order modes only represent the shape of the current spatial distribution and are decoupled from the total current amplitude.

[0072] Step 3: Offline Data Acquisition and Model Calibration. Calibration experiments were conducted in a laboratory environment. Multiple sets of data were acquired by adjusting the test conditions, and the true value of the internal chip current was recorded simultaneously. s and external coil measurement value y s The true modal coefficients are obtained by projecting the internal current into the modal space using the least squares method. a s Subsequently, a linear mapping model from the external measurement value y to the modal coefficient a is established, and the modal inversion matrix G and the bias vector b are obtained through regression analysis.

[0073] Testing phase: Step 1: Online mode inversion and current reconstruction. In the actual online monitoring stage, no internal sensors are required; only external coils are used to collect real-time magnetic flux signals y(t). The signals are substituted into the calibrated linear model (18) to quickly invert and obtain the current mode coefficient estimates; then, the current is calculated using the mode basis matrix (10) to reconstruct the current distribution of all chips inside the module.

[0074] Step 5: State index calculation and result visualization. Based on the reconstructed current distribution i pred Calculate the coefficient of variation of current (CV, also known as the standard deviation of current distribution) and the maximum relative deviation (D). max The current distribution (also known as maximum current offset) is used as a condition monitoring indicator to quantify and evaluate the overall non-uniformity and local extreme non-uniformity of the module current. Finally, the system maps the calculated current distribution data into a planar cloud map and calculates the corresponding quantitative indicators. The resulting monitoring results are as follows: Figure 6 and Figure 7 As shown in the figure, this result visually demonstrates the current density distribution within the module, effectively identifying abnormal states such as current bias or localized hot spots.

[0075] like Figure 6 As shown, the standard deviation of the chip current distribution CV = 0.38, and the maximum chip current offset D max =1.15, the current distribution shows a regular diamond-shaped gradual change, with no obvious current bias or local hot spots, indicating that the current sharing of the parallel chips inside the device is good. For example Figure 7 As shown, the standard deviation of the chip current distribution CV = 0.51, and the maximum chip current offset D max =1.59, compared to normal operating conditions, CV and D max All values ​​increased significantly, and a distinct warm-toned area appeared in the cloud map, indicating severe current bias and local hot spots inside the device.

[0076] It should be noted that the terms "first," "second," and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different parts. Terms such as "including" or "contains" mean that the element preceding the word covers the element listed after the word, and do not exclude the possibility of covering other elements as well.

[0077] Although operations are described in a specific order in the accompanying drawings in this disclosure, it should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the shown operations to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0078] Finally, it should be noted that the above content is only used to illustrate the technical solution of this disclosure, and is not intended to limit the scope of protection of this disclosure. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of this disclosure do not depart from the substance and scope of the technical solution of this disclosure.

Claims

1. A non-invasive current inversion method for press-fit parallel IGBT chips based on modal decoupling, characterized in that, include: Acquire non-invasive external magnetic field data of the device under test; Based on the non-invasive external magnetic field data, the modal coefficients corresponding to the external magnetic field data are obtained by inverting the model of linear mapping between the external magnetic field data and the modal coefficients. The modal coefficients are the expansion coefficient vectors of the parallel chip current distribution under the plane polynomial basis functions. The linear mapping model is constructed based on multiple sets of offline calibration data. Each set of offline calibration data includes: calibration modal coefficients determined by the invasive parallel chip current calibration data, and the corresponding non-invasive external magnetic field calibration data. Based on the modal coefficients, the internal parallel chip current distribution of the device under test is reconstructed using a pre-constructed modal basis matrix; wherein, the modal basis matrix is ​​constructed based on the planar coordinates of the center positions of each parallel chip in the device under test and pre-selected planar polynomial basis functions.

2. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 1, characterized in that, The linear mapping model includes: ， In the formula, These are the modal coefficients. External magnetic field data, This is an L×M dimensional modal inversion matrix, where L is the modal order and M is the number of external magnetic field sensors. The L-dimensional constant bias is used to compensate for the systematic errors of zero drift in the measurement channel and the integrator bias. The T in the upper right corner indicates matrix transpose.

3. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 1, characterized in that, The modal basis matrix is ​​constructed in the following manner, including: The planar coordinates of the center positions of each parallel chip inside the device under test are determined and normalized to obtain the normalized planar coordinates of each parallel chip inside the device under test. Select plane polynomial basis functions and construct an initial modal basis matrix based on the normalized plane coordinates; The column vectors in the initial modal basis matrix, except for the first column, are subjected to zero-sum processing to decouple the total current from the current distribution shape, thus obtaining the decoupled modal basis matrix.

4. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 1, characterized in that, The multiple sets of offline calibration data are obtained in the following manner: Select plane polynomial basis functions and determine the modal order of the plane polynomial basis functions; Based on the relationship between the number of sensors and the modal order, determine the minimum number of sensors corresponding to the modal order; Select any device of the same type as the device under test, arrange corresponding current sensors on the current paths of each parallel chip inside the device of the same type, and arrange magnetic field sensors along the circumferential direction outside the device of the same type with a number not less than the minimum number of sensors. Multiple sets of calibration experiments were conducted. In each set of calibration experiments, the invasive parallel chip current calibration data and the non-invasive external magnetic field calibration data of the same type of device were collected synchronously through the current sensor and the magnetic field sensor. Based on the current calibration data of the invasive parallel chip and the pre-constructed modal basis matrix of the same type of device, the corresponding calibration mode coefficients are obtained by solving the linear least squares method. The calibration mode coefficients and the non-invasive external magnetic field calibration data corresponding to each calibration experiment are used as a set of offline calibration data to obtain the multiple sets of offline calibration data.

5. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 4, characterized in that, The normal equation for the linear least squares method is: ， In the formula, Let be the modal coefficients of sample s. Here is the modal basis matrix. The current data of the invasive parallel chip for sample s is shown. The upper right corner -1 indicates the inverse matrix, and the upper right corner T indicates the matrix transpose.

6. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 1, characterized in that, The linear mapping model is constructed as follows: The non-invasive external magnetic field calibration data from the multiple sets of offline calibration data are stacked in rows to form an external magnetic field sample matrix, and the corresponding calibration mode coefficients are stacked in rows to form a mode coefficient sample matrix. Based on the external magnetic field sample matrix and the mode coefficient sample matrix, the mode inversion matrix and constant bias are obtained by solving the linear least squares method. The linear mapping model is constructed based on the mode inversion matrix and the constant bias.

7. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to claim 1, characterized in that, The internal parallel chip current distribution of the device under test can be reconstructed using the following formula: ， In the formula, For the current distribution of internal parallel chips, These are the modal coefficients. is the modal basis matrix.

8. The non-invasive current inversion method for press-fit IGBT parallel chips based on modal decoupling according to any one of claims 1-7, characterized in that, The modal order of the plane polynomial basis function is selected as second order, corresponding to a minimum number of six sensors.

9. A monitoring method for a press-fit type parallel IGBT chip, characterized in that, include: Obtain the internal parallel chip current distribution of the device under test; wherein, the internal parallel chip current distribution is reconstructed according to the non-intrusive current inversion method for press-fit IGBT parallel chips based on mode decoupling as described in any one of claims 1-8; Based on the current distribution of the internal parallel chips, the corresponding status monitoring indicators are calculated; wherein, the status monitoring indicators include the current variation coefficient and the maximum relative deviation; Based on the current variation coefficient and the maximum relative deviation, the parallel chips of the device under test are monitored.

10. The monitoring method for press-fit IGBT parallel chips according to claim 9, characterized in that, The parallel chip for monitoring the device under test includes: If the current variation coefficient or the maximum relative deviation exceeds the preset safety threshold, an aging or fault warning for the device under test is issued, and the current distribution of the internal parallel chip is mapped as a planar cloud map for visualization, thus completing the monitoring.

Images