Power module welding spot failure detection and life prediction method and system based on PINN

Abstract

The invention discloses a PINN-based power module welding spot failure detection and life prediction method and system, and belongs to the field of power module detection.The method comprises the steps that electric, thermal and mechanical force multi-physical field real-time monitoring is conducted on a welding spot, and a heat conduction equation, an improved Coffin-Manson equation and a Paris law are constructed; constructing a three-constraint PINN network architecture and setting a network training strategy; and inputting real-time monitoring data into the trained PINN network, and synchronously outputting three groups of prediction results of the welding spot residual life percentage, the critical crack size early warning value and the high-temperature dangerous area coordinate. According to the invention, the failure mechanism of the welding spot can be reflected more comprehensively and accurately, and the detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of power module detection, and particularly to a method and system for detecting solder joint failure and predicting the life of a power module based on PINN. Background Art

[0002] The failure of solder joints in power modules is a common problem in power electronic devices, involving multi-field coupling of electricity-thermal-mechanical forces. Traditional detection methods often rely on a single physical field and are difficult to comprehensively reflect the failure mechanism of solder joints. The lack of consideration of multi-physical field coupling in traditional detection methods limits the accuracy and reliability of detection results. At the same time, life prediction methods are also often troubled by problems such as inaccurate data and imprecise models. Existing life prediction models are often too simplified to accurately reflect the failure behavior of solder joints under actual working conditions. Summary of the Invention

[0003] The present invention aims to at least solve one of the technical problems existing in the prior art. For this purpose, the present invention proposes a method and system for detecting solder joint failure and predicting the life of a power module based on PINN, which can more comprehensively and accurately reflect the failure mechanism of solder joints and improve the accuracy of detection.

[0004] An embodiment of the present invention provides a method for detecting solder joint failure and predicting the life of a power module based on PINN, including the following steps: S100. Monitor the IGBT switch transient current amplitude based on a high-frequency current sensor, obtain the solder joint surface temperature field distribution based on an infrared thermal imager, and obtain the solder joint proximal temperature based on an embedded thermocouple implanted in the DBC substrate to obtain the current amplitude, temperature extreme value, temperature gradient, and number of thermal cycles; detect the solder joint plastic strain based on a micro-strain gauge mounted on the substrate adjacent to the solder joint, measure the main vibration frequency of the solder joint based on a laser Doppler vibrometer to obtain the plastic strain amplitude and the main vibration frequency; obtain the crack propagation acoustic signal based on an acoustic emission sensor and perform layer-by-layer scanning of internal defects based on a scanning acoustic microscope to obtain the crack length and the defect area ratio; S200. Calculate the heat source density through the Joule heat formula based on the current amplitude collected in step S100, and construct the boundary conditions of the three-dimensional heat conduction equation in combination with the infrared temperature field data; jointly input the temperature gradient and the plastic strain amplitude into the improved Coffin-Manson equation to calculate the thermo-mechanical fatigue damage factor; synchronously process the main vibration frequency and crack length data using the improved Paris law to obtain the crack propagation rate; use the heat source density, thermo-mechanical fatigue damage factor, and crack propagation rate as the physical constraints input of the PINN network;

[0005] The heat conduction equation is:

[0006] ;

[0007] where I is the current, ▽T is the temperature gradient, ρ is the density, is the specific heat capacity, is the thermal conductivity, and Qj is the current Joule heat term;

[0008] The improved Coffin-Manson equation is:

[0009] ;

[0010] where k is the temperature influence coefficient, ΔT is the temperature change range, T avg is the average temperature, Q is the activation energy, R is the gas constant, is the number of failure cycles, is the plastic strain amplitude, C is the thermo-mechanical coupling damage base coefficient, and n is the temperature gradient sensitivity index;

[0011] The improved Paris law is:

[0012] ;

[0013] where da / dN is the crack growth rate, N is the number of load cycles, a is the crack length, is the range of mode I stress intensity factor, is the range of mode II stress intensity factor, C(T) is the temperature-dependent material constant, m(T) is the temperature-dependent Paris exponent, η is the mixed-mode weight factor, and T is the operating temperature;

[0014] S300. Construct a three-constraint PINN network architecture: The first channel constrains the temperature field prediction through the residual term of the heat conduction equation; the second channel takes the strain amplitude and the number of thermal cycles as inputs and outputs the thermal fatigue accumulation through the Arrhenius-type damage equation; the third channel processes the crack growth data; and a thermo-mechanical coupling loss function is used to achieve multi-physical field coupling;

[0015] S400. Set the network training strategy: Use acoustic microscope data as the strong supervision signal for the crack growth branch and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balance the two types of physical loss functions through an adaptive weighting algorithm: the residual of the thermal damage equation and the residual of the crack growth equation;

[0016] S500. Input the real-time monitoring data into the trained PINN network and synchronously output three groups of prediction results: the percentage of the remaining life of the solder joint, the warning value of the critical crack size, and the coordinates of the high-temperature dangerous area; trigger a hierarchical alarm when any output value exceeds the preset threshold.

[0017] According to some embodiments of the present invention, the Arrhenius-type damage equation is:

[0018] ;

[0019] Among them, A is the material damage constant, Ta is the absolute temperature of the solder joint, and m is the strain sensitivity index.

[0020] According to some embodiments of the present invention, step S100 includes: the thermal-mechanical coupling loss function is:

[0021] ;

[0022] Among them, λ1,λ2,λ3 are adaptive weight coefficients, is the residual term of the heat conduction equation, D is the measured thermal damage factor, To predict the thermal damage factor, a is the measured crack length, To predict the crack length.

[0023] According to some embodiments of the present invention, step S100 also includes: S110, analyzing the size and solder joint layout of the power module to determine the critical solder joint area and potential failure points; S120, setting the layout positions of high-frequency current sensors, infrared thermal imagers, embedded thermocouples, micro strain gauges, laser Doppler vibrometers, acoustic emission sensors and scanning acoustic microscopes according to the distribution of critical solder joint areas and potential failure points; S130, determining the number of various sensors according to the number of solder joints and monitoring requirements.

[0024] According to some embodiments of the present invention, step S500 also includes: S510, screening and denoising the collected raw data, and eliminating outliers and invalid data; S520, scaling the cleaned data according to a certain ratio so that it falls within a specific numerical range; S530, extracting key features from the raw data based on the physical mechanism of solder joint failure and the requirements of the PINN network.

[0025] According to some embodiments of the present invention, the step S400 also includes: S410, setting the number of iterations according to the complexity of the PINN network and the scale of the training data; S420, during the training process, dynamically adjusting the learning rate according to the convergence of the network loss function, first using the first learning rate to accelerate convergence, and then using the second learning rate to fine-tune the network weights; the first learning rate is greater than the second learning rate.

[0026] According to some embodiments of the present invention, step S500 also includes: S540, setting different alarm levels according to the prediction results of the remaining life percentage of the weld, the critical crack size warning value and the coordinates of the high-temperature danger zone, including: emergency shutdown level, early warning maintenance level and monitoring attention level; S550, when the alarm level is the emergency shutdown level, immediately shut down for inspection; when the alarm level is the early warning maintenance level, arrange preventive maintenance; when the alarm level is the monitoring attention level, increase the monitoring frequency.

[0027] The method of the embodiment of the present invention has at least the following beneficial effects: By combining various sensors such as high-frequency current sensors, infrared thermal imagers, and embedded thermocouples, the embodiment of the present invention realizes the real-time monitoring of multiple physical fields of solder joints, including electricity, heat, and mechanical force; the multi-physical field monitoring method can more comprehensively and accurately reflect the failure mechanism of solder joints, improving the accuracy of detection. By using physical equations such as the improved Coffin-Manson equation and Paris law, the thermo-mechanical fatigue damage and crack propagation behavior of solder joints are more accurately described; a three-constraint PINN network architecture is constructed, and multi-physical field coupling prediction is realized through physical constraints, improving the accuracy and reliability of the prediction results. The combination of the improved physical equation and the PINN network architecture realizes the accurate prediction of key parameters such as the remaining life of solder joints and the critical crack size, enhancing the reliability of the prediction. According to the prediction results, different alarm levels are set, realizing the hierarchical management and preventive maintenance of the solder joint state, improving the maintenance efficiency and reducing the equipment failure rate.

[0028] On the other hand, the embodiment of the present invention provides a PINN-based power module solder joint failure detection and life prediction system, including: a data acquisition module for collecting the surface temperature field distribution of the solder joints of the power module, the temperature near the solder joints, the plastic strain of the solder joints, the main vibration frequency of the solder joints, the crack propagation acoustic signal, and internal defects; a data processing module for calculating the current amplitude, temperature extreme value, temperature gradient, and number of thermal cycles; calculating the plastic strain amplitude and the main vibration frequency; obtaining the crack length and the defect area ratio; calculating the heat source density based on the Joule heat formula; constructing the boundary conditions of the three-dimensional heat conduction equation by combining the infrared temperature field data; jointly inputting the temperature gradient and the plastic strain amplitude into the improved Coffin-Manson equation to calculate the thermo-mechanical fatigue damage factor; using the improved Paris law to process the data of the main vibration frequency and the crack length; a PINN network construction module for constructing a three-constraint PINN network architecture, including a temperature field prediction channel, a thermal fatigue cumulative amount output channel, and a crack propagation data processing channel; realizing multi-physical field coupling by using a thermal-mechanical coupling loss function; a network training module for setting network training strategies, including using acoustic microscope data as the strong supervision signal for the crack propagation branch and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balancing the two types of physical loss functions through an adaptive weighting algorithm: the residual of the thermal damage equation and the residual of the crack propagation equation; a prediction and alarm module for inputting the real-time monitoring data into the trained PINN network; synchronously outputting three groups of prediction results: the percentage of the remaining life of the solder joint, the warning value of the critical crack size, and the coordinates of the high-temperature dangerous area; triggering a hierarchical alarm when any output value exceeds the preset threshold.

[0029] According to some embodiments of the present invention, the data acquisition module includes: a high-frequency current sensor for monitoring the IGBT switching transient current amplitude; an infrared thermal imager for obtaining the surface temperature field distribution of the solder joint; an embedded thermocouple implanted in the DBC substrate for obtaining the temperature near the solder joint; a micro strain gauge mounted at the position of the substrate adjacent to the solder joint for detecting the plastic strain of the solder joint; a laser Doppler vibrometer for measuring the main vibration frequency of the solder joint; an acoustic emission sensor for obtaining the acoustic signal of crack propagation; and a scanning acoustic microscope for layer-by-layer scanning of internal defects.

[0030] According to some embodiments of the present invention, the PINN network module includes a heat conduction constraint branch, a thermal damage constraint branch, and a crack propagation constraint branch; the heat conduction constraint branch incorporates a three-dimensional heat conduction differential operator; the thermal damage constraint branch embeds an improved Coffin-Manson equation; the crack propagation constraint branch solidifies an improved Paris law calculation flow; the heat conduction constraint branch, the thermal damage constraint branch, and the crack propagation constraint branch achieve two-way gradient exchange through a shared hidden layer, and realize multi-physical field coupling prediction based on a physics-guided loss function.

[0031] The PINN-based power module solder joint failure detection and life prediction system according to the embodiments of the present invention has at least the following beneficial effects: the embodiments of the present invention can achieve all the technical effects of the method embodiments described above.

[0032] The additional aspects and advantages of the present invention will be partially given in the following description, partially become apparent from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the description of the embodiments in conjunction with the following drawings, wherein:

[0034] Figure 1 is a schematic flow chart of the method according to the embodiment of the present invention;

[0035] Figure 2 is a schematic block diagram of the modules of the system according to the embodiment of the present invention.

[0036] REFERENCE SIGNS:

[0037] Data acquisition module 100, data processing module 200, PINN network construction module 300, network training module 400, prediction and alarm module 500. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0039] In the description of the present invention, the meaning of "several" is one or more, the meaning of "multiple" is two or more, and understandings such as "greater than", "less than", "exceeding", etc. do not include the present number, and understandings such as "above", "below", "within", etc. include the present number. If there is a description of "first" and "second", it is only for the purpose of distinguishing technical features and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence relationship of the indicated technical features.

[0040] Referring to Figure 1 , an embodiment of the present invention provides a controller processing optimization method based on test data analysis, including the following steps:

[0041] S100. Monitor the IGBT switch transient current amplitude based on a high-frequency current sensor, obtain the surface temperature field distribution of the solder joint based on an infrared thermal imager, and obtain the temperature near the solder joint based on an embedded thermocouple implanted in the DBC substrate, so as to obtain the current amplitude, temperature extreme value, temperature gradient, and number of thermal cycles; detect the plastic strain of the solder joint based on a micro-strain gauge mounted on the substrate adjacent to the solder joint, measure the main vibration frequency of the solder joint based on a laser Doppler vibrometer, so as to obtain the plastic strain amplitude and the main vibration frequency; obtain the crack propagation acoustic signal based on an acoustic emission sensor, and perform layer-by-layer scanning of internal defects based on a scanning acoustic microscope, so as to obtain the crack length and the defect area ratio.

[0042] S200. Calculate the heat source density through the Joule heat formula based on the current amplitude collected in step S100, and combine the infrared temperature field data to construct the boundary conditions of the three-dimensional heat conduction equation; jointly input the temperature gradient and the plastic strain amplitude into the improved Coffin-Manson equation to calculate the thermo-mechanical fatigue damage factor; synchronously process the main vibration frequency and crack length data using the improved Paris law to obtain the crack propagation rate; use the heat source density, thermo-mechanical fatigue damage factor, and crack propagation rate as the physical constraints input of the PINN network.

[0043] The heat conduction equation is:

[0044] ;

[0045] where I is the current, ▽T is the temperature gradient, ρ is the density, is the specific heat capacity, is the thermal conductivity, and Qj is the current Joule heat term.

[0046] The improved Coffin - Manson equation is as follows:

[0047] ;

[0048] where k is the temperature influence coefficient, ΔT is the temperature change range, T_avg is the average temperature, Q is the activation energy, R is the gas constant, is the number of failure cycles, is the plastic strain amplitude, C is the thermo - mechanical coupling damage base coefficient, and n is the temperature gradient sensitivity index.

[0049] In this embodiment, the improved Coffin - Manson equation introduces a temperature factor to more accurately describe the thermo - mechanical fatigue behavior of solder joints under temperature - varying conditions. The improvement is mainly based on experimental data and changes in material properties. Specifically, the high - temperature dwell time and the temperature change range jointly determine the thermo - mechanical fatigue behavior of solder joints. Therefore, parameters such as the temperature influence coefficient k, the temperature change range ΔT, the average temperature T_avg, and the activation energy Q are introduced into the equation to correct the relationship between the plastic strain amplitude and the number of failure cycles. Such an improvement enables the equation to more accurately reflect the thermo - mechanical fatigue behavior of solder joints under actual working conditions.

[0050] In this embodiment, the thermo - mechanical coupling damage base coefficient reflects the damage accumulation characteristics of materials under thermo - mechanical coupling actions and is usually obtained through experimental measurements. Specifically, by conducting fatigue tests under different temperature and stress conditions, observing the relationship between the number of failure cycles and the plastic strain amplitude of solder joints, and then obtaining the value of C through data fitting. In addition, the value of C may also be affected by factors such as the solder joint material and the thickness of the interfacial IMC, so it may need to be corrected in actual applications.

[0051] In this embodiment, the temperature gradient sensitivity index is an exponential parameter that describes the degree of influence of temperature changes on the thermo - mechanical fatigue damage of solder joints. It is obtained through experimental measurements. By conducting fatigue tests under different temperature gradients, observing the relationship between the number of failure cycles of solder joints and the temperature change range, and then obtaining the value of n through data fitting. The value of n reflects the contribution degree of temperature changes to the fatigue damage of solder joints.

[0052] The improved Paris law is as follows:

[0053] ;

[0054] where da / dN is the crack growth rate, N is the number of load cycles, a is the crack length, is the range of the mode - I stress intensity factor, is the range of the mode II stress intensity factor, C(T) is the temperature-dependent material constant, m(T) is the temperature-dependent Paris exponent, η is the mixed-mode weight factor, and T is the operating temperature.

[0055] In this embodiment, the improved Paris law introduces the temperature-dependent material constant C(T), the temperature-dependent Paris exponent m(T), and the mixed-mode weight factor η to more accurately describe the crack growth behavior of solder joints under complex stress conditions. The improvement is also based on experimental data and changes in material properties. Especially for materials such as Sn-based solders, their crack growth behavior may be affected by multiple stress components, so it is necessary to introduce a mixed-mode weight factor to more accurately describe this effect. At the same time, the influence of temperature on material constants and the Paris exponent cannot be ignored, so it is necessary to determine the temperature dependence of these parameters through experimental data.

[0056] In this embodiment, the range of the mode I stress intensity factor characterizes the change range of the stress field intensity of the opening-mode crack under cyclic loading, which is defined as the difference between the maximum and minimum stress intensity factors. Through the fatigue crack growth test, combined with digital image correlation (DIC) or electrical measurement method to record the relationship between the crack length a and the number of cycles N, the range of the mode I stress intensity factor is inversely deduced. Similarly, the range of the mode II stress intensity factor is the range of the driving force for the sliding-mode crack, and the obtaining method refers to the range of the mode I stress intensity factor.

[0057] In this embodiment, the mixed-mode weight factor is used to describe the relative contributions of the two stresses to the crack growth rate under complex stress states (i.e., when both normal stress and shear stress exist). It is usually obtained through experimental measurement and data analysis. The value of η can be obtained by conducting crack growth tests under different stress ratios, observing the relationship between the crack growth rate and the range of the stress intensity factor, and then fitting the data.

[0058] In this embodiment, C(T) and m(T) are the temperature-dependent material constant and Paris exponent respectively, which describe the relationship between the crack growth rate and the range of the stress intensity factor at different temperatures. They are obtained through experimental measurement. Specifically, crack growth tests can be carried out at different temperatures, observing the relationship between the crack growth rate and the range of the stress intensity factor, and then the values of C(T) and m(T) can be obtained by fitting the data.

[0059] S300. Construct a three-constraint PINN network architecture: The first channel constrains the temperature field prediction through the residual term of the heat conduction equation; the second channel takes the strain amplitude and the number of thermal cycles as inputs and outputs the thermal fatigue cumulative amount through the Arrhenius-type damage equation; the third channel processes the crack growth data; a thermo-mechanical coupling loss function is used to achieve multi-physical field coupling.

[0060] S400, set the network training strategy: use acoustic microscope data as a strong supervision signal for the crack extension branch, and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balance two types of physical loss functions through an adaptive weighted algorithm: thermal damage equation residual and crack extension equation residual.

[0061] S500, input the real-time monitoring data into the trained PINN network, and simultaneously output three sets of prediction results: the percentage of remaining life of the solder joint, the critical crack size warning value, and the coordinates of the high-temperature danger zone; when any output value exceeds the preset threshold, a graded alarm is triggered.

[0062] In some embodiments, the Arrhenius-type damage equation is:

[0063] ;

[0064] Among them, A is the material damage constant, Ta is the absolute temperature of the solder joint, and m is the strain sensitivity index.

[0065] In some embodiments, the thermal-mechanical coupling loss function is:

[0066] ;

[0067] Among them, λ1,λ2,λ3 are adaptive weight coefficients, is the residual term of the heat conduction equation, D is the measured thermal damage factor, To predict the thermal damage factor, a is the measured crack length, To predict the crack length.

[0068] In some embodiments, step S100 also includes: S110, analyzing the size and solder joint layout of the power module to determine the critical solder joint area and potential failure points; S120, setting the layout positions of high-frequency current sensors, infrared thermal imagers, embedded thermocouples, micro strain gauges, laser Doppler vibrometers, acoustic emission sensors and scanning acoustic microscopes according to the distribution of critical solder joint areas and potential failure points; S130, determining the number of various sensors according to the number of solder joints and monitoring requirements.

[0069] In some embodiments, step S500 also includes: S510, screening and denoising the collected raw data, and eliminating outliers and invalid data; S520, scaling the cleaned data according to a certain ratio so that it falls within a specific numerical range; S530, extracting key features from the raw data based on the physical mechanism of solder joint failure and the requirements of the PINN network.

[0070] In some embodiments, step S400 further includes: S410. Setting the number of iterations according to the complexity of the PINN network and the scale of the training data; S420. During the training process, dynamically adjusting the learning rate according to the convergence of the network loss function. First, adopt a first learning rate to accelerate convergence, and then adopt a second learning rate to finely adjust the network weights; the first learning rate is greater than the second learning rate.

[0071] In some embodiments, step S500 further includes: S540. Setting different alarm levels according to the prediction results of the remaining life percentage of the solder joint, the early warning value of the critical crack size, and the coordinates of the high-temperature danger area, including: emergency shutdown level, early warning maintenance level, and monitoring attention level; S550. Immediately shutting down for inspection when the alarm level is the emergency shutdown level; arranging preventive maintenance when the alarm level is the early warning maintenance level; and increasing the monitoring frequency when the alarm level is the monitoring attention level.

[0072] Corresponding to the foregoing embodiments, the present invention also provides an embodiment of the system. For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment.

[0073] Referring to Figure 2 , an embodiment of the present invention provides a power module solder joint failure detection and life prediction system based on PINN, including:

[0074] A data acquisition module 100, configured to acquire the surface temperature field distribution of the solder joint of the power module, the temperature near the solder joint, the plastic strain of the solder joint, the main vibration frequency of the solder joint, the crack propagation sound signal, and internal defects.

[0075] A data processing module 200, configured to calculate the current amplitude, the temperature extreme value, the temperature gradient, and the number of thermal cycles; calculate the plastic strain amplitude and the main vibration frequency; obtain the crack length and the defect area ratio; calculate the heat source density based on the Joule heat formula; construct the boundary conditions of the three-dimensional heat conduction equation in combination with the infrared temperature field data; jointly input the temperature gradient and the plastic strain amplitude into the improved Coffin-Manson equation to calculate the thermo-mechanical fatigue damage factor; and process the main vibration frequency and crack length data by using the improved Paris law.

[0076] A PINN network construction module 300, configured to construct a three-constraint PINN network architecture, including a temperature field prediction channel, a thermal fatigue cumulative amount output channel, and a crack propagation data processing channel; and implement multi-physical field coupling by using a thermal-mechanical coupling loss function.

[0077] The network training module 400 is used to set network training strategies, including using acoustic microscope data as the strong supervision signal for the crack propagation branch and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balancing two types of physical loss functions, i.e., the residual of the thermal damage equation and the residual of the crack propagation equation, through an adaptive weighting algorithm.

[0078] The prediction and alarm module 500 is used to input real-time monitoring data into the trained PINN network; synchronously output three groups of prediction results, namely the percentage of the remaining life of the solder joint, the warning value of the critical crack size, and the coordinates of the high-temperature dangerous area; trigger a hierarchical alarm when any output value exceeds the preset threshold.

[0079] In some embodiments, the data acquisition module 100 includes: a high-frequency current sensor for monitoring the transient current amplitude of the IGBT switch; an infrared thermal imager for obtaining the surface temperature field distribution of the solder joint; an embedded thermocouple implanted in the DBC substrate for obtaining the temperature near the solder joint; a micro-strain gauge mounted on the substrate adjacent to the solder joint for detecting the plastic strain of the solder joint; a laser Doppler vibrometer for measuring the main vibration frequency of the solder joint; an acoustic emission sensor for obtaining the acoustic signal of crack propagation; and a scanning acoustic microscope for layer-by-layer scanning of internal defects.

[0080] In some embodiments, the PINN network construction module 300 includes a heat conduction constraint branch, a heat damage constraint branch, and a crack propagation constraint branch; the heat conduction constraint branch incorporates a three-dimensional heat conduction differential operator; the heat damage constraint branch embeds an improved Coffin-Manson equation; the crack propagation constraint branch solidifies an improved Paris law calculation flow; the heat conduction constraint branch, the heat damage constraint branch, and the crack propagation constraint branch achieve two-way gradient exchange through a shared hidden layer and realize multi-physical field coupling prediction based on a physics-guided loss function.

[0081] Although specific embodiments are described herein, those of ordinary skill in the art will recognize that many other modifications or alternative embodiments are also within the scope of the present disclosure. For example, any one of the functions and / or processing capabilities described in connection with a particular device or component can be performed by any other device or component. Additionally, although various illustrative specific implementations and architectures have been described in accordance with embodiments of the present disclosure, those of ordinary skill in the art will recognize that many other modifications to the illustrative specific implementations and architectures described herein are also within the scope of the present disclosure.

[0082] Those of ordinary skill in the art will understand that all or some of the steps and systems disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes but is not limited to RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0083] The above is a specific description of the preferred embodiment of the present application. However, the present application is not limited to the above embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present application, and these equivalent deformations or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A method for detecting the failure and predicting the life of solder joints of a power module based on PINN, characterized in that, The following steps are involved: S100, monitor the transient current amplitude of the IGBT switch based on a high-frequency current sensor, obtain the surface temperature field distribution of the solder joint based on an infrared thermal imager, and obtain the proximal temperature of the solder joint based on an embedded thermocouple implanted in the DBC substrate to obtain the current amplitude, temperature extreme value, temperature gradient and number of thermal cycles; detect the plastic strain of the solder joint based on a micro strain gauge mounted on the substrate adjacent to the solder joint, and measure the main vibration frequency of the solder joint based on a laser Doppler vibrometer to obtain the plastic strain amplitude and vibration main frequency; obtain the crack extension sound signal based on an acoustic emission sensor, and scan the internal defects in layers based on a scanning acoustic microscope to obtain the crack length and defect area ratio; S200, based on the current amplitude collected in step S100, the heat source density is calculated by the Joule heat formula, and the boundary conditions of the three-dimensional heat conduction equation are constructed in combination with the infrared temperature field data; the temperature gradient and the plastic strain amplitude are input into the improved Coffin-Manson equation to calculate the thermomechanical fatigue damage factor; The improved Paris law is used to process the vibration main frequency and crack length data to obtain the crack growth rate; the heat source density, thermomechanical fatigue damage factor and crack growth rate are used as physical constraint inputs of the PINN network; S300, construct a three-constraint PINN network architecture: the first channel constrains the temperature field prediction through the residual term of the heat conduction equation; the second channel takes the strain amplitude and the number of thermal cycles as input, and outputs the thermal fatigue accumulation through the Arrhenius damage equation; The third channel processes the crack growth data; The thermal-mechanical coupling loss function is used to achieve multi-physics coupling; S400, set the network training strategy: use acoustic microscope data as a strong supervision signal for the crack extension branch, and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balance two types of physical loss functions through an adaptive weighted algorithm: thermal damage equation residual and crack extension equation residual; S500, input the real-time monitoring data into the trained PINN network, and simultaneously output three sets of prediction results: the percentage of remaining life of the solder joint, the critical crack size warning value, and the coordinates of the high-temperature danger zone; when any output value exceeds the preset threshold, a graded alarm is triggered.

2. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 1, wherein The heat conduction equation is: ； where I is the current, ▽T is the temperature gradient, ρ is the density, is the specific heat capacity, is the thermal conductivity, and Qj is the current Joule heat term; The improved Coffin-Manson equation is: ； where k is the temperature influence coefficient, ΔT is the temperature change range, Tavg is the average temperature, Q is the activation energy, R is the gas constant, is the number of failure cycles, is the plastic strain amplitude, C is the thermo-mechanical coupling damage base coefficient, and n is the temperature gradient sensitivity index; The improved Paris law is as follows: ; where, da / dN is the crack growth rate, N is the number of load cycles, a is the crack length, is the range of mode I stress intensity factor, is the range of mode II stress intensity factor, C(T) is the temperature-dependent material constant, m(T) is the temperature-dependent Paris exponent, η is the mixed-mode weight factor, and T is the operating temperature; The Arrhenius type damage equation is: ； Among them, A is the material damage constant, Ta is the absolute temperature of the solder joint, and m is the strain sensitivity index.

3. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 2, wherein, The thermal-mechanical coupling loss function is: ； Among them, λ1, λ2, and λ3 are adaptive weight coefficients, is the residual term of the heat conduction equation, D is the measured thermal damage factor, is the predicted thermal damage factor, a is the measured crack length, is the predicted crack length.

4. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 1, characterized in that, Step S100 also includes: S110, analyzing the size and solder joint layout of the power module, and determining the critical solder joint area and potential failure points; S120. According to the distribution of key solder joint areas and potential failure points, set the layout positions of high-frequency current sensors, infrared thermal imagers, embedded thermocouples, micro strain gauges, laser Doppler vibrometers, acoustic emission sensors, and scanning acoustic microscopes; S130. Determine the number of various sensors according to the number of welding points and monitoring requirements.

5. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 1, characterized in that, Step S500 also includes: S510, screening and denoising the collected raw data, and removing outliers and invalid data; S520. Scale the cleaned data according to a certain ratio so that it falls within a specific numerical range; S530. Extract key features from the original data according to the physical mechanism of solder joint failure and the requirements of the PINN network.

6. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 1, characterized in that Step S400 further includes: S410. Set the number of iterations according to the complexity of the PINN network and the scale of the training data; S420. During the training process, dynamically adjust the learning rate according to the convergence of the network loss function. First, adopt the first learning rate to accelerate convergence, and then adopt the second learning rate to finely adjust the network weights; the first learning rate is greater than the second learning rate.

7. The method for detecting solder joint failure and predicting the life of a power module based on PINN according to claim 1, wherein Step S500 further includes: S540. Set different alarm levels according to the prediction results of the solder joint remaining life percentage, the critical crack size warning value, and the coordinates of the high-temperature dangerous area, including: emergency shutdown level, early warning maintenance level, and monitoring attention level; S550. Immediately stop the machine for inspection when the alarm level is the emergency shutdown level; arrange preventive maintenance when the alarm level is the early warning maintenance level; and increase the monitoring frequency when the alarm level is the monitoring attention level.

8. A PINN-based power module solder joint failure detection and life prediction system for performing the method according to any one of claims 1 to 7, characterized in that, It includes: A data acquisition module, which is used to acquire the surface temperature field distribution of the solder joints of the power module, the temperature near the solder joints, the plastic strain of the solder joints, the main vibration frequency of the solder joints, the crack propagation sound signal, and internal defects; A data processing module, which is used to calculate the current amplitude, the temperature extreme value, the temperature gradient, and the number of thermal cycles; Calculate the plastic strain amplitude and the main vibration frequency; obtain the crack length and the defect area ratio; Calculate the heat source density based on the Joule heat formula; Combine the infrared temperature field data to construct the boundary conditions of the three-dimensional heat conduction equation; jointly input the temperature gradient and the plastic strain amplitude into the improved Coffin-Manson equation to calculate the thermo-mechanical fatigue damage factor; use the improved Paris law to process the main vibration frequency and crack length data; A PINN network construction module, which is used to construct a three-constraint PINN network architecture, including a temperature field prediction channel, a thermal fatigue cumulative amount output channel, and a crack propagation data processing channel; Adopt a thermal-mechanical coupling loss function to achieve multi-physical field coupling; A network training module, which is used to set the network training strategy, including using acoustic microscope data as the strong supervision signal for the crack propagation branch and the number of thermal cycles as the iteration termination condition for the fatigue damage branch; balance the two types of physical loss functions through an adaptive weighting algorithm: the residual of the thermal damage equation and the residual of the crack propagation equation; A prediction and alarm module, which is used to input the real-time monitoring data into the trained PINN network; synchronously output three groups of prediction results: the solder joint remaining life percentage, the critical crack size warning value, and the coordinates of the high-temperature dangerous area; trigger a hierarchical alarm when any output value exceeds the preset threshold.

9. The PINN-based power module solder joint failure detection and life prediction system according to claim 8, wherein The data acquisition module includes: A high-frequency current sensor, which is used to monitor the IGBT switch transient current amplitude; An infrared thermal imager, which is used to obtain the surface temperature field distribution of the solder joints; An embedded thermocouple, implanted in the DBC substrate, which is used to obtain the temperature near the solder joints; A micro strain gauge, mounted on the substrate adjacent to the solder joint, which is used to detect the plastic strain of the solder joint; A laser Doppler vibrometer, which is used to measure the main vibration frequency of the solder joint; An acoustic emission sensor, used to acquire acoustic signals of crack propagation; A scanning acoustic microscope, used to perform layer-by-layer scanning of internal defects.

10. The PINN-based power module solder joint failure detection and life prediction system according to claim 8, wherein, The PINN network construction module includes a heat conduction constraint branch, a thermal damage constraint branch, and a crack propagation constraint branch; The heat conduction constraint branch incorporates a three-dimensional heat conduction differential operator; The thermal damage constraint branch embeds an improved Coffin-Manson equation; The crack propagation constraint branch solidifies an improved Paris law calculation flow; The heat conduction constraint branch, the thermal damage constraint branch, and the crack propagation constraint branch achieve bidirectional gradient exchange through a shared hidden layer, and realize multi-physical field coupling prediction based on a physics-guided loss function.

Images