An IGBT module multi-modal collaborative defect detection method and device

Abstract

The application discloses an IGBT module multi-modal collaborative defect detection method, comprising: in response to a synchronous trigger instruction, controlling a structured light optical detection unit and an ultrasonic detection unit to detect an IGBT module respectively, and synchronously acquiring surface optical data and internal ultrasonic echo data; based on hardware synchronization timing, establishing a spatial mapping relationship between the surface optical data and the internal ultrasonic echo data; based on signal-to-noise ratio, adaptively weighting and fusing optical features of the surface optical data and ultrasonic features of the internal ultrasonic echo data, and generating fusion feature data; analyzing a design file of the IGBT module to generate a position mask, correcting the fusion feature data by using the position mask, and identifying defects based on the corrected fusion feature data. The application can comprehensively improve the detection accuracy, detection rate and anti-interference ability of surface and internal defects of the IGBT module.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a method and apparatus for multimodal collaborative defect detection of IGBT modules. Background Technology

[0002] As a core component of power electronic equipment, IGBT modules are prone to surface bonding damage, scratches, poor soldering, and internal defects such as cracks, porosity, and delamination in their bonding wires, packaging layers, and internal chips, directly affecting device reliability and lifespan. Existing IGBT module defect detection methods mostly employ single-modal techniques, which have significant limitations: single optical detection can only identify surface defects and cannot penetrate materials to obtain information on hidden internal defects; single ultrasonic detection lacks sensitivity to shallow and minute surface defects and is easily affected by material interface interference. Furthermore, existing multimodal detection schemes often lack hardware-level time synchronization, leading to data drift and spatial coordinate misalignment, resulting in low fusion accuracy; and they do not incorporate IGBT module design files for false defect filtering, easily misclassifying design gaps and process references as defects, resulting in poor detection robustness; traditional data fusion only performs simple weighting and does not achieve feature-level deep fusion, leading to insufficient accuracy in defect localization and classification. In summary, existing technologies cannot achieve full-dimensional, high-precision, and low-false-detection defect detection of IGBT modules' "surface and interior," making it difficult to meet the detection needs of industrial mass production scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide a method and apparatus for multimodal collaborative defect detection of IGBT modules. This invention can comprehensively improve the detection accuracy, detection rate, and anti-interference capability of surface and internal defects in IGBT modules.

[0004] The technical solution provided by this invention is as follows: A multimodal collaborative defect detection method for IGBT modules, comprising:

[0005] In response to the synchronous trigger command, the structured light optical detection unit and the ultrasonic detection unit are controlled to detect the IGBT module respectively, and the surface optical data and internal ultrasonic echo data are acquired synchronously.

[0006] Based on hardware synchronization timing, a spatial mapping relationship between the surface optical data and the internal ultrasonic echo data is established.

[0007] Based on the signal-to-noise ratio, the optical features of the surface optical data and the ultrasonic features of the internal ultrasonic echo data are adaptively weighted and fused to generate fused feature data.

[0008] The design file of the IGBT module is parsed to generate a location mask. The location mask is used to correct the fused feature data, and defects are identified based on the corrected fused feature data.

[0009] The generation of the synchronization trigger instruction in the above method satisfies timing constraints:

[0010] in, This is the trigger time of the structured light optical detection unit. The trigger time of the ultrasonic detection unit. As the reference time, For hardware transmission delay, and .

[0011] The aforementioned method, wherein establishing the spatial mapping relationship includes:

[0012] The ultrasonic scanning coordinates of the internal ultrasonic echo data are mapped to the optical image coordinates of the surface optical data using a coordinate mapping algorithm. The mapping formula is as follows:

[0013] ;

[0014] in, Optical image coordinates, For ultrasound scan coordinates, This is the scaling factor. This is the offset.

[0015] The aforementioned method uses the following formula for adaptive weighted fusion:

[0016] ;

[0017] in, To fuse feature signals, It is an optical characteristic signal. It is a characteristic signal of ultrasound. and The weights are optical signal weights and ultrasonic signal weights, respectively, and the weight values ​​are adaptively determined based on the signal-to-noise ratio of each modal signal.

[0018] The aforementioned method, wherein the defect identification includes:

[0019] Based on the corrected fused feature data, the comprehensive defect probability is calculated using a decision fusion algorithm, as shown in the formula:

[0020] ;

[0021] in, To comprehensively assess the probability of defects, Confidence level for optical defects. Confidence level of ultrasonic defect. and Let be the decision weight coefficient, and .

[0022] The aforementioned method, which parses the design file of the IGBT module to generate a location mask, includes:

[0023] The design file is parsed to obtain the legal device region of the IGBT module;

[0024] A location mask matrix is ​​generated, wherein the value of the location mask matrix within the detection area is a first preset value, and the value within the non-detection area is a second preset value.

[0025] The aforementioned method, wherein the correction of the fused feature data using the location mask employs the following formula:

[0026] ;

[0027] in, The corrected confidence level for the defect. The fused feature data before correction. The position mask matrix in coordinates The value at that location.

[0028] The aforementioned method, after the synchronous acquisition of surface optical data, further includes:

[0029] The optical images acquired from multiple fields of view are weighted and fused to eliminate single-view occlusion areas, and high-pass filtering is used to remove ambient light interference.

[0030] The aforementioned method, after synchronously acquiring the internal ultrasound echo data, further includes:

[0031] The internal ultrasound echo data is adaptively filtered using the following formula:

[0032] ;

[0033] in, This is the filtered signal. The original signal, The current ultrasound signal-to-noise ratio, The reference signal-to-noise ratio is 1.

[0034] A multimodal collaborative defect detection device for IGBT modules, comprising:

[0035] The synchronous triggering system is configured to generate synchronous triggering commands.

[0036] The structured light optical detection unit is configured to detect the IGBT module and acquire surface optical data in response to the synchronous triggering command;

[0037] An ultrasonic detection unit is configured to detect the IGBT module and acquire internal ultrasonic echo data in response to the synchronous trigger command.

[0038] The data acquisition unit is configured to transmit the surface optical data and the internal ultrasonic echo data;

[0039] The processing unit is configured to execute the method as described above.

[0040] Compared with existing technologies, this invention employs a signal detection and synchronous triggering system to achieve precise alignment of the dual-modal data time axis, completely solving the problem of dynamic detection data drift. Structured light optics covers minute surface defects, while ultrasonic waves cover hidden internal defects, enabling full-level defect detection from the surface to the interior of the IGBT module. Based on an adaptive signal-to-noise ratio weighted algorithm, the detection rate is significantly improved in complex industrial environments. Standardized formulas enable quantitative calculation of defect morphology, depth, and location, achieving micron-level detection accuracy. Multi-optical-mechanical collaborative imaging solves the problems of optical path obstruction and redundant information, adapting to IGBT module mass production testing scenarios. An ultrasonic adaptive filtering algorithm filters out interface interference from the packaging layer, and multi-optical-mechanical redundancy elimination eliminates optical path overlap errors, significantly improving detection stability. PIM position masking filters out design-related false defects, and adaptive thresholds adapt to different defect types, reducing the false positive rate by over 90%. A decision-level fusion algorithm integrates dual-modal confidence, improving defect classification accuracy by 30% compared to single-modal methods. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the overall structure of the IGBT module multimodal collaborative defect detection device according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram illustrating the principle of multi-optical-mechanical collaborative structured light imaging according to an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the multimodal collaborative defect detection method for IGBT modules according to an embodiment of the present invention.

[0044] Figure Labels

[0045] 1. Single-mode laser; 2. Liquid crystal diffuser; 3. Half-wave plate; 4. Beam expander system; 5. Mirror; 6. Quarter-wave plate; 7. First mirror; 8. Second mirror; 9. Area array camera; 10. IGBT module; 11-14. Multi-field projection optical engine; 15. Ultrasonic probe; 16-17. Ring light source; 18. Light source controller; 19. Synchronous triggering system; 20. Data acquisition unit; 21. Processing unit. Detailed Implementation

[0046] The present invention will be further described below with reference to embodiments, but these embodiments are not intended to limit the scope of the invention.

[0047] Example 1: As Figure 3 As shown, this embodiment provides a multimodal collaborative defect detection method for IGBT modules. This method aims to address the limitations of single-modal detection, low fusion accuracy due to asynchronous multimodal data, and false defect misjudgment caused by a lack of prior design knowledge in existing technologies. Through collaborative operations of hardware-level synchronous triggering, spatiotemporal registration, adaptive fusion, and false defect filtering, it achieves high-precision defect detection with low false positives.

[0048] Specifically, the method includes the following steps:

[0049] In step S100, in response to the synchronous trigger command, the structured light optical detection unit and the ultrasonic detection unit are controlled to detect the IGBT module 10 respectively, and the surface optical data and internal ultrasonic echo data are acquired synchronously.

[0050] In this step, the synchronization trigger command is generated and sent by the synchronization trigger system 19. Its core function is to ensure the precise alignment of the structured light optical detection unit and the ultrasonic detection unit on the time axis. Specifically, the generation of the synchronization trigger command satisfies timing constraints:

[0051] ,in, This is the trigger time for the structured light optical detection unit. This is the trigger time of the ultrasonic detection unit. As the reference time, For hardware transmission delay, and .

[0052] Combination Figure 1 As shown, the synchronous triggering system 19 serves as a unified time reference source, at the reference reference time. Generate the original trigger signal by accurately measuring and compensating for hardware transmission delay. This makes the triggering time of the optical detection unit... Triggering time of the ultrasonic detection unit Strictly equal. By controlling the time interval to within 500 nanoseconds, it can be ensured that the deviation between optical and ultrasonic data in the time dimension is limited to the micrometer-level spatial displacement error range under high-speed detection cycles, thereby completely eliminating the data drift problem in dynamic detection. It should be understood that although this embodiment preferably uses a hardware triggering method to obtain the highest synchronization accuracy, in other embodiments, software triggering or a combination of software and hardware triggering methods can also be used, as long as the above timing constraints can be met.

[0053] The surface optical data acquired by the area array camera 9 satisfies the following acquisition formula:

[0054] ;

[0055] in, The feature signal corresponding to the surface optical data of the IGBT module 10 acquired by the area array camera 9. The intensity of light reflected from the surface is determined by the intensity of the light source. With surface reflectivity Decide: , For spatial resolution, The sampling interval is defined by this formula. This formula establishes a quantitative relationship between the optical signal and the surface reflection characteristics of the IGBT module 10, providing a data foundation for subsequent feature extraction and fusion.

[0056] After simultaneously acquiring surface optical data, preprocessing is required to improve data quality. Specifically, weighted fusion is performed on the optical images acquired from multiple fields of view to eliminate single-view occlusion areas, and high-pass filtering is used to remove ambient light interference. Figure 2 As shown, since the surface of the IGBT module 10 typically has a complex three-dimensional structure (such as bonding wires, heat sinks, etc.), optical imaging from a single viewing angle is prone to producing shadows or occlusions. This embodiment employs a multi-field projection optical engine to project structured light from different angles and acquire images. Weighted fusion processing is performed on overlapping areas, and for occluded areas under a single viewing angle, effective image information from other viewing angles is used to fill in the gaps. Furthermore, redundant pixels in the multi-field overlapping areas are normalized to remove duplicate features.

[0057] ;

[0058] ;

[0059] Where median is the median filter. Redundancy elimination factor (e.g.) This process filters out redundant information in overlapping optical paths. Then, a high-pass filtering algorithm is used to process the fused image. Since ambient light typically represents a low-frequency background signal, while structured light stripes and defect features represent high-frequency signals, high-pass filtering can effectively remove ambient light interference. Furthermore, the surface height distribution of the IGBT module 10 is reconstructed using a phase-shifting method.

[0060] ;

[0061] in, For surface height distribution, The pitch of the projection grating. , , , The light intensity value after fringe phase shift is used to obtain high-precision three-dimensional surface topography information. Further, based on the reconstructed three-dimensional surface topography information, edge recognition and surface quality assessment are performed on the optical image to extract two-dimensional features of surface defects. Edge recognition detects gradient abrupt change regions on the surface topography of the IGBT module 10, locating key structural contours such as bond line edges and chip boundaries. Surface quality assessment comprehensively evaluates the overall quality status of the IGBT module 10 surface by integrating indicators such as surface roughness and texture uniformity. Through the above processing, the system can extract comprehensive surface defect features from optical data, including edge contours, surface quality indicators, and three-dimensional topography information, providing high-quality optical feature input for subsequent multimodal fusion.

[0062] After synchronously acquiring the internal ultrasound echo data, preprocessing of the ultrasound data is also required. Specifically, adaptive filtering is applied to the internal ultrasound echo data, and the filtering formula is as follows:

[0063] ;

[0064] in, This is the filtered signal. The original signal, The current ultrasound signal-to-noise ratio, The reference signal-to-noise ratio is used. When ultrasound penetrates the more than 10 layers of an IGBT module, it is easily interfered with by factors such as material interface reflection and grain scattering. This embodiment introduces an adaptive filtering mechanism, which adjusts the signal-to-noise ratio based on the current signal-to-noise ratio. Far below the benchmark signal-to-noise ratio When the exponential term approaches 1 and the filter coefficient approaches 0, the system significantly attenuates the signal, effectively suppressing noise in the low signal-to-noise ratio (SNR) frequency band; conversely, when the SNR is close to or higher than the reference value, the signal is preserved. It should be understood that the ultrasonic echo signal within the IGBT module material satisfies the wave equation:

[0065] ;

[0066] in, For sound wave displacement, The speed at which sound waves propagate in the IGBT material. For the duration of transmission.

[0067] Defect depth can be calculated based on ultrasonic echo time delay:

[0068] ;

[0069] in, For defect depth, The time difference between ultrasonic emission and echo reception is used to accurately locate internal defects.

[0070] In addition, the amplitude of the echo signal acquired by the ultrasonic probe 15 satisfies:

[0071] ;

[0072] in, It is an ultrasonic echo signal. The formula establishes a quantitative relationship between the ultrasonic echo signal and the echo amplitude, where 15 represents the echo signal amplitude of the ultrasonic probe.

[0073] After adaptive filtering and denoising, the system further performs time-frequency analysis and echo feature extraction on the ultrasonic signal to obtain the depth and amplitude characteristics of the internal defects. Time-frequency analysis transforms the ultrasonic echo signal to the time-frequency domain, revealing the distribution characteristics of the defect echo in the time and frequency dimensions, and distinguishing the spectral response modes of different types of defects. Echo feature extraction extracts key parameters such as echo amplitude and time delay from the filtered signal to comprehensively characterize the size, location, and severity of the internal defects.

[0074] Step S200: Based on hardware synchronization timing, establish a spatial mapping relationship between surface optical data and internal ultrasonic echo data.

[0075] After acquiring two sets of time-synchronized data, the data coordinate systems of the optical and ultrasonic detection units typically differ in physical location, field of view, and resolution, resulting in a direct overlap. Therefore, this step requires establishing a spatial mapping relationship between the two. Specifically, a coordinate mapping algorithm is used to map the ultrasonic scanning coordinates of the internal ultrasonic echo data to the optical image coordinates of the surface optical data. The mapping formula is as follows:

[0076] ;

[0077] in, Optical image coordinates, For ultrasound scan coordinates, This is the scaling factor. Offset. Scaling factor. Offset used to adjust the resolution difference between two coordinate systems This is used to compensate for the fixed deviation of the two detection units in their physical installation positions. In practical applications, and Parameters are typically determined beforehand through the system calibration process. Using this linear mapping formula, the system can accurately project the coordinates of internal defects detected by ultrasound onto the optical image coordinate system, achieving pixel-level alignment between surface topography information and internal structural information. Furthermore, in more general spatial registration and fusion, a three-dimensional coordinate fusion method can also be employed.

[0078] ;

[0079] in, For optical image pixel coordinates, The coordinates and defect depth for ultrasonic testing. This is a spatial coordinate transformation function that achieves precise alignment between surface coordinates and internal depth coordinates.

[0080] Step S300: Based on the signal-to-noise ratio, adaptive weighted fusion of the optical features of the surface optical data and the ultrasonic features of the internal ultrasonic echo data is performed to generate fused feature data.

[0081] This step is crucial for improving detection accuracy. The adaptive weighted fusion uses the following formula:

[0082] ;

[0083] in, To fuse feature signals, It is an optical characteristic signal. It is a characteristic signal of ultrasound. and These are the weights for the optical signal and the ultrasonic signal, respectively, and the weight values ​​are adaptively determined based on the signal-to-noise ratio of each modal signal. Specifically, the weights are calculated as follows:

[0084] ;

[0085] ;

[0086] in, Let be the signal-to-noise ratio of the i-th mode signal. For signal power, This represents noise power. This means that modes with higher signal-to-noise ratios (SNR) contribute more to the fusion features. When the SNR of an optical signal is high, it is given a higher weight; conversely, when the SNR of an ultrasonic signal is high, the weight of the ultrasonic feature is increased. This adaptive mechanism can effectively suppress interference from low-quality signals and highlight the features of high-quality signals, thereby generating more accurate and robust fusion feature data. It should be understood that although this embodiment preferably uses the above-described adaptive SNR weighting method, other dynamic weight determination methods based on signal quality can also be used in other embodiments, as long as they can automatically adjust the fusion strategy according to the actual data quality.

[0087] Through the aforementioned spatial registration and adaptive weighted fusion, the system deeply fuses surface optical features and internal ultrasonic features at the pixel level to establish a three-dimensional defect distribution model for IGBT module 10. This model uses the optical image coordinates as the reference plane, mapping the internal defect depth information obtained from ultrasonic detection to the corresponding surface coordinates, forming a complete three-dimensional defect distribution map covering both surface morphology and internal structure, providing comprehensive data support for subsequent defect diagnosis.

[0088] Step S400: parse the design file of IGBT module 10 to generate a position mask, use the position mask to correct the fused feature data, and identify defects based on the corrected fused feature data.

[0089] The design file of IGBT module 10 typically contains information such as the device's legal areas, design vacancies, and process reference holes. In actual inspection, these non-inspection areas are often misjudged as defects, leading to a higher false alarm rate. Specifically, the design file is parsed to obtain the legal areas of the IGBT module 10; a position mask matrix is ​​generated, where the values ​​in the detection areas are a first preset value, and the values ​​in the non-inspection areas are a second preset value. The design file is usually a CAD drawing or a GERBER file. The system uses image processing algorithms or a document parsing interface to read the layer information in the design file, identifying the valid areas requiring defect detection (i.e., legal device areas), such as the chip surface and bonding wire solder joints; and simultaneously identifying areas that do not require detection or are allowed to have gaps (i.e., non-inspection areas), such as positioning holes, heat dissipation holes, and silkscreen marking areas. Based on the above identification results, the system generates a position mask matrix with the same resolution as the optical image.

[0090] Furthermore, the location mask is used to correct the fused feature data using the following formula:

[0091] ;

[0092] in, The corrected confidence level for the defect. The fused feature data before correction. For the position mask matrix in coordinates The value at that location.

[0093] In a preferred embodiment of this example, the first preset value is set to 1, and the second preset value is set to 0 or a value between 0 and 1 (e.g., 0.7). When the coordinates are within the detection area, The fused feature data is fully preserved; when the coordinates are located outside the detection area, This could be an attenuation coefficient, where the fused feature data at this location is forcibly cleared to zero or attenuated, effectively eliminating false defect signals caused by design gaps, process reference holes, etc. It should be understood that setting the second preset value to a value between 0 and 1 allows for a certain margin of error in the edge region, avoiding edge misjudgments caused by alignment errors between the mask and the actual object.

[0094] After correcting the fused feature data using a location mask, the system identifies defects based on the corrected fused feature data. Specifically, based on the corrected fused feature data, a decision fusion algorithm is used to calculate the comprehensive defect probability, as shown in the formula:

[0095] ;

[0096] in, To comprehensively assess the probability of defects, Confidence level for optical defects. Confidence level of ultrasonic defect. and Let be the decision weight coefficient, and Optical inspection is highly sensitive to surface defects (such as scratches, stains, and bond line damage), while ultrasonic inspection has a unique advantage in penetrating and detecting internal defects (such as porosity, delamination, and cold solder joints). This embodiment employs a differentiated decision-making strategy: if the initial feature indicates a surface defect, the optical decision weight coefficient is adaptively increased. Reduce the ultrasonic decision weight coefficient (For example , Conversely, if the initial characteristics indicate internal defect features, the ultrasonic decision weight coefficient is adaptively increased. Reduce optical decision weight coefficient (For example , Through this defect type-based weight adaptive adjustment mechanism, the system can adopt the optimal judgment strategy for different types of defects.

[0097] Furthermore, an adaptive threshold algorithm can be executed during defect diagnosis:

[0098] ;

[0099] in, The current defect detection threshold is set as follows: This is a general benchmark threshold. For defect type coefficients (e.g., surface defects) Internal defects This enables accurate identification of different defect types, avoiding missed detections or misjudgments.

[0100] Through the synergistic effect of steps S100 to S400, this embodiment achieves a closed-loop process from data acquisition, spatiotemporal alignment, feature fusion to false defect filtering, effectively solving technical challenges such as multimodal data drift, low fusion accuracy, and false defect misjudgment, significantly improving the accuracy and reliability of IGBT module defect detection. Finally, the system automatically outputs a detection report and a defect distribution visualization. The detection report includes quantitative information such as defect type, location, size, and confidence level. The defect distribution visualization, based on a three-dimensional defect distribution model, intuitively presents the defect distribution status on the surface and inside of the IGBT module 10, providing a basis for quality judgment and process optimization.

[0101] Example 2: As Figure 1 As shown, this embodiment provides a multimodal collaborative defect detection device for IGBT modules. This device serves as the hardware carrier for the detection method described in Embodiment 1 above. Through the collaborative operation of various functional units, it achieves high-precision, low-false-judgment detection of surface and internal defects in the IGBT module 10. Specifically, the device includes: a synchronous triggering system 19, a structured light optical detection unit, an ultrasonic detection unit, a data acquisition unit (i.e., a data acquisition device) 20, and a processing unit 21 (i.e., a host computer).

[0102] The synchronous triggering system 19 is configured to generate synchronous trigger commands. Specifically, the synchronous triggering system 19 is the time reference source and control core of the entire device. In terms of hardware implementation, it can use a high-precision FPGA or ARM processor as the main control chip to generate nanosecond-level trigger pulses through hardware circuitry. This system is connected to the single-mode laser 1, the light source controller 18, and the data acquisition unit 20 via high-frequency shielded cables to ensure the stability and low latency of the trigger signal transmission, with a timing deviation ≤1μs. By sending synchronous trigger commands, the synchronous triggering system 19 can ensure strict alignment of optical imaging and ultrasonic scanning on the time axis, satisfying the requirements of Embodiment 1 above. The timing constraints are used to solve the data drift problem in dynamic detection.

[0103] The structured light optical detection unit is configured to detect the IGBT module 10 and acquire surface optical data in response to a synchronous trigger command. Combined with... Figure 1 and Figure 2As shown, the structured light optical detection unit includes a single-mode laser 1, a liquid crystal diffuser 2, a half-wave plate 3, a beam expander 4, a reflector 5, a quarter-wave plate 6, a first reflector 7, a second reflector 8, multi-field projection optical engines 11-14, an area array camera 9, a ring light source 16-17, and a light source controller 18. The specific optical propagation path is as follows: the laser beam emitted by the single-mode laser 1 first undergoes homogenization processing through the liquid crystal diffuser 2, and then its polarization state is adjusted by the half-wave plate 3 to match subsequent optical elements; subsequently, the beam enters the beam expander 4 for beam expansion and shaping, forming a uniform line or area spot; after beam expansion, the beam is redirected by the reflector 5, its polarization state is adjusted by the quarter-wave plate 6, and then its optical path is deflected and its angle fine-tuned by the slightly tiltable first reflector 7 and second reflector 8, finally projected onto the multi-field projection optical engines 11-14. The multi-field projection optical engines 11-14 are symmetrically distributed around the IGBT module 10 and are used to project structured light stripes onto the workpiece surface. Ring lights 16-17 are positioned on both sides of workpiece 10, and their on / off state and brightness are controlled by light source controller 18 to provide auxiliary illumination. Area array camera 9 is located directly above workpiece 10 and is used to receive structured light reflection images modulated by the workpiece surface, thereby acquiring optical data containing surface morphology information. It should be understood that although... Figure 1 The image shows four multi-field projection optical engines 11-14, but in other embodiments, the number and arrangement angle of the optical engines can be adjusted according to the specific structural features of the IGBT module, as long as single-view occlusion can be eliminated and the area to be detected can be covered.

[0104] The ultrasonic testing unit is configured to detect the IGBT module 10 and acquire internal ultrasonic echo data in response to a synchronous trigger command. Specifically, the ultrasonic testing unit mainly includes an ultrasonic probe 15 and its driving circuit. The ultrasonic probe 15 is located on one side or the bottom of the IGBT module 10, the specific location depending on the testing process requirements. Under the control of the synchronous trigger command, the ultrasonic probe 15 emits high-frequency ultrasonic pulses into the IGBT module 10 and receives echo signals reflected back from internal interfaces or defects. Due to the strong penetrating power and sensitivity to internal defects of ultrasound, this unit can effectively detect hidden defects such as pores, delamination, and cracks inside the module, compensating for the inability of optical detection to penetrate materials.

[0105] The data acquisition unit 20 is configured to transmit surface optical data and internal ultrasonic echo data. As a data transmission intermediary, the data acquisition unit 20 is connected to the area array camera 9, the ultrasonic probe 15, and the processing unit 21. For optical data, the data acquisition unit 20 typically uses a high-speed Camera Link or GigE Vision interface to transmit high-resolution image data in real time; for ultrasonic echo data, a high-sampling-rate analog-to-digital converter module is used to convert the analog echo signal into a digital signal before uploading. The data acquisition unit 20 ensures the integrity and synchronization of multimodal data during transmission, providing a reliable data source for subsequent fusion processing.

[0106] The processing unit 21 is configured to execute the method described in Embodiment 1 above. The processing unit 21 is typically a high-performance industrial computer or server, integrating a data fusion module and a defect analysis module. The processing unit 21 connects to the data acquisition unit 20, receiving the uploaded optical images and ultrasonic echo signals. First, it performs hardware synchronization timing verification and spatial coordinate mapping to establish the correspondence between surface data and internal data. Then, it performs adaptive signal-to-noise ratio weighted fusion to generate fused feature data. Finally, it parses the IGBT module design file to generate a position mask, corrects the fused feature data, and identifies defects. By running a preset algorithm program, the processing unit 21 transforms abstract detection logic into specific hardware control instructions and data processing results, achieving full automation from data acquisition to defect determination.

[0107] Through the above hardware architecture configuration, this embodiment constructs a complete IGBT module multimodal collaborative defect detection device. The synchronous triggering system 19 solves the time synchronization problem of multimodal data, the structured light optical detection unit and the ultrasonic detection unit achieve full-dimensional coverage of the surface and interior, and the data acquisition unit 20 and the processing unit 21 ensure efficient data flow and intelligent processing. These units cooperate with each other to support the technical effects of the aforementioned method embodiment, significantly improving the accuracy and robustness of IGBT module defect detection.

[0108] Example 3: To verify the practical application effect of the IGBT module multimodal collaborative defect detection method and device provided by the present invention, this example selects an FF450R12ME4 type IGBT module as the detection object, and describes the detection process in detail with specific parameters. It should be understood that the specific parameters in this example are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0109] During the system preparation and calibration phase, the device is first initialized and configured. The hardware transmission delay of the synchronous trigger system 19 is then set. The value is 300 ns, which satisfies... Timing constraints are implemented to ensure synchronized triggering of the structured light optical detection unit and the ultrasonic detection unit on a microsecond timescale. The exposure time of the area array camera 9 is set to 100 μs to adapt to the production line inspection cycle; the frequency of the ultrasonic probe 15 is set to 15 MHz to balance penetration depth and resolution. The grating spacing projected by the structured light optical detection unit... It is set to 0.2mm for high-precision surface topography reconstruction.

[0110] During the multimodal data acquisition phase, the synchronous triggering system 19 sends a trigger command, and the structured light optical detection unit and the ultrasonic detection unit start simultaneously. Multi-field projection optical engines 11-14 project structured light stripes onto the surface of the IGBT module 10, and the area array camera 9 simultaneously acquires surface optical data. At the same time, the ultrasonic probe 15 scans the interior of the module at a scanning speed of 5 mm / s to acquire internal ultrasonic echo data. The data acquisition unit 20 uploads both types of data to the processing unit 21 in real time.

[0111] In the data processing and fusion stage, processing unit 21 first executes a coordinate mapping algorithm to map the ultrasound scan coordinates to the optical image coordinate system, and the scaling factor is adjusted. With offset The data was obtained through pre-calibration. For the surface optical data, a multi-field weighted fusion algorithm was used to remove bond line occlusion areas, and high-pass filtering was used to remove ambient light interference. Further edge recognition and surface quality assessment were performed on the optical images to extract two-dimensional features of surface defects. For the internal ultrasonic echo data, an adaptive filtering algorithm was applied, with a baseline signal-to-noise ratio... The signal-to-noise ratio (SNR) was set to 20 dB to effectively filter out reflection noise from the encapsulation layer interface. Subsequently, time-frequency analysis and echo feature extraction were performed to obtain the depth and amplitude characteristics of internal defects. Then, the system calculated the fused feature data based on an adaptive weighted fusion formula for the signal-to-noise ratio and established a three-dimensional defect distribution model of the IGBT module 10. Under the actual test conditions of this embodiment, due to good lighting conditions, the optical signal SNR was superior to the ultrasonic signal, and the system automatically calculated the optical weights. The value is 0.6, and the ultrasound weight is... With a value of 0.4, comprehensive extraction of surface scratches and internal pore features was achieved.

[0112] During the defect identification and output stage, processing unit 21 parses the design file of the FF450R12ME4 IGBT module and generates a location mask matrix. For the effective area of ​​the chip, the mask value... Set to 1; for non-detection areas such as positioning holes, the mask value is... The value is set to 0.7 to attenuate the false defect signal in this region. After correcting the fused feature data using a location mask, the system employs a decision fusion algorithm to calculate the overall defect probability. For detected surface bonding damage, the system adaptively adjusts the decision weight coefficients. It is 0.8. The accuracy is set to 0.2, thus accurately determining the type of surface defect. Finally, the system automatically outputs an inspection report and a visualization of the defect distribution. The inspection report includes quantitative information such as defect type, location, size, and confidence level, providing a basis for quality assessment and process optimization.

[0113] Actual test results show that, using the detection scheme of this embodiment, the defect location accuracy of the FF450R12ME4 IGBT module reaches ±5μm, the false defect filtering rate is ≥99%, and the overall defect detection rate is ≥99.5%. Compared with traditional single optical or ultrasonic detection schemes, this invention effectively solves problems such as multimodal data drift, low fusion accuracy, and false defect misjudgment by using hardware-level synchronous triggering, adaptive signal-to-noise ratio weighted fusion, and design file-driven position mask correction, significantly improving the accuracy and reliability of IGBT module defect detection.

[0114] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. It should be understood that the specific models and parameters of the structured light optical detection unit and ultrasonic detection unit described in the embodiments of the present invention, as well as the specific values ​​of the synchronous triggering timing, coordinate mapping, and weighted fusion formulas, are merely illustrative of the feasibility of the technical solution of the present invention and do not constitute a limitation on the scope of protection of the present invention. Any person skilled in the art, within the scope of the technology disclosed in the present invention, can easily conceive of variations or substitutions based on the core concepts of dual-modal hardware synchronous triggering, signal-to-noise ratio adaptive fusion, and design file-driven mask correction. For example, using different types of optical sensors, adjusting the weight calculation model of the fusion algorithm, or applying it to other power devices with similar detection requirements. These variations or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multimodal collaborative defect detection method for IGBT modules, characterized in that, include: In response to the synchronous trigger command, the structured light optical detection unit and the ultrasonic detection unit are controlled to detect the IGBT module respectively, and the surface optical data and internal ultrasonic echo data are acquired synchronously. Based on hardware synchronization timing, a spatial mapping relationship between the surface optical data and the internal ultrasonic echo data is established. Based on the signal-to-noise ratio, the optical features of the surface optical data and the ultrasonic features of the internal ultrasonic echo data are adaptively weighted and fused to generate fused feature data. The design file of the IGBT module is parsed to generate a location mask. The location mask is used to correct the fused feature data, and defects are identified based on the corrected fused feature data.

2. The method according to claim 1, characterized in that, The generation of the synchronous trigger instruction satisfies the timing constraints: ； in, This is the trigger time of the structured light optical detection unit. This is the trigger time of the ultrasonic detection unit. As the reference time, For hardware transmission delay, and .

3. The method according to claim 1, characterized in that, The establishment of the spatial mapping relationship includes: The ultrasonic scanning coordinates of the internal ultrasonic echo data are mapped to the optical image coordinates of the surface optical data using a coordinate mapping algorithm. The mapping formula is as follows: ； in, Optical image coordinates, For ultrasound scan coordinates, This is the scaling factor. This is the offset.

4. The method according to claim 1, characterized in that, The adaptive weighted fusion adopts the following formula: ； in, To fuse feature signals, It is an optical characteristic signal. It is a characteristic signal of ultrasound. and The weights are optical signal weights and ultrasonic signal weights, respectively, and the weight values ​​are adaptively determined based on the signal-to-noise ratio of each modal signal.

5. The method according to claim 1, characterized in that, The identification defects include: Based on the corrected fused feature data, the comprehensive defect probability is calculated using a decision fusion algorithm, as shown in the formula: ； in, To comprehensively assess the probability of defects, Confidence level for optical defects. Confidence level of ultrasonic defect. and Let be the decision weight coefficient, and .

6. The method according to claim 1, characterized in that, Parsing the design file of the IGBT module to generate a location mask includes: The design file is parsed to obtain the legal device region of the IGBT module; A location mask matrix is ​​generated, wherein the value of the location mask matrix within the detection area is a first preset value, and the value within the non-detection area is a second preset value.

7. The method according to claim 6, characterized in that, The fused feature data is corrected using the location mask according to the following formula: ； in, The corrected confidence level for the defect. The fused feature data before correction. The position mask matrix in coordinates The value at that location.

8. The method according to claim 1, characterized in that, Following the synchronous acquisition of surface optical data, the method further includes: The optical images acquired from multiple fields of view are weighted and fused to eliminate single-view occlusion areas, and high-pass filtering is used to remove ambient light interference.

9. The method according to claim 1, characterized in that, After the synchronous acquisition of internal ultrasound echo data, the following is also included: The internal ultrasound echo data is adaptively filtered using the following formula: ； in, This is the filtered signal. The original signal, The current ultrasound signal-to-noise ratio, The reference signal-to-noise ratio is 1.

10. A multimodal collaborative defect detection device for IGBT modules, characterized in that, include: The synchronous triggering system is configured to generate synchronous triggering commands. The structured light optical detection unit is configured to detect the IGBT module and acquire surface optical data in response to the synchronous triggering command; An ultrasonic detection unit is configured to detect the IGBT module and acquire internal ultrasonic echo data in response to the synchronous trigger command. The data acquisition unit is configured to transmit the surface optical data and the internal ultrasonic echo data; The processing unit is configured to perform the method as described in any one of claims 1 to 9.

Images