Bonded aluminum wire production toughness optimization method and system based on industrial test

Abstract

The present application relates to the technical field of bonding wire production, and particularly to a bonding aluminum wire production toughness optimization method and system based on industrial testing. By performing mechanical physical toughness testing and semiconductor packaging conductive testing on the bonding aluminum wire sample, the easy-to-deteriorate toughness abnormal points are located by combining wavelet packet energy spectrum analysis of the conductive current signal, and the chaotic disturbance factors in the low stability test sample are dynamically recovered and corrected under the corresponding production working condition state. In the constructed multi-modal toughness feature sequence space, the toughness deficiency features and the toughness excessive features of the abnormal points are decoupled and reconstructed to generate positive and negative test compensation samples, which are respectively directed to the annealing time and temperature parameters of the annealing process and the strain force and single pass reduction rate in the wire drawing process for optimization. The present application can realize accurate tracing of different toughness deterioration sources and process adaptive optimization, improve the plastic stability, mechanical strength and long-term reliability of the bonding aluminum wire.

Description

Technical Field

[0001] This invention relates to the field of bonding wire production technology, and in particular to a method and system for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing. Background Technology

[0002] As a crucial conductive material in semiconductor packaging for achieving electrical connections between chips and external circuits, the toughness of bonding aluminum wire directly impacts the bonding quality, solder joint reliability, and long-term operational stability of semiconductor devices. In actual production, bonding aluminum wire undergoes multiple precision processing steps, including melting, drawing, and annealing. Process parameters such as drawing strain, diameter reduction rate control, and annealing temperature and timing significantly affect the internal grain structure, dislocation distribution, and residual stress state of the aluminum wire, potentially leading to insufficient toughness, excessive softening, or easy degradation during long-term service. To ensure that bonding aluminum wire meets high-reliability packaging requirements, industrial production typically employs industrial-grade testing methods such as mechanical tensile testing, bending testing, and packaging conductivity current detection to comprehensively evaluate its toughness indicators, and combines these with changes in conductivity to identify potential material defects.

[0003] However, existing methods for optimizing the toughness of bonded aluminum wire production primarily rely on empirical adjustments to overall production parameters based on statistical results from testing. This lacks in-depth analysis of the correlation between transient conductivity anomalies and localized mechanical toughness instability, making it difficult to accurately pinpoint key processing nodes causing toughness degradation. Actual production environments are influenced by multiple factors, including equipment vibration, process fluctuations, and environmental disturbances, resulting in complex and nonlinear production conditions. Traditional static parameter compensation or empirical correction methods are insufficient to effectively describe the dynamic disturbance states during production, leading to limited accuracy in correcting abnormal conditions. Furthermore, existing process optimization methods typically treat toughness anomalies uniformly, lacking differentiated compensation mechanisms for wire drawing strain control and annealing process adjustments. This makes it difficult to implement precise optimization of the production line based on the source of the anomaly, easily leading to over- or under-adjustment of processes, affecting the consistency, yield, and long-term reliability of bonded aluminum wire products. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a method and system for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing, comprising the following steps: S01: Quantitatively extract bonded aluminum wire samples for industrial-grade toughness index mechanical and physical testing and semiconductor packaging conductivity testing, obtain physical toughness test parameters and conductivity current test signals, locate suspicious toughness performance test sites with abnormal transient currents based on wavelet packet energy spectrum analysis of conductivity current test signals, test the mechanical toughness of suspicious toughness performance test sites based on physical toughness test parameters, and obtain the toughness-unqualified products and easily deteriorated toughness anomalies of bonded aluminum wire samples. S02: Obtain the actual toughness processing condition when the bonding aluminum wire production line processes the abnormal point of easily deteriorated toughness. Calculate the disturbed deviation mode of the production line by dynamically tracking the attraction deviation of the actual toughness processing condition under the chaos of actual production. Based on the disturbed deviation mode, dynamically correct different chaotic factors in the low-stability test sample and generate a recovery correction vector. S03: Construct a multimodal toughness feature sequence space when the bonded aluminum wire production line processes products with unqualified toughness. In the multimodal toughness feature sequence space, decouple the local features of easily deteriorated toughness anomalies with respect to insufficient toughness and excessive toughness. Reconstruct the decoupled local features according to the recovery correction vector to generate a positive compensation sample matrix or a negative compensation sample matrix. S04: If the test compensation result output is a positive compensation sample matrix, then the actual annealing process is expanded and corrected based on the positive compensation sample matrix. The unreasonable actual annealing progress timing and annealing temperature of easily deteriorated toughness abnormal points are located and analyzed to optimize the annealing process of the production line. S05: If the test compensation result output is a negative compensation sample matrix, then the expected expansion correction of the actual wire drawing process is performed based on the negative compensation sample matrix, the abnormal wire drawing strain of the production line processing easily deteriorated toughness abnormal point is analyzed, and the process optimization of single-pass diameter reduction rate balance is performed.

[0006] Furthermore, S01 specifically includes the following steps: By quantitatively extracting bonded aluminum wire samples processed on the bonded aluminum wire production line, the extracted bonded aluminum wire samples are subjected to mechanical and physical tests of industrial-grade toughness indicators and packaging conductivity tests of power semiconductors, and the physical toughness test parameters and conductivity test signals of the bonded aluminum wire test samples within a predetermined time period are obtained. The preset test strategy input during power semiconductor package conductivity testing is obtained. The circuit modulation reference law followed by the package conductivity test is obtained by retrieving the preset test strategy through the semiconductor package knowledge graph. Based on the circuit modulation reference law, a spectrum detection framework for the current test signal is constructed. Based on big data, the prior current signal characteristics of the semiconductor conductivity abnormality caused by the non-compliance of the toughness of the bonded aluminum wire are obtained. The conductivity current test signal is fitted to the spectrum detection framework to generate the time-series current flux spectrum curve. Based on the prior current signal characteristics, the current flux frequency band is reconstructed by wavelet packet decomposition of the time-series current flux spectrum curve to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of non-compliance of the toughness of the bonded aluminum wire sample. According to the preset test strategy, the bonded aluminum wire sample is divided into k sub-test segments. The transient current singular inflection point and its conductance energy kurtosis corresponding to each sub-test segment on the time-series current flux spectrum curve are obtained by wavelet packet energy spectrum identification. At the same time, the conductance energy kurtosis range of different prior current signal characteristics is obtained based on big data. If the conductivity energy kurtosis is within any conductivity energy kurtosis interval, then the sub-test segment where the singular inflection point of the transient current is located is designated as the test site for the suspected toughness performance of the bonded aluminum wire sample. Based on the regional performance distribution of physical toughness test parameters, a physical toughness thermodynamic model diagram of the bonded aluminum wire sample with respect to each sub-test segment is constructed. The physical toughness thermodynamic colorimetric value corresponding to the suspected toughness performance test position is extracted through the physical toughness thermodynamic model diagram. Based on the production index requirements of bonded aluminum wire, a pre-defined expected thermochromic gamut is set for the physical toughness frequency. If the physical toughness thermochromic value cannot be found in the expected thermochromic gamut, the bonded aluminum wire sample is marked as a product with unqualified toughness, and this suspected toughness performance test position is marked as an abnormal point of easily deteriorated toughness.

[0007] Furthermore, the method of obtaining prior current signal characteristics based on big data to determine the semiconductor conductivity abnormality caused by the non-compliance of the bonding aluminum wire toughness, fitting the conductivity current test signal to a spectrum detection framework to generate a time-series current flux spectrum curve, and reconstructing the current flux frequency band by wavelet packet decomposition of the time-series current flux spectrum curve based on the prior current signal characteristics to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of non-compliance of the bonding aluminum wire sample toughness, specifically includes the following steps: Based on big data, different prior current signal characteristics of power semiconductor packaging caused by unqualified toughness in the production of bonding aluminum wire are obtained. The conductivity test signal is time-fitted in the spectrum monitoring framework to form the time-series current flux spectrum curve of bonding aluminum wire packaging conductivity test. A wavelet basis transform algorithm is introduced to set the wavelet packet basis function of the abnormal current characteristics of power semiconductor based on the prior current signal characteristics. Based on the wavelet packet basis function, the signal characteristics of the time-series current flux spectrum curve are decomposed by wavelet packet, and the multiple frequency band link nodes of the current flux on the bonded aluminum wire sample and their corresponding frequency band node coefficients are output. Based on the frequency band node coefficients, the corresponding frequency band signals of each frequency band link node on the time-series current flux spectrum curve are reconstructed. During the reconstruction process, the flux sampling energy of each frequency band signal is calculated to obtain the new frequency band energy index. All new frequency band energy indices are summed and the energy ratio is calculated to generate the wavelet packet energy spectrum of the conduction current test signal under the premise that the toughness of the bonded aluminum wire sample is unqualified.

[0008] Furthermore, S02 specifically includes the following steps: The production timestamps of the gradually appearing easily deteriorating toughness anomalies when the bonding aluminum wire production line processes products with unqualified toughness are defined as the toughness degradation source production timestamps. By retrieving the production control logs of bonded aluminum wire processing, the actual toughness processing conditions of the bonded aluminum wire production line at the time stamp of the toughness degradation source processing point were obtained from the output of different multimodal process data. The distribution of state points affected by different types of chaotic factors was also retrieved from the actual toughness processing conditions of the bonded aluminum wire production line. Among them, chaotic factors include equipment fluctuations, noise offset, system errors, batch differences and environmental disturbances. The underlying processing control architecture, coupled dynamic logic of multimodal processes, and chaotic factor characteristics of the bonded aluminum wire production line are obtained. The process production phase space domain of bonded aluminum wire is established by combining the underlying processing control architecture, coupled dynamic logic, and chaotic factor characteristics. Based on big data and experience processing cases, several historical steady-state operation trajectory templates were obtained for the actual toughness processing conditions of the bonded aluminum wire production line after processing easily deteriorated toughness anomalies under the premise of suppressing or rejecting various types of chaotic factors. The baseline trajectory convergence simulation of the ideal toughness processing attraction under the condition of maintaining low chaos in the initial state of the bonded aluminum wire production line was carried out using the historical steady-state operation trajectory templates, and the target attraction domain boundary model was generated. Based on the distribution of state points, the mapping is injected into the process production phase space to dynamically track the actual toughness processing conditions. Based on the tracking results, the real-time attractor trajectory of the actual toughness processing conditions caused by the row bonding aluminum wire production line being disturbed by chaotic factors is determined. Calculate the attraction deviation index of the real-time attractor trajectory from the boundary model of the target attraction domain, and identify and analyze the disturbance deviation mode of chaotic factors in the bonding aluminum wire processing production based on the attraction deviation index. By retrieving data and experience-based case studies, the structural deviation direction and correction gain scale that induce the deterioration of the toughness of bonded aluminum wire production due to the perturbation deviation mode are obtained. Based on the structural deviation direction and correction gain scale, the perturbation of the chaotic factor is dynamically corrected, and a recovery correction vector is generated and superimposed on the control system of the bonded aluminum wire production line.

[0009] Furthermore, S03 specifically includes the following steps: By obtaining the multimodal toughness features of the bonded aluminum wire production line at different process stages during the processing of products with substandard toughness from the production control logs, based on the output of the detection images, a multimodal toughness feature sequence space is constructed. A depth encoder is introduced, and a mutual information attention network is embedded in the intermediate layer of the depth encoder. After embedding, the intermediate layer is used to extract features from the multimodal toughness feature sequence space to obtain the local toughness feature map of the toughness-deficient product in each process stage of the sequential processing. Obtain the production jump response boundary points between the time sequences of each process stage, preset the handover and convergence function according to the process echelon of the production jump response boundary points, and perform aggregation operation on the local toughness feature maps of all process stage time sequences based on the handover and convergence function to generate a global summary matrix of the multimodal toughness feature sequence space. The actual process parameters and real-time toughness parameters of easily deteriorated toughness anomalies in the bonding aluminum wire production line at the process stage are obtained. Based on big data and experience processing cases, the actual process parameters, real-time toughness parameters and chaotic factor characteristics are retrieved to obtain the toughness insufficiency and toughness over-characterization of the real-time toughness process corresponding to the easily deteriorated toughness anomalies caused by different chaotic factors in the bonding aluminum wire production line. A sample discrimination network based on mutual information theory is introduced. Positive sample pairs with local resilience features are constructed based on the resilience deficiency characterization, and negative sample pairs with local resilience features are constructed based on the resilience over-characterization. The positive sample pairs and negative samples are combined in the sample discrimination network to perform feature representation modeling and generate a mutual information discriminator. Based on the global summary matrix as index information, the local resilience feature map is imported into the mutual information discriminator for sample probability discrimination and analysis, so as to obtain the mutual information probability value of different local resilience features belonging to positive sample pairs or negative sample pairs. If the mutual information probability value is greater than the maximum mutual information probability threshold, then this local resilience feature is labeled as a chaotic resilience feature and local decoupling is performed. The chaotic resilience feature after local decoupling is reconstructed and modeled according to the recovery correction vector to obtain a positive compensation sample matrix or a negative compensation sample matrix.

[0010] Furthermore, S04 specifically includes the following steps: If the test compensation result shows a positive compensation sample matrix, then the annealing decision service framework of the bonded aluminum wire production line is obtained, and the full process information of the standard annealing process for processing products with unqualified toughness in the bonded aluminum wire production line is retrieved. The actual annealing process planning set for products with unqualified toughness is obtained by acquiring the full process information of the standard annealing process within a preset time period from the production control log. The process planning subset corresponding to the easily deteriorated toughness anomaly points is extracted from the actual annealing planning set and defined as a type of compensation slice item. The positive compensation sample matrix is ​​introduced into the type of compensation slice item for expansion and correction of the toughness performance dimension. To obtain the ideal toughness index of the bonded aluminum wire sample, the entire process information of the standard annealing process, the ideal toughness index, and the expanded compensation slice item are input into the annealing decision service framework for background decision-making. The output is the expected annealing process planning set for the bonded aluminum wire production line to avoid or eliminate easily deteriorated toughness anomalies by ideal standardized annealing processing. Using the full-process information of the standard annealing process as the time-series baseline, the expected annealing time-series intervals of different annealing progress steps and each step length of the annealing process at the time-series baseline are extracted through the expected annealing process planning set. The actual annealing progress status sequence and its temperature distribution data of the bonded aluminum wire production line based on the actual annealing process planning set time-series of the production process at the point of easy deterioration of toughness are obtained. Based on different annealing follow-up lengths, the actual annealing progress state sequence is divided into n effective annealing timing windows. If the actual annealing period of an effective annealing timing window is shorter than the expected annealing time interval, then the effective annealing timing window is marked as an annealing timing pullback node. The temperature rise gradient index of each annealing timing retraction node is obtained by acquiring temperature distribution data. The actual process parameters of the annealing timing retraction node corresponding to the temperature rise gradient index of the bonded aluminum wire production line being less than the preset threshold are extracted. Based on the parameter deviation between the expected annealing process planning set and the actual process parameters, the annealing process of the production line is optimized for insufficient toughness, and the first production toughness optimization scheme is obtained.

[0011] Furthermore, S05 specifically includes the following steps: If the test compensation result shows a negative compensation sample matrix, then retrieve the full process information of the standard wire drawing process for products with unqualified toughness from the bonded aluminum wire production line. By processing the actual wire drawing process planning set of products with unqualified toughness through the full process information of the standard wire drawing process, the process planning subset corresponding to the easily deteriorated toughness anomaly point is extracted and defined as the second type of compensation slice item. The negative compensation sample matrix is ​​introduced into the second type of compensation slice item for the expansion and correction of the toughness performance dimension. Based on the full process information of the standard wire drawing process, ideal toughness index and expanded second-class compensation slice items, a decision is made to output the expected wire drawing process planning set for the ideal standardized wire drawing process of the bonded aluminum wire production line to eliminate easily deteriorated toughness anomalies. Based on the concept of spatiotemporal alignment and combined with hashing algorithms, the spatiotemporal hash degree of the wire drawing operations and timing of the actual wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the first spatiotemporal hash degree; the spatiotemporal hash degree of the wire drawing operations and timing of the expected wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the second spatiotemporal hash degree. The strain distribution pattern of easily deteriorated toughness anomalies is obtained by obtaining the production control log. The strain values ​​of different drawing distribution blocks are obtained based on the strain distribution pattern. The hash deviation of the first spatiotemporal hash degree compared with the second spatiotemporal hash degree is calculated to obtain the hash deviation value. If the strain value is greater than the tolerance strain threshold of the ideal toughness index, the drawing distribution block is marked as an abnormal local drawing block. At this time, the strain value is compensated and corrected based on the hash deviation value to obtain the correct strain value of the abnormal local drawing block. Based on big data, a positive correlation mechanism model of the wire drawing process state and strain force change is obtained. The correct strain force value is simulated through the positive correlation mechanism model to obtain the current single-pass diameter reduction rate of easily deteriorated toughness anomaly points. If the current single-pass reduction rate reaches the ideal single-pass reduction rate threshold, then the ideal wire drawing process parameters with the correct strain value are output. Based on the parameter deviation between the ideal wire drawing process parameters and the actual wire drawing process parameters, the wire drawing process of the production line is optimized for toughness overload, and a second production toughness optimization scheme is obtained.

[0012] The second aspect of the present invention provides a bonding aluminum wire production toughness optimization system based on industrial-grade testing. The system includes: a memory, a processor, and a communication interface. The memory includes a bonding aluminum wire production toughness optimization method program based on industrial-grade testing. The communication interface is used for data connection communication between the memory and the processor. When the bonding aluminum wire production toughness optimization method program is executed by the processor, it implements any of the bonding aluminum wire production toughness optimization method steps described in the present invention.

[0013] This invention addresses the technical deficiencies in the prior art, and its beneficial technical effects are as follows: This invention constructs a toughness testing compensation mechanism based on industrial-grade mechanical and physical test parameters, packaged conductive test signals, and actual production conditions. It dynamically tracks and corrects chaotic deviations caused by equipment disturbances, process fluctuations, and complex processing environments during the production of bonded aluminum wire. In the multimodal toughness feature sequence space, it accurately decouples and compensates for the insufficient toughness features and excessive toughness features of easily deteriorated toughness anomalies, generating positive and negative compensation sample matrices with high realism and high discriminability, thereby achieving accurate identification of different toughness degradation mechanisms. Furthermore, by using a positive compensation sample matrix, key parameters such as annealing timing and annealing temperature in the annealing process are optimized to avoid embrittlement and decreased plasticity of aluminum wire due to insufficient annealing. By using a negative compensation sample matrix, the wire drawing strain and single-pass diameter reduction rate in the wire drawing process are balanced and controlled to avoid excessive softening and strength reduction of aluminum wire due to abnormal wire drawing. Ultimately, this achieves high-precision detection of toughness state, source tracing of toughness defects, and adaptive optimization control of key annealing-wire drawing processes during the production of bonded aluminum wire. This effectively improves the plastic deformation capacity, mechanical strength, solder joint connection reliability, and long-term service stability of bonded aluminum wire, and reduces the risk of early failure of semiconductor packaging devices. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0015] Figure 1 A first method flowchart for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing is shown. Figure 2 A second method flowchart for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing is shown. Figure 3 A system framework diagram of a bonding aluminum wire production toughness optimization system based on industrial-grade testing is shown. Detailed Implementation

[0016] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0018] The first aspect of this invention provides a method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing, such as... Figure 1 As shown, it includes the following steps: S01: Quantitatively extract bonded aluminum wire samples for industrial-grade toughness index mechanical and physical testing and semiconductor packaging conductivity testing, obtain physical toughness test parameters and conductivity current test signals, locate suspicious toughness performance test sites with abnormal transient currents based on wavelet packet energy spectrum analysis of conductivity current test signals, test the mechanical toughness of suspicious toughness performance test sites based on physical toughness test parameters, and obtain the toughness-unqualified products and easily deteriorated toughness anomalies of bonded aluminum wire samples. S02: Obtain the actual toughness processing condition when the bonding aluminum wire production line processes the abnormal point of easily deteriorated toughness. Calculate the disturbed deviation mode of the production line by dynamically tracking the attraction deviation of the actual toughness processing condition under the chaos of actual production. Based on the disturbed deviation mode, dynamically correct different chaotic factors in the low-stability test sample and generate a recovery correction vector. S03: Construct a multimodal toughness feature sequence space when the bonded aluminum wire production line processes products with unqualified toughness. In the multimodal toughness feature sequence space, decouple the local features of easily deteriorated toughness anomalies with respect to insufficient toughness and excessive toughness. Reconstruct the decoupled local features according to the recovery correction vector to generate a positive compensation sample matrix or a negative compensation sample matrix. S04: If the test compensation result output is a positive compensation sample matrix, then the actual annealing process is expanded and corrected based on the positive compensation sample matrix. The unreasonable actual annealing progress timing and annealing temperature of easily deteriorated toughness abnormal points are located and analyzed to optimize the annealing process of the production line. S05: If the test compensation result output is a negative compensation sample matrix, then the expected expansion correction of the actual wire drawing process is performed based on the negative compensation sample matrix, the abnormal wire drawing strain of the production line processing easily deteriorated toughness abnormal point is analyzed, and the process optimization of single-pass diameter reduction rate balance is performed.

[0019] Furthermore, S01 specifically includes the following steps: By quantitatively extracting bonded aluminum wire samples processed on the bonded aluminum wire production line, the extracted bonded aluminum wire samples are subjected to mechanical and physical tests of industrial-grade toughness indicators and packaging conductivity tests of power semiconductors, and the physical toughness test parameters and conductivity test signals of the bonded aluminum wire test samples within a predetermined time period are obtained. The preset test strategy input during power semiconductor package conductivity testing is obtained. The circuit modulation reference law followed by the package conductivity test is obtained by retrieving the preset test strategy through the semiconductor package knowledge graph. Based on the circuit modulation reference law, a spectrum detection framework for the current test signal is constructed. Based on big data, the prior current signal characteristics of the semiconductor conductivity abnormality caused by the non-compliance of the toughness of the bonded aluminum wire are obtained. The conductivity current test signal is fitted to the spectrum detection framework to generate the time-series current flux spectrum curve. Based on the prior current signal characteristics, the current flux frequency band is reconstructed by wavelet packet decomposition of the time-series current flux spectrum curve to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of non-compliance of the toughness of the bonded aluminum wire sample. According to the preset test strategy, the bonded aluminum wire sample is divided into k sub-test segments. The transient current singular inflection point and its conductance energy kurtosis corresponding to each sub-test segment on the time-series current flux spectrum curve are obtained by wavelet packet energy spectrum identification. At the same time, the conductance energy kurtosis range of different prior current signal characteristics is obtained based on big data. If the conductivity energy kurtosis is within any conductivity energy kurtosis interval, then the sub-test segment where the singular inflection point of the transient current is located is designated as the test site for the suspected toughness performance of the bonded aluminum wire sample. Based on the regional performance distribution of physical toughness test parameters, a physical toughness thermodynamic model diagram of the bonded aluminum wire sample with respect to each sub-test segment is constructed. The physical toughness thermodynamic colorimetric value corresponding to the suspected toughness performance test position is extracted through the physical toughness thermodynamic model diagram. Based on the production index requirements of bonded aluminum wire, a pre-defined expected thermochromic gamut is set for the physical toughness frequency. If the physical toughness thermochromic value cannot be found in the expected thermochromic gamut, the bonded aluminum wire sample is marked as a product with unqualified toughness, and this suspected toughness performance test position is marked as an abnormal point of easily deteriorated toughness.

[0020] It is important to note that bonding aluminum wires, as crucial conductive interconnects between power semiconductor chips and external circuits, directly impact the stress distribution, conductive path stability, and fatigue resistance of the bonding interface under long-term exposure to high currents, thermal cycling, mechanical vibration, and environmental stress. Without industrial-grade testing before shipment, it is difficult to detect potential failure risks caused by abnormal material toughness, structural defects, or manufacturing process fluctuations. This leads to aluminum wires with insufficient or excessive toughness being directly used in the packaging process. The lack of mechanical and physical performance verification alongside conductive testing during packaging makes it impossible to effectively assess the tensile strength, fatigue resistance, and vibration resistance of the aluminum wires. This makes the devices more prone to performance degradation, intermittent failures, or sudden failures during long-term operation, shortening the lifespan of power semiconductor devices. Therefore, by quantitatively sampling and conducting industrial-grade toughness tests on the bonded aluminum wires processed on the production line, including two major testing systems—power semiconductor packaging conductivity testing and mechanical performance testing—precise coupling verification of the bonded aluminum wires' performance in both the energized operation dimension and the physical level is achieved. This identifies abnormal toughness regions on the bonded aluminum wires, characterized by anomalies in current pathways and low mechanical properties, providing reliable support for the reshaping of chaotic effects and toughness characteristics in subsequent aluminum wire production and processing. Specifically, by timing-fitting the conductivity test signal obtained from the power semiconductor packaging conductivity test into a spectrum detection framework based on the circuit modulation reference law, a time-series current flux spectrum curve is generated. This spectrum detection framework ensures that the analysis of the conductivity signal conforms to the operating rules of the packaged circuit. Next, the time-series current flux spectrum curve is reconstructed by wavelet packet decomposition using the prior current signal characteristics of semiconductor conductivity abnormality caused by the non-compliance of the bonding aluminum wire toughness. Fine-grained frequency domain features associated with the premise of toughness defects in the bonding aluminum wire are extracted from the complex current signal. Wavelet packet energy spectra reflecting the energy distribution characteristics of the packaged conductivity flux in different frequency bands on the bonding aluminum wire sample are generated, so as to effectively amplify and accurately characterize the potential conductivity abnormality information.

[0021] Furthermore, based on a pre-defined testing strategy, the bonded aluminum wire samples during the testing process are divided into multiple testing regions, i.e., sub-test segments. By combining wavelet packet energy spectroscopy, transient current singular inflection points and corresponding conductance energy kurtosis in different testing regions of the time-series current flux spectrum curve are identified. The conductance energy kurtosis interval formed by prior anomalous samples is introduced as a judgment criterion. The transient current singular inflection point refers to the turning point on the time-series current flux spectrum curve where the curvature energy distribution undergoes a rapid oscillation within a very short time due to abrupt changes in the conductive path, material microstructure, or contact interface degradation. The conductance energy kurtosis describes the concentration and sharpness of the conductive energy in the spectrum curve at the transient current singular inflection point. Its function is to establish a correlation mapping between current anomaly characteristics and toughness defects, achieve precise location of anomalous test segments, and measure the transient inflection force of conductive current energy caused by the aluminum wire material properties. If the conductivity energy kurtosis falls within any conductivity energy kurtosis range, it indicates that the instantaneous anomaly in the current flux measured by the conductivity test of the bonded aluminum wire sample closely matches the current anomaly characteristics of a semiconductor conductivity anomaly caused by a certain toughness failure. This suggests that the current signal anomaly is highly likely caused by a certain negative toughness property, and therefore it is temporarily classified as a suspected toughness object in the conductivity current dimension, i.e., a suspected toughness property test site. Subsequently, a physical toughness thermodynamic model diagram is constructed based on the regional spatial performance distribution of the physical toughness test parameters, and the thermodynamic colorimetric values ​​of the suspected test sites are extracted to visualize the abstract toughness parameters. This provides a global and intuitive reflection of the local toughness strength and uniformity distribution characteristics of the sample aluminum wire material, achieving cross-verification of the electrical anomaly interface and the mechanical performance interface. If the desired thermochromatic gamut does not yield the physical toughness thermochromatic value, it indicates that the mechanical toughness performance of the suspected toughness performance test site is excluded from the reasonable toughness performance and frequency of occurrence specified in the production index requirements. This means that the local material of this bonded aluminum wire sample has both abnormal conductivity and physical toughness, which is an aluminum wire material point that is extremely prone to deterioration. Therefore, this sample is unqualified, and this test site is marked as a toughness anomaly point that is prone to deterioration. This achieves the synergistic judgment of abnormal conductivity of bonded aluminum wire and material toughness failure, accurately identifying potential toughness anomaly areas and quality risk points.

[0022] In summary, this method enables industrial-grade linkage testing of the mechanical and physical properties of bonded aluminum wires with the conductivity of semiconductor packaging. It achieves a deep correlation analysis between the toughness defects of bonded aluminum wires and the conductivity anomalies in packaging, accurately identifying localized toughness anomalies that are difficult to detect with traditional methods. This allows for early location of critical locations prone to degradation and failure, providing a foundation of abnormal data for subsequent toughness production optimization. Ultimately, this improves the conductivity reliability, mechanical stability, and long-term service life of bonded aluminum wires in power semiconductor packaging products.

[0023] Furthermore, the method of obtaining prior current signal characteristics based on big data to determine the semiconductor conductivity abnormality caused by the non-compliance of the bonding aluminum wire toughness, fitting the conductivity current test signal to a spectrum detection framework to generate a time-series current flux spectrum curve, and reconstructing the current flux frequency band by wavelet packet decomposition of the time-series current flux spectrum curve based on the prior current signal characteristics to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of non-compliance of the bonding aluminum wire sample toughness, specifically includes the following steps: Based on big data, different prior current signal characteristics of power semiconductor packaging caused by unqualified toughness in the production of bonding aluminum wire are obtained. The conductivity test signal is time-fitted in the spectrum monitoring framework to form the time-series current flux spectrum curve of bonding aluminum wire packaging conductivity test. A wavelet basis transform algorithm is introduced to set the wavelet packet basis function of the abnormal current characteristics of power semiconductor based on the prior current signal characteristics. Based on the wavelet packet basis function, the signal characteristics of the time-series current flux spectrum curve are decomposed by wavelet packet, and the multiple frequency band link nodes of the current flux on the bonded aluminum wire sample and their corresponding frequency band node coefficients are output. Based on the frequency band node coefficients, the corresponding frequency band signals of each frequency band link node on the time-series current flux spectrum curve are reconstructed. During the reconstruction process, the flux sampling energy of each frequency band signal is calculated to obtain the new frequency band energy index. All new frequency band energy indices are summed and the energy ratio is calculated to generate the wavelet packet energy spectrum of the conduction current test signal under the premise that the toughness of the bonded aluminum wire sample is unqualified.

[0024] It should be noted that, for the construction of the wavelet packet energy spectrum under the premise of unqualified toughness of the bonded aluminum wire sample, specifically, the marginal constraint of the aluminum wire sample's unqualification is fixed by using the a priori abnormal current of the power semiconductor package's conductivity anomaly, i.e., the a priori current signal characteristics. By utilizing these a priori features, wavelet packet basis functions corresponding to the semiconductor package's conductivity anomaly current characteristics are constructed in the wavelet basis transform algorithm. Different wavelet packet bases can highlight the signal characteristic patterns of different a priori current types, such as spikes, impulses, or periodic components. Through the wavelet packet basis functions, the frequency characteristics, non-stationarity, and transient variation patterns of the measured signal can be potentially identified, determining the time-frequency resolution and frequency band division of the conduction current signal, improving the frequency resolution and fine-grained feature extraction capability of the time-series current flux spectrum curve, and enhancing the identifiability of instantaneous current flux anomaly information. The time-series current flux spectrum curves were decomposed at multiple levels using wavelet packet basis functions of different prior types to obtain a series of frequency band link nodes of the current flux in the bonded aluminum wire sample. These frequency band link nodes belong to the fine recursive components of the low-frequency and high-frequency spectra of the current flux. At this point, the coupling phenomenon between the frequency components is effectively weakened, forming a chain-like subband structure with a very clear frequency affiliation. Through the quantification of the frequency band node coefficients of the full-band decomposition, the frequency substructure of the abnormal conductivity of the power semiconductor package caused by the unqualified toughness of the bonded aluminum wire in the full frequency band was refined. This fully reveals the hidden abrupt frequency characteristics and local energy change laws inside the conduction current signal, providing structured data support for subsequent frequency band energy analysis.

[0025] By independently reconstructing the signal components of each frequency band link node using frequency band node coefficients, the true waveform information of each frequency band in the time domain is restored. Each reconstructed signal contains only the circuit energy test information within the abnormal conductivity frequency range when the bonding aluminum wire production toughness is unqualified, which can clearly reflect the dynamic energy changes within the corresponding frequency range. By performing square accumulation on the reconstructed signals of each frequency band, the energy carried by each frequency band is calculated, quantifying the energy information intensity of different conductivity frequency regions. The energy proportion reflects the distribution of the current signal energy of the bonding aluminum wire sample in different frequency intervals, revealing the importance and activity of frequency components. Finally, the energy proportions of each frequency band are arranged in frequency order to generate a wavelet packet energy spectrum structure, which helps to quickly reveal local abnormal path patterns in conductivity current testing and effectively improves the interpretability of bonding aluminum wire test results.

[0026] Furthermore, the S02, as Figure 2 As shown, the specific steps include: The production timestamps of the gradually appearing easily deteriorating toughness anomalies when the bonding aluminum wire production line processes products with unqualified toughness are defined as the toughness degradation source production timestamps. By retrieving the production control logs of bonded aluminum wire processing, the actual toughness processing conditions of the bonded aluminum wire production line at the time stamp of the toughness degradation source processing point were obtained from the output of different multimodal process data. The distribution of state points affected by different types of chaotic factors was also retrieved from the actual toughness processing conditions of the bonded aluminum wire production line. Among them, chaotic factors include equipment fluctuations, noise offset, system errors, batch differences and environmental disturbances. The underlying processing control architecture, coupled dynamic logic of multimodal processes, and chaotic factor characteristics of the bonded aluminum wire production line are obtained. The process production phase space domain of bonded aluminum wire is established by combining the underlying processing control architecture, coupled dynamic logic, and chaotic factor characteristics. Based on big data and experience processing cases, several historical steady-state operation trajectory templates were obtained for the actual toughness processing conditions of the bonded aluminum wire production line after processing easily deteriorated toughness anomalies under the premise of suppressing or rejecting various types of chaotic factors. The baseline trajectory convergence simulation of the ideal toughness processing attraction under the condition of maintaining low chaos in the initial state of the bonded aluminum wire production line was carried out using the historical steady-state operation trajectory templates, and the target attraction domain boundary model was generated. Based on the distribution of state points, the mapping is injected into the process production phase space to dynamically track the actual toughness processing conditions. Based on the tracking results, the real-time attractor trajectory of the actual toughness processing conditions caused by the row bonding aluminum wire production line being disturbed by chaotic factors is determined. Calculate the attraction deviation index of the real-time attractor trajectory from the boundary model of the target attraction domain, and identify and analyze the disturbance deviation mode of chaotic factors in the bonding aluminum wire processing production based on the attraction deviation index. By retrieving data and experience-based case studies, the structural deviation direction and correction gain scale that induce the deterioration of the toughness of bonded aluminum wire production due to the perturbation deviation mode are obtained. Based on the structural deviation direction and correction gain scale, the perturbation of the chaotic factor is dynamically corrected, and a recovery correction vector is generated and superimposed on the control system of the bonded aluminum wire production line.

[0027] It should be noted that the production of bonded aluminum wire is affected by a series of uncertainties, such as equipment fluctuations, environmental drift, and batch differences. This leads to significant dispersion in the toughness characterization of bonded aluminum wire, blurred performance boundaries, and a high risk of fatigue failure. Consequently, this results in complex coupling of process parameters, amplified performance errors, and difficulties in optimization and control. To address this, by dynamically correcting the noise offset, systematic errors, and environmental disturbances of low-stability test samples (easily deteriorated toughness anomalies) in products failing toughness tests on the production line, non-target interference components in the toughness of bonded aluminum wire production can be effectively eliminated. This restores the original toughness performance of defective bonded aluminum wire products, improves the ability of industrial-grade test data to characterize the true toughness state of bonded aluminum wire, and ensures the accuracy of toughness indicator defect testing and compensation.

[0028] Specifically, this study acquires actual operating condition data of easily deteriorated toughness anomalies generated when a bonded aluminum wire production line processes products with substandard toughness, as well as the distribution of these state points under different chaotic perturbations. The state point distribution characterization represents the spatiotemporal distribution pattern and evolution trend of the actual toughness processing conditions under different chaotic perturbations, effectively revealing the perturbation propagation mechanism in the formation of easily deteriorated toughness anomalies. Actual toughness processing conditions include, but are not limited to, wire drawing speed, annealing temperature, tension fluctuations, vibration amplitude, current changes, and ambient temperature and humidity. Multimodal process data includes bonding pressure, surface resistivity, grain size, and mechanical property testing information. Furthermore, a process production phase space domain reflecting the operating characteristics and changing trends of the bonded aluminum wire production line under different actual operating conditions is constructed, anchoring the true state trajectory carrier of toughness production dynamics. By leveraging big data and empirical case studies, a historical steady-state operational trajectory template of the bonded aluminum wire production line was obtained after processing easily deteriorated toughness anomalies under conditions unaffected by chaotic factors. This template serves as a simulation set to describe the boundary relationship between the stable and unstable regions of the bonded aluminum wire production line and the ideal toughness processing conditions. It defines a crucial critical baseline for the transition from a toughness-stable to agile production conditions. Through the establishment of a target attraction domain boundary model, the toughness performance range and its time-varying patterns under stable conditions of the production line can be clearly characterized. The distribution of toughness benchmark features in the stable, transition, and escape regions is detailed, and the dynamic evolution characteristics and stability margin information near the attraction domain boundary are extracted. Subsequently, the distribution of state points mapped in the process production phase space is used to dynamically track the attraction domain region corresponding to different state points and their actual toughness processing conditions. This reveals the potential instability trend of the production line deviating from steady-state production conditions when processing easily deteriorated toughness anomalies under different chaotic factor conditions. Further calculations are performed to determine the attraction deviation of the real-time attractor trajectory from the boundary model of the target attraction domain. The attraction deviation measures the degree of deviation of the current working state of the abnormal point in the production line from the stable equilibrium state. It can accurately locate the low-stability sample area disturbed by chaotic factors, providing a basis for tracing and identification for subsequent dynamic correction decisions.

[0029] Furthermore, S03 specifically includes the following steps: By obtaining the multimodal toughness features of the bonded aluminum wire production line at different process stages during the processing of products with substandard toughness from the production control logs, based on the output of the detection images, a multimodal toughness feature sequence space is constructed. A depth encoder is introduced, and a mutual information attention network is embedded in the intermediate layer of the depth encoder. After embedding, the intermediate layer is used to extract features from the multimodal toughness feature sequence space to obtain the local toughness feature map of the toughness-deficient product in each process stage of the sequential processing. Obtain the production jump response boundary points between the time sequences of each process stage, preset the handover and convergence function according to the process echelon of the production jump response boundary points, and perform aggregation operation on the local toughness feature maps of all process stage time sequences based on the handover and convergence function to generate a global summary matrix of the multimodal toughness feature sequence space. The actual process parameters and real-time toughness parameters of easily deteriorated toughness anomalies in the bonding aluminum wire production line at the process stage are obtained. Based on big data and experience processing cases, the actual process parameters, real-time toughness parameters and chaotic factor characteristics are retrieved to obtain the toughness insufficiency and toughness over-characterization of the real-time toughness process corresponding to the easily deteriorated toughness anomalies caused by different chaotic factors in the bonding aluminum wire production line. A sample discrimination network based on mutual information theory is introduced. Positive sample pairs with local resilience features are constructed based on the resilience deficiency characterization, and negative sample pairs with local resilience features are constructed based on the resilience over-characterization. The positive sample pairs and negative samples are combined in the sample discrimination network to perform feature representation modeling and generate a mutual information discriminator. Based on the global summary matrix as index information, the local resilience feature map is imported into the mutual information discriminator for sample probability discrimination and analysis, so as to obtain the mutual information probability value of different local resilience features belonging to positive sample pairs or negative sample pairs. If the mutual information probability value is greater than the maximum mutual information probability threshold, then this local resilience feature is labeled as a chaotic resilience feature and local decoupling is performed. The chaotic resilience feature after local decoupling is reconstructed and modeled according to the recovery correction vector to obtain a positive compensation sample matrix or a negative compensation sample matrix.

[0030] It should be noted that although the method can effectively remove the influence of external interference factors on the test data and improve the authenticity and stability of the data, the corrected samples may still contain complex feature interweaving phenomena caused by fluctuations in production processes, differences in material microstructure, changes in equipment status, and the coupling effect of multiple source process parameters. This causes toughness-related features and non-toughness-related features to be mixed, affecting the accurate characterization of toughness state and reducing the compensation accuracy of industrial-grade testing. Therefore, this method decouples and reconstructs toughness features in the process feature space, separating the core features reflecting the true toughness change law of bonded aluminum wire from multi-dimensional process information, weakening the linkage interference of redundant features and outlier noise, and strengthening the intrinsic correlation and structural integrity between toughness features through feature reconstruction. This allows for the determination of the toughness anomaly type of bonded aluminum wire samples, generating compensation samples with high specificity, high consistency, and high discriminativeness, especially enhancing the differentiated expression of bonded aluminum wire in plastic deformation capacity, tensile strength, and fatigue durability at different production stages.

[0031] Specifically, the multimodal toughness features of products with substandard toughness processed at different stages of the bonded aluminum wire production line are sequentially linked to construct a multimodal toughness feature sequence space representing the toughness evolution trajectory, thus fixing the exploration basis. Since the modal data from different process stages have a high temporal dependence on the degradation of toughness anomalies, a mutual information attention network is embedded in the intermediate layer of the deep encoder. This mutual information attention network is used to mine the potential correlations between the modal features of different production toughness processes. The mutual information mechanism strengthens the expression of feature information highly correlated with toughness degradation, suppresses noise interference from irrelevant chaotic factors, and improves the chaotic correlation representation ability of local toughness features. After obtaining the local toughness features of each process stage, attention aggregation processing is performed to form a global toughness feature vector. The production jump response boundary point is the temporal response boundary point for jumping from one process stage to another. The pre-defined crossover convergence function defines the attention allocation ratio for the toughness defect features of each process stage. The global summary matrix of the generated multimodal toughness feature sequence space describes the high-level semantic outline and global structural interpretation of the overall input toughness feature samples, ensuring the understanding of the decoupling chain and improving the global decoupling capture accuracy of chaotic factor perturbation features. Furthermore, by establishing positive sample pairs based on insufficient toughness representation and negative sample pairs based on excessive toughness representation, and using a sample discriminant network based on mutual information theory to construct a mutual information discriminator that arbitrates the information correlation between the two, a discriminant boundary between different toughness feature states is created through contrastive learning. This learns the essential differences between normal and abnormal toughness evolution laws, thereby improving the ability to identify abnormal features of toughness degradation. Using the global summary matrix as index information, each local toughness feature map is input into a mutual information discriminator for probabilistic discriminant analysis. The mutual information probability value measures the probability score of whether the insufficient toughness or excessive toughness characterization originates from a certain local feature, quantifying the confidence strength between local features and abnormal toughness patterns, and achieving accurate screening and credibility assessment of chaotic perturbation features in the entire process space. If the mutual information probability value is greater than the maximized mutual information probability threshold, it indicates that the probability bias of this local toughness feature belonging to insufficient toughness or excessive toughness characterization is huge, identifying it as a toughness defect feature region caused by chaotic factor perturbation at easily deteriorated toughness anomaly points. The toughness features distorted by chaotic factor interference are separated from the original feature space, and feature compensation and reconstruction are performed based on the recovery correction vector, thereby forming compensation samples with high consistency, high separability, and high representativeness. Decoupling and reconstruction are then performed based on the recovery correction vector, realizing in-depth mining and accurate modeling of the formation mechanism of easily deteriorated toughness anomalies in bonded aluminum wires. It can construct a high-quality sample matrix with a clear compensation direction, providing a reliable basis for subsequent toughness performance optimization, process parameter adjustment, and product quality stability control.

[0032] Furthermore, S04 specifically includes the following steps: If the test compensation result shows a positive compensation sample matrix, then the annealing decision service framework of the bonded aluminum wire production line is obtained, and the full process information of the standard annealing process for processing products with unqualified toughness in the bonded aluminum wire production line is retrieved. The actual annealing process planning set for products with unqualified toughness is obtained by acquiring the full process information of the standard annealing process within a preset time period from the production control log. The process planning subset corresponding to the easily deteriorated toughness anomaly points is extracted from the actual annealing planning set and defined as a type of compensation slice item. The positive compensation sample matrix is ​​introduced into the type of compensation slice item for expansion and correction of the toughness performance dimension. To obtain the ideal toughness index of the bonded aluminum wire sample, the entire process information of the standard annealing process, the ideal toughness index, and the expanded compensation slice item are input into the annealing decision service framework for background decision-making. The output is the expected annealing process planning set for the bonded aluminum wire production line to avoid or eliminate easily deteriorated toughness anomalies by ideal standardized annealing processing. Using the full-process information of the standard annealing process as the time-series baseline, the expected annealing time-series intervals of different annealing progress steps and each step length of the annealing process at the time-series baseline are extracted through the expected annealing process planning set. The actual annealing progress status sequence and its temperature distribution data of the bonded aluminum wire production line based on the actual annealing process planning set time-series of the production process at the point of easy deterioration of toughness are obtained. Based on different annealing follow-up lengths, the actual annealing progress state sequence is divided into n effective annealing timing windows. If the actual annealing period of an effective annealing timing window is shorter than the expected annealing time interval, then the effective annealing timing window is marked as an annealing timing pullback node. The temperature rise gradient index of each annealing timing retraction node is obtained by acquiring temperature distribution data. The actual process parameters of the annealing timing retraction node corresponding to the temperature rise gradient index of the bonded aluminum wire production line being less than the preset threshold are extracted. Based on the parameter deviation between the expected annealing process planning set and the actual process parameters, the annealing process of the production line is optimized for insufficient toughness, and the first production toughness optimization scheme is obtained.

[0033] It should be noted that if the test compensation output is a positive compensation sample matrix, it indicates that the easily deteriorated toughness anomalies in the toughness-unacceptable products have insufficient toughness attributes. The main manifestation of insufficient toughness in bonded aluminum wire is embrittlement. Once embrittlement occurs, the plasticity of the aluminum wire decreases, and it cannot undergo sufficient plastic deformation under stress. This makes it extremely easy for cracks to appear near the bonding points during the packaging process, or even for it to break directly. Alternatively, the probability of failure increases after high-temperature aging, which can easily lead to a decrease in solder joint reliability and premature wire breakage. The main reason for the toughness embrittlement of bonded aluminum wire is improper annealing process, which manifests as a decrease in annealing temperature and a shortening of annealing time. To address this, this method expands and corrects the annealing process planning of the entire standard annealing process of the bonded aluminum wire production line by using a positive compensation sample matrix. This yields the expected annealing process planning under the premise of avoiding chaotic factor disturbances during the production of bonded aluminum wire products with unacceptable toughness, which can effectively eliminate the probability of easily deteriorated toughness anomalies. Furthermore, based on the expected annealing process planning set, the annealing follow-up length and its corresponding expected annealing time interval are obtained with the entire standard annealing process as the time reference. This expected annealing time interval clarifies the idealized advancement rhythm of different annealing processes targeting the easily deteriorated toughness anomaly point. For example, the annealing temperature increases from 550℃ to 600℃ after an expected time interval, and then regresses after another expected time interval, forming an annealing rhythm time sequence chain targeting the easily deteriorated toughness anomaly point. In addition, each annealing follow-up length defines the ideal jump timing of the annealing production speed. Therefore, based on the annealing follow-up length, the actual annealing progress state sequence is divided into several effective annealing timing windows, anchoring a reasonable annealing process intervention time transition benchmark framework.

[0034] If the actual annealing period within the effective annealing window is shorter than the expected annealing sequence, it indicates that the annealing time corresponding to the annealing progress state occurring within the effective annealing window is significantly lower than the expected annealing sequence. This suggests insufficient, shortened, or reverted annealing time, which can lead to insufficient elimination of work hardening during subsequent aluminum wire cold drawing, resulting in high residual stress within the aluminum wire, reduced elongation, and consequently, decreased toughness. Therefore, the abnormal annealing progress node for products with unacceptable toughness in the bonded aluminum wire production line can be identified, i.e., the annealing timing reversal node. Ultimately, based on the temperature rise gradient index at the easily deteriorated toughness abnormal node, the process parameters leading to insufficient annealing time and subsequent toughness embrittlement are locked at the annealing timing reversal node. The expected annealing process planning set is used as the ideal annealing criterion for optimization, effectively improving the internal grain structure and stress distribution of the bonded aluminum wire. This avoids abnormal grain growth, structural embrittlement, and mechanical property degradation caused by excessively low annealing temperature or unreasonable annealing timing, achieving the effect of suppressing and repairing insufficient toughness in bonded aluminum wire production, reducing product performance dispersion, and improving batch consistency and finished product qualification rate.

[0035] Furthermore, S05 specifically includes the following steps: If the test compensation result shows a negative compensation sample matrix, then retrieve the full process information of the standard wire drawing process for products with unqualified toughness from the bonded aluminum wire production line. By processing the actual wire drawing process planning set of products with unqualified toughness through the full process information of the standard wire drawing process, the process planning subset corresponding to the easily deteriorated toughness anomaly point is extracted and defined as the second type of compensation slice item. The negative compensation sample matrix is ​​introduced into the second type of compensation slice item for the expansion and correction of the toughness performance dimension. Based on the full process information of the standard wire drawing process, ideal toughness index and expanded second-class compensation slice items, a decision is made to output the expected wire drawing process planning set for the ideal standardized wire drawing process of the bonded aluminum wire production line to eliminate easily deteriorated toughness anomalies. Based on the concept of spatiotemporal alignment and combined with hashing algorithms, the spatiotemporal hash degree of the wire drawing operations and timing of the actual wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the first spatiotemporal hash degree; the spatiotemporal hash degree of the wire drawing operations and timing of the expected wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the second spatiotemporal hash degree. The strain distribution pattern of easily deteriorated toughness anomalies is obtained by obtaining the production control log. The strain values ​​of different drawing distribution blocks are obtained based on the strain distribution pattern. The hash deviation of the first spatiotemporal hash degree compared with the second spatiotemporal hash degree is calculated to obtain the hash deviation value. If the strain value is greater than the tolerance strain threshold of the ideal toughness index, the drawing distribution block is marked as an abnormal local drawing block. At this time, the strain value is compensated and corrected based on the hash deviation value to obtain the correct strain value of the abnormal local drawing block. Based on big data, a positive correlation mechanism model of the wire drawing process state and strain force change is obtained. The correct strain force value is simulated through the positive correlation mechanism model to obtain the current single-pass diameter reduction rate of easily deteriorated toughness anomaly points. If the current single-pass reduction rate reaches the ideal single-pass reduction rate threshold, then the ideal wire drawing process parameters with the correct strain value are output. Based on the parameter deviation between the ideal wire drawing process parameters and the actual wire drawing process parameters, the wire drawing process of the production line is optimized for toughness overload, and a second production toughness optimization scheme is obtained.

[0036] It should be noted that if the test compensation output is a negative compensation sample matrix, it proves that the easily deteriorated toughness anomalies in the product with unqualified toughness exhibit excessive toughness. The bonded aluminum wire is in a softened state, with excessive plasticity and insufficient strength. Under external force, it is prone to excessive deformation, causing unstable bonding area morphology, excessively deep indentations, or localized necking of the aluminum wire. In severe cases, this can lead to dimensional deviations in the bonding structure and an increased risk of fatigue damage during subsequent service. It can also easily reduce the mechanical support capacity of the bonded aluminum wire, affecting the stability of the wire arc and the reliability of the package. An unreasonable wire drawing process can cause the bonded aluminum wire to soften, typically manifested as a reduced single-pass reduction rate and insufficient drawing strain, a decreased processing cherry blossom effect, and decreased aluminum wire strength while increasing plasticity. To address this, this method expands and corrects the wire drawing process planning of the entire standard wire drawing process of the bonded aluminum wire production line using a negative compensation sample matrix, obtaining a desired wire drawing process planning set that effectively eliminates the probability of easily deteriorated toughness anomalies. Since the actual wire drawing process planning set is the actual drift process of the standard wire drawing process running on the bonding aluminum wire production line to achieve the ideal toughness index, while the desired wire drawing process planning set is the accurate target process of the bonding aluminum wire production line executing the standard wire drawing process under the constraint of the ideal toughness index, it can be seen that the optimization of the wire drawing process is essentially to correct the wire drawing operation and wire drawing sequence of the actual wire drawing process planning to be equal to or infinitely close to the desired wire drawing process planning set. Therefore, by calculating the spatiotemporal hash degree of the two relative to the standard wire drawing process, the actual-desired process drift is quantified. Then, based on the hash degree offset of the first spatiotemporal hash degree relative to the second spatiotemporal hash degree, the compensation scale of this process drift is quantified, and a process optimization correction term for the wire drawing production of easily deteriorated toughness anomalies is constructed.

[0037] By acquiring the strain distribution data of bonded aluminum wire samples, and considering that drawing errors can lead to insufficient strain, the method determines whether the strain value exceeds the tolerance strain threshold of the ideal toughness index. If it does, it indicates that the local strain at the point of abnormal toughness degradation exceeds the allowable strain threshold of the ideal toughness index, representing strain expansion or contraction caused by local drawing. This is identified as an abnormal local drawing section, and the correctness of the strain is corrected using hash deviation values. The drawing process of bonded aluminum wire is positively correlated with strain changes. For example, continuous drawing of 5mm may increase (decrease) the strain by 1N. Therefore, this method uses a positive correlation mechanism model to simulate the correct strain value to obtain ideal drawing process parameters. During the process, the single-pass diameter reduction rate is used as a constraint index to ensure the rationality of the process parameters and avoid the probability of repeated decreases in the single-pass diameter reduction rate during subsequent bonded aluminum wire drawing. This method can effectively improve the internal microstructure and mechanical property distribution of aluminum wire during the drawing process, improve the strength and toughness matching level of aluminum wire, and ensure the structural stability of the bonding interface and the consistency of the weld points.

[0038] The second aspect of this invention provides a toughness optimization system for bonded aluminum wire production based on industrial-grade testing, such as... Figure 3 As shown, the system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 includes a program for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing. The communication interface 303 is used for data connection and communication between the memory 301 and the processor 302. When the program for optimizing the toughness of bonded aluminum wire production is executed by the processor 302, it implements any of the steps of the method described above.

[0039] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included 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 method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing, characterized in that, Includes the following steps: S01: Quantitatively extract bonded aluminum wire samples for industrial-grade toughness index mechanical and physical testing and semiconductor packaging conductivity testing, obtain physical toughness test parameters and conductivity current test signals, locate suspicious toughness performance test sites with abnormal transient currents based on wavelet packet energy spectrum analysis of conductivity current test signals, test the mechanical toughness of suspicious toughness performance test sites based on physical toughness test parameters, and obtain the toughness-unqualified products and easily deteriorated toughness anomalies of bonded aluminum wire samples. S02: Obtain the actual toughness processing condition when the bonding aluminum wire production line processes the abnormal point of easily deteriorated toughness. Calculate the disturbed deviation mode of the production line by dynamically tracking the attraction deviation of the actual toughness processing condition under the chaos of actual production. Based on the disturbed deviation mode, dynamically correct different chaotic factors in the low-stability test sample and generate a recovery correction vector. S03: Construct a multimodal toughness feature sequence space when the bonded aluminum wire production line processes products with unqualified toughness. In the multimodal toughness feature sequence space, decouple the local features of easily deteriorated toughness anomalies with respect to insufficient toughness and excessive toughness. Reconstruct the decoupled local features according to the recovery correction vector to generate a positive compensation sample matrix or a negative compensation sample matrix. S04: If the test compensation result output is a positive compensation sample matrix, then the actual annealing process is expanded and corrected based on the positive compensation sample matrix. The unreasonable actual annealing progress timing and annealing temperature of easily deteriorated toughness abnormal points are located and analyzed to optimize the annealing process of the production line. S 05: If the test compensation result output is a negative compensation sample matrix, then the expected expansion and correction of the actual wire drawing process is performed based on the negative compensation sample matrix. The abnormal wire drawing strain of the production line processing point that is prone to deterioration and toughness abnormality is analyzed, and the process optimization of single-pass diameter reduction rate balance is carried out.

2. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 1, characterized in that, S01 specifically includes the following steps: By quantitatively extracting bonded aluminum wire samples processed on the bonded aluminum wire production line, the extracted bonded aluminum wire samples are subjected to mechanical and physical tests of industrial-grade toughness indicators and packaging conductivity tests of power semiconductors, and the physical toughness test parameters and conductivity test signals of the bonded aluminum wire test samples within a predetermined time period are obtained. The preset test strategy input during power semiconductor package conductivity testing is obtained. The circuit modulation reference law followed by the package conductivity test is obtained by retrieving the preset test strategy through the semiconductor package knowledge graph. Based on the circuit modulation reference law, a spectrum detection framework for the current test signal is constructed. Based on big data, the prior current signal characteristics of the semiconductor conductivity abnormality caused by the non-compliance of the toughness of the bonded aluminum wire are obtained. The conductivity current test signal is fitted to the spectrum detection framework to generate the time-series current flux spectrum curve. Based on the prior current signal characteristics, the current flux frequency band is reconstructed by wavelet packet decomposition of the time-series current flux spectrum curve to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of non-compliance of the toughness of the bonded aluminum wire sample. According to the preset test strategy, the bonded aluminum wire sample is divided into k sub-test segments. The transient current singular inflection point and its conductance energy kurtosis corresponding to each sub-test segment on the time-series current flux spectrum curve are obtained by wavelet packet energy spectrum identification. At the same time, the conductance energy kurtosis range of different prior current signal characteristics is obtained based on big data. If the conductivity energy kurtosis is within any conductivity energy kurtosis interval, then the sub-test segment where the singular inflection point of the transient current is located is designated as the test site for the suspected toughness performance of the bonded aluminum wire sample. Based on the regional performance distribution of physical toughness test parameters, a physical toughness thermodynamic model diagram of the bonded aluminum wire sample with respect to each sub-test segment is constructed. The physical toughness thermodynamic colorimetric value corresponding to the suspected toughness performance test position is extracted through the physical toughness thermodynamic model diagram. Based on the production index requirements of bonded aluminum wire, a pre-defined expected thermochromic gamut is set for the physical toughness frequency. If the physical toughness thermochromic value cannot be found in the expected thermochromic gamut, the bonded aluminum wire sample is marked as a product with unqualified toughness, and this suspected toughness performance test position is marked as an abnormal point of easily deteriorated toughness.

3. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 2, characterized in that, The method involves obtaining prior current signal characteristics based on big data to identify semiconductor conductivity abnormalities caused by substandard toughness of bonded aluminum wires. The conductivity current test signal is then fitted to a spectrum detection framework to generate a time-series current flux spectrum curve. Based on the prior current signal characteristics, the time-series current flux spectrum curve is reconstructed using wavelet packet decomposition to generate the wavelet packet energy spectrum of the conductivity current test signal under the premise of substandard toughness in the bonded aluminum wire sample. Specifically, this includes the following steps: Based on big data, different prior current signal characteristics of power semiconductor packaging caused by unqualified toughness in the production of bonding aluminum wire are obtained. The conductivity test signal is time-fitted in the spectrum monitoring framework to form the time-series current flux spectrum curve of bonding aluminum wire packaging conductivity test. A wavelet basis transform algorithm is introduced to set the wavelet packet basis function of the abnormal current characteristics of power semiconductor based on the prior current signal characteristics. Based on the wavelet packet basis function, the signal characteristics of the time-series current flux spectrum curve are decomposed by wavelet packet, and the multiple frequency band link nodes of the current flux on the bonded aluminum wire sample and their corresponding frequency band node coefficients are output. Based on the frequency band node coefficients, the corresponding frequency band signals of each frequency band link node on the time-series current flux spectrum curve are reconstructed. During the reconstruction process, the flux sampling energy of each frequency band signal is calculated to obtain the new frequency band energy index. All new frequency band energy indices are summed and the energy ratio is calculated to generate the wavelet packet energy spectrum of the conduction current test signal under the premise that the toughness of the bonded aluminum wire sample is unqualified.

4. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 1, characterized in that, S02 specifically includes the following steps: The production timestamps of the gradually appearing easily deteriorating toughness anomalies when the bonding aluminum wire production line processes products with unqualified toughness are defined as the toughness degradation source production timestamps. By retrieving the production control logs of bonded aluminum wire processing, the actual toughness processing conditions of the bonded aluminum wire production line at the time stamp of the toughness degradation source processing point were obtained from the output of different multimodal process data. The distribution of state points affected by different types of chaotic factors was also retrieved from the actual toughness processing conditions of the bonded aluminum wire production line. Among them, chaotic factors include equipment fluctuations, noise offset, system errors, batch differences and environmental disturbances. The underlying processing control architecture, coupled dynamic logic of multimodal processes, and chaotic factor characteristics of the bonded aluminum wire production line are obtained. The process production phase space domain of bonded aluminum wire is established by combining the underlying processing control architecture, coupled dynamic logic, and chaotic factor characteristics. Based on big data and experience processing cases, several historical steady-state operation trajectory templates were obtained for the actual toughness processing conditions of the bonded aluminum wire production line after processing easily deteriorated toughness anomalies under the premise of suppressing or rejecting various types of chaotic factors. The baseline trajectory convergence simulation of the ideal toughness processing attraction under the condition of maintaining low chaos in the initial state of the bonded aluminum wire production line was carried out using the historical steady-state operation trajectory templates, and the target attraction domain boundary model was generated. Based on the distribution of state points, the mapping is injected into the process production phase space to dynamically track the actual toughness processing conditions. Based on the tracking results, the real-time attractor trajectory of the actual toughness processing conditions caused by the row bonding aluminum wire production line being disturbed by chaotic factors is determined. Calculate the attraction deviation index of the real-time attractor trajectory from the boundary model of the target attraction domain, and identify and analyze the disturbance deviation mode of chaotic factors in the bonding aluminum wire processing production based on the attraction deviation index. By retrieving data and experience-based case studies, the structural deviation direction and correction gain scale that induce the deterioration of the toughness of bonded aluminum wire production due to the perturbation deviation mode are obtained. Based on the structural deviation direction and correction gain scale, the perturbation of the chaotic factor is dynamically corrected, and a recovery correction vector is generated and superimposed on the control system of the bonded aluminum wire production line.

5. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 1, characterized in that, S03 specifically includes the following steps: By obtaining the multimodal toughness features of the bonded aluminum wire production line at different process stages during the processing of products with substandard toughness from the production control logs, based on the output of the detection images, a multimodal toughness feature sequence space is constructed. A depth encoder is introduced, and a mutual information attention network is embedded in the intermediate layer of the depth encoder. After embedding, the intermediate layer is used to extract features from the multimodal toughness feature sequence space to obtain the local toughness feature map of the toughness-deficient product in each process stage of the sequential processing. Obtain the production jump response boundary points between the time sequences of each process stage, preset the handover and convergence function according to the process echelon of the production jump response boundary points, and perform aggregation operation on the local toughness feature maps of all process stage time sequences based on the handover and convergence function to generate a global summary matrix of the multimodal toughness feature sequence space. The actual process parameters and real-time toughness parameters of easily deteriorated toughness anomalies in the bonding aluminum wire production line at the process stage are obtained. Based on big data and experience processing cases, the actual process parameters, real-time toughness parameters and chaotic factor characteristics are retrieved to obtain the toughness insufficiency and toughness over-characterization of the real-time toughness process corresponding to the easily deteriorated toughness anomalies caused by different chaotic factors in the bonding aluminum wire production line. A sample discrimination network based on mutual information theory is introduced. Positive sample pairs with local resilience features are constructed based on the resilience deficiency characterization, and negative sample pairs with local resilience features are constructed based on the resilience over-characterization. The positive sample pairs and negative samples are combined in the sample discrimination network to perform feature representation modeling and generate a mutual information discriminator. Based on the global summary matrix as index information, the local resilience feature map is imported into the mutual information discriminator for sample probability discrimination and analysis, so as to obtain the mutual information probability value of different local resilience features belonging to positive sample pairs or negative sample pairs. If the mutual information probability value is greater than the maximum mutual information probability threshold, then this local resilience feature is labeled as a chaotic resilience feature and local decoupling is performed. The chaotic resilience feature after local decoupling is reconstructed and modeled according to the recovery correction vector to obtain a positive compensation sample matrix or a negative compensation sample matrix.

6. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 1, characterized in that, S04 specifically includes the following steps: If the test compensation result shows a positive compensation sample matrix, then the annealing decision service framework of the bonded aluminum wire production line is obtained, and the full process information of the standard annealing process for processing products with unqualified toughness in the bonded aluminum wire production line is retrieved. The actual annealing process planning set for products with unqualified toughness is obtained by acquiring the full process information of the standard annealing process within a preset time period from the production control log. The process planning subset corresponding to the easily deteriorated toughness anomaly points is extracted from the actual annealing planning set and defined as a type of compensation slice item. The positive compensation sample matrix is ​​introduced into the type of compensation slice item for expansion and correction of the toughness performance dimension. To obtain the ideal toughness index of the bonded aluminum wire sample, the entire process information of the standard annealing process, the ideal toughness index, and the expanded compensation slice item are input into the annealing decision service framework for background decision-making. The output is the expected annealing process planning set for the bonded aluminum wire production line to avoid or eliminate easily deteriorated toughness anomalies by ideal standardized annealing processing. Using the full-process information of the standard annealing process as the time-series baseline, the expected annealing time-series intervals of different annealing progress steps and each step length of the annealing process at the time-series baseline are extracted through the expected annealing process planning set. The actual annealing progress status sequence and its temperature distribution data of the bonded aluminum wire production line based on the actual annealing process planning set time-series of the production process at the point of easy deterioration of toughness are obtained. Based on different annealing follow-up lengths, the actual annealing progress state sequence is divided into n effective annealing timing windows. If the actual annealing period of an effective annealing timing window is shorter than the expected annealing time interval, then the effective annealing timing window is marked as an annealing timing pullback node. The temperature rise gradient index of each annealing timing retraction node is obtained by acquiring temperature distribution data. The actual process parameters of the annealing timing retraction node corresponding to the temperature rise gradient index of the bonded aluminum wire production line being less than the preset threshold are extracted. Based on the parameter deviation between the expected annealing process planning set and the actual process parameters, the annealing process of the production line is optimized for insufficient toughness, and the first production toughness optimization scheme is obtained.

7. The method for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing according to claim 1, characterized in that, S05 specifically includes the following steps: If the test compensation result shows a negative compensation sample matrix, then retrieve the full process information of the standard wire drawing process for products with unqualified toughness from the bonded aluminum wire production line. By processing the actual wire drawing process planning set of products with unqualified toughness through the full process information of the standard wire drawing process, the process planning subset corresponding to the easily deteriorated toughness anomaly point is extracted and defined as the second type of compensation slice item. The negative compensation sample matrix is ​​introduced into the second type of compensation slice item for the expansion and correction of the toughness performance dimension. Based on the full process information of the standard wire drawing process, ideal toughness index and expanded second-class compensation slice items, a decision is made to output the expected wire drawing process planning set for the ideal standardized wire drawing process of the bonded aluminum wire production line to eliminate easily deteriorated toughness anomalies. Based on the concept of spatiotemporal alignment and combined with hashing algorithms, the spatiotemporal hash degree of the wire drawing operations and timing of the actual wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the first spatiotemporal hash degree; the spatiotemporal hash degree of the wire drawing operations and timing of the expected wire drawing process planning set relative to the full process information of the standard wire drawing process is calculated and defined as the second spatiotemporal hash degree. The strain distribution pattern of easily deteriorated toughness anomalies is obtained by obtaining the production control log. The strain values ​​of different drawing distribution blocks are obtained based on the strain distribution pattern. The hash deviation of the first spatiotemporal hash degree compared with the second spatiotemporal hash degree is calculated to obtain the hash deviation value. If the strain value is greater than the tolerance strain threshold of the ideal toughness index, the drawing distribution block is marked as an abnormal local drawing block. At this time, the strain value is compensated and corrected based on the hash deviation value to obtain the correct strain value of the abnormal local drawing block. Based on big data, a positive correlation mechanism model of the wire drawing process state and strain force change is obtained. The correct strain force value is simulated through the positive correlation mechanism model to obtain the current single-pass diameter reduction rate of easily deteriorated toughness anomaly points. If the current single-pass reduction rate reaches the ideal single-pass reduction rate threshold, then the ideal wire drawing process parameters with the correct strain value are output. Based on the parameter deviation between the ideal wire drawing process parameters and the actual wire drawing process parameters, the wire drawing process of the production line is optimized for toughness overload, and a second production toughness optimization scheme is obtained.

8. A toughness optimization system for bonded aluminum wire production based on industrial-grade testing, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory includes a program for optimizing the toughness of bonded aluminum wire production based on industrial-grade testing. The communication interface is used for data connection communication between the memory and the processor. When the program for optimizing the toughness of bonded aluminum wire production is executed by the processor, it implements the steps of the method for optimizing the toughness of bonded aluminum wire production as described in any one of claims 1-7.

Images