Artificial intelligence-based gold bonding wire fracture risk detection method and system

Abstract

The invention discloses a gold bonding wire fracture risk detection method and system based on artificial intelligence, and relates to the technical field of gold bonding wire fracture detection.The gold bonding wire fracture risk detection method comprises the steps that multi-mode signals generated in the bonding process are collected, and a dynamic correlation model between the multi-mode signals and fracture risks is established in combination with technological parameters; extracting local abnormal features in the multi-modal signal, and performing priority adjustment on the importance of the local abnormal features based on a dynamic correlation model; and classifying the local abnormal features after priority adjustment, identifying a fracture risk signal based on a dynamic correlation model, and generating a real-time detection result. According to the method, closed-loop optimization is formed through multi-modal signal acquisition, dynamic correlation model construction, local abnormal feature priority adjustment and fracture risk signal accurate identification and positioning, and the precision, real-time performance and industrial adaptability of gold bonding wire fracture risk detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gold bonding wire fracture detection, and in particular to a gold bonding wire fracture risk detection method and system based on artificial intelligence. Background Art

[0002] With the rapid development of microelectronic packaging technology, gold bonding wires, as a key interconnect material, have been widely used in semiconductor device manufacturing. Gold bonding wires establish electrical connections between chips and lead frames through bonding equipment, and their quality and stability are directly related to the electrical performance and service life of the chip. However, gold bonding wires are very prone to breakage under high-frequency vibration, thermal stress, and complex process parameters, resulting in circuit connection failure or even loss of function of the entire device. To ensure the quality of bonding, traditional technologies usually use a single signal monitoring method, such as detecting mechanical stress anomalies through vibration signals, or detecting material property changes through acoustic signals. However, these methods can only analyze a single feature and cannot fully reflect the multi-dimensional dynamic changes in the bonding process. Especially under complex working conditions, signal coupling phenomena can easily lead to misjudgment or missed judgment. In addition, the correlation between process parameters and signal features has not been fully explored in existing technologies, resulting in the accuracy and real-time performance of risk prediction being difficult to meet industrial needs.

[0003] In recent years, the application of artificial intelligence and multimodal data analysis technology in the field of intelligent manufacturing has continued to deepen, providing new solutions for improving the accuracy of risk detection of bonded wire fracture. For example, signal classification technology based on machine learning algorithms has been used to analyze complex signal characteristics, but existing methods mostly focus on the classification and pattern recognition of static signals and cannot effectively cope with the dynamic changes of signals in the time dimension. In addition, some technologies simply combine vibration signals with process parameters to establish risk prediction models, but lack linkage analysis and priority optimization for multimodal signal characteristics, resulting in poor robustness of the model and inefficient resource utilization. More importantly, existing technologies are usually unable to accurately locate the source of abnormal signals or provide targeted process adjustment suggestions after risk signal identification, resulting in a lack of guidance and practical application value in the detection results. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the present invention provides a bonding wire breakage risk detection method and system based on artificial intelligence, which can solve the problems mentioned in the background technology.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for detecting the risk of bonded gold wire fracture based on artificial intelligence, comprising: collecting multimodal signals generated during the bonding process, and establishing a dynamic correlation model between the multimodal signals and the fracture risk in combination with process parameters; the multimodal signals include vibration signals and acoustic signals; the process parameters include bonding pressure, bonding temperature and vibration frequency; extracting local abnormal features in the multimodal signals, and adjusting the priority of the importance of the local abnormal features based on the dynamic correlation model; classifying the local abnormal features after priority adjustment, identifying the fracture risk signal based on the dynamic correlation model, and generating real-time detection results.

[0007] As a preferred solution of the artificial intelligence-based bonding wire fracture risk detection method described in the present invention, wherein: a dynamic association model between the multimodal signal and the fracture risk is established in combination with the process parameters, including the following steps: determining the multimodal signal characteristics based on the multimodal signal, and generating a relationship data model between the multimodal signal characteristics and the process parameters by associating the multimodal signal characteristics with the process parameters; based on the relationship data model between the multimodal signal characteristics and the process parameters, time-calibrating the multimodal signal characteristics and the process parameters through a time synchronization mechanism to generate a time correspondence between the multimodal signal characteristics and the process parameters; based on the time correspondence between the multimodal signal characteristics and the process parameters, using a regression analysis method to generate a characterization model for fracture risk, and constructing a dynamic association model between fracture risk, multimodal signal characteristics and process parameters based on the characterization model for fracture risk.

[0008] As a preferred scheme of the artificial intelligence-based bonding wire fracture risk detection method described in the present invention, wherein: the multimodal signal characteristics include the amplitude of the vibration signal, the spectral distribution characteristics of the vibration signal and the energy change of the acoustic signal; the relationship data model between the multimodal signal characteristics and the process parameters includes the relationship data model between the bonding pressure and the amplitude of the vibration signal, the relationship data model between the bonding temperature and the energy change of the acoustic signal, and the relationship data model between the vibration frequency and the spectral distribution characteristics of the vibration signal; the characterization model of the fracture risk constructs a dynamic correlation model between the fracture risk, the multimodal signal characteristics and the process parameters by mapping the time correspondence between the multimodal signal characteristics and the process parameters to the risk factor space.

[0009] The characterization model of the fracture risk is expressed as:

[0010]

[0011] Where R(t) is the fracture risk; F i (t) is the i-th multimodal signal feature; Pj (t) is the jth process parameter; β0 is the bias term, which represents the basic value of the fracture risk; β i and γ j is the regression coefficient of multimodal signal characteristics and process parameters; ε(t) is the random error term.

[0012] Combine R(t) with multimodal signal features and process parameters to form a dynamic feature vector V R (t), with the dynamic feature vector V R (t) is taken as input, and the dynamic correlation model M is obtained through mapping to predict the fracture risk value R(t+1) at the next moment, or to analyze the dynamic change trend of fracture risk.

[0013] As a preferred scheme of the artificial intelligence-based bonding wire breakage risk detection method described in the present invention, wherein: the local abnormal characteristics in the multimodal signal include local abnormal characteristics of the vibration signal and the acoustic signal; the local abnormal characteristics of the vibration signal include amplitude peak value and spectrum distribution abnormal points; the local abnormal characteristics of the acoustic signal include frequency drift and high-frequency energy changes.

[0014] The priority of the importance of the local abnormal features is adjusted based on the dynamic association model, including the following steps: based on the dynamic association model, the extracted local abnormal features F ai Calculate the feature importance and assign each local abnormal feature F ai Associated with process parameters:

[0015] I(F ai )=W(F ai );

[0016] W(F ai )=G(F ai )·W(F i );

[0017] Among them, I(F ai ) is the local abnormal feature F ai The importance of W(F ai ) is the weight of the local abnormal feature; G(F ai ) is the local significance correction factor; W(F i ) is F i The weight of (t).

[0018] According to I(F ai ) prioritizes local abnormal features: when I(F ai ) is higher than the first preset threshold T1, it is a high priority; when the second preset threshold T2 is less than or equal to I(F ai ), and I(F ai) is less than or equal to the first preset threshold T1, it is medium priority; when I(F ai ) is lower than the second preset threshold T2, it is a low priority.

[0019] As a preferred solution of the artificial intelligence-based bonding wire breakage risk detection method described in the present invention, wherein: an optimized local abnormal feature sequence is generated according to the priority division result, and the collection strategy of the multimodal signal related to the local abnormal feature is dynamically adjusted: if the proportion of high-priority local abnormal features exceeds the third preset threshold R1, the collection frequency of the process parameters related to the high-priority features is increased, and the collection interval time is shortened; if the proportion of high-priority local abnormal features is between the fourth preset threshold R2 and the third preset threshold R1, the current process parameter collection frequency is maintained, and the timing information of the medium priority features is preferentially recorded; if the proportion of high-priority local abnormal features is lower than the fourth preset threshold R2, the process parameter collection frequency related to the low-priority features is reduced, and the basic information of the low-priority features is recorded at the same time.

[0020] As a preferred solution of the artificial intelligence-based bonding wire fracture risk detection method described in the present invention, wherein: the local abnormal features after priority adjustment are classified, and the fracture risk signal is identified based on the dynamic association model, including the following steps: using a density-based clustering algorithm to classify the local abnormal features after priority adjustment to generate a feature classification result; the feature classification result includes a high-risk category and a conventional category. If the boundary of a local abnormal feature category in the feature classification result exceeds the statistical range of the dynamic association model, it is marked as a high-risk category, and the timing information and associated parameters are recorded preferentially; the fracture risk signal of the classified local abnormal features is identified based on the dynamic association model; the high-risk category and conventional category features in the classification result are respectively input into the dynamic association model, and the fracture risk R(t) corresponding to each category is calculated in combination with the process parameters; for the local abnormal features belonging to the high-risk category, if R(t) is greater than the fifth preset threshold T R , it is marked as a fracture risk signal, and its dynamic feature attributes and associated process parameters are recorded; for local abnormal features belonging to the conventional category, if R(t) is greater than the fifth preset threshold T R , it is marked as a potential fracture risk signal; if R(t) is less than or equal to the fifth preset threshold T R , it is marked as a normal signal.

[0021] As a preferred solution of the artificial intelligence-based gold bonding wire fracture risk detection method of the present invention, wherein: the abnormal signal source is located for the identified fracture risk signal; combined with W(F i ) and W(F ai), determine the main local abnormal features of the fracture risk signal, and generate positioning results. Specifically, if the fracture risk signal is associated with multiple local abnormal features, then according to W(F ai ) and the correlation between process parameters, from I(F ai ) the highest local abnormal feature, and select the signal source in turn, and record the corresponding process parameter position and state; if the I(F ai ) ratio exceeds the preset value, the positioning information of the corresponding signal source is directly output, and the associated signal source positioning information is output; the positioning result and the process parameter change trend are combined to generate process parameter suggestions for adjustment, and the positioning result and the process parameter suggestions for adjustment are output as the detection result.

[0022] To further solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based bonding wire fracture risk detection system, comprising: a data processing module, used to collect multimodal signals generated during the bonding process, and establish a dynamic correlation model between the multimodal signals and fracture risks in combination with process parameters; a feature optimization module, used to extract local abnormal features in the multimodal signals, and adjust the priority of the importance of the local abnormal features based on the dynamic correlation model; a risk assessment module, used to classify the local abnormal features after priority adjustment, identify fracture risk signals based on the dynamic correlation model, and generate real-time detection results.

[0023] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the bonding wire breakage risk detection method based on artificial intelligence as described above are implemented.

[0024] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the above-mentioned method for detecting the risk of gold bonding wire breakage based on artificial intelligence are implemented.

[0025] Beneficial effects of the present invention: The present invention collects multimodal signals and establishes a dynamic correlation model in combination with process parameters, thereby realizing the quantitative correlation between multi-dimensional signal features and process parameters, and comprehensively improving the adaptability and accuracy of signal analysis; by extracting local abnormal features in multimodal signals, combining the dynamic correlation model to calculate feature importance and adjust priorities, the feature sequence is optimized and resource allocation and analysis efficiency are dynamically balanced, thereby focusing on key fracture risk features; in the fracture risk signal identification stage, a density clustering algorithm is used to classify the local abnormal features after priority adjustment, and the dynamic correlation model is combined to achieve accurate identification of fracture risk signals and location of abnormal signal sources, while providing real-time process parameter adjustment suggestions, forming a closed-loop optimization from signal acquisition to risk assessment. The overall solution solves the problems of insufficient multimodal data correlation, poor real-time performance, and inaccurate signal positioning in the prior art, greatly improving the accuracy, real-time performance, and industrial adaptability of gold bonding wire fracture risk detection, and providing innovative solutions for multimodal signal analysis, intelligent process optimization, and risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0027] Figure 1 This is a schematic diagram of the overall process of a method for detecting the risk of gold bonding wire fracture based on artificial intelligence proposed by the present invention;

[0028] Figure 2 A diagram of computer equipment in an artificial intelligence-based bonding wire breakage risk detection method proposed in the present invention. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a bonding wire breakage risk detection method based on artificial intelligence.

[0032] In related technologies, the monitoring method based on a single signal cannot fully reflect the multi-dimensional dynamic changes in the bonding process, which can easily lead to misjudgment or missed judgment; the correlation between process parameters and signal characteristics has not been fully explored, affecting the accuracy and real-time performance of risk prediction; the linkage analysis and feature priority optimization of multimodal signals are insufficient, resulting in poor model robustness and low resource utilization efficiency; in addition, the existing technology makes it difficult to accurately locate the source of abnormal signals and provide effective process adjustment suggestions, making the practicality and guidance of the detection results less practical.

[0033] The present application provides an effective solution to the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the artificial intelligence-based bonding wire breakage risk detection method.

[0034] Figure 1 The overall flow chart of a method for detecting the risk of gold bonding wire fracture based on artificial intelligence is shown, including:

[0035] S1: Collect multimodal signals generated during the bonding process and establish a dynamic correlation model between multimodal signals and fracture risk in combination with process parameters;

[0036] S2: Extract local abnormal features from multimodal signals and adjust the priority of the importance of local abnormal features based on the dynamic association model;

[0037] S3: Classify the local abnormal features after priority adjustment, identify fracture risk signals based on the dynamic association model, and generate real-time detection results.

[0038] Next, this embodiment will elaborate on S1 to S3 one by one:

[0039] S1: Collect the multimodal signals generated during the bonding process and establish a dynamic correlation model between the multimodal signals and the fracture risk in combination with the process parameters.

[0040] S1.1: Collect multimodal signals generated during the bonding process.

[0041] In industrial sites, the operation of bonding equipment is usually accompanied by strong mechanical vibration and background noise. Therefore, it is necessary to collect signals in real time through highly sensitive vibration sensors and acoustic sensors to overcome the interference of the on-site environment and obtain high-quality original signal data. In the present invention, the original signal data includes vibration signals and acoustic signals, which are called multimodal signals. Collecting multimodal signals can not only improve the sensitivity of fracture risk detection, but also provide information at different physical levels (such as mechanical properties and material stress changes), laying the foundation for the establishment and analysis of subsequent dynamic correlation models.

[0042] Vibration signals and acoustic signals are important physical indicators that reflect the dynamic behavior of bonding and changes in the internal structure of the gold wire. The vibration signal is generated by the mechanical vibration of the bonding equipment, which directly reflects the dynamic characteristics of the contact surface between the bonding head and the gold wire. For example, the risk of gold wire breakage may increase due to abnormal vibration; while the acoustic signal is composed of the sound wave signal caused by the generation of microcracks or changes in internal stress distribution when the gold wire material is subjected to high-frequency energy input, and is particularly sensitive to capturing transient characteristics before fracture. In order to ensure high-quality signal acquisition, the present invention adopts a variety of high-sensitivity sensors, among which the vibration signal is collected by a piezoelectric acceleration sensor, which is arranged at a key position of the bonding equipment to ensure accurate perception of high-frequency vibrations; the acoustic signal is collected by an ultrasonic sensor or an acoustic emission sensor, which is arranged close to the bonding area to capture the tiny sound wave energy changes inside the gold wire material to the greatest extent.

[0043] During the acquisition process, a data acquisition system with a high sampling rate (such as ≥50kHz) is used to record multimodal signals in real time and synchronously, and hardware trigger technology is combined to ensure the time alignment of vibration signals and acoustic signals, providing support for subsequent data fusion. At the same time, in order to reduce the background noise interference in the industrial environment, a filtering algorithm is introduced in the acquisition process to perform real-time denoising on the signal, including the joint application of low-pass filters and band-pass filters to further enhance the clarity and reliability of the signal. In addition, the collected signals are normalized to make different physical quantities comparable in subsequent analysis, thereby providing a high-quality data foundation for the feature extraction stage. In practical applications, this process can adapt to complex industrial environments, such as strong mechanical vibrations or high noise conditions, to ensure the comprehensiveness and accuracy of multimodal signals, and thus provide effective support for subsequent fracture risk detection.

[0044] S1.2: Establish a dynamic correlation model between multimodal signals and fracture risk in combination with process parameters.

[0045] S1.2.1: Determine the multimodal signal characteristics based on the multimodal signal, and generate a relationship data model between the multimodal signal characteristics and the process parameters by associating the multimodal signal characteristics with the process parameters.

[0046] The multimodal signal features include the amplitude of the vibration signal, the spectrum distribution characteristics of the vibration signal, and the energy change of the acoustic signal. The relationship data model between the multimodal signal features and the process parameters includes the relationship data model between the bonding pressure and the amplitude of the vibration signal, the relationship data model between the bonding temperature and the energy change of the acoustic signal, and the relationship data model between the vibration frequency and the spectrum distribution characteristics of the vibration signal.

[0047] Furthermore, determining the characteristics of multimodal signals based on multimodal signals is the basic link in building a dynamic correlation model. Multimodal signals include vibration signals and acoustic signals, both of which are collected in real time by high-sensitivity sensors. The vibration signal is generated by the bonding head during mechanical vibration. Its amplitude reflects the magnitude of the vibration force, while the spectrum distribution characteristics show the distribution of vibration energy in different frequency bands, which can capture the changes in dynamic characteristics caused by bonding pressure or equipment abnormalities. The acoustic signal is composed of high-frequency sound waves in the bonding process. Its energy change characteristics are associated with the internal stress changes of the gold wire material, which can reveal potential cracks or micro-fracture characteristics.

[0048] In order to extract multimodal signal features, the vibration signal is first decomposed into frequency domain signals through fast Fourier transform, and its spectrum distribution characteristics are calculated; at the same time, the change of amplitude over time is analyzed by short-time Fourier transform to capture the abnormal characteristics of transient vibration. For acoustic signals, the distribution of acoustic energy in the time-frequency plane is analyzed by continuous wavelet transform to determine the change trend of acoustic energy at key time points. These extracted multimodal signal features include the amplitude of the vibration signal, the spectrum distribution characteristics and the energy change of the acoustic signal, which constitute a complete set of multimodal signal features.

[0049] In the present invention, the extracted multimodal signal features are associated with process parameters to generate a relational data model. First, the multimodal signal features and process parameters are preprocessed, for example, the amplitude and spectral distribution characteristics of the vibration signal are normalized to eliminate the differences between different signal magnitudes; the bonding pressure, bonding temperature and vibration frequency in the process parameters are processed by sliding average to smooth the dynamic fluctuations of the data. Subsequently, based on the preprocessed data, a statistical regression analysis method is used to establish the relationship between the multimodal signal features and the process parameters. For example, the association between the amplitude of the vibration signal and the bonding pressure is modeled by linear regression to analyze the response of the vibration dynamic characteristics to pressure; the nonlinear relationship between the high-frequency energy change of the acoustic signal and the bonding temperature is used to reveal the influence of temperature fluctuations on the internal stress of the material; the association between the spectral distribution characteristics of the vibration signal and the vibration frequency is described by the characteristic mapping method The modulation effect of the frequency parameter on the signal bandwidth. These methods based on statistical regression and feature mapping generate relationship data models between vibration signal amplitude and bonding pressure, acoustic signal energy change and bonding temperature, vibration signal spectrum distribution and vibration frequency, and constitute a relationship data model between multimodal signal characteristics and process parameters, providing quantitative basic data for subsequent time calibration and fracture risk characterization.

[0050] S1.2.2: Based on the relationship data model between multimodal signal characteristics and process parameters, the multimodal signal characteristics and process parameters are time-calibrated through a time synchronization mechanism to generate a time correspondence between the multimodal signal characteristics and the process parameters.

[0051] It should be noted that in actual industrial applications, the time correspondence between multimodal signal features and process parameters is the key to dynamic analysis. Through the time synchronization mechanism, the multimodal signal features and process parameter data are calibrated on the acquisition time axis to solve the technical problem of time alignment between multi-source data. Specifically, the multimodal signal features are matched with the measurement data of the process parameters through high-precision timestamps. For example, during the acquisition process, each vibration signal or acoustic signal data is attached with a hardware-triggered timestamp, which is synchronously associated with the timestamp recorded by the process parameters to ensure that the signal and process data are completely consistent in the time dimension.

[0052] In the time synchronization mechanism, in order to solve the time offset problem that may be caused by the difference in sampling rates of different sensors, the present invention introduces a dynamic interpolation algorithm. On the time axis, the missing time points are interpolated and estimated to compensate for the data errors caused by uneven sampling. At the same time, through the timing analysis algorithm based on dynamic time warping (DTW), the time series of multimodal signal features and process parameters are aligned to ensure that the time features of different data sources are completely aligned. In addition, in order to enhance the reliability of time calibration, the present invention also combines adaptive time window technology to select the optimal time window range in the dynamic process to capture the slight time changes between signal features and process parameters.

[0053] Through the above time synchronization and calibration process, the time correspondence between multimodal signal features and process parameters is finally generated. This time correspondence provides complete spatiotemporal dimension information for the generation of subsequent fracture risk characterization models, enabling the model to more accurately describe the dynamic change trend of fracture risk and significantly improve the adaptability and robustness of the overall detection.

[0054] S1.2.3: Based on the time correspondence between multimodal signal characteristics and process parameters, a regression analysis method is used to generate a characterization model for fracture risk. Based on the characterization model for fracture risk, a dynamic correlation model among fracture risk, multimodal signal characteristics and process parameters is constructed.

[0055] Among them, the characterization model of fracture risk is used to describe the changing trend of fracture risk and its dynamic association with multimodal signal characteristics and process parameters. The characterization model of fracture risk maps the time correspondence between multimodal signal characteristics and process parameters to the risk factor space, and constructs a dynamic association model between fracture risk, multimodal signal characteristics and process parameters.

[0056] In this embodiment, firstly, the fracture risk characterization model generates a risk function by quantifying the correlation between the fracture risk and the multimodal signal characteristics and the process parameters, so as to predict the fracture risk value in real time.

[0057] The characterization model of fracture risk is expressed as:

[0058]

[0059] Where R(t) is the fracture risk (i.e., fracture risk factor), which represents the risk value predicted at time t and is used to quantify the fracture risk under the current working conditions; F i (t) is the i-th feature in the multimodal signal feature set; P j (t) is the jth process parameter in the process parameter set; β0 is the bias term, which represents the basic value of the fracture risk; β i and γ jis the regression coefficient of the multimodal signal characteristics and process parameters, indicating their influence weight on the fracture risk; ε(t) is the random error term, indicating the noise or unknown factors not captured by the model.

[0060] Through the regression coefficient β i and γ j , quantifying the degree to which different characteristics and process parameters affect the risk of fracture. For example, when β i When γ is larger, it means that the corresponding multimodal signal features have a more significant contribution to R(t). j The smaller the value, the weaker the influence of the corresponding process parameters.

[0061] Secondly, the fracture risk characterization model is verified and optimized, and the accuracy and adaptability of the model are improved through feature selection and parameter adjustment. The feature importance ranking method is used to screen the multimodal signal features and process parameters, and the multimodal feature set and process parameter set that contribute the most to R(t) are selected. During the optimization process, regularization techniques such as ridge regression or LASSO regression are used to adjust the regression coefficient β. i and γ j Constraints are imposed to prevent overfitting and improve the generalization ability of the model. In addition, the prediction performance of the fracture risk characterization model is evaluated by cross-validation method, including the calculation of error indicators (such as mean square error MSE and absolute error MAE) to ensure the prediction accuracy and stability of the model in more complex industrial scenarios. The optimized fracture risk characterization model can not only improve the prediction ability of fracture risk, but also reflect the relative contribution of each multimodal feature to fracture risk through weight distribution.

[0062] Next, based on the fracture risk characterization model, the time correspondence between multimodal signal features and process parameters is mapped to a dynamic feature vector, and a dynamic correlation model is constructed to analyze the dynamic change trend of fracture risk in real time. Specifically, R(t) is combined with multimodal signal features and process parameters to form a dynamic feature vector V R (t):

[0063] V R (t)=[R(t),F1(t),F2(t),...,F m (t),P1(t),P2(t),...,P n (t)];

[0064] Among them, F m (t) represents the mth multimodal feature in the multimodal signal feature set; P n (t) The nth process parameter in the process parameter set.

[0065] With dynamic feature vector V R(t) is used as input, and the dynamic correlation model M is obtained through mapping to predict the fracture risk value R(t+1) at the next moment, or to analyze the dynamic change trend of fracture risk:

[0066] R(t+1)=M(V R (t));

[0067] The full expansion of this formula is:

[0068] R(t+1)=M(V R (t))=M[R(t),F1(t),F2(t),...,F m (t),P1(t),P2(t),...,P n (t)];

[0069] It should be noted that the specific form of the dynamic correlation model M can be linear regression, nonlinear regression or other complex mapping models, depending on the actual relationship between the fracture risk factor and the dynamic feature vector. For example, when the relationship between the fracture risk factor and the multimodal signal characteristics and process parameters is relatively simple and can be directly described by linear weights, a linear regression model is selected; when there is a significant nonlinear relationship between the fracture risk factor and the multimodal signal characteristics and process parameters, but the nonlinear relationship can be described by a preset function form, a nonlinear regression model is selected; when the relationship between the fracture risk factor and the multimodal signal characteristics and process parameters is very complex and difficult to describe in an explicit function form, a machine learning model such as a support vector machine or a neural network is selected; when the dynamic trend of the fracture risk factor over time is more significant and time dependence is an important factor in prediction, a time series model is selected; when the relationship between the fracture risk factor and the multimodal signal characteristics and process parameters is complex and needs to be combined with the dynamic association of the time dimension, a hybrid model can be selected.

[0070] Preferably, in step S1, the present invention solves the three core problems in the prior art through the joint modeling of multimodal signal features and process parameters, the introduction of a time calibration mechanism, and a formulated relationship model: insufficient integration of multimodal signal features and process parameters, difficulty in data time alignment, and unclear quantitative relationship of risk factors. Specifically, the present invention combines the key features of vibration signals and acoustic signals with process parameters, generates a quantitative relational data model through a statistical regression method, and significantly enhances the comprehensiveness and accuracy of risk assessment. At the same time, through dynamic interpolation and dynamic time warping methods, the problem of aligning multi-source data in the time dimension is solved to ensure spatiotemporal consistency. At the formula level, a clear regression structure is adopted to quantify the contribution of features and parameters to risk factors through regression coefficients. The above improvements significantly improve the adaptability and reliability of fracture risk analysis, and provide a scientific and effective solution for risk monitoring under complex working conditions in actual scenarios.

[0071] S2: Extract local abnormal features from multimodal signals and prioritize the importance of local abnormal features based on a dynamic association model.

[0072] S2.1: Extract local abnormal features of vibration signals and acoustic signals.

[0073] Among them, the local abnormal characteristics of the vibration signal are extracted by short-time Fourier transform (STFT) to extract the transient spectrum energy distribution, including the amplitude peak and the spectrum distribution abnormal points. The local abnormal characteristics of the acoustic signal are extracted by wavelet packet decomposition to extract the energy change characteristics of key time points, including the frequency drift and the mutation point of high-frequency energy, that is, the high-frequency energy change.

[0074] S2.2: Based on the dynamic association model, the feature importance of the extracted local abnormal features is calculated, and each local abnormal feature is associated with the process parameters, which is expressed as:

[0075] I(F ai )=W(F ai );

[0076] W(F ai )=G(F ai )·W(F i );

[0077] Among them, I(F ai ) is the local abnormal feature F ai The importance of W(F ai ) is the weight of the local abnormal feature, which can be obtained through the internal mechanism of the dynamic association model (such as SHAP value, feature contribution analysis, etc.); G(F ai ) is the local significance correction factor; W(F i ) is F i The weight of (t).

[0078] It should be noted that W(F i )=|β i |. G(F ai ) is used to quantify the local abnormality feature F ai Relative to F i The significance of the change in (t). It is calculated based on the dynamic change amplitude of the local feature within the local signal range. The specific formula is designed as follows:

[0079] For the local abnormal characteristics of the vibration signal, the local significance correction factor G(F ai ) is defined as:

[0080]

[0081] Among them, ΔAai is the amplitude change of the local abnormal feature (the difference between the local peak and the average value); max(A) is the maximum amplitude of the global peak in the vibration signal.

[0082] For the local abnormal features of the acoustic signal, the local significance correction factor G(F ai ) is defined as:

[0083]

[0084] Where, ΔE ai is the energy change of the local abnormal feature (the difference between the local high-frequency energy and the reference energy); max(E) is the maximum value of the global energy in the acoustic signal.

[0085] Therefore, the weight of the local feature of the vibration signal is expressed as:

[0086]

[0087] The weight of the local feature of the acoustic signal is expressed as:

[0088]

[0089] Preferably, the dynamic mapping of local abnormal features and global features is achieved through the global feature weight mechanism and local significance correction factor of the dynamic association model, solving the problems of difficult quantification of local abnormal feature priority and insufficient correlation between global and local features in the prior art. The feature importance calculation formula introduces the local significance correction factor into the feature sorting, accurately quantifies the dynamic contribution of local features, and directly associates it with the fracture risk model, overcoming the defects of the prior art that relies on empirical settings and static models. Under complex working conditions, S2.2 greatly improves the real-time and accuracy of local abnormal feature sorting, provides high-quality input for the precise prediction of subsequent fracture risks, and enhances the adaptability and scientificity of the model.

[0090] S2.3: Priority adjustment and dynamic collection strategy.

[0091] According to I(F ai ) Prioritize local abnormal features:

[0092] When I(F ai ) is higher than the first preset threshold T1, it is a high priority;

[0093] When the second preset threshold T2 is less than or equal to I(F ai ), and I(F ai ) is less than or equal to the first preset threshold T1, it is medium priority;

[0094] When I(F ai) is lower than the second preset threshold T2, it is a low priority.

[0095] Furthermore, an optimized local abnormal feature sequence is generated according to the priority division result, and the acquisition strategy of the multimodal signal related to the local abnormal feature is dynamically adjusted:

[0096] If the proportion of high-priority local abnormal features exceeds the third preset threshold R1, the collection frequency of process parameters related to the high-priority features is increased and the collection interval is shortened;

[0097] If the proportion of high-priority local abnormal features is between the fourth preset threshold R2 and the third preset threshold R1, the current process parameter acquisition frequency is maintained, and the time series information of the medium-priority features is preferentially recorded;

[0098] If the proportion of high-priority local abnormal features is lower than the fourth preset threshold R2, the frequency of collecting process parameters related to low-priority features is reduced, and basic information of the low-priority features is recorded at the same time.

[0099] The optimized local abnormal feature sequence is used to input the characterization model of fracture risk in real time as the calculation basis of R(t) to realize real-time analysis of fracture risk change trend.

[0100] Preferably, in step S2, the present invention solves the technical problems of insufficient correlation between local features and global features, unclear feature priority division, and static acquisition strategy in the prior art by extracting local abnormal features of multimodal signals, calculating importance, and adjusting priority. The prior art usually adopts a global feature analysis method, ignoring the dynamic impact of local abnormal features on fracture risk, and it is difficult to adapt to the real-time changes of feature priority under complex working conditions. At the same time, the traditional technology lacks a unified feature importance quantification index, and the feature sorting relies on empirical settings, which cannot provide high-quality feature sequence input for fracture risk prediction. The present invention overcomes the above shortcomings through the following improvements: 1) Proposing a feature importance calculation formula for local abnormal features, and integrating a dynamic association model to quantify the importance of local features, and clarifying the priority division standard; 2) In combination with a dynamic acquisition strategy, by adjusting the acquisition frequency related to the multimodal signal features, the dynamic allocation and optimization of acquisition resources are realized. The above improvement realizes the deep correlation between local abnormal features and the fracture risk characterization model, making the priority sorting of local features more accurate and the feature sequence optimization more efficient. Its outstanding effects include: first, significantly improving the real-time and accuracy of fracture risk analysis under complex working conditions; second, reducing the resource usage of feature acquisition and enhancing the adaptive ability of the model; third, providing high-quality data support for real-time prediction of fracture risk. The technical solution of the present invention has significant advantages in complex and changeable bonding processes, with strong adaptability and outstanding innovation.

[0101] S3: Classify the local abnormal features after priority adjustment, identify fracture risk signals based on the dynamic association model, and generate real-time detection results.

[0102] Specifically, the detection results include fracture risk level, abnormal signal source location information, and process parameter recommendations for adjustment.

[0103] S3.1: Using the density-based clustering algorithm, the local abnormal features after priority adjustment are classified according to V R (t), I(F ai ) and perform classification to generate feature classification results.

[0104] Among them, the feature classification results include high-risk categories and conventional categories. The high-risk category includes amplitude anomaly category, frequency drift category and energy distribution anomaly category. Feature classification is based on the dynamic range and distribution law of local anomaly features in the optimized local anomaly feature sequence, including amplitude range, upper and lower limits of frequency drift and characteristic points of energy distribution. If the boundary of a local anomaly feature category in the feature classification result exceeds the statistical range of the dynamic association model, it is marked as a high-risk category, and the timing information and associated parameters are recorded first.

[0105] S3.2: Identify fracture risk signals of classified local abnormal features based on the dynamic association model; input the high-risk category and conventional category features in the classification results into the dynamic association model respectively, and calculate the fracture risk R(t) corresponding to each category in combination with the process parameters.

[0106] The risk signal identification rules are:

[0107] For local abnormal features belonging to the high-risk category, if R(t) is greater than the fifth preset threshold T R , it is marked as a fracture risk signal, and its dynamic characteristic attributes and associated process parameters are recorded;

[0108] For local abnormal features belonging to the conventional category, if R(t) is greater than the fifth preset threshold T R , it is marked as a potential fracture risk signal;

[0109] If R(t) is less than or equal to the fifth preset threshold T R , it is marked as a normal signal.

[0110] All identification results (fracture risk signals, potential fracture risk signals, normal signals) record their timing information and serve as input for subsequent signal source positioning; local abnormal features of high-risk categories are processed first to ensure the real-time and accuracy of fracture risk signal identification.

[0111] S3.3 locates the abnormal signal source of the identified fracture risk signal; combined with W(F i) and W(F ai ), determine the main local abnormal characteristics of the fracture risk signal, and generate positioning results.

[0112] Specifically, if the fracture risk signal is associated with multiple local abnormal features, then according to W(F ai ) and the correlation between process parameters, from I(F ai ) The signal sources are selected in turn from the highest local abnormal feature, and the corresponding process parameter positions and states are recorded;

[0113] If the I(F ai ) exceeds the preset value, the positioning information of the corresponding signal source is directly output, and the associated signal source positioning information is output;

[0114] The positioning results and the trend of process parameter changes are combined to generate process parameter suggestions for adjustment, and the signal source positioning results and adjustment suggestions are output as detection results.

[0115] In summary, the present invention collects multimodal signals and establishes a dynamic correlation model in combination with process parameters, thereby realizing the quantitative correlation between multi-dimensional signal features and process parameters, and comprehensively improving the adaptability and accuracy of signal analysis; by extracting local abnormal features in multimodal signals, combining the dynamic correlation model to calculate feature importance and adjust priorities, the feature sequence is optimized and resource allocation and analysis efficiency are dynamically balanced, thereby focusing on key fracture risk features; in the fracture risk signal identification stage, a density clustering algorithm is used to classify the local abnormal features after priority adjustment, and the dynamic correlation model is combined to achieve accurate identification of fracture risk signals and location of abnormal signal sources, while providing real-time process parameter adjustment suggestions, forming a closed-loop optimization from signal acquisition to risk assessment. The overall solution solves the problems of insufficient multimodal data correlation, poor real-time performance, and inaccurate signal positioning in the prior art, greatly improving the accuracy, real-time performance, and industrial adaptability of gold bonding wire fracture risk detection, and providing innovative solutions for multimodal signal analysis, intelligent process optimization, and risk assessment.

[0116] Embodiment 2 is an embodiment of the present invention, which provides an artificial intelligence-based bonding wire breakage risk detection system, which is mainly composed of a data processing module, a feature optimization module and a risk assessment module.

[0117] Among them, the data processing module collects the multimodal signals generated during the bonding process, and establishes a dynamic correlation model between the multimodal signals and the fracture risk in combination with the process parameters; the feature optimization module extracts the local abnormal features in the multimodal signals, and adjusts the priority of the importance of the local abnormal features based on the dynamic correlation model; the risk assessment module is used to classify the local abnormal features after priority adjustment, identify the fracture risk signals based on the dynamic correlation model, and generate real-time detection results.

[0118] Example 3, reference Figure 2 , is an embodiment of the present invention, which is different from the previous embodiment in that: if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0119] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0120] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0121] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0122] Example 4 is an embodiment of the present invention, which provides a method for detecting the risk of gold bonding wire breakage based on artificial intelligence. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0123] This experiment aims to verify the effectiveness and innovation of the present invention, and to achieve accurate detection of fracture risks and process optimization suggestions through joint analysis of vibration signals and acoustic signals.

[0124] First, an aluminum alloy bonding wire of model AL-1 (diameter 25μm) was selected as the experimental object, and the signal was collected with a high-precision multimodal sensor system. The sensors include a high-frequency ultrasonic sensor (sampling frequency 500kHz) and a triaxial acceleration sensor (sampling frequency 200kHz), which are used to capture the acoustic signal and vibration signal during the bonding process respectively. The experimental equipment uses an automatic bonding machine WB-100, and three sets of process parameters are set: bonding pressure (5080gf), bonding temperature (100200℃) and vibration frequency (20~40kHz). Each set of parameters corresponds to the bonding process of 10 bonding wires.

[0125] In the experiment, the vibration signal and acoustic signal of each bonding gold wire were first collected by a high-sensitivity multimodal sensor, and the data with a sampling time of 10 seconds were recorded. Subsequently, the collected multimodal signals were subjected to wavelet energy analysis and spectral decomposition to extract local abnormal features such as amplitude peak, frequency drift, and energy distribution coefficient. The dynamic correlation model is used to calculate the importance of the features and prioritize them to form an optimized sequence of local abnormal features, and the fracture risk is further calculated to perform real-time classification and abnormal signal source location. Finally, according to the risk assessment results, the corresponding process parameters are adjusted (such as reducing the bonding pressure or adjusting the vibration frequency) to verify the process optimization effect of the invention.

[0126] Table 1 Bonding wire fracture risk detection test data

[0127]

[0128] As can be seen from Table 1, by simultaneously collecting vibration signals and acoustic signals and extracting local abnormal features (such as vibration signal peak value, acoustic signal energy distribution coefficient, frequency drift, etc.), the inventive technical solution successfully realizes multi-dimensional dynamic monitoring of the risk of gold wire breakage. For example, in the high-risk cases of sample numbers 4 and 6, the abnormal linkage features of frequency drift and acoustic signal energy distribution coefficient are accurately captured, providing a scientific basis for subsequent risk assessment. Secondly, by calculating the risk of breakage, the risk factor of sample 6 in the table reaches 1.15, which is identified as high risk, and the risk of breakage is successfully eliminated by optimizing process parameters (reducing pressure to 55gf and adjusting vibration frequency to 30kHz), verifying the guiding value of dynamic correlation model in process optimization. The inventive technical solution classifies the feature categories and locates the signal source when R(t) exceeds the threshold, and can make targeted suggestions for process adjustment. For example, after the risk factor exceeds the threshold, sample numbers 1 and 4 optimize the process parameters by reducing the bonding pressure, thereby avoiding the risk of breakage. This targeted adjustment not only improves the detection accuracy, but also significantly reduces resource waste.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting the risk of gold bonding wire fracture based on artificial intelligence, characterized in that: include: Collecting multimodal signals generated during the bonding process, and establishing a dynamic correlation model between the multimodal signals and the risk of fracture in combination with process parameters; the multimodal signals include vibration signals and acoustic signals; the process parameters include bonding pressure, bonding temperature and vibration frequency; Extracting local abnormal features from the multimodal signal, and adjusting the priority of the importance of the local abnormal features based on the dynamic association model; The local abnormal features after priority adjustment are classified, fracture risk signals are identified based on the dynamic association model, and real-time detection results are generated.

2. The method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to claim 1, characterized in that: A dynamic correlation model between the multimodal signal and the fracture risk is established in combination with the process parameters, comprising the following steps: Determining a multimodal signal feature based on the multimodal signal, and generating a relationship data model between the multimodal signal feature and the process parameter by associating the multimodal signal feature with the process parameter; Based on the relationship data model between the multimodal signal characteristics and the process parameters, the multimodal signal characteristics and the process parameters are time-calibrated through a time synchronization mechanism to generate a time correspondence between the multimodal signal characteristics and the process parameters; Based on the time correspondence between the multimodal signal characteristics and the process parameters, a characterization model of fracture risk is generated using a regression analysis method, and a dynamic correlation model among fracture risk, multimodal signal characteristics and process parameters is constructed based on the characterization model of fracture risk.

3. The method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to claim 2, characterized in that: The multimodal signal features include the amplitude of the vibration signal, the spectrum distribution characteristics of the vibration signal and the energy change of the acoustic signal; The relationship data model between the multimodal signal characteristics and the process parameters includes a relationship data model between bonding pressure and the amplitude of the vibration signal, a relationship data model between bonding temperature and energy change of the acoustic signal, and a relationship data model between vibration frequency and spectrum distribution characteristics of the vibration signal; The characterization model of the fracture risk maps the time correspondence between the multimodal signal characteristics and the process parameters to the risk factor space, thereby constructing a dynamic correlation model between the fracture risk, the multimodal signal characteristics and the process parameters; The characterization model of the fracture risk is expressed as: Where R(t) is the fracture risk; F i (t) is the i-th multimodal signal feature; P j (t) is the jth process parameter; β0 is the bias term, which represents the basic value of the fracture risk; β i and γ j is the regression coefficient of multimodal signal characteristics and process parameters; ε(t) is the random error term; Combine R(t) with multimodal signal features and process parameters to form a dynamic feature vector V R (t), with the dynamic feature vector V R (t) is taken as input, and the dynamic correlation model M is obtained through mapping to predict the fracture risk value R(t+1) at the next moment, or to analyze the dynamic change trend of fracture risk.

4. The method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to claim 3, characterized in that: The local abnormal features in the multimodal signal include local abnormal features of the vibration signal and the acoustic signal; The local abnormal features of the vibration signal include amplitude peaks and abnormal points of spectrum distribution; the local abnormal features of the acoustic signal include frequency drift and high-frequency energy changes; The priority of the importance of local abnormal features is adjusted based on the dynamic association model, including the following steps: Based on the dynamic association model, the extracted local abnormal feature F ai Calculate the feature importance and assign each local abnormal feature F ai Associated with process parameters: I(F ai )=W(F ai ); W(F ai )=G(F ai )·W(F i ); Among them, I(F ai ) is the local abnormal feature F ai The importance of W(F ai ) is the weight of the local abnormal feature; G(F ai ) is the local significance correction factor; W(F i ) is F i The weight of (t); According to I(F ai ) Prioritize local abnormal features: When I(F ai ) is higher than the first preset threshold T1, it is a high priority; When the second preset threshold T2 is less than or equal to I(F ai ), and I(F ai ) is less than or equal to the first preset threshold T1, it is medium priority; When I(F ai ) is lower than the second preset threshold T2, it is a low priority.

5. The method for detecting the risk of gold bonding wire fracture based on artificial intelligence according to claim 4, characterized in that: Generate an optimized local abnormal feature sequence based on the priority division results, and dynamically adjust the acquisition strategy of multimodal signals related to local abnormal features: If the proportion of high-priority local abnormal features exceeds the third preset threshold R1, the collection frequency of process parameters related to the high-priority features is increased and the collection interval is shortened; If the proportion of high-priority local abnormal features is between the fourth preset threshold R2 and the third preset threshold R1, the current process parameter acquisition frequency is maintained, and the time series information of the medium-priority features is preferentially recorded; If the proportion of high-priority local abnormal features is lower than the fourth preset threshold R2, the frequency of collecting process parameters related to low-priority features is reduced, and basic information of the low-priority features is recorded at the same time.

6. The method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to claim 5, characterized in that: Classifying the local abnormal features after priority adjustment and identifying fracture risk signals based on the dynamic association model includes the following steps: Using a density-based clustering algorithm, the local abnormal features after priority adjustment are classified to generate feature classification results; The feature classification results include high-risk categories and regular categories. If the boundary of a local abnormal feature category in the feature classification results exceeds the statistical range of the dynamic association model, it is marked as a high-risk category, and the time series information and associated parameters are recorded first; Based on the dynamic association model, the classified local abnormal features are used to identify fracture risk signals; the high-risk category and conventional category features in the classification results are respectively input into the dynamic association model, and the fracture risk R(t) corresponding to each category is calculated in combination with the process parameters; For local abnormal features belonging to the high-risk category, if R(t) is greater than the fifth preset threshold T R , it is marked as a fracture risk signal, and its dynamic characteristic attributes and associated process parameters are recorded; For local abnormal features belonging to the conventional category, if R(t) is greater than the fifth preset threshold T R , it is marked as a potential fracture risk signal; If R(t) is less than or equal to the fifth preset threshold T R , it is marked as a normal signal.

7. The method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to claim 6, characterized in that: The abnormal signal source is located for the identified fracture risk signal; combined with W(F i ) and W(F ai ), determine the main local abnormal characteristics of the fault risk signal, and generate positioning results, specifically: If the fracture risk signal is associated with multiple local abnormal features, then according to W(F ai ) and the correlation between process parameters, from I(F ai ) The signal sources are selected in turn from the highest local abnormal feature, and the corresponding process parameter positions and states are recorded; If the I(F ai ) exceeds the preset value, the positioning information of the corresponding signal source is directly output, and the associated signal source positioning information is output; The positioning results and the trend of process parameter changes are combined to generate process parameter suggestions for adjustment, and the positioning results and the process parameter suggestions for adjustment are output as detection results.

8. A system using the artificial intelligence-based gold bonding wire breakage risk detection method as claimed in any one of claims 1 to 7, characterized in that: include: A data processing module, used to collect multimodal signals generated during the bonding process, and to establish a dynamic correlation model between the multimodal signals and the risk of fracture in combination with process parameters; A feature optimization module, used for extracting local abnormal features in the multimodal signal, and adjusting the priority of the importance of the local abnormal features based on the dynamic association model; The risk assessment module is used to classify the local abnormal features after priority adjustment, identify fracture risk signals based on the dynamic association model, and generate real-time detection results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting the risk of gold bonding wire breakage based on artificial intelligence according to any one of claims 1 to 7 are implemented.

Images