Method and System for Denoising Vibration Signals of Flip Chips

Abstract

This invention relates to the field of signal processing technology, and discloses a method and system for denoising vibration signals from a flip-chip, comprising: acquiring and segmenting the vibration signal of the flip-chip response under external ultrasonic excitation; constructing an initial sparse dictionary based on the segmented vibration signal; iteratively updating the sparse dictionary; adjusting the sparsity parameter in real time with the number of iterations and performing soft-threshold denoising on the residual term during the iterative update process; introducing a feature projection matrix to jointly model each segment of the vibration signal, and constructing a joint optimization model based on the updated sparse dictionary and the corresponding sparse coefficients; alternately optimizing the sparse coefficients and the feature projection matrix in the joint optimization model to obtain the optimal solution for the sparse coefficients; and reconstructing the vibration signal by combining the updated sparse dictionary and the optimal solution for the sparse coefficients to obtain the denoised vibration signal. This invention can effectively extract transient features and enhance the expression of key features, suppress noise, and improve denoising robustness and reconstruction accuracy.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method and system for denoising vibration signals from flip chips. Background Technology

[0002] Flip-chip packaging has become an ideal packaging solution for high-end core chips due to its high lead density, short interconnect paths, low resistance, and excellent high-frequency performance. However, as packaging density continues to increase, solder joint size and spacing are constantly decreasing, and most are hidden inside the package, making solder joint defect detection significantly more difficult. Once a solder joint defect exists, it will seriously affect the electrical performance, thermal performance, and mechanical reliability of the packaged product, thereby threatening the stability and lifespan of the overall system. Therefore, to ensure the reliability of flip-chips, effective inspection of their solder joints is necessary. However, the fact that solder joints are usually hidden between the chip and the substrate makes solder joint defect detection difficult for flip-chips.

[0003] To detect solder joint defects, existing technologies include non-destructive testing methods that combine ultrasonic excitation and vibration response analysis. This approach offers advantages such as non-contact, non-destructive operation, ease of operation, and fast response, and has been increasingly applied to the detection of solder joint defects in flip chips. This method utilizes an ultrasonic transducer to apply ultrasonic excitation to the chip surface, while simultaneously acquiring the vibration signal from the chip surface using a Doppler laser vibration meter. Defects within the chip's internal structure alter its local stiffness, causing changes in the vibration response. By analyzing the detected vibration signals, flip chip defects can be effectively detected.

[0004] However, in actual testing, due to the complex packaging structure of flip chips, the small size of defects, the significant edge effect, and the presence of multi-source noise (material grain noise, system noise, etc.), the effective signal is easily submerged, making it impossible to effectively extract key features. At the same time, the multi-source noise and the transient features of defects are coupled with each other, resulting in a significant reduction in the signal-to-noise ratio of defect features, which leads to limited signal denoising accuracy and poor robustness. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method and system for denoising vibration signals of flip chips, which can effectively extract transient features and enhance the expression of key features, suppress noise, and improve denoising robustness and reconstruction accuracy.

[0006] To address the aforementioned technical problems, this invention provides a method for denoising vibration signals from a flip chip, comprising:

[0007] The vibration signal of the flip chip under test under external ultrasonic excitation is acquired and processed in segments. An initial sparse dictionary is constructed based on the segmented vibration signal, and the sparse dictionary is iteratively updated. During the iterative update of the sparse dictionary, the sparsity parameter is adjusted in real time with the number of iterations, and the residual term is denoised by soft thresholding in combination with the real-time adjusted sparsity parameter.

[0008] A feature projection matrix is ​​introduced to jointly model each vibration signal segment, and a joint optimization model is constructed based on the updated sparse dictionary and the corresponding sparse coefficients.

[0009] The optimal solution for the sparse coefficients is obtained by alternately optimizing the sparse coefficients and the feature projection matrix in the joint optimization model.

[0010] By combining the updated sparse dictionary and the optimal solution of the sparse coefficients, the vibration signal is reconstructed, and the denoised vibration signal is obtained.

[0011] Furthermore, the sparsity parameter is adjusted in real time with the number of iterations, and the calculation method for the sparsity parameter adjusted in real time is as follows:

[0012] ,

[0013] in, As the initial factor, max iter The maximum number of iterations, iter This represents the current iteration number. η For the first iter The sparsity parameter is adjusted in real time during each iteration.

[0014] Furthermore, the soft-threshold denoising of the residual term, which is combined with the real-time adjusted sparsity parameter, specifically involves:

[0015] In the iter In the iteration, the th element in the current sparse dictionary after soft-threshold denoising... j 0 The residual term for the column atoms is:

[0016] ,

[0017] in, For the current sparse dictionary, the first... j 0 The residual terms of the atoms in the column. The first denoised word in the current sparse dictionary after soft thresholding j 0 The residual terms of the atoms in the column. sign ( ) is a sign function. η For the first iter The sparsity parameter is adjusted in real time during the next iteration; According to ηCalculated coefficients ; For dynamic amplitude parameters, To The noise variance obtained from noise estimation.

[0018] Furthermore, the introduction of a feature projection matrix to jointly model each vibration signal segment, and the construction of a joint optimization model based on the updated sparse dictionary and corresponding sparse coefficients, specifically involves:

[0019] Based on the updated sparse dictionary and corresponding sparse coefficients, a standard sparse regression model for the vibration signal is constructed. The optimization objective of the standard sparse regression model for the vibration signal is:

[0020] ,

[0021] in, Y It is a vibration signal. D For the updated sparse dictionary, The sparse coefficients are the values ​​corresponding to the updated sparse dictionary. It is the F-norm. The sparse weights are L1 norm. It is an L1 norm;

[0022] A feature projection matrix is ​​introduced, and an optimization objective with the feature projection matrix is ​​established. The feature projection matrix is ​​used to assist the learning process of sparse coefficients.

[0023] By integrating the optimization objective of the standard sparse regression model of the vibration signal and the optimization objective of the feature projection matrix, a joint optimization model is obtained.

[0024] Furthermore, the optimization objective of the feature projection matrix is:

[0025] ,

[0026] in, V For the characteristic projection matrix, T For transpose operation, It is an L2,1 norm. The sparse weights are L2,1 norm.

[0027] Furthermore, the joint optimization model is as follows:

[0028] ,

[0029] in, For the i-th atom in the updated sparse dictionary, Let m be the L2 norm, and m be the number of atoms in the updated sparse dictionary. To control projection consistency weights.

[0030] Furthermore, the alternating optimization of the sparse coefficients and feature projection matrix in the joint optimization model specifically involves:

[0031] The updated sparse dictionary and feature projection matrix are used as fixed values. The first auxiliary variable is introduced and the first augmented Lagrangian function is constructed.

[0032] The updated sparse dictionary and updated sparse coefficients are used as fixed values. A second auxiliary variable is introduced and a second augmented Lagrange function is constructed.

[0033] The updated sparse coefficients and eigenprojection matrix are solved using the alternating direction multiplier method.

[0034] Furthermore, the first augmented Lagrangian function is:

[0035] ,

[0036] in, Z As the first auxiliary variable, For the first augmented Lagrangian function, U As the first dual variable, The first penalty parameter;

[0037] The second augmented Lagrangian function is:

[0038] ,

[0039] in, Z 1 is the second auxiliary variable. For the second augmented Lagrangian function, W As the second dual variable, ρ 1 is the second penalty parameter.

[0040] Furthermore, the method of using alternating direction multipliers to solve for the updated sparse coefficients and eigenprojection matrix specifically involves:

[0041] The updated method for calculating sparsity coefficients is as follows:

[0042] ,

[0043] in, The sparse coefficients are obtained from the (k+1)th iteration. The first auxiliary variable obtained in the (k+1)th iteration is... Let be the first dual variable obtained in the (k+1)th iteration. It is the identity matrix. This is a soft thresholding function;

[0044] The method for solving the characteristic projection matrix is ​​as follows:

[0045] ,

[0046] in, The feature projection matrix obtained in the (k+1)th iteration is... For the i-th row of the feature projection matrix obtained in the (k+1)th iteration, , The number of rows in the feature projection matrix; The second auxiliary variable is obtained from the (k+1)th iteration. For the i-th row of the second auxiliary variable obtained in the (k+1)-th iteration, , The number of rows in the matrix representing the second auxiliary variable; The second dual variable is obtained by solving the problem in the k-th iteration. For the i-th row of the second dual variable obtained in the k-th iteration, , Let be the number of rows in the matrix of the second dual variable;

[0047] The updated sparse coefficients and feature projection matrix are solved alternately until the maximum number of iterations is reached or the iteration ends. The sparse coefficients obtained at this point are taken as the optimal solution for the sparse coefficients.

[0048] The present invention also provides a flip-chip vibration signal denoising system, comprising:

[0049] The signal acquisition module is used to acquire and process the vibration signal of the flip chip under test under external ultrasonic excitation.

[0050] The sparse dictionary module constructs an initial sparse dictionary based on the segmented vibration signal and iteratively updates the sparse dictionary. During the iterative update of the sparse dictionary, the sparsity parameter is adjusted in real time with the number of iterations, and the residual term is denoised using a soft threshold based on the real-time adjusted sparsity parameter.

[0051] The denoising model building module is used to introduce the feature projection matrix to jointly model each vibration signal segment and build a joint optimization model based on the updated sparse dictionary and the corresponding sparse coefficients.

[0052] The sparse coefficient optimization module is used to alternately optimize the sparse coefficients and feature projection matrix in the joint optimization model to obtain the optimal solution for the sparse coefficients.

[0053] The signal denoising module is used to reconstruct the vibration signal by combining the updated sparse dictionary and the optimal solution of the sparse coefficients to obtain the denoised vibration signal.

[0054] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0055] This invention constructs a sparse dictionary and adjusts the sparsity parameters in real time during the iterative update process of the sparse dictionary. It also performs soft thresholding denoising on the residual terms to suppress the interference of noise on atomic updates, thereby enhancing the dictionary's ability to sparsely represent transient features and effectively extracting key features. At the same time, by constructing a joint optimization model to jointly optimize the sparse coding and feature selection process, it achieves enhanced expression of key features and effective suppression of noise, improving denoising robustness and reconstruction accuracy, and effectively improving the denoising effect on the ultrasonic excitation vibration signal of flip chips. Attached Figure Description

[0056] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0057] Figure 1 This is a flowchart of a method in a preferred embodiment of the present invention.

[0058] Figure 2 The figures show simulation results of different methods for noise reduction of flip chip vibration signals in a preferred embodiment of the present invention.

[0059] Figure 3 The figure shows the simulation results of the SNR values ​​of the vibration signal reconstructed using different methods under different noise intensities in a preferred embodiment of the present invention.

[0060] Figure 4 The figure shows the simulation results of the RMSE values ​​of the vibration signals reconstructed using different methods under different noise intensities in a preferred embodiment of the present invention. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0062] Reference Figure 1 As shown, this invention discloses a method for denoising vibration signals from a flip chip, comprising the following steps:

[0063] S1: Acquire the vibration signal of the flip chip under test under external ultrasonic excitation and preprocess it. The vibration signal is one-dimensional time series data containing solder joint status information and noise components.

[0064] S1-1: In this embodiment, the method for obtaining the vibration signal is as follows: the flip chip under test with solder joint defects is placed on an air-floating vibration isolation platform, a sinusoidal sweep frequency signal is emitted by a signal generator, amplified by a power amplifier, and then excited by an air-coupled capacitive ultrasonic transducer; a scanning Doppler laser vibrometer is used to measure the vibration on the surface of the flip chip under test to obtain the vibration signal on the upper surface of the flip chip under test.

[0065] S1-2: In this embodiment, the method for preprocessing the original vibration signal is as follows: the vibration signal is windowed and then segmented: the windowed vibration signal is divided into n equal-length signal slices for subsequent dictionary training and sparse modeling.

[0066] S2: Construct an initial sparse dictionary based on the segmented vibration signal, and iteratively update the sparse dictionary.

[0067] S2-1: Randomly extract samples from the segmented vibration signal to construct an initial sparse dictionary. In this embodiment, the initial sparse dictionary is an overcomplete dictionary.

[0068] S2-2: Based on the initial sparse dictionary, the vibration signal is sparsely represented, resulting in the following sparse representation model:

[0069] ,

[0070] Where Y is the vibration signal, , For the i-th signal slice, n The number of signal slices, For the initial sparse dictionary, The initial sparsity coefficients are... It is noise.

[0071] S2-3: The K-SVD algorithm is used to iteratively update the sparse dictionary in the alternating iterations of sparse representation and sparse dictionary updates. In this embodiment, a dynamic sparsity contraction mechanism is designed to be introduced during the iterative update process of the sparse dictionary. Specifically, the sparsity parameter is adjusted in real time with the number of iterations, and the residual terms are denoised using a soft threshold based on the real-time adjusted sparsity parameter. This can suppress the interference of noise on the atomic update process of the dictionary and improve the sparse representation capability of transient features.

[0072] The calculation method for the real-time adjusted sparsity parameter is as follows:

[0073] ,

[0074] in, For initial factors, The value should be set according to the actual situation. iter max is the current iteration number. iter The maximum number of iterations, η For the first iter The sparsity parameter is adjusted in real time during each iteration.

[0075] In the iter In the iteration, the th element in the current sparse dictionary after soft-threshold denoising... j 0 The residual terms of the column atoms are:

[0076] ,

[0077] in, For the current sparse dictionary, the first... j 0 The residual terms of the atoms in the column. The first denoised word in the current sparse dictionary after soft thresholding j 0 The residual terms of the atoms in the column. sign ( ) is a sign function. η For the first iter The sparsity parameter is adjusted in real time during the next iteration; According to η Calculated coefficients ; For dynamic amplitude parameters, To The noise variance is obtained by performing noise estimation. In this embodiment, , For the residual terms The signal energy obtained by variance estimation. , MAD ( ) indicates the absolute deviation of the median.

[0078] S3: Introduce the feature projection matrix to jointly model each vibration signal segment, and construct a joint optimization model based on the updated sparse dictionary and the corresponding sparse coefficients.

[0079] S3-1: Based on the updated sparse dictionary and the corresponding sparse coefficients (i.e. (After S2-2 iterations) Construct a standard sparse regression model for the vibration signal. The optimization objective of the standard sparse regression model for the vibration signal is:

[0080] ,

[0081] in, Y It is a vibration signal. D For the updated sparse dictionary, The sparse coefficients are the values ​​corresponding to the updated sparse dictionary. It is the F-norm. The sparse weights are L1 norm. It is the L1 norm; in this optimization objective expression, Limit the reconstruction error of vibration signals on the sparse dictionary to ensure the integrity of the basic structure; This is an L1 regularization term used to improve sparsity and suppress redundant expression.

[0082] S3-2: Introducing the feature projection matrix, the optimization objective with the feature projection matrix is:

[0083] ,

[0084] in, V For the characteristic projection matrix, T For transpose operation, It is an L2,1 norm. The sparse weights are L2,1 norm; in the optimization objective expression of this feature projection matrix, constraint The distribution makes it closer to the signal structure after noise suppression; prompt V The sparse coefficients retain only the most representative features, thereby further eliminating redundancy and noise components. In this embodiment, the initial feature projection matrix is ​​a randomly initialized matrix. The feature projection matrix enhances the expressive power of the sparse coefficients for key signal features and further suppresses redundancy and noise interference. The feature projection matrix is ​​used to assist the learning process of the sparse coefficients to improve their expressive power for key transient features.

[0085] S3-3: Integrating the optimization objective of the standard sparse regression model of the vibration signal and the optimization objective of the model with characteristic projection matrix, the optimization objective of the joint optimization model, i.e., the dual sparse regression model, is as follows:

[0086] ,

[0087] in, For the i-th atom in the updated sparse dictionary, It is the L2 norm. m In this embodiment, the number of atoms in the updated sparse dictionary is set to the number of atoms in the sparse dictionary. m =2 n ; To control the projection consistency weight, in this embodiment... .

[0088] S4: The dual sparse regression model is solved iteratively using the Alternating Direction Method of Multipliers (ADMM). By alternately optimizing the sparse coefficients and feature projection matrix in the joint optimization model, the optimal solution of the sparse coefficients is obtained, thereby improving the ability of the sparse coefficients to express defect information and suppressing background noise interference.

[0089] S4-1: Using the updated sparse dictionary and feature projection matrix as fixed values, introduce the first auxiliary variable and construct the first augmented Lagrangian function as follows:

[0090] ,

[0091] in, Z As the first auxiliary variable, For the first augmented Lagrangian function, U As the first dual variable, This is the first penalty parameter.

[0092] S4-2: Using the updated sparse dictionary and updated sparse coefficients as fixed values, introduce a second auxiliary variable and construct the second augmented Lagrangian function as follows:

[0093] ,

[0094] in, Z 1 is the second auxiliary variable. For the second augmented Lagrangian function, W As the second dual variable, ρ 1 is the second penalty parameter.

[0095] S4-3: Solve for the updated sparse coefficients and eigenprojection matrix using the alternating direction multiplier method.

[0096] The updated method for calculating sparsity coefficients is as follows:

[0097] ,

[0098] in, The sparse coefficients are obtained from the (k+1)th iteration. The first auxiliary variable obtained in the (k+1)th iteration is... Let be the first dual variable obtained in the (k+1)th iteration. It is the identity matrix. Soft threshold function;

[0099] The method for solving the characteristic projection matrix is ​​as follows:

[0100]

[0101] in, The feature projection matrix obtained in the (k+1)th iteration is... For the i-th row of the feature projection matrix obtained in the (k+1)th iteration, , The number of rows in the feature projection matrix; The second auxiliary variable is obtained from the (k+1)th iteration. For the i-th row of the second auxiliary variable obtained in the (k+1)-th iteration, , The number of rows in the matrix representing the second auxiliary variable; The second dual variable is obtained by solving the problem in the k-th iteration. For the i-th row of the second dual variable obtained in the k-th iteration, , Let be the number of rows in the matrix of the second dual variable;

[0102] The updated sparse coefficients and feature projection matrix are solved alternately, and the solution is gradually converged by fixing one variable in each round to optimize the other variable, until the maximum number of iterations is reached or the iteration ends. The sparse coefficients obtained at this point are taken as the optimal solution for the sparse coefficients.

[0103] S5: By combining the updated sparse dictionary and the optimal solution of the sparse coefficients to reconstruct the vibration signal, the denoised vibration signal is obtained as follows:

[0104] ,

[0105] in, The reconstructed vibration signal, i.e., the denoised vibration signal. This is the optimal solution for the sparse coefficients. include n A reconstructed vibration signal slice is generated, and all the reconstructed signal slices are spliced ​​together to form a complete denoised vibration signal.

[0106] Through the sparse reconstruction process in this invention, background noise can be effectively suppressed while ensuring the integrity of the signal structure, the ability to identify defect-related transient features can be improved, and higher-fidelity signal restoration can be achieved.

[0107] This invention also discloses a flip-chip vibration signal denoising system, comprising:

[0108] The signal acquisition module is used to acquire and process the vibration signal of the flip chip under test under external ultrasonic excitation.

[0109] The sparse dictionary module constructs an initial sparse dictionary based on the segmented vibration signal and iteratively updates the sparse dictionary. During the iterative update of the sparse dictionary, the sparsity parameter is adjusted in real time with the number of iterations, and the residual term is denoised using a soft threshold based on the real-time adjusted sparsity parameter.

[0110] The denoising model building module is used to introduce the feature projection matrix to jointly model each vibration signal segment and build a joint optimization model based on the updated sparse dictionary and the corresponding sparse coefficients.

[0111] The sparse coefficient optimization module is used to alternately optimize the sparse coefficients and feature projection matrix in the joint optimization model to obtain the optimal solution for the sparse coefficients.

[0112] The signal denoising module is used to reconstruct the vibration signal by combining the updated sparse dictionary and the optimal solution of the sparse coefficients to obtain the denoised vibration signal.

[0113] The present invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for denoising vibration signals of flip chips.

[0114] The present invention also discloses an apparatus including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for denoising flip-chip vibration signals.

[0115] This invention effectively improves the distinguishability and signal-to-noise ratio of flip-chip vibration signals in complex noise environments. Compared with existing sparse representation methods, it has significant advantages in robustness, feature preservation, and denoising accuracy. Specifically, the advantages of this invention are:

[0116] 1. This invention uses a dynamic sparse contraction mechanism to adaptively denoise the residual terms in the dictionary update stage with a soft threshold. By adaptively adjusting the sparsity parameter through the iterative process, the interference of background noise on the dictionary atomic update is suppressed, thereby enhancing the dictionary's ability to represent transient features.

[0117] 2. By constructing a dual sparse regression model that jointly optimizes sparse coefficients and feature projection matrices, the trade-off between noise suppression and feature preservation in the sparse regression model is achieved. The projection matrix guides the sparse representation to be optimized in a cleaner direction, effectively improving the accuracy and stability of noise suppression.

[0118] 3. Reconstructing vibration signals based on dynamic sparse contraction mechanism and dual sparse regression model can effectively improve denoising robustness and reconstruction accuracy, and effectively improve the denoising effect on ultrasonic excitation vibration signals of flip chips.

[0119] To further demonstrate the advantages of this invention, this embodiment uses the method of this invention and existing technologies such as Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Ensemble Empirical Mode Decomposition (EEMD), K-SVD standard sparse reconstruction method (K-SVD method), and classical L1 regularized sparse representation method (L1 method) to denoise the vibration signal of a flip chip under -5dB noise. The spectrum diagrams after denoising by different methods are shown below. Figure 2 As shown, Figure 2 In this context, CEEMDAN represents the spectrum after noise reduction using the CEEMDAN method. Figure 2 In this context, EEMD represents the spectrum after noise reduction using the EEMD method. Figure 2 In this context, K-SVD represents the spectrum after noise reduction using the standard K-SVD sparse reconstruction method. Figure 2 In this context, L1 represents the spectrum after noise reduction using the classic L1 regularized sparse representation method. Figure 2 The present invention in this paper represents the spectrum after noise reduction using the method of the present invention.

[0120] from Figure 2 It can be seen that although the CEEMDAN method can achieve multimodal decomposition, the first-order resonant frequency is significantly attenuated due to mode aliasing, and the core features are submerged by noise. The EEMD method uses noise-assisted decomposition, but the unified noise addition and averaging strategy ignores local sparsity, resulting in the loss of 3rd-5th order high-frequency components. It is also susceptible to square wave noise interference, which misclassifies the spurious peaks caused by square wave noise as feature frequencies, reducing the accuracy of feature extraction. When using the K-SVD method, the sparse representation of noise and effective signal is not sufficiently separable, resulting in a high proportion of noise atoms in the learned dictionary and obvious noise residue in the reconstructed signal. Although the L1 method can highlight the resonant frequency well, there is still some noise interference in the low amplitude region. In contrast, the method of this invention is more effective in suppressing local noise, can completely preserve the main features of the signal, and has the highest similarity between the reconstructed signal and the original signal. In summary, this method can stably extract key features under multi-source noise interference and has higher robustness and stability.

[0121] To verify the applicability of the method of this invention under different noise intensities, this embodiment applies noise of different intensities to the simulated vibration signal and quantitatively compares the noise reduction effects of each method. The signal-to-noise ratio (SNR) and root mean square error (RMSE) are used to evaluate the quality of the reconstructed signal. Generally, a higher SNR value and a lower RMSE indicate a better noise reduction effect. Figure 3 The SNR values ​​of vibration signals reconstructed using different methods under different noise intensities are shown. Figure 4 The RMSE values ​​are for vibration signals reconstructed using different methods under different noise intensities.

[0122] from Figure 3 It can be seen that as the noise intensity increases, the SNR values ​​of the reconstructed signals of all methods show a decreasing trend. Compared with the CEEMD, K-SVD, and L1 methods, the method of this invention can achieve a higher reconstructed SNR under the same input SNR conditions, demonstrating a significant advantage. Figure 4 It can be seen that as the noise intensity increases, the RMSE value, i.e. the reconstruction error, of each method increases, but the method of the present invention always maintains the lowest error under each SNR condition.

[0123] The simulation experiments above show that the present invention has higher robustness and stability, and higher reconstruction accuracy compared with the signal reconstruction methods in the prior art, thus proving the beneficial effects of the present invention.

[0124] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for denoising vibration signals from a flip chip, characterized in that, include: The vibration signal of the flip chip under test under external ultrasonic excitation is acquired and segmented. An initial sparse dictionary is constructed based on the segmented vibration signal, and the sparse dictionary is iteratively updated. During the iterative update of the sparse dictionary, the sparsity parameter is adjusted in real time with the number of iterations, and the residual terms are denoised using soft thresholding in combination with the real-time adjusted sparsity parameter. A feature projection matrix is ​​introduced to jointly model each vibration signal segment. A joint optimization model is constructed based on the updated sparse dictionary and the corresponding sparse coefficients. The joint optimization model is as follows: ， in, Y It is a vibration signal. D For the updated sparse dictionary, The sparse coefficients are the values ​​corresponding to the updated sparse dictionary. It is the F-norm. The sparse weights are L1 norm. V For the characteristic projection matrix, T For transpose operation, It is an L1 norm. For the i-th atom in the updated sparse dictionary, Let m be the L2 norm, and m be the number of atoms in the updated sparse dictionary. To control the projection consistency weight, It is an L2,1 norm. The sparse weights are L2,1 norm. The optimal solution for the sparse coefficients is obtained by alternately optimizing the sparse coefficients and the feature projection matrix in the joint optimization model. By combining the updated sparse dictionary and the optimal solution of the sparse coefficients, the vibration signal is reconstructed, and the denoised vibration signal is obtained.

2. The method for denoising vibration signals of flip-chip according to claim 1, characterized in that: The sparsity parameter is adjusted in real time with the number of iterations. The calculation method for the sparsity parameter is as follows: ， in, As the initial factor, max iter The maximum number of iterations, iter This represents the current iteration number. η For the first iter The sparsity parameter is adjusted in real time during each iteration.

3. The method for denoising vibration signals of flip-chip according to claim 1, characterized in that: The soft thresholding denoising of the residual term, combined with the real-time adjusted sparsity parameter, is specifically as follows: In the iter In the iteration, the th element in the current sparse dictionary after soft-threshold denoising... j 0 The residual terms of the column atoms are: ， in, For the current sparse dictionary, the first... j 0 The residual terms of the atoms in the column. The first denoised word in the current sparse dictionary after soft thresholding j 0 The residual terms of the atoms in the column. sign ( ) is a sign function. η For the first iter The sparsity parameter is adjusted in real time during the next iteration; According to η Calculated coefficients ; For dynamic amplitude parameters, To The noise variance obtained from noise estimation.

4. The method for denoising vibration signals from flip-chip chips according to claim 1, characterized in that: The introduction of a feature projection matrix to jointly model each vibration signal segment, and the construction of a joint optimization model based on the updated sparse dictionary and corresponding sparse coefficients, are as follows: Based on the updated sparse dictionary and corresponding sparse coefficients, a standard sparse regression model for the vibration signal is constructed. The optimization objective of the standard sparse regression model for the vibration signal is: ； A feature projection matrix is ​​introduced, and an optimization objective with the feature projection matrix is ​​established. The feature projection matrix is ​​used to assist the learning process of sparse coefficients. By integrating the optimization objective of the standard sparse regression model of the vibration signal and the optimization objective of the feature projection matrix, a joint optimization model is obtained.

5. The method for denoising vibration signals of a flip chip according to claim 4, characterized in that: The optimization objective of the feature projection matrix is: 。 6. The method for denoising vibration signals of a flip chip according to claim 1, characterized in that: The alternating optimization of the sparse coefficients and feature projection matrix in the joint optimization model is specifically as follows: The updated sparse dictionary and feature projection matrix are used as fixed values. The first auxiliary variable is introduced and the first augmented Lagrangian function is constructed. The updated sparse dictionary and updated sparse coefficients are used as fixed values. A second auxiliary variable is introduced and a second augmented Lagrange function is constructed. The updated sparse coefficients and eigenprojection matrix are solved using the alternating direction multiplier method.

7. The method for denoising vibration signals of a flip chip according to claim 6, characterized in that: The first augmented Lagrangian function is: ， in, Z As the first auxiliary variable, For the first augmented Lagrangian function, U As the first dual variable, The first penalty parameter; The second augmented Lagrangian function is: ， in, Z 1 is the second auxiliary variable. For the second augmented Lagrangian function, W As the second dual variable, ρ 1 is the second penalty parameter.

8. The method for denoising vibration signals of a flip chip according to claim 7, characterized in that: The method of using alternating direction multipliers to solve for the updated sparse coefficients and eigenprojection matrix is ​​as follows: The updated method for solving the sparsity coefficients is as follows: ， in, The sparse coefficients are obtained from the (k+1)th iteration. The first auxiliary variable obtained in the (k+1)th iteration is... Let be the first dual variable obtained in the (k+1)th iteration. It is the identity matrix. This is a soft thresholding function; The method for solving the characteristic projection matrix is ​​as follows: ， in, The feature projection matrix obtained in the (k+1)th iteration is... For the i-th row of the feature projection matrix obtained in the (k+1)th iteration, , The number of rows in the feature projection matrix; The second auxiliary variable is obtained from the (k+1)th iteration. For the i-th row of the second auxiliary variable obtained in the (k+1)th iteration, , The number of rows in the matrix representing the second auxiliary variable; The second dual variable is obtained by solving the problem in the k-th iteration. For the i-th row of the second dual variable obtained in the k-th iteration, , Let be the number of rows in the matrix of the second dual variable; The updated sparse coefficients and feature projection matrix are solved alternately until the maximum number of iterations is reached or the iteration ends. The sparse coefficients obtained at this point are taken as the optimal solution for the sparse coefficients.

9. A flip-chip vibration signal denoising system, characterized in that, include: The signal acquisition module is used to acquire and process the vibration signal of the flip chip under test under external ultrasonic excitation. The sparse dictionary module constructs an initial sparse dictionary based on the segmented vibration signal and iteratively updates the sparse dictionary. During the iterative update of the sparse dictionary, the sparsity parameter is adjusted in real time with the number of iterations, and the residual terms are denoised using soft thresholding in combination with the real-time adjusted sparsity parameter. The denoising model construction module is used to introduce a feature projection matrix to jointly model each vibration signal segment. Based on the updated sparse dictionary and the corresponding sparse coefficients, a joint optimization model is constructed. The joint optimization model is as follows: ， in, Y It is a vibration signal. D For the updated sparse dictionary, The sparse coefficients are the values ​​corresponding to the updated sparse dictionary. It is the F-norm. The sparse weights are L1 norm. V For the characteristic projection matrix, T For transpose operation, It is an L1 norm. For the i-th atom in the updated sparse dictionary, Let m be the L2 norm, and m be the number of atoms in the updated sparse dictionary. To control the projection consistency weight, It is an L2,1 norm. The sparse weights are L2,1 norm. The sparse coefficient optimization module is used to alternately optimize the sparse coefficients and feature projection matrix in the joint optimization model to obtain the optimal solution for the sparse coefficients. The signal denoising module is used to reconstruct the vibration signal by combining the updated sparse dictionary and the optimal solution of the sparse coefficients to obtain the denoised vibration signal.

Images