Flip chip vibration signal denoising method and system
Patent Information
- Application Number
- CN202311829619.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-28
- Publication Date
- 2026-10-30
- Estimated Expiration
- 2043-12-28
AI Technical Summary
[0004]为克服现有技术中倒装芯片缺陷振动信号易受噪声影响，缺陷特征不明显，振动信号局部稀疏度不同等问题，本发明提供了一种基于能量熵局部重加权卷积稀疏编码的倒装芯片振动信号去噪方法，包括以下步骤：
[0070]1、针对现有技术中倒装芯片振动信号不同局部块使用相同的正则化参数，忽略了不同局部块的稀疏度差异，从而导致对振动信号去噪性能的降低的问题，本发明提出重加权模型对正则化参数进行调整。
Smart Images
Figure CN117786322B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of flip chip vibration signal denoising, and in particular to a flip chip vibration signal denoising method and system. Background Technology
[0002] The widespread application of integrated circuits has placed higher demands on chip packaging, with advanced packaging technologies developing towards high density and ultra-fine pitch. Flip chip and solder joint interconnect technology, with its high integration and high performance, is widely used in the manufacture of electronic devices. However, as packaging density continues to increase and bump size and spacing decrease, thermal mismatch between the chip and substrate during flip chip packaging leads to interface deformation and solder joint defects, such as missing solder joints, voids, and cracks. These defects can cause flip chip failure. Therefore, effective solder joint inspection is crucial to ensuring the reliability of flip chips. However, the solder joints are hidden between the chip and the substrate, making accurate detection of solder joint defects in flip chips extremely challenging.
[0003] Non-destructive testing (NDT) techniques combining ultrasonic excitation and vibration analysis have been applied to defect detection in flip-chip solder joints due to their advantages such as non-contact, non-damaging, convenience, and fast response. This method generates ultrasonic waves using an ultrasonic transducer, projects them onto the surface of the flip chip for ultrasonic excitation, and measures the vibration velocity response of the chip surface using a Doppler vibrometer. Because internal defects cause changes in the stiffness of the flip chip, the velocity signals of the forced vibrations of the flip chip change accordingly. By analyzing the detected vibration signals, defects in the flip chip can be effectively detected. However, in actual testing, the ultrasonically excited vibration signals are affected by factors such as material grain noise and detection system noise, which greatly affect the accuracy of flip-chip defect detection. Therefore, noise reduction processing of the vibration signals is an important prerequisite for improving detection accuracy. Summary of the Invention
[0004] To overcome the problems of existing technologies, such as the susceptibility of flip chip defect vibration signals to noise, unclear defect characteristics, and varying local sparsity of vibration signals, this invention provides a method for denoising flip chip vibration signals based on energy entropy locally reweighted convolutional sparse coding, comprising the following steps:
[0005] S1: Acquire the vibration signal of the fault defect of the sample chip;
[0006] S2: Based on the vibration signal of the fault defect, construct a convolutional sparse coding denoising model;
[0007] S3: Based on the convolutional sparse coding denoising model, a local reweighted weight matrix is constructed using the local energy entropy method. The regularization parameters of different localities of the fault defect vibration signal are adaptively adjusted to obtain the energy entropy local reweighted convolutional sparse coding model.
[0008] S4: The energy entropy locally reweighted convolutional sparse coding model is trained in multiple alternating iterations using an accelerated convolutional dictionary learning method to obtain the optimal sparse coefficients and the corresponding convolutional dictionary.
[0009] S5: Based on the optimal sparse coefficients and the corresponding convolution dictionary, the optimized energy entropy local reweighted convolution sparse coding model is used to denoise the multiple signal slices constructed in the noisy vibration signal, and the denoised signal slices are reconstructed to obtain the complete denoised vibration signal.
[0010] In step S3, the step of constructing the local reweighted weight matrix includes:
[0011] S31: Divide the fault vibration signal into N slices. Let the i-th slice s i Energy E i for:
[0012]
[0013] Where q represents the i-th slice s i The qth data point;
[0014] S32: Based on each slice s i Energy E i Construct energy entropy H i :
[0015] H i =-p i logp i (2)
[0016] in, E represents the energy for normalization, and E represents the total energy of all slices.
[0017] S33: Based on the energy entropy H i Construct a locally reweighted weight matrix for:
[0018]
[0019] Where k represents the number of iterations, k = 1, 2, ..., K, and K represents the total number of iterations; It is a locally sparse vector The j-th sparse coefficient; ∈ represents a preset parameter to prevent the denominator from being zero;
[0020] S34: Based on the aforementioned local reweighting weights A locally reweighted weight matrix is formed.
[0021] In one embodiment of the present invention, step S2, the step of constructing the convolutional sparse coding denoising model, includes:
[0022] S21: Uncorrelated Gaussian white noise is defined as noise, and the expression for the fault defect vibration signal is:
[0023] y = x + s (4)
[0024] Where x is the noiseless fault defect vibration signal with estimation, and s is zero-mean Gaussian white noise;
[0025] S22: The vibration signal of the fault defect The idealized representation is the sum of the convolutions of sparse coefficients and dictionary filters. With the aim of minimizing the error between the actual vibration signal and the idealized vibration signal, a convolutional sparse coding denoising model is established as follows:
[0026]
[0027] Where * represents a two-dimensional convolution operation, d i Z represents the i-th dictionary filter, where i = 1, 2, ..., m; i Represents the i-th dictionary filter d i The corresponding sparsity coefficient; λ represents the regularization parameter, used to determine the weights of data precision and coefficient sparsity in the objective function;
[0028] S23: Consider the global dictionary D as a concatenation of m striped cyclic matrices, where each striped cyclic matrix has a strip of width n. The global dictionary D is composed of local dictionaries D of size n×m. L It consists of all shifted versions, where the filter As columns of the global dictionary D, and the global sparse vector Γ is composed of all interleaved concatenated sparse coefficients Given these components, the above convolutional sparse coding denoising model can be transformed into:
[0029]
[0030] S24: Divide the fault vibration signal into N slices. The global sparse vector is correspondingly decomposed into N non-overlapping m-dimensional local sparse vectors. Each local sparse vector α i Defined as a needle, it is determined by the local sparse vector α. iand local dictionary D L To express a global convolutional sparse coding denoising model:
[0031]
[0032] in, Indicates that D L α i An operator that places a value at the i-th position and fills the rest with zeros;
[0033] S25: By using the i-th slice s i D represents L α i The final convolutional sparse coding denoising model is as follows:
[0034]
[0035] In one embodiment of the present invention, the local sparse vector α i The update method is as follows: update the local sparse vector using the block coordinate descent algorithm. each needle α i Consider each coordinate block as a subset of the global sparse vector Γ, and optimize each coordinate block of the global sparse vector Γ sequentially; where each needle α i The update calculation method is as follows:
[0036]
[0037] Among them, R i Indicates no needle α i Contributing residual signal, Indicates that D L α i An operator that places a value at the i-th position and fills the rest with zeros.
[0038] In one embodiment of the present invention, in step S3, the weight for the next iteration is calculated based on the value of the current sparse coefficient solution, and the weight after iteration is used to adjust different local sparse coefficients α. i The regularization parameter λ gives the energy entropy locally reweighted convolutional sparse coding model as follows:
[0039]
[0040] Among them, D L Let W represent the local dictionary, λ represent the regularization parameter, and W represent the local reweighted weight matrix, with the weights on the diagonal being positive numbers w1, w2, ..., w... m The values in other positions are zero.
[0041] In one embodiment of the present invention, step S4, obtaining the optimal convolutional dictionary, includes:
[0042] S41: Set the learning rate η and the number of subspaces t, and use the convolution dictionary obtained from the previous iteration. Perform a linear combination as the convolution dictionary result of the current iteration. Right now:
[0043]
[0044] Where g(·) represents the affine mapping function, and η represents the learning rate. Indicates updating the convolution dictionary gradient value at time,
[0045] S42: By fixing the convolution dictionary result of the current iteration The sparse coefficients are obtained by solving the energy entropy locally reweighted convolutional sparse coding model.
[0046] S43: Fixing the sparse coefficients, when the maximum number of iterations or the absolute error is less than a set threshold, the optimal value of the convolution dictionary is estimated by forming an iterative linear combination.
[0047]
[0048] Among them, c k This represents the sparsity coefficient.
[0049] In one embodiment of the present invention, the sparsity coefficients are obtained as follows:
[0050] Let a sparsity coefficient sparsity coefficient c k The solution process can be expressed as:
[0051]
[0052] Since g(·) represents the affine mapping function, we obtain the sparse coefficient vector c. k The solution formula can be transformed into:
[0053]
[0054] Let R k =[r k-t ,...,r k [ ] is the residual matrix at the k-th iteration, where In order to form Searching for the point with the minimum residual within the subspace of iteration t+1, we have:
[0055]
[0056] Therefore, the sparsity coefficient c is obtained. k :
[0057]
[0058] In one embodiment of the present invention, after each update of the convolution dictionary, the convolution dictionary D is... L Normalize:
[0059]
[0060] in, This represents the operator that projects dictionary atoms onto the unit sphere, where η is the learning rate.
[0061] Based on the same inventive concept, the present invention also provides a flip-chip vibration signal denoising system, which includes the following modules:
[0062] The signal acquisition module is used to acquire the vibration signals of fault defects in the sample chip;
[0063] The denoising model construction module is used to construct a convolutional sparse coding denoising model based on the fault defect vibration signal;
[0064] The denoising model optimization module is used to construct a local reweighted weight matrix based on the convolutional sparse coding denoising model by using local energy entropy, and adaptively adjust the regularization parameters of different localities of the fault defect vibration signal to obtain the energy entropy local reweighted convolutional sparse coding model.
[0065] The model iteration module is used to perform multiple iterations of training on the energy entropy locally reweighted convolutional sparse coding model using an accelerated convolutional dictionary learning method to obtain the optimal sparse coefficients and the corresponding convolutional dictionary.
[0066] The signal denoising module is used to denoise the new signal block constructed from the noisy vibration signal using the optimal sparse coefficients and the corresponding convolution dictionary, and to reconstruct the denoised signal block to obtain the complete denoised vibration signal.
[0067] The present invention also provides an electronic device, 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 the aforementioned flip chip vibration signal denoising method.
[0068] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned flip-chip vibration signal denoising method.
[0069] The technical solution of the present invention has the following advantages compared with the prior art:
[0070] 1. In view of the problem that the existing technology uses the same regularization parameter for different local blocks of the vibration signal of flip chip, ignoring the sparsity difference of different local blocks, thus resulting in a reduction in the denoising performance of the vibration signal, the present invention proposes a reweighted model to adjust the regularization parameter.
[0071] 2. In order to control the problem of local noise suppression leading to ineffective removal of local noise, the local energy entropy method is used to adjust the reweighting coefficients of each local area and construct a local reweighting weight matrix.
[0072] 3. For convolutional dictionary learning, this invention utilizes Anderson's accelerated extrapolation method to accelerate the stochastic gradient descent algorithm, and linearly combines the historical iteration information of the convolutional dictionary to accelerate the learning of the convolutional dictionary and improve the solution accuracy of the dictionary. Attached Figure Description
[0073] 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...
[0074] Figure 1 This is a flowchart of the flip chip vibration signal denoising method described in this invention;
[0075] Figure 2 This is a structural diagram of the flip-chip vibration signal denoising system described in this invention;
[0076] Instruction manual illustrations:
[0077] 101. Signal Acquisition Module; 102. Denoising Model Construction Module; 103. Denoising Model Optimization Module; 104. Model Iteration Module; 105. Signal Denoising Module. Detailed Implementation
[0078] 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.
[0079] Example 1
[0080] Reference Figure 1As shown, this invention provides a method for denoising flip-chip vibration signals based on energy entropy local reweighted convolutional sparse coding. This method effectively solves the problem of reduced accuracy of the acquired flip-chip vibration signals due to noise from the acquisition system, material grains, etc., during the acquisition process. The denoising method includes the following steps:
[0081] S1: Acquire the vibration signal of the fault defect of the sample chip;
[0082] S2: Based on the vibration signal of the fault defect, construct a convolutional sparse coding denoising model;
[0083] S3: Based on the convolutional sparse coding denoising model, in order to control the problem of local noise suppression leading to the inability to effectively remove local noise, the local energy entropy method is adopted to adjust the reweighting coefficients of each locality, construct a local reweighting weight matrix, and adaptively adjust the regularization parameters of different localities of the fault defect vibration signal to obtain the energy entropy local reweighted convolutional sparse coding model.
[0084] S4: The energy entropy locally reweighted convolutional sparse coding model is trained repeatedly in alternating iterations using an accelerated convolutional dictionary learning method to obtain the optimal sparse coefficients and the corresponding convolutional dictionary.
[0085] S5: Based on the optimal sparse coefficients and the corresponding convolution dictionary, the multiple signal slices constructed from the noisy vibration signal are denoised using the optimized energy entropy local reweighted convolution sparse coding model, and the denoised signal slices are reconstructed to obtain the complete denoised vibration signal.
[0086] In this embodiment, the method for obtaining the vibration signal of the fault defect of the sample chip is as follows:
[0087] With the aid of an MC001-YR2010 image measuring instrument, some solder joints and pads were manually removed from the sample chips using fine-tipped tweezers to obtain two typical defects: cold solder joints and missing solder joints. The locations of the solder joint defects were then marked. To obtain a solder joint defect such as cracks, thermal cycling experiments were conducted on the sample chips using an ETS medium-to-large temperature shock test chamber, and the solder joint condition was observed using an acoustic scanning microscope. The sample chips with solder joint defects were placed on an air-floating vibration isolation platform; the sample chips were excited using an air-coupled capacitive ultrasonic transducer; and the vibration of the sample chip surface was measured using a scanning Doppler laser vibrometer to obtain the vibration signal of the chip surface.
[0088] Step S2, the steps for constructing the convolutional sparse coding denoising model include:
[0089] S21: Uncorrelated Gaussian white noise is defined as noise, and the expression for the fault defect vibration signal is:
[0090] y=x+s (1)
[0091] where x is the noiseless fault defect vibration signal to be estimated, and s is zero-mean Gaussian white noise;
[0092] S22: The fault defect vibration signal may be ideally represented as the sum of convolutions of m sparse coefficients each with a length of N and m dictionary filters each with a length of n<<N For the purpose of minimizing the error between the actual vibration signal and the idealized vibration signal, a convolutional sparse coding denoising model is established as follows:
[0093]
[0094] where * denotes a two-dimensional convolution operation, d i denotes the i-th dictionary filter, Z i denotes the sparse coefficient corresponding to d i , and λ denotes a regularization parameter;
[0095] S23: The global dictionary D is regarded as the concatenation of m banded circulant matrices, wherein each banded circulant matrix has a band with a width of n<<N, and the global dictionary D consists of all shifted versions of a local dictionary D with the size of n×m L , wherein the filters serve as the columns of the global dictionary D, and the global sparse vector Γ is composed of all interleaved and concatenated sparse coefficients , then the convolutional dictionary learning problem can be rewritten as:
[0096]
[0097] S24: The fault defect vibration signal is divided into N slices, and the global sparse vector Γ is decomposed into N non-overlapping m-dimensional local sparse vectors Each local sparse vector α i is defined as a needle, then the global convolutional sparse coding denoising model is expressed by the local sparse vector α i and the local dictionary D L :
[0098]
[0099] where denotes an operator that places D L α i at the i-th position and fills the rest with zeros;
[0100] S25: By representing the i-th slice s i as D L αi The final convolutional sparse coding denoising model is as follows:
[0101]
[0102] Specifically, the local sparse vector α i The update method is as follows: update the local sparse vector using the block coordinate descent algorithm. each needle α i Consider each coordinate block as a subset of the global sparse vector Γ, and optimize each coordinate block of the global sparse vector Γ sequentially; where each needle α i The update calculation method is as follows:
[0103]
[0104] Among them, R i Indicates no needle α i Contributing residual signal, Indicates that D L α i An operator that places a value at the i-th position and fills the rest with zeros.
[0105] Step S3, the step of constructing the local reweighted weight matrix includes:
[0106] S31: Divide the fault vibration signal into N slices. Let the i-th slice s i Energy E i for:
[0107]
[0108] Where q represents the i-th slice s i The qth data point;
[0109] S32: Based on each slice s i Energy E i Construct energy entropy H i :
[0110] H i =-p i logp i (8)
[0111] in, E represents the energy for normalization, and E represents the total energy of all slices.
[0112] S33: Based on the energy entropy H i Construct locally reweighted weights for:
[0113]
[0114] Where k represents the number of iterations, k = 1, 2, ..., K, and K represents the total number of iterations; It is a locally sparse vector The j-th sparse coefficient; ∈ represents a preset parameter to prevent the denominator from being zero;
[0115] S34: Based on the aforementioned local reweighting weights A locally reweighted weight matrix is formed.
[0116] Calculate the weights for the next iteration based on the current sparse coefficient solution, and adjust the different local sparse coefficients α using the locally reweighted weights after the iteration. i The regularization parameter λ yields the energy entropy locally reweighted convolutional sparse coding model, which is expressed as:
[0117]
[0118] Among them, D L Let λ represent the local dictionary, λ represent the regularization parameter, and W represent the local reweighted weight matrix, with the weights on the diagonal being positive numbers w1, w2, ..., w m The values in other positions are zero.
[0119] Step S4, the step of obtaining the optimal convolutional dictionary, includes:
[0120] S41: Set the learning rate η and the number of subspaces t, and use the convolution dictionary obtained from the previous iteration. Perform a linear combination as the convolution dictionary result of the current iteration. Right now:
[0121]
[0122] Where g(·) represents the affine mapping function, and η represents the learning rate. Indicates updating the convolution dictionary gradient value at time,
[0123] S42: By fixing the convolution dictionary result of the current iteration Solving the energy entropy locally reweighted convolutional sparse coding model yields the sparse coefficients c. k :
[0124] Let a sparsity coefficient sparsity coefficient c k The solution process can be expressed as:
[0125]
[0126] Since g(·) represents the affine mapping function, we obtain the sparse coefficient vector c. k The solution formula can be transformed into:
[0127]
[0128] Let R k =[r k-t ,...,r k [ ] is the residual matrix at the k-th iteration, where In order to form Searching for the point with the minimum residual within the subspace of iteration t+1, we have:
[0129]
[0130] Therefore, the sparsity coefficient c is obtained. k :
[0131]
[0132] S43: Fixing the sparse coefficients, when the maximum number of iterations or the absolute error is less than a set threshold, the optimal value of the convolution dictionary is estimated by forming an iterative linear combination.
[0133]
[0134] Among them, c k This represents the sparsity coefficient.
[0135] In this embodiment, the convolutional dictionary is normalized after each update:
[0136]
[0137] in, This represents the operator that projects dictionary atoms onto the unit sphere, where η is the learning rate.
[0138] In summary, this invention utilizes Anderson's accelerated extrapolation method to speed up the stochastic gradient descent algorithm, linearly combining the historical iteration information of the convolutional dictionary to accelerate the learning of the convolutional dictionary and improve the solution accuracy. Simulation and actual flip-chip vibration signal test results demonstrate that the proposed denoising method can effectively remove noise from the flip-chip vibration signal and accelerate the learning of the convolutional dictionary, showing greater competitiveness and superiority compared to existing convolutional sparse coding denoising algorithms.
[0139] Example 2
[0140] like Figure 2As shown, based on the same inventive concept as the flip-chip vibration signal denoising method described in Embodiment 1, the present invention also provides a flip-chip vibration signal denoising system, which includes the following modules:
[0141] The signal acquisition module 101 is used to acquire the vibration signal of the fault defect of the sample chip;
[0142] The denoising model construction module 102 is used to construct a convolutional sparse coding denoising model based on the fault defect vibration signal;
[0143] The denoising model optimization module 103 is used to construct a local reweighted weight matrix based on the convolutional sparse coding denoising model by using the local energy entropy method, and adaptively adjust the regularization parameters of different localities of the fault defect vibration signal to obtain the energy entropy local reweighted convolutional sparse coding model.
[0144] The model iteration module 104 is used to perform multiple iterations of training on the energy entropy locally reweighted convolutional sparse coding model using an accelerated convolutional dictionary learning method to obtain the optimal sparse coefficients and the corresponding convolutional dictionary.
[0145] The signal denoising module 105 is used to denoise the new signal block constructed from the noisy vibration signal using the optimal sparse coefficients and the corresponding convolution dictionary, and to reconstruct the denoised signal block to obtain the complete denoised vibration signal.
[0146] Example 3
[0147] The present invention also provides an electronic device, 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 the flip chip vibration signal denoising method described in Embodiment 1.
[0148] Example 4
[0149] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the flip-chip vibration signal denoising method described in Embodiment 1.
[0150] 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.
[0151] 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.
[0152] 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.
[0153] 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.
[0154] 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: S1: Acquire the vibration signal of the fault defect of the sample chip; S2: Based on the vibration signal of the fault defect, construct a convolutional sparse coding denoising model; S3: Based on the convolutional sparse coding denoising model, a local reweighted weight matrix is constructed using the local energy entropy method. The regularization parameters of different localities of the fault defect vibration signal are adaptively adjusted to obtain the energy entropy local reweighted convolutional sparse coding model. S4: Perform multiple iterations of training on the energy entropy locally reweighted convolutional sparse coding model to obtain the optimal sparse coefficients and the corresponding optimal convolution dictionary; S5: Using the optimal sparse coefficients and the corresponding optimal convolution dictionary, the multiple signal slices constructed in the noisy vibration signal are denoised using the optimized energy entropy local reweighted convolution sparse coding model, and the denoised signal slices are reconstructed to obtain the complete denoised vibration signal. In step S3, the step of constructing the local reweighted weight matrix includes: S31: Divide the fault vibration signal into N slices. , order the slice energy for: （1） in, Indicates the first slice The One data point; S32: Based on each slice energy Constructing energy entropy : （2） in, For energy that has been normalized, E The total energy of all slices; S33: Based on the energy entropy Construct a locally reweighted weight matrix for: （3） in, Indicates the number of iterations. , Indicates the total number of iterations; It is a locally sparse vector The Middle One sparse coefficient; This indicates a preset parameter to prevent the denominator from being zero; S34: Based on the aforementioned local reweighting weights This forms a locally reweighted weight matrix; In step S3, the steps to obtain the energy entropy locally reweighted convolutional sparse coding model include: calculating the weights for the next iteration based on the values of the current sparse coefficient solutions, and adjusting different local sparse coefficients using the weights after iteration. regularization parameters The energy entropy locally reweighted convolutional sparse coding model is expressed as follows: in, This represents a local dictionary. Represents the regularization parameter; This represents a locally reweighted weight matrix, with the weights on the diagonal being positive. All other positions are zero; Indicates there is no needle Contributed residual signal; Indicates will Placed in the An operator that identifies positions and fills the rest with zeros.
2. The method for denoising vibration signals of flip-chip according to claim 1, characterized in that: Step S2, the steps for constructing the convolutional sparse coding denoising model include: S21: Uncorrelated Gaussian white noise is defined as noise, and the expression for the fault defect vibration signal is: （4） in, For noiseless fault defect vibration signals with estimation, Zero-mean Gaussian white noise; S22: The vibration signal of the fault defect The idealized representation is the sum of the convolutions of sparse coefficients and dictionary filters. With the aim of minimizing the error between the actual vibration signal and the idealized vibration signal, a convolutional sparse coding denoising model is established as follows: （5） in, This represents a two-dimensional convolution operation. Indicates the first A dictionary filter, ; Indicates the first dictionary filters The corresponding sparsity coefficient; This represents the regularization parameter, used to determine the weights of data precision and coefficient sparsity in the objective function; S23: Consider the global dictionary D as a concatenation of m striped cyclic matrices, where each striped cyclic matrix has a strip of width n. The global dictionary D consists of a series of strips of size n. Local dictionary It consists of all shifted versions, where the filter As columns of the global dictionary D, and the global sparse vector Γ is composed of all interleaved concatenated sparse coefficients The above convolutional sparse coding denoising model is then transformed into: （6） S24: Divide the fault vibration signal into N slices. The global sparse vector is correspondingly decomposed into N non-overlapping m-dimensional local sparse vectors. Each local sparse vector Defined as a needle, it is determined by local sparse vectors. and local dictionary To express a global convolutional sparse coding denoising model: （7） in, Indicates will Placed in the An operator that identifies positions and fills the rest with zeros; S25: By passing the first slice express The final convolutional sparse coding denoising model is as follows: （8）。 3. The method for denoising vibration signals of flip-chip according to claim 2, characterized in that: The local sparse vector The update method is as follows: update the local sparse vector using the block coordinate descent algorithm. , each needle Consider it as taken from a global sparse vector The coordinate blocks, and respectively for each global sparse vector in sequence. The coordinate blocks are optimized; where each needle The update calculation method is as follows: （9） in, Indicates there is no needle Contributing residual signal, , Indicates will Placed in the An operator that identifies positions and fills the rest with zeros.
4. The method for denoising vibration signals of a flip chip according to claim 1, characterized in that: Step S4, the step of obtaining the optimal convolutional dictionary, includes: S41: Set the learning rate The number of subspaces t, using the convolution dictionary obtained from the previous iteration. Perform a linear combination as the convolution dictionary result of the current iteration. ,Right now: （11） in, Represents an affine mapping function. Indicates the learning rate. Indicates updating the convolution dictionary gradient value at time, ; S42: By fixing the convolution dictionary result of the current iteration The sparse coefficients are obtained by solving the energy entropy locally reweighted convolutional sparse coding model. S43: Fix the sparse coefficients, and when the maximum number of iterations or the absolute error is less than a set threshold, estimate the optimal convolution dictionary by forming an iterative linear combination. : （12） in, c k This represents a sparse coefficient vector.
5. The method for denoising vibration signals of a flip chip according to claim 4, characterized in that: The sparsity coefficients are obtained as follows: Let a sparsity coefficient sparsity coefficient The solution process is expressed as follows: （13） because This represents the affine mapping function, thus yielding a sparse coefficient vector. The solution formula is transformed into: （14） make For the first The residual matrix at the next iteration, where In order to form ,exist If we search for the point with the minimum residual within the subspace of each iteration, then: （15） Therefore, the sparsity coefficient is obtained. c k : （16）。 6. The method for denoising vibration signals of a flip chip according to claim 4, characterized in that: After each update of the convolution dictionary, the convolution dictionary is... Normalize: （17） in, , This represents an operator that projects dictionary atoms onto the unit sphere. This is the learning rate.
7. A flip-chip vibration signal denoising system, used to implement the flip-chip vibration signal denoising method according to any one of claims 1 to 6, characterized in that, include: The signal acquisition module is used to acquire the vibration signals of fault defects in the sample chip; The denoising model construction module is used to construct a convolutional sparse coding denoising model based on the fault defect vibration signal; The denoising model optimization module is used to construct a local reweighted weight matrix based on the convolutional sparse coding denoising model by using local energy entropy, and adaptively adjust the regularization parameters of different localities of the fault defect vibration signal to obtain the energy entropy local reweighted convolutional sparse coding model. The model iteration module is used to perform multiple iterations of training on the energy entropy locally reweighted convolutional sparse coding model using an accelerated convolutional dictionary learning method, so as to obtain the optimal sparse coefficients and the corresponding optimal convolutional dictionary. The signal denoising module is used to denoise the new signal block constructed from the noisy vibration signal using the optimal sparse coefficients and the corresponding optimal convolution dictionary, and to reconstruct the denoised signal block to obtain the complete denoised vibration signal.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the flip chip vibration signal denoising method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the flip chip vibration signal denoising method as described in any one of claims 1 to 6.