Electronic component detection system and method

Abstract

The invention discloses an electronic component detection system and method, and relates to the technical field of electronic component detection. Image features, functional parameter features and environmental response features are quantized, a unified feature data set is generated, component states are classified in combination with a support vector machine (SVM), normal, suspicious and potential fault components are effectively distinguished, and the detection accuracy is improved. Marking a detection priority according to a classification result in combination with a production demand; in addition, through classification error rate analysis and loss function optimization, the model prediction accuracy is improved; based on the detection priority, an improved multi-target scheduling algorithm MOEA / D is adopted, detection time and detection cost are brought into optimization targets, and a resource allocation strategy is dynamically adjusted; high-priority components are allocated to core detection resources, and low-priority components use shared resources or are arranged in off-peak periods, so that the resource utilization rate and the scheduling efficiency are effectively improved; and the weight vector is dynamically adjusted to further enhance the adaptability of the scheduling algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic component detection, and particularly to an electronic component detection system and method. Background Art

[0002] Due to the characteristics of miniaturization, high integration and complexity of electronic components, the quality requirements have become extremely strict.

[0003] The current detection methods mainly rely on manual inspection or vision detection technology based on fixed rules. Manual inspection has low efficiency and is difficult to meet the needs of large-scale applications; while rule-based vision detection has limited recognition accuracy when facing components with complex geometric structures or tiny defects, and problems such as missed detection and false detection often occur.

[0004] In addition, factories often face urgent demands for components during the production process. However, the existing pipeline-style batch detection mode lacks flexibility and is difficult to adjust quickly. In the case of urgent production tasks, manual intervention is required for adjustment, which not only easily disrupts the original order of detected products, but also reduces the flexibility of production mode adjustment and cannot meet the requirements of rapid delivery. Therefore, there is an urgent need for a more flexible and targeted electronic component detection solution to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an electronic component detection system and method to solve

[0007] the problem that the existing pipeline-style batch detection mode lacks flexibility and is difficult to adjust quickly, and reduces the adaptability of the production mode for rapid delivery and flexible adjustment in the case of urgent production tasks.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, an embodiment of the present invention provides an electronic component detection system, which includes,

[0010] a detection module for collecting data and detecting components;

[0011] The detection module includes a vision acquisition module, a non-contact electrical performance detection module, and an environmental test module;

[0012] The vision acquisition module is used to collect the appearance data of components, and an in-built camera device combines with a deep learning model to generate appearance image data,

[0013] A non-contact electrical performance detection module, which is used to collect the electrical performance parameters of components, including current, voltage, and impedance curves during operation, and uses non-contact electromagnetic sensors to collect electrical performance.

[0014] An environmental test module, which is used to collect the performance response data of components under high and low temperature and vibration conditions, simulate extreme environmental conditions, and record the performance drift curve of components.

[0015] A scheduling module, which is used to generate a resource scheduling plan according to the status classification and detection priority of components.

[0016] A quality traceability module, which is used to trace, analyze, and optimize the detection results and scheduling plans.

[0017] In a second aspect, the present invention provides a method for detecting electronic components, including:

[0018] Step S1, through the collaborative work of the visual acquisition module, non-contact electrical performance detection module, and environmental test module, collect the basic data of the components, including the appearance image, electrical performance parameters, and environmental response data of the components.

[0019] Step S2, perform feature extraction on the basic data, and uniformly convert the image features, functional parameter features, and environmental response features of the components into a feature data set.

[0020] Use the support vector machine SVM to classify the status of the components using the feature data set into three categories: normal, suspicious, and potential failure. According to the status classification results and production requirements, mark the detection priority of the components as high priority and low priority.

[0021] Step S3, based on the detection priority in step S2, use an improved multi-objective evolutionary algorithm MOEA / D to generate a resource scheduling plan.

[0022] Step S4, perform multi-channel synchronous detection on the detection production line according to the resource scheduling plan; and upload the detection results of all channels to the quality traceability module.

[0023] Step S5, analyze the detection results in step S4 and the scheduling plan in step S3, and use the analysis results to adjust the detection model and scheduling algorithm.

[0024] As a preferred solution of the method for detecting electronic components according to the present invention, wherein: the electrical performance parameters include current, voltage, and impedance curves during operation;

[0025] The environmental response data includes changes in the operating state under high and low temperature and vibration conditions.

[0026] As a preferred solution of the electronic component detection method described in the present invention, wherein: in step S2, the extracted features include image features, functional parameter features, and environmental response features;

[0027] The image features include shape, texture, and defect features. The functional parameter features are electrical response curve features, and the environmental response features are performance drift curves.

[0028] As a preferred solution of the electronic component detection method described in the present invention, wherein: the step of performing feature extraction on the basic data and uniformly converting the image features, functional parameter features, and environmental response features of the components into a feature data set is

[0029] Converting the image features, functional parameter features, and environmental response features of the components into a unified feature data set. The feature matrix calculation formula is:

[0030] F i,j =w 1 ·f I (I i,j )+w 2 ·f P (P i,j )+w 3 ·f E (E i,j ),

[0031] where F i,j is the j-th eigenvalue of the i-th component, i is the index of the component, ranging from 1 to n, where n is the total number of components, j is the index of the feature, ranging from 1 to m, where m is the total number of features, I i,j is the j-th image information of the i-th component, P i,j is the j-th functional parameter of the i-th component, E i,j is the j-th environmental response parameter of the i-th component, f I , f P , f E are the feature extraction functions representing the image, functional parameter, and environmental response respectively, w 1 , w 2 , w 3 are the feature weighting factors, satisfying w 1 +w 2 +w 3 =1;

[0032] The image feature extraction formula is:

[0033] f I (I)=ReLU(W I I + b I ),

[0034] Among them, f I (I) is the result of image feature extraction, ReLU is the rectified linear unit function, defined as ReLU(x) = max(0, x), W I is the convolutional kernel weight matrix, * is the convolution operator, I is the input image matrix, b I is the convolutional bias value;

[0035] The parameter feature extraction formula is:

[0036]

[0037] Among them, f P (P) is the statistical feature of the functional parameter, P k is the value of the k-th functional parameter, k is the functional parameter index, ranging from 1 to m, where m is the total number of functional parameters, is the mean value of the functional parameter, defined as

[0038]

[0039] The environmental response feature extraction formula is:

[0040]

[0041] Among them, f E (E) is the environmental response feature, E(t) is the time-domain environmental signal, t is the time variable, ranging from 0 to T, where T is the observation time, and α is the time decay coefficient.

[0042] As a preferred solution of the electronic component detection method described in the present invention, wherein: the step of using the support vector machine SVM to classify the components according to the feature dataset into three categories: normal, suspicious, and potential failure, and marking the detection priorities of the components as high priority and low priority according to the status classification results combined with production requirements is,

[0043] Using the support vector machine SVM to classify the feature dataset, the classification function is:

[0044]

[0045] Among them, g(x) is the result of the classification function, taking values of {-1, 0, 1} corresponding to potential failure, suspicious, and normal respectively, x is the data point to be classified, x i is the i-th support vector, i is the support vector index, ranging from 1 to n, where n is the total number of support vectors, α i is the coefficient of the i-th support vector, y i is the class label of the i-th support vector, K(x i, x) is a kernel function used to map the high-dimensional space, and b is the classification bias term;

[0046] The classification objective is defined as:

[0047]

[0048] where α is the Lagrange multiplier of the support vector, x j is the j-th support vector, j is the support vector index, and the range is from 1 to n;

[0049] Priority marking is performed. Combining the classification results and production requirements, the detection priority is marked. The priority calculation formula is:

[0050] P ij = ω · g(F i,j ) + (1 - ω) · R ij ,

[0051] where P ij is the j-th priority of the i-th component, g(F i,j ) is the classification result, R ij is the production demand impact factor, and ω is the priority weight coefficient, with the range of [0, 1].

[0052] As a preferred solution of the electronic component detection method described in the present invention, wherein: in the resource scheduling plan:

[0053] Core detection resources are allocated to high-priority components;

[0054] Low-priority components use shared resources or are scheduled for detection during off-peak hours.

[0055] As a preferred solution of the electronic component detection method described in the present invention, wherein: based on the detection priority in step S2, the steps of generating a resource scheduling plan using the improved multi-objective scheduling algorithm MOEA / D are as follows,

[0056] Build optimization objectives, including detection time T(x) and detection cost C(x). The total objective is:

[0057] min f(x) = ω 1 · T(x) + ω 2 · C(x),

[0058] where f(x) is the objective function value, T(x) is the detection time, C(x) is the detection cost, and ω 1 , ω 2 are the objective weight factors, satisfying ω 1 + ω 2 = 1;

[0059] Add resource constraints, and the constraint formula is:

[0060]

[0061] Among them, h(x) is the resource occupancy constraint, and R i (x) is the resource occupancy of the i-th task, i is the task index, and R max is the maximum available value of the resource;

[0062] Based on MOEA / D, perform scheduling adjustment to generate weight vectors, and the generation formula is:

[0063]

[0064] Among them, λ i is the weight of the i-th sub-problem, and N is the total number of sub-problems;

[0065] Define the objective function of the sub-problem, and the objective function formula is:

[0066] g(x|λ i ) = max{λ i ·T(x), (1 - λ i )·C(x)},

[0067] Among them, g(x|λ i ) is the objective function after the i-th weight decomposition,

[0068] The parameter update formula is:

[0069]

[0070] Among them, x t is the solution of the t-th iteration, η is the learning rate, is the gradient of the objective function, and t is the iteration index.

[0071] As a preferred solution of the electronic component detection method described in the present invention, among them: the content analyzed in step S5 includes:

[0072] The resource utilization rate and efficiency of the scheduling plan;

[0073] If the scheduling efficiency is low, adjust the resource allocation weight and re-perform the detection.

[0074] As a preferred solution of the electronic component detection method described in the present invention, among them: the step of analyzing the detection result of step S4 and the scheduling plan of step S3, and using the analysis result to adjust the detection model and the scheduling algorithm is,

[0075] Based on the detection results of step S4 and the scheduling plan of step S3, analyze the system performance, with a focus on classification accuracy and scheduling efficiency;

[0076] Classification performance analysis, the analysis formula is:

[0077]

[0078] Among them, A is the detection accuracy, TP is the true positive class sample in the detection result, TN is the true negative class sample in the detection result, FP is the sample misjudged as the positive class in the detection result, and FN is the sample misjudged as the negative class in the detection result.

[0079] Classification error rate, the classification formula is:

[0080]

[0081] Among them, E c is the classification error rate;

[0082] Conduct scheduling performance analysis.

[0083] The scheduling efficiency calculation formula is:

[0084]

[0085] Among them, E is the resource scheduling efficiency, U i is the resource utilization of the i-th task, T total is the total scheduling time, i is the task index, ranging from 1 to n, where n is the total number of tasks.

[0086] The scheduling delay calculation formula is:

[0087]

[0088] Among them, D is the scheduling delay, T i is the completion time of the i-th task, T deadline is the set deadline of the task;

[0089] According to the detection result R d and the classification performance metrics, adjust the detection model. The update formula for the model loss function is:

[0090]

[0091] Among them, L is the loss function of the detection model, N is the total number of training samples, y i is the true class label of the i-th sample, p i is the predicted probability of the i-th sample.

[0092] The parameter adjustment formula is:

[0093]

[0094] Among them, θ t is the model parameter of the t-th iteration, η is the learning rate,

[0095] is the gradient of the loss function with respect to the parameter,

[0096] Optimize the scheduling algorithm by combining the scheduling efficiency E and the delay D. The optimization formula is:

[0097]

[0098] Among them, is the weight of the (t + 1)-th iteration, is the weight of the t-th iteration, β is the weight adjustment rate, Δλ i is the weight update amount, and the calculation formula is:

[0099] Δλ i = γ(T(x) - T target ) + (1 - γ)(C(x) - C target ),

[0100] Among them, T(x) is the total time of the current scheduling, C(x) is the total cost of the current scheduling, T target , C target are the target time and target cost, and γ is the time-cost balance factor.

[0101] The beneficial effects of the present invention are as follows: In the present invention, the image features, functional parameter features, and environmental response features are quantified to generate a unified feature data set. Combining with the support vector machine SVM, the states of components are classified, effectively distinguishing normal, suspicious, and potentially faulty components, and marking the detection priorities according to the classification results in combination with production requirements; in addition, through the classification error rate analysis and loss function optimization, the prediction accuracy of the model is significantly improved, misjudgment is reduced, and resource waste is reduced.

[0102] In the present invention, based on the detection priorities, an improved multi-objective evolutionary algorithm for decomposition (MOEA / D) is adopted, the detection time and detection cost are incorporated into the optimization objectives, and the resource allocation strategy is dynamically adjusted; high-priority components are allocated to core detection resources, while low-priority components use shared resources or are arranged during off-peak hours, effectively improving the resource utilization rate and scheduling efficiency; and the weight vector is dynamically adjusted to further enhance the adaptability of the scheduling algorithm and reduce task delay and resource conflicts.

[0103] Based on the analysis of the detection results and the scheduling plan, the present invention focuses on evaluating the classification accuracy, resource utilization efficiency, and scheduling delay metrics. After discovering problems, it dynamically adjusts the feature extraction method or classification algorithm of the detection model and optimizes the parameters of the scheduling algorithm to continuously improve the system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0105] Figure 1 It is a schematic structural diagram of the electronic component detection system of the present invention.

[0106] Figure 2 It is a schematic flow diagram of the electronic component detection method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0107] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0108] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0109] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or selectively exclusive embodiment with other embodiments.

[0110] Embodiment 1, referring to Figure 1 and Figure 2 , this embodiment provides an electronic component detection system, including:

[0111] A detection module for collecting and detecting data of components;

[0112] The detection module includes a visual acquisition module, a non-contact electrical performance detection module, and an environmental test module;

[0113] The visual acquisition module is used to collect the appearance data of components, and the built-in camera device combines with the deep learning model to generate appearance image data.

[0114] A non-contact electrical performance detection module, which is used to collect the electrical performance parameters of components, including the current, voltage and impedance curves during operation, and uses a non-contact electromagnetic sensor to collect electrical performance.

[0115] An environmental test module, which is used to collect the performance response data of components under high and low temperature and vibration conditions, simulate extreme environmental conditions, and record the performance drift curve of components.

[0116] A scheduling module, which is used to generate a resource scheduling plan according to the status classification and detection priority of components.

[0117] A quality traceability module, which is used to trace, analyze and optimize the detection results and scheduling plans.

[0118] This embodiment also provides a method for detecting electronic components, including:

[0119] Step S1, through the collaborative work of the visual acquisition module, non-contact electrical performance detection module and environmental test module, collect the basic data of the components, including the appearance image, electrical performance parameters and environmental response data of the components.

[0120] The electrical performance parameters include the current, voltage and impedance curves during operation.

[0121] The environmental response data includes the changes in the operating state under high and low temperature and vibration conditions.

[0122] Step S2, perform feature extraction on the basic data, and uniformly transform the image features, functional parameter features and environmental response features of the components into a feature dataset.

[0123] Use the support vector machine SVM to classify the status of components using the feature dataset, which is divided into three categories: normal, suspicious and potential failure. According to the status classification results and combined with production requirements, mark the detection priority of the components, which is divided into high priority and low priority.

[0124] In step S2, the extracted features include image features, functional parameter features and environmental response features.

[0125] The image features include shape, texture, defect features, the functional parameter features are the electrical response curve features, and the environmental response features are the performance drift curves.

[0126] The step of performing feature extraction on the basic data and uniformly transforming the image features, functional parameter features and environmental response features of the components into a feature dataset is

[0127] Transform the image features, functional parameter features and environmental response features of the components into a unified feature dataset, and the feature matrix calculation formula is:

[0128] F i,j = w 1 · f I (I i,j ) + w 2 · f P (P i,j ) + w 3 · f E (E i,j ),

[0129] where F i,j is the j-th eigenvalue of the i-th component, i is the index of the component, ranging from 1 to n, where n is the total number of components, j is the index of the feature, ranging from 1 to m, where m is the total number of features, I i,j is the j-th image information of the i-th component, P i,j is the j-th functional parameter of the i-th component, E i,j is the j-th environmental response parameter of the i-th component, f I , f P , f E are feature extraction functions representing image, functional parameter, and environmental response respectively, w 1 , w 2 , w 3 are feature weighting factors, satisfying w 1 + w 2 + w 3 = 1;

[0130] The image feature extraction formula is:

[0131] f I (I) = ReLU(W I I + b I ),

[0132] where f I (I) is the image feature extraction result, ReLU is the rectified linear unit function, defined as ReLU(x) = max(0, x), W I is the convolutional kernel weight matrix, * is the convolution operator, I is the input image matrix, b I is the convolutional bias value;

[0133] The parameter feature extraction formula is:

[0134]

[0135] where f P (P) is the statistical feature of the functional parameter, P k is the k-th functional parameter value, k is the functional parameter index, ranging from 1 to m, where m is the total number of functional parameters, is the mean of the functional parameters, defined as

[0136]

[0137] The formula for extracting environmental response characteristics is:

[0138]

[0139] where f E (E) is the environmental response characteristic, E(t) is the time-domain environmental signal, t is the time variable, ranging from 0 to T, where T is the observation time, and α is the time decay coefficient;

[0140] Using the support vector machine SVM, the components are classified into three categories: normal, suspicious, and potential failure according to the feature dataset. According to the status classification results and production requirements, the detection priorities of the components are marked, divided into high priority and low priority. The steps are as follows:

[0141] Using the support vector machine SVM to classify the feature dataset, and the classification function is:

[0142]

[0143] where g(x) is the result of the classification function, taking values of {-1, 0, 1}, corresponding to potential failure, suspicious, and normal respectively. x is the data point to be classified, and x i is the i-th support vector, i is the support vector index, ranging from 1 to n, where n is the total number of support vectors, and α i is the coefficient of the i-th support vector, y i is the class label of the i-th support vector, and K(x i , x) is the kernel function, used to map to a high-dimensional space, and b is the classification bias term;

[0144] The classification objective is defined as:

[0145]

[0146] where α is the Lagrange multiplier of the support vector, and x j is the j-th support vector, and j is the support vector index, ranging from 1 to n;

[0147] Perform priority marking. Combining the classification results and production requirements, mark the detection priority. The priority calculation formula is:

[0148] P ij = ω · g(F i,j ) + (1 - ω) · R ij ,

[0149] where P ij is the j-th priority of the i-th component, and g(Fi,j ) is the classification result, R ij is the production demand impact factor, ω is the priority weight coefficient, and the range is [0, 1];

[0150] Specifically, analyze the accuracy rate and error rate of the detection results, and adjust the classification model based on the analysis results to effectively improve the classification accuracy by iteratively optimizing the loss function.

[0151] Step S3, based on the detection priority in step S2, use the improved multi-objective evolutionary algorithm MOEA / D to generate a resource scheduling plan;

[0152] In the resource scheduling plan:

[0153] Allocate core detection resources to high-priority components;

[0154] Low-priority components use shared resources or are scheduled for detection during off-peak hours;

[0155] The steps to generate a resource scheduling plan using the improved multi-objective evolutionary algorithm MOEA / D based on the detection priority in step S2 are as follows:

[0156] Construct optimization objectives, including detection time T(x) and detection cost C(x), and the overall objective is:

[0157] minf(x) = ω 1 ·T(x) + ω 2 ·C(x),

[0158] where f(x) is the objective function value, T(x) is the detection time, C(x) is the detection cost, ω 1 , ω 2 is the objective weight factor, satisfying ω 1 + ω 2 = 1;

[0159] Add resource constraints, and the constraint formula is:

[0160]

[0161] where h(x) is the resource occupancy constraint, R i (x) is the resource occupancy of the i-th task, i is the task index, R max is the maximum available value of the resource;

[0162] Based on MOEA / D, perform scheduling adjustments to generate a weight vector, and the generation formula is:

[0163]

[0164] where λ iis the weight of the i-th sub-problem, and N is the total number of sub-problems;

[0165] Define the sub-problem objective function, and the objective function formula is:

[0166] g(x|λ i ) = max{λ i ·T(x), (1 - λ i )·C(x)},

[0167] where g(x|λ i ) is the objective function after the i-th weight decomposition,

[0168] The parameter update formula is:

[0169]

[0170] where x t is the solution at the t-th iteration, η is the learning rate, is the gradient of the objective function, and t is the iteration index;

[0171] Specifically, the efficiency of the scheduling performance and the evaluation delay can effectively reflect the execution quality of the scheduling plan; and a multi-objective scheduling algorithm is dynamically adjusted based on the weight update formula to further reduce the task delay.

[0172] Step S4, perform multi-channel synchronous detection on the detection pipeline according to the resource scheduling plan; and upload the detection results of all channels to the quality traceability module;

[0173] Step S5, analyze the detection results of Step S4 and the scheduling plan of Step S3, and the analysis results are used to adjust the detection model and the scheduling algorithm;

[0174] The content analyzed in Step S5 includes:

[0175] The resource utilization rate and efficiency of the scheduling plan;

[0176] If the scheduling efficiency is low, adjust the resource allocation weight and re-perform the detection;

[0177] The steps of analyzing the detection results of Step S4 and the scheduling plan of Step S3, and using the analysis results to adjust the detection model and the scheduling algorithm are,

[0178] Based on the detection results of Step S4 and the scheduling plan of Step S3, analyze the system performance, and focus on the classification accuracy and the scheduling efficiency;

[0179] Classification performance analysis, and the analysis formula is:

[0180]

[0181] Among them, A is the detection accuracy rate, TP is the true positive class sample in the detection result, TN is the true negative class sample in the detection result, FP is the sample misjudged as the positive class in the detection result, and FN is the sample misjudged as the negative class in the detection result.

[0182] The classification error rate, and the classification formula is:

[0183]

[0184] Among them, E c is the classification error rate;

[0185] Perform scheduling performance analysis.

[0186] The scheduling efficiency calculation formula is:

[0187]

[0188] Among them, E is the resource scheduling efficiency, U i is the resource utilization of the i-th task, T total is the total scheduling time, i is the task index, ranging from 1 to n, where n is the total number of tasks.

[0189] The scheduling delay calculation formula is:

[0190]

[0191] Among them, D is the scheduling delay, T i is the completion time of the i-th task, T deadline is the set deadline of the task;

[0192] According to the detection result R d and the classification performance index, adjust the detection model, and the update formula of the model loss function is:

[0193]

[0194] Among them, L is the loss function of the detection model, N is the total number of training samples, y i is the true class label of the i-th sample, p i is the predicted probability of the i-th sample.

[0195] The parameter adjustment formula is:

[0196]

[0197] Among them, θ t is the model parameter at the t-th iteration, η is the learning rate,

[0198] is the gradient of the loss function with respect to the parameter.

[0199] Optimize the scheduling algorithm by combining the scheduling efficiency E and the latency D. The optimization formula is as follows:

[0200]

[0201] where is the weight of the (t + 1)-th iteration, λ i t is the weight of the t-th iteration, β is the weight adjustment rate, and Δλ i is the weight update amount, and its calculation formula is as follows:

[0202] Δλ i = γ(T(x) - T target ) + (1 - γ)(C(x) - C target ),

[0203] where T(x) is the total time of the current scheduling, C(x) is the total cost of the current scheduling, T target and C target are the target time and target cost, and γ is the time-cost balance factor;

[0204] Specifically, establish a closed-loop optimization mechanism by combining detection and scheduling feedback, which is more adaptable to the actual scenario requirements and has better practicability and scalability.

[0205] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An electronic component detection system, characterized in that: include, Detection module, used for data collection and detection of components; Detection module, including visual acquisition module, non-contact electrical performance detection module and environmental test module; The visual acquisition module is used to collect the appearance data of components. The built-in camera device combines the deep learning model to generate appearance image data. The non-contact electrical performance detection module is used to collect the electrical performance parameters of components, including the current, voltage and impedance curves during operation, using non-contact electromagnetic sensors to collect electrical performance. Environmental test module, used to collect performance response data of components under high and low temperature and vibration conditions, simulate extreme environmental conditions, and record performance drift curves of components; A scheduling module is used to generate a resource scheduling plan based on the status classification and detection priority of components; The quality traceability module is used to trace, analyze and optimize test results and scheduling plans.

2. An electronic component detection method, based on the electronic component detection system according to claim 1, characterized in that: include: Step S1, collecting basic data of components, including appearance images, electrical performance parameters and environmental response data of components, through the collaborative work of the visual acquisition module, the non-contact electrical performance detection module and the environmental test module; Step S2, extracting features from the basic data, and converting the image features, functional parameter features, and environmental response features of the components into a feature data set; Support vector machine (SVM) is used to classify the status of components using feature data sets into three categories: normal, suspicious, and potential failure. According to the status classification results and production needs, the detection priority of components is marked and divided into high priority and low priority. Step S3, based on the detection priority of step S2, using the improved multi-objective scheduling algorithm MOEA / D to generate a resource scheduling plan; Step S4, perform multi-channel synchronous testing on the testing pipeline according to the resource scheduling plan; and upload the test results of all channels to the quality traceability module; Step S5, analyzing the detection result of step S4 and the scheduling plan of step S3, and the analysis result is used to adjust the detection model and the scheduling algorithm.

3. An electronic component detection method as claimed in claim 2, characterized in that: The electrical performance parameters include current, voltage and impedance curves during operation; The environmental response data includes operating state changes under high and low temperature and vibration conditions.

4. An electronic component detection method as claimed in claim 3, characterized in that: In step S2, the extracted features include image features, functional parameter features and environmental response features; Image features include shape, texture, and defect features, functional parameter features are electrical response curve features, and environmental response features are performance drift curves.

5. An electronic component detection method as claimed in claim 4, characterized in that: The step of extracting features from the basic data and converting the image features, functional parameter features and environmental response features of the components into a feature data set is as follows: The image features, functional parameter features and environmental response features of components are converted into a unified feature data set. The feature matrix calculation formula is: F i,j =w1·f I (I i,j )+w2·f P (P i,j )+w3·f E (E i,j ), Among them, F i,j is the jth feature value of the i-th component, i is the index of the component, ranging from 1 to n, where n is the total number of components, j is the index of the feature, ranging from 1 to m, where m is the total number of features, I i,j is the jth image information of the ith component, P i,j is the jth functional parameter of the ith component, E i,j is the jth environmental response parameter of the ith component, f I ,f P ,f E are feature extraction functions representing images, functional parameters and environmental responses respectively, w1, w2, w3 are feature weighting factors, satisfying w1+w2+w3=1; The image feature extraction formula is: f I (I)=ReLU(W I I+b I ), Among them, f I (I) is the image feature extraction result, ReLU is the linear rectification function, defined as ReLU(x)=max(0,x), W I is the convolution kernel weight matrix, * is the convolution operator, I is the input image matrix, b I is the convolution bias value; The parameter feature extraction formula is: Among them, f P (P) is the statistical characteristic of the functional parameter, P k is the kth function parameter value, k is the function parameter index, ranging from 1 to m, where m is the total number of function parameters, is the function parameter mean, defined as The environmental response feature extraction formula is: Among them, f E (E) is the environmental response characteristic, E(t) is the time domain environmental signal, t is the time variable ranging from 0 to T, where T is the observation time, and α is the time attenuation coefficient.

6. An electronic component detection method as claimed in claim 5, characterized in that: The support vector machine (SVM) is used to classify the status of components using feature data sets into three categories: normal, suspicious and potential faults. According to the status classification results combined with production requirements, the detection priority of the components is marked and divided into high priority and low priority. The steps are as follows: The feature data set is classified using support vector machine SVM, and the classification function is: Among them, g(x) is the result of the classification function, and its value is {-1, 0, 1}, which corresponds to potential failure, suspicious, and normal respectively. x is the data point to be classified, and x i is the i-th support vector, i is the support vector index, ranging from 1 to n, where n is the total number of support vectors, α i is the coefficient of the i-th support vector, y i is the category label of the i-th support vector, K(x i ,x) is the kernel function used to map high-dimensional space, and b is the classification bias term; The classification target is defined as: Among them, α is the Lagrange multiplier of the support vector, x j is the jth support vector, j is the support vector index, ranging from 1 to n; Priority marking is performed. Combined with the classification results and production requirements, the detection priority is marked. The priority calculation formula is: P ij =ω·g(F i,j )+(1-ω)·R ij , Among them, P ij is the jth priority of the ith component, g(F i,j ) is the classification result, R ij is the production demand influencing factor, ω is the priority weight coefficient, and the range is [0,1].

7. An electronic component detection method as claimed in claim 6, characterized in that: In the resource scheduling plan: High-priority components are allocated core testing resources; Low priority components use shared resources or are scheduled for testing during off-peak hours.

8. An electronic component detection method as claimed in claim 7, characterized in that: The step of generating a resource scheduling plan using the improved multi-objective scheduling algorithm MOEA / D based on the detection priority of step S2 is as follows: Construct optimization objectives, including detection time T(x) and detection cost C(x). The overall goal is: min f(x)=ω1·T(x)+ω2·C(x), Where f(x) is the objective function value, T(x) is the detection time, C(x) is the detection cost, ω1, ω2 are the target weight factors, satisfying ω1+ω2=1; Add resource constraints. The constraint formula is: Among them, h(x) is the resource occupancy constraint, R i (x) is the resource usage of the i-th task, i is the task index, R max is the maximum available value of the resource; Based on MOEA / D, the scheduling adjustment is performed to generate the weight vector. The generation formula is: Among them, λ i is the weight of the ith subproblem, and N is the total number of subproblems; Define the sub-problem objective function, the objective function formula is: g(x|λ i )=max{λ i ·T(x),(1-λ i )·C(x)}, Among them, g(x|λ i ) is the objective function after the i-th weight decomposition, The parameter update formula is: Among them, x t is the solution of the tth iteration, η is the learning rate, is the gradient of the objective function and t is the iteration index.

9. An electronic component detection method as claimed in claim 8, characterized in that: The contents analyzed in step S5 include: Resource utilization and efficiency of scheduling plans; If the scheduling efficiency is low, adjust the resource allocation weight and re-test.

10. An electronic component detection method as claimed in claim 9, characterized in that: The step of analyzing the detection result of step S4 and the scheduling plan of step S3, and using the analysis result to adjust the detection model and the scheduling algorithm is as follows: Based on the detection result of step S4 and the scheduling plan of step S3, the system performance is analyzed, focusing on the classification accuracy and scheduling efficiency; Classification performance analysis, the analysis formula is: Among them, A is the detection accuracy, TP is the true positive sample in the detection result, TN is the true negative sample in the detection result, FP is the sample mistakenly judged as positive in the detection result, and FN is the sample mistakenly judged as negative in the detection result. Classification error rate, classification formula is: Among them, E c is the classification error rate; Perform scheduling performance analysis, The calculation formula for scheduling efficiency is: Among them, E is the resource scheduling efficiency, U i is the resource utilization of the ith task, T total is the total scheduling time, i is the task index, ranging from 1 to n, where n is the total number of tasks, The scheduling delay calculation formula is: Where D is the scheduling delay, T i is the completion time of the i-th task, T deadline Set deadlines for tasks; According to the test results R d The detection model is adjusted based on the classification performance index, and the model loss function update formula is: Among them, L is the loss function of the detection model, N is the total number of training samples, and y i is the true category label of the i-th sample, p i is the predicted probability of the i-th sample, The parameter adjustment formula is: Among them, θ t is the model parameter of the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters, The scheduling algorithm is optimized by combining scheduling efficiency E and delay D. The optimization formula is: in, is the weight of the t+1th iteration, is the weight of the tth iteration, β is the weight adjustment rate, Δλ i is the weight update amount, and the calculation formula is: Dl i =γ(T(x)-T target )+(1-γ)(C(x)-C target ), Among them, T(x) is the total time of the current schedule, C(x) is the total cost of the current schedule, and T target , C target are the target time and target cost, and γ is the time and cost balance factor.

Images