PCBA circuit board automatic detection method, device and equipment

Abstract

The invention relates to the technical field of PCBA circuit boards, and discloses a PCBA circuit board automatic detection method, device and equipment. The method comprises the following steps: clustering welding point image data, component position data and electrical characteristic data of a PCBA circuit board to obtain a circuit characteristic data set; performing defect detection and Bayesian probability calculation based on the circuit characteristic data set to obtain component failure probability data; singular value decomposition and feature space mapping are carried out on the voltage data, the current data and the temperature data of the PCBA circuit board to obtain circuit parameter state data; performing constraint optimization on the circuit control quantity set according to the circuit parameter state data to obtain operation parameter control data; coupling optimization is carried out based on the component failure probability data and the operation parameter control data, the circuit protection control parameters are obtained, the misjudgment rate and the omission ratio of PCBA circuit board defect detection are reduced, and the component failure risk is accurately evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of PCBA circuit boards, and particularly to an automatic detection method, device, and equipment for PCBA circuit boards. Background Art

[0002] Traditional PCBA detection methods mainly rely on manual visual inspection and single electrical characteristic testing. This method is not only inefficient but also easily affected by human factors, making it difficult to handle complex defect types and dynamic working environments.

[0003] On the other hand, existing PCBA operating parameter control methods often use fixed models and preset thresholds. This control strategy is sensitive to component parameter changes and external interference, making it difficult to ensure the stable operation of the system under complex working conditions. Especially under the influence of factors such as component aging and temperature changes, traditional control methods cannot effectively predict and prevent potential failure risks. Summary of the Invention

[0004] The present invention provides an automatic detection method, device, and equipment for PCBA circuit boards, which reduces the misjudgment rate and missed detection rate of detection and realizes the accurate assessment of component failure risks.

[0005] In a first aspect, the present invention provides an automatic detection method for a PCBA circuit board. The automatic detection method for the PCBA circuit board includes: Clustering the solder joint image data, component position data, and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; Performing defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; Performing singular value decomposition and feature space mapping on the voltage data, current data, and temperature data of the PCBA circuit board to obtain circuit parameter state data; Constraining and optimizing the circuit control quantity set according to the circuit parameter state data to obtain operating parameter control data; Performing coupled optimization based on the component failure probability data and the operating parameter control data to obtain circuit protection control parameters.

[0006] In a second aspect, the present invention provides an automatic detection device for a PCBA circuit board. The automatic detection device for the PCBA circuit board includes: A clustering module for clustering the solder joint image data, component position data, and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; A detection module for performing defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; A mapping module, configured to perform singular value decomposition and feature space mapping on voltage data, current data, and temperature data of the PCBA circuit board to obtain circuit parameter status data; A constraint optimization module, configured to perform constraint optimization on a set of circuit control quantities according to the circuit parameter status data to obtain operating parameter control data; A coupling optimization module, configured to perform coupling optimization based on the component failure probability data and the operating parameter control data to obtain circuit protection control parameters.

[0007] A third aspect of the present invention provides an automatic detection device for a PCBA circuit board, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the automatic detection device for the PCBA circuit board to execute the above-mentioned automatic detection method for the PCBA circuit board.

[0008] In the technical solution provided by the present invention, through the combination of weighted K-means clustering and feature space mapping, the false judgment rate and missed detection rate of detection are significantly reduced. By adopting the Kriging surrogate model and the Bayesian probability calculation method, the accurate assessment of the component failure risk is realized, and through the online parameter update mechanism, the prediction accuracy of the model is guaranteed. The parameter status prediction method based on singular value decomposition and subspace mapping overcomes the disadvantage of strong parameter dependence of the traditional physical model and improves the adaptability of the system to parameter changes. The finite control set optimization strategy combined with the branch and bound algorithm realizes the solution of the optimal control sequence under the condition of limited computing resources and guarantees the real-time performance and executability of the control instructions. By establishing a detection-control coupling optimization mechanism, the defect detection result and the control strategy are organically combined to form a complete closed-loop protection system, effectively improving the overall reliability and safety of the system. By adopting the hierarchical protection decision-making and dynamic threshold adjustment method, the system can adaptively adjust the control strategy according to different risk levels, effectively preventing the cascading failure risk of components. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] 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 some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1 It is a schematic flowchart of the automatic detection method for a PCBA circuit board provided by an embodiment of the present application; Figure 2 It is a schematic block diagram of the structure of the automatic detection device for a PCBA circuit board provided by an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of the automatic detection device for PCBA circuit boards provided by the embodiments of the present application. Specific embodiments

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] The flowchart shown in the accompanying drawings is only an example illustration, and does not necessarily include all contents and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change based on the actual situation.

[0013] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless clearly indicated otherwise in the context, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0014] It should be further understood that the term " / and" as used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0015] Next, some embodiments of the present application will be described in detail in conjunction with the accompanying drawings. Without conflict, the features in the following embodiments and the embodiments can be combined with each other.

[0016] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of the automatic detection method for PCBA circuit boards provided by the embodiments of the present application. As Figure 1 shown, the automatic detection method for PCBA circuit boards provided by the embodiments of the present application includes steps S100 to step S500.

[0017] Step S100: Cluster the image data of the welding points, the component position data, and the electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; It can be understood that the execution subject of the present invention can be an automatic detection device for PCBA circuit boards, or a terminal or a server. Specifically, it is not limited here. In the embodiments of the present invention, the server is taken as the execution subject for illustration.

[0018] Specifically, denoising and normalization processing are performed on the image data of the solder joints, component position data, and electrical characteristic data of the PCBA circuit board to obtain preprocessed data. During the denoising process, methods such as adaptive filtering, mean filtering, or Gaussian filtering are used to remove the noise in the image data. For the electrical characteristic data, wavelet transform or Fourier transform is used to eliminate the high-frequency interference during the measurement process. To process data from different sources on the same scale, normalization processing is carried out to make all data vary within the same range, thereby reducing the influence between different physical quantities and improving the accuracy of clustering calculation. The weight coefficients of the preprocessed data are calculated to ensure that the influence degree of different types of data on the final result during the clustering process meets the requirements of the actual circuit characteristics. The weight coefficient calculation is based on principal component analysis or information entropy calculation, and different weights are assigned by analyzing the importance of the data. The preprocessed data is multiplied by the corresponding weight coefficients to obtain a weighted data set. The weighted data set is input into the K-means clustering algorithm for iterative calculation to determine the initial cluster center data. During the execution of the K-means algorithm, several initial cluster centers are randomly selected, then the Euclidean distance from each data point to these centers is calculated, and the data points are classified into the corresponding cluster centers according to the minimum distance principle. Subsequently, the new center points of each category are calculated, and this process is repeated continuously until the cluster centers converge or reach the preset number of iterations. The K-means++ algorithm is used to optimize the initial cluster centers to make their distribution more reasonable and improve the stability and convergence speed of clustering. After obtaining the initial cluster center data, the stability index of the clustering is calculated to determine the optimal number of clusters. The stability index is evaluated by the silhouette coefficient, CH index, or DBI. Among them, the silhouette coefficient is used to measure the tightness within each cluster and the separation between different clusters. The CH index measures the clustering quality by calculating the ratio of the within-group dispersion to the between-group dispersion, while the DBI index is used to measure the similarity between different clusters. By comprehensively considering these indicators, it is determined whether the current number of clusters is reasonable, and based on this, the number of clusters of the K-means algorithm is optimized to make the clustering result more stable and reasonable, obtaining the optimized data of the number of clusters. Based on the optimized data of the number of clusters, dynamic clustering is performed on the weighted data set to optimize the clustering process and make it more in line with the actual characteristics of the PCBA circuit board. The dynamic clustering method combines hierarchical clustering or Gaussian mixture model, and automatically adjusts the number of clusters according to the data distribution characteristics, so that the clustering process no longer depends on a fixed number of clusters, but can be adaptively adjusted according to the data characteristics, thereby avoiding the overfitting or underfitting problems that occur when the number of clusters of the traditional K-means algorithm is inappropriate, and making the clustering result more representative. Feature space mapping is performed on the clustering feature vector data to reduce the dimension of the data and extract the core information.Feature space mapping uses methods such as principal component analysis, t-SNE, or kernel principal component analysis to map high-dimensional data to a lower-dimensional feature space, making the data structure more compact and more interpretable. After feature space mapping, a circuit feature data set is finally obtained.

[0019] Step S200: Perform defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; Specifically, perform feature importance ranking and screening on the circuit feature dataset to ensure that the feature point data used for subsequent analysis has the highest representativeness and effectiveness. During this process, adopt feature selection algorithms, such as the information entropy method based on information gain, the LASSO method based on L1 regularization, or the feature importance evaluation method based on random forest, calculate the weights of all features, and arrange them in descending order according to their contribution degrees. Select the most representative feature points from the sorted feature set, and perform stratified sampling on the weight data of these feature points to ensure data balance and avoid affecting the stability of the model in the case of uneven data distribution. At the same time, combined with data similarity analysis, use calculation methods such as Euclidean distance, cosine similarity, or Mahalanobis distance to screen out feature points with higher correlations to form the final target feature point data. Perform data standardization and regularization processing on the target feature point data to eliminate the influence of data with different dimensions on the model and improve the convergence and generalization ability of the model. Standardization adopts the Z-score standardization or the minimum-maximum normalization method to ensure that the mean and variance of all feature point data are within a reasonable range. At the same time, to prevent overfitting problems, perform regularization processing, including L1 regularization and L2 regularization, where L1 regularization is used to further sparsify the feature set, and L2 regularization suppresses the complexity of the model to prevent overfitting. Based on the standardized and regularized target feature point data, construct an initial Kriging surrogate model in combination with the radial basis kernel function. This model is a statistical method based on spatial interpolation and is suitable for the modeling and optimization of nonlinear systems. When constructing the initial Kriging surrogate model, define the basis function, covariance structure, and hyperparameters to ensure its better fitting effect on the circuit feature dataset. Substitute the first surrogate model parameters of the initial Kriging surrogate model into the maximum likelihood estimation function to optimize the model parameters so that it can more accurately describe the probability distribution of the PCBA circuit feature data. The maximum likelihood estimation function adjusts the model parameters by maximizing the likelihood function value of the sample data so that it can achieve the best fit on the given circuit feature dataset. By iteratively calculating the gradient of the maximum likelihood estimation, adjust the kernel function parameters, covariance parameters, and noise parameters of the Kriging surrogate model to finally obtain the second surrogate model parameters. Based on the optimized second surrogate model parameters, perform structural optimization and online update on the initial Kriging surrogate model to improve the self-adaptability and dynamic response ability of the model. During the structural optimization process, find the optimal kernel function parameters through grid search, Bayesian optimization, or genetic algorithm, and evaluate the generalization performance of the model based on the cross-validation method. At the same time, considering that the defect modes of the PCBA circuit board change over time and environmental conditions, perform online update of the model parameters, adopt adaptive filtering or incremental learning methods, so that the Kriging surrogate model can continuously optimize its own structure as the data stream is input to form the final target Kriging surrogate model.Based on this objective Kriging surrogate model, dynamic defect classification is performed on the circuit feature dataset. The trained surrogate model is used to predict the defect categories of different PCBA circuit boards, such as problems like solder joint dry joints, short circuits, open circuits, overheating, etc. And the classification boundary is continuously updated according to the distribution of the feature space to ensure the accuracy and robustness of the classification results, obtaining the dynamic defect detection results. Bayesian probability analysis is carried out on the dynamic defect detection results to determine the probability distribution of different defect types. Bayesian probability analysis is a reasoning method based on conditional probability. According to the existing defect detection data, the probability of each defect occurrence is calculated, and the prediction results are continuously updated by combining the prior distribution and the posterior distribution. For example, for a detected solder joint defect, its probability under different environmental conditions is calculated through Bayes' theorem and corrected by combining historical data to obtain a more realistic defect probability distribution. Based on this defect type probability distribution data, component cascade failure analysis is carried out to evaluate the fault propagation effect between different components. For example, if the failure probability of a certain capacitor is relatively high, it will affect the connected resistors and power supply modules, resulting in a larger range of circuit failures. By constructing a cascade failure model, the causal relationship between components is analyzed, and the system-level failure probability of the entire PCBA circuit is calculated based on Markov chains, causal graph analysis, or Bayesian networks. Through this cascade failure analysis method, the failure probability data of each component in the PCBA circuit board are calculated.

[0020] Step S300: Perform singular value decomposition and feature space mapping on the voltage data, current data, and temperature data of the PCBA circuit board to obtain circuit parameter state data; Specifically, the voltage data, current data, and temperature data of the PCBA circuit board are divided into sampling periods to ensure the time consistency and integrity of the data. During the operation of the PCBA circuit board, voltage, current, and temperature data are collected at uneven time intervals. To accurately establish a state model, data are aligned in time sequence through interpolation methods or signal resampling techniques, enabling all data to be calculated on the same time scale. Meanwhile, according to the physical characteristics of the system, a reasonable sampling period is set to ensure that both dynamic changes can be captured and calculation redundancy caused by excessive data sampling is avoided. Through this process, a multi-dimensional time series matrix containing voltage, current, and temperature, namely the operation parameter time series matrix, is constructed. Each column of this matrix represents data at different time points, while each row represents different types of physical parameters. The operation parameter time series matrix is subjected to singular value decomposition operation to extract the main feature subspace of the data. Singular value decomposition is a dimensionality reduction method that maps high-dimensional data to a low-dimensional space while retaining the main information of the data. Therefore, it is used in the state analysis of the PCBA circuit board to remove redundant information and highlight the main influencing factors. By performing singular value decomposition on the time series matrix, it is decomposed into a singular value matrix, a left eigenvector matrix, and a right eigenvector matrix. The larger singular values in the singular value matrix represent the main change patterns of the data, while the smaller singular values are regarded as noise or minor perturbations. By screening the most important singular value components, characteristic subspace data are constructed to describe the main operation characteristics of the PCBA circuit board under different working conditions. The characteristic subspace data are input into a state recursive predictor for parameter recursive calculation to obtain state transition matrix data. The state recursive predictor uses methods such as Kalman filtering, long short-term memory network (LSTM), or autoregressive integrated moving average model (ARIMA). Through these recursive algorithms, based on the historical change patterns of the characteristic subspace data, the future circuit operation trend is inferred, and a state transition matrix for modeling is generated. This matrix describes the circuit state evolution relationship between different time steps and reflects the dynamic change characteristics of the operation parameters of the PCBA circuit board. The state transition matrix data are scanned with a sliding window to analyze the state changes in different time periods and perform prediction error compensation in combination with historical data. The role of the sliding window scan is to calculate the state transition law within the window by setting a time window of a certain length and gradually slide the window to obtain the characteristic changes of the entire time series, extracting potential abnormal signals or change trends. Historical prediction deviation compensation is carried out. The least squares method is used for regression analysis of the error, or a recursive neural network is used to learn the error pattern to calculate the error correction data. The error correction data are input into a prediction period adaptor for dynamic calculation to optimize the prediction result and ensure its consistency with the actual operation state. The prediction period adaptor adaptively adjusts the length of the prediction window according to the change trend of the circuit characteristic data, improving the modeling ability for non-stationary time series.Through this method, it is ensured that a longer prediction period is adopted when the operating state is stable, while the prediction period is shortened during state mutations to enhance the real-time response ability of the model. Deviation analysis is performed on the dynamically calculated prediction data and a preset reference trajectory to calculate parameter deviation data, which is used to measure the difference between the operating state of the current PCBA circuit board and the ideal operating state, thereby identifying abnormal trends or potential fault risks. Based on the parameter deviation data, mapping transformation and state reconstruction are performed on the feature subspace data to finally obtain circuit parameter state data. During the mapping transformation process, manifold learning methods such as t-SNE or locally linear embedding are used to project the data from the high-dimensional space to the low-dimensional state space, making state modeling more intuitive and stable. At the same time, state reconstruction optimizes the relationship between feature variables, enabling the data to more accurately reflect the operating mechanism of the PCBA circuit board. Modeling is performed through principal component regression or generalized linear models to ensure the physical rationality and stability of the state data. Circuit parameter state data is obtained.

[0021] Step S400: Constrained optimization is performed on the circuit control quantity set according to the circuit parameter state data to obtain operation parameter control data; Specifically, a multi-objective optimization function containing parameter tracking error and control energy consumption is constructed based on the circuit parameter status data. The parameter tracking error measures the deviation between the current circuit state and the desired state, while the control energy consumption involves the overall consumption of voltage, current, and power during operation. When constructing the optimization function, the weighted summation method or the Pareto optimal method is adopted to ensure that the two can be balanced during the optimization process, avoiding unreasonable changes in other parameters caused by the optimization of a single objective. To ensure that the optimization function conforms to the physical constraints of the PCBA circuit board, it is constraint-matched with voltage limits, current limits, and power limits. These constraint conditions are obtained from actual circuit design parameters or historical operation data statistics and are introduced into the optimization solution process in the form of linear constraints or non-linear constraints to form control constraint condition data. Based on the control constraint condition data, the prediction time domain is divided to ensure the real-time performance and applicability of the optimization solution. The prediction time domain division splits the optimization problem into multiple discrete time steps according to the dynamic characteristics of the circuit system. Each time step corresponds to an optimization calculation period to ensure the continuous optimization and real-time adjustment of the control parameters. The fixed time step method is used for the division of the prediction time domain, so that the optimization calculation is carried out within a preset time interval. For the PCBA circuit board with fast dynamic change characteristics, the adaptive time step method is adopted, that is, the optimization period is adjusted based on the circuit state change rate, so that the calculation frequency is higher in the high-dynamic stage and lower in the stable stage, thereby reducing unnecessary calculation burdens. After the division is completed, the initial value of the optimization sequence is set to obtain the initial value data of the optimization sequence. After the initial value data of the optimization sequence is determined, the branch and bound search algorithm is used for finite control set optimization to find the optimal control quantity sequence data. The branch and bound search algorithm is a global search method suitable for discrete optimization problems. By recursively dividing the problem space into multiple sub-problems and performing boundary estimation in each sub-problem, impossible solutions are excluded, thus accelerating the search process for the optimal solution. When applying this algorithm, the discrete value set of the control variable is defined, and the objective function and constraint conditions are set. Then, through the heuristic search method, the search range is gradually narrowed, and finally the optimal control quantity sequence data is obtained. This sequence data can ensure that the circuit system achieves the optimal control effect while satisfying the physical constraints. The number of control switches of the optimal control quantity sequence data is constrained to reduce unnecessary switching operations and improve the reliability of the system. By setting the control switch threshold or introducing a penalty term, meaningless switching operations are reduced during the optimization process, and necessary adjustments can still be made when the circuit state changes suddenly. Due to the physical characteristics of the circuit system, some control variables will exceed the safe range, so saturation protection is carried out to ensure that the values of all control variables operate within the safe range. For example, through hardware limiting or software limiting methods, voltage, current, or power exceeding the rated range is avoided to prevent equipment damage or function failure.After the control quantity is corrected, dynamic compensation calculation is performed on it to improve the robustness and adaptability of the control system. The goal of dynamic compensation calculation is to fine-tune the control quantity according to the real-time changes in the circuit operating state to correct the control deviation caused by modeling errors, external disturbances, or unmodeled dynamics. Through these compensation calculations, the final control data of the operating parameters is obtained.

[0022] The control constraint condition data is divided into time steps to ensure that the time scale of the optimization calculation matches the dynamic changes of the circuit system. According to the circuit response characteristics of the PCBA circuit board, the entire optimization time domain is divided into multiple discrete time periods by means of fixed time step division or adaptive step adjustment, and the state recurrence calculation is performed within each time step to generate the initial value data of the optimization sequence. The state recurrence calculation adopts the method of dynamic system modeling, and the circuit state information at the current moment is transmitted to the next moment through the recurrence relationship, so that the optimization sequence can more accurately describe the dynamic evolution process of the circuit system, thereby ensuring that the optimization algorithm can perform calculations based on reasonable initial values. The initial value data of the optimization sequence is input into the upper bound calculation function of the branch and bound algorithm to calculate the upper bound value, so as to determine the initial constraint of the search range. In the branch and bound search algorithm, the upper bound value is used to limit the range of the search space and avoid waste of invalid calculations. The Lagrangian relaxation method or the heuristic search method is used to estimate the upper bound of the initial optimization sequence and obtain the upper bound threshold data. This data is used to constrain the range of the optimal solution during the search process, making the subsequent search process more targeted and improving the calculation efficiency of the algorithm. When performing the branch and bound search, the branch nodes are traversed in breadth-first order, and the objective function values of each node are calculated to obtain the lower bound value data. The relaxation optimization method, such as linear programming or quadratic programming, is used to estimate the lower bound of the objective function, and the active node set data is constructed. This set is used to store the search nodes that may still contain the optimal solution at present to ensure the controllability of the search process. The lower bound value in the active node set data is compared with the upper bound threshold data, and the pruning operation is performed to reduce the search space and improve the calculation efficiency. The pruning method adopts the α-β pruning or the dynamic pruning algorithm, and the pruning strategy is dynamically adjusted according to the updated upper and lower bounds during the search process to ensure that the algorithm does not lose potential optimal solutions while reducing the amount of calculation. After pruning, based on the data of the nodes to be branched, the node with the smallest lower bound value is selected for binary splitting and the upper bound threshold data is updated to ensure the further optimization of the search process. Binary splitting is one of the core steps of the branch and bound algorithm. By splitting the current optimal node into two sub-nodes, the solution space can be searched more finely and gradually converge to the optimal solution. When performing binary splitting, the binary division or integer programming method is adopted, so that each new node can be calculated within the constraint range of the original search space, ensuring that the finally searched solution meets the circuit constraint conditions. After completing the binary splitting, the newly added node data is input into the depth-first traversal search process to optimize the search space, so that the algorithm can gradually approach the optimal solution. By continuously performing the depth-first traversal and continuously optimizing the upper bound threshold during the process of the search space gradually converging, the algorithm can find the optimal control quantity sequence data. The termination condition of the depth-first traversal is that when the difference between the upper bound and the lower bound of the search space is less than a certain preset threshold, or when the search depth reaches the maximum allowable value, the algorithm stops iterating and outputs the final optimal solution.

[0023] Step S500: Based on the component failure probability data and the operation parameter control data, perform coupling optimization to obtain circuit protection control parameters.

[0024] Specifically, classify the failure modes and divide the risk levels of component failure probability data to accurately evaluate the potential failure risks of each component under different operating conditions. The process of failure mode classification is based on historical failure data, experimental test data, and statistical analysis methods. The failure types are divided into common categories such as solder joint open circuit, short circuit, capacitor breakdown, MOSFET overheating failure, etc. Methods such as Bayesian inference, fault tree analysis, or Markov process modeling are used to calculate the occurrence probabilities of each failure mode. Conduct risk level division, that is, according to the severity, impact range, and possible consequences of component failure, use the analytic hierarchy process or fuzzy logic method to divide the failure risk into low risk, medium risk, and high risk levels, and generate component risk assessment data. Construct a hierarchical protection decision table according to different risk levels and associate it with the operating parameter control data to ensure the accuracy and pertinence of the control strategy. The construction of the hierarchical protection decision table defines different levels of protection measures based on the criticality and risk level of the components. Associating and matching the protection decision with the operating parameter control data means judging the corresponding protection level according to the current parameter states such as current, voltage, and temperature, and mapping it to the corresponding control strategy to form control strategy mapping data. Conduct dynamic threshold calculation and response timing analysis on the control strategy mapping data to ensure that the protection measures can be triggered at the optimal time point and avoid false alarms or overprotection problems. The core of dynamic threshold calculation is to conduct real-time assessment of the health status of components according to the operating characteristics and historical trends of the circuit, and adaptively adjust the trigger threshold according to environmental changes. For example, use the exponentially weighted moving average or autoregressive integrated moving average method to calculate the dynamic fluctuation range of key electrical parameters, and optimize the threshold online based on the Bayesian update mechanism. To ensure the timeliness of protection control, conduct response timing analysis, that is, based on the dynamic characteristics of the control system, calculate the time delay from the occurrence of a fault to the execution of the protection measure, and optimize the trigger order and time window of the protection action through timing optimization methods (such as finite state machine model or timing diagram analysis) to ensure that the control response does not interfere with normal operation prematurely and can be executed in time before the failure occurs, generating protection timing control data. Based on the protection timing control data, conduct segmented adjustment and amplitude limiting processing on the operating parameter control data to prevent system instability or control failure caused by excessive adjustment of control parameters. The method of segmented adjustment adopts an adaptive control strategy, that is, different adjustment strategies are set in different operating intervals. For example, in the low-risk interval, a linear proportional adjustment method is used, while in the high-risk interval, a non-linear exponential adjustment method is used, so that the control parameters can transition more smoothly to the safe range. The goal of amplitude limiting processing is to ensure that the adjusted key parameters such as voltage and current do not exceed the safe range. For example, based on the constraint optimization algorithm or the anti-saturation technology of the PID controller, limit and filter the control signal to avoid sudden parameter jumps from impacting the circuit system, forming parameter adjustment scheme data.Perform state transition calculations based on parameter adjustment scheme data to ensure that emergency protection measures can be executed in extreme cases, thereby ensuring the safe operation of the circuit system. The core of the state transition calculation is to judge whether it is necessary to switch from the normal operation mode to the derated operation mode, redundant backup mode or emergency shutdown mode based on the failure risk of components and the current operating parameters, and optimize the state transition strategy based on the Markov decision process or reinforcement learning algorithm, so that the system maximizes its availability while meeting the reliability requirements. When performing emergency protection, verify the safety margin of the protection strategy to ensure that the protection measures will not be mis-triggered or cause cascading failures. For example, based on the Monte Carlo simulation method, calculate the reliability of the protection strategy under different fault scenarios, and adjust the safety margin parameters according to the analysis results to improve the anti-interference ability of the system. Perform redundant backup processing on the emergency protection data to prevent the protection mechanism from failing due to single-point failures or data loss. The redundant backup method uses a dual-channel control or distributed storage mechanism, so that key protection parameters can be synchronized between multiple backup nodes and automatically switch to the backup control strategy when a failure occurs. Obtain the circuit protection control parameters.

[0025] In the embodiment of the present invention, through the combination of weighted K-means clustering and feature space mapping, the false positive rate and missed detection rate of detection are significantly reduced. The Kriging surrogate model and Bayesian probability calculation method are used to achieve accurate assessment of the component failure risk, and the prediction accuracy of the model is ensured through the online parameter update mechanism. The parameter state prediction method based on singular value decomposition and subspace mapping overcomes the disadvantage of strong parameter dependence of the traditional physical model and improves the adaptability of the system to parameter changes. The finite control set optimization strategy combined with the branch and bound algorithm realizes the solution of the optimal control sequence under the condition of limited computing resources, ensuring the real-time performance and executability of the control instructions. By establishing a detection-control coupling optimization mechanism, the defect detection results are organically combined with the control strategy to form a complete closed-loop protection system, effectively improving the overall reliability and safety of the system. The hierarchical protection decision-making and dynamic threshold adjustment method enable the system to adaptively adjust the control strategy according to different risk levels, effectively preventing the cascading failure risk of components.

[0026] In a specific embodiment, the process of executing step S100 may specifically include the following steps: Perform noise reduction and normalization processing on the solder joint image data, component position data, and electrical characteristic data of the PCBA circuit board to obtain preprocessed data; Calculate the weight coefficients for the preprocessed data to obtain a weighted data set; Input the weighted data set into the K-means clustering algorithm for iterative calculation to obtain the initial clustering center data, and calculate the stability index for the initial clustering center data to obtain the optimized clustering number data; Optimize the data according to the number of clusters, perform dynamic clustering on the weighted data set to obtain cluster feature vector data, and perform feature space mapping on the cluster feature vector data to obtain a circuit feature data set.

[0027] Specifically, perform noise reduction and normalization on the solder joint image data, component position data, and electrical characteristic data of the PCBA circuit board to reduce noise interference and ensure the comparability of different types of data. The solder joint image data is affected by factors such as light, reflection, and dust, and Gaussian filtering or adaptive median filtering is used to reduce noise. For example, assume that the solder joint image data is represented as , where and are pixel coordinates, then the calculation formula for Gaussian filtering is:

[0028] where, is the two-dimensional Gaussian kernel function, defined as:

[0029] where, is the standard deviation, which determines the smoothness of the filtering. A larger will result in a stronger blurring effect, while a smaller can better retain image details. Preprocess the component position data. This type of data is structured data represented by two-dimensional or three-dimensional coordinates. The mean normalization method is used to normalize the value range of all data to between [-1,1] or [0,1] to reduce the influence of scale differences. Assume that the component position data consists of samples, and each sample has features, then the normalization calculation method is as follows:

[0030] where, and are the minimum and maximum values of all data respectively. For electrical characteristic data, such as resistance, capacitance, inductance, voltage, current, etc., the order of magnitude of these data varies greatly. The Z-score standardization method is used to ensure that the data mean of different physical quantities is zero and the standard deviation is one. Let the electrical characteristic data set be , then its standardization calculation formula is:

[0031] where, is the sample mean, is the sample standard deviation. The processed data can maintain the relative distribution of the data unchanged while eliminating the dimension difference. After completing the data preprocessing, the weight coefficient is calculated to ensure that in the process of cluster analysis, the influence of various data on the final clustering result conforms to the importance of the actual circuit characteristics. The weight coefficient calculation uses the information entropy method or principal component analysis. Suppose the entropy value of a certain data feature is calculated as follows:

[0032] where, is the probability that the data in the th group appears. The higher the entropy value, the greater the amount of information of this feature, and the weight should be increased accordingly. The calculation of the final weighted data set is expressed as:

[0033] where, is the weight vector, and each represents the importance of different data features. The weighted data set is input into the K-means clustering algorithm for iterative calculation to determine the initial clustering center data. The basic process of K-means clustering includes: randomly selecting data points as the initial clustering centers . Calculate the Euclidean distance from each data point to the clustering center:

[0034] and assign it to the nearest clustering center. Update the clustering center:

[0035] where, is the number of samples belonging to the th cluster. Repeat the above steps until the clustering center no longer changes or meets the convergence condition. After obtaining the initial clustering center data, calculate the stability index of the clustering to optimize the number of clusters. Calculate through the "silhouette coefficient":

[0036] where, represents the average distance of the data point in its belonging cluster, and represents the average distance of the data point to the nearest other cluster. When The closer it is to 1, the better the clustering effect of the data point. Optimize the data according to the number of clusters, and perform dynamic clustering on the weighted data set, that is, adaptively adjust the number of clusters under different data feature distribution conditions to ensure that the clustering process can accurately reflect the topological structure and operating characteristics of the circuit. Through the Gaussian mixture model, the expectation-maximization (EM) algorithm is used to optimize the clustering center. Map the clustering feature vector data to the feature space. Through principal component analysis, the calculation process is as follows. Calculate the data covariance matrix:

[0037] Calculate the eigenvalues and eigenvectors of the covariance matrix:

[0038] Select the eigenvectors corresponding to the largest several eigenvalues as the dimensionality reduction matrix Project the data:

[0039] Through calculation, the circuit feature data set is finally obtained.

[0040] In a specific embodiment, the process of executing step S200 may specifically include the following steps: Sort and screen the feature importance of the circuit feature data set to obtain the feature point weight data, and perform hierarchical sampling and data similarity analysis on the feature point weight data to obtain the target feature point data; Perform data standardization and regularization processing on the target feature point data, and construct an initial Kriging surrogate model in combination with the radial basis kernel function; Substitute the first surrogate model parameters of the initial Kriging surrogate model into the maximum likelihood estimation function to adjust the model parameters and obtain the second surrogate model parameters; Based on the second surrogate model parameters, perform structural optimization and online update of the model parameters on the initial Kriging surrogate model to obtain the target Kriging surrogate model, and perform dynamic defect classification on the circuit feature data set through the target Kriging surrogate model to obtain the dynamic defect detection result; Perform Bayesian probability analysis on the dynamic defect detection result to obtain the defect type probability distribution data, and perform component cascade failure analysis on the defect type probability distribution data to obtain the component failure probability data.

[0041] Specifically, perform importance analysis on the circuit feature data set to determine which features have a high contribution to defect detection and failure prediction. The feature importance ranking uses the entropy weight method based on information gain, and the information gain calculation method is as follows:

[0042] Among them, is the th feature in the circuit feature dataset, is the defect category label, representing the entropy of the defect category, and the calculation method is:

[0043] where is the probability of category , and is the conditional entropy under the given feature . The higher the information gain value, the greater the importance of the feature in distinguishing defect types. Therefore, it is used as the basis for feature screening. After calculating the feature weight data, stratified sampling is performed, that is, the feature points are distributed and sampled according to different weight intervals to ensure the balance of the dataset. Combining data similarity analysis, the Euclidean distance is used to calculate the similarity between feature points:

[0044] where represent two different feature samples, represents the feature dimension. If the distance between two feature points is small, it is considered that their contributions in the classification task are similar. Combine or select the feature point with larger information volume to reduce data redundancy and improve calculation efficiency, forming the target feature point data. Perform data standardization and regularization processing on the target feature point data to ensure the unity of the numerical scale of the data and avoid overfitting problems. Standardization uses the Z-score standardization method, and its calculation formula is as follows:

[0045] where is the feature mean, is the standard deviation. Regularization processing uses the L2 regularization method, that is:

[0046] where is the loss function, is the predicted value, is the true value, is the model parameter, is the regularization coefficient. By imposing constraints on , it prevents the overfitting problem caused by too high model complexity. Based on the standardized and regularized data, combined with the radial basis kernel function, an initial Kriging surrogate model is constructed. This model is a spatial interpolation method, and its prediction function is expressed as:

[0047] where is the global trend term, is the weight coefficient, is the radial basis kernel function:

[0048] where, is the scale parameter of the kernel function, which determines how the similarity between data points affects the prediction result. Substitute the first surrogate model parameter of the Kriging surrogate model into the maximum likelihood estimation function for parameter adjustment, so that the parameters of the model can more accurately describe the probability distribution of the circuit characteristic data. The goal of maximum likelihood estimation is to adjust the model parameters by maximizing the likelihood function value of the sample data, so that it can achieve the best fit on the given data. Its calculation formula is:

[0049] where, is the likelihood function, is the data point under the given model parameters The probability density. The gradient descent method is used to solve the optimal parameters to obtain the optimized second surrogate model parameters. Based on the second surrogate model parameters, the initial Kriging surrogate model is structurally optimized and updated online to improve its self-adaptability and dynamic response ability. The optimization method uses the Bayesian optimization method, that is, by sampling to update the posterior distribution of the surrogate model, so that it can continuously optimize the prediction accuracy with the input of new data, forming the target Kriging surrogate model. Based on this model, dynamic defect classification is performed on the circuit characteristic data set, that is, by learning the mapping relationship between circuit characteristics and defect types, classification prediction is performed on new data points to obtain dynamic defect detection results. After obtaining the dynamic defect detection results, Bayesian probability analysis is performed to determine the probability distribution of different defect types. Bayesian probability analysis is a reasoning method based on conditional probability. According to the existing defect detection data, the probability of each defect occurrence is calculated, and the prediction results are continuously updated in combination with the prior distribution and posterior distribution. For example, for a detected solder joint defect, its posterior probability is calculated by Bayes' theorem:

[0050] where, is the defect category after observing the defect feature The probability of, is the probability of the feature appearing under the condition of known defect categories is the prior probability of the defect category, is the feature The total probability of occurrence. Through this calculation, defect type probability distribution data is obtained. To evaluate the mutual influence between different defect types, component cascade failure analysis is performed on the defect type probability distribution data to study the fault propagation relationship between components. Component cascade failure analysis uses the Markov chain modeling method, and its state transition matrix is calculated as follows:

[0051] where, represents the probability that a component transfers from state to state . By calculating this transition matrix, the failure probability of other related components after a certain component fails is predicted, and complete component failure probability data is obtained.

[0052] In a specific embodiment, the process of executing step S300 may specifically include the following steps: Perform sampling period division and time series alignment on the voltage data, current data, and temperature data of the PCBA circuit board to obtain an operating parameter time series matrix; Perform singular value decomposition operation on the operating parameter time series matrix to obtain eigen-subspace data, and input the eigen-subspace data into a state recurrence predictor for parameter recursive calculation to obtain state transition matrix data; Perform a sliding window scan on the state transition matrix data and perform historical prediction deviation compensation to obtain error correction data; Input the error correction data into a prediction period adaptor for dynamic calculation, perform deviation analysis with a preset reference trajectory to obtain parameter deviation data, and perform mapping transformation and state reconstruction on the eigen-subspace data according to the parameter deviation data to obtain circuit parameter state data.

[0053] Specifically, perform sampling period division and time series alignment on the voltage data, current data, and temperature data of the PCBA circuit board to construct an operating parameter time series matrix. Since the voltage, current, and temperature data in the PCBA circuit board are usually collected by multiple sensors, resampling methods and interpolation methods are used for time series alignment. Let respectively represent the voltage, current, and temperature data collected at time , and the standard sampling period is , then the new sampling time series is defined as:

[0054] where, is the initial time, is the total number of sampling points. For unevenly sampled data points, linear interpolation is used:

[0055] Or spline interpolation is used to fill in the missing data points to obtain a complete time series data matrix of voltage, current, and temperature:

[0056] Among them, represents the final time series matrix of operating parameters. Each column corresponds to the data of one time step, and each row corresponds to the measured values of different physical quantities. Singular value decomposition is performed on the time series matrix to extract the feature subspace data. Singular value decomposition is a dimensionality reduction method that decomposes the high-dimensional data matrix into:

[0057] Among them, is the left singular vector matrix, representing the orthogonal basis of the data; is the diagonal matrix, and the singular values in it represent the main change patterns of the data; is the right singular vector matrix, representing the time pattern of the data. To extract the most important feature subspace data, select the eigenvectors corresponding to the first singular values:

[0058] Among them, is the feature subspace data after dimensionality reduction, which is determined by the cumulative singular value contribution rate. Input the feature subspace data into the state recursive predictor to perform parameter recursive calculation and obtain the state transition matrix data. The state recursive predictor adopts a state space model, assuming that the system state is described by the following equation:

[0059]

[0060] Among them, is the state transition matrix, representing the evolution relationship of the system over time; is the control input matrix, is the external input; is the observation matrix, is the measured value; are the process noise and measurement noise respectively. The state transition matrix is recursively estimated by the least squares method:

[0061] After solving this optimization problem, The optimal estimated value is used to construct the state recurrence equation. The state transition matrix data is scanned by a sliding window to capture short-term dynamic changes and compensate for historical prediction errors. The process of sliding window scanning is to set a window size , and then calculate the state change trend within the window:

[0062] If the observed state and the predicted state have a large error:

[0063] Then error correction is required, and the correction term is calculated by the exponentially weighted moving average method:

[0064] Among them, is the smoothing coefficient. The error correction data is input into the prediction period adaptor for dynamic calculation and deviation analysis with the preset reference trajectory. Assume the ideal operating trajectory is , then the deviation calculation formula is as follows:

[0065] If the deviation exceeds the allowable range, the system parameters need to be adjusted. According to the parameter deviation data, mapping transformation and state reconstruction are performed on the feature subspace data. The mapping transformation uses principal component regression (PCR):

[0066] Among them, is the regression coefficient matrix. After the above steps, the complete circuit parameter state data is finally obtained.

[0067] In a specific embodiment, the process of executing step S400 may specifically include the following steps: Construct a multi-objective optimization function including parameter tracking error and control energy consumption based on the circuit parameter state data, and perform constraint matching between the multi-objective optimization function and voltage limit, current limit, and power limit to obtain control constraint condition data; Based on the control constraint condition data, perform prediction time domain division to obtain the initial value data of the optimization sequence, and input the initial value data of the optimization sequence into the branch and bound search algorithm for finite control set optimization to obtain the optimal control quantity sequence data; Perform control switching times constraint and saturation protection on the optimal control quantity sequence data to obtain control quantity correction data, and perform dynamic compensation calculation on the control quantity correction data to obtain operation parameter control data.

[0068] Specifically, considering the core optimization objectives, namely, minimizing the parameter tracking error and minimizing the control energy consumption, where the parameter tracking error measures the deviation between the current circuit state and the desired target, and the control energy consumption involves the consumption of voltage, current, and power. Therefore, a multi-objective optimization function is defined as follows:

[0069] where, is the optimization objective function; is the time of the actual circuit state; is the target reference trajectory; is the control input vector (including voltage, current, etc.); and are the weight coefficients used to balance the tracking error and the control energy consumption, is the optimization time domain length. To ensure that the optimization process meets the physical constraints of the circuit, the multi-objective optimization function is constrained and matched with the voltage limit, current limit, and power limit. Assume that the maximum allowable voltage, current, and power of the PCBA circuit board are , and , respectively. Then the constraint conditions are expressed as:

[0070] To avoid sudden changes in the circuit during the control process, the rate of change of the control input also needs to be constrained, such as restricting the change speeds of the voltage and current:

[0071] where, and are the maximum allowable change amplitudes. After the above constraint matching, the control constraint condition data is obtained for subsequent optimization calculations. On this basis, the prediction time domain is divided to ensure that the time scale of the optimization calculation adapts to the dynamic changes of the circuit system. The goal of the prediction time domain division is to define an optimization window , and solve the optimal control strategy within this window. Assume that the optimization calculation is discretized with a time step . Then the initial value data of the optimization sequence is expressed as:

[0072] where, is the state transition matrix; is the control input matrix; is the circuit state at the next moment. To solve the optimization sequence, a branch and bound search algorithm is used for finite control set optimization. Branch and bound is a search method for solving integer optimization problems, which decomposes the original problem into multiple sub-problems and reduces the computational complexity through pruning. Define the discrete value set of the control input , and establish upper and lower bounds in the search space:

[0073] where and are the lower and upper bounds of the objective function respectively. Use breadth-first search to traverse the branch nodes and calculate the objective function values of each node:

[0074] Store all the calculated into the active node set, and perform pruning according to the upper and lower bound relationships, that is, if the lower bound value of a certain node is higher than the currently found optimal upper bound value, then directly discard this branch, thereby reducing the amount of calculation. After the search algorithm converges iteratively, the optimal control quantity sequence data is obtained. After obtaining the optimal control quantity sequence, perform control switching times constraint and saturation protection. The constraint objective of the control switching times is to reduce the frequent state switching of the circuit in a short time, thereby improving the system stability. Let be the change times of the control signal, then define a maximum allowable switching threshold :

[0075] If the calculated switching times exceed the threshold , then smooth the control signal through a low-pass filter or an adaptive adjustment strategy to avoid system oscillation. The objective of saturation protection is to ensure that all control variables remain within a safe range, for example, limit the amplitude of the optimal control quantity:

[0076] where and are the minimum and maximum allowable values of the control variable. To improve the dynamic adaptability of the control system, perform dynamic compensation calculation on the control quantity correction data to correct the control deviation caused by modeling errors, external disturbances or system nonlinear effects. Adopt the method of proportional integral derivative (PID) compensation or adaptive feedforward compensation. For example, when using PID compensation, the corrected control quantity is expressed as:

[0077] where is the proportional gain; is the integral gain; is differential gain; is the tracking error. If adaptive feedforward compensation is adopted, the error compensation function is learned by training a neural network model to optimize the control quantity further:

[0078] After dynamic compensation, the operation parameter control data is finally obtained, ensuring that the circuit control of the PCBA circuit board achieves optimal control performance while meeting the limiting conditions such as power, temperature, and current.

[0079] In a specific embodiment, the process of performing steps to divide the prediction time domain based on the control constraint condition data, obtaining the initial data of the optimization sequence, and inputting the initial data of the optimization sequence into the branch and bound search algorithm for finite control set optimization to obtain the optimal control quantity sequence data may specifically include the following steps: Perform time step division and state recurrence calculation on the control constraint condition data to obtain the initial data of the optimization sequence; Input the initial data of the optimization sequence into the upper bound calculation function of the branch and bound algorithm to calculate the upper bound value and obtain the upper bound threshold data; Perform breadth-first traversal on the branch nodes, and calculate the lower bound value data of each node through the objective function to obtain the active node set data; Compare the lower bound value in the active node set data with the upper bound threshold data, and perform pruning operation to obtain the data of the node to be branched; Based on the data of the node to be branched, select the node with the minimum lower bound value for binary splitting branch, update the upper bound threshold data to obtain the new node data, and perform depth traversal on the new node data until the search space converges to obtain the optimal control quantity sequence data.

[0080] Specifically, perform time step division on the control constraint condition data to ensure that the optimization process conforms to the dynamic characteristics of the circuit system. Let the control input of the circuit system be , and the state variable be . The state evolution equation of the system is expressed as:

[0081] Among them, is the state transition matrix, describing the change of the circuit state over time, is the control input matrix, determining the influence of the input on the state . For a discrete system, the time step Divided according to the dynamic response characteristics of the system. For example, for a power control system with fast response, a time step of milliseconds is required, while for a power system operating in a steady state, a time step of seconds or longer is used. The division of the time step follows:

[0082] where is the optimized time step, is the initial time. Perform state recursion calculation on the divided time steps to obtain the initial value data of the optimized sequence. The process of state recursion is based on predicting the future changes of the state based on the dynamic system model. For example:

[0083] where represents the noise term, assuming it follows a zero-mean normal distribution, that is , where is the process noise covariance matrix. In this way, an initial optimized sequence is constructed, that is , as the input for subsequent optimization calculations. Input the initial value data of the optimized sequence into the upper bound calculation function of the branch and bound algorithm to calculate the upper bound threshold data. The branch and bound algorithm is a search method for optimization problems, calculating an upper bound value and a lower bound value , and using these bounds to prune during the search process, thereby accelerating the solution of the optimal solution. Define the optimization objective function:

[0084] where is the optimization objective function, representing the state tracking error and control energy consumption; is the target reference trajectory; is the weight coefficient, controlling the relative importance of the control error and energy consumption. To calculate the upper bound, use a heuristic method to solve a feasible solution to the optimization problem and use the objective function value of this solution as the upper bound value:

[0085] where is a feasible control input sequence that satisfies the constraint conditions. This upper bound value is used to determine whether to continue exploring a certain branch during the subsequent search process. Perform a breadth-first traversal of the branch nodes to calculate the objective function values of each node and obtain the lower bound value data. The basic idea of breadth-first traversal is to start from the root node (i.e., the initial optimized state), expand each possible combination of control inputs in turn, and calculate its objective function value . For each node , calculate the lower bound value:

[0086] Among them, is a finite set of control inputs. All calculated lower bound values are stored in the active node set, which contains all branch nodes that may still lead to the optimal solution. The lower bound values in the active node set data are compared with the upper bound threshold data, and pruning operations are performed. The principle of pruning is that if the lower bound value of a certain node is greater than the current upper bound value:

[0087] then directly discard this branch because it cannot contain a better solution. Pruning can significantly reduce the search space and improve the computational efficiency. For the remaining data of the nodes to be branched, optimization calculations are performed. In the pruned active node set, select the node with the smallest lower bound value for binary splitting, that is, further divide it into two child nodes, and each child node corresponds to a different control input selection. For each newly generated node, calculate its objective function value and update the upper bound threshold data, that is:

[0088] Among them, is the objective function value of the newly generated control input sequence. If is less than the current upper bound value, then update the upper bound threshold, indicating that a better feasible solution has been found. Perform a depth traversal on the newly added node data, that is, gradually expand the branches downward until the search space converges. During the depth traversal process, each time select the current optimal control sequence for further optimization until all possible branches have been explored, or the lower bound values of all remaining nodes are greater than the current optimal upper bound value, that is:

[0089] At this time, the algorithm terminates and returns the current optimal control quantity sequence data , where: .

[0090] In a specific embodiment, the process of executing step S500 may specifically include the following steps: Classify the failure modes and divide the risk levels of the component failure probability data to obtain component risk assessment data; Construct a hierarchical protection decision table based on the component risk assessment data and perform an associated match with the operating parameter control data to obtain control strategy mapping data; Perform dynamic threshold calculation and response timing analysis on the control strategy mapping data to obtain protection timing control data; Based on the protection timing control data, perform segmented adjustment and amplitude limiting processing on the operating parameter control data to obtain parameter adjustment scheme data; Based on the parameter adjustment scheme data, perform state switching calculations to obtain emergency protection data, and conduct safety margin verification and redundant backup processing on the emergency protection data to obtain circuit protection control parameters.

[0091] Specifically, analyze the failure modes of components and calculate the failure probability based on statistical data or physical modeling methods. Assume the failure probability of a certain component is , which can be modeled by Poisson distribution or Weibull distribution. Among them, Poisson distribution is applicable to random failures, while Weibull distribution is applicable to life-related failures. Assume the time The number of failures within follows Poisson distribution, then the failure probability is expressed as:

[0092] where, is the failure rate, is the operating time. If modeled by Weibull distribution, the failure probability is:

[0093] where, is the characteristic life parameter, is the shape parameter, which determines the growth trend of the failure. According to these calculation methods, classify the failure modes of different components. For example, the failure modes of capacitors include capacitance drift and breakdown short circuit, the failure modes of resistors include open circuit or increased resistance value, and the failure modes of MOSFETs involve thermal failure or gate breakdown. Divide the risk level according to the component failure probability data to evaluate its impact on the system. Define the risk level as:

[0094] where, is the failure consequence coefficient, indicating the degree of influence of the component failure on the circuit system. For example, the failure of a key power management IC causes the entire system to collapse, so its value is relatively high, while the failure of an ordinary bypass capacitor has less impact, and its value is relatively low. Divide the risk level according to the calculated value. After completing the component risk assessment, construct a hierarchical protection decision table and associate it with the operating parameter control data for matching to generate control strategy mapping data. For example, in the hierarchical protection decision table, if the risk level of a certain resistor is medium, adjust its current load to reduce power consumption:

[0095] where, is the derating factor. If the MOSFET is in a high-risk state, perform active voltage limiting control:

[0096] Among them, is the voltage limiting coefficient. Perform dynamic threshold calculation and response timing analysis on the control strategy mapping data to optimize the protection control timing. The goal of dynamic threshold calculation is to ensure that the protection measures are triggered at the appropriate time, avoiding false alarms or delayed responses. Assume that the temperature is one of the key parameters, and its dynamic threshold adopts "Exponentially Weighted Moving Average (EWMA)"

[0097]

[0098] Among them, is the smoothing factor, which controls the weights of new and old data. The response timing analysis is based on the "Finite State Machine (FSM)" model, defining different protection trigger conditions and time windows. For example, if the current exceeds the set threshold and the duration exceeds the set value, then the protection mode is triggered:

[0099] After completing the response timing analysis, perform segmented adjustment and amplitude limiting processing on the operating parameter control data based on the protection timing control data to ensure the safe operation of the circuit. Assume that the circuit parameters (such as voltage, current, power) are affected by the protection strategy, and its adjustment scheme is expressed as:

[0100] Among them, is a piecewise function, which optimizes the adjustment process of circuit parameters based on PID control, fuzzy control or adaptive filtering algorithm. Perform state switching calculation based on the parameter adjustment scheme data to dynamically adjust between different operating modes. For example, when the system enters a high-risk state, switch to the derating mode:

[0101] Among them, and are the derating factors respectively, ensuring that the circuit operates in a low-power mode and reducing overheating and electrical stress. In order to ensure the reliability of the protection control parameters, perform safety margin verification and redundant backup processing on the emergency protection data. The goal of safety margin verification is to ensure that the protection strategy does not affect normal operation. For example, evaluate the distribution of voltage, current and power through Monte Carlo simulation:

[0102] Among them, represents the parameter The probability exceeding the safety threshold is the allowable probability of extreme events (such as 0.01). The redundant backup process adopts a dual-channel protection mechanism, that is: the main-channel protection strategy is mainly adjusted based on real-time monitoring data; the backup-channel protection strategy automatically switches when the main channel fails, such as

[0103] Through the above calculations, complete circuit protection control parameters are obtained to ensure that the PCBA circuit board has intelligent adaptive protection capabilities during operation.

[0104] Please refer to Figure 2 , Figure 2 which is a schematic block diagram of the PCBA circuit board automatic detection device 200 provided by the embodiment of the present application. As Figure 2 shown, the PCBA circuit board automatic detection device 200 includes: A clustering module 210, configured to cluster the solder joint image data, component position data, and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; A detection module 220, configured to perform defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; A mapping module 230, configured to perform singular value decomposition and feature space mapping on the voltage data, current data, and temperature data of the PCBA circuit board to obtain circuit parameter state data; A constraint optimization module 240, configured to perform constraint optimization on the circuit control quantity set according to the circuit parameter state data to obtain operation parameter control data; A coupling optimization module 250, configured to perform coupling optimization based on the component failure probability data and the operation parameter control data to obtain circuit protection control parameters.

[0105] Through the collaborative cooperation of the above-mentioned various components, through the combination of weighted K-means clustering and feature space mapping, the false positive rate and missed detection rate of detection are significantly reduced. By adopting the Kriging surrogate model and Bayesian probability calculation method, the accurate assessment of the failure risk of components is realized, and through the online parameter update mechanism, the prediction accuracy of the model is guaranteed. The parameter state prediction method based on singular value decomposition and subspace mapping overcomes the shortcoming of the strong dependence of traditional physical models on parameters and improves the adaptability of the system to parameter changes. The finite control set optimization strategy combined with the branch and bound algorithm realizes the solution of the optimal control sequence under the condition of limited computing resources, ensuring the real-time performance and executability of control instructions. By establishing a detection-control coupling optimization mechanism, the defect detection results are organically combined with the control strategy to form a complete closed-loop protection system, effectively improving the overall reliability and safety of the system. By adopting the hierarchical protection decision-making and dynamic threshold adjustment method, the system can adaptively adjust the control strategy according to different risk levels, effectively preventing the cascading failure risk of components.

[0106] Please refer to Figure 3 , Figure 3 FIG. is a schematic block diagram of the structure of the PCBA circuit board automatic detection device 300 provided by an embodiment of the present application. The PCBA circuit board automatic detection device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a device bus 303. Among them, the memory 302 may include a non-volatile storage medium and an internal memory.

[0107] The non-volatile storage medium can store a computer program. The computer program includes program instructions. When the program instructions are executed by the processor 301, the processor 301 can be enabled to execute any one of the above PCBA circuit board automatic detection methods.

[0108] The processor 301 is used to provide computing and control capabilities to support the operation of the entire PCBA circuit board automatic detection device 300.

[0109] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can be enabled to execute any one of the above PCBA circuit board automatic detection methods.

[0110] Those skilled in the art can understand that Figure 3 the structure shown in

[0111] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0112] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described PCBA circuit board automatic detection device 300 can refer to the corresponding process of the foregoing PCBA circuit board automatic detection method, and will not be described herein again.

[0113] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0114] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0115] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A PCBA circuit board automatic detection method, characterized in that: include: Cluster the soldering point image data, component location data, and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; Perform defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; Performing singular value decomposition and feature space mapping on the voltage data, current data and temperature data of the PCBA circuit board to obtain circuit parameter state data; Performing constraint optimization on a circuit control quantity set according to the circuit parameter state data to obtain operating parameter control data; Based on the component failure probability data and the operating parameter control data, coupling optimization is performed to obtain circuit protection control parameters.

2. The PCBA circuit board automatic detection method according to claim 1, characterized in that: The clustering of the soldering point image data, component position data and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set includes: Perform noise reduction and normalization on the soldering point image data, component position data and electrical characteristic data of the PCBA circuit board to obtain pre-processed data; Calculating weight coefficients for the preprocessed data to obtain a weighted data set; Inputting the weighted data set into the K-means clustering algorithm for iterative calculation to obtain initial cluster center data, and performing stability index calculation on the initial cluster center data to obtain cluster quantity optimization data; The weighted data set is dynamically clustered according to the cluster quantity optimization data to obtain cluster feature vector data, and feature space mapping is performed on the cluster feature vector data to obtain a circuit feature data set.

3. The PCBA circuit board automatic detection method according to claim 1, characterized in that: The defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data includes: Sorting and screening the circuit feature data set according to feature importance to obtain feature point weight data, and performing stratified sampling and data similarity analysis on the feature point weight data to obtain target feature point data; Performing data standardization and regularization processing on the target feature point data, and building an initial Kriging proxy model in combination with a radial basis kernel function; Bringing the first proxy model parameters of the initial Kriging proxy model into the maximum likelihood estimation function to adjust the model parameters, thereby obtaining the second proxy model parameters; Based on the second proxy model parameters, the initial Kriging proxy model is structurally optimized and the model parameters are updated online to obtain a target Kriging proxy model, and the circuit feature data set is dynamically classified by the target Kriging proxy model to obtain a dynamic defect detection result; A Bayesian probability analysis is performed on the dynamic defect detection result to obtain defect type probability distribution data, and a component cascading failure analysis is performed on the defect type probability distribution data to obtain component failure probability data.

4. The PCBA circuit board automatic detection method according to claim 1, characterized in that: The voltage data, current data and temperature data of the PCBA circuit board are subjected to singular value decomposition and feature space mapping to obtain circuit parameter state data, including: The voltage data, current data and temperature data of the PCBA circuit board are divided into sampling periods and time-series aligned to obtain an operating parameter time-series matrix; Performing a singular value decomposition operation on the operating parameter timing matrix to obtain characteristic subspace data, and inputting the characteristic subspace data into a state recursive predictor to perform parameter recursive calculation to obtain state transfer matrix data; Performing a sliding window scan on the state transfer matrix data and performing historical prediction deviation compensation to obtain error correction data; The error correction data is input into the prediction cycle adaptor for dynamic calculation, and deviation analysis is performed with the preset reference trajectory to obtain parameter deviation data, and the characteristic subspace data is mapped and transformed and the state is reconstructed according to the parameter deviation data to obtain circuit parameter state data.

5. The PCBA circuit board automatic detection method according to claim 1, characterized in that: The constrained optimization of the circuit control quantity set according to the circuit parameter state data to obtain the operation parameter control data includes: Based on the circuit parameter state data, a multi-objective optimization function including parameter tracking error and control energy consumption is constructed, and the multi-objective optimization function is constraint matched with voltage limit, current limit and power limit to obtain control constraint condition data; Based on the control constraint condition data, the prediction time domain is divided to obtain the optimization sequence initial value data, and the optimization sequence initial value data is input into the branch and bound search algorithm to perform finite control set optimization to obtain the optimal control amount sequence data; The optimal control quantity sequence data is subjected to control switching times constraint and saturation protection to obtain control quantity correction data, and the control quantity correction data is subjected to dynamic compensation calculation to obtain operation parameter control data.

6. The PCBA circuit board automatic detection method according to claim 5, characterized in that: The prediction time domain is divided based on the control constraint condition data to obtain the optimization sequence initial value data, and the optimization sequence initial value data is input into the branch and bound search algorithm to perform finite control set optimization to obtain the optimal control amount sequence data, including: Performing time step division and state recursive calculation on the control constraint condition data to obtain optimization sequence initial value data; Inputting the initial value data of the optimized sequence into the upper bound calculation function of the branch and bound algorithm to calculate the upper bound value, and obtaining the upper bound threshold data; Perform breadth-first traversal on the branch nodes, and obtain the lower bound value data of each node through objective function calculation to obtain the active node set data; Compare the lower bound value in the active node set data with the upper bound threshold data, and perform a pruning operation to obtain the node data to be branched; Based on the node data to be branched, the node with the minimum lower boundary value is selected for binary splitting, and the upper boundary threshold data is updated to obtain newly added node data, and the newly added node data is deeply traversed until the search space converges to obtain the optimal control amount sequence data.

7. The PCBA circuit board automatic detection method according to claim 1, characterized in that: The coupling optimization based on the component failure probability data and the operating parameter control data to obtain the circuit protection control parameters includes: Classifying the failure modes and risk levels of the component failure probability data to obtain component risk assessment data; Constructing a hierarchical protection decision table according to the component risk assessment data, and performing correlation matching with the operating parameter control data to obtain control strategy mapping data; Performing dynamic threshold calculation and response timing analysis on the control strategy mapping data to obtain protection timing control data; Based on the protection timing control data, the operation parameter control data is segmented and limited to obtain parameter adjustment scheme data; A state switching calculation is performed based on the parameter adjustment scheme data to obtain emergency protection data, and safety margin verification and redundant backup processing are performed on the emergency protection data to obtain circuit protection control parameters.

8. A PCBA circuit board automatic detection device, characterized in that: The method for automatically detecting a PCBA circuit board according to any one of claims 1 to 7 comprises: The clustering module is used to cluster the soldering point image data, component location data and electrical characteristic data of the PCBA circuit board to obtain a circuit feature data set; A detection module, used to perform defect detection and Bayesian probability calculation based on the circuit feature data set to obtain component failure probability data; A mapping module is used to perform singular value decomposition and feature space mapping on the voltage data, current data and temperature data of the PCBA circuit board to obtain circuit parameter state data; A constraint optimization module, used to perform constraint optimization on a circuit control quantity set according to the circuit parameter state data to obtain operating parameter control data; A coupling optimization module is used to perform coupling optimization based on the component failure probability data and the operating parameter control data to obtain circuit protection control parameters.

9. A PCBA circuit board automatic detection device, characterized in that: The PCBA circuit board automatic detection device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instruction in the memory so that the PCBA circuit board automatic detection device executes the PCBA circuit board automatic detection method according to any one of claims 1 to 7.

Images