Wave soldering quality improving method and system

Abstract

The invention discloses a wave-soldering quality improving method and system, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: collecting the process and sensor data of wave-soldering equipment and a welding spot morphology image, adding a space-time identifier, and transmitting the image to an edge calculation node through an encryption channel; based on a process difference degree evaluation result, performing dynamic weighted fusion on the multi-dimensional feature vectors to generate cross-factory common features, and storing the cross-factory common features in a federal feature library; according to the feature retrieval result set and the rule engine ready state code, generating an optimal parameter combination and converting the optimal parameter combination into a control instruction; and after the control instruction is executed, calculating a deviation value between the actual morphology feature vector and the standard morphology feature vector, adjusting a cross-factory common feature fusion weight, and synchronizing the cross-factory common feature fusion weight to a federal feature library. According to the method, efficient alignment and fusion of multi-source heterogeneous process features are realized through a cross-factory common feature dynamic weighted fusion mechanism and adaptive weight distribution driven by a process difference degree evaluation result.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for improving wave soldering quality. Background Art

[0002] Modern wave soldering process quality control has gradually evolved into an intelligent system driven by multi-source data, integrating with the Industrial Internet of Things to enable process parameter monitoring and solder joint quality assessment. The goal is to deploy multimodal sensor networks (such as thermocouple arrays and laser rangefinders) to collect process parameters such as the soldering temperature field and the state of the molten tin bath in real time, and to use machine vision to extract solder joint topographic features (such as wetting angle and pad coverage). In recent years, the prevalence of edge computing architectures has made it possible to shift data processing forward. Federated learning also provides privacy-preserving solutions for multi-factory data collaboration. Some studies have attempted to use distribution difference metrics such as KL divergence for cross-domain knowledge transfer.

[0003] Existing technologies face the problem of cross-domain feature fusion mismatch in multi-factory collaboration scenarios. Traditional federated learning frameworks directly aggregate node parameters without considering the shift in process feature distribution caused by the heterogeneity of equipment across different production lines. Static weight allocation mechanisms are unable to dynamically adapt to changes in process variability. There is a timing mismatch between process parameter optimization and morphological feature feedback, and delayed feature library updates lead to a mismatch between control command generation and real-time operating conditions, seriously affecting the convergence speed and stability of optimization. Summary of the Invention

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

[0005] Therefore, the present invention provides a wave soldering quality improvement method to solve the cross-domain feature fusion mismatch and real-time feedback lag problems in the dynamic optimization of wave soldering process parameters.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for improving the quality of wave soldering, which comprises collecting process and sensor data and solder joint topography images of wave soldering equipment, adding spatiotemporal identifiers, and transmitting them to an edge computing node via an encrypted channel;

[0008] In the edge computing node, the process, sensor data and solder joint topography images are analyzed to extract multidimensional feature vectors, calculate the KL divergence of the feature distribution of each factory cluster, and generate process difference evaluation results;

[0009] Based on the process difference evaluation results, the multi-dimensional feature vectors are dynamically weighted and fused to generate cross-factory common features, which are then stored in the federated feature library.

[0010] Load the target common feature vector and the associated standard morphology feature vector from the federated feature library, initialize the cross-factory association rule engine, and generate a feature retrieval result set and a rule engine ready status code;

[0011] Generate the optimal parameter combination and convert it into control instructions based on the feature retrieval result set and the rule engine ready status code;

[0012] After executing the control instruction, the deviation between the actual morphology feature vector and the standard morphology feature vector is calculated, and the cross-factory common feature fusion weights are adjusted and synchronized to the federal feature library.

[0013] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the transmission to the edge computing node through the encrypted channel is carried out in the following specific steps:

[0014] The collected process, sensor data and solder joint topography images are divided into blocks according to the data source type to generate a data block set;

[0015] Add a time and space identifier to each data block in the data block set, select a dynamic encryption strategy according to the data block type for encryption, and transmit the encrypted data block to the edge computing node.

[0016] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the process difference evaluation result is generated, the specific steps are as follows:

[0017] Decrypt the encrypted process, sensor data, and solder joint topography images in the edge computing node;

[0018] Parse the decrypted process data to extract the process feature vector, parse the decrypted sensor data to extract the state feature vector, parse the decrypted solder joint topography image to extract the topography feature vector;

[0019] Normalize the process, state and morphology feature vectors and splice them into multi-dimensional feature vectors;

[0020] The multi-dimensional feature vectors are clustered among factories to generate factory cluster division results. The KL divergence of the feature distribution between each factory cluster is calculated to generate the process difference evaluation results.

[0021] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the generation of cross-factory common features and storage in the federal feature library are as follows:

[0022] According to the results of factory cluster division, the multi-dimensional feature vector set of each factory cluster is extracted, the information entropy of each factory cluster is calculated, and the information entropy difference factor is generated;

[0023] Based on the process difference evaluation results and the information entropy difference factor, the fusion weight of each factory cluster is obtained through a dynamic weight function;

[0024] Based on the multidimensional feature vector set of each factory cluster, the direction of the multidimensional feature vector is dynamically adjusted, and weighted fusion is performed according to the fusion weight of each factory cluster to generate cross-factory common features;

[0025] Common features across factories are stored in the federated feature library according to time decay rules and dynamic weight priorities.

[0026] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the generating of the feature retrieval result set and the rule engine ready status code is carried out in the following specific steps:

[0027] Load the target common feature vectors from the federated feature library according to timestamp and dynamic weight priority, and match the associated standard morphological feature vector set through local sensitive hashing;

[0028] Based on the set of standard morphological feature vectors and the target equipment operating parameters, the cross-factory association rule engine is initialized to construct a dynamic constraint space.

[0029] In the dynamic constraint space, feature similarity is compared through the spatiotemporal decay integral function, and the standard morphological feature vectors that meet the preset confidence threshold are screened to generate a feature retrieval result set;

[0030] Generate a rule engine ready status code based on the non-emptiness determination results of the feature retrieval result set and the dynamic constraint space.

[0031] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the optimal parameter combination is generated and converted into a control instruction, the specific steps are as follows:

[0032] Based on the rule engine ready status code, activate the multi-threaded traversal channel in the dynamic constraint space;

[0033] In the multi-threaded traversal channel, the parameter space clustering of the feature retrieval result set is performed to generate candidate parameter combinations, and the optimal parameter combination is selected through nonlinear ratio constraints;

[0034] The optimal parameter combination is converted into binary control instructions through instruction compilation rules.

[0035] As a preferred solution of the wave soldering quality improvement method of the present invention, wherein: the adjustment of the cross-factory common feature fusion weights is synchronized to the federal feature library, the specific steps are as follows:

[0036] After executing the control instruction, the actual solder joint morphology image is collected, the actual morphology feature vector is extracted, and the deviation between the actual morphology vector and the expected morphology vector is calculated by combining the kernel function spatial integration;

[0037] Based on the deviation, a dynamic weight optimization algorithm is used to adjust the cross-factory common feature fusion weight to generate the adjusted cross-factory common feature fusion weight;

[0038] The adjusted cross-factory common feature fusion weights are distributed to the federated feature library for synchronization through the multi-factory data synchronization protocol.

[0039] In a second aspect, the present invention provides a wave soldering quality improvement system, comprising a data acquisition module, a feature analysis module, a dynamic fusion module, a federated loading module, a parameter optimization module, and an update synchronization module;

[0040] The data collection module is used to collect process data, sensor data and solder joint morphology images of wave soldering equipment, add time and space identifiers, and transmit them to the edge computing node through an encrypted channel;

[0041] The feature analysis module is used to analyze process, sensor data and solder joint topography images in the edge computing node, extract multidimensional feature vectors, calculate the KL divergence of the feature distribution of each factory cluster, and generate process difference evaluation results;

[0042] The dynamic fusion module is used to perform dynamic weighted fusion on the multi-dimensional feature vectors based on the process difference evaluation results, generate cross-factory common features, and store them in the federal feature library;

[0043] The federated loading module is used to load the target common feature vector and the associated standard morphology feature vector from the federated feature library, initialize the cross-factory association rule engine and generate a feature retrieval result set and a rule engine ready status code;

[0044] The parameter optimization module is used to generate the optimal parameter combination and convert it into control instructions based on the feature retrieval result set and the rule engine ready status code;

[0045] The update synchronization module is used to calculate the deviation between the actual morphology feature vector and the standard morphology feature vector after executing the control instruction, and adjust the cross-factory common feature fusion weight to synchronize it to the federal feature library.

[0046] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, any step of the method for improving the quality of wave soldering as described in the first aspect of the present invention is implemented.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for improving the quality of wave soldering as described in the first aspect of the present invention is implemented.

[0048] The beneficial effects of the present invention are: through the dynamic weighted fusion mechanism of common features across factories and the adaptive weight allocation driven by process difference evaluation results, efficient alignment and fusion of multi-source heterogeneous process features are achieved; based on the KL divergence of factory cluster distribution differences and combined with the spatiotemporal attenuation factor, a dynamic weight function is constructed to reduce the common feature extraction error; the incremental update strategy of the federated feature library cooperates with the multi-threaded retrieval channel to support the rapid generation of process parameter optimization instructions under real-time working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 Flowchart of the method for improving wave soldering quality.

[0051] Figure 2 Block diagram of a system for improving the quality of wave soldering.

[0052] Figure 3 Flowchart for transmitting encrypted data blocks to edge computing nodes.

[0053] Figure 4 Flowchart for generating process variability assessment results. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented with reference to other methods different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places in this specification does not necessarily refer to the same embodiment, nor does it necessarily refer to separate or selective embodiments that are mutually exclusive with the referenced embodiment.

[0057] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for improving wave soldering quality, comprising the following steps:

[0058] S1. Collect the process, sensor data and solder joint morphology images of the wave soldering equipment, add time and space identifiers, and transmit them to the edge computing node through an encrypted channel.

[0059] Furthermore, the collected process, sensor data and solder joint topography images are divided into blocks according to the data source type to generate a data block set;

[0060] It should be noted that process data refers to the set of temperature curve, solder flow rate, conveyor speed and pressure parameters generated during the operation of wave soldering equipment;

[0061] Sensor data refers to the real-time monitoring physical quantity data stream collected by temperature sensors, flow rate sensors, and pressure sensors;

[0062] A solder joint topography image refers to a two-dimensional digital image of the solder joint surface wetting angle, pad coverage area, and microstructural features captured by a multispectral industrial camera, which includes imaging data in the visible and near-infrared bands.

[0063] Specifically, the process, sensor data and solder joint morphology images of the wave soldering equipment are physically isolated and stored according to the data source type. The process data is divided into continuous data segments according to the time series, and the sensor data is divided into independent storage units according to the acquisition channel number. The solder joint morphology image generates a unique index based on the combination of the image acquisition device identifier and the timestamp to generate a data block set.

[0064] Add a time and space identifier to each data block in the data block set, select a dynamic encryption strategy according to the data block type for encryption, and transmit the encrypted data block to the edge computing node.

[0065] Specifically, a spatiotemporal identifier consisting of a geographic location code and a coordinated universal time stamp is attached to the process data block, sensor data block and solder joint morphology image block in the data block set respectively; the process data block is symmetrically encrypted using the AES-256 algorithm, the sensor data block is asymmetrically encrypted using the RSA-2048 algorithm, and the solder joint morphology image block is encrypted using pixel replacement based on chaotic mapping; the encrypted process data block, sensor data block and solder joint morphology image block are respectively transmitted to the edge computing node through the quantum key distribution channel.

[0066] S2. In the edge computing node, the process, sensor data and solder joint morphology images are analyzed, the multi-dimensional feature vector is extracted, the KL divergence of the feature distribution of each factory cluster is calculated, and the process difference evaluation results are generated.

[0067] Furthermore, the encrypted process, sensor data and solder joint topography images in the edge computing node are decrypted;

[0068] Specifically, the process data block is decrypted using the symmetric key of the AES-256 algorithm, the sensor data block is decrypted using the private key of the RSA-2048 algorithm, and the solder joint morphology image block is restored to its original image data through the inverse pixel permutation operation of the chaotic mapping.

[0069] Parse the decrypted process data to extract the process feature vector, parse the decrypted sensor data to extract the state feature vector, parse the decrypted solder joint topography image to extract the topography feature vector;

[0070] Specifically, a time series wavelet transform is performed on the decrypted process data to extract the frequency domain energy distribution of the temperature curve and the solder flow rate change gradient as the process feature vector; a sliding window statistical method is applied to the decrypted sensor data to calculate the mean, variance and kurtosis of each acquisition channel to generate a state feature vector; morphological processing and region segmentation are performed on the decrypted solder joint morphology image to extract the wetting angle, pad coverage area and microscopic pore distribution geometric parameters as the morphology feature vector.

[0071] Normalize the process, state and morphology feature vectors and splice them into multi-dimensional feature vectors;

[0072] Specifically, the minimum-maximum normalization method is used for the process feature vector to map the frequency domain energy distribution of the temperature curve and the solder flow rate change gradient to the (0,1) interval; the Z-score normalization method is used for the state feature vector to process the mean, variance and kurtosis parameters of the sensor data; the wetting angle, pad coverage area and micropore distribution geometric parameters in the morphology feature vector are subjected to deviation normalization processing; the normalized process feature vector, the standardized state feature vector and the deviation-normalized morphology feature vector are linearly spliced ​​in the order of process feature vector, state feature vector and morphology feature vector to form a multidimensional feature vector containing temperature frequency domain energy, flow rate gradient, sensor statistics and geometric parameters.

[0073] The multi-dimensional feature vectors are clustered among factories to generate factory cluster division results. The KL divergence of the feature distribution between each factory cluster is calculated to generate the process difference evaluation results.

[0074] Specifically, the OPTICS density clustering algorithm is used to perform density clustering analysis on the spliced ​​multidimensional feature vectors containing temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters. The core objects and boundary points of the factory cluster are determined by calculating the reachable distance of the multidimensional feature vectors, and the factory cluster division results are generated according to the density reachability.

[0075] Specifically, the KL divergence of the feature distribution between each factory cluster is calculated to generate the process difference evaluation result, which is expressed as:

[0076]

[0077] Where, D(C i ||C j ) represents the target factory cluster C i With reference to factory cluster C j KL divergence of feature distribution, C i represents the target factory cluster, C j represents the reference plant cluster, i represents the target plant cluster index, j represents the reference plant cluster index, k represents the category index of the feature, Indicates the cumulative sum of all feature categories, w k represents the dynamic weight of the k-th feature, represents the full domain integral, p i,k (x) represents the target factory cluster C i The probability density function of the k-th feature, ln represents the natural logarithm function, p j,k (x) represents the reference factory cluster C j The probability density function of the k-th class feature in , γ represents the time decay factor, and Δt represents the target factory cluster C i With reference to factory cluster C j The time difference of the data collection time, dx represents the differential of the integral variable x, α represents the covariance adjustment coefficient, δ(C i ,C j ) represents the target factory cluster C i With reference to factory cluster C j Feature covariance, δ represents the feature covariance.

[0078] It should be noted that the covariance adjustment coefficient is derived from the need to suppress the redundant correlation of features between factory clusters. It is optimized and determined in the interval [0.1, 1.0] through grid search, and the example value is 0.5; the default value of the time decay factor is 0.1.

[0079] S3. Based on the process difference evaluation results, the multi-dimensional feature vectors are dynamically weighted and fused to generate common features across factories and store them in the federal feature library.

[0080] Furthermore, based on the results of factory cluster division, the multidimensional feature vector set of each factory cluster is extracted, the information entropy of each factory cluster is calculated, and the information entropy difference factor is generated;

[0081] It should be noted that based on the factory cluster division results generated by the OPTICS density clustering algorithm, the factory node indexes contained in each factory cluster are traversed, and the multidimensional feature vectors belonging to the same factory cluster are screened from the multidimensional feature vector set containing temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters. All multidimensional feature vectors belonging to the same factory cluster are arranged in the order of factory node numbers to form a multidimensional feature vector set corresponding to each factory cluster, which is used to calculate the joint information entropy of each factory cluster.

[0082] Specifically, the multidimensional feature vector set of each factory cluster is extracted, and the information entropy of each factory cluster is calculated. The expression is:

[0083]

[0084] In the formula, H(C l ) represents factory cluster C l The information entropy of l represents the factory cluster index, ∫ Ω represents the integral within the domain Ω, Ω represents the domain, p l,k (x) represents factory cluster C l The probability density function of the k-th class feature, λ represents the gradient penalty coefficient, Indicates traversal and summation of all feature dimensions, m represents the feature dimension index, and M represents the total feature dimension. Indicates factory cluster C l The partial derivative of the KL divergence of the specific value of the m-th dimension feature, KL l Indicates factory cluster C l KL divergence of x m Represents the specific value of the m-th dimension feature.

[0085] It should be noted that the gradient penalty coefficient is derived from the need to suppress mutation noise in the feature dimension. It is determined through cross-validation optimization and the example value is 0.2, which can effectively smooth the interference of feature fluctuations on the information entropy calculation while retaining the true process difference signal.

[0086] It should be noted that the joint information entropy of each factory cluster is calculated based on the multidimensional feature vector set of each factory cluster, the information entropy is normalized by the maximum and minimum values, and the exponential decay term is calculated in combination with the KL divergence in the process difference evaluation results. The normalized information entropy is multiplied by the exponential decay term to generate an information entropy difference factor that reflects the uncertainty of the factory cluster characteristic distribution and the process difference.

[0087] Based on the process difference evaluation results and the information entropy difference factor, the fusion weight of each factory cluster is obtained through a dynamic weight function;

[0088] Specifically, the KL divergence and information entropy difference factor in the process difference evaluation results are input into the dynamic weight function. The dynamic weight function assigns negative weights to the information entropy difference factor through exponential operation. The weighted sum of the KL divergence and the information entropy difference factor is converted into a probability distribution through normalization processing to generate the fusion weight of each factory cluster.

[0089] It should be noted that the larger the fusion weight value, the higher the contribution of the corresponding factory cluster in the fusion of cross-factory common features.

[0090] Based on the multidimensional feature vector set of each factory cluster, the direction of the multidimensional feature vector is dynamically adjusted, and weighted fusion is performed according to the fusion weight of each factory cluster to generate cross-factory common features;

[0091] Specifically, the temperature-frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters in the multidimensional feature vector set corresponding to each factory cluster are multiplied by the fusion weight respectively, and the direction and amplitude of the multidimensional feature vector in each feature dimension are adjusted. The adjusted multidimensional feature vector is weighted and summed item by item according to the feature dimension to generate the fused temperature-frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters. The fusion results are combined in the order of process characteristics, state characteristics and morphology characteristics to generate common features across factories.

[0092] Common features across factories are stored in the federated feature library according to time decay rules and dynamic weight priorities.

[0093] It should be noted that the time decay rule refers to the exponential decay of historical data weights over time calculated based on the timestamps stored in common features across factories. The longer the time difference, the greater the decay, which is used to reduce the impact weight of obsolete data in the federated feature library.

[0094] Dynamic weight priority refers to sorting common features across factories according to the fusion weight value generated by the dynamic weight optimization algorithm. The higher the fusion weight value, the higher the storage priority, ensuring that high-contribution features are stored and retrieved first in the federated feature library. When storing, the time-attenuated weight is combined with the dynamic weight priority to generate the final storage order, and the federated feature library metadata index table is updated.

[0095] Specifically, the temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters of the common features across factories are extracted, the current timestamp is associated with the fusion weight value generated by the dynamic weight optimization algorithm, the weight attenuation calculation of the historically stored common features across factories is performed based on the time attenuation rule, the storage priority is determined by sorting the fusion weight values ​​from high to low, the timestamps, fusion weights and feature parameters of the common features across factories are encapsulated into data blocks, the data blocks are written into the designated storage partition of the federated feature library through the multi-factory data synchronization protocol, and the metadata index table of the federated feature library is updated to record the latest storage location and version information.

[0096] S4. Load the target common feature vector and the associated standard morphology feature vector from the federated feature library, initialize the cross-factory association rule engine and generate a feature retrieval result set and a rule engine ready status code.

[0097] Furthermore, the target common feature vectors are loaded from the federated feature library according to timestamp and dynamic weight priority, and the associated standard morphological feature vector set is matched by locality sensitive hashing;

[0098] It should be noted that the target common feature vector refers to the cross-factory common feature generated by weighted fusion of the process feature vectors, state feature vectors and morphology feature vectors of multiple factory clusters through a dynamic weight optimization algorithm. It includes the standardized fusion results of temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters, and is stored in the federated feature library for feature retrieval and process parameter matching. When loading, the latest and highest-weighted feature data is filtered based on timestamps and dynamic weight priorities, and rapid process optimization is achieved through the associated set of standard morphology feature vectors matched by local sensitive hashing.

[0099] Specifically, target common feature vectors that meet the timestamp range are screened according to the metadata index table of the federated feature library, and the temperature-frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters in the target common feature vector are sorted from high to low according to the dynamic weight priority. A local sensitive hash function is applied to the loaded target common feature vector to generate a hash signature, and candidate matches with the set of standard morphological feature vectors are retrieved in the local sensitive hash table based on the hash signature. The Hamming distance between the candidate matches and the target common feature vector is calculated, and the standard morphological feature vector whose Hamming distance is lower than the preset Hamming distance threshold is screened.

[0100] It should be noted that the preset Hamming distance threshold is determined based on the dimension of the feature vector, the hash signature length and the actual mismatch tolerance, and the example value is 10.

[0101] Based on the set of standard morphological feature vectors and the target equipment operating parameters, the cross-factory association rule engine is initialized to construct a dynamic constraint space.

[0102] Specifically, the wetting angle, pad coverage area and micro-pore distribution geometric parameters are extracted from the set of standard morphological feature vectors, combined with the temperature, humidity and voltage data in the operating parameters of the target equipment, and a multi-dimensional constraint boundary is generated through the KL divergence in the process difference evaluation results and the fusion weight in the dynamic weight optimization algorithm. The multi-dimensional constraint boundary is superimposed on the statistical distribution range of the standard morphological feature vector to form a multi-dimensional dynamic constraint space that integrates process characteristics, state characteristics, morphological characteristics and operating parameters. The accuracy of the constraint space boundary is optimized through local sensitive hash matching to complete the initialization of the cross-factory association rule engine.

[0103] In the dynamic constraint space, feature similarity is compared through the spatiotemporal decay integral function, and the standard morphological feature vectors that meet the preset confidence threshold are screened to generate a feature retrieval result set;

[0104] Specifically, a multidimensional constraint boundary integrating process features, state features, morphology features and operating condition parameters is loaded in the dynamic constraint space, and the temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters collected in real time by the target equipment are input into the time-space attenuation integral function. The time-space attenuation integral function calculates the weight attenuation factor of the historical data based on the timestamp difference, and combines the KL divergence and fusion weight in the dynamic constraint space to perform multidimensional feature similarity integral calculation. The similarity scores of the wetting angle, pad coverage area and micropore distribution geometric parameters in the set of standard morphology feature vectors are compared, and the standard morphology feature vectors whose similarity scores exceed the preset confidence threshold are screened. The matched standard morphology feature vectors and the associated process parameters and operating condition parameters are encapsulated into a feature retrieval result set.

[0105] Generate a rule engine ready status code based on the non-emptiness determination results of the feature retrieval result set and the dynamic constraint space.

[0106] Specifically, check whether there are valid parameters for the number of standard morphological feature vector entries in the feature retrieval result set and the multidimensional constraint boundary of the dynamic constraint space. If the feature retrieval result set contains at least one matching standard morphological feature vector and the multidimensional constraint boundary range of the dynamic constraint space is not empty, then the rule engine ready status code generated is a success identifier. If the feature retrieval result set is empty or the multidimensional constraint boundary range of the dynamic constraint space is empty, then the rule engine ready status code generated is an error identifier.

[0107] S5. Generate the optimal parameter combination based on the feature retrieval result set and the rule engine ready status code and convert it into a control instruction.

[0108] Furthermore, based on the rule engine ready status code, a multi-threaded traversal channel in the dynamic constraint space is activated;

[0109] Specifically, when the ready status code of the rule engine is a success identifier, the multi-threaded traversal channel activation instruction in the dynamic constraint space is triggered, the thread pool resources are initialized according to the multi-dimensional constraint boundary parameters of the dynamic constraint space, the process characteristics, state characteristics, morphology characteristics and working condition parameters within the multi-dimensional constraint boundary are divided into multiple parallel processing subspaces, the thread task queue is allocated according to the distribution density of the standard morphology feature vector in the feature retrieval result set, the multi-threaded traversal channel is started to perform local sensitive hash matching and time-space attenuation integral calculation on the feature parameters in each subspace, the calculation results of each subspace are summarized through the thread synchronization mechanism to generate global feature optimization parameters, and the activation and execution of the multi-threaded traversal channel are completed.

[0110] It should be noted that when the ready status code of the rule engine is a success identifier, the multi-threaded traversal channel activation instruction in the dynamic constraint space is terminated, the loaded target common feature vector and standard morphological feature vector set are rolled back to the previous version data block of the federated feature library, and the error status code is broadcast to all factory nodes through the multi-factory data synchronization protocol. The current dynamic constraint space is marked as invalid in the metadata index table of the federated feature library, triggering the re-initialization process of the cross-factory association rule engine, reloading the target common feature vector of the second priority within the timestamp range according to the predefined retry strategy, and re-executing the local sensitive hash matching and dynamic constraint space construction. If the consecutive retry failures exceed the preset number of times, the manual intervention alarm signal is triggered, and the error log is written into the audit trail record of the federated feature library.

[0111] In the multi-threaded traversal channel, the parameter space clustering of the feature retrieval result set is performed to generate candidate parameter combinations, and the optimal parameter combination is selected through nonlinear ratio constraints;

[0112] Specifically, the standard morphological feature vectors and associated process parameters in the feature retrieval result set are mapped to the multidimensional parameter space of the dynamic constraint space. The OPTICS density clustering algorithm is used to perform density clustering analysis on the temperature-frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters in the multidimensional parameter space to identify candidate parameter combination clusters in high-density areas. The confidence scores of the candidate parameter combinations are calculated based on the KL divergence of the fusion weights and process difference evaluation results in the dynamic weight optimization algorithm. The candidate parameter combinations that meet the requirements are screened through nonlinear ratio constraints. The screened candidate parameter combinations are sorted from high to low according to the confidence scores, and the candidate parameter combination with the highest confidence score is extracted as the optimal parameter combination to generate the optimal parameter combination.

[0113] The optimal parameter combination is converted into binary control instructions through instruction compilation rules.

[0114] Specifically, the mapping relationship between temperature frequency domain energy, flow velocity gradient, sensor statistics and geometric parameters and the communication protocol field of the device controller is defined, the temperature frequency domain energy is converted into the controller's 8-bit floating-point binary representation through table lookup mapping and piecewise linear encoding method, the flow velocity gradient is converted into a high-precision 16-bit fixed-point binary string through an iterative closed-loop correction algorithm, the sensor statistics and geometric parameters are block-encoded into Modbus register addresses and data values ​​that can be recognized by the device, the temperature frequency domain energy floating-point binary segment, the flow velocity gradient fixed-point binary segment, the sensor statistics register address binary segment and the geometric parameter data value binary segment are bit-spliced ​​in the protocol order, and a CRC-16 checksum is attached to generate a complete binary control instruction frame, which is then written into the instruction queue of the device controller through the high-speed industrial bus to complete the conversion.

[0115] S6. After executing the control instruction, calculate the deviation between the actual morphology feature vector and the standard morphology feature vector, adjust the cross-factory common feature fusion weights and synchronize them to the federal feature library.

[0116] Furthermore, after executing the control instruction, the actual solder joint morphology image is collected, the actual morphology feature vector is extracted, and the deviation between the actual morphology vector and the expected morphology vector is calculated by combining the kernel function spatial integration;

[0117] Specifically, the surface image of the solder joint after executing the control instruction is collected by an industrial camera, and the morphological segmentation algorithm is applied to remove the image background noise and light interference. The frequency domain energy distribution characteristics of the solder joint contour are extracted based on wavelet transform, and the contrast and entropy of the surface texture are calculated by gray-level co-occurrence matrix. The geometric parameters of the wetting angle and the pad coverage area are extracted based on the Canny edge detection algorithm. The number and distribution density of microscopic pores are statistically analyzed using binary connected domain analysis. The frequency domain energy distribution characteristics, surface texture statistical characteristics, wetting angle, pad coverage area and microscopic pore distribution geometric parameters are combined in the preset order of the morphological feature vector to generate an actual morphological feature vector containing multidimensional morphological parameters.

[0118] It should be noted that the preset order refers to the arrangement rules of the frequency domain energy distribution characteristics, surface texture statistical characteristics, wetting angle, pad coverage area and microscopic pore distribution geometric parameters in the actual morphological feature vector according to the dimensions of process characteristics, state characteristics and morphological characteristics.

[0119] Specifically, the deviation between the actual shape vector and the expected shape vector is calculated by combining the kernel function spatial integration. The expression is:

[0120]

[0121] Where S represents the deviation between the actual shape vector and the expected shape vector, Indicates that in the morphological feature definition domain The integral within, φ(g) represents the spatial attenuation kernel function at position g, g represents the position, exp represents the natural exponential function, ε k Represents the difference sensitivity coefficient of the k-th category feature, f 1,k (g) represents the value of the k-th feature of the actual shape vector at position g, f1 represents the actual shape vector, f 2,k (g) represents the value of the kth feature of the expected shape vector at position g, f2 represents the expected shape vector, σ represents the Gaussian kernel bandwidth, p1(g) represents the joint probability density of the actual shape vector at position g, p1 represents the joint probability density, p2(g) represents the reference probability density of the expected shape vector at position g, p2 represents the reference probability density, β represents the variance suppression factor, Var(g) represents the feature fluctuation variance at position g, and dg represents the three-dimensional spatial differential unit.

[0122] Based on the deviation, a dynamic weight optimization algorithm is used to adjust the cross-factory common feature fusion weight to generate the adjusted cross-factory common feature fusion weight;

[0123] Specifically, the deviation between the actual morphology vector calculated by the spatial integral of the kernel function and the expected morphology vector is input into the dynamic weight optimization algorithm. The dynamic weight optimization algorithm performs nonlinear transformation on the deviation through an exponential function to generate a weight correction factor. The weight adjustment ratio of each factory cluster is calculated by combining the KL divergence and the information entropy difference factor of the process difference evaluation results. The weight correction factor and the weight adjustment ratio are multiplied item by item according to the dimensions of process characteristics, state characteristics and morphology characteristics. The product result is normalized to generate the adjusted cross-factory common feature fusion weight.

[0124] The adjusted cross-factory common feature fusion weights are distributed to the federated feature library for synchronization through the multi-factory data synchronization protocol.

[0125] Specifically, the process feature weights, state feature weights and morphology feature weights in the adjusted cross-factory common feature fusion weights are encapsulated into a data packet according to the field format defined by the multi-factory data synchronization protocol, and a timestamp, version number and CRC-16 checksum are added. The data packet is broadcast to all factory nodes through the distributed message queue of the multi-factory data synchronization protocol. After receiving the data packet, each factory node verifies the CRC-16 integrity and parses the process feature weights, state feature weights and morphology feature weights. The parsing results are written into the weight field of the cross-factory common feature vector according to the index rules of the federal feature library storage partition, and the weight version information corresponding to the timestamp in the federal feature library metadata index table is updated. The status confirmation mechanism of the multi-factory data synchronization protocol is used to confirm that the weight update of all factory nodes is completed, and the global synchronization of the cross-factory common feature fusion weights is completed.

[0126] This embodiment also provides a wave soldering quality improvement system, including: a data acquisition module, a feature analysis module, a dynamic fusion module, a federated loading module, a parameter optimization module and an update synchronization module; the data acquisition module is used to collect the process, sensor data and solder joint morphology images of the wave soldering equipment, add time and space identifiers, and transmit them to the edge computing node through an encrypted channel; the feature analysis module is used to analyze the process, sensor data and solder joint morphology images in the edge computing node, extract multidimensional feature vectors, calculate the KL divergence of the feature distribution of each factory cluster, and generate a process difference evaluation result; the dynamic fusion module is used to, based on the process difference evaluation result, Dynamically weighted fusion of multi-dimensional feature vectors is performed to generate cross-factory common features and store them in the federal feature library; the federal loading module is used to load the target common feature vectors and associated standard morphological feature vectors from the federal feature library, initialize the cross-factory association rule engine and generate a feature retrieval result set and a rule engine ready status code; the parameter optimization module is used to generate the optimal parameter combination and convert it into control instructions based on the feature retrieval result set and the rule engine ready status code; the update synchronization module is used to calculate the deviation between the actual morphological feature vector and the standard morphological feature vector after executing the control instruction, and adjust the cross-factory common feature fusion weight to synchronize it to the federal feature library.

[0127] This embodiment also provides a computer device suitable for the wave soldering quality improvement method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the wave soldering quality improvement method proposed in the above embodiment.

[0128] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented via WIFI, an operator network, NFC (near field communication), or a reference technology. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball, or touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse.

[0129] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for improving the quality of wave soldering as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0130] In summary, the present invention achieves efficient alignment and fusion of multi-source heterogeneous process features through: a dynamic weighted fusion mechanism of common features across factories, adaptive weight allocation driven by process difference evaluation results, and constructs a dynamic weight function based on the KL divergence of factory cluster distribution differences combined with the spatiotemporal attenuation factor to reduce the common feature extraction error; the incremental update strategy of the federated feature library cooperates with the multi-threaded retrieval channel to support the rapid generation of process parameter optimization instructions under real-time working conditions.

[0131] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for improving wave soldering quality, characterized in that: include, Collect process and sensor data of wave soldering equipment and solder joint topography images, add time and space identifiers, and transmit them to edge computing nodes through encrypted channels; In the edge computing node, the process, sensor data and solder joint topography images are analyzed to extract multidimensional feature vectors, calculate the KL divergence of the feature distribution of each factory cluster, and generate process difference evaluation results; Based on the process difference evaluation results, the multi-dimensional feature vectors are dynamically weighted and fused to generate cross-factory common features, which are then stored in the federated feature library. Load the target common feature vector and the associated standard morphology feature vector from the federated feature library, initialize the cross-factory association rule engine, and generate a feature retrieval result set and a rule engine ready status code; Generate the optimal parameter combination and convert it into control instructions based on the feature retrieval result set and the rule engine ready status code; After executing the control instruction, the deviation between the actual morphology feature vector and the standard morphology feature vector is calculated, and the cross-factory common feature fusion weights are adjusted and synchronized to the federal feature library.

2. The method for improving wave soldering quality according to claim 1, wherein: The transmission to the edge computing node through the encrypted channel is as follows: The collected process, sensor data and solder joint topography images are divided into blocks according to the data source type to generate a data block set; Add a time and space identifier to each data block in the data block set, select a dynamic encryption strategy according to the data block type for encryption, and transmit the encrypted data block to the edge computing node.

3. The method for improving wave soldering quality according to claim 2, wherein: The specific steps of generating the process difference evaluation result are as follows: Decrypt the encrypted process, sensor data, and solder joint topography images in the edge computing node; Parse the decrypted process data to extract the process feature vector, parse the decrypted sensor data to extract the state feature vector, parse the decrypted solder joint topography image to extract the topography feature vector; Normalize the process, state and morphology feature vectors and splice them into multi-dimensional feature vectors; The multi-dimensional feature vectors are clustered among factories to generate factory cluster division results. The KL divergence of the feature distribution between each factory cluster is calculated to generate the process difference evaluation results.

4. The method for improving wave soldering quality according to claim 3, wherein: The steps for generating cross-factory common features and storing them in the federated feature library are as follows: According to the results of factory cluster division, the multi-dimensional feature vector set of each factory cluster is extracted, the information entropy of each factory cluster is calculated, and the information entropy difference factor is generated; Based on the process difference evaluation results and the information entropy difference factor, the fusion weight of each factory cluster is obtained through a dynamic weight function; Based on the multidimensional feature vector set of each factory cluster, the direction of the multidimensional feature vector is dynamically adjusted, and weighted fusion is performed according to the fusion weight of each factory cluster to generate cross-factory common features; Common features across factories are stored in the federated feature library according to time decay rules and dynamic weight priorities.

5. The method for improving wave soldering quality according to claim 4, wherein: The specific steps for generating the feature retrieval result set and the rule engine ready status code are as follows: Load the target common feature vectors from the federated feature library according to timestamp and dynamic weight priority, and match the associated standard morphological feature vector set through local sensitive hashing; Based on the set of standard morphological feature vectors and the target equipment operating parameters, the cross-factory association rule engine is initialized to construct a dynamic constraint space. In the dynamic constraint space, feature similarity is compared through the spatiotemporal decay integral function, and the standard morphological feature vectors that meet the preset confidence threshold are screened to generate a feature retrieval result set; Generate a rule engine ready status code based on the non-emptiness determination results of the feature retrieval result set and the dynamic constraint space.

6. The method for improving wave soldering quality according to claim 5, wherein: The specific steps of generating the optimal parameter combination and converting it into control instructions are as follows: Based on the rule engine ready status code, activate the multi-threaded traversal channel in the dynamic constraint space; In the multi-threaded traversal channel, the parameter space clustering of the feature retrieval result set is performed to generate candidate parameter combinations, and the optimal parameter combination is selected through nonlinear ratio constraints; The optimal parameter combination is converted into binary control instructions through instruction compilation rules.

7. The method for improving wave soldering quality according to claim 6, wherein: The specific steps for adjusting the cross-factory common feature fusion weights and synchronizing them to the federated feature library are as follows: After executing the control instruction, the actual solder joint morphology image is collected, the actual morphology feature vector is extracted, and the deviation between the actual morphology vector and the expected morphology vector is calculated by combining the kernel function spatial integration; Based on the deviation, a dynamic weight optimization algorithm is used to adjust the cross-factory common feature fusion weight to generate the adjusted cross-factory common feature fusion weight; The adjusted cross-factory common feature fusion weights are distributed to the federated feature library for synchronization through the multi-factory data synchronization protocol.

8. A wave soldering quality improvement system, based on the wave soldering quality improvement method according to any one of claims 1 to 7, characterized in that: Including data collection module, feature analysis module, dynamic fusion module, federated loading module, parameter optimization module and update synchronization module; The data collection module is used to collect process data, sensor data and solder joint morphology images of wave soldering equipment, add time and space identifiers, and transmit them to the edge computing node through an encrypted channel; The feature analysis module is used to analyze process, sensor data and solder joint topography images in the edge computing node, extract multidimensional feature vectors, calculate the KL divergence of the feature distribution of each factory cluster, and generate process difference evaluation results; The dynamic fusion module is used to perform dynamic weighted fusion on the multi-dimensional feature vectors based on the process difference evaluation results, generate cross-factory common features, and store them in the federal feature library; The federated loading module is used to load the target common feature vector and the associated standard morphology feature vector from the federated feature library, initialize the cross-factory association rule engine and generate a feature retrieval result set and a rule engine ready status code; The parameter optimization module is used to generate the optimal parameter combination and convert it into control instructions based on the feature retrieval result set and the rule engine ready status code; The update synchronization module is used to calculate the deviation between the actual morphology feature vector and the standard morphology feature vector after executing the control instruction, and adjust the cross-factory common feature fusion weight to synchronize it to the federal feature library.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for improving the quality of wave soldering according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for improving the quality of wave soldering according to any one of claims 1 to 7 are implemented.

Images