Welding quality data protection method based on differential privacy

Abstract

The invention discloses a differential privacy-based welding quality data protection method, which comprises the following steps of: 1, collecting and preprocessing welding quality data to obtain a multi-mode welding quality data set; 2, performing feature extraction on the multi-modal welding quality data set; 3, carrying out privacy protection processing on the welding quality feature set through an improved Rappor differential privacy algorithm; 4, constructing a privacy welding quality map; 5, constructing an adjacent matrix and a Laplacian matrix, and performing eigenvalue decomposition to obtain a welding quality privacy eigenvector set; sixthly, Markov decision processing is conducted on the welding quality privacy feature vector set, and a welding quality privacy protection strategy is generated; and 7, real-time welding quality monitoring is executed, and a welding quality evaluation report is generated. According to the method, the improved Rappor differential privacy algorithm and the Markov decision are combined, and welding quality monitoring and privacy protection are achieved.

Description

Technical Field

[0001] This invention relates to the field of data privacy protection technology, and in particular to a method for protecting welding quality data based on differential privacy. Background Technology

[0002] Welding, an indispensable part of industrial production, is widely used in the manufacturing and maintenance of various mechanical equipment, structural components, and pipelines. With the development of automated welding technology, the monitoring and control of the welding process has gradually become data-driven and intelligent. Especially in welding tasks requiring high precision and high quality, real-time monitoring of welding quality and timely adjustment of welding parameters have become key factors in improving welding quality. However, traditional welding quality monitoring methods still have some limitations, particularly in areas such as data privacy protection, data analysis accuracy, and real-time feedback during the welding process.

[0003] Most existing welding quality monitoring technologies rely on sensors and image acquisition devices to collect key parameters (such as temperature, current, voltage, and gas flow) in real time during the welding process and analyze the data using pre-set algorithms. These technologies typically employ simple statistical analysis methods to evaluate welding quality; however, these methods often fail to adequately consider the balance between welding quality and privacy protection. In many cases, the data from sensors and image acquisition devices may involve sensitive information about users or the factory, making data privacy a significant challenge. Existing technologies often fail to effectively address how to ensure data privacy while guaranteeing welding quality, especially during data transmission, storage, and analysis.

[0004] Existing welding quality data processing methods typically employ traditional feature extraction and pattern recognition techniques, neglecting the complex temporal relationships and interrelationships between multi-dimensional data during the welding process. Current technologies struggle to handle the data correlations within multimodal welding quality datasets, leading to inaccurate data analysis results and consequently affecting the accuracy and real-time performance of welding quality assessments. While some methods have been optimized by incorporating machine learning and data mining techniques, most have failed to effectively integrate features from different data types (such as sensor data and welding image data), thus failing to fully extract the potential information from the data.

[0005] Therefore, how to provide a method for protecting welding quality data based on differential privacy is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a welding quality data protection method based on differential privacy. This invention combines an improved Rapper differential privacy algorithm with multimodal data analysis to resolve the contradiction between welding quality monitoring and data privacy protection. It protects welding data privacy while optimizing welding parameters through feature extraction and Markov decision models, improving the accuracy and real-time performance of welding quality assessment. Dynamically adjusting sensor sampling frequency and accuracy ensures efficient quality control and privacy protection, providing a secure and intelligent welding quality management solution.

[0007] A welding quality data protection method based on differential privacy according to an embodiment of the present invention includes the following steps:

[0008] Step 1: Collect and preprocess welding quality data from multiple sensors and image acquisition devices to obtain a multimodal welding quality dataset with a unified structure;

[0009] Step 2: Extract features from the multimodal welding quality dataset to obtain a welding quality feature set;

[0010] Step 3: Perform privacy protection processing on the welding quality feature set using the improved Rappor differential privacy algorithm to obtain a privacy feature vector set;

[0011] Step 4: Construct a privacy-preserving weld quality map based on the privacy feature vector set;

[0012] Step 5: Based on the privacy-preserving weld quality graph, construct the adjacency matrix and Laplacian matrix, and perform eigenvalue decomposition to obtain the privacy-preserving feature vector set of weld quality;

[0013] Step 6: Perform Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy;

[0014] Step 7: Based on the welding quality privacy protection strategy, perform real-time welding quality monitoring and generate a welding quality assessment report.

[0015] Optionally, the welding quality data includes sensor data and welding images. The sensor data includes temperature data, current data, voltage data, and gas flow data. The welding images include temperature distribution images of the welding area and weld morphology images. The preprocessing includes outlier removal, missing value imputation, timestamp alignment, and normalization of different types of data to obtain a multimodal welding quality dataset with a unified structure.

[0016] Optionally, step two specifically includes:

[0017] Statistical features are calculated from sensor data, and the temporal variation trend of sensor data during the welding process is analyzed by sliding window time series analysis. The statistical features include the mean, standard deviation, maximum, minimum, range, kurtosis, and skewness of temperature, current, voltage, and gas flow during the welding process.

[0018] Temperature distribution features during the welding process are extracted from temperature distribution images by thermal imaging image analysis. These temperature distribution features represent changes in heat flow during welding.

[0019] Weld features are extracted from weld morphology images by edge detection and texture analysis. These features include weld morphology, welding trajectory deviation, crack features, and porosity features during the welding process.

[0020] The statistical characteristics, time-varying trend characteristics, temperature distribution characteristics, and weld morphology characteristics of the sensor data are combined to obtain the welding quality feature set.

[0021] Optionally, the improved Rapper differential privacy algorithm is specifically as follows:

[0022] Discretize each welding quality feature data point in the welding quality feature set to generate discrete data points;

[0023] The discretization process involves dividing the original data into multiple discrete intervals, each of which corresponds to a discrete category label. For each data point in the original data, the value is mapped to the corresponding discrete category label according to the interval in which it is located.

[0024] Calculate the probability distribution difference between the datasets in two adjacent discrete intervals. If the probability distribution difference is less than the preset differential privacy threshold, then the discretization process meets the differential privacy requirements.

[0025] If the difference in probability distributions is greater than or equal to the preset differential privacy threshold, then the discretization process is repeated until the discretization process meets the differential privacy requirements.

[0026] For each discrete data point in a discrete interval, a random response probability is set, and a random number is generated for each discretized data point;

[0027] If the generated random number is less than the probability of a random response, return the original data value;

[0028] If the generated random number is greater than or equal to the random response probability, a noise value is randomly generated from the predetermined Laplace noise distribution and returned.

[0029] The standard deviation of the noise value is the ratio of the feature sensitivity to the privacy budget for each data point. The privacy budget is determined to have a maximum value based on industry standards and company policies, and different privacy budgets are allocated for different data categories. The feature sensitivity is the maximum range of variation for the corresponding discrete data point.

[0030] All privacy-preserved discrete data points are classified and grouped according to the same feature category;

[0031] For each category of discrete data points, aggregate them separately and perform a chi-square test to generate a privacy-preserving dataset;

[0032] The aggregated privacy-preserving dataset is vectorized by mapping each privacy-preserving data point to a feature vector and then performing Z-score standardization to obtain a privacy feature vector set.

[0033] Optionally, step four specifically involves:

[0034] Each privacy feature vector in the privacy feature vector set is used as a node in the privacy welding quality graph;

[0035] Calculate the Euclidean distance between every two pairs of privacy welding quality map nodes. The Euclidean distance is used to measure the similarity between two privacy feature vectors. The Euclidean distance value is the sum of the squares of the differences between each feature value in the corresponding privacy feature vector and the square root of the sum.

[0036] If the Euclidean distance value is less than the preset distance value, an edge connection is established between the corresponding nodes, and the weight of the edge is the corresponding Euclidean distance value.

[0037] Optionally, step five specifically includes:

[0038] For each pair of nodes in the privacy welding quality graph, if an edge exists, the corresponding element in the adjacency matrix is ​​set to the weight value of the corresponding edge.

[0039] If there are no edges between nodes, the corresponding element in the adjacency matrix is ​​set to 0, where 0 indicates that there is no connection between nodes;

[0040] Calculate the degree of each node and build a degree matrix. The degree is the number of edges connected to the node. The degree matrix is ​​a diagonal matrix, and each diagonal element in the degree matrix represents the degree of the corresponding node.

[0041] The difference between the degree matrix and the adjacency matrix is ​​used to obtain the corresponding Laplacian matrix;

[0042] The Laplacian matrix is ​​decomposed into eigenvalues ​​to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvectors represent the potential structural features of the privacy welding quality map, and the eigenvalues ​​represent the importance of the corresponding eigenvectors.

[0043] Welding quality privacy feature vector set is constructed by selecting the eigenvectors corresponding to the first few smallest eigenvalues.

[0044] Optionally, the Markov decision process specifically includes:

[0045] Set the privacy feature vector with the earliest timestamp in the privacy welding quality graph as the test start node and use it as the current state node;

[0046] Based on the current state node, a state transition space is constructed. The state transition space consists of all directly adjacent nodes of the current state node in the privacy welding quality graph, representing the possible next hop state.

[0047] Assign a state transition probability to each neighboring node in the state transition space. The state transition probability is the ratio of the reciprocal of the Euclidean distance between the current state node and its neighboring nodes to the sum of the reciprocals of the Euclidean distances of all directly adjacent nodes.

[0048] Based on the state transition probability, the optimal next-hop node is selected in the state transition space as the current state transition node and added to the test path sequence. The optimal next-hop node is the neighboring node with the highest state transition probability.

[0049] The optimal next-hop node is selected as the new current state node. The state transition space is built and the state transition operation is executed iteratively until the test path length reaches the predetermined maximum path length, at which point the iteration stops.

[0050] The output consists of a test start node and multiple state transition nodes, which represent the execution order of the welding quality privacy protection strategy.

[0051] Optionally, step seven specifically includes:

[0052] Based on a welding quality privacy protection strategy, welding quality monitoring is performed, and the sampling frequency and accuracy of the sensors are adjusted in real time.

[0053] The quality indicators in the welding process are evaluated through statistical analysis, including temperature fluctuation range, current and voltage fluctuation range, and weld morphology.

[0054] Generate a welding quality assessment report, which includes the trends of welding temperature, current and voltage, sensor accuracy and the rate of change of sampling frequency.

[0055] The beneficial effects of this invention are:

[0056] The differential privacy-based welding quality data protection method proposed in this invention effectively solves the data privacy protection problem while ensuring the accuracy of welding quality monitoring. Existing technologies often fail to adequately balance accurate welding quality assessment and data privacy protection, leading to potential leakage risks during data transmission, storage, and analysis. This invention introduces an improved Rapper differential privacy algorithm to perform privacy protection processing on welding quality data, ensuring that sensitive data such as temperature, current, and voltage involved in the welding process are effectively protected, avoiding the risk of data leakage. Simultaneously, a multimodal data fusion method is employed, combining sensor data and welding image data to overcome the shortcomings of inaccurate data feature extraction and insufficient data correlation in existing technologies, thereby improving the accuracy and real-time performance of welding quality assessment. By constructing a privacy-preserving welding quality map, Laplace matrix, and Markov decision process model, combined with welding quality privacy protection strategies, intelligent adjustment of welding parameters is successfully achieved. By real-time monitoring of key parameters during the welding process and dynamically adjusting the sampling frequency and accuracy of sensors, changes during the welding process can be more accurately grasped, and welding quality assessment reports can be generated in a timely manner, providing a basis for decision-making. This intelligent monitoring system, which combines privacy protection and quality optimization, not only improves the quality control of the welding process but also ensures data privacy security, meeting the current needs of industrial data protection and intelligent welding. Therefore, this invention has significant innovation and application value in welding quality monitoring and privacy protection, providing the welding industry with a safer and more efficient quality management method. Attached Figure Description

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0058] Figure 1 This is an overall flowchart of a welding quality data protection method based on differential privacy proposed in this invention;

[0059] Figure 2 This is a schematic diagram of the improved Rappor differential privacy algorithm processing steps of a welding quality data protection method based on differential privacy proposed in this invention.

[0060] Figure 3 This is a schematic diagram of the Markov decision process for a welding quality data protection method based on differential privacy proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figure 1-3 A method for protecting welding quality data based on differential privacy includes the following steps:

[0063] Step 1: Collect and preprocess welding quality data from multiple sensors and image acquisition devices to obtain a multimodal welding quality dataset with a unified structure;

[0064] Step 2: Extract features from the multimodal welding quality dataset to obtain a welding quality feature set;

[0065] Step 3: Perform privacy protection processing on the welding quality feature set using the improved Rappor differential privacy algorithm to obtain a privacy feature vector set;

[0066] Step 4: Construct a privacy-preserving weld quality map based on the privacy feature vector set;

[0067] Step 5: Based on the privacy-preserving weld quality graph, construct the adjacency matrix and Laplacian matrix, and perform eigenvalue decomposition to obtain the privacy-preserving feature vector set of weld quality;

[0068] Step 6: Perform Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy;

[0069] Step 7: Based on the welding quality privacy protection strategy, perform real-time welding quality monitoring and generate a welding quality assessment report.

[0070] This step ensures the privacy of welding quality data through an improved Rapper differential privacy algorithm. The method improves the accuracy and privacy of data analysis by extracting features from multimodal datasets and combining privacy-preserving welding quality maps and atlas analysis, thus avoiding the risk of data leakage. Furthermore, optimizing the welding quality privacy protection strategy using a Markov decision model further enhances the efficiency of generating real-time welding quality monitoring and evaluation reports, providing an innovative solution for intelligent welding processes and data security.

[0071] In this embodiment, the welding quality data includes sensor data and welding images. The sensor data includes temperature data, current data, voltage data, and gas flow data. The welding images include temperature distribution images of the welding area and weld morphology images. The preprocessing includes outlier removal, missing value imputation, timestamp alignment, and normalization of different types of data to obtain a multimodal welding quality dataset with a unified structure.

[0072] This step, by preprocessing sensor data and welding images, ensures the high quality and consistency of the multimodal welding quality dataset. It eliminates noise and bias from the data acquisition process, improving data accuracy and reliability. This process lays a solid foundation for subsequent feature extraction, privacy protection, and quality assessment, ensuring the precision of welding quality monitoring and the effectiveness of privacy protection.

[0073] In this embodiment, step two specifically includes:

[0074] Statistical features are calculated from sensor data, and the temporal variation trend of sensor data during the welding process is analyzed by sliding window time series analysis. The statistical features include the mean, standard deviation, maximum, minimum, range, kurtosis, and skewness of temperature, current, voltage, and gas flow during the welding process.

[0075] Temperature distribution features during the welding process are extracted from temperature distribution images by thermal imaging image analysis. These temperature distribution features represent changes in heat flow during welding.

[0076] Weld features are extracted from weld morphology images by edge detection and texture analysis. These features include weld morphology, welding trajectory deviation, crack features, and porosity features during the welding process.

[0077] The statistical characteristics, time-varying trend characteristics, temperature distribution characteristics, and weld morphology characteristics of the sensor data are combined to obtain the welding quality feature set.

[0078] This step, by calculating the statistical and temporal trends of sensor data, enables real-time monitoring of fluctuations in key parameters such as temperature and current during the welding process, identifying potential problems. Simultaneously, analysis of thermal imaging and weld morphology images allows for in-depth extraction of temperature distribution and weld morphology characteristics during welding, accurately identifying welding defects such as cracks and porosity. This multi-dimensional, multi-modal data processing approach provides a more comprehensive and accurate foundation for welding quality monitoring, offering reliable data support for subsequent privacy protection and welding quality optimization decisions.

[0079] In this embodiment, the improved Rapper differential privacy algorithm is specifically as follows:

[0080] Discretize each welding quality feature data point in the welding quality feature set to generate discrete data points;

[0081] The discretization process involves dividing the original data into multiple discrete intervals, each of which corresponds to a discrete category label. For each data point in the original data, the value is mapped to the corresponding discrete category label according to the interval in which it is located.

[0082] Calculate the probability distribution difference between the datasets in two adjacent discrete intervals. If the probability distribution difference is less than the preset differential privacy threshold, then the discretization process meets the differential privacy requirements.

[0083] If the difference in probability distributions is greater than or equal to the preset differential privacy threshold, then the discretization process is repeated until the discretization process meets the differential privacy requirements.

[0084] For each discrete data point in a discrete interval, a random response probability is set, and a random number is generated for each discretized data point;

[0085] If the generated random number is less than the probability of a random response, return the original data value;

[0086] If the generated random number is greater than or equal to the random response probability, a noise value is randomly generated from the predetermined Laplace noise distribution and returned.

[0087] The standard deviation of the noise value is the ratio of the feature sensitivity to the privacy budget for each data point. The privacy budget is determined to have a maximum value based on industry standards and company policies, and different privacy budgets are allocated for different data categories. The feature sensitivity is the maximum range of variation for the corresponding discrete data point.

[0088] All privacy-preserved discrete data points are classified and grouped according to the same feature category;

[0089] For each category of discrete data points, aggregate them separately and perform a chi-square test to generate a privacy-preserving dataset;

[0090] The aggregated privacy-preserving dataset is vectorized by mapping each privacy-preserving data point to a feature vector and then performing Z-score standardization to obtain a privacy feature vector set.

[0091] This step utilizes an improved Rapper differential privacy algorithm to effectively protect the privacy of welding quality data and prevent the leakage of sensitive information. By discretizing the welding quality feature data, injecting noise, and employing a random response mechanism, privacy protection is ensured while maintaining data validity. Furthermore, the use of chi-square tests and Z-score standardization ensures that the privacy-protected dataset possesses high quality and reliability. This method, while guaranteeing data privacy, provides precise data support for welding quality monitoring and optimization decisions, thereby improving the safety and quality control level of the welding process.

[0092] In this embodiment, step four specifically includes:

[0093] Each privacy feature vector in the privacy feature vector set is used as a node in the privacy welding quality graph;

[0094] Calculate the Euclidean distance between every two pairs of privacy welding quality map nodes. The Euclidean distance is used to measure the similarity between two privacy feature vectors. The Euclidean distance value is the sum of the squares of the differences between each feature value in the corresponding privacy feature vector and the square root of the sum.

[0095] If the Euclidean distance value is less than the preset distance value, an edge connection is established between the corresponding nodes, and the weight of the edge is the corresponding Euclidean distance value.

[0096] This step constructs a privacy-preserving weld quality graph, transforming privacy-preserving feature vectors from the weld quality data into graph nodes. Euclidean distance is calculated to measure the similarity between data points, ensuring the privacy of the weld data. This method effectively captures the relationships between different weld quality data and establishes node connections based on similarity, optimizing the structured representation of the data. While satisfying privacy requirements, the design of the graph structure and edge weights enhances the accuracy and efficiency of data analysis, providing a more precise basis for weld quality optimization decisions.

[0097] In this embodiment, step five specifically includes:

[0098] For each pair of nodes in the privacy welding quality graph, if an edge exists, the corresponding element in the adjacency matrix is ​​set to the weight value of the corresponding edge.

[0099] If there are no edges between nodes, the corresponding element in the adjacency matrix is ​​set to 0, where 0 indicates that there is no connection between nodes;

[0100] Calculate the degree of each node and build a degree matrix. The degree is the number of edges connected to the node. The degree matrix is ​​a diagonal matrix, and each diagonal element in the degree matrix represents the degree of the corresponding node.

[0101] The difference between the degree matrix and the adjacency matrix is ​​used to obtain the corresponding Laplacian matrix;

[0102] The Laplacian matrix is ​​decomposed into eigenvalues ​​to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvectors represent the potential structural features of the privacy welding quality map, and the eigenvalues ​​represent the importance of the corresponding eigenvectors.

[0103] Welding quality privacy feature vector set is constructed by selecting the eigenvectors corresponding to the first few smallest eigenvalues.

[0104] This step, by constructing adjacency and degree matrices, clearly represents the relationships and connection strengths between nodes. The calculation of the Laplacian matrix and its eigenvalue decomposition further reveal the global structure of the graph, effectively capturing the implicit patterns in the welding data. Selecting the eigenvector corresponding to the smallest eigenvalue to construct a privacy-preserving feature vector set ensures data privacy protection while enhancing the effectiveness of welding quality monitoring and decision support. This method provides a new solution for the privacy protection and optimization of welding quality data.

[0105] In this embodiment, the Markov decision processing specifically includes:

[0106] Set the privacy feature vector with the earliest timestamp in the privacy welding quality graph as the test start node and use it as the current state node;

[0107] Based on the current state node, a state transition space is constructed. The state transition space consists of all directly adjacent nodes of the current state node in the privacy welding quality graph, representing the possible next hop state.

[0108] Assign a state transition probability to each neighboring node in the state transition space. The state transition probability is the ratio of the reciprocal of the Euclidean distance between the current state node and its neighboring nodes to the sum of the reciprocals of the Euclidean distances of all directly adjacent nodes.

[0109] Based on the state transition probability, the optimal next-hop node is selected in the state transition space as the current state transition node and added to the test path sequence. The optimal next-hop node is the neighboring node with the highest state transition probability.

[0110] The optimal next-hop node is selected as the new current state node. The state transition space is built and the state transition operation is executed iteratively until the test path length reaches the predetermined maximum path length, at which point the iteration stops.

[0111] The output consists of a test start node and multiple state transition nodes, which represent the execution order of the welding quality privacy protection strategy.

[0112] This step optimizes the decision-making process for welding quality monitoring and privacy protection by applying Markov decision processes to the generation of welding quality privacy protection strategies. The starting node is set based on the feature vector of the earliest timestamp in the privacy-preserving welding quality graph. The state transition probabilities of adjacent nodes are calculated using Euclidean distance, and the optimal next-hop node is selected, thus achieving optimal path selection in the welding quality data. This method ensures that welding parameters are precisely adjusted with each state transition during the welding process, achieving the best balance between welding quality and privacy protection. The iteratively generated test path sequences provide privacy protection strategies for different welding processes, effectively improving the accuracy of data analysis and decision-making.

[0113] In this embodiment, step seven specifically includes:

[0114] Based on a welding quality privacy protection strategy, welding quality monitoring is performed, and the sampling frequency and accuracy of the sensors are adjusted in real time.

[0115] The quality indicators in the welding process are evaluated through statistical analysis, including temperature fluctuation range, current and voltage fluctuation range, and weld morphology.

[0116] Generate a welding quality assessment report, which includes the trends of welding temperature, current and voltage, sensor accuracy and the rate of change of sampling frequency.

[0117] This step, by dynamically adjusting the sensor's sampling frequency and accuracy, enables more precise acquisition of key data during the welding process and optimizes data acquisition efficiency as needed. Combined with statistical analysis methods, it provides real-time evaluation of quality indicators such as temperature, current, and voltage during the welding process, ensuring that problems are promptly identified and corrected. Simultaneously, the generated welding quality assessment report details the trends in welding parameters, providing real-time monitoring results and analysis to help optimize the welding process and ensure both quality and privacy protection.

[0118] Example 1:

[0119] To verify the feasibility of this invention in practice, it was applied to a modern automobile manufacturing plant where an automated welding system is responsible for welding various parts of the car body together. The cars produced in this plant require high-precision welding quality standards, and due to the industry's stringent privacy requirements, the processing of welding data must ensure privacy and prevent leakage. Traditional welding quality monitoring methods typically rely on sensor data collection and analysis based on simple algorithms; however, these methods are weak in privacy protection and are prone to data leakage risks. Against this backdrop, the differential privacy-based welding quality data protection method proposed in this invention is implemented. By introducing an improved Rapper differential privacy algorithm, it provides more secure and efficient privacy protection for sensor data during the welding process.

[0120] During the welding process at this factory, sensors and image acquisition equipment collect real-time sensor data such as welding temperature, current, voltage, and gas flow rate, as well as temperature distribution images and weld morphology images of the welding area. Each set of data undergoes preprocessing, including outlier removal, missing value imputation, timestamp alignment, and normalization, to generate a unified multimodal welding quality dataset. These datasets contain key parameters at every moment during the welding process. In traditional methods, this data is directly used for quality assessment and adjustment of welding parameters without effective privacy protection. In the implementation of this invention, this data is protected for privacy using an improved Rapper differential privacy algorithm, ensuring the privacy of the welding data.

[0121] For each data point in the welding quality feature set, it is first discretized, mapping it to multiple discrete intervals, each corresponding to a discrete category label. Then, the difference in probability distribution between adjacent discrete intervals is calculated to determine if differential privacy requirements are met. If the distribution difference exceeds a preset differential privacy threshold, the system readjusts the discretization process until privacy protection requirements are met. Privacy protection for each data point is ensured through a random response mechanism and noise injection. In this way, the system can provide reliable data support for subsequent quality assessment and strategy optimization while ensuring data privacy. During the construction of the privacy-preserving welding quality graph, each privacy feature vector corresponds to a node, and the edges between nodes represent the similarity between data features. The edge weights are measured by calculating Euclidean distance. The privacy-preserving welding quality graph provides a structured way to analyze the relationships between data. By constructing adjacency matrices and Laplacian matrices and performing eigenvalue decomposition, a set of welding quality privacy feature vectors is obtained. These feature vector sets contain all the important features of the welding process and are used for subsequent welding quality privacy protection strategy generation.

[0122] In its practical application at this factory, the Markov decision model generates a welding quality privacy protection strategy based on a privacy-preserving welding quality map and feature vector set. This strategy allows the system to adjust welding parameters such as temperature, current, and gas flow rate in real time during the welding process, maximizing welding quality while ensuring data privacy. The real-time monitoring system adjusts the sensor sampling frequency and accuracy according to the generated strategy to ensure timely response to changes during the welding process and generates a detailed quality report based on the real-time assessment of welding quality. This report not only includes the changing trends of parameters such as temperature, current, and voltage during the welding process but also the rate of change of sensor accuracy and sampling frequency, further enhancing the controllability of welding quality and ensuring data privacy.

[0123] To evaluate the privacy protection effectiveness of the method of this invention, comparative experiments were conducted. The data compared included temperature fluctuations, gas flow rate changes, and the accuracy of welding defect detection during the welding process.

[0124] Table 1. Comparison of experimental data between the traditional method and the method of this invention.

[0125] parameter Gaussian noise method k-anonymity method l-diversity method Privacy blocking law Method of the present invention Welding temperature fluctuation range (°C) 5.6 5.3 4.9 5.0 2.1 Welding current fluctuation range (A) 3.5 3.3 3.1 3.2 1.4 Welding voltage fluctuation range (V) 2.3 2.0 2.2 2.1 1.0 Accuracy of welding defect detection (%) 82.0 84.0 86.0 83.0 98.0 Privacy leak rate (%) 8.0 10.0 7.0 12.0 2.0

[0126] As shown in Table 1, traditional privacy protection methods generally suffer from significant data deviations in terms of the fluctuation range of welding temperature, current, and voltage, resulting in inaccurate data. In contrast, the method of this invention, through a differential privacy algorithm, can maintain higher accuracy while protecting privacy. Regarding the accuracy of welding defect detection, traditional privacy protection methods are generally inferior to the method of this invention, especially the Gaussian noise method, whose accuracy is significantly reduced due to excessive noise addition. In terms of data privacy protection strength, the privacy leakage rate of the method of this invention is only 2%, significantly lower than that of traditional methods, indicating that this invention can provide stronger data privacy protection.

[0127] Compared with traditional privacy protection methods, the differential privacy-based welding quality data protection method of this invention can more accurately monitor welding quality and more effectively protect data privacy under the same privacy protection conditions. It is suitable for welding quality control fields with high precision requirements and involving sensitive data.

[0128] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for protecting welding quality data based on differential privacy, characterized in that, Includes the following steps: Step 1: Collect and preprocess welding quality data from multiple sensors and image acquisition devices to obtain a multimodal welding quality dataset with a unified structure; Step 2: Extract features from the multimodal welding quality dataset to obtain a welding quality feature set; Step 3: Perform privacy protection processing on the welding quality feature set using the improved Rappor differential privacy algorithm to obtain a privacy feature vector set; Step 4: Construct a privacy-preserving weld quality map based on the privacy feature vector set; Step 5: Based on the privacy-preserving weld quality graph, construct the adjacency matrix and Laplacian matrix, and perform eigenvalue decomposition to obtain the privacy-preserving feature vector set of weld quality; Step 6: Perform Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy; Step 7: Based on the welding quality privacy protection strategy, perform real-time welding quality monitoring and generate a welding quality assessment report.

2. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, The welding quality data includes sensor data and welding images. The sensor data includes temperature data, current data, voltage data, and gas flow data. The welding images include temperature distribution images of the welding area and weld morphology images. The preprocessing includes outlier removal, missing value imputation, timestamp alignment, and normalization of different types of data to obtain a multimodal welding quality dataset with a unified structure.

3. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, Step two specifically involves: Statistical features are calculated from sensor data, and the temporal variation trend of sensor data during the welding process is analyzed by sliding window time series analysis. The statistical features include the mean, standard deviation, maximum, minimum, range, kurtosis, and skewness of temperature, current, voltage, and gas flow during the welding process. Temperature distribution features during the welding process are extracted from temperature distribution images by thermal imaging image analysis. These temperature distribution features represent changes in heat flow during welding. Weld features are extracted from weld morphology images by edge detection and texture analysis. These features include weld morphology, welding trajectory deviation, crack features, and porosity features during the welding process. The statistical characteristics, time-varying trend characteristics, temperature distribution characteristics, and weld morphology characteristics of the sensor data are combined to obtain the welding quality feature set.

4. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, The improved Rapper differential privacy algorithm is specifically as follows: Discretize each welding quality feature data point in the welding quality feature set to generate discrete data points; The discretization process involves dividing the original data into multiple discrete intervals, each of which corresponds to a discrete category label. For each data point in the original data, the value is mapped to the corresponding discrete category label according to the interval in which it is located. Calculate the probability distribution difference between the datasets in two adjacent discrete intervals. If the probability distribution difference is less than the preset differential privacy threshold, then the discretization process meets the differential privacy requirements. If the difference in probability distributions is greater than or equal to the preset differential privacy threshold, then the discretization process is repeated until the discretization process meets the differential privacy requirements. For each discrete data point in a discrete interval, a random response probability is set, and a random number is generated for each discretized data point; If the generated random number is less than the probability of a random response, return the original data value; If the generated random number is greater than or equal to the random response probability, a noise value is randomly generated from the predetermined Laplace noise distribution and returned. The standard deviation of the noise value is the ratio of the feature sensitivity to the privacy budget for each data point. The privacy budget is determined to have a maximum value based on industry standards and company policies, and different privacy budgets are allocated for different data categories. The feature sensitivity is the maximum range of variation for the corresponding discrete data point. All privacy-preserved discrete data points are classified and grouped according to the same feature category; For each category of discrete data points, aggregate them separately and perform a chi-square test to generate a privacy-preserving dataset; The aggregated privacy-preserving dataset is vectorized by mapping each privacy-preserving data point to a feature vector and then performing Z-score standardization to obtain a privacy feature vector set.

5. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, Step four specifically involves: Each privacy feature vector in the privacy feature vector set is used as a node in the privacy welding quality graph; Calculate the Euclidean distance between every two pairs of privacy welding quality map nodes. The Euclidean distance is used to measure the similarity between two privacy feature vectors. The Euclidean distance value is the sum of the squares of the differences between each feature value in the corresponding privacy feature vector and the square root of the sum. If the Euclidean distance value is less than the preset distance value, an edge connection is established between the corresponding nodes, and the weight of the edge is the corresponding Euclidean distance value.

6. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, Step five specifically involves: For each pair of nodes in the privacy welding quality graph, if an edge exists, the corresponding element in the adjacency matrix is ​​set to the weight value of the corresponding edge. If there are no edges between nodes, the corresponding element in the adjacency matrix is ​​set to 0, where 0 indicates that there is no connection between nodes; Calculate the degree of each node and build a degree matrix. The degree is the number of edges connected to the node. The degree matrix is ​​a diagonal matrix, and each diagonal element in the degree matrix represents the degree of the corresponding node. The difference between the degree matrix and the adjacency matrix is ​​used to obtain the corresponding Laplacian matrix; The Laplacian matrix is ​​decomposed into eigenvalues ​​to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvectors represent the potential structural features of the privacy welding quality map, and the eigenvalues ​​represent the importance of the corresponding eigenvectors. Welding quality privacy feature vector set is constructed by selecting the eigenvectors corresponding to the first few smallest eigenvalues.

7. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, The Markov decision process is specifically as follows: Set the privacy feature vector with the earliest timestamp in the privacy welding quality graph as the test start node and use it as the current state node; Based on the current state node, a state transition space is constructed. The state transition space consists of all directly adjacent nodes of the current state node in the privacy welding quality graph, representing the possible next hop state. Assign a state transition probability to each neighboring node in the state transition space. The state transition probability is the ratio of the reciprocal of the Euclidean distance between the current state node and its neighboring nodes to the sum of the reciprocals of the Euclidean distances of all directly adjacent nodes. Based on the state transition probability, the optimal next-hop node is selected in the state transition space as the current state transition node and added to the test path sequence. The optimal next-hop node is the neighboring node with the highest state transition probability. The optimal next-hop node is selected as the new current state node. The state transition space is built and the state transition operation is executed iteratively until the test path length reaches the predetermined maximum path length, at which point the iteration stops. The output consists of a test start node and multiple state transition nodes, which represent the execution order of the welding quality privacy protection strategy.

8. The welding quality data protection method based on differential privacy according to claim 1, characterized in that, Step seven specifically involves: Based on a welding quality privacy protection strategy, welding quality monitoring is performed, and the sampling frequency and accuracy of the sensors are adjusted in real time. The quality indicators in the welding process are evaluated through statistical analysis, including temperature fluctuation range, current and voltage fluctuation range, and weld morphology. Generate a welding quality assessment report, which includes the trends of welding temperature, current and voltage, sensor accuracy and the rate of change of sampling frequency.

Images