CN-122020719-A - Welding quality data protection method based on differential privacy

CN122020719ACN 122020719 ACN122020719 ACN 122020719ACN-122020719-A

Abstract

The invention discloses a welding quality data protection method based on differential privacy, which comprises the following steps of collecting and preprocessing welding quality data to obtain a multi-mode welding quality data set, carrying out feature extraction on the multi-mode welding quality data set, carrying out privacy protection processing on the welding quality feature set through an improved Rappor differential privacy algorithm, constructing a privacy welding quality graph, constructing an adjacent matrix and a Laplace matrix, carrying out feature value decomposition to obtain a welding quality privacy feature vector set, carrying out Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy, and carrying out real-time welding quality monitoring and generating a welding quality assessment report. The invention combines the improved Rappor differential privacy algorithm with the Markov decision to realize welding quality monitoring and privacy protection.

Inventors

FAN YISONG
ZOU MENGMENG

Assignees

中国科学院合肥物质科学研究院

Dates

Publication Date: 20260512
Application Date: 20260207

Claims (8)

1. The welding quality data protection method based on differential privacy is characterized by comprising the following steps of: firstly, acquiring and preprocessing welding quality data from a plurality of sensors and image acquisition equipment to obtain a multi-mode welding quality data set with a uniform structure; step two, extracting characteristics of the multi-mode welding quality data set to obtain a welding quality characteristic set; performing privacy protection processing on the welding quality feature set through an improved Rappor differential privacy algorithm to obtain a privacy feature vector set; step four, constructing a privacy welding quality diagram based on the privacy feature vector set; constructing an adjacent matrix and a Laplacian matrix based on the privacy welding quality diagram, and decomposing the characteristic values to obtain a welding quality privacy characteristic vector set; Carrying out Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy; and seventhly, based on the welding quality privacy protection strategy, performing real-time welding quality monitoring and generating a welding quality evaluation report.
2. The welding quality data protection method based on differential privacy according to claim 1 is characterized in that the welding quality data comprise sensor data and welding images, the sensor data comprise temperature data, current data, voltage data and gas flow data, the welding images comprise temperature distribution images and welding seam morphology images of welding areas, and the preprocessing comprises the steps of performing outlier rejection, missing value filling, time stamp alignment and normalization on different types of data respectively to obtain a multi-mode welding quality data set with a unified structure.
3. The method for protecting welding quality data based on differential privacy according to claim 1, wherein the step two is specifically: Calculating statistical characteristics from the sensor data, and analyzing time variation trend characteristics of the sensor data in the welding process through a sliding window time sequence, wherein the statistical characteristics comprise the mean value, standard deviation, maximum value, minimum value, polar deviation, kurtosis and skewness of temperature, current, voltage and gas flow in the welding process; extracting temperature distribution characteristics in the welding process from the temperature distribution image through thermal imaging image analysis, wherein the temperature distribution characteristics represent the heat flow change of welding; Extracting weld features from the weld morphology image through edge detection and texture analysis, wherein the weld features comprise weld morphology, welding track deviation, crack features and gas hole features in the welding process; And combining the statistical characteristics, the time variation trend characteristics and the temperature distribution characteristics of the sensor data and the welding seam morphological characteristics to obtain a welding quality characteristic set.
4. The welding quality data protection method based on differential privacy according to claim 1, wherein the improved Rappor differential privacy algorithm specifically comprises: discretizing each welding quality characteristic data point in the welding quality characteristic set to generate discrete data points; the discretization processing process is to divide the original data into a plurality of discrete intervals, wherein each discrete interval corresponds to a discrete class label, and each data point of the original data is mapped to the corresponding discrete class label according to the interval where the value is located; Calculating probability distribution differences of data sets in two adjacent discrete intervals, and if the probability distribution differences are smaller than a preset differential privacy threshold value, discretizing to meet the differential privacy requirement; If the probability distribution difference is greater than or equal to a preset differential privacy threshold, carrying out discretization again until the discretization meets the differential privacy requirement; setting a random response probability for discrete data points in each discrete interval, and generating a random number for each discretized data point; If the generated random number is smaller than the random response probability, returning to the original data value; Randomly generating a noise value from a predetermined Laplace noise distribution and returning if the generated random number is greater than or equal to the random response probability; the standard deviation of the noise value is the ratio of the characteristic sensitivity of each data point to the privacy budget, the privacy budget determines the maximum value according to the industry standard and the company policy, different privacy budgets are allocated for different data types, and the characteristic sensitivity is the maximum variation range of the corresponding discrete data points; Classifying all the discrete data points after privacy protection, and grouping according to the same characteristic category; respectively aggregating discrete data points of each category, and carrying out chi-square inspection to generate a privacy protection data set; And carrying out vectorization processing on the aggregated privacy-preserving data set, mapping each privacy-preserving data point onto a feature vector, and carrying out Z-score standardization to obtain a privacy feature vector set.
5. The welding quality data protection method based on differential privacy as set forth in claim 1, wherein the fourth step is specifically: taking each privacy feature vector in the privacy feature vector set as one node in the privacy welding quality graph; Calculating Euclidean distance between every two pairs of privacy welding quality map nodes, wherein the Euclidean distance is used for measuring the similarity of two privacy feature vectors, the Euclidean distance value is the sum of the squares of the difference value of each feature value in the corresponding privacy feature vector, and taking the square root; if the Euclidean distance value is smaller than the preset distance value, 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 as set forth in claim 1, wherein the fifth step is specifically: for each pair of nodes in the privacy welding quality graph, if edges exist, setting the corresponding elements in the adjacent matrix as weight values of the corresponding edges; if no edge exists between the nodes, setting the corresponding element in the adjacent matrix to 0, wherein 0 indicates that no connection exists between the nodes; Calculating the degree of each node and establishing a degree matrix, wherein the degree is the number of edges connected with 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 degree matrix and the adjacent matrix are subjected to difference to obtain a corresponding Laplacian matrix; Decomposing the characteristic values of the Laplace matrix to obtain a group of characteristic values and corresponding characteristic vectors, wherein the characteristic vectors represent potential structural characteristics of the privacy welding quality map, and the characteristic values represent the importance of the corresponding characteristic vectors; And selecting feature vectors corresponding to the first few minimum feature values of the feature values to construct a welding quality privacy feature vector set.
7. The welding quality data protection method based on differential privacy according to claim 1, wherein the markov decision process is specifically: setting a privacy feature vector with the earliest timestamp in the privacy welding quality diagram as a test starting node and taking the privacy feature vector as a current state node; according to the current state node, a state transition space is constructed, wherein the state transition space consists of all direct adjacent nodes of the current state node in the privacy welding quality diagram and represents a possible next-hop state; Allocating a state transition probability for each adjacent node in a state transition space, wherein the state transition probability is the ratio of the inverse Euclidean distance between the current state node and the adjacent node to the sum of the inverse Euclidean distances of all the direct adjacent nodes; Based on the state transition probability, selecting an optimal next-hop node in a state transition space as a current state transition node, and adding the optimal next-hop node into a test path sequence, wherein the optimal next-hop node is an adjacent node with the maximum state transition probability; taking the selected optimal next-hop node as a new current state node, repeatedly iterating to construct a state transition space and executing state transition operation until the test path length reaches a preset maximum path length, and stopping iterating; and outputting a test path sequence formed by the test starting node and a plurality of state transition nodes, wherein the test path sequence represents the execution sequence of the welding quality privacy protection strategy.
8. The welding quality data protection method based on differential privacy according to claim 1, wherein the step seven specifically comprises: Based on a welding quality privacy protection strategy, performing welding quality monitoring, and adjusting the sampling frequency and the accuracy of the sensor in real time; Evaluating quality indexes in the welding process through statistical analysis, wherein the quality indexes comprise a temperature fluctuation range, a current and voltage fluctuation range and a welding seam morphology; a weld quality assessment report is generated that includes the welding temperature, the trend of change in current and voltage, the sensor accuracy, and the rate of change of the sampling frequency.

Description

Welding quality data protection method based on differential privacy Technical Field The invention relates to the technical field of data privacy protection, in particular to a welding quality data protection method based on differential privacy. Background Welding is an indispensable ring in industrial production, and is widely applied to the manufacture and maintenance of various mechanical equipment, structural parts and pipelines. With the development of automated welding technology, the monitoring and control of the welding process has gradually achieved data and intelligence. Especially in the welding task with high precision and high quality requirement, how to monitor the welding quality in real time and adjust the welding parameters in time becomes a key factor for improving the welding quality. However, the conventional welding quality monitoring method still has some limitations, especially in terms of data privacy protection, data analysis accuracy, and real-time feedback during welding. The existing welding quality monitoring technology mostly relies on sensors and image acquisition equipment to acquire key parameters (such as temperature, current, voltage, gas flow and the like) in the welding process in real time, and performs data analysis through a preset algorithm. These techniques typically employ simple statistical analysis methods to evaluate weld quality, however, these methods often fail to adequately account for the balance between weld quality and privacy protection. In many cases, the data of the sensor and the image acquisition device may involve sensitive information of the user or the factory, and thus the data privacy problem becomes a non-negligible challenge. The prior art often fails to effectively solve the problem of how to ensure the privacy of data, especially in the data transmission, storage and analysis processes, while ensuring the welding quality. The conventional welding quality data processing method generally adopts the traditional feature extraction and pattern recognition technology, and ignores the complex time sequence relationship and the interaction between multidimensional data in the welding process. In the prior art, the data relevance in the multi-mode welding quality data set is difficult to process, so that the data analysis result is not accurate enough, and the accuracy and the instantaneity of welding quality evaluation are further affected. While some methods are optimized by introducing machine learning and data mining techniques, most methods fail to efficiently fuse features of different types of data (e.g., sensor data and welding image data), failing to fully mine the underlying information of the data. Therefore, how to provide a welding quality data protection method based on differential privacy is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a welding quality data protection method based on differential privacy, which combines an improved Rappor differential privacy algorithm with multi-mode data analysis to solve the contradiction between welding quality monitoring and data privacy protection. The welding data privacy is protected, and meanwhile, the welding parameters are optimized through feature extraction and a Markov decision model, so that the accuracy and the instantaneity of welding quality assessment are improved. The sampling frequency and the precision of the sensor are dynamically adjusted, so that efficient quality control and privacy protection are ensured, and a safe and intelligent welding quality management scheme is provided. According to the embodiment of the invention, the welding quality data protection method based on differential privacy comprises the following steps: firstly, acquiring and preprocessing welding quality data from a plurality of sensors and image acquisition equipment to obtain a multi-mode welding quality data set with a uniform structure; step two, extracting characteristics of the multi-mode welding quality data set to obtain a welding quality characteristic set; performing privacy protection processing on the welding quality feature set through an improved Rappor differential privacy algorithm to obtain a privacy feature vector set; step four, constructing a privacy welding quality diagram based on the privacy feature vector set; constructing an adjacent matrix and a Laplacian matrix based on the privacy welding quality diagram, and decomposing the characteristic values to obtain a welding quality privacy characteristic vector set; Carrying out Markov decision processing on the welding quality privacy feature vector set to generate a welding quality privacy protection strategy; and seventhly, based on the welding quality privacy protection strategy, performing real-time welding quality monitoring and generating a welding quality evaluation report. Optionally, the welding quality data comprises sensor data and welding images,