A metal welding powder detection method and system based on data fusion

Through multi-detection technology and data fusion algorithm, weld powder production process is monitored and adjusted in real time, and abnormal data is identified and processed, which solves the problems of randomness and insufficient monitoring in traditional welding powder quality evaluation and decision-making, and achieves efficient quality control and production Optimization reduces production costs and defect rate.

CN118883883BActive Publication Date: 2024-12-06JIANGSU JINZHUO NEW MATERIAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411394003.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-08
Publication Date
2024-12-06
Estimated Expiration
2044-10-08

AI Technical Summary

Technical Problem

Traditional metal welding powder quality assessment and production decisions lack randomness and innovation, and lack effective quality monitoring and control methods, which leads to the inability to take timely measures before the quality problems are expanded, increasing production costs and defect rates.

Method used

Using the metal welding powder detection method based on data fusion, multi-dimensional material characteristic data is collected through multi-detection technology, and using the co-evolution fusion algorithm and the adaptive extreme isolation forest algorithm to integrate and analyze the data, monitor and adjust production process parameters in real time, identify and process Outliers, dynamically adjust data source weights, optimize quality assessment and production decisions.

Benefits of technology

Improve data consistency and randomness and innovation of decision-making in metal welding powder production, timely identify and deal with quality problems, reduce defect rate and cost, ensure product quality meets standards, improve overall product reliability and market competition force.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118883883B_ABST
    Figure CN118883883B_ABST
Patent Text Reader

Abstract

The present invention discloses a metal welding powder detection method and system based on data fusion, which is used in the detection field. The metal welding powder detection method includes the following steps: obtaining the metal welding powder data to be detected and setting performance indicators; using multiple detection technologies to respectively collect multi-dimensional material property data of the metal welding powder data, and integrating the multi-dimensional material property data according to the co-evolution fusion algorithm to obtain a material property data set; using an adaptive extreme difference isolation forest algorithm to analyze the material property data set, predict the quality of the metal welding powder, and evaluate the quality of the metal welding powder; feeding back the predicted results to the production line in real time to adjust the production process parameters of the metal welding powder. By identifying specific raw materials or production parameters related to low performance, the present invention can adjust the supply chain management or production strategy to improve the overall performance and reliability of the final product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of detection, and in particular, to a metal welding powder detection method and system based on data fusion. Background Art

[0002] Metal welding powder usually refers to powdered materials used in welding or brazing. Its main function is to help melt and combine metals during welding. Welding powder is widely used in welding technology, especially in specific welding environments and materials. Metal welding powder can be classified according to its use and chemical composition. Common classifications include: brazing welding powder, silver welding powder, iron-based welding powder and aluminum-based welding powder.

[0003] Among them, brazing powder is usually used in the brazing process, and its melting point is lower than the melting point of the metal to be welded. Brazing powder contains flux, which can help remove oxides on the surface of the joint, reduce surface tension, and make the filler metal flow and fill more easily; silver solder powder usually contains silver and other metals such as copper and zinc, and is used for brazing at high temperatures. Silver solder powder has good mechanical strength and corrosion resistance and is suitable for welding precision equipment and electronic components; iron-based solder powder is mainly used for welding iron and steel. Iron-based solder powder can help improve the strength and wear resistance of the weld; aluminum-based solder powder is mainly used for welding aluminum materials. Due to the lower melting point of aluminum, aluminum-based solder powder is specially designed to adapt to the physical properties of aluminum.

[0004] In traditional data fusion methods, there may be a lack of effective data consistency optimization and dynamic weight adjustment mechanisms, so that during the data fusion process, the weights of each data source may not be automatically adjusted according to its performance and the latest data, thereby affecting the accuracy and efficiency of the analysis, and may fail to effectively utilize the advantages of each data source, resulting in a waste of resources. In traditional metal welding powder quality assessment and production decision-making, there may be a lack of randomness and innovation. In addition, the lack of effective quality monitoring and control measures makes it impossible to take timely measures before quality problems expand, thereby increasing production costs and defect rates.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In order to overcome the above problems, the present invention aims to propose a metal welding powder detection method and system based on data fusion, with the aim of solving the problems of lack of randomness and innovation in traditional metal welding powder quality assessment and production decision-making. In addition, the lack of effective quality monitoring and control means leads to the inability to take timely measures before quality problems expand, thereby increasing production costs and defect rate problems.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a metal welding powder detection method based on data fusion is provided, and the metal welding powder detection method comprises the following steps:

[0009] S1. Obtain the metal welding powder data to be tested, determine the test target according to the preset requirements, and set the performance index according to the test target;

[0010] S2, using multi-detection technology to collect multi-dimensional material property data of metal welding powder data, and integrating the multi-dimensional material property data according to the co-evolutionary fusion algorithm to obtain a material property data set;

[0011] S3. Use the adaptive extreme isolation forest algorithm to analyze the material property data set, predict the quality of the metal welding powder, and compare it with the performance index to evaluate the quality of the metal welding powder;

[0012] S4. Build a real-time data monitoring terminal, and according to the data monitoring terminal, feed back the predicted results to the production line in real time to adjust the production process parameters of metal welding powder;

[0013] The material property data set is analyzed using the adaptive range isolation forest algorithm to predict the quality of metal welding powder and compare it with the performance index. The quality evaluation of metal welding powder includes the following steps:

[0014] S31. Statistically analyze the unified material property data set to obtain data distribution, and use the isolation forest algorithm to identify outliers in the data set through random subsampling and range analysis;

[0015] S32, constructing an isolation tree model using the screened outlier data, and optimizing the isolation tree model by adjusting parameters of the isolation tree model;

[0016] S33, calculating the score of each outlier point, and performing correlation analysis between the score and the performance index of the metal welding powder to evaluate the influence of the outlier point;

[0017] S34. Analyze the correlation between outlier scores and performance indicators based on the correlation analysis results between outlier scores and performance indicators, and formulate production quality improvement measures for metal welding powder based on the analysis results.

[0018] Optionally, using multiple detection technologies to respectively collect multi-dimensional material property data of metal welding powder data, and using data fusion technology to integrate the multi-dimensional material property data to obtain a material property data set includes the following steps:

[0019] S21, selecting multiple detection technologies suitable for collecting metal welding powder characteristic data, and setting detection parameters;

[0020] S22, preparing the metal welding powder data to be tested, setting the testing conditions, running the multi-testing technology, and collecting the multi-dimensional material property data of the metal welding powder;

[0021] S23, preprocessing the multi-dimensional material property data of the metal welding powder;

[0022] S24. Use a co-evolutionary fusion algorithm to integrate the pre-processed multi-dimensional material property data into a unified material property data set.

[0023] Optionally, integrating the pre-processed multi-dimensional material property data into a unified material property data set using a co-evolutionary fusion algorithm comprises the following steps:

[0024] S241. quantifying the differences between data sources in the multidimensional material property data by calculating the Manhattan distance using known benchmark data;

[0025] S242, according to the calculation results, using the membership function to assign initial weights reflecting corresponding performance to each data source, and dynamically adjust the inertia weight;

[0026] S243, monitor the output of each data source in real time, and dynamically adjust the weight according to the latest performance of the data source using the membership function and the inertia weight adaptive rule;

[0027] S244, using the dynamically adjusted weights, calculating the weighted average of the outputs of each data source, and using the roulette wheel method to select a data source that meets the preset fitness;

[0028] S245, evaluate the quality of the fused material property dataset, and use Manhattan distance to evaluate the consistency between the fused dataset and the preset target; if the evaluation result is satisfactory, output the fused dataset; if the evaluation result is unsatisfactory, return to S241 and optimize the weight adjustment process until the preset target is met.

[0029] Optionally, the formula for quantifying the differences between data sources in the multi-dimensional material property data by calculating the Manhattan distance is:

[0030] ;

[0031] In the formula, Represents the data source i With data source j The distance between the provided datasets;

[0032] Represents the data source i The k Material property values ​​of dimensions;

[0033] Represents the data source j Thek Material property values ​​of dimensions;

[0034] d represents the number of material property dimensions included in each data set;

[0035] Represents the data source i Datasets;

[0036] Represents the data source j of the dataset.

[0037] Optionally, real-time monitoring of the output of each data source and dynamically adjusting the weight according to the latest performance of the data source using the membership function and the inertia weight adaptive rule include the following steps:

[0038] S2421. Quantify the contribution of each data source based on performance indicators;

[0039] S2422, calculating the membership score of each data source using a membership function;

[0040] S2423. Allocate an initial weight to each data source according to the membership score;

[0041] S2424. Setting and implementing a dynamic adjustment mechanism to dynamically adjust the weight of each data source based on real-time performance feedback from the data source;

[0042] The formula for calculating the membership score of each data source using the membership function is:

[0043] ;

[0044] In the formula, μ i Represents the data source i The membership score of

[0045] p i Represents the data source i performance index value;

[0046] a represents the slope coefficient;

[0047] b represents the intercept coefficient.

[0048] Optionally, using the dynamically adjusted weights to calculate the weighted average of the outputs of each data source, and using the roulette wheel method to select a data source that meets the preset fitness includes the following steps:

[0049] S2441. Collect the output data of all data sources, determine the weight of each data source based on the historical performance of the data source, multiply the output of each data source by its corresponding weight, and calculate the sum of all weighted outputs to obtain a weighted average as a comprehensive data point;

[0050] S2442, setting a fitness threshold, and using a fitness function to calculate a fitness score for each data source;

[0051] S2443, calculating the relative probability of each data source based on the fitness score and the comprehensive data points of all data sources;

[0052] S2444, randomly selecting a data source according to the relative probability of the data source using a roulette algorithm;

[0053] S2445. Select a starting point from the data source list and start accumulating the relative probabilities of the data sources. When the accumulated relative probability exceeds or equals the randomly generated value, stop accumulating to obtain the selected data source.

[0054] Optionally, performing statistical analysis on the unified material property data set to obtain data distribution, and using the isolation forest algorithm to identify outliers in the data set through random subsampling and range analysis includes the following steps:

[0055] S311. Analyze the unified material property data set and identify outliers by calculating statistics and evaluating data distribution patterns;

[0056] S312. Use a visual interface to display the distribution of outliers, relationships and correlations between variables, and identify outliers and multicollinearity issues;

[0057] S313, performing random sub-sampling of outliers by using an isolation forest algorithm, calculating the range of features in the random sub-sampling, and identifying outliers in the material property data set according to a set threshold;

[0058] S314. Based on the recognition results, all outliers are located, and the positions and related characteristic values ​​of the outliers in the material property data set are recorded.

[0059] Optionally, the formula for calculating the range of a feature in a random subsampling is:

[0060] ;

[0061] In the formula, Samples max Indicates the maximum value of a specified dimension in a material property data set;

[0062] Samples min Indicates the minimum value of a specified dimension in a material property data set;

[0063] SubSamples max Indicates the maximum value of a specified dimension in a subset of material property data obtained by random sampling;

[0064] SubSamples min Indicates the minimum value of the specified dimension in the sample subset;

[0065] E Represents the range of the feature in the random subsample.

[0066] Optionally, calculating the score of each outlier point and performing correlation analysis between the score and the performance index of the metal welding powder, and evaluating the influence of the outlier point includes the following steps:

[0067] S331, using the optimized isolation tree model to calculate the abnormality score of each outlier point, reflecting the degree of its deviation from normal data;

[0068] S332, for each identified outlier point, collecting metal welding powder performance data related to the outlier point;

[0069] S333. Use statistical methods to quantify the correlation between outlier scores and performance indicators, and identify the performance variables most affected by outliers;

[0070] S334. Display the relationship between the abnormality score of each outlier and the performance index through a visualization platform, and evaluate the relationship between the outlier and the performance index.

[0071] According to another aspect of the present invention, there is also provided a metal welding powder detection system based on data fusion, the metal welding powder detection system comprising: a detection target and performance index setting module, a multi-dimensional data integration module, a quality prediction and evaluation module and a real-time monitoring and production adjustment module;

[0072] The detection target and performance index setting module, the multi-dimensional data integration module, the quality prediction and evaluation module, and the real-time monitoring and production adjustment module are connected in sequence;

[0073] The detection target and performance index setting module is used to obtain the metal welding powder data to be detected, determine the detection target according to the preset requirements, and set the performance index according to the detection target;

[0074] A multi-dimensional data integration module is used to collect multi-dimensional material property data of metal welding powder data by using multiple detection technologies, and integrate the multi-dimensional material property data according to the co-evolution fusion algorithm to obtain a material property data set;

[0075] The quality prediction and evaluation module is used to analyze the material property data set using the adaptive extreme difference isolation forest algorithm, predict the quality of the metal welding powder, and compare it with the performance index to evaluate the quality of the metal welding powder;

[0076] The real-time monitoring and production adjustment module is used to build a real-time data monitoring terminal, and based on the data monitoring terminal, the predicted results are fed back to the production line in real time to adjust the production process parameters of the metal welding powder.

[0077] Compared with the prior art, this application has the following beneficial effects:

[0078] 1. The present invention uses a variety of detection technologies to obtain characteristic data of metal welding powder from multiple dimensions and construct a comprehensive and multidimensional data set. The multidimensional data set provides more comprehensive information support for analysis and decision-making, reduces the deviation and error that may be caused by relying on a single data source, and integrates data from different technologies through a co-evolutionary fusion algorithm, which can not only optimize the consistency of data, but also ensure that each data source automatically adjusts its influence in the analysis according to its latest performance through a dynamic weight adjustment mechanism. By using the dynamic adjustment of membership functions and inertia weights, the weight of each data source is adjusted according to its real-time performance, which can effectively utilize the advantages of each data source and avoid waste of resources. By using a roulette algorithm to select data sources while maintaining diversity, it is possible to balance exploration and utilization in comprehensive analysis, which increases the randomness and innovation of the decision-making process. In the production of metal welding powder, by integrating and optimizing multidimensional data, the production process can be more accurately monitored and controlled, process parameters can be adjusted in time, and quality problems can be prevented in advance.

[0079] 2. The adaptive extreme difference isolation forest algorithm of the present invention can effectively identify and isolate outliers, which is crucial for early identification of abnormal parameters or conditions that may lead to quality problems. Through timely identification and analysis of these outliers, measures can be taken before the problem expands, thereby reducing the defect rate and cost in production. By comparing with performance indicators, the quality of each production batch can be accurately evaluated, providing instant feedback on product quality, allowing manufacturers to adjust and optimize production processes to ensure that each batch of products meets predetermined quality standards.

[0080] 3. The present invention utilizes the analysis results of the isolation forest algorithm in combination with actual performance data to make production decisions more scientifically. By identifying specific raw materials or production parameters associated with low performance, supply chain management or production strategies can be adjusted to improve the overall performance and reliability of the final product. Through accurate quality prediction and effective exception management, the overall reliability of the product can be significantly improved and the failure rate in the market can be reduced. Data integration and result sharing in the analysis process promote cross-departmental collaboration. For example, quality control, R&D and production departments can jointly participate in the decision-making process to ensure that every link from material selection to final product testing meets quality standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0082] Figure 1 is a flow chart of a metal welding powder detection method based on data fusion according to an embodiment of the present invention;

[0083] Figure 2 It is a principle block diagram of a metal welding powder detection system based on data fusion according to an embodiment of the present invention.

[0084] In the figure:

[0085] 1. Testing target and performance index setting module; 2. Multi-dimensional data integration module; 3. Quality prediction and evaluation module; 4. Real-time monitoring and production adjustment module. DETAILED DESCRIPTION

[0086] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0087] According to an embodiment of the present invention, a metal welding powder detection method and system based on data fusion are provided.

[0088] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a metal welding powder detection method based on data fusion is improved, and the metal welding powder detection method includes the following steps:

[0089] S1. Obtain the metal welding powder data to be tested, determine the test target according to the preset requirements, and set the performance index according to the test target.

[0090] It should be explained that metal welding powder samples should be collected directly from the production line, or historical production data should be obtained from the database; cooperate with the quality management team to determine the inspection targets, including welding performance, chemical stability, particle size distribution, etc., based on market demand, customer feedback and historical problem records, and set specific performance indicators based on the inspection targets, such as hardness testing, particle size analysis, chemical composition analysis, etc. Each indicator should have clear test standards and expected values ​​to ensure that all testing processes comply with industry standards and the company's internal quality control process; use data acquisition terminals, such as sensors and scanning equipment, to automatically record information such as batches, production dates, and chemical compositions of welding powders; use meeting discussions, historical data analysis software or customer management systems to help determine the key performance requirements of welding powders, rely on professional testing and analysis equipment, such as hardness testers, laser particle size analyzers, spectrometers, etc., to accurately measure various performance indicators, and use database management systems to record the results of each test for subsequent data analysis and traceability.

[0091] S2. Use multi-detection technology to collect multi-dimensional material property data of metal welding powder data respectively, and integrate the multi-dimensional material property data according to the co-evolutionary fusion algorithm to obtain a material property data set.

[0092] Preferably, using multiple detection technologies to respectively collect multi-dimensional material property data of metal welding powder data, and using data fusion technology to integrate the multi-dimensional material property data to obtain a material property data set includes the following steps:

[0093] S21. Select multiple detection technologies suitable for collecting metal welding powder characteristic data and set detection parameters.

[0094] S22. Prepare the metal welding powder data to be tested, set the testing conditions, run the multi-testing technology, and collect the multi-dimensional material property data of the metal welding powder.

[0095] S23. Preprocessing the multi-dimensional material property data of the metal welding powder.

[0096] It needs to be explained that it is necessary to select detection technologies that can effectively evaluate the multi-dimensional characteristics of metal welding powder, such as X-ray fluorescence spectroscopy, scanning electron microscopy, differential scanning calorimetry, etc., and select appropriate detection technologies based on the properties of metal welding powder (such as chemical composition, particle size, thermal stability). Collaborate with equipment suppliers to ensure that the technology is suitable for welding powder characterization. For each detection technology, set the detection parameters such as temperature, time, energy, etc. according to the manufacturer's recommendations and past experience to obtain the best detection results. Ensure that the metal welding powder sample is pure and free of impurities. Pretreatment steps such as screening and drying may be involved to standardize the sample conditions. According to the parameters set in S21, adjust the operating conditions of the detection equipment, including environmental control (such as temperature, humidity) and specific settings of the equipment (such as scanning rate, resolution). Run multiple detection technologies to collect multi-dimensional material property data of metal welding powder according to the preset parameters and conditions. Preliminary inspection of the data to remove obvious errors or outliers, such as data caused by instrument reading errors or sample processing errors. Standardize the data generated by different technologies to ensure that the data are comparable under the same scale. For example, convert the percentage data of chemical composition into a unified format. Combine the data obtained from different detection technologies into a comprehensive data set to provide a complete perspective for subsequent analysis and interpretation.

[0097] S24. Use a co-evolutionary fusion algorithm to integrate the pre-processed multi-dimensional material property data into a unified material property data set.

[0098] Preferably, integrating the pre-processed multi-dimensional material property data into a unified material property data set using a co-evolutionary fusion algorithm comprises the following steps:

[0099] S241. Using known benchmark data, quantify the differences between data sources in multidimensional material property data by calculating Manhattan distance.

[0100] Preferably, the formula for quantifying the differences between data sources in the multi-dimensional material property data by calculating the Manhattan distance is:

[0101] ;

[0102] In the formula, Represents the data source i With data source j The distance between the provided datasets;

[0103] Represents the data source i The k Material property values ​​of dimensions;

[0104] Represents the data source j The k Material property values ​​of dimensions;

[0105] d represents the number of material property dimensions included in each data set;

[0106] Represents the data source i Datasets;

[0107] Represents the data source j of the dataset.

[0108] S242. According to the calculation results, a membership function is used to assign an initial weight reflecting the corresponding performance to each data source, and the inertia weight is dynamically adjusted.

[0109] Preferably, real-time monitoring of the output of each data source, using the membership function and the inertia weight adaptive rule, and dynamically adjusting the weight according to the latest performance of the data source includes the following steps:

[0110] S2421. Quantify the contribution of each data source based on performance indicators;

[0111] S2422, calculating the membership score of each data source using a membership function;

[0112] S2423. Allocate an initial weight to each data source according to the membership score;

[0113] S2424. Setting and implementing a dynamic adjustment mechanism to dynamically adjust the weight of each data source based on real-time performance feedback from the data source;

[0114] The formula for calculating the membership score of each data source using the membership function is:

[0115] ;

[0116] In the formula, μ i Represents the data source i The membership score of

[0117] p i Represents the data source i performance index value;

[0118] a represents the slope coefficient;

[0119] b represents the intercept coefficient.

[0120] It should be explained that the contribution of each data source to the overall analysis goal is evaluated, for example, by comparing the accuracy, reliability and data integrity of each data source, using statistical analysis software to compare data, such as calculating standard deviation, mean, error range, etc., so as to quantify the contribution of each data source, defining a membership function, which outputs a score based on the performance indicators of the data source (such as accuracy, timeliness and data consistency) to reflect the relative importance of the data source. This function can be set based on professional knowledge and experience, or automatically adjusted and optimized through machine learning algorithms. Based on the membership score calculated in step S2422, an initial weight is assigned to each data source. The initial weight reflects the relative importance of each data source in the comprehensive analysis. The weight allocation can be automatically completed by software to ensure that the weight allocation is fair and can dynamically reflect the performance changes of each data source. The weight is dynamically adjusted according to the real-time performance feedback from the data source to respond to the changes in the data source performance. A monitoring system is implemented to track the output and performance of each data source in real time, and the weight is adjusted using predetermined rules or algorithms (such as sliding window average, exponential decay factor, etc.). Real-time data processing and analysis platforms, such as Apache Kafka or Apache Spark, can be used to monitor performance and dynamically adjust weights.

[0121] The co-evolutionary fusion algorithm is a particle swarm algorithm based on adaptive inertia weights, which includes a membership function and its adaptive rules. By applying the concept of co-evolutionary fusion to the particle swarm optimization algorithm (PSO) and integrating the membership function and its adaptive adjustment rules, an efficient optimization algorithm is realized, which not only enhances the algorithm's search ability and adaptability, but also promotes diversity and global optimization efficiency. The following is the expanded content:

[0122] Co-evolution refers to the interaction and influence between multiple components or individuals during the operation of the algorithm, which jointly promotes the evolution of the entire system. In particle swarm optimization, multiple populations (subpopulations) can be introduced. These subpopulations can evolve independently for different tasks or different stages of the algorithm, and then exchange or merge information through a certain strategy, which increases the overall flexibility and adaptability of the algorithm. The membership function is used in fuzzy logic to quantify the degree of membership of an element to a certain set. In the particle swarm optimization algorithm, the membership function is used to dynamically adjust the behavior of particles, such as the adjustment of inertia weight. By setting different membership rules, the search behavior can be dynamically adjusted according to the distance between the particle and the optimal solution, so that the algorithm can switch adaptively between global exploration and local fine search. By integrating the idea of ​​co-evolution into the particle swarm algorithm and using the membership function for adaptive adjustment, a more powerful optimization algorithm can be created. For example, in multi-objective optimization problems, different subpopulations can focus on different goals and exchange information to collaboratively find a solution that meets all goals, effectively handling complex optimization problems, especially in engineering design problems that need to consider multiple performance indicators or constraints at the same time.

[0123] S243. Monitor the output of each data source in real time, and use the membership function and inertia weight adaptive rules to dynamically adjust the weight according to the latest performance of the data source.

[0124] S244. Calculate the weighted average of the outputs of each data source using the dynamically adjusted weights, and use the roulette wheel method to select a data source that meets the preset fitness.

[0125] Preferably, using the dynamically adjusted weights to calculate the weighted average of the outputs of each data source, and using the roulette wheel method to select a data source that meets the preset fitness includes the following steps:

[0126] S2441. Collect the output data of all data sources, determine the weight of each data source based on the historical performance of the data source, multiply the output of each data source by its corresponding weight, and calculate the sum of all weighted outputs to obtain a weighted average as a comprehensive data point;

[0127] S2442, setting a fitness threshold, and using a fitness function to calculate a fitness score for each data source;

[0128] S2443, calculating the relative probability of each data source based on the fitness score and the comprehensive data points of all data sources;

[0129] S2444, randomly selecting a data source according to the relative probability of the data source using a roulette algorithm;

[0130] S2445. Select a starting point from the data source list and start accumulating the relative probabilities of the data sources. When the accumulated relative probability exceeds or equals the randomly generated value, stop accumulating to obtain the selected data source.

[0131] S245, evaluate the quality of the fused material property dataset, and use Manhattan distance to evaluate the consistency between the fused dataset and the preset target; if the evaluation result is satisfactory, output the fused dataset; if the evaluation result is unsatisfactory, return to S241 and optimize the weight adjustment process until the preset target is met.

[0132] It should be explained that the output of each data source is integrated through a weighted average, so that the data source that contributes more to the overall analysis has a larger proportion in the comprehensive result. The output data of all data sources are collected, and the weight of each data source is determined according to the historical performance of the data source (such as accuracy, reliability, etc.). The output of each data source is multiplied by its corresponding weight, and then these weighted outputs are added to obtain a weighted average. The fitness of each data source is evaluated, that is, the fitness or relevance of its output. A fitness threshold is set, which is determined according to the analysis requirements and the performance indicators of the data source. A fitness function is developed that can calculate each The fitness score of each data source reflects the degree of match between its output and the analysis target. Based on the fitness score of each data source, its probability of being selected relative to other data sources is calculated, and the sum of the fitness scores of all data sources is calculated. For each data source, the proportion of its fitness score to the sum of the fitness scores of all data sources is calculated to obtain its relative probability. According to the relative probability, one or more data sources are randomly selected. Using the roulette algorithm, a starting point is randomly selected, and the relative probability of each data source is accumulated from the data source list until the accumulated probability exceeds or equals the randomly generated value. At this time, the accumulation is stopped, and the selected point is the selected data source.

[0133] The specific implementation steps for calculating the relative probability of each data source based on the fitness score and the comprehensive data points of all data sources are as follows:

[0134] Assess each data source’s fitness score, typically based on factors such as data quality, accuracy, completeness, and timeliness;

[0135] Normalize the fitness scores of each data source to convert them into probabilities by dividing each score by the sum of all scores;

[0136] Use the standardized fitness score as the weight of each data source to calculate its relative probability in the comprehensive data points. The relative probability can be expressed as the ratio of the standardized fitness score of each data source to the total fitness score;

[0137] When conducting comprehensive analysis or data fusion, the calculated relative probability is used as a weight to adjust or influence the influence of each data source in the final analysis result.

[0138] The specific implementation steps of using the roulette algorithm to randomly select a data source based on the relative probability of the data source are as follows:

[0139] Calculate the probability of each data source being selected based on its reliability, accuracy or other relevant performance indicators. The probability should be based on the historical performance of the data source and expert evaluation;

[0140] Ensuring that the sum of the probabilities from all data sources is equal to 1 can be achieved by adding all the probabilities together and adjusting the individual probabilities to ensure that the sum of the probabilities is 1;

[0141] Set up a random number generator to generate a random number between 0 and 1. This random number will be used to simulate the stop position of the pointer in the roulette wheel;

[0142] Generate a random number between 0 and 1;

[0143] Traverse each data source and accumulate its probability value;

[0144] When the accumulated probability value is greater than or equal to the random number, the current data source is selected;

[0145] If the accumulated probability value is less than the random number, continue to the next data source until a data source that meets the conditions is found;

[0146] Simulate multiple roulette selections to obtain selection results from multiple data sources.

[0147] S3. Use the adaptive extreme difference isolation forest algorithm to analyze the material property data set, predict the quality of metal welding powder, and compare it with the performance indicators to evaluate the quality of metal welding powder.

[0148] The material property data set is analyzed using the adaptive range isolation forest algorithm to predict the quality of metal welding powder and compare it with the performance index. The quality evaluation of metal welding powder includes the following steps:

[0149] S31. Perform statistical analysis on the unified material property data set to obtain data distribution, and use the isolation forest algorithm to identify outliers in the data set through random subsampling and range analysis.

[0150] Preferably, statistical analysis is performed on the unified material property data set to obtain data distribution, and outliers in the data set are identified by random subsampling and range analysis using an isolation forest algorithm, including the following steps:

[0151] S311. Analyze the unified material property data set and identify outliers by calculating statistics and evaluating data distribution patterns;

[0152] S312. Use a visual interface to display the distribution of outliers, relationships and correlations between variables, and identify outliers and multicollinearity issues;

[0153] S313, performing random sub-sampling of outliers by using an isolation forest algorithm, calculating the range of features in the random sub-sampling, and identifying outliers in the material property data set according to a set threshold;

[0154] S314. Based on the recognition results, all outliers are located, and the positions and related characteristic values ​​of the outliers in the material property data set are recorded.

[0155] Preferably, the formula for calculating the range of features in random subsampling is:

[0156] ;

[0157] In the formula, Samples max Indicates the maximum value of a specified dimension in a material property data set;

[0158] Samples min Indicates the minimum value of a specified dimension in a material property data set;

[0159] SubSamples max Indicates the maximum value of a specified dimension in a subset of material property data obtained by random sampling;

[0160] SubSamples min Indicates the minimum value of the specified dimension in the sample subset;

[0161] E Represents the range of the feature in the random subsample.

[0162] It should be explained that possible outliers should be identified by calculating statistics (such as mean, median, standard deviation, etc.) and evaluating data distribution patterns (such as normal distribution, skewed distribution, etc.). Statistical software or programming languages ​​(such as R, Python, etc.) should be used for data analysis, including visualization methods such as box plots and histograms to assist in identifying outliers. Visualization tools should be used to display the distribution of outliers, the relationship between variables, and correlations in order to identify outliers and potential multicollinearity problems. Scatter plots, correlation matrix plots, heat maps, etc. should be used to display the relationship between variables and their contribution to outliers. Professional data visualization software (such as Tableau, PowerBI, etc.) or programming libraries (such as matplotlib, seaborn, etc.) should be used to analyze data. ) to create interactive charts, perform random sub-sampling of outliers through the isolation forest algorithm, and calculate the range of features in the sub-sampling, identify outliers in the material property dataset according to the set threshold, the isolation forest algorithm isolates each sample by building multiple decision trees, and short paths indicate possible outliers, use machine learning libraries (such as IsolationForest in scikit-learn) to implement the isolation forest algorithm, set the appropriate sample size and number of trees, locate all outliers based on the recognition results of S313, and record the location and related feature values ​​of the outliers in the material property dataset, and create a detailed outlier report, including the location of the outliers, the variables affected, and their specific values ​​in the dataset.

[0163] S32. Building an isolation tree model using the screened outlier data, and optimizing the isolation tree model by adjusting parameters of the isolation tree model.

[0164] S33. Calculate the score of each outlier point, and perform correlation analysis between the score and the performance index of the metal welding powder to evaluate the influence of the outlier point.

[0165] Preferably, calculating the score of each outlier point and performing correlation analysis between the score and the performance index of the metal welding powder, and evaluating the influence of the outlier point includes the following steps:

[0166] S331, using the optimized isolation tree model to calculate the abnormality score of each outlier point, reflecting the degree of its deviation from normal data;

[0167] S332, for each identified outlier point, collecting metal welding powder performance data related to the outlier point;

[0168] S333. Use statistical methods to quantify the correlation between outlier scores and performance indicators, and identify the performance variables most affected by outliers;

[0169] S334. Display the relationship between the abnormality score of each outlier and the performance index through a visualization platform, and evaluate the relationship between the outlier and the performance index.

[0170] It should be explained that the isolation tree model is constructed using the screened outlier data. The isolation tree is an algorithm specially designed for outlier detection. It "isolates" observations by randomly selecting features and split values, and adjusts model parameters to optimize performance, including the number of trees, sample size, etc., to enhance the generalization ability of the model on new data. Cross-validation and other techniques are used to evaluate the model performance under different parameter settings, and the optimal parameters are selected. The optimized isolation tree model is used to calculate an abnormality score for each outlier, which reflects the degree of deviation from normal data points. For each identified outlier, relevant metal welding powder performance data is collected, such as hardness, particle size distribution, chemical composition, etc. Statistical methods, such as the Pearson correlation coefficient or the Spearman rank correlation coefficient, are used to quantify the correlation between the outlier score and each performance indicator, which helps to identify the performance variables most affected by outliers. Visualization tools (such as Tableau, PowerBI, etc.) are used to display the relationship between the abnormality score of the outlier and the performance indicators, which helps to intuitively evaluate the specific impact of outliers on performance indicators, as well as potential quality control issues.

[0171] The optimized isolation tree model is used to calculate the abnormality score of each outlier point, reflecting the degree of its deviation from normal data. The specific implementation steps are as follows:

[0172] Make sure the dataset you are using has been preprocessed, including scaling and handling missing values, to make it suitable for model input, and load an already optimized (e.g., parameter tuning such as the number of trees and sample size) isolation tree model;

[0173] Input the data into the isolation tree model, which evaluates each data point and outputs its path length in the structure;

[0174] The anomaly score is calculated using the path length output by the model. Generally, the shorter the path, the more likely the data point is an outlier.

[0175] The anomaly score is usually calculated by the following formula: ,in, Represents data points x The expected value of the path length, c ( n ) represents the adjustment factor, which is related to the sample size n The closer the score is to 1, the more likely it is an outlier, and the closer the score is to 0, the more likely it is a normal data point. The abnormal score of each data point is recorded to facilitate subsequent analysis and decision support.

[0176] S34. Analyze the correlation between outlier scores and performance indicators based on the correlation analysis results between outlier scores and performance indicators, and formulate production quality improvement measures for metal welding powder based on the analysis results.

[0177] It should be explained that statistical methods, such as correlation coefficient analysis (Pearson or Spearman), are used to determine the strength and direction of the correlation between the outlier score and each performance indicator, identify which performance indicators have a significant positive or negative correlation with the outlier score, identify the outlier factors that have the greatest impact on performance based on the correlation analysis results, determine the causes of these factors, such as impure raw materials, improper handling or equipment problems, and formulate specific improvement measures for the identified key problems and factors. The improvement measures may include improving the raw material screening standards, optimizing the control parameters in the production process, or maintaining and upgrading the equipment. Consider introducing new monitoring and detection technologies to detect and reduce anomalies in the production process in real time, implement improvement measures into the production process, and set evaluation indicators to monitor the effectiveness of these measures. Regularly review the effectiveness of improvement measures to ensure continuous improvement, and adjust measures based on new data and feedback. Establish a mechanism for continuous improvement, including regular data analysis and review meetings, to ensure continuous improvement in production quality, promote cross-departmental cooperation, and ensure information flow and collaborative work between quality control, production and R&D departments.

[0178] S4. Build a real-time data monitoring terminal, and based on the data monitoring terminal, feed back the predicted results to the production line in real time to adjust the production process parameters of metal welding powder.

[0179] It needs to be explained that a real-time data monitoring terminal should be designed and implemented to capture and analyze key data on the production line, such as temperature, pressure, particle size distribution, etc. Sensors and data acquisition equipment should be used to collect data in real time and transmit the data to the central monitoring terminal through the network. Machine learning models or statistical analysis tools should be integrated to analyze the collected data in real time and predict potential problems or performance deviations. The model can be trained based on historical data to identify patterns and trends and predict future production conditions. According to the results of data analysis, process parameters on the production line can be automatically or manually adjusted, such as temperature adjustment, pressure adjustment, etc. Ensure that the feedback mechanism has low latency and high reliability to respond in real time to any abnormalities or predicted changes detected by the monitoring. An intuitive user interface should be designed for operators to monitor real-time data and predicted results, and control options should be provided to manually adjust production parameters. The interface should include real-time charts of key performance indicators, alarm systems, and historical data analysis functions. The effectiveness of the system should be tested in a production environment to ensure accurate data collection and analysis, as well as the effectiveness of the feedback mechanism. Production line operators should be trained to ensure that they can effectively use the monitoring system for daily operations and emergency response.

[0180] Specific examples are as follows:

[0181] A manufacturing company specializing in high-performance welding materials was faced with the problem of fluctuating welding powder quality and needed to ensure that its products met the strict aerospace industry standards. The specific implementation steps were as follows:

[0182] 1) Data acquisition and target determination;

[0183] Data source: The company obtains real-time solder powder data from automatic sensors and scanning equipment on its production line, including batch number, production date and chemical composition.

[0184] Inspection objectives: By working with the quality management team, key inspection objectives were determined, including consistency of welding performance, uniformity of grain size, and stability of chemical composition.

[0185] Performance indicators: Specific performance indicators are set, such as the average diameter of the particles must not exceed 30 microns, the aluminum content in the chemical composition must be within 1.5%,

[0186] 2) Collection and integration of multi-dimensional material property data;

[0187] Application of multiple detection technologies: chemical composition analysis using X-ray fluorescence spectroscopy, particle size distribution analysis using scanning electron microscopy, thermal stability evaluation using differential scanning calorimetry,

[0188] Data fusion: The co-evolutionary fusion algorithm is used to integrate these multi-dimensional data, which uses membership functions and adaptive weight adjustment strategies to ensure that the data of each detection technology is reasonably used and integrated.

[0189] 3) Quality prediction and control;

[0190] Outlier analysis: The adaptive range isolation forest algorithm is used to analyze the fused material property data set to identify outliers that do not conform to the regular data pattern.

[0191] Performance evaluation: compare the outlier data with the performance indicators, and use data visualization methods to show the distribution of outliers and their potential impact on performance indicators.

[0192] Quality improvement measures: Based on the outlier analysis results, adjust key parameters in the production process, such as raw material screening and heat treatment temperature, to reduce quality deviations.

[0193] 4) Real-time monitoring and feedback mechanism;

[0194] Monitoring system establishment: deploy real-time data monitoring system on the production line to collect key operating parameters and performance data.

[0195] Prediction and adjustment: Analyze data in real time, predict quality trends, and automatically adjust process parameters based on the prediction results to ensure that product quality meets preset standards.

[0196] Operator training: Train production line operators to ensure they can use the monitoring system effectively and respond promptly to feedback and alarms provided by the system.

[0197] According to another embodiment of the present invention, Figure 2 As shown, a metal welding powder detection system based on data fusion is also provided, and the metal welding powder detection system includes: a detection target and performance index setting module 1, a multi-dimensional data integration module 2, a quality prediction and evaluation module 3 and a real-time monitoring and production adjustment module 4;

[0198] The detection target and performance index setting module 1, the multi-dimensional data integration module 2, the quality prediction and evaluation module 3 and the real-time monitoring and production adjustment module 4 are connected in sequence;

[0199] The detection target and performance index setting module 1 is used to obtain the metal welding powder data to be detected, determine the detection target according to the preset requirements, and set the performance index according to the detection target;

[0200] The multi-dimensional data integration module 2 is used to collect multi-dimensional material property data of metal welding powder data respectively by using multi-detection technology, and integrate the multi-dimensional material property data according to the co-evolution fusion algorithm to obtain a material property data set;

[0201] Quality prediction and evaluation module 3, used to analyze the material property data set using the adaptive extreme difference isolation forest algorithm, predict the quality of the metal welding powder, and compare it with the performance index to evaluate the quality of the metal welding powder;

[0202] The real-time monitoring and production adjustment module 4 is used to construct a real-time data monitoring terminal, and according to the data monitoring terminal, feed back the predicted results to the production line in real time to adjust the production process parameters of the metal welding powder.

[0203] In summary, with the help of the above-mentioned technical scheme of the present invention, the present invention uses a variety of detection technologies to obtain the characteristic data of metal welding powder from multiple dimensions and construct a comprehensive and multidimensional data set. The multidimensional data set provides more comprehensive information support for analysis and decision-making, reduces the deviation and error that may be caused by relying on a single data source, and integrates data from different technologies through a co-evolutionary fusion algorithm. It can not only optimize the consistency of the data, but also ensure that each data source automatically adjusts its influence in the analysis according to its latest performance through a dynamic weight adjustment mechanism. By using the dynamic adjustment of the membership function and the inertia weight, the weight of each data source is adjusted according to its real-time performance, which can effectively utilize the advantages of each data source and avoid waste of resources. By maintaining diversity while selecting data sources through a roulette algorithm, it is possible to balance exploration and utilization in comprehensive analysis, which increases the randomness and innovation of the decision-making process. In the production of metal welding powder, by integrating and optimizing multidimensional data, the production process can be more accurately monitored and controlled, process parameters can be adjusted in time, and quality problems can be prevented in advance; the adaptive extreme difference isolation forest algorithm of the present invention can effectively Identifying and isolating outliers is crucial for early identification of abnormal parameters or conditions that may lead to quality problems. By timely identifying and analyzing these outliers, measures can be taken before the problem expands, thereby reducing the defect rate and cost in production. By comparing with performance indicators, the quality of each production batch can be accurately evaluated, providing instant feedback on product quality, allowing manufacturers to adjust and optimize production processes to ensure that each batch of products meets the predetermined quality standards. The present invention uses the analysis results of the isolation forest algorithm in combination with actual performance data to make production decisions more scientifically. By identifying specific raw materials or production parameters associated with low performance, supply chain management or production strategies can be adjusted to improve the overall performance and reliability of the final product. Through accurate quality prediction and effective exception management, the overall reliability of the product can be significantly improved and the failure rate in the market can be reduced. Data integration and result sharing in the analysis process promote cross-departmental collaboration, such as quality control, R&D and production departments can jointly participate in the decision-making process to ensure that every link from material selection to final product testing meets quality standards.

[0204] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A metal welding powder detection method based on data fusion, characterized in that: The metal welding powder detection method comprises the following steps: S1. Obtain the metal welding powder data to be tested, determine the test target according to the preset requirements, and set the performance index according to the test target; S2, using multi-detection technology to collect multi-dimensional material property data of metal welding powder data, and integrating the multi-dimensional material property data according to the co-evolutionary fusion algorithm to obtain a material property data set; S3, using an adaptive extreme difference isolation forest algorithm to analyze a material property data set, predict the quality of the metal welding powder, and compare it with the performance index to evaluate the quality of the metal welding powder; S4. Build a real-time data monitoring terminal, and according to the data monitoring terminal, feed back the predicted results to the production line in real time to adjust the production process parameters of metal welding powder; The method of analyzing the material property data set using the adaptive range isolation forest algorithm, predicting the quality of the metal welding powder, and comparing it with the performance index to evaluate the quality of the metal welding powder includes the following steps: S31. Statistically analyze the unified material property data set to obtain data distribution, and use the isolation forest algorithm to identify outliers in the data set through random subsampling and range analysis; S32, constructing an isolation tree model using the screened outlier data, and optimizing the isolation tree model by adjusting parameters of the isolation tree model; S33, calculating the score of each outlier point, and performing correlation analysis between the score and the performance index of the metal welding powder to evaluate the influence of the outlier point; S34. Analyze the correlation between the outlier score and the performance index according to the correlation analysis results between the outlier score and the performance index, and formulate production quality improvement measures for metal welding powder according to the analysis results; The method of collecting multi-dimensional material property data of metal welding powder data by using multiple detection technologies and integrating the multi-dimensional material property data by using data fusion technology to obtain a material property data set includes the following steps: S21, selecting multiple detection technologies suitable for collecting metal welding powder characteristic data, and setting detection parameters; S22, preparing the metal welding powder data to be tested, setting the testing conditions, running the multi-testing technology, and collecting the multi-dimensional material property data of the metal welding powder; S23, preprocessing the multi-dimensional material property data of the metal welding powder; S24, integrating the pre-processed multi-dimensional material property data into a unified material property data set using a co-evolutionary fusion algorithm; The method of integrating the pre-processed multi-dimensional material property data into a unified material property data set by using the co-evolutionary fusion algorithm comprises the following steps: S241. quantifying the differences between data sources in the multidimensional material property data by calculating Manhattan distance using known benchmark data; S242, according to the calculation results, using the membership function to assign initial weights reflecting corresponding performance to each data source, and dynamically adjust the inertia weight; S243, monitor the output of each data source in real time, and dynamically adjust the weight according to the latest performance of the data source using the membership function and the inertia weight adaptive rule; S244, using the dynamically adjusted weights, calculating the weighted average of the outputs of each data source, and using the roulette wheel method to select a data source that meets the preset fitness; S245, evaluating the quality of the fused material property data set, and using Manhattan distance to evaluate the consistency between the fused data set and the preset target; if the evaluation result is satisfactory, outputting the fused data set; if the evaluation result is unsatisfactory, returning to S241, optimizing the weight adjustment process until the preset target is met; The formula for quantifying the differences between data sources in multi-dimensional material property data by calculating the Manhattan distance is: ; In the formula, Represents the data source i With data source j The distance between the provided datasets; Represents the data source i The k Material property values ​​of dimensions; Represents the data source j The k Material property values ​​of dimensions; d represents the number of material property dimensions included in each data set; Represents the data source i Datasets; Represents the data source j Datasets; The real-time monitoring of the output of each data source and the use of the membership function and the inertia weight adaptive rule to dynamically adjust the weight according to the latest performance of the data source include the following steps: S2421. Quantify the contribution of each data source based on performance indicators; S2422, calculating the membership score of each data source using a membership function; S2423. Allocate an initial weight to each data source according to the membership score; S2424. Setting and implementing a dynamic adjustment mechanism to dynamically adjust the weight of each data source based on real-time performance feedback from the data source; The formula for calculating the membership score of each data source using the membership function is: ; In the formula, μ i Represents the data source i The membership score of p i Represents the data source i performance index value; a represents the slope coefficient; b represents the intercept coefficient; The method of calculating the weighted average of the outputs of each data source by using the dynamically adjusted weights and selecting the data source that meets the preset fitness by using the roulette wheel method comprises the following steps: S2441. Collect the output data of all data sources, determine the weight of each data source based on the historical performance of the data source, multiply the output of each data source by its corresponding weight, and calculate the sum of all weighted outputs to obtain a weighted average as a comprehensive data point; S2442, setting a fitness threshold, and using a fitness function to calculate a fitness score for each data source; S2443, calculating the relative probability of each data source based on the fitness score and the comprehensive data points of all data sources; S2444, randomly selecting a data source according to the relative probability of the data source using a roulette algorithm; S2445. Select a starting point from the data source list and start accumulating the relative probabilities of the data sources. When the accumulated relative probability exceeds or equals the randomly generated value, stop accumulating to obtain the selected data source.

2. A metal welding powder detection method based on data fusion according to claim 1, characterized in that: The statistical analysis of the unified material property data set to obtain data distribution and identify outliers in the data set by random subsampling and range analysis using the isolation forest algorithm includes the following steps: S311. Analyze the unified material property data set and identify outliers by calculating statistics and evaluating data distribution patterns; S312. Use a visual interface to display the distribution of outliers, relationships and correlations between variables, and identify outliers and multicollinearity issues; S313, performing random sub-sampling of outliers by using an isolation forest algorithm, calculating the range of features in the random sub-sampling, and identifying outliers in the material property data set according to a set threshold; S314. Based on the recognition results, all outliers are located, and the positions and related characteristic values ​​of the outliers in the material property data set are recorded.

3. A metal welding powder detection method based on data fusion according to claim 2, characterized in that: The formula for calculating the range of features in random subsampling is: ; In the formula, Samples max Indicates the maximum value of a specified dimension in a material property data set; Samples min Indicates the minimum value of a specified dimension in a material property data set; SubSamples max Indicates the maximum value of a specified dimension in a subset of material property data obtained by random sampling; SubSamples min Indicates the minimum value of the specified dimension in the sample subset; E Represents the range of the feature in the random subsample.

4. The metal welding powder detection method based on data fusion according to claim 3 is characterized in that: The calculation of the score of each outlier point, and the correlation analysis between the score and the performance index of the metal welding powder, and the evaluation of the influence of the outlier point include the following steps: S331, using the optimized isolation tree model to calculate the abnormality score of each outlier point, reflecting the degree of its deviation from normal data; S332, for each identified outlier point, collecting metal welding powder performance data related to the outlier point; S333. Use statistical methods to quantify the correlation between outlier scores and performance indicators, and identify the performance variables most affected by outliers; S334. Display the relationship between the abnormality score of each outlier and the performance index through a visualization platform, and evaluate the relationship between the outlier and the performance index.

5. A metal welding powder detection system based on data fusion, used to implement the metal welding powder detection method based on data fusion as described in any one of claims 1 to 4, characterized in that: The metal welding powder detection system includes: detection target and performance index setting module, multi-dimensional data integration module, quality prediction and evaluation module and real-time monitoring and production adjustment module; The detection target and performance index setting module, the multi-dimensional data integration module, the quality prediction and evaluation module, and the real-time monitoring and production adjustment module are connected in sequence; The detection target and performance index setting module is used to obtain the metal welding powder data to be detected, determine the detection target according to the preset requirements, and set the performance index according to the detection target; A multi-dimensional data integration module is used to collect multi-dimensional material property data of metal welding powder data by using multiple detection technologies, and integrate the multi-dimensional material property data according to the co-evolution fusion algorithm to obtain a material property data set; A quality prediction and evaluation module, for analyzing a material property data set using an adaptive extreme difference isolation forest algorithm, predicting the quality of the metal welding powder, and comparing it with the performance index to evaluate the quality of the metal welding powder; The real-time monitoring and production adjustment module is used to build a real-time data monitoring terminal, and based on the data monitoring terminal, the predicted results are fed back to the production line in real time to adjust the production process parameters of the metal welding powder.