Real-time data analysis and optimization system for wide voltage non-phase sequence welding process
By deploying high-precision sensors and improved Bagging algorithms in wide-voltage phaseless sequence welding machines, combined with adaptive differential evolution methods, the problems of welding data collection and parameter adjustment are solved, real-time monitoring and automatic optimization of the welding process are realized, and the improvement of the improvement of the welding process is improved. Welding efficiency and consistency.
Patent Information
- Application Number
- CN202411124446.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-16
AI Technical Summary
The existing wide voltage phaseless sequence welding machines lack data collection and analysis during the welding process, lack real-time monitoring and automatic optimization, resulting in unstable welding quality and difficulty in adapting to variable welding conditions, affecting production efficiency and product consistency.
Welding data is collected in real time with high-precision sensors, and data analysis is performed through improved Bagging algorithm, and welding parameters are dynamically adjusted in combination with adaptive differential evolution methods to ensure the real-time and efficient welding process.
It improves the comprehensiveness and prediction accuracy of welding data, reduces welding defects, significantly improves welding efficiency and product consistency, and realizes an intelligent welding production process.
Smart Images

Figure CN118650253B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of data analysis, and in particular to a real-time data analysis and optimization system for a wide voltage non-phase sequence electric welding process. Background Art
[0002] Wide voltage phase-sequence free welding machine is a welding device designed to operate at different voltages and frequencies, and it can adapt to different power input conditions. The large amount of data generated in the existing welding process has not been effectively collected and analyzed, lacking real-time monitoring and automatic optimization. The welding quality is greatly affected by human factors, and consistency is difficult to ensure, which limits the ability to improve the process and predict faults. The parameters of the existing wide voltage phase-sequence free welding machine are difficult to adapt to the changing welding conditions. When welding workpieces of special materials and thicknesses, the preset parameters of the welding machine may not be sufficient to provide the best welding effect, resulting in a decrease in welding quality and affecting the production rhythm. Summary of the invention
[0003] In view of the above situation, the present invention provides a real-time data analysis and optimization system for a wide voltage phase-sequence-free electric welding process. In view of the problem that a large amount of data generated during the welding process cannot be effectively collected and analyzed, and lacks real-time monitoring and automatic optimization, high-precision sensors are deployed at the welding site to continuously capture welding data during the welding process to ensure the comprehensiveness and real-time nature of the data. Subsequently, an improved Bagging (Bootstrap aggregating, also known as bagging algorithm) algorithm is used for data analysis. The improved Bagging algorithm constructs multiple parallel base classifiers to independently learn different subsets in the welding data, reduce the risk of overfitting, improve prediction accuracy, effectively extract signals from noise, identify welding states, and evaluate welding quality. In view of the problem that the parameters of existing wide voltage phase-sequence-free electric welding machines are difficult to adapt to the changing welding conditions, an improved adaptive differential evolution method is used to dynamically adjust the welding parameters according to the welding conditions, and the output current, voltage and temperature of the electric welding machine are dynamically adjusted according to the real-time welding conditions, so that the electric welding machine maintains high performance and high-precision welding quality in various complex environments, thereby significantly improving welding efficiency and product consistency, while reducing welding defects caused by improper parameters, and realizing an intelligent welding production process.
[0004] The wide voltage non-phase sequence electric welding process real-time data analysis and optimization system provided by the present invention comprises a data acquisition module, a signal conditioning module, a data processing and analysis module, an optimization algorithm module, a remote monitoring and maintenance module and a user interface module;
[0005] The data acquisition module uses sensors to collect current, voltage, temperature, welding speed and gas flow data in real time during welding and outputs analog signals;
[0006] The signal conditioning module converts the analog signal output by the sensor into a digital signal, and performs preprocessing to remove noise and interference to obtain preprocessed data;
[0007] The data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the pre-processed data, identify the welding state and quality, and obtain welding parameters;
[0008] The optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters;
[0009] The remote monitoring and maintenance module supports remote access and maintenance, allowing technical support personnel to perform troubleshooting and software updates, and generates warning information for generated faults to provide early warning;
[0010] The user interface module displays real-time data, welding parameters and warning information during the welding process, allowing the user to monitor and control the welding process.
[0011] Furthermore, the data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the data, specifically including the following steps:
[0012] Step S1: Initialization, using the preprocessed data to construct an initial data set and setting the number of TN base classifiers, including support vector machine, K nearest neighbor, decision tree and logistic regression;
[0013] Step S2: feature selection, using mutual information metric to perform feature selection on the initial data set, remove irrelevant or redundant features, generate a data set with reduced features, and divide it into a training set and a test set;
[0014] Step S3: Pre-classification: Use 10-fold cross validation to train the base classifier on the training set to generate classification results, and use the majority voting method to determine the overall classification accuracy. According to the classification results, the training set is divided into two subsets, namely, subset TD containing correctly classified samples and subset MT containing incorrectly classified samples;
[0015] Step S4: Dataset creation and identification, create K data sets BD, identify out-of-bag samples in each data set BD, and generate a validation set;
[0016] Step S5: In the training phase, all base classifiers are trained independently on each data set BD, the performance of all base classifiers is tested on the validation set, and the base classifier with the best performance on each data set BD is selected as the best classifier to obtain K best classifiers;
[0017] Step S6: In the testing phase, all the best classifiers are used to classify the test set, and the classification results of K best classifiers are obtained;
[0018] Step S7: classification prediction, using majority voting method to combine the classification results of all the best classifiers to obtain the final prediction category;
[0019] Furthermore, in step S4, the data set is created and identified, specifically including the following steps:
[0020] Step S41: Data set creation, randomly resampling samples from subset TD, and obtaining all samples from subset MT, repeating the sampling K times, and creating K data sets BD;
[0021] Step S42: Identify out-of-bag samples. Identify the samples in the subset TD that are not included in each data set BD as out-of-bag samples, and use the out-of-bag samples as a validation set.
[0022] Furthermore, the optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters, specifically including the following steps:
[0023] Step M1: parameter updating mechanism, using the parameter updating mechanism of the successful parameter storage mechanism to optimize the welding parameters and dynamically adjust the key parameters in the differential evolution algorithm;
[0024] Step M2: Proportional adjustment mechanism, using L evolutionary algorithms to optimize welding parameters and key parameters, and adjusting the use frequency of each evolutionary algorithm to generate mutation vectors;
[0025] Step M3: Monitor the stagnation state, define the stagnation threshold, use the number of iterations and fitness advantage indicators as the stagnation criteria, and when the stagnation threshold is reached, generate a random number β, and use the dimension search and population restart strategy to optimize the parameters according to the random number β;
[0026] Step M4: Dimension search. When the random number β is greater than 0.7, the dimensional search strategy is used to generate the mutation vector. The formula used is as follows:
[0027] ;
[0028] ;
[0029] In the formula, is a natural number, Represents the dimension, represents the location of the current optimal solution, is a random function, and They represent the upper and lower bounds of the search space respectively, represents the mutation vector, is the modular operation function, is the current iteration number, Indicates the number of dimensions;
[0030] Step M5: Population restart. When the random number β is not greater than 0.7, the population restart strategy is used to generate a mutation vector.
[0031] Step M6: Iterative optimization, set the maximum number of iterations, repeat steps M1 to M5 until the maximum number of iterations is reached, and output the final optimal solution;
[0032] Furthermore, in step M1, the parameter updating mechanism specifically includes the following steps:
[0033] Step M11: Initialize the population and population parameters, randomly generate the initial population, individuals, number of individuals and positions, set the search space, and use the greedy algorithm to divide individuals into parent individuals and offspring individuals;
[0034] Step M12: Initialize the repository, create a successful cross rate library and a successful scale factor library, the successful cross rate library and the successful scale factor library contain the cross rate and the scale factor;
[0035] Step M13: Evaluate the offspring, use the differential evolution algorithm to evaluate the offspring individuals. If the fitness value of the offspring individual is less than that of the parent individual, it is considered that the offspring individual has a beneficial evolution and is called a successful individual.
[0036] Step M14: storing the success parameters, and accumulating the crossover rate and scale factor of the successful individuals into the success crossover rate library and the success scale factor library according to the fitness value optimization ratio;
[0037] Step M15: bilateral distribution, using the bilateral distribution method to calculate the distribution of the crossover rate and the scale factor, and optimize the search ability of successful individuals in the search space;
[0038] Step M16: Update the distribution of the crossover rate and the scale factor, use the normal distribution model to update the crossover rate, and use the maximum likelihood estimation model to update the scale factor;
[0039] Step M17: Mixing distribution and range constraints, setting the threshold of the scale factor, using Beta distribution to mix the distribution of the scale factor, if the value of the scale factor exceeds the set threshold, using Cauchy distribution to redistribute the scale factor, and constraining the crossover rate and the value of the scale factor to [0,1];
[0040] Step M18: Mutation vector generation, using the current best strategy to generate a mutation vector, and using the updated crossover rate and scale factor to control the strength and direction of the mutation vector;
[0041] Furthermore, in step M5, the population is restarted, specifically including the following steps:
[0042] Step M51: Reset the population and parameters, reset the population and population parameters, and use the Gaussian distribution strategy to generate mutation vectors;
[0043] Step M52: Construct a covariance matrix. Use the current optimal solution and the population mean to calculate the covariance matrix. The formula used is as follows:
[0044] ;
[0045] In the formula, Indicates a point in time, Indicates at a point in time The optimal solution vector of represents the population mean vector, is the transpose symbol, Indicates percentage, is the population size, For rounding operation, represents the covariance matrix;
[0046] Step M53: Characteristic analysis, calculate the eigenvalues and eigenvectors of the covariance matrix, the formula used is as follows:
[0047] ;
[0048] In the formula, To solve the operation of eigenvalue and eigenvector, is the eigenvector matrix, is the eigenvalue matrix;
[0049] Step M54: Update the mutation vector. Use Gaussian distribution and eigenvector matrix to update the mutation vector. The formula used is as follows:
[0050] ;
[0051] In the formula, Indicates Individuals at a point in time The position vector of is a random perturbation vector generated using a Gaussian distribution, represents the transformation of the position vector into the feature space, represents the standard deviation of the Gaussian distribution, is the scale factor, and are two randomly selected individual vectors, Indicates Individuals at a point in time The mutation vector after feature space transformation;
[0052] Step M55: Update the velocity vector. The formula used is as follows:
[0053] ;
[0054] In the formula, Indicates Individuals at a point in time The velocity vector.
[0055] The beneficial effects achieved by the present invention using the above scheme are as follows:
[0056] (1) In order to solve the problem that a large amount of data generated during the welding process cannot be effectively collected and analyzed, and there is a lack of real-time monitoring and automatic optimization, high-precision sensors are deployed at the welding site to continuously capture welding data during the welding process to ensure the comprehensiveness and real-time nature of the data. Then, an improved Bagging algorithm is used for data analysis. The improved Bagging algorithm constructs multiple parallel base classifiers to independently learn different subsets of welding data, reduce the risk of overfitting, improve prediction accuracy, effectively extract signals from noise, identify welding status, and evaluate welding quality.
[0057] (2) In view of the problem that the parameters of existing wide-voltage phase-sequenceless welding machines are difficult to adapt to the changing welding conditions, an improved adaptive differential evolution method is used to dynamically adjust the welding parameters. The output current, voltage and temperature of the welding machine are dynamically adjusted according to the real-time welding conditions. The improved adaptive differential evolution method integrates the parameter update mechanism, proportional adjustment mechanism, dimensional search and population restart strategy. Through continuous learning and self-improvement, the welding machine can maintain high performance and high-precision welding quality in various complex environments. It automatically corrects according to real-time feedback to adapt to the optimal welding strategy under different materials, thicknesses and environmental conditions, thereby significantly improving welding efficiency and product consistency. At the same time, it reduces welding defects caused by improper parameters and realizes an intelligent welding production process. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] Figure 1 A schematic diagram of a wide voltage non-phase sequence electric welding process real-time data analysis and optimization system provided by the present invention;
[0059] Figure 2 This is a flowchart of the improved Bagging algorithm;
[0060] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0062] Example 1, see Figure 1 The wide voltage non-phase sequence electric welding process real-time data analysis and optimization system provided by the present invention includes a data acquisition module, a signal conditioning module, a data processing and analysis module, an optimization algorithm module, a remote monitoring and maintenance module and a user interface module;
[0063] The data acquisition module uses sensors to collect current, voltage, temperature, welding speed and gas flow data in real time during welding and outputs analog signals;
[0064] The signal conditioning module converts the analog signal output by the sensor into a digital signal, and performs preprocessing to remove noise and interference to obtain preprocessed data;
[0065] The data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the pre-processed data, identify the welding state and quality, and obtain welding parameters;
[0066] The optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters;
[0067] The remote monitoring and maintenance module supports remote access and maintenance, allowing technical support personnel to perform troubleshooting and software updates, and generates warning information for generated faults to provide early warning;
[0068] The user interface module displays real-time data, welding parameters and warning information during the welding process, allowing the user to monitor and control the welding process.
[0069] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the data. Specifically, the following steps are included:
[0070] Step S1: Initialization, using the preprocessed data to construct an initial data set and setting the number of TN base classifiers, including support vector machine, K nearest neighbor, decision tree and logistic regression;
[0071] Step S2: feature selection, using mutual information metric to perform feature selection on the initial data set, remove irrelevant or redundant features, generate a data set with reduced features, and divide it into a training set and a test set;
[0072] Step S3: Pre-classification: Use 10-fold cross validation to train the base classifier on the training set to generate classification results, and use the majority voting method to determine the overall classification accuracy. According to the classification results, the training set is divided into two subsets, namely, subset TD containing correctly classified samples and subset MT containing incorrectly classified samples;
[0073] Step S4: Dataset creation and identification, create K data sets BD, identify out-of-bag samples in each data set BD, and generate a validation set;
[0074] Step S5: In the training phase, all base classifiers are trained independently on each data set BD, the performance of all base classifiers is tested on the validation set, and the base classifier with the best performance on each data set BD is selected as the best classifier to obtain K best classifiers;
[0075] Step S6: In the testing phase, all the best classifiers are used to classify the test set, and the classification results of K best classifiers are obtained;
[0076] Step S7: classification prediction, using majority voting method to combine the classification results of all the best classifiers to obtain the final prediction category;
[0077] In step S4, the data set is created and identified, specifically including the following steps:
[0078] Step S41: Dataset creation: randomly resample samples from subset TD and obtain all samples from subset MT, repeat sampling K times, create K data sets BD, K data sets BD = {data set BD1, data set BD2, ..., data set BD k ;
[0079] Step S42: Identify out-of-bag samples. Identify the samples in the subset TD that are not included in each data set BD as out-of-bag samples, and use the out-of-bag samples as a validation set.
[0080] In order to solve the problems that a large amount of data generated during the welding process cannot be effectively collected and analyzed, and there is a lack of real-time monitoring and automatic optimization, high-precision sensors are deployed at the welding site to continuously capture the welding data during the welding process to ensure the comprehensiveness and real-time nature of the data. Subsequently, an improved Bagging algorithm is used for data analysis. The improved Bagging algorithm constructs multiple parallel base classifiers to independently learn different subsets of the welding data, reduce the risk of overfitting, improve prediction accuracy, effectively extract signals from noise, identify welding status, and evaluate welding quality.
[0081] Embodiment 3: This embodiment is based on the above embodiment. The optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters. Specifically, the following steps are included:
[0082] Step M1: parameter updating mechanism, using the parameter updating mechanism of the successful parameter storage mechanism to optimize the welding parameters and dynamically adjust the key parameters in the differential evolution algorithm;
[0083] Step M2: Proportional adjustment mechanism, using L evolutionary algorithms to optimize welding parameters and key parameters, and adjusting the use frequency of each evolutionary algorithm to generate mutation vectors;
[0084] Step M3: Monitor the stagnation state, define the stagnation threshold, use the number of iterations and fitness advantage indicators as the stagnation criteria, and when the stagnation threshold is reached, generate a random number β, and use the dimension search and population restart strategy to optimize the parameters according to the random number β;
[0085] Step M4: Dimension search. When the random number β is greater than 0.7, the dimensional search strategy is used to generate the mutation vector. The formula used is as follows:
[0086] ;
[0087] ;
[0088] In the formula, is a natural number, Represents the dimension, represents the location of the current optimal solution, is a random function, and They represent the upper and lower bounds of the search space respectively, represents the mutation vector, is the modular operation function, is the current iteration number, Indicates the number of dimensions;
[0089] Step M5: Population restart. When the random number β is not greater than 0.7, the population restart strategy is used to generate a mutation vector.
[0090] Step M6: Iterative optimization, set the maximum number of iterations, repeat steps M1 to M5 until the maximum number of iterations is reached, and output the final optimal solution.
[0091] Embodiment 4: This embodiment is based on the above embodiment. In step M1, the parameter updating mechanism specifically includes the following steps:
[0092] Step M11: Initialize the population and population parameters, randomly generate the initial population, individuals, number of individuals and positions, set the search space, and use the greedy algorithm to divide individuals into parent individuals and offspring individuals;
[0093] Step M12: Initialize the repository, create a successful cross rate library and a successful scale factor library, the successful cross rate library and the successful scale factor library contain the cross rate and the scale factor;
[0094] Step M13: Evaluate the offspring, use the differential evolution algorithm to evaluate the offspring individuals. If the fitness value of the offspring individual is less than that of the parent individual, it is considered that the offspring individual has a beneficial evolution and is called a successful individual.
[0095] Step M14: storing the success parameters, and accumulating the crossover rate and scale factor of the successful individuals into the success crossover rate library and the success scale factor library according to the fitness value optimization ratio;
[0096] Step M15: bilateral distribution, using the bilateral distribution method to calculate the distribution of the crossover rate and the scale factor, and optimize the search ability of successful individuals in the search space;
[0097] Step M16: Update the distribution of the crossover rate and the scale factor, use the normal distribution model to update the crossover rate, and use the maximum likelihood estimation model to update the scale factor;
[0098] Step M17: Mixing distribution and range constraints, setting the threshold of the scale factor, using Beta distribution to mix the distribution of the scale factor, if the value of the scale factor exceeds the set threshold, using Cauchy distribution to redistribute the scale factor, and constraining the crossover rate and the value of the scale factor to [0,1];
[0099] Step M18: Mutation vector generation, use the current best strategy to generate a mutation vector, and use the updated crossover rate and scale factor to control the strength and direction of the mutation vector.
[0100] Embodiment 5: This embodiment is based on the above embodiment. In step M2, the ratio adjustment mechanism specifically includes the following steps:
[0101] Step M21: assigning strategy ratios, taking L evolutionary algorithms as L strategies, and assigning an initial ratio to each strategy, the sum of the ratios being equal to 1, the L evolutionary algorithms including genetic algorithm, differential evolution and distribution estimation algorithm;
[0102] Step M22: Evaluation criteria: for each strategy, if the fitness value of the mutation vector is better than that of the parent individual, it indicates that the strategy of the corresponding welding parameters and key parameters is effective; otherwise, the strategy of the corresponding welding parameters and key parameters is invalid;
[0103] Step M23: Record the number of wins. If the fitness of the mutation vector of the three strategies is better than that of the parent generation, the number of wins of the strategy increases.
[0104] Step M24: Calculate the winning rate. Calculate the winning rate of each strategy using the following formula:
[0105] ;
[0106] In the formula, is the number of iteration cycles, is a natural number, is the number of subgroups using different strategies, represents the mutation vector, Represents the parent individual, To calculate the number of elements that satisfy the conditions in brackets, Indicates the winning rate;
[0107] Step M25: Adjust the ratio, take the strategy with the highest winning rate as the optimal strategy, and the strategy with the lowest winning rate as the worst strategy, increase the ratio of the optimal strategy, and reduce the ratio of the worst strategy. If the ratio of the worst strategy is already the smallest, then reduce the ratio of the suboptimal strategy.
[0108] Embodiment 6: This embodiment is based on the above embodiment. In step M5, the population is restarted, which specifically includes the following steps:
[0109] Step M51: Reset the population and parameters, reset the population and population parameters, and use the Gaussian distribution strategy to generate mutation vectors;
[0110] Step M52: Construct a covariance matrix. Use the current optimal solution and the population mean to calculate the covariance matrix. The formula used is as follows:
[0111] ;
[0112] In the formula, Indicates time, Indicates at a point in time The optimal solution vector of represents the population mean vector, is the transpose symbol, Indicates percentage, is the population size, For rounding operation, represents the covariance matrix;
[0113] Step M53: Characteristic analysis, calculate the eigenvalues and eigenvectors of the covariance matrix, the formula used is as follows:
[0114] ;
[0115] In the formula, To solve the operation of eigenvalue and eigenvector, is the eigenvector matrix, is the eigenvalue matrix;
[0116] Step M54: Update the mutation vector. Use Gaussian distribution and eigenvector matrix to update the mutation vector. The formula used is as follows:
[0117] ;
[0118] In the formula, Indicates Individuals at a point in time The position vector of is a random perturbation vector generated using a Gaussian distribution, represents the transformation of the position vector into the feature space, represents the standard deviation of the Gaussian distribution, is the scale factor, and are two randomly selected individual vectors, Indicates Individuals at a point in time The mutation vector after feature space transformation;
[0119] Step M55: Update the velocity vector. The formula used is as follows:
[0120] ;
[0121] In the formula, Indicates Individuals at a point in time The velocity vector.
[0122] Embodiment 7, this embodiment is based on the above embodiment. In the specific implementation, the Windows 10 operating system is used, the CPU is Intel (R) Xeon (R) Gold 6248R, the operating frequency is 3.00 GHz, and the memory is 256GB for implementation;
[0123] In the specific implementation process of the welding machine parameters, the population size was set to 100 and the number of iterations was 1000. Four test functions, namely, unimodal function, multimodal function, mixed function and combined function, were used to test the performance of the improved adaptive differential evolution method. Each test function was run 50 times, and the performance of the algorithm was evaluated based on the mean and standard deviation of the error, and the optimization effect of the wide voltage phase-sequence free welding process was obtained.
[0124] Aiming at the problem that the parameters of existing wide-voltage phase-sequenceless welding machines are difficult to adapt to the changing welding conditions, an improved adaptive differential evolution method is adopted to dynamically adjust the welding parameters. The output current, voltage and temperature of the welding machine are dynamically adjusted according to the real-time welding conditions. The improved adaptive differential evolution method integrates parameter update mechanism, proportional adjustment mechanism, dimension search and population restart strategy. Through continuous learning and self-improvement, the welding machine can maintain high performance and high-precision welding quality in various complex environments. It automatically corrects according to real-time feedback to adapt to the optimal welding strategy under different materials, thicknesses and environmental conditions, thereby significantly improving welding efficiency and product consistency, while reducing welding defects caused by improper parameters and realizing intelligent welding production process.
[0125] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0126] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. Real-time data analysis and optimization system for wide voltage non-phase sequence welding process, characterized by: It includes data acquisition module, signal conditioning module, data processing and analysis module, optimization algorithm module, remote monitoring and maintenance module and user interface module; The data acquisition module uses sensors to collect welding process data in real time and outputs analog signals; The signal conditioning module converts the analog signal output by the sensor into a digital signal and performs preprocessing to obtain preprocessing data; The data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the pre-processed data to obtain welding parameters; The optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters; The remote monitoring and maintenance module supports remote access and maintenance, generates warning information for generated faults, and issues early warnings; The user interface module displays real-time data, welding parameters and warning information during the welding process; The data processing and analysis module uses an improved Bagging algorithm to perform real-time analysis on the data, specifically including the following steps: Step S1: Initialization, using preprocessed data to construct an initial data set and setting the number of TN base classifiers; Step S2: feature selection, using mutual information metric to perform feature selection on the initial data set, generating a data set with reduced features, and dividing it into a training set and a test set; Step S3: Pre-classification: Use 10-fold cross validation to train the base classifier on the training set to generate classification results, and use the majority voting method to determine the overall classification accuracy. According to the classification results, the training set is divided into two subsets, namely, subset TD containing correctly classified samples and subset MT containing incorrectly classified samples; Step S4: Dataset creation and identification, create K data sets BD, identify out-of-bag samples in each data set BD, and generate a validation set; Step S5: In the training phase, all base classifiers are trained independently on each data set BD, the performance of all base classifiers is tested on the validation set, and the base classifier with the best performance on each data set BD is selected as the best classifier to obtain K best classifiers; Step S6: In the testing phase, all the best classifiers are used to classify the test set, and the classification results of K best classifiers are obtained; Step S7: classification prediction, using majority voting method to combine the classification results of all the best classifiers to obtain the final prediction category; In step S4, the data set is created and identified, specifically including the following steps: Step S41: Data set creation, randomly resampling samples from subset TD, and obtaining all samples from subset MT, repeating the sampling K times, and creating K data sets BD; Step S42: Identify out-of-bag samples, identify the samples in the subset TD that are not included in each data set BD as out-of-bag samples, and use the out-of-bag samples as a validation set; The optimization algorithm module uses an improved adaptive differential evolution method to dynamically adjust welding parameters, specifically including the following steps: Step M1: parameter updating mechanism, using the parameter updating mechanism of the successful parameter storage mechanism to optimize the welding parameters and dynamically adjust the key parameters in the differential evolution algorithm; Step M2: Proportional adjustment mechanism, using L evolutionary algorithms to optimize welding parameters and key parameters, and adjusting the use frequency of each evolutionary algorithm to generate mutation vectors; Step M3: Monitor the stagnation state, define the stagnation threshold, use the number of iterations and fitness advantage indicators as the stagnation criteria, and when the stagnation threshold is reached, generate a random number β, and use the dimension search and population restart strategy to optimize the parameters according to the random number β; Step M4: Dimension search, when the random number β is greater than 0.7, the dimensional search strategy is used to generate the mutation vector; Step M5: Population restart. When the random number β is not greater than 0.7, the population restart strategy is used to generate a mutation vector. Step M6: Iterative optimization, set the maximum number of iterations, repeat steps M1 to M5 until the maximum number of iterations is reached, and output the final optimal solution; In step M1, the parameter updating mechanism specifically includes the following steps: Step M11: Initialize the population and population parameters, randomly generate the initial population, individuals, number of individuals and positions, set the search space, and use the greedy algorithm to divide individuals into parent individuals and offspring individuals; Step M12: Initialize the repository, create a successful cross rate library and a successful scale factor library, the successful cross rate library and the successful scale factor library contain cross rates and scale factors; Step M13: Evaluate the offspring, use the differential evolution algorithm to evaluate the offspring individuals. If the fitness value of the offspring individual is less than that of the parent individual, it is considered that the offspring individual has a beneficial evolution and is called a successful individual. Step M14: storing the success parameters, and accumulating the crossover rate and scale factor of the successful individuals into the success crossover rate library and the success scale factor library according to the fitness value optimization ratio; Step M15: bilateral distribution, using the bilateral distribution method to calculate the distribution of the crossover rate and the scale factor, and optimize the search ability of successful individuals in the search space; Step M16: Update the distribution of the crossover rate and the scale factor, use the normal distribution model to update the crossover rate, and use the maximum likelihood estimation model to update the scale factor; Step M17: Mixing distribution and range constraints, setting the threshold of the scale factor, using Beta distribution to mix the distribution of the scale factor, if the value of the scale factor exceeds the set threshold, using Cauchy distribution to redistribute the scale factor, and constraining the crossover rate and the value of the scale factor to [0,1]; Step M18: Mutation vector generation, using the current best strategy to generate a mutation vector, and using the updated crossover rate and scale factor to control the strength and direction of the mutation vector; In step M5, the population is restarted, which specifically includes the following steps: Step M51: Reset the population and parameters, reset the population and population parameters, and use the Gaussian distribution strategy to generate mutation vectors; Step M52: construct a covariance matrix, and calculate the covariance matrix using the current optimal solution and the population mean; Step M53: characteristic analysis, calculating the eigenvalues and eigenvectors of the covariance matrix; Step M54: updating the mutation vector, using Gaussian distribution and eigenvector matrix to update the mutation vector; Step M55: Update the velocity vector.
