CN115563549B - Welding defect cause diagnosis method and system and electronic equipment - Google Patents

Abstract

The invention relates to a welding defect cause diagnosis method, a system and an electronic device, wherein the method comprises the steps of establishing a fuzzy diagnosis rule and a cause diagnosis neural network database, combining the fuzzy diagnosis rule and the cause diagnosis neural network database to perform defect diagnosis, performing front item expansion based on a traditional fuzzy rule, adding a threshold, performing fuzzy reasoning by using a sample rule selected from the fuzzy diagnosis rule, obtaining a target diagnosis result according to the sample rule if the actual credibility is greater than a preset threshold, and performing complementary processing by using the cause diagnosis neural network if the actual credibility is not present, so as to obtain an ideal target diagnosis result. Compared with the traditional fuzzy neural network, the fuzzy neural network and the fuzzy rule processing method based on the fuzzy neural network combine two theories, namely, the fuzzy neural network has the advantage that fuzzy reasoning can process uncertainty knowledge, also has the advantages of self-learning, high efficiency and high precision of the neural network, greatly improves the reliability of the reasoning process, and is more suitable for welding the scene of multiple influencing factors.

Description

A method, system and electronic equipment for diagnosing causes of welding defects

Technical field

The present invention relates to the technical field of welding defect diagnosis, and in particular to a welding defect cause diagnosis method, system and electronic equipment.

Background technique

Analysis of the causes of welding defects has always been a difficulty in the field of welding. Due to the various types of defects and the complex causes of defect formation, defect cause diagnosis relies too much on manual experience, resulting in extremely low efficiency and reliability. To solve this problem, welding expert systems with intelligent diagnostic functions are gradually being applied.

For research on the cause detection of welding defects, the Chinese patent "A Welding Defect Analysis System and Method Based on Big Data" (CN109447403A) proposes a welding defect analysis system based on big data. However, due to the quality of the selected data set, the quality of the selected data set is directly affected. It affects the diagnosis of the causes of final welding defects, and the knowledge base does not have the corresponding self-learning ability and lacks versatility. The Chinese patent "A Welding Defect Diagnosis Method Based on Sparse Matrix" (CN105223275A) proposes a sparse matrix welding defect diagnosis method. However, the database sources obtained by this method are self-made standard samples with single prefabricated defects. For multiple The detection of defective non-standard weldments is not applicable.

It can be known from the above research that the existing technology is mostly based on production rules for deductive reasoning, that is, inference based on existing samples and existing data sets, which cannot well reflect the complex relationship between defect characteristics, process parameters and causes of welding defects. Nonlinear relationship, low diagnostic accuracy, not suitable for nonlinear and multi-classification problems such as welding defect diagnosis and identification, and cannot handle uncertainty knowledge well.

Contents of the invention

In view of this, it is necessary to provide a welding defect cause diagnosis method, system and electronic equipment to solve the problem that the existing welding defect cause diagnosis technology relies too much on existing samples and cannot handle the non-linearity of welding defect diagnosis well. , multi-classification, problems with uncertain knowledge.

In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

In a first aspect, the present invention provides a method for diagnosing the causes of welding defects, including:

Determine the evaluation factors and the cause of the defect, and establish multiple fuzzy diagnostic rules for characterizing the fuzzy relationship between the evaluation factors and the cause of the defect based on the fuzzy rule extending the preceding item;

Establish a cause diagnosis neural network model, the input parameters of the cause diagnosis neural network model include welding defect types and the evaluation factors, and the output parameters of the cause diagnosis neural network model include diagnosis results;

Obtain the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed;

Based on the evaluation factor data, select matching fuzzy diagnostic rules from multiple fuzzy diagnostic rules as sample rules, and perform fuzzy inference on the sample rules to obtain the actual credibility of each sample rule;

Determine whether there is a sample rule whose actual credibility is greater than the preset threshold. If so, obtain a target diagnosis result based on the defect cause of the sample rule whose actual credibility is greater than the preset threshold. If If not, the target diagnosis result is obtained according to the cause diagnosis neural network model.

Further, in determining the evaluation factors and defect causes, a plurality of fuzzy diagnostic rules for characterizing the fuzzy relationship between the evaluation factors and the defect causes are established based on the fuzzy rule extending the preceding item, including:

Establish a welding feature information database, and obtain the evaluation factors and the defect causes based on the welding feature information database;

According to the evaluation factors, a plurality of the fuzzy diagnosis rules are established based on the fuzzy rule extending the preceding item;

Each of the fuzzy diagnosis rules includes an antecedent, a consequent, a theoretical credibility and the preset threshold, the antecedent includes a plurality of the evaluation factors, and the consequent includes the cause of the defect.

Further, based on the evaluation factor data, a matching fuzzy diagnostic rule is selected from a plurality of the fuzzy diagnostic rules as a sample rule, and fuzzy inference is performed on the sample rule to obtain the result of each sample rule. Actual credibility, including:

Based on the evaluation factor data, select a matching fuzzy diagnostic rule from a plurality of the diagnostic rules as the sample rule;

According to the actual welding defect type, based on the analytic hierarchy process, the weight coefficient of each evaluation factor for the actual welding defect type is obtained;

According to the evaluation factor data, the membership degree of each evaluation factor in the sample rule is obtained;

According to the weight coefficient and the membership degree, the actual credibility of each sample rule is calculated.

Further, according to the actual welding defect type, based on the analytic hierarchy process, the weight coefficient of each evaluation factor for the actual welding defect type is obtained, including:

According to the actual welding defect type, establish a pairwise comparison matrix of the evaluation factors;

Perform a consistency test on the pairwise comparison matrix, and adjust the pairwise comparison matrix according to the test results;

According to the adjusted pairwise comparison matrix, obtain the eigenvector of the adjusted pairwise comparison matrix;

The feature vector is normalized to obtain the weight coefficient of each evaluation factor for the actual welding defect type.

Further, it is determined whether there is a sample rule whose actual credibility is greater than a preset threshold. If so, based on the cause of defects of the sample rule whose actual credibility is greater than the preset threshold, we obtain Target diagnosis result, if not, obtain the target diagnosis result based on the cause diagnosis neural network model, including:

Sort the sample rules according to the numerical value of the actual credibility to obtain a sample rule sequence;

Determine whether there is a sample rule whose actual credibility is greater than its corresponding preset threshold. If so, select a set number of samples with the highest actual credibility based on the sample rule sequence. According to the regular defect causes, the target diagnosis result is obtained. If not, the evaluation factor data and the actual welding defect type are input into the cause diagnosis neural network model, and the target diagnosis result is obtained.

Further, the establishment of a cause diagnosis neural network model includes:

Establish an initial BP neural network model, the input parameters of the initial BP neural network model include a plurality of the evaluation factors, and the output parameters of the initial BP neural network model include the diagnostic results;

Obtain the initial parameters of the initial BP neural network model, and optimize them through the PSO-BP algorithm to obtain the optimized parameters;

The initial BP neural network model is optimized according to the optimization parameters to obtain the cause diagnosis neural network model.

Further, the initial parameters of the initial BP neural network model are obtained and optimized through the PSO-BP algorithm to obtain optimized parameters, including:

Obtain initial parameters of the initial BP neural network model, where the initial parameters include weights, thresholds and loss functions of neurons;

According to the weights and thresholds of the neurons, establish a solution space and a particle population, and initialize the position of the particle population in the solution space;

According to the loss function, a fitness function is established;

According to the fitness function, optimize the position of the particle population in the solution space;

Obtain the optimized position of the particle population in the solution space and obtain the optimization parameters.

In a second aspect, the present invention also provides a welding defect cause diagnosis system, including:

A rule establishment module for determining evaluation factors and defect causes, and establishing multiple fuzzy diagnostic rules for characterizing the fuzzy relationship between the evaluation factors and the defect causes based on the fuzzy rules extending the preceding item;

A network building module is used to establish a cause diagnosis neural network model. The input parameters of the cause diagnosis neural network model include welding defect types and the evaluation factors. The output parameters of the cause diagnosis neural network model include diagnosis results;

A data acquisition module, used to obtain the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed;

A fuzzy inference module, configured to select a matching fuzzy diagnostic rule as a sample rule from a plurality of the fuzzy diagnostic rules based on the evaluation factor data, and perform fuzzy inference on the sample rule to obtain each of the sample rules. actual credibility;

A result diagnosis module is used to determine whether there is a sample rule whose actual credibility is greater than a preset threshold. If so, based on the cause of the defect of the sample rule whose actual credibility is greater than the preset threshold, The target diagnosis result is obtained. If not, the target diagnosis result is obtained according to the neural network model.

In a third aspect, the present invention also provides an electronic device including a memory and a processor, wherein,

Memory, used to store programs;

The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the welding defect cause diagnosis method in any of the above implementations.

The invention provides a welding defect cause diagnosis method, system and electronic equipment. The method establishes fuzzy diagnosis rules and cause diagnosis neural network databases, and combines the two to jointly diagnose welding defects. Specifically, it is first based on The traditional fuzzy rule expands the previous item and adds a threshold. The sample rules selected from the fuzzy diagnostic rules are used for fuzzy inference. If there is a sample rule whose actual credibility is greater than the preset threshold, it can be obtained based on the sample rule. If the target diagnosis result does not exist, it can be considered that the existing fuzzy diagnosis rules cannot handle the existing facts. At this time, using the cause diagnosis neural network for supplementary processing, the ideal target diagnosis result can be obtained. Compared with the existing technology, the present invention combines fuzzy rules and neural network theories, so that the diagnosis process not only has the advantages of fuzzy reasoning that can handle uncertain knowledge, but also has the advantages of neural network self-learning, high efficiency and high accuracy. It greatly improves the reliability of the reasoning process. Compared with the traditional fuzzy neural network model, it is more suitable for scenarios with multiple influencing factors such as welding.

Description of the drawings

Figure 1 is a method flow chart of an embodiment of a welding defect cause diagnosis method provided by the present invention;

Figure 2 is a schematic diagram of the welding feature information database established in one embodiment of the welding defect cause diagnosis method provided by the present invention;

Figure 3 is a method flow chart of an embodiment of step S104 in Figure 1;

Figure 4 is a method flow chart of an embodiment of step S302 in Figure 3;

Figure 5 is a schematic structural diagram of an embodiment of a welding defect cause diagnosis system provided by the present invention;

Figure 6 is a schematic structural diagram of the electronic device provided by the present invention.

Detailed ways

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

In the description of this application, "plurality" means two or more than two, unless otherwise explicitly and specifically limited.

Reference herein to "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.

The invention provides a welding defect cause diagnosis method, system and electronic equipment. On the one hand, it uses fuzzy theory to improve the ability to process uncertain knowledge; on the other hand, it uses the connection mechanism of neural networks to deal with facts that fuzzy reasoning cannot handle. Supplementary reasoning has the advantages of both theories, which is helpful for diagnosing the causes of welding defects when rules are missing or the matching degree is not high. It has important engineering practical significance for promoting the automation of welding processes.

The present invention provides a welding defect cause diagnosis method, system and electronic equipment, which are described separately below.

As shown in Figure 1, a specific embodiment of the present invention discloses a method for diagnosing the causes of welding defects, which includes:

S101. Determine the evaluation factors and defect causes, and establish multiple fuzzy diagnostic rules for characterizing the fuzzy relationship between the evaluation factors and the defect causes based on the fuzzy rule extending the preceding item;

S102. Establish a cause diagnosis neural network model. The input parameters of the cause diagnosis neural network model include welding defect types and the evaluation factors. The output parameters of the cause diagnosis neural network model include diagnosis results;

S103. Obtain the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed;

S104. Based on the evaluation factor data, select matching fuzzy diagnostic rules as sample rules from multiple fuzzy diagnostic rules, and perform fuzzy inference on the sample rules to obtain the actual credibility of each sample rule. Spend;

S105. Determine whether there is a sample rule whose actual credibility is greater than the preset threshold. If so, obtain a target diagnosis result based on the defect cause of the sample rule whose actual credibility is greater than the preset threshold. , if not, obtain the target diagnosis result based on the cause diagnosis neural network model.

The invention provides a welding defect cause diagnosis method, system and electronic equipment. The method establishes fuzzy diagnosis rules and cause diagnosis neural network databases, and combines the two to jointly diagnose welding defects. Specifically, it is first based on The traditional fuzzy rule expands the previous item and adds a threshold. The sample rules selected from the fuzzy diagnostic rules are used for fuzzy inference. If there is a sample rule whose actual credibility is greater than the preset threshold, it can be obtained based on the sample rule. If the target diagnosis result does not exist, it can be considered that the existing fuzzy diagnosis rules cannot handle the existing facts. At this time, using the cause diagnosis neural network for supplementary processing, the ideal target diagnosis result can be obtained. Compared with the existing technology, the present invention combines fuzzy rules and neural network theories, so that the diagnosis process not only has the advantages of fuzzy reasoning that can handle uncertain knowledge, but also has the advantages of neural network self-learning, high efficiency and high accuracy. It greatly improves the reliability of the reasoning process. Compared with the traditional fuzzy neural network model, it is more suitable for scenarios with multiple influencing factors such as welding.

It should be noted that the evaluation factors in this embodiment include any influencing factors that can cause welding defects, and the cause of defects refers to the specific quantification of influencing factors that form the defects to be diagnosed, which indicates the specific reasons for the formation of defects. The diagnostic results include the cause of the defect and the diagnostic conclusions that can be deduced based on the cause of the defect, such as the possible causes of the defect, the possibility of each cause, and the process optimization plan.

Specifically, as a preferred embodiment, step S101 in this embodiment determines the evaluation factors and the defect causes, and establishes a multi-dimensional fuzzy relationship between the evaluation factors and the defect causes based on the fuzzy rule that extends the preceding item. A fuzzy diagnosis rule, specifically including:

Establish a welding feature information database, and obtain the evaluation factors and the defect causes based on the welding feature information database;

According to the evaluation factors, multiple fuzzy diagnostic rules are established based on the fuzzy rules extending the preceding term, each of the fuzzy diagnostic rules includes the preceding term, the subsequent term, the theoretical credibility and the preset threshold, and the The former term includes a plurality of the evaluation factors, and the latter term includes the causes of defects.

Compared with the traditional fuzzy rule model, the welding causes derived from this are more in line with the diagnostic needs of actual welding.

Figure 2 shows the welding feature information database established in this embodiment, which can be directly obtained from the existing technology, and then use testing and other methods according to actual needs to determine the required evaluation factors from the welding feature information database.

In this embodiment, the determined evaluation factors include: welding method C ₀ , relative center distance of defects _{C 1} , shortest distance between two defects C 2 , defect aspect ratio C ₃ , plate thickness C ₄ , welding current C ₅ , welding voltage C _6. Welding speed C ₇ , wire feeding speed C ₈ , gas flow C ₉ ten factors.

After determining the evaluation factors, the fuzzy diagnostic rules can be established. In this embodiment, the C-F model based on fuzzy rules is used to design the fuzzy diagnostic rules and further expanded. Specifically, the previous item is expanded and the membership degree of each condition is increased. . The fuzzy diagnosis rules in this embodiment are designed as follows:

IF E ₁ (α ₁ ,η ₁ )AND E ₂ (α ₂ ,η ₂ )…AND E _n (α _n ,η _n )THEN H(CF(H),λ)

Among them, E ₁ , E ₂ ,..., E _n all correspond to various known facts, that is, the actual situation of each evaluation factor; α ₁ , α ₂ ,..., α _n are weight coefficients, that is, the influence of each fact on the cause of defects. The degree of influence is guaranteed in this embodiment η ₁ , η ₂ ,…, η _n represents the membership degree of the fact, CF(H) represents the theoretical credibility of the entire rule, and λ is the preset threshold of this rule. Among the above parameters, the weight coefficient and membership degree will be calculated accordingly according to the actual situation in subsequent steps.

The following table is part of the fuzzy diagnostic rules established in this embodiment:

Defect Cause Diagnosis Partial Rules

Inference is carried out through the above fuzzy diagnostic rules, and compared with the traditional fuzzy rule model, it is more suitable for scenarios such as welding with multiple influencing factors, improves the accuracy of judgment, can provide reasonable analysis solutions, and effectively replaces manual experts. It saves a lot of time and labor costs, helps operators reduce the impact of welding defects in a timely and effective manner, and improves the automation of welding production. At the same time, as a preferred embodiment, step S102 in this embodiment is to establish a cause diagnosis neural network model. The input parameters of the cause diagnosis neural network model include welding defect types and the evaluation factors. The cause diagnosis neural network model The output parameters include diagnostic results, specifically including:

Establish an initial BP neural network model, the input parameters of the initial BP neural network model include a plurality of the evaluation factors, and the output parameters of the initial BP neural network model include diagnostic results;

Obtain the initial parameters of the initial BP neural network model, and optimize them through the PSO-BP algorithm to obtain the optimized parameters;

The initial BP neural network model is optimized according to the optimization parameters to obtain the cause diagnosis neural network model.

Specifically, in a preferred embodiment, in the above process, the initial parameters of the initial BP neural network model are obtained and optimized through the PSO-BP algorithm to obtain the optimized parameters, which specifically includes:

Obtain initial parameters of the initial BP neural network model, where the initial parameters include weights, thresholds and loss functions of neurons;

According to the weights and thresholds of the neurons, establish a solution space and a particle population, and initialize the position of the particle population in the solution space;

According to the loss function, a fitness function is established;

According to the fitness function, optimize the position of the particle population in the solution space;

Obtain the optimized position of the particle population in the solution space and obtain the optimization parameters.

In order to understand the above process more conveniently, the present invention also provides a more specific embodiment, in which:

(1) Collect welding experimental data, and establish a single hidden layer BP neural network welding defect analysis model as the initial BP neural network model based on the process test data. The input parameters of the initial BP neural network model include multiple evaluation factors. The initial BP neural network model The output parameters of the network model include diagnostic results;

(2) In order to further improve the convergence efficiency and accuracy of the model, the initial values and parameters of the initial BP neural network model are updated and improved through the improved PSO-BP algorithm using the initial parameters of the initial BP neural network. The specific process includes:

① Use the initial parameters of the initial BP neural network to set the specific parameters of the particle swarm algorithm, set the weights and thresholds of each neuron in the initial BP neural network structure as position variables of the particles, and establish based on the weights and thresholds of the neurons. Solution space and particle population; set the loss function of the initial BP neural network to the fitness function of the particles, and initialize the population of particles, that is, initialize the position of the particle population in the solution space;

②For each example, the fitness _fi of the particle can be calculated and compared with the minimum fitness f _min of the particle. If _fi <f _min , the updated fitness will take the value of The minimum fitness f _min of the particle, the corresponding position is the best position _Pi ;

③The particle swarm searches the entire state space by updating its own speed and position, and by tracking the individual extreme value (the optimal solution with the best fitness found by the particle itself) and the global extreme value (the best fitness currently found by the entire population) The optimal solution) updates itself until the optimal solution is found;

The evolutionary formula of particle speed and position in the above process is:

In the formula, ω is the nonlinear dynamic change inertia weight; c ₁ and c ₂ are learning factors; r ₁ and r ₂ are random numbers with values between [0,1]; p _in (k) represents the individual position of the particle Extreme value; p _gn (k) represents the extreme value of the global position of the particle; x _in (k) is the position of particle i at time k; x _in (k+1) is the position of particle i at time k+1; v _in ( k) is the speed of particle i at time k; v _in (k+1) is the speed of particle i at time k+1;

Among them, the above nonlinear dynamic change inertia weight ω is expressed as:

Among them, ω _min is the minimum inertia weight; ω _max is the maximum inertia weight, l ₁ (t) is a nonlinear function, l ₂ (t) is a linear function, f _i is the fitness of the current particle, and f _min is the particle Minimum fitness. l ₁ (t), l ₂ (t) are determined by the following formula:

In the formula, t is the current iteration number of the model, and t _max is the maximum iteration number of the model;

④ Calculate the minimum fitness f _g =min{f ₁ , f ₂ ,..., f _M } of the entire particle swarm, and determine whether the current number of iterations reaches the set maximum number of iterations. If it reaches the set maximum number of iterations, the iteration will stop, otherwise go to ② to continue. renew;

⑤ Output the neural network parameters P _g = (p _g1 , p _g2 ,..., p _gD ) determined by the entire particle swarm as the initial value of the neural network structure. Then the particle swarm algorithm is combined with the gradient descent algorithm, and dynamic weight factors are used to further iteratively update the neural network parameters. The initial parameters of the neural network determined by the improved PSO-BP algorithm are output, and the corresponding parameters of the initial BP neural network are replaced to optimize the initial BP neural network model and obtain the cause diagnosis neural network model.

The following table shows the specific values of some other parameters required in the above process of this embodiment:

Adaptive PSO-BP algorithm parameter setting based on dynamic weight

In the neural network structure of this embodiment, the network parameter update method retains the original gradient descent algorithm, and introduces dynamic weight factors to combine the PSO algorithm and the BP algorithm to take advantage of different algorithms at different stages of iteration to improve Training speed and iteration efficiency. Use the trained PSO-BP neural network model to supplement the facts that cannot be determined by fuzzy rule reasoning, determine the cause of welding defects, and modify the model based on the evaluation results, such as using the output results as new fuzzy diagnosis rules, so that The entire method has the characteristics of self-learning, which effectively makes up for the problem of existing technologies that rely too much on existing samples, and effectively improves the training efficiency and diagnostic accuracy of the model. It combines rule reasoning and model reasoning to form a rule-model collaborative reasoning mechanism. Compared with traditional neural networks or fuzzy neural networks, it has increased self-learning capabilities, is more suitable for scenarios with multiple influencing factors such as welding, and improves the reliability of the reasoning method.

After the fuzzy diagnosis rules and the cause diagnosis neural network model are established, the actual diagnosis can be started. First, step S103 needs to be performed to obtain the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed, where the evaluation factor data is The actual value corresponding to the evaluation factor of the defect.

Further, as shown in FIG. 3 , as a preferred embodiment, step S104 in this embodiment is to select a matching fuzzy diagnostic rule from a plurality of the fuzzy diagnostic rules as a sample rule based on the evaluation factor data. And perform fuzzy inference on the sample rules to obtain the actual credibility of each sample rule, which specifically includes:

S301. Based on the evaluation factor data, select a matching fuzzy diagnosis rule from a plurality of the diagnosis rules as the sample rule;

S302. According to the actual welding defect type, based on the analytic hierarchy process, obtain the weight coefficient of each evaluation factor for the actual welding defect type;

S303. According to the evaluation factor data, obtain the membership degree of each evaluation factor in the sample rule;

S304: Calculate the actual credibility of each sample rule based on the weight coefficient and the membership degree.

The above process is the main step of fuzzy inference in this embodiment. It can be understood that because fuzzy inference is an existing technology that can be understood by those skilled in the relevant field, in practice, other fuzzy inference methods can also be used to obtain credibility. .

As a preferred embodiment, this embodiment first eliminates obviously unreasonable rules from multiple fuzzy diagnosis rules, selects appropriate rules as sample rules, and then proceeds to step S302. Specifically, please refer to Figure 4. Step S302 in: According to the actual welding defect type, based on the analytic hierarchy process, obtain the weight coefficient of each evaluation factor for the actual welding defect type, specifically including:

S401. Establish a pairwise comparison matrix of the evaluation factors according to the actual welding defect type;

S402. Perform a consistency test on the pairwise comparison matrix, and adjust the pairwise comparison matrix according to the test results;

S403. Obtain the eigenvector of the adjusted pairwise comparison matrix according to the adjusted pairwise comparison matrix;

S404: Normalize the feature vector to obtain the weight coefficient of each evaluation factor for the actual welding defect type.

The present invention also provides a more specific embodiment to illustrate the above processes S401 to S404:

According to the actual welding type, the relative importance of each evaluation factor is determined by experts in the field, and then compared in pairs. The relative comparison scale of 1 to 5 is used for comparison as shown in the following table:

relative comparison scale

Using the above relative comparison scale, the pairwise comparison matrix A is obtained as follows:

The calculation of each element in matrix A is calculated according to the following formula:

Among them, i, j, k = 0, 1, 2, 3,..., n.

According to the obtained pairwise comparison matrix A, the eigenvector corresponding to the largest eigenvalue λ _max is obtained as the weight vector ω, that is, Aω=λ _max . The weight vector is normalized to obtain the final weight coefficient.

Since different evaluation factors have different effects on different types of defect characteristics, the above process needs to be flexibly set according to the actual welding defect type. In this embodiment, the "welding penetration" defect is taken as an example to compare the importance of different evaluation factors on the causes of welding defects. property, the pairwise comparison matrix A is obtained as follows:

After obtaining the comparison matrix, it needs to be tested for consistency to verify its rationality. The comparison process can be carried out by calculating the consistency index CR:

It can be seen from the above formula that the established pairwise comparison matrix has reasonable weight distribution for relevant parameters and has high credibility, and the weight coefficient can be calculated. In this embodiment, the final weight coefficient vector β of “welding penetration” obtained based on the above matrix is:

β＝(0.25,0.05,0.05,0.06,0.09,0.13,0.13,0.09,0.09,0.06)

The above process is only a weight coefficient for an actual welding defect type. In practice, different welding defect types need to separately establish a pairwise comparison matrix and calculate the weight coefficient. In this embodiment, there are ten influencing factors C _i , corresponding to four different types of defects S _i , and the final defect weight coefficient table is obtained as shown in the following table.

Defect weight coefficient table

After the weight coefficient is determined, steps S303 and S304 can be performed, in which the membership function is used to determine the membership degree of each fact based on the evaluation factor data. The specific process is an existing technology and will not be explained in detail here. Then step S304 is performed. In this embodiment, the actual credibility λ _g of each fuzzy diagnosis rule can be calculated by the following formula:

For ease of understanding, the present invention also provides a specific embodiment to illustrate the calculation process of the above-mentioned actual credibility. The evaluation factor data obtained in this example are: welding method M "pulse arc welding", relative center distance of the defect l ₁ "3mm", shortest distance l ₂ "1.2mm" between two defects, defect aspect ratio δ "1.1" , plate thickness b "6mm", welding current I "142A", welding voltage U "19.8V", welding speed V _h "0.36m/min", wire feed speed V _s "9.0m/min", gas flow rate V _q "25L". The actual welding defect type is "pores". Through the above process, it can be seen that the weight coefficient S ₃ of each evaluation factor for defects such as "pores" is 0.25, 0.05, 0.05, 0.06, 0.08, 0.06, 0.06, 0.13, 0.13, 0.13, and then The actual credibility λ _g1 , λ g2 , and λ _g3 of the above three fuzzy diagnostic rules (i.e., rule 1, rule 2, and _rule 3 in the previous table) can be obtained:

λ _g1 =0.95×(0.25×1+0.05×1+0.05×1+0.06×1+0.08×1+0.06×0.9+0.06×0.9+0.13×1+0.13×1+0.13×1)=0.9386

λ _g2 =0.95×(0.25×1+0.05×1+0.05×1+0.06×1+0.08×1+0.06×0.9+0.06×0.9+0.13×1+0.13×1+0.13×0)=0.8151

λ _g3 =0.90×(0.25×1+0.05×1+0.05×1+0.06×1+0.08×1+0.06×0.9+0.06×0.9+0.13×0.5+0.13×1+0.13×0.5)=0.7722

After calculating the actual credibility, step S105 can be performed to determine whether there is a sample rule whose actual credibility is greater than the preset threshold. If so, based on all the sample rules whose actual credibility is greater than the preset threshold, Describe the defect causes of the sample rules and obtain the target diagnosis result. If not, obtain the target diagnosis result according to the neural network model. As a preferred embodiment, this step in this embodiment specifically includes:

Sort the sample rules according to the numerical value of the actual credibility to obtain a sample rule sequence;

Determine whether there is a sample rule whose actual credibility is greater than its corresponding preset threshold. If so, select a set number of samples with the highest actual credibility based on the sample rule sequence. According to the regular defect causes, the target diagnosis result is obtained. If not, the evaluation factor data and the actual welding defect type are input into the cause diagnosis neural network model, and the target diagnosis result is obtained.

In this embodiment, by comparing the above three actual credibility with the preset threshold λ of each rule, it can be obtained that the comprehensive credibility of rule 1 λ _g1 = 0.9386 is greater than the threshold λ = 0.88, and the other two The λ _g of the rule is smaller than λ, so only rule 1 is activated and used. Finally, it can be concluded based on rule 1 that the cause of the diagnostic defect may be excessive gas flow, and its credibility is 0.9386.

It can be understood that in this embodiment, each actual credibility is compared with the preset threshold of its corresponding rule itself. In practice, the same preset threshold can also be set uniformly for comparison. When there is a rule with λ _g ≥ λ, it means that the defect cause diagnosis result corresponding to the rule is credible. After all the rules are retrieved, if there are multiple fuzzy diagnosis rules whose actual credibility is greater than the preset threshold, then it can Select a set number (for example, three) of fuzzy diagnosis rules from them and use them as the final diagnosis results.

When the comprehensive credibility λg of all rules is less than the threshold λ, it means that the existing fuzzy diagnosis rules are not credible, and the neural network model can be further selected for reasoning. The input features are normalized and imported into the model, and the trained cause diagnosis neural network model is used for reasoning. The network model will output a reasonable cause of welding defects in a specific form (such as the probability of each defect cause), and then the user will The model output results are evaluated, and finally a defect cause diagnosis plan or a model correction report is obtained.

In order to verify the application effect of the above method in actual production, the present invention randomly selected 5 test samples in the enterprise's tank production. The test samples are shown in the table below.

test sample

During the test, the system gave inference results based on fuzzy rules for Samples 2 to 5, as shown in the table below. The true value of the defect cause is the actually obtained evaluation factor data in the previous article, and the credibility of the rule inference is Actual credibility in the previous article:

True value of defect cause and rule inference result

For sample 1, there is a situation where the comprehensive credibility of the rule λ _g is lower than the preset threshold λ, so the cause diagnosis neural network model is used for auxiliary reasoning. The true value of the defect cause and the network output results are shown in the following table:

Actual values of defect causes and model output results

Taking the diagnosis of aluminum alloy welding pore defects (sample 1) in the enterprise as an example for testing, it can be seen that through the assisted diagnosis of the adaptive PSO-BP neural network model based on dynamic weights, it is inferred that the input pore defects are 99.51% likely The problem is caused by excessive welding speed, 0.38% possibility is caused by excessive gas flow, and 0.11% possibility is caused by too small gas flow.

In order to better implement the welding defect cause diagnosis method in the embodiment of the present invention, based on the welding defect cause diagnosis method, correspondingly, please refer to Figure 5. Figure 5 is an embodiment of the welding defect cause diagnosis system provided by the present invention. The structural schematic diagram of a welding defect cause diagnosis system 500 provided by an embodiment of the present invention includes:

The rule establishment module 510 is used to determine evaluation factors and defect causes, and establish multiple fuzzy diagnostic rules for characterizing the fuzzy relationship between the evaluation factors and the defect causes based on the fuzzy rules of the extended preceding term;

The network establishment module 520 is used to establish a cause diagnosis neural network model. The input parameters of the cause diagnosis neural network model include welding defect types and the evaluation factors, and the output parameters of the cause diagnosis neural network model include diagnosis results;

The data acquisition module 530 is used to obtain the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed;

The fuzzy inference module 540 is configured to select a matching fuzzy diagnostic rule as a sample rule from a plurality of the fuzzy diagnostic rules based on the evaluation factor data, and perform fuzzy inference on the sample rule to obtain each of the samples. the actual credibility of the rules;

The result diagnosis module 550 is used to determine whether there is a sample rule whose actual credibility is greater than a preset threshold. If so, determine the defect causes of the sample rule based on the actual credibility greater than the preset threshold. , obtain the target diagnosis result, if not, obtain the target diagnosis result according to the neural network model.

It should be noted here that the corresponding system 500 provided by the above embodiments can implement the technical solutions described in the above method embodiments. The specific implementation principles of each of the above modules or units can be found in the corresponding content in the above method embodiments. Here No longer.

In addition, the present invention also provides an embodiment of a welding defect cause diagnosis system, which system includes:

Database module: The function of retrieving and querying data on materials, processes and defect identification features required for welding saves the cost of data storage and the time cost of querying data. At the same time, system administrators are given authority to add, delete, and maintain data within the system;

Defect identification module: Users can upload X-ray films to the system, and the system will use super-resolution reconstructed convolutional neural networks to identify the X-ray films to obtain relevant characteristics of welding defects, such as defect type, area, length and width ratio;

Defect cause diagnosis module: Combine the data in the database and defect identification features to diagnose the cause of welding defects. The system will use the welding defect cause diagnosis method described in this article to determine the cause of the welding defect with the highest confidence;

Process optimization module: The system designed in this article includes two optimization methods. Mode 1 starts from the defect cause diagnosis level. After the cause of the defect is found, a process optimization plan is given based on the cause of the defect to obtain a defect-free weld; Mode 2 Starting from the welding seam performance, the multi-objective particle swarm algorithm (MOPSO) is used to perform multi-objective optimization of the welding seam performance, and the welding seam performance indicators that meet industrial needs are obtained by regulating the process parameters;

User rights management module: For the security of the database, user rights are divided into system administrators, system operators and ordinary users. System administrators can use the user rights management module to grant relevant permissions to all users in the system; system operators can add, delete, and maintain the database; ordinary users can only search and consult the database, but cannot perform other operations on the database.

Further, please refer to FIG. 6 , which is a schematic structural diagram of an electronic device according to an embodiment of the present invention. Based on the above welding defect cause diagnosis method, the present invention also provides a welding defect cause diagnosis equipment 600, that is, the above-mentioned electronic device. The welding defect cause diagnosis equipment 600 can be a mobile terminal, a desktop computer, a notebook, a palmtop computer, and a server. and other computing equipment. The welding defect cause diagnosis equipment 600 includes a processor 610, a memory 620 and a display 630. FIG. 6 only shows some components of the welding defect cause diagnosis equipment, but it should be understood that implementation of all the components shown is not required, and more or less components may be implemented instead.

In some embodiments, the memory 620 may be an internal storage unit of the welding defect cause diagnosis device 600 , such as a hard disk or memory of the welding defect cause diagnosis device 600 . In other embodiments, the memory 620 may also be an external storage device of the welding defect cause diagnosis device 600 , such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), or a secure digital device equipped on the welding defect cause diagnosis device 600 . (Secure Digital, SD) card, Flash Card, etc. Further, the memory 620 may also include both an internal storage unit of the welding defect cause diagnosis device 600 and an external storage device. The memory 620 is used to store application software and various data installed in the welding defect cause diagnosis equipment 600 , such as program codes for installing the welding defect cause diagnosis equipment 600 . The memory 620 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a welding defect cause diagnosis program 640 is stored in the memory 620, and the welding defect cause diagnosis program 640 can be executed by the processor 610, thereby implementing the welding defect cause diagnosis methods in various embodiments of the present application.

In some embodiments, the processor 610 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 620, for example, to perform welding defect causation. Diagnostic methods, etc.

In some embodiments, the display 630 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. The display 630 is used to display information on the welding defect cause diagnosis device 600 and to display a visual user interface. The components 610 - 630 of the welding defect cause diagnosis apparatus 600 communicate with each other through the system bus.

In one embodiment, when the processor 610 executes the welding defect cause diagnosis program 640 in the memory 620, the above steps in the welding defect cause diagnosis method are implemented.

The invention provides a welding defect cause diagnosis method, system and electronic equipment. The method establishes fuzzy diagnosis rules and cause diagnosis neural network databases, and combines the two to jointly diagnose welding defects. Specifically, it is first based on The traditional fuzzy rule expands the previous item and adds a threshold. The sample rules selected from the fuzzy diagnostic rules are used for fuzzy inference. If there is a sample rule whose actual credibility is greater than the preset threshold, it can be obtained based on the sample rule. If the target diagnosis result does not exist, it can be considered that the existing fuzzy diagnosis rules cannot handle the existing facts. At this time, using the cause diagnosis neural network for supplementary processing, the ideal target diagnosis result can be obtained. Compared with the existing technology, the present invention combines fuzzy rules and neural network theories, so that the diagnosis process not only has the advantages of fuzzy reasoning that can handle uncertain knowledge, but also has the advantages of neural network self-learning, high efficiency and high accuracy. It greatly improves the reliability of the reasoning process. Compared with the traditional fuzzy neural network model, it is more suitable for scenarios with multiple influencing factors such as welding.

The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or modifications within the technical scope disclosed in the present invention. All substitutions are within the scope of the present invention.

Claims (7)

1. A welding defect cause diagnosis method, comprising:

determining an evaluation factor and a defect cause, and establishing a plurality of fuzzy diagnosis rules for representing fuzzy relations between the evaluation factor and the defect cause based on fuzzy rules of an expansion front;

establishing a cause diagnosis neural network model, wherein input parameters of the cause diagnosis neural network model comprise welding defect types and evaluation factors, and output parameters of the cause diagnosis neural network model comprise diagnosis results;

acquiring actual welding defect type and evaluation factor data corresponding to defects to be diagnosed;

based on the evaluation factor data, selecting a matched fuzzy diagnosis rule from a plurality of fuzzy diagnosis rules as a sample rule, and performing fuzzy reasoning on the sample rule to obtain the actual credibility of each sample rule;

Judging whether the sample rule with the actual credibility larger than a preset threshold exists or not, if so, obtaining a target diagnosis result according to the defect cause of the sample rule with the actual credibility larger than the preset threshold, and if not, obtaining the target diagnosis result according to the cause diagnosis neural network model;

wherein the determining the evaluation factor and the defect cause, establishing a plurality of fuzzy diagnosis rules for representing fuzzy relations between the evaluation factor and the defect cause based on the fuzzy rules of the expansion foreterm, comprises the following steps:

establishing a welding characteristic information base, and obtaining the evaluation factors and the defect causes according to the welding characteristic information base;

establishing a plurality of fuzzy diagnosis rules based on fuzzy rules of the extension antecedents according to the evaluation factors;

each fuzzy diagnosis rule comprises a front term, a rear term, theoretical credibility and the preset threshold value, wherein the front term comprises a plurality of evaluation factors, and the rear term comprises the defect cause;

the fuzzy diagnosis rule is designed by adopting a C-F model based on the fuzzy rule, and the antecedents of the fuzzy diagnosis rule are expanded, so that the membership degree of each condition is increased;

The step of selecting a matched fuzzy diagnosis rule from a plurality of fuzzy diagnosis rules as a sample rule based on the evaluation factor data, and performing fuzzy reasoning on the sample rule to obtain the actual credibility of each sample rule, wherein the step of obtaining the actual credibility of each sample rule comprises the following steps:

selecting a matched fuzzy diagnosis rule from a plurality of diagnosis rules as the sample rule based on the evaluation factor data;

according to the actual welding defect type, obtaining a weight coefficient of each evaluation factor for the actual welding defect type based on an analytic hierarchy process;

obtaining the membership degree of each evaluation factor in the sample rule according to the evaluation factor data;

and calculating the actual credibility of each sample rule according to the weight coefficient and the membership degree.

2. The welding defect cause diagnosis method according to claim 1, wherein the obtaining a weight coefficient of each evaluation factor for the actual welding defect type based on a hierarchical analysis method according to the actual welding defect type comprises:

establishing a pair comparison matrix of the evaluation factors according to the actual welding defect type;

Consistency test is carried out on the paired comparison matrixes, and the paired comparison matrixes are adjusted according to test results;

obtaining the feature vector of the pair of adjusted comparison matrixes according to the pair of adjusted comparison matrixes;

and normalizing the feature vector to obtain a weight coefficient of each evaluation factor for the actual welding defect type.

3. The method according to claim 1, wherein the determining whether the sample rule having the actual reliability greater than the preset threshold exists, if so, obtaining a target diagnosis result according to a defect cause of the sample rule having the actual reliability greater than the preset threshold, and if not, obtaining a target diagnosis result according to the cause diagnosis neural network model, includes:

sequencing the sample rule according to the numerical value of the actual credibility to obtain a sample rule sequence;

judging whether the actual credibility of the sample rule is larger than the corresponding preset threshold value, if so, selecting the defect causes of the set number of the sample rules with the highest actual credibility according to the sample rule sequence to obtain the target diagnosis result, and if not, inputting the evaluation factor data and the actual welding defect type into the cause diagnosis neural network model to obtain the target diagnosis result.

4. The welding defect cause diagnosis method according to claim 1, wherein the establishing a cause diagnosis neural network model includes:

establishing an initial BP neural network model, wherein the input parameters of the initial BP neural network model comprise a plurality of evaluation factors, and the output parameters of the initial BP neural network model comprise the diagnosis results;

acquiring initial parameters of the initial BP neural network model, and optimizing through a PSO-BP algorithm to obtain optimized parameters;

and optimizing the initial BP neural network model according to the optimization parameters to obtain the causal diagnosis neural network model.

5. The welding defect cause diagnosis method according to claim 4, wherein the obtaining initial parameters of the initial BP neural network model and optimizing by a PSO-BP algorithm to obtain optimized parameters comprises:

acquiring initial parameters of the initial BP neural network model, wherein the initial parameters comprise weights, thresholds and loss functions of neurons;

establishing a solution space and a particle population according to the weight and the threshold of the neuron, and initializing the position of the particle population in the solution space;

Establishing an adaptability function according to the loss function;

optimizing the position of the particle population in the solution space according to the fitness function;

and acquiring the position of the optimized particle population in the solution space to obtain the optimization parameters.

6. A welding defect cause diagnosis system, comprising:

the rule establishing module is used for determining an evaluation factor and a defect cause, and establishing a plurality of fuzzy diagnosis rules for representing fuzzy relations between the evaluation factor and the defect cause based on fuzzy rules of an expansion front term;

the network establishment module is used for establishing a cause diagnosis neural network model, wherein the input parameters of the cause diagnosis neural network model comprise welding defect types and evaluation factors, and the output parameters of the cause diagnosis neural network model comprise diagnosis results;

the data acquisition module is used for acquiring the actual welding defect type and evaluation factor data corresponding to the defect to be diagnosed;

the fuzzy reasoning module is used for selecting a matched fuzzy diagnosis rule from a plurality of fuzzy diagnosis rules as a sample rule based on the evaluation factor data, and carrying out fuzzy reasoning on the sample rule to obtain the actual credibility of each sample rule;

The result diagnosis module is used for judging whether the sample rule with the actual credibility being larger than a preset threshold exists or not, if yes, obtaining a target diagnosis result according to the defect cause of the sample rule with the actual credibility being larger than the preset threshold, and if not, obtaining the target diagnosis result according to the neural network model;

wherein the determining the evaluation factor and the defect cause, establishing a plurality of fuzzy diagnosis rules for representing fuzzy relations between the evaluation factor and the defect cause based on the fuzzy rules of the expansion foreterm, comprises the following steps:

establishing a welding characteristic information base, and obtaining the evaluation factors and the defect causes according to the welding characteristic information base;

establishing a plurality of fuzzy diagnosis rules based on fuzzy rules of the extension antecedents according to the evaluation factors;

each fuzzy diagnosis rule comprises a front term, a rear term, theoretical credibility and the preset threshold value, wherein the front term comprises a plurality of evaluation factors, and the rear term comprises the defect cause;

the fuzzy diagnosis rule is designed by adopting a C-F model based on the fuzzy rule, and the antecedents of the fuzzy diagnosis rule are expanded, so that the membership degree of each condition is increased;

The step of selecting a matched fuzzy diagnosis rule from a plurality of fuzzy diagnosis rules as a sample rule based on the evaluation factor data, and performing fuzzy reasoning on the sample rule to obtain the actual credibility of each sample rule, wherein the step of obtaining the actual credibility of each sample rule comprises the following steps:

selecting a matched fuzzy diagnosis rule from a plurality of diagnosis rules as the sample rule based on the evaluation factor data;

according to the actual welding defect type, obtaining a weight coefficient of each evaluation factor for the actual welding defect type based on an analytic hierarchy process;

obtaining the membership degree of each evaluation factor in the sample rule according to the evaluation factor data;

and calculating the actual credibility of each sample rule according to the weight coefficient and the membership degree.

7. An electronic device comprising a memory and a processor, wherein,

the memory is used for storing programs;

the processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps in the welding defect cause diagnosis method according to any one of the preceding claims 1 to 5.