A method and device for realizing sustainable optimization of adjustment parameters of aerospace products

Abstract

The application provides a method and device for realizing sustainable optimization of adjustment parameters of aerospace products, and the method comprises the following steps: classifying adjustment parameters of aerospace products based on a pre-constructed parameter classifier to obtain a parameter classification result; inputting adjustment parameters with abnormal parameter classification results into a pre-established fault classification and identification model to obtain a fault classification and identification result; inputting the fault classification and identification result into a performance judgment network to obtain a performance judgment result corresponding to the adjustment parameters; when the performance judgment result comprises a fault appearance, determining an abnormal probability of the adjustment parameters according to a mapping relationship between the adjustment parameters and the fault appearance of the aerospace product; determining an optimization interval of the adjustment parameters with the maximum abnormal probability according to application debugging data of the aerospace product; and debugging the aerospace product according to the optimization interval. The scheme of the application realizes intelligent judgment of the performance of the aerospace product and intelligent positioning of faults, and optimizes the parameters according to the application debugging data.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of precision assembly and adjustment, and particularly relates to a method and device for realizing sustainable optimization of assembly and adjustment parameters of aerospace products. BACKGROUND

[0002] The aerospace product is a highly complex optoelectromechanical terminal guidance system, and whole-machine debugging corresponds to multiple key performance indicators. Each possible fault can cause the multiple key performance indicators to be abnormal, and the multiple key performance indicators have certain coupling relationships. It is difficult to establish an indirect analytical mapping relationship between assembly and adjustment parameter abnormalities and fault causes, and the positioning of determining whether the aerospace product is qualified and the fault problem often occupies most of the working hours of the aerospace product.

[0003] The existing performance indicator data qualified range of the aerospace product can ensure that the aerospace product is qualified in performance, but the various data may not be in the parameter interval with the optimal performance. Moreover, the existing method for realizing sustainable optimization of assembly and adjustment parameters of the aerospace product has the problem that the qualification determination does not well meet the performance indicator requirements, so it is necessary to optimize the debugging parameters of the aerospace product, and more reliable data and fault analysis are needed to further improve the performance of the aerospace product. SUMMARY

[0004] The purpose of the embodiment of the application is to provide a method and device for realizing sustainable optimization of assembly and adjustment parameters of aerospace products, thereby solving the problem of how to optimize the debugging of the assembly and adjustment parameters of the aerospace product in the prior art.

[0005] In order to achieve the above purpose, the embodiment of the application provides a method for realizing sustainable optimization of assembly and adjustment parameters of aerospace products, comprising:

[0006] Classifying the assembly and adjustment parameters of the aerospace product based on a pre-constructed parameter classifier to obtain a parameter classification result;

[0007] Inputting the assembly and adjustment parameter with the parameter classification result into a pre-established fault classification and identification model to obtain a fault classification and identification result;

[0008] Inputting the fault classification and identification result into a performance judgment network to obtain a performance judgment result corresponding to the assembly and adjustment parameter;

[0009] When the performance judgment result includes a fault manifestation, determining an abnormal probability of the assembly and adjustment parameter according to a mapping relationship between the assembly and adjustment parameter of the aerospace product and the fault manifestation;

[0010] Determining an optimization interval of the assembly and adjustment parameter with the maximum abnormal probability according to application debugging data of the aerospace product;

[0011] According to the optimization interval, debugging the aerospace product.

[0012] Optionally, the method further comprises:

[0013] According to the application debugging data, it is determined that the parameter classification result is an optimization interval of the normal installation and adjustment parameter.

[0014] Optionally, the method further comprises:

[0015] When the performance judgment result is performance qualified, according to the application debugging data, it is determined that the performance judgment result is an optimization interval of the installation and adjustment parameter of the performance qualified.

[0016] Optionally, the method further comprises:

[0017] The normal installation and adjustment parameters of the aerospace product of the past batches of the batch production model are expanded to obtain an expanded data set;

[0018] The support vector machine classifier is trained by using the expanded data set to obtain the installation and adjustment parameter classifier.

[0019] Optionally, the normal installation and adjustment parameters of the aerospace product of the past batches of the batch production model are expanded to obtain an expanded data set, comprising:

[0020] Based on the cycle generative adversarial network, the normal installation and adjustment parameters are expanded to obtain abnormal installation and adjustment parameters;

[0021] According to the normal installation and adjustment parameters and the abnormal installation and adjustment parameters, the expanded data set is obtained.

[0022] Optionally, the method further comprises:

[0023] Based on the first deep neural network model, the performance indicators and fault data of the aerospace product of the past batches of the batch production model are pre-trained to obtain a pre-training model;

[0024] Based on the association rule mining algorithm, a mapping relationship between the installation and adjustment parameters and the fault manifestations of the aerospace product is obtained.

[0025] According to the mapping relationship and the pre-training model, the fault classification identification model is obtained.

[0026] Optionally, the method further comprises:

[0027] According to the abnormal installation and adjustment parameters of the aerospace product of the past batches of the batch production model, the fault classification identification model is fine-tuned.

[0028] Optionally, the fault classification identification result is input into the performance judgment network to obtain a performance judgment result corresponding to the installation and adjustment parameter, comprising:

[0029] The performance judgment network is used to obtain a plurality of occurrence probabilities of different performance judgment results corresponding to the fault classification recognition result.

[0030] The plurality of occurrence probabilities are sorted to obtain a performance judgment result with the maximum occurrence probability.

[0031] Optionally, when the performance judgment result includes a fault manifestation, an abnormal probability of a mounting and adjusting parameter is determined according to a mapping relationship between the mounting and adjusting parameter and the fault manifestation of the aerospace product, and the abnormal probability of the mounting and adjusting parameter includes:

[0032] The abnormal probability of each mounting and adjusting parameter corresponding to different fault manifestations is calculated according to a Bayesian algorithm and the mapping relationship.

[0033] Optionally, an optimization interval of the mounting and adjusting parameter with the maximum abnormal probability is determined according to application debugging data of the aerospace product, and the optimization interval includes:

[0034] The application debugging data is input into a second deep neural network model constructed in advance to determine the optimization interval of the mounting and adjusting parameter with the maximum abnormal probability.

[0035] The second deep neural network model is obtained by training application debugging success data and application debugging fault data.

[0036] The embodiment of the application further provides a device for realizing sustainable optimization of mounting and adjusting parameters of an aerospace product, and the device includes:

[0037] A classification module is configured to classify mounting and adjusting parameters of the aerospace product based on a pre-constructed parameter classifier to obtain a parameter classification result.

[0038] An input module is configured to input the mounting and adjusting parameter with the abnormal parameter classification result into a pre-established fault classification recognition model to obtain a fault classification recognition result.

[0039] A judgment module is configured to input the fault classification recognition result into a performance judgment network to obtain a performance judgment result corresponding to the mounting and adjusting parameter.

[0040] A first determination module is configured to determine an abnormal probability of the mounting and adjusting parameter according to a mapping relationship between the mounting and adjusting parameter and a fault manifestation of the aerospace product when the performance judgment result includes the fault manifestation.

[0041] A second determination module is configured to determine an optimization interval of the mounting and adjusting parameter with the maximum abnormal probability according to application debugging data of the aerospace product.

[0042] A debugging module is configured to debug the aerospace product according to the optimization interval.

[0043] Optionally, the device further includes:

[0044] The third determining module is configured to determine, according to the application debugging data, an optimization interval of the parameter classified as a normal installation and adjustment parameter.

[0045] Optionally, the device further comprises:

[0046] The fourth determining module is configured to, when the performance judgment result is performance qualified, determine, according to the application debugging data, an optimization interval of the installation and adjustment parameter of which the performance judgment result is performance qualified.

[0047] Optionally, the device further comprises:

[0048] The first obtaining module is configured to expand normal installation and adjustment parameters of the aerospace product of the past batch of the batch production type, and obtain an expanded data set.

[0049] The first training module is configured to train a support vector machine classifier by using the expanded data set, and obtain the installation and adjustment parameter classifier.

[0050] Optionally, the first obtaining module is specifically configured to:

[0051] The normal installation and adjustment parameters are expanded based on a cycle generative adversarial network, and abnormal installation and adjustment parameters are obtained.

[0052] The expanded data set is obtained according to the normal installation and adjustment parameters and the abnormal installation and adjustment parameters.

[0053] Optionally, the device further comprises:

[0054] The second training module is configured to pre-train, based on a first deep neural network model, performance indicators and fault data of the aerospace product of the past batch of the batch production type, and obtain a pre-training model.

[0055] The first obtaining module is configured to obtain a mapping relationship between installation and adjustment parameters and fault manifestations of the aerospace product based on an association rule mining algorithm.

[0056] The second obtaining module is configured to obtain the fault classification and identification model according to the mapping relationship and the pre-training model.

[0057] Optionally, the device further comprises:

[0058] The third training module is configured to fine-tune train the fault classification and identification model according to abnormal installation and adjustment parameters of the aerospace product of the past batch of the batch production type.

[0059] Optionally, the judging module is specifically configured to:

[0060] The performance judgment network is used to obtain a plurality of occurrence probabilities of different performance judgment results corresponding to the fault classification and identification result.

[0061] The plurality of occurrence probabilities are sorted to obtain a performance judgment result with the largest occurrence probability.

[0062] Optionally, the first determining module is specifically configured to:

[0063] According to the Bayes algorithm and the mapping relationship, an abnormal probability of each installation and adjustment parameter corresponding to different fault manifestations is calculated.

[0064] Optionally, the second determining module is specifically configured to:

[0065] The application debugging data is input into a second pre-constructed deep neural network model to determine an optimization interval of the installation and adjustment parameter with the largest abnormal probability.

[0066] The second deep neural network model is obtained by training application debugging success data and application debugging failure data.

[0067] The embodiment of the application further provides an electronic device, including a processor, a memory, and a program stored in the memory and executable on the processor, and when the program is executed by the processor, the steps of the method for realizing sustainable optimization of installation and adjustment parameters of a space product are implemented.

[0068] The embodiment of the application further provides a readable storage medium, and the readable storage medium stores a program, and when the program is executed by a processor, the steps of the method for realizing sustainable optimization of installation and adjustment parameters of a space product are implemented.

[0069] The above technical solution of the application has at least the following beneficial effects:

[0070] In the above scheme, the installation and adjustment parameters of the space product are classified based on a pre-constructed parameter classifier to obtain a parameter classification result; the installation and adjustment parameters with abnormal parameter classification results are input into a pre-established fault classification and identification model to obtain a fault classification and identification result; the fault classification and identification result is input into a performance judgment network to obtain a performance judgment result corresponding to the installation and adjustment parameters; when the performance judgment result includes a fault manifestation, the abnormal probability of the installation and adjustment parameter is determined according to the mapping relationship between the installation and adjustment parameters and the fault manifestation of the space product; the optimization interval of the installation and adjustment parameter with the largest abnormal probability is determined according to the application debugging data of the space product; and the space product is debugged according to the optimization interval, so as to realize qualified judgment of the space product and intelligent positioning of fault problems, and the installation and adjustment parameters are continuously debugged to more quickly and better meet the performance index requirements, and the sustainable optimization of the installation and adjustment parameters is realized. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1A step schematic diagram of a method for realizing sustainable optimization of aerospace product installation and adjustment parameters according to an embodiment of the present application;

[0072] Figure 2 A schematic diagram of a cyclic generative adversarial network according to an embodiment of the present application;

[0073] Figure 3 A second schematic diagram of a cyclic generative adversarial network according to an embodiment of the present application;

[0074] Figure 4 A schematic diagram of a first deep neural network model according to an embodiment of the present application;

[0075] Figure 5 An application flow schematic diagram of a method for realizing sustainable optimization of aerospace product installation and adjustment parameters according to an embodiment of the present application;

[0076] Figure 6 A block diagram of a device for realizing sustainable optimization of aerospace product installation and adjustment parameters according to an embodiment of the present application;

[0077] Figure 7 A block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0078] To make the technical problems, technical solutions and advantages of the present application clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0079] Embodiments of the present application provide a method and device for realizing sustainable optimization of aerospace product installation and adjustment parameters to solve the problem of how to optimize and debug aerospace product installation and adjustment parameters in the prior art.

[0080] As shown in Figure 1 The present application provides a method for realizing sustainable optimization of aerospace product installation and adjustment parameters, comprising:

[0081] Step 101, based on a pre-constructed parameter classifier, classifying installation and adjustment parameters of an aerospace product to obtain a parameter classification result;

[0082] It should be noted that the embodiments of the present application are applicable to performance debugging of aerospace products, and are also applicable to other products that need to optimize performance parameters. Here, the parameter classification result includes normal installation and adjustment parameters or abnormal installation and adjustment parameters.

[0083] Step 102, inputting the parameter classification result of the abnormal installation and adjustment parameters into a pre-established fault classification and identification model to obtain a fault classification and identification result;

[0084] It should be noted that the embodiment of the present application can also input the abnormal assembly parameter into the pre-established fault tree to obtain a fault classification recognition result, so as to determine whether the assembly parameter of the aerospace product is qualified or faulty, and identify the type of the fault parameter when the fault occurs.

[0085] In step 103, the fault classification recognition result is input into a performance judgment network to obtain a performance judgment result corresponding to the assembly parameter.

[0086] It should be noted that the embodiment of the present application comprehensively classifies the parameters and the fault classification recognition model, inputs the abnormal assembly parameter into the performance judgment network, analyzes the corresponding performance judgment result, and intelligently locates the fault problem.

[0087] In step 104, when the performance judgment result includes a fault appearance, the abnormal probability of the assembly parameter is determined according to the mapping relationship between the assembly parameter and the fault appearance of the aerospace product.

[0088] Here, the performance judgment result includes different fault appearances corresponding to performance qualification and performance disqualification. By determining the abnormal probability of each assembly parameter, the fault cause can be accurately located.

[0089] In step 105, the optimization interval of the assembly parameter with the maximum abnormal probability is determined according to the application debugging data of the aerospace product.

[0090] It should be noted that the application debugging data of the embodiment of the present application is the debugging data of multiple groups of missiles of multiple batches of aerospace products, and can also be applied to other products. The application debugging data of different products can be different.

[0091] In step 106, the aerospace product is debugged according to the optimization interval.

[0092] Here, according to the optimization interval, the deviation of the assembly parameter of the aerospace product is determined, and the aerospace product is further debugged, which shortens the time of manually checking the fault, saves the computing resources, quickly achieves the optimization effect of the assembly parameter, and continuously optimizes the quality performance index of the aerospace product.

[0093] The embodiment of the application classifies the adjustment parameters of the aerospace product based on a pre-constructed parameter classifier to obtain a parameter classification result; inputs the adjustment parameter with the abnormal parameter classification result into a pre-constructed fault classification and identification model to obtain a fault classification and identification result; inputs the fault classification and identification result into a performance judgment network to obtain a performance judgment result corresponding to the adjustment parameter; when the performance judgment result includes a fault manifestation, determines an abnormal probability of the adjustment parameter according to a mapping relationship between the adjustment parameter and the fault manifestation of the aerospace product; determines an optimization interval of the adjustment parameter with the maximum abnormal probability according to application debugging data of the aerospace product; and debugs the aerospace product according to the optimization interval, so as to realize the qualified judgment of the aerospace product and the intelligent positioning of the fault problem, and continuously debug the adjustment parameter to more quickly and better meet the performance index requirement and realize the sustainable optimization of the adjustment parameter.

[0094] Optionally, the method further comprises:

[0095] According to the application debugging data, the optimization interval of the adjustment parameter with the normal parameter classification result is determined.

[0096] It should be noted that the optimization interval of the optimal adjustment parameter is determined, so that the qualified judgment of the aerospace product is more quickly and better to meet the performance index requirement.

[0097] Optionally, the method further comprises:

[0098] When the performance judgment result is performance qualified, the optimization interval of the adjustment parameter with the performance qualified performance judgment result is determined according to the application debugging data.

[0099] Here, even if the performance judgment result is performance qualified, the performance index does not necessarily achieve the optimal effect, the optimal optimization interval of the adjustment parameter is determined according to the application debugging data, so as to perform the optimization debugging.

[0100] Optionally, the method further comprises:

[0101] The normal adjustment parameters of the aerospace product of the past batches of the batch production type are expanded to obtain an expanded data set;

[0102] The expanded data set is used to train the SVM (Support Vector Machine) classifier to obtain the adjustment parameter classifier.

[0103] It should be noted that the test data set of the aerospace product can also be used to train the SVM classifier, so that the SVM classifier has the ability to identify the normal adjustment parameter and the abnormal adjustment parameter.

[0104] Optionally, the normal adjustment parameters of the aerospace product of the past batches of the batch production type are expanded to obtain an expanded data set, comprising:

[0105] Based on CycleGAN (Recurrent Generative Adversarial Network), normal assembly and adjustment parameters are expanded to obtain abnormal assembly and adjustment parameters;

[0106] The expanded dataset is obtained based on normal and abnormal assembly parameters.

[0107] It should be noted that the embodiments of the present invention are achieved through, as follows: Figure 2 The CycleGAN shown completes data augmentation, i.e., from X to Y. Specifically, X represents normal setup parameters and Y represents abnormal setup parameters.

[0108] Traditional GANs (Generative Adversarial Networks) belong to a one-way generation mode, that is, the input is the X domain and the output is the Y domain, and the direction is X→Y. The forward generation network X→Y→X in CycleGAN can be seen as an extension of the one-way generation mode, which restores the generated Y domain to the X domain through the reverse generation network. Conversely, the reverse generation network Y→X→Y first transforms the sample input Y domain to the X domain, and then restores it to the Y domain.

[0109] exist Figure 2 CycleGAN, as described above, is accomplished through the collaboration of two GAN networks: the forward generator network consists of generators G and F, and a discriminator D. Y Composition, discriminator D Y Used to determine whether the generated data belongs to the Y domain; the reverse generation network consists of generators F and G and a discriminator D. X Composition. The goal of generators G and F is to generate data that the discriminator of the other's data domain will recognize as real data.

[0110] The CycleGAN includes two loss functions: adversarial loss and cycle consistency loss.

[0111] Specifically, the explanation of countermeasures against losses is as follows:

[0112] CycleGAN uses least squares loss instead of cross-entropy loss in its adversarial loss, and also designs a cycle consistency loss to complement the bidirectional generation process.

[0113] like Figure 3 As shown, the generator G strives to make the generated data G(x) resemble the data in the Y domain as much as possible, while the discriminator D... Y This involves trying to distinguish between the data G(x) and the actual data y as much as possible. Assume the data distribution follows this pattern:

[0114] x~Pdata(x),x i ∈X;

[0115] y~Pdata(y),y i ∈Y;

[0116] For generator G and discriminator DY The adversarial loss function of G, D and (X, Y) is as follows:

[0117] L GAN (G, D Y , X, Y) = E y～Pdata(y) [log D Y (y)] + E x～Pdata(x) [log (1-D Y G(x))];

[0118] Similarly, the adversarial loss function of the generator F and the discriminator D X is as follows:

[0119] L GAN (F, D X , Y, X) = E x～Pdata(x) [log D X (x)] + E y～Pdata(y) [log (1-D X F(y))];

[0120] The cross-entropy loss reflects the difference between the real data distribution and the generated data distribution by using the size of the entropy value. If the cross-entropy loss is smaller, it indicates that the two distributions are closer, and the generation effect is better. Cross-entropy can effectively classify true and false samples, but it cannot help the positive samples far from the decision boundary to continue optimization, resulting in problems such as diffusion of the generated network in gradient update. Therefore, the embodiment of the present application adopts the least square loss. The loss will punish the false samples deviating from the decision boundary, and then provide the direction for gradient descent. As Figure 3 shown in the CycleGAN, a, b and c take values of 1, 0 and 1 respectively, and the standard formula of the least square loss is as follows:

[0121]

[0122]

[0123] Specifically, the cycle consistency loss is explained as follows:

[0124] Only the cycle network cannot guarantee that the input domain is close to the target domain. Because the generator and the discriminator network need to be trained independently, if the generator G does not learn the content related to the X domain at all, but directly generates data from the Y domain and outputs to the discriminator D for discrimination, it is enough to deceive the discriminator. Such a learning process is obviously invalid. In order to cooperate with the design of the bidirectional network, the CycleGAN uses a consistency loss. Cycle consistency refers to mapping an element from one domain to another domain, and then mapping it to another domain, which should produce a sample close to the original element. As Figure 3As shown, assuming that the input X domain data x is finally recovered as x' after being converted by the generator G and F, in order to ensure that the input data can be as consistent as possible after being processed by the two networks twice, the two are constrained, and the specific expression is x→G(x)→F(G(x))≈x. For the reverse generation process, it is ensured that the output y' approaches the input y, and the specific expression is y→F(y)→G(F(y))≈y, so the cycle consistency loss function is as follows:

[0125] L cyc (G,F)=E x～Pdata(x) p||F(G(x))-x||1]+E y～Pdata(y) [||G(F(y))-y||2]。

[0126] Optionally, further comprising:

[0127] Based on the first deep neural network model, the performance indicators and fault data of the past batches of batch production models of the aerospace product are pre-trained to obtain a pre-trained model;

[0128] Based on the association rule mining algorithm, a mapping relationship between the adjustment parameters and fault manifestations of the aerospace product is obtained;

[0129] According to the mapping relationship and the pre-trained model, a fault classification and identification model is obtained.

[0130] First, as shown in Figure 4 , the first deep neural network model can be a TextRNN deep neural network model, and the process is embedding layer (embedding layer)→BiLSTM (bidirectional long short-term memory network)→concat layer (concat layer)→FC layer (full connection layer)→softmax layer (softmax activation function layer, as the output layer).

[0131] Specifically, the hidden state of the forward LSTM (long short-term memory network) at the last time step is taken, then spliced, and then a multi-classification is performed through a softmax layer; or the hidden state of the forward LSTM at each time step is taken, the two hidden states at each time step are spliced, then the average of the spliced hidden states at all time steps is taken, and then a multi-classification is performed through a softmax layer.

[0132] It should be noted that the above structure can also add parameter regularization methods, such as dropout or L2 regularization, and BatchNormalization (batch normalization) can also be added to prevent overfitting and accelerate model training. After pre-training, a classification model, i.e. a pre-trained model, is output, which preliminarily classifies the types of fault data.

[0133] Then, the association rule mining algorithm can be the Apriori algorithm, and the Apriori algorithm is used for learning the association rules of the adjustment parameters and the fault manifestations of the aerospace product. Here, it is assumed that there are three adjustment parameter combinations A, B and C in total, and two fault manifestations M and N, so the support is:

[0134]

[0135]

[0136]

[0137]

[0138]

[0139]

[0140] For example, support(A, M) means that the proportion of the transaction containing the adjustment parameter A and the fault manifestation M in all transactions.

[0141] Further, the confidence is:

[0142] confidence(M←A)=P(M|A)=P(AM) / P(A);

[0143] confidence(M←B)=P(M|B)=P(BM) / P(B);

[0144] confidence(M←C)=P(M|C)=P(CM) / P(C);

[0145] confidence(N←A)=P(N|A)=P(AN) / P(A);

[0146] confidence(N←B)=P(N|B)=P(BN) / P(B);

[0147] confidence(N←C)=P(N|C)=P(CN) / P(C);

[0148] For example, confidence(M←A) means the proportion of the transaction containing the fault manifestation M in the transaction containing the adjustment parameter A.

[0149] Further, the lift is:

[0150] lift(M←A)=confidence(M←A) / P(M);

[0151] lift(M←B) = confidence(M←B) / P(M) ;

[0152] lift(M←C) = confidence(M←C) / P(M) ;

[0153] lift(N←A) = confidence(N←A) / P(N) ;

[0154] lift(N←B) = confidence(N←B) / P(N) ;

[0155] lift(N←C) = confidence(N←C) / P(N) ;

[0156] For example, lift(M←A) means the proportion of the transactions containing both the adjustment parameter A and the failure manifestation M to the proportion of the transactions containing the failure manifestation.

[0157] The adjustment parameters and the failure manifestations of the space product and the support threshold are input into the association rule mining algorithm, so as to output the maximum frequent K-item set, that is, the association rule between the adjustment parameters and the failure manifestations of the space product.

[0158] Specifically, the steps of obtaining the output frequent K-item set are as follows:

[0159] Step one: scan the entire data set to obtain all the data that has appeared, as a candidate frequent 1-item set. k = 1, and the frequent 0-item set is an empty set.

[0160] Step two: mine the frequent k-item set.

[0161] Scan the data to calculate the support of the candidate frequent k-item set.

[0162] Remove the data set with a support lower than the threshold in the candidate frequent k-item set to obtain the frequent k-item set. If the obtained frequent k-item set is empty, directly return the set of frequent k-1-item set as the result. If the obtained frequent k-item set has only one item, directly return the set of frequent k-item set as the result.

[0163] Based on the frequent k-item set, connect to generate a candidate frequent k+1-item set.

[0164] Step three: let k = k + 1, and re-execute step two.

[0165] Finally, input the association analysis rule obtained by the association rule mining algorithm into the pre-trained model based on the first deep neural network model, so as to obtain a fault classification and identification model, which can effectively associate the adjustment parameters and the failure manifestations of the space product.

[0166] Optionally, further comprising:

[0167] According to the abnormal assembly and adjustment parameters of the aerospace product of the past batches of the batch production model, the fault classification and identification model is fine-tuned and trained.

[0168] Here, the abnormal assembly and adjustment parameters of the aerospace product are used to fine-tune and train the fault classification and identification model, so that the fault classification and identification result of the fault classification and identification model is more accurate and efficient.

[0169] Optionally, in step 103, the fault classification and identification result is input into the performance judgment network to obtain a performance judgment result corresponding to the assembly and adjustment parameter, including:

[0170] Through the performance judgment network, a plurality of occurrence probabilities of different performance judgment results corresponding to the fault classification and identification result are obtained;

[0171] The plurality of occurrence probabilities are sorted to obtain a performance judgment result with the largest occurrence probability.

[0172] It should be noted that the performance judgment network can be a Bayesian network, which judges the specific performance judgment result corresponding to the fault classification and identification result, including fault manifestations A, B and C corresponding to performance qualified and unqualified.

[0173] By comparing the occurrence probabilities of different performance judgment results, the performance judgment result with the largest occurrence probability is selected as the final judgment result, so as to judge whether the aerospace product is performance qualified and the specific fault manifestation type corresponding to performance unqualified.

[0174] P(result=qualified)*P(param1|qualified)*P(param2|qualified)*P(param3|qualified)…=P(qualified);

[0175] P(result=faultA)*P(param1|faultA)*P(param2|faultA)*P(param3|faultA)…=P(faultA);

[0176] P(result=faultB)*P(param1|faultB)*P(param2|faultB)*P(param3|faultB)…=P(faultB);

[0177] P(result=faultC)*P(param1|faultC)*P(param2|faultC)*P(param3|qualified)…=P(faultC)。

[0178] Optionally, in step 104, when the performance judgment result includes a fault manifestation, the abnormal probability of the assembly and adjustment parameter is determined according to the mapping relationship between the assembly and adjustment parameter and the fault manifestation of the aerospace product established in advance, including:

[0179] Based on the Bayesian algorithm and mapping relationship, calculate the abnormal probability of each assembly parameter corresponding to different fault manifestations.

[0180] It should be noted that if the performance assessment result of an aerospace product is unqualified, i.e., a faulty product, then based on the identified fault symptoms (fault type), a Bayesian algorithm and mapping relationship are used to calculate the anomaly probability of each assembly and adjustment parameter corresponding to different fault symptoms. For example, when fault symptom A is determined, a Bayesian algorithm based on conditional probability is used to calculate the maximum probability. This calculates the probability of certain assembly and adjustment parameters of the system becoming abnormal given that fault symptom A has occurred. The calculated anomaly probabilities are then sorted in descending order. The performance index corresponding to the assembly and adjustment parameter with the highest anomaly probability has the greatest impact on the occurrence of fault symptom A, thus quickly locating the assembly and adjustment parameter corresponding to the highest anomaly probability.

[0181] For example, under the condition that fault symptom A occurs, the abnormal probability of assembly parameter 1 is as follows:

[0182] P(parameter1|faultA)=P(parameter1,faultA) / P(faultA)=P(faultA|parameter1)*P(parameter1) / P(faultA).

[0183] Optionally, in step 105, based on the application and debugging data of the aerospace products, the optimization range of the assembly and debugging parameters with the highest probability of anomalies is determined, including:

[0184] The application debugging data is input into a pre-built second deep neural network model to determine the optimization range of the assembly parameter with the highest anomaly probability;

[0185] The second deep neural network model was trained using application debugging success data and application debugging failure data.

[0186] It should be noted that this second deep neural network model can be the same as or different from the first deep neural network model. Multiple batches of application debugging data are input into this second deep neural network model. For example, missile firing data is used, including successful and unsuccessful firing data. This allows for the analysis and mining of the overall performance indicators of the aerospace product, forming an envelope range of optimized assembly and adjustment parameters. Based on this envelope range, the adjustable range of the assembly and adjustment parameters can be intuitively reflected, enabling users to continuously optimize and debug the aerospace product in conjunction with actual on-site handling conditions.

[0187] like Figure 5 As shown, the application process of the method for continuously optimizing the assembly and adjustment parameters of aerospace products according to an embodiment of the present invention is described as follows:

[0188] Pre-training a pre-training model based on a TextRNN deep neural network model, according to performance indicators and fault data of past batches of batch production models of the aerospace product;

[0189] Obtaining test data, which is the adjustment parameters of past batches of batch production models of the aerospace product;

[0190] Expanding the adjustment parameters to obtain an expanded data set;

[0191] Training an SVM classifier using the expanded data set to obtain a parameter classification result, the parameter classification result including normal adjustment parameters and abnormal adjustment parameters;

[0192] Obtaining a mapping relationship between the adjustment parameters and the fault manifestations of the aerospace product based on an association rule mining algorithm;

[0193] Obtaining a fault classification and identification model based on the mapping relationship and the pre-training model;

[0194] Inputting the abnormal adjustment parameters into the fault classification and identification model to obtain a fault classification and identification result;

[0195] Inputting the fault classification and identification result and the mapping relationship into a Bayesian network to obtain a performance judgment result corresponding to the abnormal adjustment parameters;

[0196] According to the Bayesian algorithm and the mapping relationship, calculating the abnormal probability of each adjustment parameter corresponding to different fault manifestations, and obtaining an adjustment parameter with the maximum abnormal probability;

[0197] Training a second deep neural network model using missile shooting success data and missile shooting fault data, and fine-tuning the model using on-site actual processing data to obtain a trained neural network;

[0198] According to the trained neural network, outputting an optimization interval corresponding to the adjustment parameter with the maximum abnormal probability;

[0199] According to the trained neural network, outputting an optimization interval of the adjustment parameter with a normal parameter classification result and an optimization interval of the adjustment parameter with a performance qualified performance judgment result;

[0200] According to the above optimization intervals, continuously optimizing and debugging the aerospace product.

[0201] In summary, the method for realizing sustainable optimization of the adjustment parameters of the aerospace product according to the embodiment of the present application analyzes and mines the overall performance indicators of the aerospace products of the batch production type and past batches by combining the application debugging data, pre-trains the performance indicators and fault data of the batch production type and past batches by using the deep neural network, fine-tunes the deep network by combining the actual processing results on site, enables the deep network to have the fault classification and identification function, establishes the mapping relationship between the overall adjustment parameters and the fault manifestations, realizes the intelligent determination of the performance of the aerospace product and the intelligent positioning of the fault problems, thereby optimizing the quality and performance indicators of the aerospace product, shortens the artificial troubleshooting time, and saves the computing resources.

[0202] As shown in Figure 6 The embodiment of the present application also provides a device for realizing sustainable optimization of adjustment parameters of aerospace products, which comprises:

[0203] The classification module 601 is configured to classify the adjustment parameters of the aerospace product based on the pre-constructed parameter classifier, and obtain a parameter classification result.

[0204] The input module 602 is configured to input the adjustment parameter with the parameter classification result being abnormal into the pre-established fault classification and identification model, and obtain a fault classification and identification result.

[0205] The judgment module 603 is configured to input the fault classification and identification result into the performance judgment network, and obtain a performance judgment result corresponding to the adjustment parameter.

[0206] The first determination module 604 is configured to determine the abnormal probability of the adjustment parameter according to the mapping relationship between the adjustment parameters and the fault manifestations of the aerospace product when the performance judgment result comprises the fault manifestations.

[0207] The second determination module 605 is configured to determine the optimization interval of the adjustment parameter with the maximum abnormal probability according to the application debugging data of the aerospace product.

[0208] The debugging module 606 is configured to debug the aerospace product according to the optimization interval.

[0209] The embodiment of the application classifies the adjustment parameters of the aerospace product based on a pre-constructed parameter classifier, obtains a parameter classification result, inputs the adjustment parameters with abnormal parameter classification results into a pre-constructed fault classification and identification model, obtains a fault classification and identification result, inputs the fault classification and identification result into a performance judgment network, obtains a performance judgment result corresponding to the adjustment parameters, determines an abnormal probability of the adjustment parameters according to a mapping relationship between the adjustment parameters and fault manifestations of the aerospace product when the performance judgment result includes the fault manifestations, determines an optimization interval of the adjustment parameters with the maximum abnormal probability according to application debugging data of the aerospace product, and debugs the aerospace product according to the optimization interval, so as to realize qualified judgment of the aerospace product and intelligent positioning of fault problems, and continuously debug the adjustment parameters to more quickly and better meet the performance index requirements and realize sustainable optimization of the adjustment parameters.

[0210] Optionally, the method further comprises:

[0211] The third determining module is configured to determine an optimization interval of the adjustment parameters with the normal parameter classification result according to the application debugging data.

[0212] Optionally, the method further comprises:

[0213] The fourth determining module is configured to determine an optimization interval of the adjustment parameters with the performance-qualified performance judgment result according to the application debugging data when the performance judgment result is performance-qualified.

[0214] Optionally, the method further comprises:

[0215] The first obtaining module is configured to expand the normal adjustment parameters of the aerospace products of the past batches of the batch-production type to obtain an expanded data set.

[0216] The first training module is configured to train the support vector machine classifier by using the expanded data set to obtain the adjustment parameter classifier.

[0217] Optionally, the first obtaining module is specifically configured to:

[0218] The normal adjustment parameters are expanded based on a recurrent generative adversarial network to obtain abnormal adjustment parameters.

[0219] The expanded data set is obtained according to the normal adjustment parameters and the abnormal adjustment parameters.

[0220] Optionally, the method further comprises:

[0221] The second training module is configured to pre-train the performance index and fault data of the aerospace products of the past batches of the batch-production type based on the first deep neural network model to obtain a pre-training model.

[0222] The first obtaining module is configured to obtain a mapping relationship between the adjustment parameters and the fault manifestations of the aerospace product based on an association rule mining algorithm.

[0223] The second obtaining module is configured to obtain the fault classification and identification model according to the mapping relationship and the pre-trained model.

[0224] Optionally, the device further comprises:

[0225] The third training module is configured to fine-tune the fault classification and identification model according to the abnormal adjustment parameters of the aerospace product of the past batches of the batch production model.

[0226] Optionally, the judging module 603 is specifically configured to:

[0227] obtain a plurality of occurrence probabilities of different performance judgment results corresponding to the fault classification and identification result through the performance judgment network;

[0228] sort the plurality of occurrence probabilities to obtain a performance judgment result with the largest occurrence probability.

[0229] Optionally, the first determining module 604 is specifically configured to:

[0230] calculate the abnormal probability of each adjustment parameter corresponding to different fault manifestations according to the Bayesian algorithm and the mapping relationship.

[0231] Optionally, the second determining module 605 is specifically configured to:

[0232] input the application debugging data into the pre-constructed second deep neural network model to determine the optimization interval of the adjustment parameter with the largest abnormal probability;

[0233] The second deep neural network model is obtained by training the application debugging success data and the application debugging fault data.

[0234] It should be noted that the device for realizing sustainable optimization of adjustment parameters of aerospace products provided by the embodiments of the present application is a device capable of executing the above-mentioned method for realizing sustainable optimization of adjustment parameters of aerospace products, and all the embodiments of the above-mentioned method for realizing sustainable optimization of adjustment parameters of aerospace products are applicable to the device, and the same or similar technical effects can be achieved.

[0235] As shown in Figure 7 The embodiments of the present application also provide an electronic device, which comprises a processor, a memory and a program stored in the memory and executable on the processor, and the program is executed by the processor to realize the steps of the above-mentioned method for realizing sustainable optimization of adjustment parameters of aerospace products.

[0236] Optionally, the electronic device further comprises a transceiver 703 for receiving and transmitting data under the control of the processor 701.

[0237] wherein, in Figure 7 The bus architecture can include any number of interconnected buses and bridges, specifically the various circuitry of the processor 701 and the memory 702 represented by one or more processors and memory. The bus architecture can also link various other circuitry such as peripheral devices, voltage regulators, and power management circuitry, all of which are well known in the art and thus, not further described herein. The bus interface provides the user interface 704. The transceiver 703 can be multiple elements, i.e., including a transmitter and a receiver, providing the means for communicating with various other devices on a transmission medium. The user interface 704 can also be an interface capable of external or internal attachment to the device as needed, including but not limited to a keypad, a display, a speaker, a microphone, a joystick, etc. The processor 701 is responsible for managing the bus architecture and general processing, and the memory 702 can store data used by the processor 701 in performing operations.

[0238] The embodiment of the present application also provides a readable storage medium, and the readable storage medium stores a program, and the program is executed by a processor to implement the steps of the method for realizing sustainable optimization of the adjustment parameters of the aerospace product.

[0239] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0240] The foregoing exemplary embodiments are described with reference made to the drawings which are provided for the purpose of explanation and illustration. They are not intended to limit the scope of the invention. Rather, these exemplary embodiments are described in order to enable others skilled in the art to embody the application. As will be understood by those familiar with the art, the application can be embodied in many different forms and should not be limited to the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided so that this disclosure will be complete and fully convey the scope of the application to those skilled in the art. In the drawings, the size and relative sizes of components can be exaggerated for clarity. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the singular articles "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Unless otherwise indicated, a value range includes the upper and lower limits of the range and any sub-ranges therebetween.

[0241] The above description is that of the preferred embodiments of the application. Various modifications and changes can be made thereto without departing from the spirit and scope of the application, which is to be understood. The scope of the application is indicated only by the subjoined claims.

Claims

1. A method for realizing sustainable optimization of the adjustment parameters of a space product, characterized in that, The method comprises the following steps: classifying the adjustment parameters of the aerospace product based on a pre-constructed parameter classifier to obtain a parameter classification result; inputting the adjustment parameters with the abnormal parameter classification result into a pre-established fault classification and identification model to obtain a fault classification and identification result; inputting the fault classification and identification result into a performance judgment network to obtain a performance judgment result corresponding to the adjustment parameters; when the performance judgment result includes a fault manifestation, determining an abnormal probability of the adjustment parameters according to a mapping relationship between the adjustment parameters and the fault manifestation of the aerospace product; determining an optimization interval of the adjustment parameters with the maximum abnormal probability according to application debugging data of the aerospace product; and debugging the aerospace product according to the optimization interval.

2. The method for realizing sustainable optimization of the adjustment parameters of a space product according to claim 1, characterized in that, The method further comprises the following steps: determining an optimization interval of the adjustment parameters with the normal parameter classification result according to the application debugging data.

3. The method for realizing sustainable optimization of the assembly parameters of a space product according to claim 1, characterized in that, The method further comprises the following steps: when the performance judgment result is qualified, determining an optimization interval of the adjustment parameters with the qualified performance judgment result according to the application debugging data.

4. The method for realizing sustainable optimization of space product alignment parameters according to claim 1, characterized in that, The method further comprises the following steps: expanding the normal adjustment parameters of the aerospace product of the past batches of the batch production type to obtain an expanded data set; training the support vector machine classifier by using the expanded data set to obtain the adjustment parameter classifier.

5. The method for realizing sustainable optimization of space product alignment parameters according to claim 4, characterized in that, The method further comprises the following steps: expanding the normal adjustment parameters of the aerospace product of the past batches of the batch production type to obtain an expanded data set, comprising: expanding the normal adjustment parameters based on a cycle generation adversarial network to obtain abnormal adjustment parameters; 6. The method for realizing sustainable optimization of space product alignment parameters according to claim 1, characterized in that, obtaining the expanded data set according to the normal adjustment parameters and the abnormal adjustment parameters. The method further comprises the following steps: pre-training the performance indicators and fault data of the aerospace product of the past batches of the batch production type based on a first deep neural network model to obtain a pre-training model; obtaining a mapping relationship between the adjustment parameters and the fault manifestations of the aerospace product based on an association rule mining algorithm; 7. The method for realizing sustainable optimization of space product alignment parameters according to claim 6, characterized in that, obtaining the fault classification and identification model according to the mapping relationship and the pre-training model. The method further comprises the following steps:

8. The method for realizing sustainable optimization of space product alignment parameters according to claim 1, characterized in that, fine-tuning the fault classification and identification model according to the abnormal adjustment parameters of the aerospace product of the past batches of the batch production type. Inputting the fault classification and identification result into a performance judgment network to obtain a performance judgment result corresponding to the adjustment parameters, comprising: obtaining multiple occurrence probabilities of different performance judgment results corresponding to the fault classification and identification result through the performance judgment network; 9. The method for realizing sustainable optimization of space product alignment parameters according to claim 1, characterized in that, sorting the multiple occurrence probabilities to obtain the performance judgment result with the maximum occurrence probability. When the performance judgment result includes a fault manifestation, determining an abnormal probability of the adjustment parameters according to a pre-established mapping relationship between the adjustment parameters and the fault manifestation of the aerospace product, comprising:

10. The method for realizing sustainable optimization of space product alignment parameters according to claim 1, characterized in that, calculating the abnormal probability of each adjustment parameter corresponding to different fault manifestations according to a Bayesian algorithm and the mapping relationship. Determining an optimization interval of the adjustment parameters with the maximum abnormal probability according to application debugging data of the aerospace product, comprising: inputting the application debugging data into a pre-constructed second deep neural network model to determine the optimization interval of the adjustment parameters with the maximum abnormal probability. The second deep neural network model is obtained by training according to application debugging success data and application debugging failure data.

11. An apparatus for realizing sustainable optimization of the adjustment parameters of a space product, characterized in that it comprises: The method comprises the steps of: The classification module is configured to classify the adjustment parameters of the aerospace product based on a pre-constructed parameter classifier to obtain a parameter classification result. The input module is configured to input the adjustment parameter with the abnormal parameter classification result into a pre-established fault classification identification model to obtain a fault classification identification result. The judgment module is configured to input the fault classification identification result into a performance judgment network to obtain a performance judgment result corresponding to the adjustment parameter. The first determination module is configured to determine an abnormal probability of the adjustment parameter according to a mapping relationship between the adjustment parameter and a fault manifestation of the aerospace product when the performance judgment result includes the fault manifestation. The second determination module is configured to determine an optimization interval of the adjustment parameter with the maximum abnormal probability according to application debugging data of the aerospace product. The debugging module is configured to debug the aerospace product according to the optimization interval.

12. An electronic device, comprising: The processor, the memory, and a program stored in the memory and executable on the processor are provided. The program is stored in the readable storage medium and executable on the processor to implement the steps of the method for realizing sustainable optimization of adjustment parameters of an aerospace product.

13. A readable storage medium, characterized by, ​

Images