A neural network-based ship weld quality sampling system and method

Abstract

The application discloses a kind of ship weld quality sampling inspection system and method based on neural network, and the sampling inspection system includes weld information acquisition module, welder qualification evaluation module, weld sampling inspection grouping module, random sampling inspection module, sampling inspection result feedback module;Sampling inspection method includes S1: information acquisition;S2: determine weld quality influence factor and weight initial value;S3: create weld sampling inspection grouping model;S4: input new weld data, group weld, for grouped weld, according to sampling inspection plan, proportionally carry out random sampling inspection;S5: update grouping module.The input weld data is divided into three categories by weld sampling inspection grouping module, respectively, one class is larger possible problem, 100% sampling inspection;Second class is possible problem, 50% sampling inspection;Three classes are non-problem, 10% sampling inspection;The proportion of sampling inspection is determined by reasonable grouping, and the efficiency of ship weld quality inspection is improved.

Description

TECHNICAL FIELD

[0001] The application relates to a quality sampling inspection method and system, in particular to a ship weld quality sampling inspection method and system based on a neural network. BACKGROUND

[0002] The shipbuilding industry is a labor-intensive heavy industry. In the whole shipbuilding process, the time for welding accounts for 30-40% of the total time, and the cost accounts for one-third of the total. The welding quality will directly affect the structural strength of the ship and the sealing property of the ship, and further affect the navigation safety of the ship. Ship welding is different from ordinary metal welding. Ship welding has the characteristics of long weld, high welding difficulty, complex welding process and high technical requirements for welders. In order to ensure the normal completion of the shipbuilding project, it is necessary to ensure the welding quality of the weld through inspection. Considering that a large number of welds will be generated during the shipbuilding process, the workload of inspecting all the welds is large, the efficiency is low, and a certain degree of waste of manpower and material resources will be caused. Therefore, in order to reduce the workload of inspection and improve the efficiency of inspection, it is necessary to develop a system and method for sampling inspection of ship weld quality. SUMMARY

[0003] In order to solve the problem of large workload and low efficiency of welding quality inspection in the prior art, the present application provides a ship weld quality sampling inspection system and method based on a neural network, and the specific scheme is as follows:

[0004] A ship weld quality sampling inspection system based on a neural network comprises a weld information acquisition module, a welder qualification evaluation module, a weld sampling grouping module, a random sampling module and a sampling result feedback module.

[0005] The weld information acquisition module is used for acquiring weld information and establishing a standard weld information database.

[0006] The welder qualification evaluation module is used for establishing a welder information database.

[0007] The weld sampling grouping module uses a neural network to construct a weld sampling grouping model.

[0008] The random sampling module completes sampling based on the weld sampling grouping module.

[0009] The sampling result feedback module is used for feeding back the sampling result to the weld sampling grouping model and updating the model.

[0010] The specific steps for constructing the weld sampling grouping model are as follows:

[0011] S1: Determine the weld quality influencing factors and the initial value of the weight by using the expert scoring method.

[0012] S2: Take the weld information and welder information as input values, train the BP neural network model through the past weld inspection data of the shipyard, and obtain the weld sampling grouping model.

[0013] The standard weld information database includes weld code, inspection type, inspection length, weld length, plate thickness, material, welding material batch, inspection category, welding head joint type, welding method, and groove type.

[0014] The welder information database includes welder basic information, welder qualification certificate, and assessment level.

[0015] The weld sampling grouping model divides the input weld data into three categories, which are:

[0016] The first category is that there may be problems, and 100% sampling is required.

[0017] The second category is that there may be problems, and 50% sampling is required.

[0018] The third category is that there are no problems, and 10% sampling is required.

[0019] A neural network-based ship weld quality management method, comprising the following steps:

[0020] S1: Collect weld information, establish a standard weld information database, and establish a welder information database;

[0021] S2: Determine the weld quality influencing factors and initial weight values using expert scoring method;

[0022] S3: Create a weld sampling grouping model based on BP neural network;

[0023] S4: Input new weld data, group the welds, and randomly sample the grouped welds according to the sampling plan in proportion;

[0024] S5: Import the obtained weld sampling results into the weld detection library for maintenance, feedback the weld sampling grouping module according to the detection results, and update the grouping module.

[0025] Expert scoring method is a way to quantify qualitative problems by collecting, processing and summarizing expert opinions. It requires the entire expert group to complete the task collectively, train the evaluation criteria and scoring details uniformly, and exclude mutual influence to ensure the effectiveness of the data and the scientificity of the evaluation results. In S2, the expert scoring method is used to determine the weld quality influencing factors and initial weight values. Through the opinions of R experts, the factors affecting the weld quality are scored, and the membership degree of each index is analyzed. The total number of expert selections is R i , and the total number of valid questionnaires is n. The membership degree expression of the evaluation index is:

[0026] ri = Ri / n

[0027] According to the membership from large to small in turn, the final evaluation index is determined as the factor set {U}, which is the welder grade, the weld length, the plate thickness, the inspection category, the welding method and the welding joint type. The index number and the index meaning are shown in the following table:

[0028]

[0029]

[0030] The evaluation result of the expert is taken as the evaluation set {V}, and the weight {A} is given according to the importance, which is shown in the following table:

[0031]

[0032] The weight of each index is determined through the weight calculation formula, and the weight calculation formula is shown as follows:

[0033]

[0034] Beneficial effects:

[0035] (1) The ship weld quality sampling system and method based on the neural network provided by the application comprises a weld information acquisition module, a welder qualification evaluation module, a weld sampling grouping module, a random sampling module and a sampling result feedback module. The weld information acquisition module is used to acquire weld information and establish a standard weld information database. The welder qualification evaluation module is used to establish a welder information database. The weld sampling grouping module uses a neural network to construct a weld sampling grouping model. The random sampling module completes sampling based on the weld sampling grouping module. The sampling result feedback module is used to feed back the sampling result to the weld sampling grouping model and update the model. The weld sampling grouping module divides the input weld data into three categories, i.e. one category with a large possibility of problems, 100% sampling; a second category with a possibility of problems, 50% sampling; and a third category without problems, 10% sampling. Reasonable grouping determines the sampling ratio, which improves the efficiency of ship weld quality inspection.

[0036] (2) The ship weld quality sampling system and method based on the neural network provided by the application is based on a BP neural network. The initial weight of the BP neural network is given by the expert scoring method, which is more in line with the actual situation of ship weld quality inspection. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 A flow chart of the ship weld quality sampling system based on the neural network.

[0038] Figure 2It is a neural network-based ship weld quality sampling method flow chart.

[0039] Figure 3 It is a structure diagram of BP neural network. DETAILED DESCRIPTION

[0040] In order to deepen the understanding of the present application, the present application will be further described below in conjunction with examples and drawings, which are only used to explain the present application and do not constitute a limitation on the protection scope of the present application.

[0041] Embodiment:

[0042] Figure 1 It is a neural network-based ship weld quality sampling system flow chart, including a weld information acquisition module, a welder qualification evaluation module, a weld sampling grouping module, a random sampling module, and a sampling result feedback module.

[0043] The weld information acquisition module is used to acquire weld information and establish a standard weld information database.

[0044] The welder qualification evaluation module is used to establish a welder information database.

[0045] The weld sampling grouping module uses neural networks to build a weld sampling grouping model.

[0046] The random sampling module completes sampling based on the weld sampling grouping module.

[0047] The sampling result feedback module is used to feed back the sampling results to the weld sampling grouping model and update the model.

[0048] Figure 2 It is a neural network-based ship weld quality sampling method flow chart, including the following steps:

[0049] S1: Collecting weld information, establishing a standard weld information database, and establishing a welder information database;

[0050] S2: Using expert scoring method to determine weld quality influencing factors and weight initial value;

[0051] S3: Creating a weld sampling grouping model based on BP neural network;

[0052] S4: Inputting new weld data, grouping the welds, and randomly sampling the grouped welds according to the sampling plan;

[0053] S5: Importing the obtained weld sampling results into the weld detection library for maintenance, feeding back the weld sampling grouping module according to the detection results, and updating the grouping module.

[0054] Figure 3 is the BP neural network structure diagram, and a weld sampling grouping model is created based on the BP neural network algorithm. A large amount of data is used to train the model to obtain a model meeting the requirements. The given training set is D={(X1, Y1), (X2, Y2), …, (Xn, Yn)}, wherein XeR, YeR, indicates that the input example is composed of d attributes, and the output is an l-dimensional real value variable. n n}, wherein X n ∈R d , Y n ∈R l , indicates that the input example is composed of d attributes, and the output is an l-dimensional real value variable.

[0055] For the ith neuron, X1, X2, …, X i are inputs of the neuron, and the inputs are usually independent variables that have a key influence on the system model, W1, W2, …, W i are connection weight values that adjust the proportion of each input. Linear weighted summation is selected to obtain the Net in neuron net input

[0056]

[0057] The output adopts the activation function sigmoid function, which can transform the input from negative infinity to positive infinity into an output between 0 and 1.

[0058]

[0059] The factor indexes influencing the weld quality are scored by 50 industry experts, the number of expert scoring is counted, the membership degree of each index is analyzed, and the membership degree values are sorted. The greater the membership degree value, the more important the evaluation index in the evaluation system. Finally, the following indexes are determined as the formal evaluation indexes:

[0060]

[0061] The evaluation results of the experts are taken as the evaluation set {V}, and the weights {A} are assigned according to the importance, and the specific values are as follows:

[0062]

[0063] The number of scoring times of 50 experts for different indexes is counted, and the specific values are shown in the following table:

[0064]

[0065]

[0066] The weight of each index is determined according to the weight calculation formula, and the weight calculation formula is as follows:

[0067] ​

[0068] The welding quality of the ship is sampled and inspected:

[0069] S1: input welding data, specific parameters as follows:

[0070]

[0071] S2: the welding sampling grouping module groups the input welding data, and the welding is output by the welding sampling grouping module into three categories, which are:

[0072] The first category is that there may be problems, 100% sampling;

[0073] The second category is that there may be problems, 50% sampling;

[0074] The third category is that there is no problem, 10% sampling;

[0075] The category of the welding data in S1 after output by the welding sampling grouping module is:

[0076]

[0077] S3: random sampling according to the sampling ratio of different categories.

[0078] As a further improvement, the above only describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A neural network-based shipweld quality sampling inspection system, characterized by, The welding seam information acquisition module, the welder qualification evaluation module, the welding seam sampling grouping module, the random sampling module and the sampling result feedback module are comprised. The welding seam information acquisition module is used for collecting welding seam information and establishing a standard welding seam information database. The welding seam information acquisition module is used for collecting welding seam information and establishing a standard welding seam information database. The welding seam sampling grouping module uses a neural network to construct a welding seam sampling grouping model. The random sampling module completes sampling based on the welding seam sampling grouping module. The sampling result feedback module is used for feeding back the sampling result to the welding seam sampling grouping model and updating the model. The specific steps of constructing the welding seam sampling grouping model are as follows: S1: The expert scoring method is used to determine the welding seam quality influencing factors and the weight initial value. S2: The welding seam information and the welder information are used as input values, and the BP neural network model is trained through the past welding seam inspection data of the shipyard to obtain the welding seam sampling grouping model. The standard welding seam information database comprises welding seam codes, inspection types, inspection lengths, welding seam lengths, plate thicknesses, material qualities, welding material batches, inspection categories, welding head joint types, welding methods and groove types. The welder information database comprises welder basic information, welder qualification certificates and assessment grades.

2. A neural network-based method for shipweld quality spot-checking, characterized in that, The specific steps are as follows: S1: Collect welding seam information, establish a standard welding seam information database and establish a welder information database. S2: The expert scoring method is used to determine the welding seam quality influencing factors and the weight initial value. S3: The welding seam sampling grouping model is created based on the BP neural network. S4: New welding seam data is input to group the welding seams, and the grouped welding seams are randomly sampled according to the sampling plan in proportion. S5: The obtained welding seam sampling result is imported into the welding seam detection library for maintenance, the welding seam sampling grouping module is fed back according to the detection result, and the grouping module is updated. The step of determining the weld quality influencing factors and weight initial values in S2 is: scoring the factor indexes influencing the weld quality through the opinions of R experts, and performing membership degree analysis on each index, and the total number of expert selections is R i , the total number of effective questionnaires is n, and the membership degree expression of the evaluation index is: ri=Ri / n The final evaluation indexes are arranged in descending order of membership to determine the final evaluation indexes as the factor set {U}. The evaluation results of the experts are used as the evaluation set {V}, and the weights {A} are assigned according to the importance. The weight calculation formula is used to determine the weight of each index.

3. The neural network-based shipweld quality sampling inspection method according to claim 2, characterized in that, The welding seam sampling grouping model divides the input welding seam data into three categories, namely, one category of 100% sampling for possible problems, one category of 50% sampling for possible problems, and one category of 10% sampling for no problems.

4. The neural network-based shipweld quality sampling inspection method according to claim 2, characterized in that, The final evaluation indexes are the welder grade, the welding seam length, the plate thickness, the inspection category, the welding method and the welding joint type.

Images