An artificial intelligence-based ship welding defect identification system

Abstract

The application provides a ship welding defect identification system based on artificial intelligence, characterized by comprising a data acquisition module, a feature extraction module, a defect identification module and a defect prediction module; the data acquisition module is used for acquiring multi-dimensional data related to ship welding defects, the feature extraction module is used for extracting dimensional data with greater relevance to welding defects in the multi-dimensional data, the defect identification module is used for identifying ship welding defects, and the defect prediction module is used for predicting possible ship welding defects based on the identification result; the system can not only identify real-time ship welding defects, but also predict possible future ship welding defects, so that users can better evaluate and take corresponding response measures for ship welding defects according to real-time identification results and prediction results.

Description

An AI-based ship welding defect identification system Technical Field

[0001] This invention relates to the field of ship welding technology, and more specifically to an artificial intelligence-based ship welding defect identification system. Background Technology

[0002] Welding is a common joining technology in shipbuilding, but various defects may occur during the welding process, such as cracks, porosity, and lack of fusion. These welding defects can seriously affect the structural strength and safety of ships, so it is crucial to detect and accurately identify ship welding defects as early as possible. The development of artificial intelligence technology has provided new solutions for the identification of ship welding defects. Machine learning algorithms can learn patterns and features from large amounts of data and perform intelligent identification and classification.

[0003] A review of publicly available technical solutions reveals several examples. For instance, CN115564249A discloses a ship welding quality control system and method, including a basic database, monitoring module, functional module, and functional processing module. This solution provides full-process quality control over the welding torch status before welding, welding machine parameters and environmental parameters during welding, and post-weld flaw detection results. It overcomes problems in existing technologies such as incomplete monitoring of welding quality elements, insufficient systematic welding control, lack of control action instructions, and lack of targets for welding quality improvement. It comprehensively ensures controllable welding parameters, improves overall welding construction quality, promptly stops ineffective welding tasks, and accurately pushes welding assistance improvement suggestions. Another typical example is CN113537621A, which discloses a big data-driven method for predicting the welding quality of thin-plate ships. This method includes the following steps: using an established IoT framework to collect and transmit real-time welding data; performing key data feature optimization on the collected welding quality influencing factors to complete the initial analysis of the welding big data. The method involves dimensionality reduction of the initial feature set; and the establishment of a BP neural network prediction model optimized by an adaptive simulated annealing particle swarm optimization algorithm. Based on the dimensionality-reduced feature set, the method outputs the prediction results for the welding quality of thin plates in ships. This approach establishes an SAPSO_BP prediction model, which improves upon the shortcomings of BP neural network prediction models, such as being prone to getting trapped in local minima, slow convergence speed, and poor robustness. This enables online monitoring of the welding process of thin plates in ships and accurate prediction of welding quality, improving the accuracy of prediction results and providing a highly reliable reference for the optimization decision-making of welding processes. The above approach can only predict ship welding defects by collecting data in real time. Users can only take relevant measures after receiving the defect identification results, and cannot predict possible ship welding defects in advance. If users can obtain prediction information related to possible ship welding defects in advance, they can better assess ship welding defects and take corresponding measures, and increase the time for users to remedy welding defects. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings by proposing an artificial intelligence-based ship welding defect identification system.

[0005] The present invention adopts the following technical solution:

[0006] An artificial intelligence-based ship welding defect identification system is characterized by comprising a data acquisition module, a feature extraction module, a defect identification module, and a defect prediction module.

[0007] The data acquisition module is used to acquire multidimensional data related to ship welding defects, the feature extraction module is used to extract the dimension data that is highly correlated with welding defects from the multidimensional data, the defect identification module is used to identify ship welding defects, and the defect prediction module predicts possible ship welding defects based on the identification results.

[0008] The data acquisition module includes a real-time data acquisition module and a historical data acquisition module. The real-time data acquisition module is used to acquire multi-dimensional data during the real-time welding process, and the historical data acquisition module is used to acquire multi-dimensional data during the historical welding process and corresponding time point defect information. The real-time data acquisition module includes a data sensing unit and a data integration unit. The data sensing unit is used to collect multi-dimensional data during the real-time welding process, and the data integration unit is used to perform calibration, formatting, and time synchronization operations on the multi-dimensional data.

[0009] The feature extraction module extracts the dimension data that is highly correlated with welding defects from the multidimensional data through the following steps:

[0010] S101: Data Preparation: Obtain multi-dimensional data from historical welding processes and corresponding time point defect information;

[0011] S102: Constructing a rough set: Based on the data obtained in the previous step, a rough set model is established. The rough set model can be represented as (U, C∪D), where U is the universe of discourse, i.e., the dataset containing multi-dimensional data from the historical welding process; C is the condition feature set, containing all dimensions in the universe of discourse U; and D is the decision feature set, containing various defect categories and their values. The defect category value is the dimension corresponding to that defect category.

[0012] S103: Extract important dimensions: Calculate the approximate classification quality of each dimension, which is calculated by the ratio of the lower approximation of each dimension to the universe of discourse; and compare these approximate classification qualities with a set extraction threshold, and take the dimensions with approximate classification qualities greater than the threshold as the dimensions with greater correlation to welding defects.

[0013] Furthermore, the defect identification module establishes a BP neural network model and uses multidimensional data with high correlation to welding defects during historical welding processes and corresponding time point defect information as a training set to train the neural network model. The trained neural network model is then used for ship welding defect identification. The input of the neural network model is multidimensional data with high correlation to welding defects, and the output is the probability of occurrence of various welding defects, namely welding cracking, welding porosity, welding slag inclusion, welding misalignment, insufficient welding transition zone, and welding deformation.

[0014] Furthermore, the specific method by which the defect prediction module predicts possible ship welding defects includes the following steps:

[0015] S301: Extract 10 sets of neural network model recognition results containing temporal sequence from the current time as the cutoff time; each set of recognition results includes the probability of welding cracks, welding holes, welding slag inclusions, welding misalignment, insufficient welding transition zone and welding deformation during ship welding;

[0016] S302: Calculate the average probability value of each welding defect across 10 sets of identification results;

[0017] S303: Extract the identification results corresponding to the two welding defects with the highest average probability values ​​in step S302;

[0018] S304: Calculate the warning value based on the data extracted in step S303.

[0019]

[0020] in, Let a be the average probability value corresponding to the welding defect with the highest average probability value among all welding defects. max The weld defect with the highest average probability value among all weld defects is the maximum value among its 10 identification results, a. min Δa is the minimum value of the welding defect with the highest average probability among all welding defects in its 10 sets of identification results; Δa is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results for the welding defect with the highest average probability among all welding defects; b max b is the maximum value of the welding defect with the second largest average probability among all welding defects in its 10 sets of identification results. min Δb is the minimum value of the welding defect with the second largest average probability value among all welding defects in its 10 sets of identification results, and Δb is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results of the welding defect with the second largest average probability value among all welding defects.

[0021] S305: Compare the warning value with the defect threshold. When the warning value is greater than the defect threshold, output the welding defect with the highest average probability value among all welding defects in step S304 as the predicted possible ship welding defect.

[0022] The beneficial effects achieved by this invention are:

[0023] The feature extraction module of this system is based on rough set theory. It measures the importance of each dimension by calculating the approximate classification quality of each dimension. This can filter out dimensions that have no significant impact on defect judgment and extract dimensions with greater discriminative and predictive capabilities. This reduces computational and storage overhead, improves the efficiency of feature selection and data analysis, and better supports subsequent tasks of identifying and predicting ship welding defects. The defect identification module identifies real-time ship welding defects, and the defect prediction module predicts potential future ship welding defects. This ensures that users can better assess ship welding defects and take appropriate countermeasures based on real-time identification and prediction results, increasing the time available for defect remediation and improving the structural strength and safety of ship welding.

[0024] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0025] Figure 1 is a schematic diagram of the overall structural framework of the present invention;

[0026] Figure 2 is a schematic diagram of the process of the feature extraction module of the present invention extracting dimensional data that are highly correlated with welding defects;

[0027] Figure 3 is a schematic diagram of the process of identifying ship welding defects by the defect identification module of the present invention;

[0028] Figure 4 is a schematic diagram of the process by which the defect prediction module of the present invention predicts welding defects in ships. Detailed Implementation

[0029] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated beforehand. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0030] Example 1.

[0031] This embodiment provides an artificial intelligence-based ship welding defect identification system, characterized in that it includes a data acquisition module, a feature extraction module, a defect identification module, and a defect prediction module;

[0032] The data acquisition module is used to acquire multidimensional data related to ship welding defects, the feature extraction module is used to extract the dimension data that is highly correlated with welding defects from the multidimensional data, the defect identification module is used to identify ship welding defects, and the defect prediction module predicts possible ship welding defects based on the identification results.

[0033] The data acquisition module includes a real-time data acquisition module and a historical data acquisition module. The real-time data acquisition module is used to acquire multi-dimensional data during the real-time welding process, and the historical data acquisition module is used to acquire multi-dimensional data during the historical welding process and corresponding time point defect information. The real-time data acquisition module includes a data sensing unit and a data integration unit. The data sensing unit is used to collect multi-dimensional data during the real-time welding process, and the data integration unit is used to perform calibration, formatting, and time synchronization operations on the multi-dimensional data.

[0034] The feature extraction module extracts the dimension data that is highly correlated with welding defects from the multidimensional data through the following steps:

[0035] S101: Data Preparation: Obtain multi-dimensional data from historical welding processes and corresponding time point defect information;

[0036] S102: Constructing a rough set: Based on the data obtained in the previous step, a rough set model is established. The rough set model can be represented as (U, C∪D), where U is the universe of discourse, i.e., the dataset containing multi-dimensional data from the historical welding process; C is the condition feature set, containing all dimensions in the universe of discourse U; and D is the decision feature set, containing various defect categories and their values. The defect category value is the dimension corresponding to that defect category.

[0037] S103: Extract important dimensions: Calculate the approximate classification quality of each dimension, which is calculated by the ratio of the lower approximation of each dimension to the universe of discourse; and compare these approximate classification qualities with a set extraction threshold, and take the dimensions with approximate classification qualities greater than the threshold as the dimensions with greater correlation to welding defects.

[0038] Furthermore, the defect identification module establishes a BP neural network model and uses multidimensional data with high correlation to welding defects during historical welding processes and corresponding time point defect information as a training set to train the neural network model. The trained neural network model is then used for ship welding defect identification. The input of the neural network model is multidimensional data with high correlation to welding defects, and the output is the probability of occurrence of various welding defects, namely welding cracking, welding porosity, welding slag inclusion, welding misalignment, insufficient welding transition zone, and welding deformation.

[0039] Furthermore, the specific method by which the defect prediction module predicts possible ship welding defects includes the following steps:

[0040] S301: Extract 10 sets of neural network model recognition results containing temporal sequence from the current time as the cutoff time; each set of recognition results includes the probability of welding cracks, welding holes, welding slag inclusions, welding misalignment, insufficient welding transition zone and welding deformation during ship welding;

[0041] S302: Calculate the average probability value of each welding defect across 10 sets of identification results;

[0042] S303: Extract the identification results corresponding to the two welding defects with the highest average probability values ​​in step S302;

[0043] S304: Calculate the warning value based on the data extracted in step S303.

[0044]

[0045] in, Let a be the average probability value corresponding to the welding defect with the highest average probability value among all welding defects. max The weld defect with the highest average probability value among all weld defects is the maximum value among its 10 identification results, a. min Δa is the minimum value of the welding defect with the highest average probability among all welding defects in its 10 sets of identification results; Δa is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results for the welding defect with the highest average probability among all welding defects; b max b is the maximum value of the welding defect with the second largest average probability among all welding defects in its 10 sets of identification results. min Δb is the minimum value of the welding defect with the second largest average probability value among all welding defects in its 10 sets of identification results, and Δb is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results of the welding defect with the second largest average probability value among all welding defects.

[0046] S305: Compare the warning value with the defect threshold. When the warning value is greater than the defect threshold, output the welding defect with the highest average probability value among all welding defects in step S304 as the predicted possible ship welding defect.

[0047] Example 2.

[0048] This embodiment should be understood to include at least all the features of any of the foregoing embodiments, and to further improve upon them;

[0049] This embodiment provides an artificial intelligence-based ship welding defect identification system, characterized in that it includes a data acquisition module, a feature extraction module, a defect identification module, and a defect prediction module;

[0050] The data acquisition module is used to acquire multidimensional data related to ship welding defects, the feature extraction module is used to extract the dimension data that is highly correlated with welding defects from the multidimensional data, the defect identification module is used to identify ship welding defects, and the defect prediction module predicts possible ship welding defects based on the identification results.

[0051] The data acquisition module includes a real-time data acquisition module and a historical data acquisition module. The real-time data acquisition module is used to acquire multi-dimensional data during the real-time welding process, and the historical data acquisition module is used to acquire multi-dimensional data during the historical welding process and corresponding time point defect information. The real-time data acquisition module includes a data sensing unit and a data integration unit. The data sensing unit is used to collect multi-dimensional data during the real-time welding process, and the data integration unit is used to perform calibration, formatting, and time synchronization operations on the multi-dimensional data.

[0052] The data sensing unit includes, but is not limited to, the following sensing devices or information acquisition devices: camera device, current sensor, voltage sensor, inductance sensor, welding gas flow sensor, etc.; these sensing devices are only examples, and other types of sensing devices can also be used to acquire multidimensional data in the ship welding process.

[0053] The system also includes a communication module for transmitting data between the data acquisition module and the feature extraction module;

[0054] The feature extraction module performs the following steps to extract data from dimensions of multidimensional data that are highly correlated with welding defects:

[0055] S101: Data Preparation: Obtain multi-dimensional data from historical welding processes and corresponding time point defect information;

[0056] S102: Constructing a rough set: Based on the data obtained in the previous step, a rough set model is established. The rough set model can be represented as (U, C∪D), where U is the universe of discourse, i.e., the dataset containing multi-dimensional data from the historical welding process; C is the condition feature set, containing all dimensions in the universe of discourse U; and D is the decision feature set, containing various defect categories and their values. The defect category value is the dimension corresponding to that defect category.

[0057] S103: Extract important dimensions: Calculate the approximate classification quality of each dimension, which is calculated by the ratio of the lower approximation of each dimension to the universe of discourse; and compare these approximate classification qualities with a set extraction threshold, and take the dimensions with approximate classification qualities greater than the threshold as the dimensions with greater correlation to welding defects.

[0058] The defect identification module identifies welding defects in ships by establishing a BP neural network model; specifically, it includes the following steps:

[0059] S201: Establish a BP neural network model and determine the number of nodes in the input layer, hidden layer, and output layer of the neural network model; the number of nodes in the input layer is the number of dimensions in the multidimensional data that is highly correlated with welding defects; the number of nodes in the output layer is 7, corresponding to the welding defect types of welding cracking, welding voids, welding slag inclusions, welding misalignment, insufficient welding transition zone, and welding deformation; the number of nodes in the hidden layer, n1, is:

[0060]

[0061] Where n is the number of feature layer nodes, m is the number of output layer nodes, and k is a constant in the range [1,10], the value of which is obtained through experimental verification;

[0062] S202: Determine the activation function: Set the activation function of the hidden layer and the output layer in this neural network model to the sigmoid function;

[0063] S203: Model Training Optimization: Use the data obtained by the historical data acquisition module as the training set to train and optimize the weights and thresholds in the neural network model; Model training optimization can be achieved through existing technologies such as gradient descent, genetic algorithm or particle swarm optimization, which will not be elaborated here;

[0064] S204: The trained neural network model is used for real-time prediction. At this time, the input of the neural network model is multi-dimensional data that is highly correlated with welding defects during the real-time welding process; the output recognition result is the probability of occurrence of the defect type corresponding to the above output layer node.

[0065] The specific method by which the defect prediction module predicts possible ship welding defects satisfies the following requirements:

[0066] S301: Extract 10 sets of neural network model recognition results containing temporal sequence from the current time as the cutoff time; each set of recognition results includes the probability of welding cracks, welding holes, welding slag inclusions, welding misalignment, insufficient welding transition zone and welding deformation during ship welding;

[0067] S302: Calculate the average probability value of each welding defect across 10 sets of identification results;

[0068] S303: Extract the identification results corresponding to the two welding defects with the highest average probability values ​​in step S302;

[0069] S304: Calculate the warning value based on the data extracted in step S303.

[0070]

[0071] in, Let a be the average probability value corresponding to the welding defect with the highest average probability value among all welding defects. max The weld defect with the highest average probability value among all weld defects is the maximum value among its 10 identification results, a. min Δa is the minimum value of the welding defect with the highest average probability among all welding defects in its 10 sets of identification results; Δa is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results for the welding defect with the highest average probability among all welding defects; b max b is the maximum value of the welding defect with the second largest average probability among all welding defects in its 10 sets of identification results. min Δb is the minimum value of the welding defect with the second largest average probability value among all welding defects in its 10 sets of identification results, and Δb is the difference between the identification result closest to the current time and the second closest identification result among the 10 sets of identification results of the welding defect with the second largest average probability value among all welding defects.

[0072] S305: Compare the warning value with the defect threshold. When the warning value is greater than the defect threshold, output the welding defect with the highest average probability value among all welding defects in step S304 as the predicted possible ship welding defect.

[0073] In the ship welding process, various welding defects are often interconnected, and different defect types may be caused by the same or similar reasons. In step S304, the relevant factors of two welding defects are considered, so as to ensure that the calculation of the warning value is closer to reality.

[0074] The system also includes an early warning module, which is connected to both the defect identification module and the defect prediction module. When the defect identification module detects a situation where the identification result exceeds a preset alarm threshold, the system sends a real-time defect early warning message to the user. When the defect prediction module outputs a predicted ship welding defect, the system sends a predicted defect early warning message to the user. The real-time defect early warning message and the predicted defect early warning message can take the form of an alarm message, SMS, email, push notification, etc., to ensure that the user can be informed in a timely manner and take appropriate measures.

[0075] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. An artificial intelligence-based ship welding defect identification system, characterized in that, The system includes a data acquisition module, a feature extraction module, a defect identification module, and a defect prediction module. The data acquisition module acquires multidimensional data related to ship welding defects. The feature extraction module extracts dimensions from the multidimensional data that are highly correlated with welding defects. The defect identification module identifies ship welding defects. The defect prediction module predicts potential ship welding defects based on the identification results. The data acquisition module includes a real-time data acquisition module and a historical data acquisition module. The real-time data acquisition module acquires multidimensional data during the real-time welding process, and the historical data acquisition module acquires historical welding process data. The real-time data acquisition module includes a data sensing unit and a data integration unit. The data sensing unit is used to collect multi-dimensional data during the real-time welding process, and the data integration unit is used to perform calibration, formatting, and time synchronization operations on the multi-dimensional data. The feature extraction module extracts the dimensional data that is highly correlated with welding defects from the multi-dimensional data through the following steps: S101: Data preparation: acquire multi-dimensional data and corresponding time-point defect information from the historical welding process; S102: Construct a rough set: based on the data acquired in the previous step, establish a rough set model, which can be represented as... ,in The domain is the dataset containing multidimensional data from the historical welding process. The conditional feature set contains the universe of discourse. All dimensions in The decision feature set includes various defect categories and defect category values; the defect category value is the dimension corresponding to the defect category; S103: Extract important dimensions: calculate the approximate classification quality of each dimension, the approximate classification quality is calculated by the ratio of the lower approximation of each dimension to the universe of discourse; and compare these approximate classification qualities with the set extraction threshold, and take the dimensions corresponding to the approximate classification quality greater than the threshold as the dimensions with greater correlation to welding defects; The defect identification module establishes a BP neural network model and trains it using multidimensional data highly correlated with welding defects from historical welding processes and corresponding time-point defect information as the training set. The trained neural network model is then used for ship welding defect identification. The input to the neural network model is multidimensional data highly correlated with welding defects, and the output is the probability of occurrence of various welding defects, namely weld cracking, weld porosity, weld slag inclusion, weld misalignment, insufficient weld transition zone, and weld deformation. The specific method by which the defect prediction module predicts possible ship welding defects is described below. The process includes the following steps: S301: Extract 10 sets of temporal neural network model recognition results from the current time backward; each set includes the probability of welding cracks, weld holes, slag inclusions, misaligned weld edges, insufficient weld transition zone, and welding deformation occurring during ship welding; S302: Calculate the average probability value of each welding defect within the 10 sets of recognition results; S303: Extract the recognition results corresponding to the two welding defects with the highest average probability values ​​from step S302; S304: Calculate the warning value based on the data extracted in step S303. ： ;in, This represents the average probability value corresponding to the welding defect with the highest average probability value among all welding defects. The weld defect with the highest average probability value among all weld defects is the maximum value among its 10 identification results. The minimum value among the 10 sets of identification results for the welding defect with the highest average probability value. The difference between the welding defect with the highest average probability value among all welding defects and the second most recent identification result among its 10 identification results; The weld defect with the second highest average probability among all weld defects has the highest value among its 10 identification results. The minimum value among the 10 identified welding defects is the welding defect with the second highest average probability value. For the welding defect with the second highest average probability value among all welding defects, the difference between the identification result closest to the current time and the second highest identification result in its 10 sets of identification results is calculated; S305: The warning value is compared with the defect threshold. When the warning value is greater than the defect threshold, the welding defect with the highest average probability value among all welding defects in step S304 is output as the predicted possible ship welding defect.

Images