Welding structure load identification method and system based on physical information neural network

Abstract

The invention discloses a welding structure load identification method and system based on a physical information neural network, and the method comprises the steps: obtaining the dynamic load data of a welding structure part, carrying out the data signal preprocessing, and obtaining the dynamic load characteristics of the welding structure part; introducing an equation based on structure dynamics, and constructing a welding structure dynamic load identification model based on a physical information neural network; and carrying out welding structure load identification on the dynamic load characteristics of the welding structural part based on the welding structure dynamic load identification model of the physical information neural network to obtain a welding structure load identification result. According to the method, the problem of multi-dimensional load identification of a complex structural member can be effectively solved, and efficient and real-time load prediction and health monitoring capabilities are provided. The welding structure load identification method and system based on the physical information neural network can be widely applied to the technical field of welding structure dynamic load identification.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic load identification of welded structural parts, and in particular to a welded structural load identification method and system based on a physical information neural network. Background Art

[0002] Welded structural parts are widely used in engineering fields such as heavy machinery, aerospace, shipbuilding, and bridges. These structural parts are usually subjected to irregular dynamic loads in complex working environments, which have a significant impact on their structural strength, fatigue life, and safety. However, the dynamic loads of complex welded structural parts are nonlinear, time-varying, and uncertain. However, the relevant technologies mainly rely on finite element analysis, physical sensor layout, and empirical modeling. However, these methods are limited by the shortcomings of high computational cost, complex sensor layout, and susceptibility to noise interference. It is difficult to reflect the actual situation of dynamic loads in real time, accurately, and economically. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a welding structure load identification method and system based on physical information neural network, which can effectively handle the problem of multi-dimensional load identification of complex structural parts and provide efficient and real-time load prediction and health monitoring capabilities.

[0004] The first technical solution adopted by the present invention is: a welding structure load identification method based on physical information neural network, comprising the following steps: Acquire dynamic load data of welded structural parts and perform data signal preprocessing to obtain dynamic load characteristics of welded structural parts; Introducing structural dynamics equations, and constructing a dynamic load identification model for welded structures based on physical information neural networks; The dynamic load identification model of welding structure based on physical information neural network is used to identify the dynamic load characteristics of welding structural parts and obtain the welding structure load identification results.

[0005] Furthermore, the step of acquiring the dynamic load data of the welded structural member and performing data signal preprocessing to obtain the dynamic load characteristics of the welded structural member specifically includes: Obtain dynamic load data of welded structural parts through sensor equipment; The dynamic load data of the welded structural parts are preprocessed to obtain the dynamic load characteristics of the welded structural parts.

[0006] Furthermore, the step of obtaining the dynamic load data of the welded structural member by means of the sensor device specifically includes: Acquire strain signal data of welded structural parts through strain sensors; Acquire acceleration signal data of the welded structural parts through an acceleration sensor; Acquire displacement signal data of welding structural parts through displacement sensors; The dynamic load data of the welded structure are obtained by combining the strain signal data of the welded structure, the acceleration signal data of the welded structure and the displacement signal data of the welded structure.

[0007] Furthermore, the step of performing data signal preprocessing on the dynamic load data of the welded structural member to obtain the dynamic load characteristics of the welded structural member specifically includes: The dynamic load data of the welded structure is filtered by a filter to obtain the dynamic load data of the welded structure after noise is removed; The dynamic load data of the welded structure after noise removal is processed by wavelet decomposition method to obtain the preliminary dynamic load characteristics of the welded structure. The preliminary dynamic load characteristics of the welded structure are normalized to obtain the dynamic load characteristics of the welded structure.

[0008] Furthermore, the step of introducing a structural dynamics equation and constructing a welding structure dynamic load identification model based on a physical information neural network specifically includes: Selecting a three-layer neural network model, wherein the three-layer neural network model includes an input layer, an output layer and a hidden layer; Introduce structural dynamics equations, velocity initial conditions and displacement initial conditions, and construct the network model loss function; Based on the three-layer neural network model and the network model loss function, a dynamic load identification model of welding structure based on physical information neural network is constructed.

[0009] Furthermore, the expression of the network model loss function is specifically as follows: ; In the above formula, represents the structural dynamics equation, represents the initial velocity condition, represents the initial displacement condition, represents the weight coefficient of the loss function, denote the mass, damping and stiffness parameters respectively, , and are acceleration, velocity and displacement responses respectively, represents the external excitation of the dynamic system, express The weight of the loss term, represents the initial time point, Indicates A point in time, represents the number of randomly sampled points, Represents the index of the sampling point, represents the number of randomly sampled boundary points, Indicates the sampling time.

[0010] Furthermore, the welding structure dynamic load identification model based on physical information neural network performs welding structure load identification on the dynamic load characteristics of the welding structure to obtain the welding structure load identification result, which specifically includes: The dynamic load characteristics of the welded structural parts are processed by data division to obtain a dynamic load characteristic training set and a dynamic load characteristic test set; The dynamic load feature training set and the dynamic load feature test set are respectively input into the welding structure dynamic load identification model of the physical information neural network to perform welding structure load identification, and the predicted load data and the actual load data are obtained; The difference data value between the predicted load data and the actual load data is obtained through the mean square error loss function; According to the difference data value, the weight and threshold of the physical information neural network dynamic load identification model of the welding structure are optimized by genetic algorithm to obtain the optimized physical information neural network dynamic load identification model of the welding structure; The dynamic load identification model of welding structure based on the optimized physical information neural network is used to identify the dynamic load characteristics of welding structural parts and obtain the welding structure load identification results.

[0011] Furthermore, the genetic algorithm guides optimization through a fitness function, and the expression of the fitness function is specifically as follows: ; In the above formula, represents the fitness function, represents the number of output nodes of the neural network, Represents the neural network The expected output of each node is Indicates The predicted output of each node is represents the coefficient, represents the absolute value function.

[0012] The second technical solution adopted by the present invention is: a welding structure load identification system based on physical information neural network, comprising: The first module is used to obtain the dynamic load data of the welded structural part and perform data signal preprocessing to obtain the dynamic load characteristics of the welded structural part; The second module is used to introduce the structural dynamics equation and build a dynamic load identification model of welded structure based on physical information neural network; The third module is used to identify the dynamic load characteristics of the welded structural parts using the welding structure dynamic load identification model based on the physical information neural network to obtain the welding structure load identification results.

[0013] The beneficial effects of the method and system of the present invention are as follows: the present invention obtains the dynamic load data of the welded structural parts and performs data signal preprocessing, utilizes sensor data-driven learning to improve the generalization ability of the model, can handle the nonlinear characteristics of the structure, does not require an accurate physical model, and adapts to complex environmental changes. It further introduces a structural dynamics equation, constructs a welding structure dynamic load identification model based on a physical information neural network, introduces a physical information method to analyze the structural dynamic response problem, and enhances the prediction accuracy and stability of the model. Finally, the welding structure dynamic load identification model based on the physical information neural network performs welding structure load identification on the dynamic load characteristics of the welded structural parts, which can effectively handle the problem of multi-dimensional load identification of complex structural parts, and provide efficient and real-time load prediction and health monitoring capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the steps of the welding structure load identification method based on physical information neural network of the present invention; Figure 2 It is a structural block diagram of a welding structure load identification system based on physical information neural network of the present invention; Figure 3 It is a schematic diagram of setting a sensor on a welding structure provided by a specific embodiment of the present invention; Figure 4 It is a schematic diagram of the topological structure of a physical information neural network provided by a specific embodiment of the present invention; Figure 5 It is a schematic diagram of the optimization process of the physical information neural network provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0016] Reference Figure 1 The present invention provides a welding structure load identification method based on physical information neural network, which comprises the following steps: S100, acquiring dynamic load data of the welded structural component and performing data signal preprocessing to obtain dynamic load characteristics of the welded structural component; S110, obtaining dynamic load data of the welded structural member through a sensor device; Specifically, the strain signal data of the welded structure is obtained by the strain sensor; the acceleration signal data of the welded structure is obtained by the acceleration sensor; the displacement signal data of the welded structure is obtained by the displacement sensor; and the dynamic load data of the welded structure is obtained by combining the strain signal data of the welded structure, the acceleration signal data of the welded structure and the displacement signal data of the welded structure.

[0017] In this embodiment, strain sensors, acceleration sensors, displacement sensors and other sensing devices are arranged on the welded structural parts to collect dynamic response signals of the structural parts under different working conditions, including strain, displacement, vibration and acceleration data, which are mainly used to monitor the stress and motion state of the structure to identify and analyze the dynamic load conditions. The strain sensor is used to measure the strain (degree of deformation) of the welded structure under different loads; the acceleration sensor is used to measure the acceleration change of the structure under dynamic loads to reflect its vibration state and impact; the displacement sensor is used to measure the displacement change of the structure under the action of external force to help analyze the overall deformation of the structure.

[0018] More specifically, sensors are integrated into key locations of the suspension arm, such as Figure 3 As shown in the figure, Q_O is the strain data collected by the upper sensor, Q_R is the strain data collected by the right sensor, Q_U is the strain data collected by the lower sensor, and Q_L is the strain data collected by the left sensor. Multiple strain gauges and accelerometer sensing devices are arranged on the suspension arm. The multiple strain gauges collect the dynamic response of the stress concentration area. According to the results of finite element simulation, they are placed at the optimal position of the suspension arm, and the sensors collect strain data in different directions. The accelerometer is used to measure the acceleration change of the structure under dynamic load, reflecting its vibration state and impact. The sensor is connected to the edge computing device through the data collector to ensure real-time transmission and processing of data.

[0019] S120, performing data signal preprocessing on the dynamic load data of the welded structural component to obtain the dynamic load characteristics of the welded structural component.

[0020] Specifically, the dynamic load data of the welded structure is filtered by a filter to obtain the dynamic load data of the welded structure after noise removal; the dynamic load data of the welded structure after noise removal is feature extracted by a wavelet decomposition method to obtain preliminary dynamic load characteristics of the welded structure; the preliminary dynamic load characteristics of the welded structure are normalized to obtain the dynamic load characteristics of the welded structure.

[0021] In this embodiment, for the collected dynamic response signal, a bandpass filter is first used to remove noise, and then the wavelet decomposition method is used to extract the characteristic information data related to the load in the signal to ensure that the extracted characteristic data can reflect the dynamic load changes of the structure. After being extracted, the characteristic information data is normalized to reduce the impact of different dimensional data on the neural network training, so that the data meets the input requirements of the neural network.

[0022] It should also be noted that the forces applied to welded structures that vary with time and external conditions include impact forces in the horizontal and vertical directions or lateral loads caused by centrifugal forces.

[0023] S200, introduce the structural dynamics equation and build a dynamic load identification model of welding structure based on physical information neural network; Specifically, a three-layer neural network model is selected, and the three-layer neural network model includes an input layer, an output layer and a hidden layer; the structural dynamics equation, the initial conditions of velocity and the initial conditions of displacement are introduced to construct a network model loss function; based on the three-layer neural network model combined with the network model loss function, a dynamic load identification model of the welding structure based on the physical information neural network is constructed.

[0024] First of all, it should be noted that with the rapid development of artificial intelligence technology, physical information neural networks have shown great potential in the fields of nonlinear system modeling, pattern recognition and signal processing. Existing studies have proved the application value of physical information neural networks in load identification. PINN embeds structural dynamics equations in the neural network loss function, combines the advantages of physical constraints and data-driven, and can efficiently and accurately identify dynamic loads without the need for a large amount of system response data. Based on this, the present invention proposes an efficient method and system for identifying dynamic loads of complex welded structural parts using physical information neural networks (PINN), in order to overcome the shortcomings of the prior art.

[0025] In this embodiment, a dynamic load identification model based on physical information neural network (PINN) is constructed, such as Figure 4 As shown. The model input is feature information data, and the output is the structural displacement response. Physical information based on the structural dynamics equation is introduced as the loss function in the output layer of the model. The loss function consists of three parts, all of which use the mean square error (MSE) to measure the neural network fitting error. The neural network selects a neural network with three layers: input layer, output layer, and hidden layer. The hidden layer can abstract the features of the input data to a higher dimension, so that these originally nonlinear features can be better linearized, and any nonlinear mapping can be achieved by appropriately increasing the number of neuron nodes.

[0026] The loss function consists of three parts, all of which use mean square error (MSE) to measure the neural network fitting error. The loss function controlled by the structural dynamics formula equation ( ), initial velocity condition ( ), and the displacement initial conditions ( ), the specific equation is as follows: ; In the above formula, represents the structural dynamics equation, represents the initial velocity condition, represents the initial displacement condition, represents the weight coefficient of the loss function, denote the mass, damping and stiffness parameters respectively, , and are acceleration, velocity and displacement responses respectively, represents the external excitation of the dynamic system, express The weight of the loss term, represents the initial time point, Indicates A point in time, represents the number of randomly sampled points, Represents the index of the sampling point, represents the number of randomly sampled boundary points, Indicates the sampling time.

[0027] S300, a welding structure dynamic load identification model based on a physical information neural network performs welding structure load identification on dynamic load characteristics of welding structure parts to obtain welding structure load identification results.

[0028] Specifically, the dynamic load characteristics of the welding structure are divided into data and processed to obtain a dynamic load characteristic training set and a dynamic load characteristic test set; the dynamic load characteristic training set and the dynamic load characteristic test set are respectively input into the welding structure dynamic load identification model of the physical information neural network for welding structure load identification to obtain predicted load data and actual load data; the difference data value between the predicted load data and the actual load data is obtained through the mean square error loss function; according to the difference data value, the weights and thresholds of the welding structure dynamic load identification model of the physical information neural network are optimized through the genetic algorithm to obtain the optimized welding structure dynamic load identification model of the physical information neural network; the dynamic load characteristics of the welding structure are used to identify the welding structure load based on the optimized physical information neural network dynamic load identification model to obtain the welding structure load identification result.

[0029] In this embodiment, the neural network selects a three-layer neural network of input layer, output layer and hidden layer. The hidden layer can abstract the features of the input data to a higher dimension, so that these originally nonlinear features can be better linearized, and any nonlinear mapping can be achieved by appropriately increasing the number of neuron nodes.

[0030] The optimization process of physical information neural network is as follows Figure 5 As shown. Import the processed data set, divide the data set into a training set and a test set, use the training set to supervise the constructed physical information neural network model, and introduce the structural dynamics formula equation control, velocity and displacement loss function, and select the optimal model structure through cross-validation. Then use the genetic algorithm (GA) to optimize the physical information neural network. Through a series of operations such as selection operation, crossover operation, mutation operation and fitness function, the weights and thresholds that meet the termination conditions are obtained, and then the model is retrained. The mean square error loss function is introduced to measure the difference between the predicted load and the actual load, and then the model is optimized to improve the recognition accuracy and generalization ability. The genetic algorithm guides the optimization process through the fitness function, and determines the optimal solution based on the error between the predicted output and the actual output. The fitness function expression is as follows: ; In the above formula, represents the fitness function, represents the number of output nodes of the neural network, Represents the neural network The expected output of each node is Indicates The predicted output of each node is represents the coefficient, Represents the absolute value function, that is, it calculates the non-negative value of a number.

[0031] After training, the model can monitor the working status of the suspension arm in real time, identify dynamic loads in real time in the actual operating environment, predict the dynamic response of loads under different working conditions, and output load size and direction information for engineering analysis.

[0032] Reference Figure 2 , a welded structure load identification system based on physical information neural network, including: The first module 201 is used to obtain dynamic load data of the welded structural member and perform data signal preprocessing to obtain dynamic load characteristics of the welded structural member; The second module 202 is used to introduce a structural dynamics equation and construct a dynamic load identification model of a welded structure based on a physical information neural network; The third module 203 is used to perform welding structure load identification on the dynamic load characteristics of the welding structure based on the welding structure dynamic load identification model of the physical information neural network to obtain the welding structure load identification result.

[0033] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0034] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for identifying loads of welded structures based on physical information neural network, characterized in that: The following steps are involved: Acquire dynamic load data of welded structural parts and perform data signal preprocessing to obtain dynamic load characteristics of welded structural parts; Introducing structural dynamics equations, and constructing a dynamic load identification model for welded structures based on physical information neural networks; The dynamic load identification model of welding structure based on physical information neural network is used to identify the dynamic load characteristics of welding structural parts and obtain the welding structure load identification results.

2. The method for identifying welded structure load based on physical information neural network according to claim 1 is characterized in that: The step of acquiring the dynamic load data of the welded structural part and performing data signal preprocessing to obtain the dynamic load characteristics of the welded structural part specifically includes: Obtain dynamic load data of welded structural parts through sensor equipment; The dynamic load data of the welded structural parts are preprocessed to obtain the dynamic load characteristics of the welded structural parts.

3. The method for identifying welded structure load based on physical information neural network according to claim 2 is characterized in that: The step of obtaining the dynamic load data of the welded structural member by means of the sensor device specifically comprises: Acquire strain signal data of welded structural parts through strain sensors; Acquire acceleration signal data of the welded structural parts through an acceleration sensor; Acquire displacement signal data of welding structural parts through displacement sensors; The dynamic load data of the welded structure are obtained by combining the strain signal data of the welded structure, the acceleration signal data of the welded structure and the displacement signal data of the welded structure.

4. The method for identifying welded structure load based on physical information neural network according to claim 3 is characterized in that: The step of performing data signal preprocessing on the dynamic load data of the welded structural member to obtain the dynamic load characteristics of the welded structural member specifically includes: The dynamic load data of the welded structure is filtered by a filter to obtain the dynamic load data of the welded structure after noise is removed; The dynamic load data of the welded structure after noise removal is processed by wavelet decomposition method to obtain the preliminary dynamic load characteristics of the welded structure. The preliminary dynamic load characteristics of the welded structure are normalized to obtain the dynamic load characteristics of the welded structure.

5. The method for identifying welded structure load based on physical information neural network according to claim 4 is characterized in that: The step of introducing a structural dynamics equation and constructing a welding structure dynamic load identification model based on a physical information neural network specifically includes: Selecting a three-layer neural network model, wherein the three-layer neural network model includes an input layer, an output layer and a hidden layer; Introduce structural dynamics equations, velocity initial conditions and displacement initial conditions, and construct the network model loss function; Based on the three-layer neural network model and the network model loss function, a dynamic load identification model of welding structure based on physical information neural network is constructed.

6. The method for identifying welded structure load based on physical information neural network according to claim 5 is characterized in that: The expression of the network model loss function is specifically as follows: ； In the above formula, represents the structural dynamics equation, represents the initial velocity condition, represents the initial displacement condition, represents the weight coefficient of the loss function, denote the mass, damping and stiffness parameters respectively, , and are acceleration, velocity and displacement responses respectively, represents the external excitation of the dynamic system, express The weight of the loss term, represents the initial time point, Indicates A point in time, represents the number of randomly sampled points, Represents the index of the sampling point, represents the number of randomly sampled boundary points, Indicates the sampling time.

7. The method for identifying loads of welding structures based on physical information neural network according to claim 6 is characterized in that: The step of performing welding structure load identification on the dynamic load characteristics of the welding structure using the physical information neural network-based welding structure dynamic load identification model to obtain the welding structure load identification result specifically includes: The dynamic load characteristics of the welded structural parts are processed by data division to obtain a dynamic load characteristic training set and a dynamic load characteristic test set; The dynamic load feature training set and the dynamic load feature test set are respectively input into the welding structure dynamic load identification model of the physical information neural network to perform welding structure load identification, and the predicted load data and the actual load data are obtained; The difference data value between the predicted load data and the actual load data is obtained through the mean square error loss function; According to the difference data value, the weight and threshold of the physical information neural network dynamic load identification model of the welding structure are optimized by genetic algorithm to obtain the optimized physical information neural network dynamic load identification model of the welding structure; The dynamic load identification model of welding structure based on the optimized physical information neural network is used to identify the dynamic load characteristics of welding structural parts and obtain the welding structure load identification results.

8. The method for identifying welded structure load based on physical information neural network according to claim 7 is characterized in that: The genetic algorithm guides the optimization through a fitness function, and the expression of the fitness function is specifically as follows: ； In the above formula, represents the fitness function, represents the number of output nodes of the neural network, Represents the neural network The expected output of each node is Indicates The predicted output of each node is represents the coefficient, represents the absolute value function.

9. The welded structure load identification system based on physical information neural network is characterized by: Includes the following modules: The first module is used to obtain the dynamic load data of the welded structural part and perform data signal preprocessing to obtain the dynamic load characteristics of the welded structural part; The second module is used to introduce the structural dynamics equation and build a dynamic load identification model of welded structure based on physical information neural network; The third module is used to identify the dynamic load characteristics of the welded structural parts using the welding structure dynamic load identification model based on the physical information neural network to obtain the welding structure load identification results.

Images