Manned submersible observation window stress prediction method based on machine learning

Abstract

The invention belongs to the technical field of electric digital data processing, and particularly relates to a manned submersible observation window stress prediction method based on machine learning, which comprises the following steps: acquiring strain and pressure data, and establishing a finite element model of an observation window; constructing a training set based on the strain and pressure data, and correspondingly constructing a machine learning hybrid model; training the machine learning hybrid model through the training set, and obtaining a stress prediction model; and predicting stress data of the observation window of the manned submersible in different environments through the stress prediction model. The stress prediction method based on data driving does not depend on a complex physical modeling process, stress distribution under complex working conditions can be predicted more quickly and accurately, calculation time and resource consumption can be greatly reduced, and a robust stress prediction result can be provided in a complex load environment; the calculation time is shortened, the prediction precision is improved, and an efficient and reliable prediction result is provided for structural safety assessment.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical digital data processing, and in particular relates to a method for predicting stress of an observation window of a manned submersible based on machine learning. Background Art

[0002] As an important tool for marine scientific research and resource development, deep-sea manned submersibles need to operate safely in extremely high-pressure environments. The observation window is one of the important components of manned submersibles. It is usually made of transparent material polymethyl methacrylate (PMMA). While providing vision for operators, it also bears great deep-sea pressure. During the submersible's dive, as the external water depth and pressure increase, the observation window will experience huge external loads, resulting in stress accumulation, especially the maximum principal stress. This stress is one of the key factors affecting the structural safety of the observation window, because when the maximum principal stress exceeds the tensile strength or compressive strength of the PMMA material, the observation window may suffer tensile or compressive failure, thereby threatening the overall safety of the submersible. Therefore, accurately predicting the maximum principal stress of the observation window to ensure that it can withstand extreme external pressure is the key to ensuring the structural integrity of the manned submersible and the safety of the occupants.

[0003] In terms of stress prediction of observation windows, traditional methods mainly rely on numerical simulation techniques such as finite element analysis (FEA). Finite element analysis is an effective means of obtaining the overall stress distribution by decomposing the structure into many small units and performing stress, strain and displacement analysis on each unit. Finite element analysis can provide accurate stress distribution of observation windows under high pressure loads to a certain extent, and is widely used in structural design and safety assessment. However, due to the complexity of the working environment of deep-sea manned submersibles, the nonlinearity and uncertainty of external loads are high, resulting in many limitations of finite element analysis in practical applications. First, the computational cost of finite element analysis is high, which usually requires a lot of computing resources and time, making it difficult to achieve real-time prediction; second, under complex load conditions, the adaptability and accuracy of the finite element model may be limited, especially under multiple dynamic loads or irregular load conditions, the traditional finite element method is difficult to provide stable and accurate prediction results. Therefore, finding a more efficient and adaptable stress prediction method under complex deep-sea conditions has become a research hotspot.

[0004] In summary, the traditional analysis method needs to be improved so that the stress distribution of the submersible observation window under complex working conditions can be predicted more quickly and accurately, especially the maximum principal stress; it can also reduce the calculation time and resource consumption, and provide robust stress prediction results in more complex load environments, greatly improving the efficiency and applicability of stress prediction. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method for predicting stress of an observation window of a manned submersible based on machine learning to solve the above problems.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows: The present invention provides a method for predicting stress of a manned submersible observation window based on machine learning, the steps comprising: Collecting strain and pressure data of an observation window of a manned submersible and establishing a finite element model of the observation window of the manned submersible; constructing a training set based on the strain and pressure data, and then constructing a machine learning hybrid model based on the training set and the finite element model; Training the machine learning hybrid model through the training set and obtaining a stress prediction model; The stress prediction model is used to predict stress data of the observation window of the manned submersible under different environments.

[0007] Furthermore, the steps include: A validation set and a test set are also constructed based on the strain and pressure data; During the training process of the machine learning hybrid model, adjusting model hyperparameters by using the validation set; After the stress prediction model is obtained, the accuracy of the stress prediction model is evaluated through the test set.

[0008] Furthermore, the accuracy evaluation method includes: At least one of a mean square error, an average mean square error, and a root mean square residual is calculated based on the predicted stress data and the stress data of the test set.

[0009] Furthermore, when the accuracy is lower than a threshold, the machine learning hybrid model is reconstructed and data expansion is performed on the training set.

[0010] Furthermore, the method of data expansion is to increase the amount of data collection and / or add simulation of finite element analysis.

[0011] Furthermore, the method of data expansion is to add simulation quantities of finite element analysis, and the steps include: Sorting the training set based on the pressure sequence, and establishing a mapping relationship between the pressure and strain data sets to obtain an initial data set; Dividing the pressure sequence into a high-density interval and a low-density interval based on a data density threshold; Inserting a first pressure point in the high-density interval based on a linear interpolation method; Inserting a second pressure point in the low-density interval based on a midpoint interpolation method; Inputting the pressure values ​​corresponding to the first and second pressure points into a finite element analysis model to obtain corresponding strain simulation data; A mapping relationship is established between the pressure values ​​corresponding to the first and second pressure points and the strain simulation data, and the pressure values ​​are inserted into the initial data set to obtain a training set after data expansion.

[0012] Further, the strain data of the observation window of the manned submersible is multi-directional strain data; The structural outer large surface of the manned submersible observation window is provided with at least one key point, and two-dimensional strain data are collected at the key point, with directions of 0° and 90°; The structural inner surface of the observation window of the manned submersible is provided with at least one key point, and three-dimensional strain data are collected at the key point, with directions of 0°, 45° and 90°.

[0013] Furthermore, the architecture of the machine learning hybrid model includes a position encoding layer, a Transformer layer, a convolutional neural network layer, and a long short-term memory network layer in sequence.

[0014] Furthermore, the architecture of the machine learning hybrid model is provided with a position encoding layer, a Transformer layer, a CNN layer (convolutional neural network layer), and an LSTM layer (long short-term memory network layer) in sequence.

[0015] Furthermore, the convolutional neural network layer includes an input layer, a convolution layer, a pooling layer and a flattening layer; The output of the convolutional neural network layer is the input of the long short-term memory network layer; The long short-term memory network layer includes a fully connected layer and an output layer; the output result of the long short-term memory network layer is the maximum principal stress.

[0016] Furthermore, the hyperparameters of the machine learning hybrid model include learning rate, number of layers and number of neurons; A dropout layer is also provided in the long short-term memory network layer.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The data-driven stress prediction method of the present invention provides a new direction for the structural safety assessment of observation windows; its data-driven method uses historical data and experimental data, does not rely on complex physical modeling processes, and can predict stress distribution under complex working conditions more quickly and accurately, especially when dealing with nonlinear and high-dimensional characteristic data. Compared with traditional numerical simulation methods, the hybrid model of machine learning can not only greatly reduce computing time and resource consumption, but also provide robust stress prediction results in more complex load environments, greatly improving the efficiency and applicability of stress prediction, reducing computing time and improving the prediction accuracy of the model, providing efficient and reliable prediction results for the structural safety assessment of manned submersible observation windows under extreme working conditions, and also providing important technical support for the design, maintenance and real-time monitoring of deep-sea manned submersibles. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0019] Figure 1 The figure is a flow chart of the method of the present invention.

[0020] Figure 2 This is a schematic diagram of the observation window structure in a specific embodiment; Figure 3 It is a schematic diagram of a simulation of a finite element model of an observation window in a specific embodiment; Figure 4 This is a schematic diagram of the pressure change of the observation window in a specific embodiment; Figure 5 It is a schematic diagram of the strain change of the observation window in a specific embodiment; Figure 6 This is a schematic diagram of a CNN-LSTM model for observation window prediction in a specific embodiment; Figure 7 This is a schematic diagram of the observation window prediction Transformer-CNN-LSTM model in a specific embodiment; Figure 8 This is a graph showing the change in loss value of the CNN-LSTM model in a specific embodiment; Fig. 9 This is a graph showing the change in loss value of the Transformer-CNN-LSTM model in a specific embodiment; Fig.10 The CNN-LSTM model predicts the maximum principal stress value and the true principal stress value in a specific embodiment; Fig.11The Transformer-CNN-LSTM model predicts the maximum principal stress value and the true principal stress value in the specific embodiment; Fig.12 This is a comparison chart of Transformer-CNN-LSTM and CNN-LSTM indicators in a specific embodiment; Fig.13 This is a graph showing changes in Transformer-CNN-LSTM indicators in a specific embodiment; Fig.14 This is a graph showing changes in CNN-LSTM indicators in a specific embodiment. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0023] It should also be noted that the methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products unless otherwise specified.

[0024] This embodiment proposes a method for predicting the stress of a manned submersible observation window based on machine learning. Figure 1 As shown, the specific steps include: 1. Collect the strain and pressure data of the observation window of the manned submersible and establish the finite element model of the observation window of the manned submersible.

[0025] Among them, this embodiment is based on Figure 2 The submersible observation window (longitudinal cross section) shown is used as an example for explanation, and includes an outer large surface (circular), an inner small surface (circular) and a lateral conical surface.

[0026] Based on the structural dimensions of the submersible observation window, a finite element model of the manned submersible observation window structure under high pressure load is constructed. Furthermore, a finite element model simulating the observation window under high pressure load in a deep-sea environment is established, which can not only be used for subsequent analysis, but also for simulation analysis such as Figure 3As shown in Figure 1. Furthermore, the finite element model is constructed based on the material properties, geometric structure and boundary conditions of the observation window. The observation window is usually made of polymethyl methacrylate (PMMA), and its material parameters (density, elastic modulus, Poisson's ratio, etc.) directly affect the stress distribution results. Therefore, when building the model, these material parameters need to be accurately set. At the same time, the geometric structure (such as thickness, curvature) and boundary conditions (such as fixation, force direction) of the observation window should also be consistent with the real environment. The external high-pressure load under the deep sea is simulated in the finite element model to reflect the actual stress conditions under deep-sea conditions. Generally, the pressure of the deep-sea environment increases with increasing depth, so an external load that conforms to the deep-sea pressure is applied in the model to simulate the stress accumulation of the observation window during the dive. The applied high-pressure load should conform to the pressure gradient of the deep-sea environment in the model so that a stress distribution that conforms to the actual environment can be generated. The model is run in a finite element analysis software (such as Abaqus) for dynamic stress analysis. Through simulation analysis, the preliminary stress distribution of the observation window under high-pressure conditions can be obtained, including the stress concentration of key points and the position of the maximum principal stress. This process provides a preliminary theoretical basis for the structural safety of the observation window and also provides a reference for subsequent experiments.

[0027] Strain and pressure data: Conduct stress deformation experiments on the observation window structure under real high-pressure environment, that is, place the observation window in an experimental device that simulates the deep-sea high-pressure environment. The device can gradually apply pressure equivalent to the deep sea to simulate the external loads on the observation window under different depth environments. In order to obtain high-precision experimental data, the pressure is gradually increased during the experiment, and the rate of pressure increase is ensured to be consistent with the rate of pressure change during the actual dive. Use strain sensors and pressure sensors to collect strain data and pressure change data of key points on the surface of the observation window, such as Figure 4 and Figure 5 As shown in the figure, it is used as the initial data source of the model. During the experiment, the pressure sensor is used to record the pressure changes in the experimental device in real time so as to be analyzed synchronously with the strain data. The acquisition frequency of the pressure sensor is consistent with that of the strain sensor to ensure that the pressure data and strain data at each time point accurately correspond. These pressure data will reflect the intensity of the external load applied to the observation window and provide key load condition information for the model.

[0028] Preferably, the strain data of the observation window of the manned submersible in this embodiment is multi-directional strain data; specifically, two key points are set on the large surface of the structure outside the observation window of the manned submersible, and two-dimensional strain data are collected at each key point, with directions of 0° and 90°, that is, two groups of strain sensors are vertically placed along the surface; thus, this embodiment performs 13,000 collections to form 4×13,000 data. The small surface inside the structure of the observation window of the manned submersible is also provided with three key points, and three-dimensional strain data are collected at each key point, with directions of 0°, 45° and 90°; thus, 9×13,000 data are formed. Combined with the pressure data, the total length of the input vector of the constructed training set is 13×13,000 and the corresponding 13,000 pressure data.

[0029] The collected multidimensional data needs to be preprocessed before being passed into the machine learning hybrid model, including data cleaning, normalization, and time series alignment. Data cleaning is used to remove abnormal data, normalization ensures that the numerical range of the data is consistent when it is input into the model, and time series alignment ensures that the strain data and pressure data at each time point are completely corresponding to improve the learning effect of the model. The processed data will form a standardized training set, which can provide clear and effective feature input for the machine learning model. By taking the multidimensional data at each time point as a training sample, a complete training data set is formed. Each sample records the stress distribution of the observation window under specific high pressure conditions. Such training samples can reflect how pressure changes affect the stress on the surface of the observation window, providing comprehensive and accurate data support for stress prediction of the machine learning model.

[0030] 2. Build a training set based on the initial data (strain and pressure data), and then build a machine learning hybrid model based on the training set and the finite element model.

[0031] In this embodiment, a hybrid model integrating Transformer, Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is designed and constructed, wherein the Convolutional Neural Network is as follows: Figure 6 As shown in Figure 7 As shown in the figure, the machine learning hybrid model also introduces a position encoding layer. Specifically, the hybrid model includes a position encoding layer, a Transformer layer, a convolutional neural network layer, and a long short-term memory network layer. The position encoding layer is used to enhance the model's sensitivity to the temporal characteristics of stress data, and the self-attention mechanism learns the pattern in the sequence by calculating the correlation between each position and other positions.

[0032] Optionally, the Transformer layer is set as follows: 2.1 Transformer layer self-attention mechanism (Self-Attention) is used to make each input vector "pay attention" to other vectors in the sequence. The formula for self-attention is as follows: First, the input sequence strain data and pressure data are mapped into queries ( Q ),key( K ) and value ( V )matrix: ; In the formula, and is a trainable weight matrix.

[0033] Compute the dot product of the query and the key and scale the result as follows: ; In the formula, It is a scaling factor to prevent the value from being too large and causing the gradient to disappear or explode. Softmax is used to normalize the result of the dot product.

[0034] Multi-Head Attention extends the self-attention mechanism to multiple heads: ; Among them, each ,and is the output projection matrix.

[0035] 2.2 Transformer layer position encoding, since Transformer does not have a sequence structure, it is necessary to add sequence order information through positional encoding. The formula of the positional encoding vector is as follows: Given location POS and dimensions i : ; In the formula, POS Indicates the position of an element in a sequence. i represents the first dimension in the encoding i indexes. d is the dimension of the encoding. Position encoding uses sine and cosine functions to ensure that the encodings between different positions are unique and maintain the relative distance information between different positions. This position encoding is applied in a fixed dimension and can help Transformer capture the position information in the sequence.

[0036] The above enables the model to automatically capture the important relationships between different positions in the sequence. Since the Transformer model is generally ineffective in predicting stress, we considered combining the advantages of Transformer position encoding, bidirectional LSTM and one-dimensional convolutional neural network to enable it to provide high-precision stress prediction in real time under high-pressure environments.

[0037] This embodiment uses CNN to extract the features of the data set, which is used to process the spatial features of multi-dimensional data. CNN can perform convolution operations on three-dimensional strain data and pressure data, extract local spatial relationships and patterns from them, help the model understand the local features of stress distribution at different observation points, and then input these features into LSTM for prediction. The CNN in this embodiment mainly includes an input layer, a convolution layer, a pooling layer, and a flattening layer. The output of CNN is mainly used as the input of LSTM, and the 1D convolution layer and the pooling layer are used to extract the spatial features of the input data. The flattening layer is set between CNN and LSTM to convert the features of CNN into a one-dimensional array as the input of LSTM.

[0038] This embodiment uses the LSTM layer to extract the long-term dependency features of the time series. Since stress data has a time series, LSTM captures the stress evolution trend and law of the observation window during the pressure change process by memorizing and processing historical information.

[0039] The spatial features extracted by CNN, the temporal dependency features extracted by LSTM, and the global dependency features extracted by Transformer are fused to form a complete feature vector. These features can fully represent the multi-dimensional information in the stress data, ensuring that the model can capture local and global stress characteristics.

[0040] Preferably, a validation set is also constructed based on the strain and pressure data, and the model hyperparameter model is adjusted through the validation set during the training process of the machine learning hybrid model. Among them, the hyperparameter adjustment includes the learning rate, the number of layers and the number of neurons to ensure the best performance of the model in the maximum principal stress prediction; the performance of different network structures on stress prediction is compared, and the network structure with higher prediction accuracy is selected. This embodiment also sets three groups of one-dimensional convolution operations and pooling operations, and then flattens the operation to two LSTM layers and the output layer for output prediction. Each LSTM is set with a dropout layer to prevent overfitting, and the dropout is set to 0.5. The hyperparameters of the model are set as follows: The learning rate (controls the step size of the model weight update, using the learning rate scheduler and the adaptive optimization algorithm Adam to dynamically adjust the learning rate to speed up convergence) is set to 0.001; The number of neurons between the convolutional layer and the pooling layer is set to 64; The convolution kernel size and pooling size are set to 1; The number of neurons in both LSTMs is 100; Relu is used as the activation function.

[0041] In addition, in order to prevent the gradient explosion during training, gradient clipping is added before training to prevent gradient explosion. The maximum range of gradient clipping is 1, and the gradient is clipped before performing the optimization step.

[0042] Therefore, the fully connected layer and output layer of LSTM obtain the prediction result of the maximum principal stress.

[0043] Finally, the machine learning hybrid model is trained using the aforementioned training set, with the target being the maximum principal stress, and a stress prediction model is obtained after training.

[0044] Optionally, during the model training process, different model structure combinations can be tried and compared, using CNN-LSTM and Transformer-CNN-LSTM for comparison, and the prediction effect of each structure can be evaluated through cross validation to find the best model architecture. Specifically, the comparison results in this embodiment are as follows: pass Figure 8 and Fig. 9 From the loss value change diagram of the CNN-LSTM model and the loss value change diagram of the Transformer-CNN-LSTM model, it can be seen that the loss value change of the Transformer-CNN-LSTM model is more stable than that of the CNN-LSTM model, which means that the Transformer-CNN-LSTM model has a better learning effect on predicting the maximum principal stress value and better stability.

[0045] Fig.10 The CNN-LSTM model in this embodiment predicts the maximum principal stress value and the true principal stress value. Fig.11 The Transformer-CNN-LSTM model in this embodiment predicts the maximum principal stress value and the true principal stress value. From these two figures, it can be seen that both models are good at predicting the maximum principal stress, but when predicting the high point of stress concentration in the observation window, the Transformer-CNN-LSTM is obviously more accurate, so the Transformer-CNN-LSTM has higher prediction accuracy than the CNN-LSTM.

[0046] Fig.12This is a comparison chart of Transformer-CNN-LSTM and CNN-LSTM indicators in this embodiment. The Transformer-CNN-LSTM model has the best effect, which is specifically manifested in that the MSE, MAE, and RSR values ​​in the CNN-LSTM model are the smallest. The smaller the MSE and MAE values, the better, and the closer the RSR value is to 0, the better.

[0047] Fig.13 This is the Transformer-CNN-LSTM indicator change diagram in this embodiment. Fig.14 This is a graph showing changes in CNN-LSTM indicators in this embodiment. It can be seen from the figure that the evaluation indicators of the improved Transformer-CNN-LSTM model have been improved, and the evaluation indicators have been reduced.

[0048] From the above, it can be seen that the deep learning algorithm verifies the feasibility of the machine learning method to predict the stress structure of the observation window. By improving and optimizing the Transformer-CNN-LSTM algorithm, the learning effect is enhanced, which is highly consistent with the experimental data.

[0049] 3. Model performance evaluation.

[0050] Optionally, in order to effectively predict stress values, this embodiment also adds a model evaluation method, and the specific steps are as follows: A test set is also constructed based on strain and pressure data; After obtaining the stress prediction model, the accuracy of the stress prediction model is evaluated through the test set.

[0051] Specifically, when evaluating the performance of the model based on the predicted stress data and the stress data of the test set, the deviation and distribution similarity between the predicted value and the true value should be considered. Therefore, this embodiment uses the mean square error (MSE), average mean square error (MAE) and root mean square residual (RSR) to compare the deviation between the true value and the predicted value of different models and evaluate the performance of the model, wherein the formula involved is as follows: ; ; ; In the formula, N is the number of values, is the true value, is the predicted value, It is the average value of the true value. The smaller the mean square error (MSE) and average mean square error (MAE), the better. The closer the root mean square residual (RSR) is to 0, the better.

[0052] The above evaluation indicators are comprehensively defined as accuracy, and corresponding thresholds are set for each indicator according to engineering needs. When any indicator is lower than the threshold, the above steps are repeated, including reconstructing the machine learning hybrid model and expanding the training set data.

[0053] Optionally, methods of data expansion include increasing the amount of data collected and adding simulations for finite element analysis.

[0054] In this embodiment, considering the difficulty of measuring data and the long preparation period, the method of data expansion is to add the simulation of finite element analysis. The specific steps include: First, since each set of data in the data set contains the corresponding pressure, the pressure is arranged in ascending order, the training set is sorted based on the pressure sequence, and the mapping relationship between the pressure and strain data sets is established to obtain the initial data set.

[0055] Then, the range of the pressure sequence (a one-dimensional array of pressures) is divided into several intervals of equal length (intervals with the same pressure interval, such as 5 MPa as an interval length, and 10 MPa-15 MPa as an interval), and then the intervals are divided into high-density intervals and low-density intervals based on the data density threshold; if there are a pressure points in interval n, and a is greater than the data density threshold k, then this interval is a high-density interval; if there are b pressure points in interval m, and b is less than the data density threshold k, then this interval is a low-density interval.

[0056] In order to increase the simulation data, the corresponding pressure points need to be determined. Therefore, different pressure point insertion rules are set according to different intervals, as follows: (1) Insert the first pressure point in the high-density interval based on the linear interpolation method; since the data sampling in the measured data does not change completely linearly, there are such high-density intervals, in which there are many pressure points and they are densely distributed. The corresponding training set generated by this method can obtain better training results. Even if the simulation results of the expanded data are distorted to a certain extent, they will generally be corrected in the training under the influence of the measured data. Therefore, a simple linear interpolation method can be used to insert several first pressure points in the interval.

[0057] (2) Insert the second pressure point based on the midpoint interpolation method in the low-density interval; Since there are fewer pressure points in such low-density intervals and they are sparsely distributed, the general linear interpolation method will cause the second pressure point to be close to the measured pressure point. If the strain variable of the simulation analysis is distorted, the closer erroneous data will cause the weight update to deviate in the neural network. Since the model adjusts the weight according to the loss function of the sample during training, the distorted data will cause the loss function to produce an erroneous gradient signal. Especially in small batch training, if the sample containing the erroneous data is combined with other correct samples in a batch, the direction of gradient descent may be "pulled off" by the erroneous sample. For example, when calculating the mean square error loss between the predicted strain value and the true value, the model will try to reduce this loss, so that the weight update is carried out in the direction of adapting to the erroneous sample, resulting in a decrease in the prediction accuracy on the correct sample. Therefore, to interpolate in this interval, it is necessary to keep the inserted second pressure point as far away from the measured pressure point as possible. The ideal method is the special midpoint interpolation method in linear interpolation, that is, each second pressure point is the same distance from the two adjacent measured pressure points.

[0058] The pressures corresponding to the first and second pressure points are sequentially input into the constructed finite element analysis model to obtain corresponding strain simulation data.

[0059] Then, the pressure and strain simulation data corresponding to the first and second pressure points are mapped and inserted back into the initial data set after the time is assigned; the time assignment can refer to the interpolation point to calculate the corresponding time, which will not be repeated here. After data insertion, an expanded data set containing real collected data and finite element simulation insertion data can be obtained, thereby forming a new training set for improving the accuracy of the stress prediction model.

[0060] 4. Output the results.

[0061] According to actual needs, environmental parameters such as pressure are substituted into the stress prediction model to predict the stress data of the observation window of the manned submersible in different environments, and this can be used to conduct further structural strength research or verify applications such as observation window design schemes.

[0062] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A method for predicting stress of a manned submersible observation window based on machine learning, characterized in that the steps include: Collecting strain and pressure data of an observation window of a manned submersible and establishing a finite element model of the observation window of the manned submersible; constructing a training set based on the strain and pressure data, and then constructing a machine learning hybrid model based on the training set and the finite element model; Training the machine learning hybrid model through the training set and obtaining a stress prediction model; The stress prediction model is used to predict stress data of the observation window of the manned submersible under different environments.

2. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 1, characterized in that: The steps also include: A validation set and a test set are also constructed based on the strain and pressure data; During the training of the machine learning hybrid model, adjusting model hyperparameters using the validation set; After the stress prediction model is obtained, the accuracy of the stress prediction model is evaluated through the test set.

3. The method for predicting stress of observation window of manned submersible based on machine learning according to claim 2, characterized in that: The accuracy evaluation method includes: At least one of a mean square error, an average mean square error, and a root mean square residual is calculated based on the predicted stress data and the stress data of the test set.

4. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 2 or 3, characterized in that: When the accuracy is lower than a threshold, the machine learning hybrid model is reconstructed and data expansion is performed on the training set.

5. The method for predicting stress of observation window of manned submersible based on machine learning according to claim 4, characterized in that: The method of data expansion is to increase the amount of data collection and / or add the simulation amount of finite element analysis.

6. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 5, characterized in that: The method for data expansion is to add the simulation amount of finite element analysis, and the steps include: Sorting the training set based on the pressure sequence, and establishing a mapping relationship between the pressure and strain data sets to obtain an initial data set; Dividing the pressure sequence into a high-density interval and a low-density interval based on a data density threshold; Inserting a first pressure point in the high-density interval based on a linear interpolation method; Inserting a second pressure point in the low-density interval based on a midpoint interpolation method; Inputting the pressure values ​​corresponding to the first and second pressure points into a finite element analysis model to obtain corresponding strain simulation data; A mapping relationship is established between the pressure values ​​corresponding to the first and second pressure points and the strain simulation data, and the pressure values ​​are inserted into the initial data set to obtain a training set after data expansion.

7. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 1, characterized in that: The strain data of the observation window of the manned submersible is multi-directional strain data; The structural outer large surface of the manned submersible observation window is provided with at least one key point, and two-dimensional strain data are collected at the key point, with directions of 0° and 90°; The structural inner surface of the observation window of the manned submersible is provided with at least one key point, and three-dimensional strain data are collected at the key point, with directions of 0°, 45° and 90°.

8. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 1, characterized in that: The architecture of the machine learning hybrid model includes a position encoding layer, a Transformer layer, a convolutional neural network layer, and a long short-term memory network layer in sequence.

9. The method for predicting stress of a manned submersible observation window based on machine learning according to claim 8, characterized in that: The convolutional neural network layer includes an input layer, a convolution layer, a pooling layer and a flattening layer; The output of the convolutional neural network layer is the input of the long short-term memory network layer; The long short-term memory network layer includes a fully connected layer and an output layer; the output result of the long short-term memory network layer is the maximum principal stress.

10. The method for predicting stress of observation window of manned submersible based on machine learning according to claim 9, characterized in that: The hyperparameters of the machine learning hybrid model include learning rate, number of layers, and number of neurons; A dropout layer is also provided in the long short-term memory network layer.

Images