Artificial Intelligence-Based Thermal Fatigue Evaluation Method and Device for Safety Hook Slings

Abstract

This application relates to the field of fatigue testing, and particularly to a method and device for thermal fatigue assessment of safety hook slings based on artificial intelligence. The method includes: obtaining the current monitoring data of the sling to be evaluated; according to the current monitoring data, obtaining the thermal fatigue prediction result of the sling to be evaluated through a thermal fatigue prediction model; wherein, the thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of sample slings; the historical monitoring data is the state data of the sample sling during actual use; the fatigue test data is the state data obtained by performing fatigue tests on the sample sling; the thermal fatigue prediction result indicates the degree of thermal fatigue of the sling to be evaluated. This application can solve the technical problem in the prior art of lacking a method to determine the thermal fatigue state of the hook sling.

Description

Technical Field

[0001] The present application relates to the field of fatigue testing, and in particular to a method and device for thermal fatigue evaluation of safety hook slings based on artificial intelligence. Background Art

[0002] During the construction and maintenance of infrastructure such as power plants and mines, materials or heavy equipment are required. As an important tool in the handling process, the safety, reliability, and durability of hook slings need to be strictly monitored and evaluated to ensure their safe operation during handling.

[0003] In actual operation, equipment thermal fatigue is an important factor leading to the failure of hook slings. Especially when hook slings are under long-term high temperature and high stress, their fatigue damage gradually accumulates, and eventually cracks, deformation, or even fracture of hook slings may occur. Among them, thermal fatigue not only affects the service life of hook slings but also may cause serious safety accidents.

[0004] In practical applications, although determining the thermal fatigue state of hook slings is crucial for their safety, there is a lack of corresponding technical solutions in the existing technology. Summary of the Invention

[0005] In view of this, the purpose of the present application is to provide a method and device for thermal fatigue evaluation of safety hook slings based on artificial intelligence to solve the technical problem of the lack of a technical solution for determining the thermal fatigue state of hook slings in the existing technology.

[0006] In the first aspect, the present application provides a method for thermal fatigue evaluation of safety hook slings based on artificial intelligence, and the method includes:

[0007] Obtain the current monitoring data of the hook sling to be evaluated;

[0008] According to the current monitoring data, obtain the thermal fatigue prediction result of the hook sling to be evaluated through a thermal fatigue prediction model;

[0009] Among them, the thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of sample hook slings; the historical monitoring data is the state data of sample hook slings during actual use; the fatigue test data is the state data obtained from fatigue tests on sample hook slings; and the thermal fatigue prediction result indicates the degree of thermal fatigue of the hook sling to be evaluated.

[0010] In the second aspect, the present application provides a device for thermal fatigue evaluation of safety hook slings based on artificial intelligence, and the device includes: a data module and a prediction module;

[0011] The data module is used to obtain the current monitoring data of the sling to be evaluated.

[0012] The prediction module is used to obtain the thermal fatigue prediction result of the sling to be evaluated through a thermal fatigue prediction model according to the current monitoring data.

[0013] Among them, the thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of sample slings; the historical monitoring data is the state data of sample slings during actual use; the fatigue test data is the state data obtained by performing fatigue tests on sample slings; the thermal fatigue prediction result indicates the thermal fatigue degree of the sling to be evaluated.

[0014] Beneficial effects:

[0015] The present application provides a method for evaluating the thermal fatigue of a safety sling based on artificial intelligence. The method includes: obtaining the current monitoring data of the sling to be evaluated; obtaining the thermal fatigue prediction result of the sling to be evaluated through a thermal fatigue prediction model according to the current monitoring data; among them, the thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of sample slings; the historical monitoring data is the state data of sample slings during actual use; the fatigue test data is the state data obtained by performing fatigue tests on sample slings; the thermal fatigue prediction result indicates the thermal fatigue degree of the sling to be evaluated.

[0016] In summary, according to the trained thermal fatigue prediction model of the present application, the relationship between the state of the sling and the degree of thermal fatigue is learned. Subsequently, the current monitoring data of the sling to be evaluated is input into the model to obtain the thermal fatigue prediction result. Therefore, the present application solves the technical problem in the prior art of lacking a method for determining the thermal fatigue state of the sling; in addition, due to the technical solution provided by the present application, the thermal fatigue state of the sling can be accurately evaluated, providing a basis for subsequent maintenance and management. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments of the present application. The following drawings only show some embodiments of the present application, so they should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a schematic flowchart of the method for evaluating the thermal fatigue of a safety sling based on artificial intelligence provided by the embodiments of the present application.

[0019] Figure 2Schematic diagram for comparing the noise suppression capabilities of different sample augmentation methods provided by the embodiments of the present application;

[0020] Figure 3 Schematic diagram for comparing the data generation capabilities of different sample augmentation methods provided by the embodiments of the present application;

[0021] Figure 4 Schematic diagram for comparing the prediction capabilities of different sample augmentation methods provided by the embodiments of the present application;

[0022] Figure 5 Schematic diagram of the structure of an artificial intelligence-based thermal fatigue evaluation device for safety hook slings provided by the embodiments of the present application. Detailed implementation manners

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0024] First, the embodiments of the present application provide an artificial intelligence-based thermal fatigue evaluation method for safety hook slings, as Figure 1 shown Figure 1 Schematic diagram of the process of the artificial intelligence-based thermal fatigue evaluation method for safety hook slings provided by the embodiments of the present application. The method includes: S110~S120, details are as follows:

[0025] S110: Obtain the current monitoring data of the hook sling to be evaluated during use.

[0026] Specifically, in the embodiments of the present application, the hook sling to be evaluated refers to the hook sling that is being used during the construction and maintenance processes of power plants and mines and needs to be evaluated (i.e., predicted) for thermal fatigue.

[0027] In actual operation, the monitoring data of the hook sling includes: the sling monitoring data of the hook sling to be evaluated during the current working process and the environmental monitoring data of the current working environment of the hook sling to be evaluated.

[0028] The monitoring data is usually collected through sensors such as temperature sensors. The sensor data can reflect various state data of the hook sling to be evaluated at the data collection moment; the sensor for collecting the sling monitoring data is installed on the hook sling, and the sensor for collecting the environmental monitoring data is set in the working environment where the hook sling is located.

[0029] Among them, the types of sensors at least include: temperature sensors, force sensors, strain gauges, acceleration sensors, and other types. All sensors are communicatively connected to a data acquisition terminal for real-time transmission of the status data obtained by real-time acquisition to a background server for storage and processing. Among them, the acceleration sensor is used to measure the vibration data of the hook sling to be evaluated. The sling monitoring data at least includes: data such as the sling temperature, stress, and deformation of the sample hook sling, and the environmental monitoring data at least includes: data such as the load of the sample hook sling and the environmental temperature of the working environment.

[0030] In actual operation, the monitoring data is stored in a database in the form of a time series, and the data storage format is a structured file format such as JSON, CSV, or HDF5.

[0031] In actual operation, since the training samples of the embodiments of the present application include the monitoring data of the sample hook sling under different environmental states, and different monitoring data can reflect different degrees of thermal fatigue, the labels of the training samples of the embodiments of the present application can be "no fatigue", "slight fatigue", "moderate fatigue", "severe fatigue", etc. indicating the degree of thermal fatigue; the annotation method is manual annotation based on expert experience, and complies with the annotation specifications of the machine learning training data set.

[0032] During the process of monitoring the hook sling to be evaluated that is being used during the construction and maintenance processes of power plants and mines, it is necessary to regularly collect monitoring data at a preset time interval; among them, the preset time interval can be determined according to actual needs, and the present application does not make specific limitations on this.

[0033] S120: According to the current monitoring data, obtain the thermal fatigue prediction result of the hook sling to be evaluated through the thermal fatigue prediction model;

[0034] Among them, the thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of the sample hook sling; the historical monitoring data is the status data collected by sensors during the actual use of the sample hook sling, and the fatigue test data is the status data obtained by performing a fatigue test on the sample hook sling; the thermal fatigue prediction result indicates the degree of thermal fatigue of the hook sling to be evaluated.

[0035] Specifically, the historical monitoring data is the status data collected by sensors in the working environment, that is, the historical monitoring data is the actual status data of the sample hook sling; the fatigue test data is the status data collected by sensors during the fatigue test in the simulated environment, that is, the fatigue test data is the simulated status data of the sample hook sling, and the fatigue test data can be used to determine the sling monitoring data of the sample hook sling under any environmental monitoring data.

[0036] In one implementation, the fatigue test data includes: conventional fatigue test data and accelerated fatigue test data; before S110, the method further includes steps (1) to (5), details are as follows:

[0037] Step (1): Obtain a training sample set of the sample hook and rigging, including historical monitoring data, conventional fatigue test data, and accelerated fatigue test data;

[0038] Among them, the conventional fatigue test data is the state data obtained from the conventional fatigue test of the sample hook and rigging, and the conventional fatigue test is a test that simulates the actual use process of the sample hook and rigging; the accelerated fatigue test data is the state data obtained from the accelerated fatigue test of the sample hook and rigging.

[0039] Specifically, both the conventional fatigue test data and the accelerated fatigue test data are data collected by sensors in a simulated working environment. The difference is that the conventional fatigue test data is collected in a simulated working environment close to the actual working environment, while the accelerated fatigue test data is collected in a simulated working environment that is quite different from the actual working environment, and is used to shorten the test time to directly obtain the state data of the sample hook and rigging during long-term operation.

[0040] Step (2): Determine the original feature vector of each original sample among the multiple original samples included in the training sample set.

[0041] Specifically, through operations such as encoding, the various device data included in each original sample are converted into the original feature vector in vector form.

[0042] Step (3): Construct new feature vectors according to the original feature vectors and the linear interpolation formula.

[0043] Specifically, first, insufficient training samples are likely to lead to poor generalization ability of models such as the thermal fatigue prediction model, and at the same time affect the prediction accuracy of models such as the thermal fatigue prediction model. Therefore, in the embodiments of the present application, "sample expansion" is achieved by constructing new feature vectors.

[0044] In addition, since the historical monitoring data, conventional fatigue test data, and accelerated fatigue test data of the sample hook rigging have different statistical distribution characteristics, such as the conventional fatigue test data may be more stable, the historical monitoring data is greatly affected by the working environment, and the environmental monitoring data included in the accelerated fatigue test data may contain extreme cases. To avoid prediction errors that may be caused by the above different statistical distribution characteristics, the embodiments of the present application use symmetric polynomials to ensure the non-linear expression ability of features, enabling features from different data sources to be fused in a high-order feature space and improving the diversity of generated samples. For example, in the historical monitoring data, the real-time fluctuation data of the bearing capacity of the hook rigging may contain high-frequency noise, and the symmetric polynomial can effectively extract stable feature patterns and avoid the influence of abnormal data on sample expansion.

[0045] In the embodiments of the present application, "sample expansion" is achieved by constructing feature vectors in the feature space.

[0046] The linear interpolation formula is as follows:

[0047] ;

[0048] ;

[0049] In the formula, represents the original feature vector; represents the newly added feature vector; represents the symmetric polynomial of; represents the th interpolation coefficient; represents the th interpolation matrix; represents the number of interpolation coefficients and interpolation matrices, the value of can be determined according to actual needs; represents the vector set selected from multiple original feature vectors; represents converting the vector into a diagonal matrix; represents the natural exponential function; represents the L2 norm.

[0050] In actual operation, the newly generated feature vectors obtained by using the linear interpolation formula for "sample augmentation" can not only fill the distribution gaps in the existing data, but also optimize the data for specific data sources. For example, accelerated fatigue test data is usually used to predict the wear of sling rigging during long-term use. The distribution of accelerated fatigue test data may deviate from the historical monitoring data to a certain extent. To improve data consistency, the linear interpolation formula uses a set of vectors randomly selected from the dataset and weights them by the L2 norm, making the newly generated feature vectors more in line with the actual working conditions. When performing sample augmentation, data points close to the limit load condition can be selected for interpolation, making the newly generated feature vectors more representative.

[0051] is determined according to the symmetric polynomial formula; the symmetric polynomial formula is shown as follows:

[0052] ;

[0053] ;

[0054] In the formula, symmetric polynomial function; represents the weight of the th term in represents the number of terms of represents the vector corresponding to the rd feature in represents the number of feature vectors included in represents the power of the vector corresponding to the th feature in , represents the highest order; represents the th power of the vector corresponding to the th feature in , , and are all positive integers.

[0055] In the embodiments of the present application, to verify the noise suppression ability of the technical solutions provided by the embodiments of the present application for data, the noise suppression abilities of different sample augmentation methods are now compared. As Figure 2 shown, Figure 2 is a schematic diagram of the comparison of the noise suppression abilities of different sample augmentation methods provided by the embodiments of the present application. Figure 2 The abscissa ofFigure 2 The ordinate is the spectral amplitude. The high-frequency noise achieved by the traditional interpolation method still exists, and there are obvious noise signals in the 40Hz - 100Hz region. However, the high-frequency noise achieved by the embodiments of the present application is significantly reduced, the curve is smoother and the key features are retained.

[0056] It should be emphasized that Figure 2 The bulge near 50Hz in is actually the main frequency component of the signal, that is, the dominant signal feature of the monitoring data. This signal distribution conforms to the normal situation because the signal generated by the embodiments of the present application is a sine wave (50Hz) and the embodiments of the present application do not remove the real signal components, but mainly eliminate the high-frequency noise.

[0057] is determined according to the first coefficient update formula; the first coefficient update formula is shown as follows:

[0058] ; ;

[0059] ; ;

[0060] In the formula, represents the updated th interpolation coefficient; represents the interpolation coefficient adjustment amount; represents the th weight determined based on the Euclidean distance of the

[0061] represents the verification index corresponding to the vector corresponding to the th feature in represents all and the average value of the Euclidean distances between; represents the regularization parameter; represents the importance coefficient of the vector corresponding to the th feature in

[0062] In actual operation, the optimization difference step of the embodiment of the present application can ensure the rationality of sample expansion. Due to the data density differences in different sources of sample data, such as historical monitoring data is usually evenly distributed, while conventional fatigue test data may be densely distributed under specific working conditions. This regularization strategy can adapt to this non-uniform distribution situation. Especially in historical monitoring data, due to environmental factors, there may be fewer data points under specific conditions. For example, the force data of rigging under high humidity or high temperature environments may be less. This regularization strategy can perform additional interpolation on such data to fill the data gaps and improve the adaptability of the model.

[0063] The calculation formula is as follows:

[0064] ; ;

[0065] ;

[0066] In the formula, represents the regularization parameter; represents the vector corresponding to the th feature in ; represents the vector corresponding to the th feature in ; and represents the Euclidean distance between and in the feature space; and represent the adjustment factor and the threshold respectively; represents the th feature in the feature vector ; represents the th feature in the feature vector ;

[0067] In actual operation, by calculating the L2 norm between data points, the quality of the generated data is ensured. For example, in historical monitoring data, there is less deformation data of hook rigging under certain extreme loads. By the L2 norm constraint, the generated data can be prevented from deviating from the true physical laws. For example, when interpolating between 20% and 80% of the rated load, if the generated data deviates from the actually measured deformation curve, data points that do not conform to the actual working conditions can be screened out by setting a threshold.

[0068] In one implementation, after constructing the newly added feature vector, it is also necessary to verify the newly added feature vector through a verification formula, and the verification formula is as follows:

[0069] ;

[0070] In the formula, represents the verification index of the newly added feature vector. In actual execution, if exceeds the preset threshold , it is considered that the newly added feature vector is unqualified and needs to be discarded.

[0071] In the embodiments of the present application, to verify the superiority of the newly added feature vectors generated by the technical solutions provided in the embodiments of the present application, the similarity between the newly added feature vectors generated by different sample augmentation methods and the original feature vectors is now compared. As Figure 3 shown, Figure 3 is a schematic diagram of the comparison of the data generation capabilities of different sample augmentation methods provided in the embodiments of the present application. Figure 3 The abscissa of Figure 3 indicates different data augmentation methods (i.e., sample augmentation methods), Figure 3 and the ordinate is the L2 norm. The L2 norm is used to measure the error. A lower L2 norm indicates that the newly added feature vector is closer to the original feature vector, thereby improving the credibility of the model. The experimental results show that the technical solutions provided in the embodiments of the present application can generate high-quality newly added feature vectors more accurately.

[0072] Step (4): Combine the original feature vector and the newly added feature vector to obtain a feature vector set.

[0073] Specifically, step (3) is repeatedly executed to continuously achieve "sample augmentation" until the predetermined data quantity requirement is met; in actual execution, when the total number of the original feature vectors and the newly added feature vectors included in the feature vector set is greater than or equal to 2000, step (3) is stopped.

[0074] In actual operation, to verify the influence of different sample augmentation methods on the thermal fatigue prediction model and further prove that the sample augmentation method provided in the embodiments of the present application can improve the classification accuracy of the model, a comparison of the classification accuracy of the rigging fatigue prediction is now carried out. As Figure 4 shown, Figure 4 is a schematic diagram of the comparison of the prediction capabilities of different sample augmentation methods provided in the embodiments of the present application. Figure 4 The abscissa of Figure 4The vertical coordinate is the classification accuracy. The experimental results show that the newly added feature vectors obtained by the sample augmentation method provided in the embodiments of the present application are closer to the actual working conditions, enabling the thermal fatigue prediction model to better distinguish different fatigue states.

[0075] In actual operation, after sample augmentation, it is also necessary to preprocess the augmented feature vectors to provide structured and high-quality input data for the thermal fatigue prediction model. Among them, the preprocessing at least includes: data denoising, outlier detection, and data normalization operations. Data denoising is used to eliminate environmental interference, outlier detection ensures data integrity, and data normalization optimizes the data distribution.

[0076] In actual operation, after preprocessing, key indicators reflecting the thermal fatigue characteristics of the hooking rigging can also be extracted and constructed from the feature data obtained after preprocessing, and high-quality feature vectors can be constructed to improve the classification accuracy of the thermal fatigue prediction model, and feature dimension reduction technology can be used to reduce the feature dimension to prevent overfitting. In addition, an intelligent evaluation model can also be constructed for the feature data after preprocessing to achieve accurate prediction and risk warning of the thermal fatigue prediction model.

[0077] Step (5): According to the feature vector set, perform iterative training on the neural network model to obtain a thermal fatigue prediction model.

[0078] Specifically, during the iterative training process, when the training stop condition is reached, the thermal fatigue prediction model can be obtained.

[0079] In one implementation, step (5) includes: step (5.1) to step (5.2), details are as follows:

[0080] Step (5.1): According to the feature vector set, determine the reconstructed feature data through a feature dimension reduction model. The neural network structure of the feature dimension reduction model is an autoencoder. The autoencoder includes: an encoder, a quantum state network layer, and a decoder connected in sequence. The quantum state network layer is used to compress the low-dimensional feature data output by the encoder to obtain compressed feature data, and the compressed feature data is used to be input into the decoder, so that the decoder decodes the compressed feature data to obtain the reconstructed feature data.

[0081] The existing autoencoder only includes: an encoder (encoder) and a decoder (decoder). The encoder E is used to convert the feature data with a higher dimension into the feature data with a lower dimension, and the decoder D is used to decode the low-dimensional feature representation to obtain the reconstructed feature data. In the embodiments of the present application, a quantum state network layer is inserted between the encoder E and the decoder D of the autoencoder to process and store the quantum state information of the low-dimensional feature data, thereby improving the efficiency and quality of feature compression.

[0082] In practical applications, the feature dimensionality reduction model is obtained by iteratively training an autoencoder; in actual execution, when the number of iterations of the iterative training reaches 1000, the training stops.

[0083] In one implementation, during the iterative training of the autoencoder, the network parameters of the encoder E and the decoder D need to be initialized first, as detailed below:

[0084] ; ;

[0085] In the formula, represents the initial weight matrices of the encoder E and the decoder D; represents the initial bias. In actual execution, can be set to 0; and respectively represent the number of neurons included in the input layer and the output layer in the autoencoder.

[0086] After the network parameters are initialized, the weight matrix and bias of the encoder E are updated according to their respective parameter update formulas during the iterative training. In actual operation, after the network parameters are initialized, it is possible to make different types of data have a relatively balanced activation distribution when input into the autoencoder, avoiding gradient disappearance or explosion caused by differences in the numerical ranges of different data sources. For example, the values of conventional fatigue test data may be relatively large, while the historical monitoring data is relatively stable. Initialization helps to balance the data distributions of the above data.

[0087] The parameter update formulas corresponding to the weight matrix and bias of the encoder E are as follows:

[0088] ; ;

[0089]

[0090] ;

[0091] ;

[0092] In the formula, represents the weight matrix used by the encoder E in the th training; represents the weight matrix used by the encoder E in the th training; represents the bias used by the encoder E in the th training; represents the bias used by the encoder E in the th training; Represents the learning rate of the preset encoder E;

[0093] Represents the value of the loss function determined based on the Represents the th training of the autoencoder; th feature vector input to the encoder E during the Represents the th training; reconstructed feature data output by the decoder D corresponding to

[0094] Represents the preset regularization coefficient. In actual execution, can be set to 0.3; Represents the weight coefficient of the preset quantum fidelity term. In actual execution, can be set to 0.4; Represents the fidelity between the quantum state data and the reconstructed state data . The quantum state data is the data obtained after quantum encoding of the low-dimensional feature data by the quantum state network layer during the th training. The reconstructed state data is the compressed feature data obtained after quantum decoding of by the quantum state network layer during the th training;

[0095] Represents the Sigmoid activation function; Represents the weight matrix used by the th network layer in the decoder D during the th training. The number of network layers included in the decoder D is ; Represents the bias used by the th network layer in the decoder D during the th training; Represents the output data output by the th network layer in the decoder D during the th training; Represents the multiplication between elements; Represents the feature importance coefficient used during the th training, indicating the importance degree of the features included in ;

[0096] In actual operation, historical monitoring data may be affected by environmental noise, resulting in large fluctuations in some data. In the embodiments of the present application, the gradient descent method is used to optimize the encoder parameters to reduce the data error caused by dimensionality reduction.

[0097] In actual operation, since the monitoring data involves sensor timing data (such as stress, tensile force), material test data (such as fatigue strength), and accelerated test data (such as wear rate under high-temperature environment), the Sigmoid activation function can be used to ensure that the data is mapped to a controllable range, avoiding instability in network training caused by extreme values. For example, there may be mutation points in the tensile test data of a sample hanger rigging, and the Sigmoid function can smooth the changes in the above data, making the features after data dimensionality reduction more stable.

[0098] The calculation formula of is as follows:

[0099] ;

[0100] In the formula, represents the Softmax function; represents in the th training, the output data of the th network layer in the encoder E (i.e., the low-dimensional feature data mentioned above), and the number of network layers included in the encoder E is ; and respectively represent the weight matrix and bias used in the th network layer in the encoder E in the th training.

[0101] In one implementation, the process of the quantum state network layer processing the low-dimensional feature data output by the encoder E in the th training includes the following steps:

[0102] 1) The quantum encoding function in the quantum state network layer encodes the low-dimensional feature data onto qubits to obtain quantum state data ; The quantum encoding function is as follows:

[0103] ;

[0104] In the formula, represents the quantum encoding function; represents the superposition of quantum states; represents the number of qubits in the quantum state network layer; represents the th training, the low-dimensional feature data output by the th network layer in the encoder E the th element;

[0105] represents the quantum encoding unit gate applied to each low-dimensional feature data. In actual execution, it is implemented by combining Hadamard gates and Pauli gates;

[0106] Among them, during the process of the encoder E encoding the feature vector, the process of processing the input data in each 1 network layer is as follows:

[0107]

[0108] In the formula, represents the output data output by the th network layer in the encoder E during the th training; represents the output data output by the th network layer in the encoder E during the th training, ; and respectively represent the weight matrix and bias used in the th network layer in the encoder E during the th training.

[0109] In actual operation, since the monitoring data includes multi-modal information (such as multi-dimensional data from different sensors), conventional dimensionality reduction methods may lose the key relevance of some data. For example, in conventional fatigue test data, there may be a complex non-linear relationship between the stress deformation data of metal materials and the environmental temperature and humidity. Quantum state multiplexing can, through the quantum superposition property, enable multiple feature information to be encoded onto the quantum state simultaneously, thereby retaining key information during the compression process.

[0110] 2) The quantum decoding function in the quantum state network layer decodes the quantum state data to obtain the compressed feature data; the quantum decoding function is as follows:

[0111] = ;

[0112] In the formula, represents the quantum decoding function; represents the input data of the first network layer in the decoder D during the th training, that is, the compressed feature data; represents the reconstructed state data during the th training in the Data of qubits Represents the unit gate in the decoder D, which is set corresponding to and is used to convert the reconstructed state data into compressed feature data. In actual operation, due to inevitable information loss during the dimensionality reduction of monitoring data, quantum decoding restores the data through inverse quantum gate operations and combines measurement operations to achieve data conversion. For on-site real-time monitoring data, such as vibration signals and stress signals, the decoding stage further screens out the most representative features through feature importance calculation, thereby reducing redundant features and improving the quality of data reconstruction.

[0113]

[0114] Step (5.2): According to the reconstructed feature data, an iterative training is performed on the neural network model to obtain a thermal fatigue prediction model.

[0115] In one implementation, the neural network structure of the thermal fatigue prediction model is an extreme learning machine. The network parameters updated during the iterative training of the extreme learning machine include: the number of neurons included in the hidden layer and the weight matrix of the output layer.

[0116] In practical applications, the input data used for training the extreme learning machine is the reconstructed feature data output by the trained feature dimensionality reduction model.

[0117] During the iterative training of the extreme learning machine, it is necessary to first initialize the number of neurons included in the hidden layer and the weight matrix of the output layer of the extreme learning machine (not shown below). The details are as follows:

[0118] ;

[0119] In the formula, represents the number of neurons included in the hidden layer of the extreme learning machine;

[0120] represents the minimum value function; represents the dimension of the reconstructed feature data.

[0121] After initializing the number of neurons included in the hidden layer and the weight matrix of the output layer of the extreme learning machine, the number of neurons included in the hidden layer of the extreme learning machine is updated according to the first parameter update formula during the training process; the weight matrix of the output layer of the extreme learning machine is updated according to the second parameter update formula.

[0122] ​In actual operation, the dimension of the input data directly affects the initial number of neurons in the hidden layer, and the prediction of the thermal fatigue prediction model is usually affected by various factors including stress, temperature, and material properties. Therefore, key fatigue influencing factors will be retained during feature reduction to ensure that the extreme learning machine can effectively learn the fatigue damage pattern.

[0123] The first parameter update formula is as follows:

[0124] ;

[0125] In the formula, represents the number of neurons included in the hidden layer of the extreme learning machine in the th training; represents the number of neurons included in the hidden layer of the extreme learning machine in the th training;

[0126] represents the number of neurons that can be increased in the hidden layer of the preset extreme learning machine during each training process. In actual implementation, the value can be set to 5, that is, if the prediction accuracy rate of the extreme learning machine does not reach the target accuracy rate, 5 neurons will be automatically added to the hidden layer of the extreme learning machine until the prediction accuracy rate of the extreme learning machine tends to be stable or reaches the quantity threshold, and the quantity threshold can be set to 50;

[0127] represents the prediction accuracy rate of the extreme learning machine determined based on the th training, represents the target accuracy rate of the extreme learning machine; represents the indicator function. When the prediction accuracy rate of the extreme learning machine does not reach the target accuracy rate, its value is 1, otherwise, its value is 0;

[0128] In actual operation, if the prediction accuracy rate of the thermal fatigue prediction model is low, additional nodes can be added to improve the classification ability of the thermal fatigue prediction model for this category, which helps to avoid the model ignoring minority categories and improve the overall classification performance.

[0129] The second parameter update formula is as follows:

[0130] ;

[0131] In the formula, represents the weight matrix of the output layer of the extreme learning machine in the th training; represents the output matrix of the hidden layer of the extreme learning machine; represents the transpose of the output matrix of the hidden layer of the extreme learning machine; denotes the target output matrix; denotes a preset regularization parameter. In actual execution, can be set to 0.3; denotes the identity matrix.

[0132] The loss function used by the thermal fatigue prediction model during iterative training is as follows:

[0133] ;

[0134] In the formula, denotes the loss function value of the extreme learning machine determined based on the th training; denotes a preset regularization parameter. In actual execution, can be set to 0.3; denotes the weight matrix of the output layer of the extreme learning machine in the th training.

[0135] In actual execution, if the loss function value of the extreme learning machine determined based on the th training is less than the preset loss function threshold, will be retained, indicating the end of the training of the output layer part of the extreme learning machine. Otherwise, the th training will be performed.

[0136] In actual operation, the embodiment of the present application uses the quantum fidelity term to measure the degree of information loss during the dimensionality reduction process. Since the accelerated fatigue test data often contains key failure modes, if the dimensionality reduction leads to the loss of important modes, it will affect the accuracy of equipment life prediction. By using the quantum fidelity term, the similarity between the reconstructed data and the original data can be improved during the training process. For example, in the dimensionality reduction task of accelerated fatigue test data, this optimization strategy can ensure that the main change trends of materials under different stress states are still retained after data compression, thereby improving the accuracy of fault diagnosis.

[0137] In actual operation, in actual application, if the label of the thermal fatigue prediction model is a binary classification of 0-1 (low fatigue vs. high fatigue), then a smaller regularization parameter (such as 0.1-0.3) can be used in the training process of the thermal fatigue prediction model to ensure the clarity of the classification boundary. If it is a multi-level fatigue classification (such as low, medium, and high fatigue damage), then the regularization parameter needs to be appropriately increased (such as 0.5) to enhance the generalization ability of the thermal fatigue prediction model and reduce overfitting.

[0138] In one implementation, since the dimensionality-reduced data used for training the extreme learning machine has been processed by a feature dimensionality reduction model, before using the dimensionality-reduced data to train the extreme learning machine, it is also necessary to supplement "technical features" to the dimensionality-reduced data; the supplementation of "technical features" is determined according to a feature construction formula;

[0139] The feature construction formula is as follows:

[0140] ; ;

[0141] In the formula, represents the vector of reconstructed feature data; represents the feature vector to be added to each ; represents feature extraction based on finite element analysis; represents the feature vector merging operation; , and represent the results of dimensionality reduction according to different dimensional benchmarks based on the principal component analysis method, represents the first finite element vector, represents the second finite element vector, represents the third finite element vector.

[0142] In actual operation, finite element analysis can simulate the stress distribution and deformation of the rigging during long-term operation, and extract the most representative fatigue features after dimensionality reduction by principal component analysis. For example, the residual stress distribution of the hook rigging under different loads can be generated by finite element analysis and added as an additional feature to the extreme learning machine model, enabling the thermal fatigue prediction model to more accurately distinguish between high-fatigue and low-fatigue categories.

[0143] In actual operation, the performance of the thermal fatigue prediction model can also be verified and continuously improved to ensure that the model has high accuracy and good generalization ability in actual applications; among them, performance verification and continuous improvement include cross-validation, performance index calculation, error analysis, and an online update mechanism. Cross-validation is used to evaluate the stability of the thermal fatigue prediction model, performance index calculation provides a quantitative evaluation basis, error analysis helps to locate the deficiencies of the model, and the online update mechanism enables the thermal fatigue prediction model to adapt to new data and dynamically adjust; in addition, the evaluation results and model prediction results of the thermal fatigue prediction model can also be presented to decision-makers and maintenance personnel in an intuitive way to support fire safety management and preventive maintenance.

[0144] In summary, the technical solutions in the embodiments of the present application have the following advantages compared with the prior art:

[0145] 1) Adopt a multilinear interpolation algorithm with symmetric polynomial embedding, make full use of the physical, monitoring, and simulation data of the rigging, achieve high-quality sample augmentation, improve the generalization ability of the thermal fatigue prediction model, and reduce the overfitting problem of the thermal fatigue prediction model;

[0146] 2) Adopt an autoencoder neural network based on quantum state multiplexing, use the quantum state multiplexing layer to reduce information loss, and improve the compression efficiency and reconstruction quality of the thermal fatigue characteristics of the rigging.

[0147] 3) Adopt an extreme learning machine algorithm based on dynamic hidden layer node allocation, dynamically adjust the structure of the classification model according to the complexity of the state data, improve the classification accuracy, and make it adapt to the fatigue assessment requirements under different working conditions.

[0148] Second, this application provides a thermal fatigue assessment device for safety hook rigging based on artificial intelligence, as Figure 5 shown, Figure 5 is a schematic structural diagram of the thermal fatigue assessment device for safety hook rigging based on artificial intelligence provided by an embodiment of this application. The device includes: a data module 310 and a prediction module 320;

[0149] The data module 310 is used to obtain the current monitoring data of the hook rigging to be evaluated;

[0150] The prediction module 320 is used to obtain the thermal fatigue prediction result of the hook rigging to be evaluated through the thermal fatigue prediction model according to the current monitoring data;

[0151] Among them, the thermal fatigue prediction model is obtained by training with the historical monitoring data and fatigue test data of the sample hook rigging; the historical monitoring data is the state data of the sample hook rigging during actual use; the fatigue test data is the state data obtained from the fatigue test of the sample hook rigging; the thermal fatigue prediction result indicates the thermal fatigue degree of the hook rigging to be evaluated.

[0152] In one implementation, the fatigue test data includes: conventional fatigue test data and accelerated fatigue test data; the prediction module 320 is further used to obtain a training sample set of the sample hook rigging including historical monitoring data, conventional fatigue test data, and accelerated fatigue test data;

[0153] Among them, the conventional fatigue test data is the state data obtained from the conventional fatigue test of the sample hook rigging, and the conventional fatigue test is a test process that simulates the actual use process of the sample hook rigging; the accelerated fatigue test data is the state data obtained from the accelerated fatigue test of the sample hook rigging;

[0154] The prediction module 320 is further used to determine the original feature vector of each original sample in the multiple original samples included in the training sample set;

[0155] The prediction module 320 is further configured to construct new feature vectors according to the original feature vectors and the linear interpolation formula;

[0156] The prediction module 320 is further configured to combine the original feature vectors and the new feature vectors to obtain a feature vector set;

[0157] The prediction module 320 is further configured to train the thermal fatigue prediction model according to the feature vector set.

[0158] In one implementation, the linear interpolation formula is as follows:

[0159] ;

[0160] ;

[0161] In the formula, represents the original feature vector; represents the new feature vector; represents the symmetric polynomial of; represents the th interpolation coefficient; represents the th interpolation matrix; represents the number of interpolation coefficients and interpolation matrices; represents the vector set selected from multiple original feature vectors; represents converting the vector into a diagonal matrix; represents the natural exponential function; represents the L2 norm.

[0162] In one implementation, is determined according to the symmetric polynomial formula; the symmetric polynomial formula is as follows:

[0163] ;

[0164] ;

[0165] In the formula, symmetric polynomial function; represents the weight of the th term in; represents the number of terms of; represents the vector corresponding to the th feature in; represents the number of feature vectors included in; represents The power of the vector corresponding to the th feature, , represents the highest order; represents the th power of the vector corresponding to the th feature in

[0166] In one implementation, the prediction module 320 is further configured to determine reconstructed feature data according to the feature vector set through a feature dimensionality reduction model;

[0167] wherein, the neural network structure of the feature dimensionality reduction model is an autoencoder; the autoencoder includes: an encoder, a quantum state network layer, and a decoder connected in sequence. The quantum state network layer is configured to compress the low-dimensional feature data output by the encoder to obtain compressed feature data, and the compressed feature data is used to be input into the decoder, so that the decoder decodes the compressed feature data to obtain reconstructed feature data;

[0168] The prediction module 320 is further configured to iteratively train the neural network model according to the reconstructed feature data to obtain a thermal fatigue prediction model.

[0169] In one implementation, the weight matrix and bias of the encoder are updated according to their respective corresponding parameter update formulas during the iterative training process. The parameter update formulas corresponding to the weight matrix and bias of the encoder are as follows:

[0170] ;

[0171] ;

[0172] ;

[0173] ;

[0174] In the formula, represents the weight matrix used by the encoder in the th training; represents the weight matrix used by the encoder in the th training; represents the bias used by the encoder in the th training; represents the bias used by the encoder in the th training; represents the preset learning rate of the encoder;

[0175] represents the loss function value determined based on the th training of the autoencoder; denotes the th feature vector input to the encoder during the th training; denotes the reconstructed feature data output by the decoder corresponding to during the th training; denotes a preset regularization coefficient; denotes the weight coefficient of a preset quantum fidelity term;

[0176] denotes the quantum state data and the reconstructed state data ; the fidelity between the quantum state data is the data obtained by quantum encoding the low-dimensional feature data by the quantum state network layer during the th training, and the reconstructed state data is the compressed feature data obtained by quantum decoding by the quantum state network layer during the th training; denotes the number of samples, denotes the F-norm;

[0177] denotes the Sigmoid activation function; denotes the weight matrix used by the th network layer in the decoder during the th training; the number of network layers included in the decoder is ; denotes the bias used by the th network layer in the decoder during the th training; denotes the output data output by the th network layer in the decoder during the th training; denotes the multiplication between elements; denotes the feature importance coefficient used during the th training, indicating the importance degree of the features included in ;

[0178] In one implementation, the neural network structure of the thermal fatigue prediction model is an extreme learning machine; the number of neurons included in the hidden layer of the extreme learning machine is updated according to the first parameter update formula during the iterative training process, and the first parameter update formula is as follows:

[0179] ;

[0180] In the formula, represents the number of neurons included in the hidden layer of the extreme learning machine in the th training; represents the number of neurons included in the hidden layer of the extreme learning machine in the th training; represents the number that the neurons included in the hidden layer of the preset extreme learning machine can increase in each training process;

[0181] represents the prediction accuracy rate of the extreme learning machine determined based on the th training; represents the target accuracy rate of the extreme learning machine; represents an indicator function, which is 1 when the prediction accuracy rate of the extreme learning machine does not reach the target accuracy rate, and 0 otherwise.

[0182] In one implementation, the loss function used by the thermal fatigue prediction model in the iterative training process is as follows:

[0183] ；

[0184] In the formula, represents the loss function value of the extreme learning machine determined based on the th training; represents a preset regularization parameter; represents the weight matrix of the output layer of the extreme learning machine in the th training.

[0185] Third, the present application also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of S110~S120 provided in the above embodiments are implemented.

[0186] Fourth, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps of S110~S120 in the above embodiments are executed.

[0187] Fifth, the computer program product provided by the present application includes a computer-readable storage medium storing program codes. The instructions included in the program codes can be used to execute the methods in the foregoing method embodiments. For specific implementation, reference can be made to the steps of S110~S120 in the method embodiments, which will not be elaborated here.

[0188] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some communication interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical or other forms.

[0189] In addition, the units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0190] Furthermore, in each embodiment of this application, the various functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0191] It should be noted that if the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0192] In this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0193] The above are only the embodiments of the present application and are not intended to limit the protection scope of the present application. For those skilled in the art, various modifications and changes can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A thermal fatigue assessment method for safety hook rigging based on artificial intelligence, characterized in that: The method comprises: Obtain current monitoring data for the hook rigging to be evaluated; According to the current monitoring data, a thermal fatigue prediction result of the hook rigging to be evaluated is obtained through a thermal fatigue prediction model; The thermal fatigue prediction model is obtained by training with historical monitoring data and fatigue test data of sample hook rigging; the historical monitoring data is the state data of the sample hook rigging during actual use; the fatigue test data is the state data obtained by fatigue testing the sample hook rigging; the thermal fatigue prediction result indicates the degree of thermal fatigue of the hook rigging to be evaluated; The fatigue test data includes: conventional fatigue test data and accelerated fatigue test data; before obtaining the current monitoring data of the hook rigging to be evaluated, the method further includes: Acquire a training sample set including the historical monitoring data, the conventional fatigue test data and the accelerated fatigue test data about the sample hook rigging; The conventional fatigue test data is the state data obtained by performing a conventional fatigue test on the sample hook rigging, and the conventional fatigue test is a test process that simulates the actual use process of the sample hook rigging; the accelerated fatigue test data is the state data obtained by performing an accelerated fatigue test on the sample hook rigging; Determining an original feature vector of each of the original samples in the plurality of original samples included in the training sample set; Constructing a new feature vector according to the original feature vector and the linear interpolation formula; Combining the original feature vector and the newly added feature vector to obtain a feature vector set; According to the feature vector set, training is performed on the thermal fatigue prediction model; The step of training the thermal fatigue prediction model according to the feature vector set includes: Determining, according to the feature vector set, reconstructing feature data through a feature dimension reduction model; The neural network structure of the feature dimensionality reduction model is an autoencoder; the autoencoder comprises: an encoder, a quantum state network layer and a decoder connected in sequence, the quantum state network layer is used to compress the low-dimensional feature data output by the encoder to obtain compressed feature data, the compressed feature data is used to input into the decoder, and the decoder is made to decode the compressed feature data to obtain reconstructed feature data; According to the reconstructed characteristic data, an iterative training is performed on a neural network model to obtain the thermal fatigue prediction model.

2. The method according to claim 1, characterized in that The linear interpolation formula is as follows: ； ； In the formula, represents the original feature vector; represents the newly added feature vector; express Symmetric polynomial of ; Indicates interpolation coefficients; Indicates interpolation matrix; Represents the number of interpolation coefficients and interpolation matrices; represents a vector set selected from the plurality of original feature vectors; Indicates converting a vector into a diagonal matrix; represents the natural exponential function; represents the L2 norm.

3. The method according to claim 2, characterized in that Said is determined according to a symmetric polynomial formula; the symmetric polynomial formula is as follows: ； ； In the formula, Symmetric polynomial functions; express Middle The weight of the item, express The number of items; express The corresponding A vector of features; express the number of eigenvectors included; express Corresponding to The power of the vector of features, , Indicates the highest order; express Corresponding to The vector of features Power.

4. The method according to claim 1, characterized in that: The weight matrix and bias of the encoder are updated according to their corresponding parameter update formulas during the iterative training process. The parameter update formulas corresponding to the weight matrix and bias of the encoder are as follows: ； ； ； ； In the formula, Indicated in The weight matrix used by the encoder in this training; Indicated in The weight matrix used by the encoder in this training; Indicated in The bias used by the encoder in this training; Indicated in The bias used by the encoder in this training; Indicates the preset encoder learning rate; Indicates that based on The loss function value determined by training the encoder once; Indicated in The first input to the encoder in the training feature vectors; Indicated in The decoder output in the training corresponds to Reconstructed feature data; Represents the preset regularization coefficient; Represents the weight coefficient of the preset quantum fidelity term; Representing quantum state data and reconstructed state data The fidelity between the quantum state data For the In the training, the quantum state network layer obtains the data after quantum encoding the low-dimensional feature data, and reconstructs the state data For the In this training, the quantum state network layer targets The compressed feature data obtained by quantum decoding, is the sample size, represents the F-norm; Represents the Sigmoid activation function; Indicated in In the decoder of the training The weight matrix used by the network layers, the number of network layers included in the decoder is ; Indicated in In the decoder of the training The bias used by each network layer; Indicated in In the decoder of the training The output data output by each network layer; Represents multiplication between elements; Indicated in The feature importance coefficient used in training, indicating The importance of the features contained in .

5. The method according to claim 1, characterized in that The neural network structure of the thermal fatigue prediction model is an extreme learning machine; the number of neurons included in the hidden layer of the extreme learning machine is updated according to the first parameter update formula during the iterative training process, and the first parameter update formula is as follows: ； In the formula, Indicated in The number of neurons included in the hidden layer of the extreme learning machine during training; Indicated in The number of neurons included in the hidden layer of the extreme learning machine during training; Indicates the number of neurons included in the hidden layer of the preset extreme learning machine that can be increased during each training process; Indicates that based on The prediction accuracy of the extreme learning machine determined by the training; represents the target accuracy of the extreme learning machine; Represents the indicator function, which has a value of 1 when the prediction accuracy of the extreme learning machine does not reach the target accuracy, otherwise it has a value of 0.

6. The method according to claim 5, characterized in that The loss function used by the thermal fatigue prediction model in the iterative training process is as follows: ； In the formula, Indicates that based on The loss function value of the extreme learning machine determined by the training; represents the preset regularization parameter; Indicated in The weight matrix of the output layer of the extreme learning machine in this training.

7. An artificial intelligence-based safety hook rigging thermal fatigue assessment device, characterized in that: For implementing the method of claim 1, the device comprises: a data module and a prediction module; The data module is used to obtain current monitoring data of the hook rigging to be evaluated; The prediction module is used to obtain the thermal fatigue prediction result of the hook rigging to be evaluated through a thermal fatigue prediction model according to the current monitoring data; Among them, the thermal fatigue prediction model is obtained by training on historical monitoring data and fatigue test data of sample hook rigging; the historical monitoring data is the status data of the sample hook rigging during actual use; the fatigue test data is the status data obtained by fatigue testing the sample hook rigging; the thermal fatigue prediction result indicates the degree of thermal fatigue of the hook rigging to be evaluated.

8. The device according to claim 7, characterized in that The fatigue test data includes: conventional fatigue test data and accelerated fatigue test data; The prediction module is further used to obtain a training sample set including the historical monitoring data, the conventional fatigue test data and the accelerated fatigue test data about the sample hook rigging; The conventional fatigue test data is the state data obtained by performing a conventional fatigue test on the sample hook rigging, and the conventional fatigue test is a test process that simulates the actual use process of the sample hook rigging; the accelerated fatigue test data is the state data obtained by performing an accelerated fatigue test on the sample hook rigging; The prediction module is further used to determine the original feature vector of each of the multiple original samples included in the training sample set; The prediction module is further used to construct a newly added feature vector according to the original feature vector and a linear interpolation formula; The prediction module is further used to combine the original feature vector and the newly added feature vector to obtain a feature vector set; The prediction module is further used to train the thermal fatigue prediction model according to the feature vector set.

Images