Bulb tube service life prediction method and device, equipment and storage medium

Abstract

The invention provides a bulb tube service life prediction method and device, equipment and a storage medium, and relates to the technical field of medical equipment, and the method can well mine the degradation information of data in a time dimension through a multi-scale feature extraction network, can mine the degradation information of data in a space dimension through a two-dimensional convolutional network, and improves the prediction accuracy of the service life of a bulb tube. Data mining is carried out in multiple dimensions, and more comprehensive degradation information is provided. A double-branch-group attention module is adopted to carry out further key feature extraction on fusion features, and the module adopts a multi-scale parallel structure to capture local and long-term dependency relationships in data, so that the double-branch-group attention module can learn richer feature representation. The target prediction model obtained through training is used for predicting the service life of the bulb tube, and the effect of improving prediction accuracy can be achieved.

Description

Technical Field

[0001] The present application relates to the technical field of medical devices, and particularly relates to a method, device, equipment, and storage medium for predicting the life of an X-ray tube. Background Art

[0002] In the current field of medical imaging, X-ray computed tomography (X-ray CT), as an important diagnostic tool, is crucial for examining human body parts such as the brain, spine, chest and abdomen, pelvis, and limbs. CT equipment not only plays a significant role in accurate diagnosis, guiding surgery, evaluating treatment effects, and disease research. However, the X-ray tube, one of the core components of CT equipment, as a high-value consumable, its performance and reliability directly affect image quality, imaging speed, and the stability of equipment operation. The CT X-ray tube system involves multiple subsystems such as mechanical and electronic systems, and its operating data has a high degree of dimensionality and complex coupling relationships. These data show significant correlations in the time series, and also show non-linear correlations in the spatial dimension. This spatial correlation is determined by the interaction of the internal structure of the CT X-ray tube, and the interdependence between various operating parameters constitutes a complex network in space.

[0003] Regarding the life prediction of CT X-ray tubes, the currently mainly used methods include methods based on expert knowledge, model-based, and data-driven methods. The method based on expert knowledge relies on experts in the field of medical imaging equipment. Through their professional knowledge, an expert system is constructed, and then a life prediction model is formed. However, this method faces the challenges of difficult acquisition and transformation of expert knowledge into codes, and usually requires long-term cooperation and development between experts and knowledge engineers. The model-based method focuses on establishing an accurate physical model, and by comparing the residuals between the model prediction values and the actual measurement values, it is judged whether the equipment is abnormal. Traditional time series prediction networks basically only focus on single-dimensional or single-scale features, and rarely can simultaneously focus on multi-scale and multi-dimensional features. Moreover, traditional life prediction models cannot capture local and long-term dependencies in the data. Summary of the Invention

[0004] The present application provides a method, device, equipment, and storage medium for predicting the life of an X-ray tube, which can improve the accuracy of X-ray tube life prediction.

[0005] A method for predicting the life of an X-ray tube provided by the present application includes: Obtaining a training sample data set, where the training sample data set includes multiple training samples, each training sample corresponds to a training label, the training label represents the remaining life of the X-ray tube corresponding to the training sample, and the training sample includes time series corresponding to multiple characteristic parameters of the X-ray tube; Input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple temporal features from the temporal features of the training samples through the multi-scale feature extraction network; Input the training samples into the two-dimensional convolutional network of the initial prediction model, and extract spatial features from the spatial features of the training samples through the two-dimensional convolutional network; Input multiple temporal features and the spatial features into the feature fusion layer of the initial prediction model to obtain fused features; Input the fused features into the dual-branch group attention module of the initial prediction model, determine attention weights and reorganized features through the dual-branch group attention module, and determine the output of the attention module according to the attention weights and the reorganized features; Input the output of the attention module into the fully connected network of the initial prediction model to obtain a life prediction label, and iteratively train the initial prediction model according to the training label and the life prediction label, and use the iteratively trained initial prediction model as the target prediction model for predicting the life of the to-be-predicted tube.

[0006] In an embodiment of the present application, the multi-scale feature extraction network includes multiple temporal convolutional networks, and the convolutional kernels of each temporal convolutional network are different. Extracting multiple temporal features from the temporal features of the training samples through the multi-scale feature extraction network includes: Extract temporal features from the training samples through each temporal convolutional network respectively to obtain multiple temporal features.

[0007] In an embodiment of the present application, inputting multiple temporal features and the spatial features into the feature fusion layer of the initial prediction model to obtain fused features includes: Perform splicing fusion on multiple temporal features and the spatial features through the feature fusion layer to obtain the fused features.

[0008] In an embodiment of the present application, inputting the fused features into the dual-branch group attention module of the initial prediction model, determining attention weights and reorganized features through the dual-branch group attention module, and determining the output of the attention module according to the attention weights and the reorganized features includes: Group the fused features according to the preset number of groups and the number of channels of the fused features to obtain multiple grouped features, and obtain reorganized features according to the multiple grouped features; Perform channel information interaction on the reorganized features to obtain channel interaction features; Perform local spatial feature extraction on the reorganized features to obtain local spatial features; Determine the first weight according to the channel cross-interaction feature, and determine the second weight according to the local spatial feature; Determine the first cross-interaction result according to the channel cross-interaction feature and the second weight; Determine the second cross-interaction result according to the local spatial feature and the first weight; Determine the sum of the first cross-interaction result and the second cross-interaction result as the attention weight; Obtain the output of the attention module according to the attention weight and the recombined feature.

[0009] In an embodiment of the present application, performing channel information cross-interaction on the recombined feature to obtain a channel cross-interaction feature includes: Perform adaptive pooling processing on the recombined feature to obtain a pooled feature map; Determine the channel cross-interaction feature according to the pooled feature map and the recombined feature.

[0010] In an embodiment of the present application, performing local spatial feature extraction on the recombined feature to obtain a local spatial feature includes: Perform local spatial feature extraction on the recombined feature by using convolution processing to obtain the local spatial feature.

[0011] In an embodiment of the present application, the feature parameters of the training sample include the cumulative number of scans, the initial value of the tube current, the maximum value of the tube current, the nominal value of the tube current, the final dose value, the initial value of the filament current, the final value of the filament current, the nominal value of the filament current, the motor drive current, and the true tube voltage.

[0012] To achieve the above object and other related objects, the present application provides a tube life prediction device, including: A data acquisition module, configured to acquire a training sample data set, where the training sample data set includes a plurality of training samples, each training sample corresponds to a training label, the training label represents the remaining life of the tube corresponding to the training sample, and the training sample includes a time series corresponding to a plurality of feature parameters of the tube; A first feature extraction module, configured to input the training sample into a multi-scale feature extraction network of an initial prediction model, and extract a plurality of time features from the time features of the training sample through the multi-scale feature extraction network; A second feature extraction module, configured to input the training sample into a two-dimensional convolutional network of the initial prediction model, and extract spatial features from the spatial features of the training sample through the two-dimensional convolutional network; A feature fusion module, configured to input a plurality of the time features and the spatial features into a feature fusion layer of the initial prediction model to obtain a fused feature; An output determination module, configured to input the fusion feature into a dual-branch group attention module of the initial prediction model, determine attention weights and reorganized features through the dual-branch group attention module, and determine an output of the attention module according to the attention weights and the reorganized features; A model training module, configured to input the output of the attention module into a fully connected network of the initial prediction model to obtain a life prediction label, and iteratively train the initial prediction model according to the training label and the life prediction label, and use the iteratively trained initial prediction model as a target prediction model for predicting the life of a to-be-predicted tube.

[0013] To achieve the above object and other related objects, the present application further provides an electronic device, where the electronic device includes: One or more processors; A memory for storing program codes executable by the processor; Wherein, the processor is configured to execute the program codes to implement the above tube life prediction method.

[0014] To achieve the above object and other related objects, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer is enabled to execute the foregoing one or more tube life prediction methods.

[0015] As described above, a tube life prediction method, device, device, and storage medium provided by the present application have the following beneficial effects: In a tube life prediction method of the present application, the method can well mine the degradation information of data in the time dimension through a multi-scale feature extraction network, and can mine the degradation information of data in the spatial dimension through a two-dimensional convolutional network, so as to realize data mining in multiple dimensions and provide more comprehensive degradation information. A dual-branch group attention module is used to further extract key features from the fusion feature. The module adopts a multi-scale parallel structure to capture local and long-term dependencies in the data, so that the dual-branch group attention module can learn richer feature representations. Using the trained target prediction model to predict the life of the tube can achieve the effect of improving the prediction accuracy.

[0016] It should be understood that the above general description and subsequent detailed description are only exemplary and explanatory, and cannot limit the present application. Description of the Drawings

[0017] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the drawings: Figure 1 is a flowchart of a tube life prediction method shown in an exemplary embodiment of the present application; Figure 2 is a schematic diagram of sliding window processing shown in an exemplary embodiment of the present application; Figure 3 is a schematic structural diagram of a temporal convolutional network provided in an embodiment of the present application; Figure 4 is a schematic structural diagram of a double-branch group attention module shown in an exemplary embodiment of the present application; Figure 5 is a flowchart of determining the output of the attention module shown in an exemplary embodiment of the present application; Figure 6 is a schematic diagram of the model architecture of an initial prediction model shown in an embodiment of the present application; Figure 7 is a structural block diagram of a tube life prediction device shown in an exemplary embodiment of the present application. Detailed implementation manners

[0018] The following will describe the implementation manners of the present application with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for explaining the present application, rather than for limiting the protection scope of the present application.

[0019] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present application in a schematic manner. Therefore, only the components related to the present application are shown in the diagrams, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, number, and ratio of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0020] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present application. However, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present application difficult to understand.

[0021] Please refer to Figure 1 , Figure 1 which is a flowchart of a tube life prediction method shown in an exemplary embodiment of the present application. Referring to Figure 1 it can be seen that the tube life prediction method may include: Step S110, obtaining a training sample data set.

[0022] Among them, the training sample data set includes a plurality of training samples, each training sample corresponds to a training label, the training label represents the remaining life of the tube corresponding to the training sample, and each training sample includes a time series corresponding to a plurality of characteristic parameters of the tube.

[0023] In an embodiment of the present application, a training sample data set can be obtained. The training sample data set includes a plurality of training samples, and each training sample includes a time series corresponding to a plurality of characteristic parameters of the tube. The training sample is composed of a series of time points, and each time point corresponds to a plurality of characteristic parameters of the tube. In the embodiment of the present application, the tube represents a CT tube. The CT (Computed Tomography) tube is one of the core components of a CT machine, and it is responsible for generating X-rays. These X-rays pass through the human body or an object and are captured by a detector and converted into an image. The quality and performance of the CT tube are directly related to the imaging quality, the reliability of the device, and the radiation safety of the patient.

[0024] In an embodiment, the characteristic parameters of the training sample include the cumulative number of scans, the initial tube current value, the maximum tube current value, the nominal tube current value, the final dose value, the initial filament current value, the final filament current value, the nominal filament current value, the motor drive current, and the actual tube voltage.

[0025] In a possible implementation manner, the process of obtaining the training sample data set may include: Step 1: During the operation of the CT tube, operating data of 33 parameters are collected. The operating data includes full life cycle data, and there are four groups of operating data for different tubes. When analyzing the operating data of the CT tube's full life cycle, it is first necessary to preprocess the data to ensure the accuracy and reliability of the analysis results. Specifically, it is necessary to identify and process blank values and singular values in the dataset. Blank values may be caused by mistakes in the data collection process or equipment failures, while singular values may be caused by measurement errors or data entry errors. By deleting these outliers, the noise in the data can be reduced, and the accuracy of subsequent analysis can be improved. Subsequently, feature selection is carried out based on expert knowledge, and 10 parameters such as the cumulative scan count, initial tube current value, maximum tube current value, nominal tube current value, final dose value, initial filament current value, final filament current value, nominal filament current value, motor drive current, and true tube voltage are selected as feature parameters, and multiple datasets can be obtained , represents the th tube, represents the th dataset corresponding to the th tube, represents 10 feature parameters.

[0026] Step 2: During the process of making labels for the CT tube operating data, an inverse order label strategy based on time series is adopted. This strategy first sorts each group of data according to the time series to ensure the timeliness of the data. Subsequently, the generation of labels follows the rule of incrementing by day from the last record in the dataset according to the timestamp until the first record, where the label value of the last record is set to 0. The last record is the record at the end of the tube's life. Each record includes the values of 10 feature parameters at the time point, and labels are obtained, where represents the th day. This method ensures the consistency and traceability of the label values of each dataset in the time series. In addition, in order to highlight the data degradation characteristics when the tube's life is about to end, all label values greater than 15 can be uniformly adjusted to 15 to highlight the data degradation characteristics when the life is about to end. Through this label generation method, a standardized and ordered label system can be provided for subsequent data analysis and model training. The labels are added to the data at each time point in the four datasets cleaned in Step 1.

[0027] Step 3: Data standardization. In order to eliminate the problem of different dimensions among different operating parameters, the embodiments of the present application use max-min standardization to perform normalization operations on the data cleaned in Step 1. The four standardized datasets are represented as , and its calculation process is as follows: ; Among them represents one of the ten characteristic parameters, represents the th data in the dataset. represents the minimum value of the th characteristic parameter, represents the maximum value of the th characteristic parameter, represents the standardized data. Each data corresponds to a time point (i.e., a timestamp).

[0028] Step 4: Sliding window processing and data denoising. Please refer to Figure 2 , which is a schematic diagram of the sliding window processing shown in an exemplary embodiment of the present application. Statistically calculate the total data length of 4 datasets , the total number of time points in 4 datasets (when the time points are divided by days, the total number of days in 4 datasets can be statistically calculated), the sliding window width , the step size ; for example, the total data length in 4 datasets , the total number of time points in 4 datasets , the sliding window width , the step size . The window data after sliding window processing for each dataset is: ; Among them, represents the value of the th characteristic parameter at the th moment.

[0029] The data in the window data has large fluctuations and cannot highlight the degradation trend existing in the data. The window data is denoised by data downsampling and data smoothing operations. The sampling interval of data downsampling , and the window data after downsampling is as follows: ; The data smoothing operation uses exponential moving average (EMA), and its iterative formula is as follows: ; Among them, is the th smoothed data of the th characteristic parameter in the window after downsampling, is the th data of the th characteristic parameter in the window after downsampling, the smoothing factor The calculation formula is as follows: is the th data of the th characteristic parameter in the window after downsampling, the smoothing factor is the th data of the th characteristic parameter in the window after downsampling, the smoothing factor is the th data of the th characteristic parameter in the window after downsampling, the smoothing factor The calculation formula is as follows: ; Considering the long-term and local characteristics of data degradation, Denote the smoothing window length. In the embodiments of the present application . The smoothed dataset is denoted as .

[0030] It should be noted that the exponential moving average smoothing technique (EMA) is used to denoise the data. By weighting the most recent points, it makes them more influential on future data points, while the influence on historical data points gradually weakens. When the data changes drastically, EMA can quickly respond and adjust its value, thus following the trend of the data, rather than lagging behind the data like the traditional moving average.

[0031] Step 5: Dataset division. Divide the 4 datasets after smoothing and denoising in Step 4 into a training set and a test set. Take the data of three tubes as the training set data, that is, the training sample dataset, namely , and the data of another tube as the test set .

[0032] It should be noted that each step of the embodiments of the present application can be executed by devices such as a server, a server cluster, a terminal, etc.

[0033] Step S120, input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple temporal features from the temporal features of the training samples through the multi-scale feature extraction network.

[0034] In an embodiment of the present application, the training samples can be input into the multi-scale feature extraction network of the initial prediction model to obtain multiple temporal features. Three temporal features can be extracted.

[0035] In one embodiment, the multi-scale feature extraction network includes multiple temporal convolutional networks, and the convolutional kernels of each temporal convolutional network are different. Multiple temporal features are extracted from the temporal features of the training samples through the multi-scale feature extraction network, including: Respectively extract the temporal features of the training samples through each temporal convolutional network to obtain multiple temporal features.

[0036] Input the training samples into the multi-scale multi-dimensional feature extraction network to obtain multiple temporal features of different scales. The multi-scale feature extraction network includes multiple temporal convolutional networks (Temporal Convolutional Network, TCN). The core components of TCN include causal convolution (Causal Convolution) and dilated convolution (Dilated Convolution). These two mechanisms enable TCN to expand the receptive field without sacrificing the sequence length. Please refer toFigure 3 , which is a schematic structural diagram of a temporal convolutional network provided by an embodiment of the present application. A TCN is usually composed of multiple causal convolutional layers, and these layers may have different dilation rates to capture patterns at different time scales. Residual connections can be added between each layer to facilitate the flow of gradients and avoid the problem of gradient vanishing in the training of deep networks.

[0037] Exemplarily, multi-scale temporal feature representation: ; ; ; Among them, indicates that the TCN branches with different-sized convolutional kernels are used to extract temporal features at different scales, represents the first temporal feature, represents the second temporal feature, is the third temporal feature. The size of the convolutional kernel in the TCN branch can be selected by the operator according to the actual situation. Exemplarily, the sizes of the convolutional kernels can be 2, 3, and 5 respectively. Figure 3 The convolutional kernel in the temporal convolutional network shown is 3.

[0038] It should be noted that the multi-scale feature extraction network can well mine the degradation information of data in the time dimension. Compared with the traditional time series prediction network that only focuses on single-dimensional or single-scale features, the method provided by the embodiment of the present application can perform data mining in multiple dimensions and provide more comprehensive degradation information.

[0039] Step S130, input the training samples into the two-dimensional convolutional network of the initial prediction model, and extract the spatial features of the training samples through the two-dimensional convolutional network.

[0040] In an embodiment of the present application, the training samples can be input into the two-dimensional convolutional network of the initial prediction model to obtain spatial features.

[0041] Exemplarily, the spatial feature representation: ; Among them, represents the two-dimensional convolutional network, represents the spatial feature.

[0042] It should be noted that the two-dimensional convolutional network can well mine the degradation information of data in the spatial dimension.

[0043] Step S140, input multiple temporal features and spatial features into the feature fusion layer of the initial prediction model to obtain the fused feature.

[0044] In one embodiment of the present application, multiple temporal features and spatial features can be input into the feature fusion layer of the initial prediction model to obtain fused features.

[0045] In one embodiment, inputting multiple temporal features and spatial features into the feature fusion layer of the initial prediction model to obtain fused features includes: splicing and fusing the multiple temporal features and spatial features through the feature fusion layer to obtain fused features.

[0046] Exemplarily, when fusing the extracted multi-scale temporal features and spatial features, the present application embodiment adopts a splicing fusion strategy, and the expression is as follows: ; where represents the fused feature.

[0047] Step S150: Input the fused feature into the dual-branch group attention module of the initial prediction model, determine the attention weights and the reorganized feature through the dual-branch group attention module, and determine the output of the attention module according to the attention weights and the reorganized feature.

[0048] In one embodiment of the present application, the fused feature can be input into the dual-branch group attention module of the initial prediction model, the attention weights and the reorganized feature are determined according to the fused feature, and then the output of the attention module is determined according to the attention weights and the reorganized feature.

[0049] Exemplarily, the dual-branch group attention adopts a multi-scale parallel structure for capturing local and long-term dependencies in the data. This design enables the model to learn richer feature representations at different scales. At the same time, the channels are learned in batches to avoid information loss caused by channel dimensionality reduction. Taking the fused feature (feature map ) as the input, represents the batch processing size, represents the number of channels, represents the width of the feature map. Then the output of the attention module can be expressed as: ; where represents the dual-branch group attention module, represents the output of the attention module.

[0050] Please refer to Figure 4 , which is a schematic structural diagram of the dual-branch group attention module shown in an exemplary embodiment of the present application. Please refer to Figure 5 , which is a flowchart of determining the output of the attention module shown in an exemplary embodiment of the present application.

[0051] In one embodiment, step S150 inputs the fusion features into the dual-branch group attention module of the initial prediction model. The process of determining the attention weights and the recombined features through the dual-branch group attention module and determining the output of the attention module based on the attention weights and the recombined features includes: Step S151: Group the fusion features according to the preset number of groups and the number of channels of the fusion features to obtain a plurality of grouped features, and obtain the recombined features based on the plurality of grouped features.

[0052] Exemplarily, divide the fusion features into groups, then the recombined features are , representing the preset number of groups.

[0053] Step S152: Perform channel information interaction on the recombined features to obtain channel interaction features.

[0054] In one embodiment, performing channel information interaction on the recombined features to obtain channel interaction features includes: performing adaptive pooling on the recombined features to obtain a pooled feature map; determining the channel interaction features based on the pooled feature map and the recombined features.

[0055] Exemplarily, the channel information interaction is implemented through 1*1 convolution to achieve interaction between channels, so as to enhance the information flow between different channels. Enabling the model to pay attention to the dependency relationships existing between different feature parameters. The recombined features pass through adaptive pooling, 1×1 convolution, activation function, and group normalization to obtain channel interaction features: ; where represents the pooled feature map, represents the adaptive pooling operation; ; where represents the channel interaction features, represents group normalization, represents the 1*1 convolution operation, represents the Sigmoid activation function.

[0056] Step S153: Extract local spatial features from the recombined features to obtain local spatial features.

[0057] In one embodiment of the present application, convolutional processing can be used to extract local spatial features from the recombined features to obtain local spatial features.

[0058] Exemplarily, local spatial feature extraction extracts local spatial features through a convolution operation with a convolution kernel size of 3*1, focusing on the local region information in the width direction. By using a larger convolution kernel, the model can capture more complex patterns in the width direction. The local spatial feature extraction is as follows: ; where represents the local spatial feature, represents the 3*1 convolution operation.

[0059] Step S154, determine the first weight according to the channel cross feature and the second weight according to the local spatial feature.

[0060] In an embodiment of the present application, an adaptive pooling operation can be performed on the transposed matrix of the channel cross feature and then the Softmax activation function can be used to obtain the first weight. An adaptive pooling operation can be performed on the transposed matrix of the local spatial feature and then the Softmax activation function can be used to obtain the second weight.

[0061] Exemplarily, the first weight can be expressed as: ; where, represents the first weight, represents the transpose.

[0062] Exemplarily, the second weight can be expressed as: ; where, represents the second weight.

[0063] Step S155, determine the first cross result according to the channel cross feature and the second weight.

[0064] In an embodiment of the present application, the product of the channel cross feature and the second weight can be determined as the first cross result.

[0065] Exemplarily, the first cross result can be expressed as: ; where, represents the first cross result.

[0066] Step S156, determine the second cross result according to the local spatial feature and the first weight.

[0067] In an embodiment of the present application, the product of the local spatial feature and the first weight can be determined as the second cross result.

[0068] Exemplarily, the second cross result can be expressed as: ; Among them, represents the second cross-intersection result.

[0069] Step S157: Determine the sum of the first cross-intersection result and the second cross-intersection result as the attention weight.

[0070] In an embodiment of the present application, the sum of the first cross-intersection result and the second cross-intersection result can be determined as the attention weight.

[0071] Exemplarily, the attention weight can be expressed as: ; Among them, represents the attention weight.

[0072] Step S158: Obtain the output of the attention module according to the attention weight and the recombined features.

[0073] In an embodiment of the present application, the Sigmoid activation weight can be used and weighted input to obtain the output of the attention module.

[0074] Exemplarily, the output of the attention module can be expressed as: ; Among them, represents the output of the attention module.

[0075] It should be noted that the dual-branch group attention module is used to further extract key features from the fused features. This module adopts a multi-scale parallel structure to capture local and long-term dependencies in the data. This design allows the model to learn richer feature representations at different scales. At the same time, the module learns channels in batches, avoiding information loss caused by channel dimensionality reduction. The traditional attention mechanism does not clearly distinguish features at different scales and does not particularly emphasize the independent processing ability between channels.

[0076] Step S160: Input the output of the attention module into the fully connected network of the initial prediction model to obtain the life prediction label, and iteratively train the initial prediction model according to the training label and the life prediction label. Use the iteratively trained initial prediction model as the target prediction model for predicting the life of the to-be-predicted tube.

[0077] In one embodiment of the present application, the output of the attention module can be input into the fully connected network of the initial prediction model to obtain a life prediction label. The difference between the life prediction label and the training label can be measured according to the loss function, and the model parameters of the initial prediction model can be updated according to the difference and the gradient update strategy. When the number of iterations meets the preset number of iterations, a target prediction model is obtained. The operator can select an appropriate loss function, batch size, number of iterations, and gradient update strategy.

[0078] Exemplarily, the life prediction label can be expressed as: ; where represents the life prediction label, represents the fully connected network.

[0079] The loss function for iterative training can adopt a piecewise weighted loss function, and the piecewise weighted loss function can include: ; where represents the number of training labels, represents the training label.

[0080] It should be noted that by adopting the piecewise weighted loss function and using different weights for the loss functions in different stages, the penalty coefficient is smaller in the stable stage and larger in the degradation stage, so that the model pays more attention to the data in the degradation stage.

[0081] The trained target prediction model can be tested using a test set, and the mean square error (MSE) can be used as an evaluation index. Its calculation formula is as follows: .

[0082] It should be noted that when using the target prediction model to predict the life of the tube to be predicted, the data to be predicted of the tube to be predicted can be obtained, and the data to be predicted is input into the target prediction model to obtain the remaining life of the tube.

[0083] Please refer to Figure 6 , which is a schematic diagram of the model architecture of the initial prediction model shown in an embodiment of the present application.

[0084] Figure 7 is a block diagram of a tube life prediction device shown in an exemplary embodiment of the present application. As Figure 7 shown, the exemplary tube life prediction device 700 includes: A data acquisition module 710, configured to acquire a training sample data set. The training sample data set includes a plurality of training samples, each training sample corresponding to a training label, where the training label represents the remaining life of the tube corresponding to the training sample, and the training sample includes a time series corresponding to a plurality of characteristic parameters of the tube.

[0085] A first feature extraction module 720, configured to input the training sample into a multi-scale feature extraction network of an initial prediction model, and extract time features of the training sample through the multi-scale feature extraction network to obtain a plurality of time features.

[0086] A second feature extraction module 730, configured to input the training sample into a two-dimensional convolutional network of the initial prediction model, and extract spatial features of the training sample through the two-dimensional convolutional network to obtain spatial features.

[0087] A feature fusion module 740, configured to input the plurality of time features and spatial features into a feature fusion layer of the initial prediction model to obtain fused features.

[0088] An output determination module 750, configured to input the fused features into a dual-branch group attention module of the initial prediction model, determine attention weights and reorganized features through the dual-branch group attention module, and determine an output of the attention module according to the attention weights and the reorganized features.

[0089] A model training module 760, configured to input the output of the attention module into a fully connected network of the initial prediction model to obtain a life prediction label, and iteratively train the initial prediction model according to the training label and the life prediction label, and use the iteratively trained initial prediction model as a target prediction model for predicting the life of the tube to be predicted.

[0090] It should be noted that the tube life prediction device provided in the above embodiment and the tube life prediction method provided in the above embodiment belong to the same concept. The specific manners in which each module and unit perform operations have been described in detail in the method embodiment, and will not be repeated here. In practical applications, the tube life prediction device provided in the above embodiment may, according to needs, allocate the above functions to different functional modules, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above. This is not limited here either.

[0091] An embodiment of the present application further provides an electronic device, including: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the tube life prediction method provided in each of the above embodiments.

[0092] Another aspect of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer is enabled to execute the tube life prediction method provided in each of the above embodiments. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist alone without being assembled into the electronic device.

[0093] Another aspect of the present application also provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the tube life prediction method provided in each of the above embodiments.

[0094] In the embodiments of the present application, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance. The terms "comprising" and "including" mentioned throughout the specification and claims are open-ended terms and should be interpreted as "including but not limited to".

[0095] The above embodiments are only used to exemplarily illustrate the principles and effects of the present application, rather than to limit the present application. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed in the present application should still be covered by the claims of the present application.

Claims

1. A method for predicting tube life, characterized in that: include: Acquire a training sample data set, the training sample data set includes a plurality of training samples, each training sample corresponds to a training label, the training label represents the remaining life of the tube corresponding to the training sample, and the training sample includes a time series corresponding to a plurality of characteristic parameters of the tube; Inputting the training sample into a multi-scale feature extraction network of an initial prediction model, and extracting the time features of the training sample through the multi-scale feature extraction network to obtain multiple time features; Inputting the training sample into the two-dimensional convolutional network of the initial prediction model, and extracting the spatial features of the training sample through the two-dimensional convolutional network to obtain spatial features; Inputting the plurality of the time features and the spatial features into a feature fusion layer of the initial prediction model to obtain fusion features; Inputting the fusion feature into the dual-branch group attention module of the initial prediction model, determining the attention weight and the reorganization feature through the dual-branch group attention module, and determining the attention module output according to the attention weight and the reorganization feature; The output of the attention module is input into the fully connected network of the initial prediction model to obtain a life prediction label, and the initial prediction model is iteratively trained based on the training label and the life prediction label, and the initial prediction model after the iterative training is used as the target prediction model for life prediction of the tube to be predicted.

2. The method for predicting tube life according to claim 1, characterized in that: The multi-scale feature extraction network includes a plurality of time convolution networks, each of which has a different convolution kernel. The time features of the training samples are extracted by the multi-scale feature extraction network to obtain a plurality of time features, including: The time features of the training samples are extracted respectively through each time convolution network to obtain multiple time features.

3. The method for predicting tube life according to claim 1, characterized in that: Inputting the plurality of the time features and the spatial features into the feature fusion layer of the initial prediction model to obtain fusion features includes: The feature fusion layer is used to concatenate and fuse a plurality of the temporal features and the spatial features to obtain the fused feature.

4. The method for predicting tube life according to claim 1, characterized in that: Inputting the fusion feature into the dual-branch group attention module of the initial prediction model, determining the attention weight and the reorganization feature through the dual-branch group attention module, and determining the attention module output according to the attention weight and the reorganization feature, including: The fused features are grouped according to a preset number of groups and the number of channels of the fused features to obtain a plurality of grouped features, and a recombined feature is obtained according to the plurality of grouped features; Performing channel information interaction on the recombined features to obtain channel interaction features; Performing local spatial feature extraction on the recombined features to obtain local spatial features; Determine a first weight according to the channel interaction feature, and determine a second weight according to the local space feature; Determine a first interaction result according to the channel interaction feature and the second weight; Determine a second interaction result according to the local spatial feature and the first weight; Determine a sum of the first interaction result and the second interaction result as an attention weight; The attention module output is obtained according to the attention weight and the reorganized feature.

5. The method for predicting tube life according to claim 4, characterized in that: Performing channel information interaction on the recombined features to obtain channel interaction features includes: Performing adaptive pooling processing on the recombined features to obtain a pooling feature map; The channel interaction feature is determined according to the pooling feature map and the reorganization feature.

6. The method for predicting tube life according to claim 4, characterized in that: Performing local spatial feature extraction on the recombined features to obtain local spatial features includes: The local spatial feature is extracted from the reorganized feature by using convolution processing to obtain the local spatial feature.

7. The method for predicting the life of a tube according to any one of claims 1 to 6, characterized in that: The characteristic parameters of the training samples include the cumulative number of scans, the initial value of the tube current, the maximum value of the tube current, the nominal value of the tube current, the final value of the dose, the initial value of the filament current, the final value of the filament current, the nominal value of the filament current, the motor drive current, and the real tube voltage.

8. A device for predicting the life of a tube, characterized in that: include: A data acquisition module is used to acquire a training sample data set, wherein the training sample data set includes a plurality of training samples, each training sample corresponds to a training label, the training label represents the remaining life of the tube corresponding to the training sample, and the training sample includes a time series corresponding to a plurality of characteristic parameters of the tube; A first feature extraction module, configured to input the training sample into a multi-scale feature extraction network of an initial prediction model, and extract the time features of the training sample through the multi-scale feature extraction network to obtain a plurality of time features; A second feature extraction module, used for inputting the training sample into the two-dimensional convolutional network of the initial prediction model, and extracting the spatial features of the training sample through the two-dimensional convolutional network to obtain spatial features; A feature fusion module, used for inputting the plurality of temporal features and the spatial features into a feature fusion layer of the initial prediction model to obtain a fusion feature; An output determination module, used for inputting the fusion feature into the dual-branch group attention module of the initial prediction model, determining the attention weight and the reorganization feature through the dual-branch group attention module, and determining the attention module output according to the attention weight and the reorganization feature; A model training module is used to input the output of the attention module into the fully connected network of the initial prediction model to obtain a life prediction label, and iteratively train the initial prediction model based on the training label and the life prediction label, and use the initial prediction model after iterative training as the target prediction model for life prediction of the tube to be predicted.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; a memory for storing program code executable by the processor; Wherein, the processor is configured to execute the program code to implement the tube life prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method for predicting the life of a tube as claimed in any one of claims 1 to 7.

Images