Fault monitoring method and device for breathing heating pipeline

Abstract

The invention discloses a fault monitoring method and device for a breathing heating pipeline, and relates to the technical field of medical equipment. The method is applied to a heating breathing pipeline matched with a breathing machine and comprises the steps that the current corresponding heating current of the breathing heating pipeline is obtained, and the heating current is the current corresponding heating current of a heating wire of each heating pipeline unit in the breathing heating pipeline; the heating current is input into the trained fault monitoring model, the distribution probability of a monitoring result corresponding to the breathing heating pipeline is obtained, and the distribution probability of the monitoring result comprises the probability of fault occurrence and the probability of normal work; and determining a fault detection result corresponding to the breathing heating pipeline according to the distribution probability of the monitoring result and a preset probability processing rule. According to the method, the fault monitoring model is obtained through deep learning of historical data, fault detection and diagnosis can be automatically carried out based on the fault monitoring model, and the accuracy and efficiency of detection are greatly improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of medical equipment, and in particular to a fault monitoring method and device for a respiratory heating circuit. Background Art

[0002] Heated breathing circuits are commonly used to connect the air circuits between ventilators and anesthesia machines and masks or endotracheal tubes. The purpose is to prevent the generation of condensed water by heating the gas in the circuit, thereby reducing discomfort to the patient. In practical applications, fault identification of heated circuit systems is a complex and critical task. Existing technologies generally manually extract electrical signals when a heating circuit fails, and manually extract statistical results from the electrical signals: such as RMS values, or manually extract the frequency domain characteristics of the current from the electrical signals to obtain feature vectors, and then use data-driven models for classification or detection.

[0003] However, after a large number of practical applications, it is found that this artificially extracted feature is subjective and limited. To a certain extent, due to the differences between data sets, it is difficult to further improve the recognition accuracy. Therefore, there is an urgent need for a method that can quickly and accurately identify heating pipeline system faults. Summary of the invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical deficiencies.

[0005] On the one hand, an embodiment of the present application provides a method for monitoring a fault of a heating breathing circuit, which is applied to a heating breathing circuit matched with a ventilator, and the method includes:

[0006] Obtain the heating current currently corresponding to the breathing heating pipeline, where the heating current is the heating current currently corresponding to the heating wire of each heating pipeline unit in the breathing heating pipeline;

[0007] The heating current is input into the trained fault monitoring model to obtain the distribution probability of the monitoring results corresponding to the breathing heating pipeline, and the distribution probability of the monitoring results includes the probability of failure and the probability of normal operation;

[0008] According to the distribution probability of the monitoring results and the preset probability processing rules, the fault detection result corresponding to the breathing heating pipeline is determined.

[0009] Optionally, the fault monitoring model includes a convolution layer, a normalized activation module, an attention mechanism module, a residual module, a fully connected layer, a Dropout layer and a Softmax layer, and the normalized activation module includes a first Batch Normalization and a first ReLU activation function.

[0010] Optionally, the convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer, and the first convolution layer, the second convolution layer and the third convolution layer are all connected with a normalized activation module, the input of the first convolution layer is the heating current, the input of the second convolution layer is the output of the normalized activation module connected after the first convolution layer, the input of the third convolution layer is the output of the normalized activation module connected after the second convolution layer, and the output of the normalized activation module connected after the third convolution layer serves as the input of the attention mechanism module.

[0011] Optionally, the residual module includes two residual blocks, each residual block includes a second BatchNormalization, a second ReLU activation function, a weight layer and a third Batch Normalization connected in sequence, the residual module input is the output of the attention mechanism module, and the residual module output is the input of the fully connected layer.

[0012] Optionally, the attention mechanism module includes a channel attention module and a spatial attention module. The output of the attention mechanism module is obtained in the following way:

[0013] Obtain the output features of the normalized activation module connected after the third convolutional layer, and input the output features into the spatial attention module to obtain the first feature;

[0014] Multiply the output feature and the first feature element by element to obtain the product feature;

[0015] The product feature is normalized and the normalized product feature is input into the channel attention module to obtain the second feature;

[0016] The second feature and the normalized product feature are multiplied element by element to obtain the output of the attention mechanism module.

[0017] Optionally, the fully connected layer includes a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence, the input of the first fully connected layer is the output of the residual module, the output of the third fully connected layer is the input of the Dropout layer, and the output of the Dropout layer is the input of the Softmax layer.

[0018] Optionally, the fault detection model is trained by:

[0019] Acquire a training data set, the training data set includes a training sample set and a test sample set, the training data set includes sample pairs, each sample pair includes sample data and a corresponding sample label, wherein the sample data includes a heating current of a heating wire of each heating circuit unit in a breathing heating circuit, and the sample label represents a labeling result of a fault detection result corresponding to the sample pair;

[0020] The initial neural network model is trained based on the training sample set until the loss function corresponding to the initial neural network model converges, thereby obtaining an initial fault monitoring model, wherein the input of the initial neural network model is the sample pair, the output is the fault detection result corresponding to the sample pair, and the value of the loss function represents the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair;

[0021] The initial fault monitoring model is tested and verified based on the test sample set. If the test verification result corresponding to the initial fault monitoring model does not meet the set verification requirements, the step of training the initial neural network model based on the training sample set is repeated until the test verification result corresponding to the initial fault monitoring model meets the set verification requirements.

[0022] Optionally, the inspection sample set includes a validation sample set and a test sample set, the sample data includes positive sample data and negative sample data, and obtaining a training data set includes:

[0023] Acquire at least one heating current of the breathing heating circuit during normal operation as at least one negative sample data;

[0024] Acquire at least one heating current of the breathing heating circuit when a fault occurs as at least one positive sample data;

[0025] At least one negative sample data and at least one positive sample data are randomly combined and divided according to the set quantity requirements to obtain a training sample set, a verification sample set and a test sample set respectively.

[0026] Optionally, the verification requirements are set to include a recall rate requirement and a precision rate requirement corresponding to each sample pair.

[0027] On the other hand, an embodiment of the present application provides a fault monitoring device for a breathing heating circuit, which is applied to a heating breathing circuit matched with a ventilator, and the device includes:

[0028] A current acquisition module is used to acquire the heating current currently corresponding to the breathing heating pipeline, where the heating current is the heating current currently corresponding to the heating wire of each heating pipeline unit in the breathing heating pipeline;

[0029] A probability acquisition module is used to input the heating current into the trained fault monitoring model to obtain the distribution probability of the monitoring results corresponding to the breathing heating pipeline, and the distribution probability of the monitoring results includes the probability of failure and the probability of normal operation;

[0030] The result determination module is used to determine the fault detection result corresponding to the breathing heating pipeline according to the distribution probability of the monitoring result and the preset probability processing rules.

[0031] On the other hand, an embodiment of the present application provides an electronic device, including a processor and a memory:

[0032] The memory is configured to store machine-readable instructions, which, when executed by the processor, cause the processor to perform any one of the methods for fault monitoring of a respiratory warming circuit.

[0033] The beneficial effects brought by the technical solution provided in the embodiments of the present application include at least:

[0034] In the embodiment of the present application, a fault monitoring model is obtained through deep learning of historical data, so that fault detection and diagnosis can be automatically performed based on the fault monitoring model, greatly improving the accuracy and efficiency of detection. In addition, it also has good scalability and adaptability, and can be applied to heating pipeline systems of different types and sizes. By continuously learning and updating the model, the method can adapt to system changes and new fault modes, ensuring continuous high efficiency in actual operation, so as to better control the temperature of the heating pipeline, improve the heating efficiency, reduce the heating power consumption of the ventilator and better prevent the generation of condensed water.

[0035] In the embodiment of the present application, since the fault monitoring model includes a residual module, that is, a residual learning mechanism is introduced, the gradient vanishing and gradient exploding problems in the deep neural network training can be effectively solved, so that the network can be trained deeper, and the weight layer and ReLU in the classic residual layer structure are reduced, the classic residual structure is simplified, the amount of calculation is reduced, and the model converges faster.

[0036] In an embodiment of the present application, the fault monitoring model also includes an attention mechanism module, and the attention mechanism module adopts a processing method that combines spatial attention and channel attention. This is an effective feature enhancement strategy that can improve the performance and generalization ability of the model without significantly increasing the computational cost.

[0037] In an embodiment of the present application, a Dropout layer is also introduced between the fully connected layer and the Softmax layer, which can randomly select a certain proportion of neurons in each training batch to temporarily invalidate their outputs, thereby reducing the network's dependence on specific neurons and preventing overfitting. This helps to improve the generalization ability of the model and enables the fault monitoring model to perform better on unseen data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1a A schematic diagram of a breathing warming circuit provided in an embodiment of the present application;

[0040] Figure 1b A schematic diagram of a flow chart of fault monitoring of a breathing heating circuit provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of the structure of a fault monitoring model provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of the structure of each residual block provided in an embodiment of the present application;

[0043] Figure 4 A schematic diagram of the mechanism of the attention mechanism module provided in the embodiment of the present application;

[0044] Figure 5 A schematic diagram of the structure of another fault monitoring model provided in an embodiment of the present application;

[0045] Figure 6 A schematic diagram of the structure of a fault monitoring device for a breathing heating circuit provided in an embodiment of the present application;

[0046] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present invention.

[0048] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0049] In order to make the objectives, technical solutions and advantages of the present application clearer, the key specific terms involved in the embodiments of the present application will be explained in conjunction with the accompanying drawings.

[0050] Heated breathing circuit: usually used to connect the air circuit between the ventilator and anesthesia machine and the mask or endotracheal tube. Its purpose is to prevent condensation by heating the gas in the tube, thereby reducing the discomfort to the patient. Figure 1a As shown, the breathing heating circuit is usually composed of multiple heating circuit units equipped with heating wires and temperature sensors. The heating wires of each heating circuit unit are independently connected to the temperature control unit. The temperature control unit for connecting the ventilator and the anesthetic agent is located below the breathing heating circuit.

[0051] The Central Processing Unit (CPU) is one of the main devices of a computer. Its main function is to interpret computer instructions and process data in computer software. In a broad sense, the CPU refers to a series of logical machines that can execute complex computer programs.

[0052] Feature engineering (English: feature engineering), also known as feature extraction (English: feature extraction) or feature discovery (English: feature discovery), is the process of extracting features (features, attributes, characteristics) from raw data using domain knowledge. Compared with providing only raw data to the machine learning process, its motivation is to use these additional features to improve the quality of the results of the machine learning process.

[0053] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0054] Specifically, Figure 1b As shown, the method can be applied to a heated breathing circuit matched with a ventilator, and the method may include:

[0055] Step S101, obtaining a heating current currently corresponding to the breathing heating pipeline, where the heating current is a heating current currently corresponding to a heating wire of each heating pipeline unit in the breathing heating pipeline.

[0056] Optionally, when monitoring whether the breathing heating circuit has a fault, the heating current currently corresponding to the heating wire of each heating circuit unit in the breathing heating circuit can be obtained, and then whether the breathing heating circuit has a fault can be further determined based on the obtained heating current. Among them, the method of obtaining the heating current can be selected according to actual needs, and the embodiment of the present application is not limited. For example, the heating current can be obtained based on an installed current sensor.

[0057] Step S102, inputting the heating current into the trained fault monitoring model to obtain the distribution probability of the monitoring result corresponding to the breathing heating pipeline, wherein the distribution probability of the monitoring result includes the probability of a fault and the probability of normal operation.

[0058] Optionally, after obtaining the heating current currently corresponding to the breathing heating circuit, the heating current can be input into the trained fault monitoring model. At this time, the fault monitoring model will output the distribution probability of the monitoring results corresponding to the breathing heating circuit, which includes the probability of failure of the breathing heating circuit and the probability of normal operation of the breathing heating circuit.

[0059] Step S103, determining the fault detection result corresponding to the breathing heating circuit according to the distribution probability of the monitoring result and the preset probability processing rule.

[0060] Furthermore, for the distribution probability of the obtained monitoring results, the obtained distribution probability can be compared with the preset probability processing rules to determine whether the breathing heating pipeline is currently faulty. Among them, the preset probability processing rules can be set by themselves according to the actual required accuracy, and the embodiments of the present application are not limited. For example, when the preset probability processing rules are set to the probability of failure being greater than the probability of normal operation, it is determined that a fault currently occurs. At this time, if the probability of failure in the distribution probability of the obtained monitoring results is greater than the probability of normal operation, it is determined that the breathing heating pipeline is currently faulty.

[0061] In the embodiment of the present application, a fault monitoring model is obtained through deep learning of historical data, so that fault detection and diagnosis can be performed automatically, greatly improving the accuracy and efficiency of detection. In addition, it has good scalability and adaptability, and can be applied to heating pipeline systems of different types and sizes. By continuously learning and updating the model, the method can adapt to system changes and new fault modes, ensuring continuous high efficiency in actual operation, so as to better control the temperature of the heating pipeline, improve the heating efficiency, reduce the heating power consumption of the ventilator and better prevent the generation of condensed water.

[0062] In an optional embodiment of the present application, the fault monitoring model includes a convolutional layer, a normalized activation module, an attention mechanism module, a residual module, a fully connected layer, a Dropout layer and a Softmax layer, and the normalized activation module includes a first BatchNormalization and a first ReLU activation function.

[0063] Optionally, the structure diagram of the fault monitoring model in the embodiment of the present application can be as follows: Figure 2 As shown, it can specifically include a convolutional layer, a normalized activation module, an attention mechanism module, a residual module, a fully connected layer, a Dropout layer and a Softmax layer, and the normalized activation module includes the first Batch Normalization and the first ReLU activation function. Among them, the input of the convolutional layer (Convolutional Layer) is the heating current (i.e., input), and is connected to the normalized activation module (i.e., BatchNormalization&ReLU), the output of the normalized activation module is connected to the attention mechanism module (i.e., Attention), the output of the attention mechanism module is connected to the residual module (Residual Block, also known as Residual Network), the output of the residual module is connected to the fully connected layer, and the Dropout layer is connected to the fully connected layer and the Softmax layer respectively.

[0064] In an optional embodiment of the present application, the convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer. The first convolution layer, the second convolution layer and the third convolution layer are all connected to a normalized activation module. The input of the first convolution layer is the heating current, the input of the second convolution layer is the output of the normalized activation module connected after the first convolution layer, the input of the third convolution layer is the output of the normalized activation module connected after the second convolution layer, and the output of the normalized activation module connected after the third convolution layer serves as the input of the attention mechanism module.

[0065] Optionally, the convolution layer may specifically include a first convolution layer, a second convolution layer, and a third convolution layer connected in sequence. Among them, the number of input channels of the first convolution layer (Conv1) is 3, the number of output channels is 64, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; the number of input channels of the second convolution layer (Conv2) is 64, the number of output channels is 128, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; the number of input channels of the third convolution layer (Conv3) is 128, the number of output channels is 256, the convolution kernel size is 3x3, the stride is 1, and the padding is 1.

[0066] Optionally, the first convolution layer, the second convolution layer and the third convolution layer are all connected with a normalization activation module (i.e., Batch Normalization & ReLU), and the input of the first convolution layer is the heating current, the input of the second convolution layer is the output of the normalization activation module connected after the first convolution layer, the input of the third convolution layer is the output of the normalization activation module connected after the second convolution layer, and the output of the normalization activation module connected after the third convolution layer is the input of the attention mechanism module.

[0067] In an optional embodiment of the present application, the residual module includes two residual blocks, each residual block includes a second Batch Normalization, a second ReLU activation function, a weight layer and a third Batch Normalization connected in sequence, the residual module input is the output of the attention mechanism module, and the residual module output is the input of the fully connected layer.

[0068] Optionally, the input of the residual module in the fault monitoring model is the output of the attention mechanism module, and the output is the input of the fully connected layer. The residual module includes two residual blocks, and the structure of each residual block can be specifically as follows: Figure 3 As shown, each residual block includes Batch Normalization (i.e. BN in the figure), ReLU activation function (i.e. ReLU in the figure), weight layer (i.e. weight in the figure) and the third Batch Normalization (i.e. BN in the figure) connected in sequence. The input of each residual block is X l , the output is X l+1 .

[0069] In the application embodiment, since the fault monitoring model includes a residual module, that is, a residual learning mechanism is introduced, the gradient vanishing and gradient exploding problems in deep neural network training can be effectively solved, so that the network can be trained deeper. The weight layer and ReLU in the classic residual layer structure are reduced, the classic residual structure is simplified, the amount of calculation is reduced, and the model converges faster. The specific benefits may include the following:

[0070] 1. Alleviate the vanishing gradient problem: Through short-circuit connections, the gradient can be passed directly from the output to the input, thereby alleviating the vanishing gradient problem and enabling the network to be trained deeper.

[0071] 2. Improve training speed: Since the gradient can be transferred directly, the network is easier to converge, thereby improving the training speed.

[0072] 3. Improve the generalization ability of the model: The residual block can learn the direct mapping relationship between input and output, so that the model can better generalize to unseen data.

[0073] 4. Reduce dependence on initialization: Since the residual block improves the stability of training, it can reduce the network’s dependence on weight initialization.

[0074] 5. Support deeper network structures: Residual blocks make it possible to build deeper network structures, thereby improving the feature extraction capabilities of the model.

[0075] 6. Improve the robustness of the model: The residual block can learn the direct mapping relationship between input and output, making the model more robust to slight changes in the input data.

[0076] 7. Reduce overfitting: The residual block can improve the generalization ability of the model, thereby reducing the risk of overfitting.

[0077] In an optional embodiment of the present application, the attention mechanism module includes a channel attention module and a spatial attention module, and the output of the attention mechanism module is obtained in the following manner:

[0078] Obtain the output features of the normalized activation module connected after the third convolutional layer, and input the output features into the spatial attention module to obtain the first feature;

[0079] Multiply the output feature and the first feature element by element to obtain the product feature;

[0080] The product feature is normalized and the normalized product feature is input into the channel attention module to obtain the second feature;

[0081] Multiply the second feature and the normalized product feature element by element to get the output of the attention mechanism module.

[0082] Optionally, the attention mechanism module may include a channel attention module (Channel Attention) and a spatial attention module (Spatial Attention). Among them, channel attention is used to pay attention to the importance of different channels, while spatial attention is used to pay attention to the importance of different spatial positions. At this time, attention can be effectively focused on information closely related to the heating current failure of the heating wire, while less attention resources are allocated to information that is weakly related to the heating current of the heating wire. Different attention can be paid to different frequency bands of the current signal in different scenarios to improve the detection sensitivity of the heating current failure problem of the heating wire.

[0083] Optional, such as Figure 4 As shown in FIG. 1 , when the output feature of the normalized activation module connected after the third convolutional layer is input into the spatial attention module as the input feature (i.e., input feature), the output of the spatial attention module obtains the first feature, and then the input feature of the spatial attention module (i.e., the original feature) and the first feature are element-wise multiplied to obtain the product feature; the product feature is further normalized, such as Softmax normalization (i.e., Figure 4 Normalization with softmax in , and input the normalized product feature into the channel attention module to obtain the second feature, and then multiply the second feature and the normalized product feature element by element to obtain the output of the attention mechanism module (i.e., output feature). At this time, the output feature of the attention mechanism module is an enhanced new feature representation.

[0084] In the embodiment of the present application, the processing method of combining spatial attention and channel attention is an effective feature enhancement strategy, which can improve the performance and generalization ability of the model without significantly increasing the computational cost, and can specifically bring the following benefits:

[0085] 1. Enhanced feature representation: Through the attention mechanism, the model can focus more on important features and suppress unimportant parts, thereby improving the representation ability of features.

[0086] 2. Improve the generalization ability of the model: The attention module can help the model learn more robust features, reduce dependence on specific features, and improve the generalization ability of the model under different data distributions.

[0087] 3. Reduce computational burden: Through the attention mechanism, the model can use computing resources more efficiently and focus on the most important features for the task, thereby reducing unnecessary calculations to a certain extent.

[0088] 4. Improved gradient flow: The attention module can provide clearer gradient signals, which helps to better propagate gradients during training and thus accelerate the convergence of the model.

[0089] 5. Improve model performance: In many model training, the introduction of attention mechanism has been shown to significantly improve model performance.

[0090] 6. Strong adaptability: The combination of spatial attention and channel attention enables the model to pay attention to both the spatial distribution and channel importance of features, which enables the model to better adapt to diverse task requirements.

[0091] 7. Easy to integrate: The attention module can usually be easily integrated into existing convolutional neural networks without large-scale modifications to the network structure.

[0092] 8. Interpretability: The attention mechanism can provide a certain degree of explainability for the model's decisions. By visualizing the attention weights, we can understand which areas or features the model focuses on when making decisions.

[0093] In an optional embodiment of the present application, the fully connected layer includes a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence, the input of the first fully connected layer is the output of the residual module, the output of the third fully connected layer is the input of the Dropout layer, and the output of the Dropout layer is the input of the Softmax layer.

[0094] Optionally, the fully connected layers in the fault monitoring module include a first fully connected layer, a second fully connected layer, and a third fully connected layer connected in sequence. The input dimension of the first fully connected layer (FC1) is 256*8*8, and the output dimension is 512. The input dimension of the second fully connected layer (FC2) is 512, and the output dimension is 256. The input dimension of the third fully connected layer (FC3) is 256, and the output dimension is 2 (for binary classification of the Softmax layer). Among them, the input of the first fully connected layer is the output of the residual module, the output of the third fully connected layer is the input of the Dropout layer, and the output of the Dropout layer is the input of the Softmax layer

[0095] The fault monitoring module in the embodiment of the present application introduces a Dropout layer between the fully connected layer and the Softmax layer. Its main function is to randomly "discard" (i.e., set to zero) a part of the neurons in the network during the training process of the model. Specifically, by introducing the Dropout layer between the fully connected layer and the Softmax layer, a certain proportion of neurons can be randomly selected in each training batch to temporarily invalidate their outputs, thereby reducing the network's dependence on specific neurons and preventing the occurrence of overfitting. This approach forces the network to learn more robust feature representations because it cannot rely on the existence of any single neuron to make predictions. In this way, Dropout helps to improve the generalization ability of the model, allowing it to perform better on new, unseen data.

[0096] In an optional embodiment of the present application, the fault monitoring model is trained in the following manner:

[0097] Acquire a training data set, the training data set includes a training sample set and a test sample set, the training data set includes sample pairs, each sample pair includes sample data and a corresponding sample label, wherein the sample data includes a heating current of a heating wire of each heating circuit unit in a breathing heating circuit, and the sample label represents a labeling result of a fault detection result corresponding to the sample pair;

[0098] The initial neural network model is trained based on the training sample set until the loss function corresponding to the initial neural network model converges, thereby obtaining an initial fault monitoring model, wherein the input of the initial neural network model is the sample pair, the output is the fault detection result corresponding to the sample pair, and the value of the loss function represents the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair;

[0099] The initial fault monitoring model is tested and verified based on the test sample set. If the test verification result corresponding to the initial fault monitoring model does not meet the set verification requirements, the step of training the initial neural network model based on the training sample set is repeated until the test verification result corresponding to the initial fault monitoring model meets the set verification requirements.

[0100] Optionally, the present application can obtain a training sample set in a training data set, the training sample set including sample pairs, each sample pair including sample data and a corresponding sample label, and the sample data refers to the heating current of the heating wire of each heating circuit unit in the respiratory heating circuit, and the corresponding sample label is the annotation result of the fault detection result corresponding to the input sample data.

[0101] Furthermore, each sample pair can be input into the initial neural network model, and the initial neural network model can output the fault detection result corresponding to each sample pair. Then, the value of the loss function corresponding to the model can be determined based on the fault detection result of each sample pair output and the labeling result corresponding to each sample pair. If the determined loss function value does not converge, the network parameters of the initial neural network model can be adjusted, and then each sample pair can be input into the initial neural network model after adjusting the parameters again, and the value of the loss function can be determined again based on the fault detection result output and the labeling result corresponding to the sample pair. If it still does not converge, the network parameters of the initial neural network model are adjusted again until the corresponding loss function value converges. Among them, the value of the loss function corresponding to the model represents the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair. When the loss function converges, it indicates that the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair meets the requirements, that is, the fault detection result of the sample pair output by the model is close to the labeling result of the sample pair.

[0102] Furthermore, when the loss function converges, the initial fault monitoring model can be obtained. At this time, the initial fault monitoring model can be tested and verified based on the test sample set in the training data set. If the test verification result corresponding to the initial fault monitoring model does not meet the set verification requirements, the step of training the initial neural network model based on the training sample set is repeated until the test verification result corresponding to the initial fault monitoring model meets the set verification requirements.

[0103] In an optional embodiment of the present application, the verification requirements are set to include the recall rate requirements and accuracy rate requirements corresponding to each sample pair.

[0104] Optionally, the verification requirements may include the recall rate requirements and accuracy requirements corresponding to each sample pair, that is, when the recall rate requirements and accuracy requirements corresponding to each sample pair meet the requirements, the model obtained at this time can be used as a fault monitoring model. Among them, the recall rate refers to the proportion of samples correctly classified by the model to the total number of samples, and the recall rate (Recall) refers to the proportion of false positives in samples that are actually positive.

[0105] In an optional embodiment of the present application, the inspection sample set includes a verification sample set and a test sample set, the sample data includes positive sample data and negative sample data, and obtaining a training data set includes:

[0106] Acquire at least one heating current of the breathing heating circuit during normal operation as at least one negative sample data;

[0107] Acquire at least one heating current of the breathing heating circuit when a fault occurs as at least one positive sample data;

[0108] At least one negative sample data and at least one positive sample data are randomly combined and divided according to the set quantity requirements to obtain a training sample set, a verification sample set and a test sample set respectively.

[0109] Optionally, the inspection sample set includes a verification sample set and a test sample set, and the sample data includes positive sample data and negative sample data. In practice, at least one heating current of the breathing heating circuit in normal operation can be obtained as at least one negative sample data, and at least one heating current of the breathing heating circuit in the event of a fault can be obtained as at least one positive sample data, and then the obtained at least one negative sample data and at least one positive sample data are randomly combined and divided into a training sample set, a verification sample set, and a test sample set. For example, the obtained positive sample data and negative sample data sets are a total of 1000, and at this time the data set can be divided into 800 sample data as a training sample set, 100 sample data as a verification sample set, and 100 sample data as a test sample set. In the present application, the training data set is obtained by randomly combining all the data, which can simulate complex situations in actual applications and help improve the generalization ability of the model.

[0110] Optionally, since heating pipeline fault monitoring is a binary classification problem, the sample data of the fault is the positive class, and the sample data of the normal working is the negative class. In the binary classification problem, the samples can be classified by comparing the predicted results and the true labels of the sample data, that is, they can be divided into true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN). For example, the English abbreviations of their names can be used to represent the corresponding sample numbers. At this time, the confusion matrix of the classification results can be further obtained as shown in the following table.

[0111]

[0112] Furthermore, after obtaining the confusion matrix corresponding to the sample data, the test sample set can be used regularly to evaluate the performance of the model. First, the performance indicators on the validation set are calculated, such as accuracy, precision, recall, F1 score, etc. Then, according to the performance indicators on the validation set, the hyperparameters of the model (such as learning rate, regularization parameter, number of network layers, etc.) are adjusted. During this period, if the performance indicators on the validation set do not improve within a certain round, the training can be stopped in advance to prevent overfitting. Furthermore, after the model training and verification are completed, the final model is evaluated using the test sample set. The performance indicators on the test sample set are usually regarded as the performance estimation of the model in practical applications. When the estimation ability meets the requirements, the final fault monitoring model can be obtained.

[0113] In an embodiment of the present application, after the fault monitoring model is trained, it will be verified and evaluated based on the test sample set to obtain the final fault monitoring model, which can help monitor the performance of the model on unseen data, thereby preventing the model from overfitting the training data, and through the performance indicators on the verification set, the hyperparameters of the model can be effectively adjusted to find the optimal model configuration.

[0114] Optionally, in order to better understand the method provided in the embodiment of the present application, the following is based on Figure 5 The fault monitoring model shown in the figure is trained and tested. The test data is input into the convolution layer (i.e. Conv) as the input feature, and then the output of the convolution layer is input into the Batch Normalization & ReLU activation function. The output of the Batch Normalization & ReLU activation function is sequentially connected to the attention mechanism module (i.e. Attention), the residual module (i.e. ResidualBlock), the fully connected layer (i.e. FC) and the Softmax layer to finally obtain the distribution probability of the monitoring result (i.e. outputfeature).

[0115] Furthermore, a comparative experiment was conducted on the heating current based on the traditional method and the method in this application, and the parameters set were shown in the following table:

[0116]

[0117] Threshold method:

[0118] Hyperparameters of the model without introducing the optimization residual module

[0119]

[0120] Introducing the hyperparameters of the model when optimizing the residual module

[0121]

[0122] Hyperparameters of the model when introducing the attention mechanism and residual structure

[0123]

[0124] Furthermore, based on the traditional method and the method in this application, the heating current fault prediction is performed, and the performance of the obtained accuracy and recall rate is shown in the following table:

[0125]

[0126] At this point, it can be seen from the above table that the method in this application can automatically extract multiple features at different levels without manually designing features, which is more in line with the actual situation, with higher accuracy and lower misrecognition rate than the threshold method, and the accuracy of the model introducing the residual module and the attention mechanism module is also further improved.

[0127] The present application embodiment provides a fault monitoring device for a breathing heating circuit, such as Figure 6 As shown, the device may include: a current acquisition module 601, a probability acquisition module 602 and a result determination module 603, wherein:

[0128] A current acquisition module is used to acquire the heating current currently corresponding to the breathing heating pipeline, where the heating current is the heating current currently corresponding to the heating wire of each heating pipeline unit in the breathing heating pipeline;

[0129] A probability acquisition module is used to input the heating current into the trained fault monitoring model to obtain the distribution probability of the monitoring results corresponding to the breathing heating pipeline, and the distribution probability of the monitoring results includes the probability of failure and the probability of normal operation;

[0130] The result determination module is used to determine the fault detection result corresponding to the breathing heating pipeline according to the distribution probability of the monitoring result and the preset probability processing rules.

[0131] Optionally, the fault monitoring model includes a convolution layer, a normalized activation module, an attention mechanism module, a residual module, a fully connected layer, a Dropout layer and a Softmax layer, and the normalized activation module includes a first Batch Normalization and a first ReLU activation function.

[0132] Optionally, the convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer, and the first convolution layer, the second convolution layer and the third convolution layer are all connected with a normalized activation module, the input of the first convolution layer is the heating current, the input of the second convolution layer is the output of the normalized activation module connected after the first convolution layer, the input of the third convolution layer is the output of the normalized activation module connected after the second convolution layer, and the output of the normalized activation module connected after the third convolution layer serves as the input of the attention mechanism module.

[0133] Optionally, the residual module includes two residual blocks, each residual block includes a second BatchNormalization, a second ReLU activation function, a weight layer and a third Batch Normalization connected in sequence, the residual module input is the output of the attention mechanism module, and the residual module output is the input of the fully connected layer.

[0134] Optionally, the attention mechanism module includes a channel attention module and a spatial attention module. The output of the attention mechanism module is obtained in the following way:

[0135] Obtain the output features of the normalized activation module connected after the third convolutional layer, and input the output features into the spatial attention module to obtain the first feature;

[0136] Multiply the output feature and the first feature element by element to obtain the product feature;

[0137] The product feature is normalized and the normalized product feature is input into the channel attention module to obtain the second feature;

[0138] Multiply the second feature and the normalized product feature element by element to get the output of the attention mechanism module.

[0139] Optionally, the fully connected layer includes a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence, the input of the first fully connected layer is the output of the residual module, the output of the third fully connected layer is the input of the Dropout layer, and the output of the Dropout layer is the input of the Softmax layer.

[0140] Optionally, the device further includes a model training module, which is a training fault monitoring model obtained by:

[0141] Acquire a training data set, the training data set includes a training sample set and a test sample set, the training data set includes sample pairs, each sample pair includes sample data and a corresponding sample label, wherein the sample data includes a heating current of a heating wire of each heating circuit unit in a breathing heating circuit, and the sample label represents a labeling result of a fault detection result corresponding to the sample pair;

[0142] The initial neural network model is trained based on the training sample set until the loss function corresponding to the initial neural network model converges, thereby obtaining an initial fault monitoring model, wherein the input of the initial neural network model is the sample pair, the output is the fault detection result corresponding to the sample pair, and the value of the loss function represents the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair;

[0143] The initial fault monitoring model is tested and verified based on the test sample set. If the test verification result corresponding to the initial fault monitoring model does not meet the set verification requirements, the step of training the initial neural network model based on the training sample set is repeated until the test verification result corresponding to the initial fault monitoring model meets the set verification requirements.

[0144] Optionally, the inspection sample set includes a validation sample set and a test sample set, the sample data includes positive sample data and negative sample data, and the model training module, when obtaining the training data set, is specifically used for:

[0145] Acquire at least one heating current of the breathing heating circuit during normal operation as at least one negative sample data;

[0146] Acquire at least one heating current of the breathing heating circuit when a fault occurs as at least one positive sample data;

[0147] At least one negative sample data and at least one positive sample data are randomly combined and divided according to the set quantity requirements to obtain a training sample set, a verification sample set and a test sample set respectively.

[0148] Optionally, the verification requirements are set to include a recall rate requirement and a precision rate requirement corresponding to each sample pair.

[0149] A fault monitoring device for a breathing heating circuit in this embodiment can execute a fault monitoring method for a breathing heating circuit shown in an embodiment of the present application. The implementation principle is similar and will not be repeated here.

[0150] An embodiment of the present application provides an electronic device, and the electronic device in the embodiment of the present application includes: a processor; and a memory, the memory is configured to store machine-readable instructions, and when the instructions are executed by the processor, the processor executes a fault monitoring method for a respiratory heating circuit.

[0151] The present application embodiment provides an electronic device, such as Figure 7 As shown, Figure 7 The electronic device shown includes: a processor 2001 and a memory 2003. The processor 2001 and the memory 2003 are connected, such as through a bus 2002. Optionally, the electronic device 2000 may also include a transceiver 2004. It should be noted that in actual applications, the transceiver 2004 is not limited to one, and the structure of the electronic device 2000 does not constitute a limitation on the embodiments of the present application.

[0152] Processor 2001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 2001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0153] The bus 2002 may include a path to transmit information between the above components. The bus 2002 may be a PCI bus or an EISA bus, etc. The bus 2002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0154] The memory 2003 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0155] The memory 2003 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 2001. The processor 2001 is used to execute the application code stored in the memory 2003 to implement Figure 6 The illustrated embodiment provides an operation of a fault monitoring device for a respiratory warming circuit.

[0156] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0157] The above is only part of the embodiments of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring a fault of a breathing heating circuit, characterized in that: The method is applied to a heated breathing circuit matched with a ventilator, and the method comprises: Obtaining a heating current currently corresponding to the breathing heating pipeline, wherein the heating current is a heating current currently corresponding to a heating wire of each heating pipeline unit in the breathing heating pipeline; Inputting the heating current into a trained fault monitoring model to obtain a distribution probability of a monitoring result corresponding to the breathing heating circuit, wherein the distribution probability of the monitoring result includes a probability of a fault occurring and a probability of normal operation; According to the distribution probability of the monitoring result and the preset probability processing rule, the fault detection result corresponding to the breathing heating pipeline is determined.

2. The method according to claim 1, characterized in that The fault monitoring model includes a convolutional layer, a normalized activation module, an attention mechanism module, a residual module, a fully connected layer, a Dropout layer and a Softmax layer, and the normalized activation module includes a first Batch Normalization and a first ReLU activation function.

3. The method according to claim 2, characterized in that The convolution layer includes a first convolution layer, a second convolution layer and a third convolution layer, and the first convolution layer, the second convolution layer and the third convolution layer are all connected with a normalized activation module, the input of the first convolution layer is the heating current, the input of the second convolution layer is the output of the normalized activation module connected after the first convolution layer, the input of the third convolution layer is the output of the normalized activation module connected after the second convolution layer, and the output of the normalized activation module connected after the third convolution layer serves as the input of the attention mechanism module.

4. The method according to claim 3, characterized in that The residual module includes two residual blocks, each of which includes a second Batch Normalization, a second ReLU activation function, a weight layer and a third Batch Normalization connected in sequence, the residual module input is the output of the attention mechanism module, and the residual module output is the input of the fully connected layer.

5. The method according to claim 4, characterized in that The attention mechanism module includes a channel attention module and a spatial attention module. The output of the attention mechanism module is obtained in the following way: Obtaining output features of a normalized activation module connected after the third convolutional layer, and inputting the output features into the spatial attention module to obtain a first feature; Multiplying the output feature and the first feature element by element to obtain a product feature; Normalizing the product feature, and inputting the normalized product feature into the channel attention module to obtain a second feature; The second feature and the normalized product feature are multiplied element by element to obtain the output of the attention mechanism module.

6. The method according to claim 2, characterized in that The fully connected layer includes a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence, the input of the first fully connected layer is the output of the residual module, the output of the third fully connected layer is the input of the Dropout layer, and the output of the Dropout layer is the input of the Softmax layer.

7. The method according to claim 1, characterized in that The fault monitoring model is trained in the following way: Acquire a training data set, wherein the training data set includes a training sample set and a test sample set, wherein the training data set includes sample pairs, each sample pair includes sample data and a corresponding sample label, wherein the sample data includes a heating current of a heating wire of each heating circuit unit in a breathing heating circuit, and the sample label represents a labeling result of a fault detection result corresponding to the sample pair; The initial neural network model is trained based on the training sample set until the loss function corresponding to the initial neural network model converges, thereby obtaining an initial fault monitoring model, wherein the input of the initial neural network model is a sample pair, the output is a fault detection result corresponding to the sample pair, and the value of the loss function represents the difference between the fault detection result corresponding to each sample pair output by the model and the labeling result corresponding to each sample pair; The initial fault monitoring model is tested and verified based on the test sample set. If the test verification result corresponding to the initial fault monitoring model does not meet the set verification requirements, the step of training the initial neural network model based on the training sample set is repeated until the test verification result corresponding to the initial fault monitoring model meets the set verification requirements.

8. The method according to claim 7, characterized in that The inspection sample set includes a validation sample set and a test sample set, the sample data includes positive sample data and negative sample data, and the obtaining of a training data set includes: Acquire at least one heating current of the breathing heating circuit during normal operation as at least one negative sample data; Acquire at least one heating current of the breathing heating circuit when a fault occurs as at least one positive sample data; The at least one negative sample data and the at least one positive sample data are randomly combined and divided according to a set quantity requirement to obtain the training sample set, the verification sample set and the test sample set respectively.

9. The method according to claim 7, characterized in that: The set verification requirements include the recall rate requirements and accuracy rate requirements corresponding to each sample pair.

10. A fault monitoring device for a breathing heating pipeline, characterized in that: The fault monitoring device is applied to a heated breathing circuit matched with a ventilator, and the fault monitoring device comprises: A current acquisition module, used for acquiring a heating current currently corresponding to the breathing heating pipeline, wherein the heating current is a heating current currently corresponding to a heating wire of each heating pipeline unit in the breathing heating pipeline; A probability acquisition module, used for inputting the heating current into a trained fault monitoring model to obtain a distribution probability of a monitoring result corresponding to the breathing heating pipeline, wherein the distribution probability of the monitoring result includes a probability of a fault and a probability of normal operation; The result determination module is used to determine the fault detection result corresponding to the breathing heating pipeline according to the distribution probability of the monitoring result and the preset probability processing rule.

Images