Training method and device of medical accelerator fault prediction model and medium thereof

Abstract

The invention relates to a training method and device for a medical accelerator fault prediction model and a medium. The method comprises the following steps: acquiring first medical accelerator data; preprocessing the first medical accelerator data to obtain second medical accelerator data; inputting the second medical accelerator data into a fault prediction neural network at least comprising a gating circulation layer to obtain a fault diagnosis result; and training the fault prediction neural network in a back propagation mode based on the fault diagnosis result and the label in the corresponding first medical accelerator data. By preprocessing the time sequence data of the medical accelerator and inputting the preprocessed data into the fault prediction neural network, the technology can capture the time sequence characteristics in the equipment operation, and performs back propagation training in combination with the fault diagnosis result and the tag, thereby optimizing the prediction performance of the fault prediction neural network, and improving the accuracy of the fault prediction. The equipment fault prediction precision is improved, the potential equipment damage risk is reduced, and the service life of the equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the field of medical accelerator diagnosis, and specifically provides a training method, a device and a medium for a medical accelerator fault prediction model. Background Art

[0002] Among the existing medical accelerator fault diagnosis technologies, most methods focus on detection and repair after a device failure occurs, usually after the failure has occurred. However, once a failure occurs, it may have a significant negative impact on device performance and other aspects, so timely and effective early warning is particularly important.

[0003] However, the existing technology cannot predict the failure of medical accelerators.

[0004] Accordingly, the art needs a new medical accelerator fault prediction solution to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects, the present invention is proposed to solve or at least partially solve the technical problem in the prior art that a medical accelerator cannot be given a fault warning.

[0006] In a first aspect, the present invention provides a method for training a medical accelerator fault prediction model, which comprises obtaining first medical accelerator data; preprocessing the first medical accelerator data to obtain second medical accelerator data; inputting the second medical accelerator data into a fault prediction neural network including at least a gated loop layer to obtain a fault diagnosis result; and training the fault prediction neural network by back propagation based on the fault diagnosis result and a label corresponding to the first medical accelerator data.

[0007] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, wherein the fault prediction neural network includes at least a first gated loop layer and a second gated loop layer, wherein the second medical accelerator data is time series data, and the second medical accelerator data is input into a fault prediction neural network including at least a gated loop layer to obtain a fault diagnosis result, including: obtaining a first medical accelerator feature vector based on the second medical accelerator data; inputting the first medical accelerator feature vector into the first gated loop layer and the second gated loop layer respectively to obtain a first time series sequence vector and a second time series sequence vector, wherein the first gated training layer and the second gated loop layer process the time series data along different time paths; fusing the first time series sequence vector and the second time series sequence vector to obtain a second medical accelerator feature vector; and the subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result.

[0008] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the time path includes a forward time sequence and a reverse time sequence, and wherein the first gated training layer and the second gated loop layer process the time series data along different time paths, including: when the first gated loop layer processes the first medical accelerator feature vector, the first medical accelerator feature vector is processed in the forward time sequence of the first medical accelerator feature vector to obtain a first time sequence vector; and when the second gated loop layer processes the first medical accelerator feature vector, the first medical accelerator feature vector is processed in the reverse time sequence of the first medical accelerator feature vector to obtain a second time sequence vector.

[0009] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the fault prediction neural network also includes a one-dimensional convolution layer, and obtaining the first medical accelerator feature vector based on the second medical accelerator data includes: inputting the second medical accelerator data into the one-dimensional convolution layer to obtain the first medical accelerator feature vector, wherein the first medical accelerator feature vector is a one-dimensional vector.

[0010] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the subsequent layer structure of the fault prediction neural network includes at least an attention module layer, and the subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result, including: inputting the second medical accelerator feature vector into the attention module layer to obtain a fault diagnosis result.

[0011] As an alternative or supplement to the above solution, in a method according to an embodiment of the present invention, the first medical accelerator data at least includes: current data and / or voltage data, internal pressure, internal vacuum degree, internal temperature and internal vibration.

[0012] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the first medical accelerator data includes data of multiple categories, and preprocessing the first medical accelerator data to obtain second medical accelerator data includes: performing denoising and standardization processing on the first medical accelerator data to obtain the second medical accelerator data, wherein the standardization processing is used to eliminate dimensional differences between data of different categories in the first medical accelerator data.

[0013] As an alternative or supplement to the above solution, in a method according to an embodiment of the present invention, the first medical accelerator data at least includes data of the medical accelerator when it is normal and data of the medical accelerator within a period of time before a failure; or, the first medical accelerator data at least includes data of the medical accelerator when it is normal, data of the medical accelerator within a period of time before a failure, and data of the medical accelerator when the failure occurs.

[0014] In a second aspect, the present invention provides a medical accelerator fault diagnosis method, comprising: acquiring first real-time medical accelerator data; preprocessing the first real-time medical accelerator data to obtain second real-time medical accelerator data; inputting the second real-time medical accelerator data into a trained fault prediction neural network that at least includes a gated loop layer to obtain a fault diagnosis result.

[0015] In a third aspect, a control device is provided, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of computer programs, wherein the computer programs are suitable for being loaded and run by the processor to execute the training method for a medical accelerator failure prediction model described in any one of the technical solutions of the training method for a medical accelerator failure prediction model.

[0016] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of computer programs are stored in the computer-readable storage medium, wherein the computer programs are suitable for being loaded and run by a processor to execute the method for training a medical accelerator failure prediction model as described in any one of the technical solutions of the method for training a medical accelerator failure prediction model.

[0017] The above one or more technical solutions of the present invention have at least one or more of the following Beneficial effects:

[0018] In the technical solution of the present invention, by preprocessing the timing data of the medical accelerator and inputting the preprocessed data into the fault prediction neural network, the technology can efficiently capture the timing characteristics of the equipment operation, combine the fault diagnosis results with the labels, and perform back propagation training, thereby optimizing the prediction performance of the fault prediction neural network, improving the accuracy of equipment fault prediction, reducing the potential risk of equipment damage and extending the equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0020] Figure 1is a flowchart of main steps of a method for training a medical accelerator fault prediction model according to an embodiment of the present invention;

[0021] Figure 2 is a schematic diagram of the structure of a fault prediction neural network of a training method for a medical accelerator fault prediction model according to an embodiment of the present invention;

[0022] Figure 3 is a schematic diagram of a process for obtaining a fault diagnosis result of a method for training a medical accelerator fault prediction model according to an embodiment of the present invention;

[0023] Figure 4 4 is a flow chart of a method for predicting medical accelerator failure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0025] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as computer programs, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store computer programs, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0026] Embodiment 1:

[0027] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart of the main steps of a method for training a medical accelerator fault prediction model according to an embodiment of the present invention. Figure 1 As shown, the training method of the medical accelerator fault prediction model in the embodiment of the present invention mainly includes the following steps S10 to S40.

[0028] Step S10: Acquire first medical accelerator data.

[0029] In this embodiment, the first medical accelerator data is collected time series data.

[0030] In one embodiment, the first medical accelerator data at least includes data of the medical accelerator when it is normal, data of the medical accelerator when it fails, and data of the medical accelerator in a period of time before the failure. In the present technology, the period of time before the failure refers to the time period before the failure occurs, wherein the failure refers to the medical accelerator issuing a failure alarm, wherein the failure alarm indicates that the medical accelerator is in a failure state at this time. The failure alarm depends on the failure alarm system, which is a system preset in the medical accelerator, and can only identify and alarm when the medical accelerator fails. When the medical accelerator issues a failure alarm, it means that the medical accelerator has failed. If it continues to be used, it may cause huge errors (i.e., inaccurate data) or further damage the machine.

[0031] In one embodiment, in order to ensure the comprehensiveness and accuracy of the acquired medical accelerator data, in this embodiment, various operating parameters of each subsystem of the medical accelerator are collected, including but not limited to current, voltage, temperature, vibration, pressure, vacuum, rotation speed, etc. These parameters can reflect the behavioral characteristics of the medical accelerator equipment under different operating conditions from different angles. In particular, the data therein will change accordingly in a period of time before the failure occurs. Since the causes of the failure of the medical accelerator are various, some of the data therein may be abnormal in a period of time before a single failure occurs, while some of the data is normal. For example, in a period of time before a certain failure occurs, the current and voltage data are abnormal at this time, and are different from the normal data, but the vibration data is the same as the normal data. In a period of time before another failure occurs, the vibration data is different from the normal data, and an abnormality occurs, while the current and voltage data are the same as the normal data.

[0032] Therefore, unlike most neural network inputs, the fault prediction of the first medical accelerator data, wherein the first medical accelerator data at least includes: current data and / or voltage data, internal pressure, internal vacuum degree, internal temperature and internal vibration. In this embodiment, the abnormality of a single parameter is often insufficient to fully reflect the fault state of the device. Therefore, the joint monitoring of multiple parameters can more comprehensively and accurately diagnose the operating state of the device, especially predict the fault.

[0033] Furthermore, when each single data of the medical accelerator data such as current data and / or voltage data, internal pressure, internal vacuum degree, internal temperature and internal vibration reaches a certain state (for example, exceeds a certain value), the acquisition system will not consider it a fault at this time, but when multiple values ​​thereof reach a certain state, although the standard of system failure has not been reached at this time, it is already a standard precursor. However, different faults have different precursors, and the accuracy of setting the engineering quantity is not high if it is only based on experience. However, in this embodiment, by analyzing the above-mentioned first medical accelerator data through the neural network method, the state law corresponding to these data before the fault can be found, thereby predicting the fault.

[0034] In this embodiment, by jointly monitoring current data and / or voltage data, internal pressure, internal vacuum, internal temperature and internal vibration, the overall operating status of the medical accelerator can be more comprehensively understood. The first medical accelerator data provides rich information for equipment fault diagnosis from multiple perspectives such as electrical, mechanical, thermal and environmental conditions. The correlation and combined analysis between different parameters can help identify early warning signals before a fault occurs, thereby achieving more accurate fault prediction.

[0035] Therefore, this step realizes all-round monitoring of the complex system of the medical accelerator through comprehensive collection and analysis of multiple parameters, which not only improves the accuracy of fault prediction, but also provides reliable data support for preventive maintenance. At the same time, multi-category data helps the subsequent neural network to make predictions from another angle, that is, the prediction of abnormal components. The medical accelerator is composed of multiple components, most of which are inside the medical accelerator. When the medical accelerator fails, it needs to be manually dismantled to slowly find which component of the medical accelerator has failed, which undoubtedly greatly prolongs the maintenance time. However, in this embodiment, the fault prediction neural network can be trained through multi-category data, so that the fault prediction neural network can eventually identify abnormal components.

[0036] Step S20: preprocessing the first medical accelerator data to obtain second medical accelerator data.

[0037] In this embodiment, when the first medical accelerator data includes multiple categories of data (such as the first medical accelerator data at least includes: current data and / or voltage data, internal pressure, internal vacuum, internal temperature, internal vibration and other categories), each category of data is preprocessed. The preprocessing at least includes denoising and standardization of the first medical accelerator data, wherein the standardization is used to eliminate the dimensional differences of different categories of data in the first medical accelerator data. Furthermore, the second medical accelerator data after the standardization is made to meet the input requirements of the subsequent fault prediction neural network.

[0038] In one embodiment, when processing complex time series data such as the first medical accelerator data, the first medical accelerator data as raw data is usually affected by noise and other interference factors, which may come from the instability of the device sensor, the interference of the external environment, or the anomalies in the data acquisition process. The unprocessed raw data may contain errors, deviations, or abnormal points, which will affect the accuracy of the prediction model if they are not removed. Therefore, the preprocessing in this embodiment first improves the data quality by denoising.

[0039] In one embodiment, a wavelet analysis algorithm is used for denoising. Wavelet analysis can decompose each category of data in the first medical accelerator data into different frequency components, remove high-frequency noise in the first medical accelerator data by filtering, and retain effective low-frequency signals. This method can accurately retain important features in the first medical accelerator data while reducing erroneous signals caused by equipment noise or external interference. In this embodiment, by using a wavelet analysis algorithm, the clarity and accuracy of the data are ensured, thereby enhancing the reliability of the subsequent analysis process.

[0040] Next, after the denoising process is completed, the data is standardized. Standardization is the process of converting feature data of different dimensions into a unified scale to facilitate input into the neural network for training and reasoning. Since various parameters in the First Medical Accelerator data, such as current, voltage, temperature, vibration, etc., have different physical magnitudes and numerical ranges, if standardization is not performed, some features may have too much or too little impact on the model during the training process, thereby affecting the convergence speed and prediction accuracy of the model. Through standardization, these feature data of different scales can be converted to the same range (for example, normalized to between 0 and 1 or with a mean of 0 and a variance of 1), thereby ensuring that the weight of each category of data in the neural network training is balanced, and avoiding a certain category of data dominating the model training process.

[0041] In this embodiment, after the standardization process, each category of data in the first medical accelerator data will be converted into a unified format to ensure that it meets the input requirements of the fault prediction neural network.

[0042] In this embodiment, by performing preprocessing operations, the noise in the first medical accelerator data is effectively removed, ensuring the accuracy of the data. At the same time, the standardization process converts the data into a consistent numerical range to ensure that it meets the input requirements of the subsequent fault prediction neural network. At the same time, since all categories of data in the first medical accelerator data are standardized and converted into a unified scale, the neural network can treat different categories of data equally in the subsequent training of the fault prediction neural network, so that it can distinguish the importance of different types of data during training, rather than distinguishing them at the beginning. Therefore, standardization is very important for the fault prediction neural network.

[0043] Step S30: inputting the second medical accelerator data into a fault prediction neural network including at least a gated loop layer to obtain a fault diagnosis result.

[0044] In one embodiment, the second medical accelerator data is preprocessed and input into a fault prediction neural network, which has been pre-trained and includes at least one gated recurrent layer. The gated recurrent layer is a neural network layer structure (gated recurrent unit, GRU for short) set based on a gated recurrent mechanism. The gated recurrent layer is a neural network layer that is good at processing time series data and can effectively capture the time dependencies in time series data (such as the second medical accelerator data). The gated recurrent layer can retain important historical information when processing long sequence data through its internal gating mechanism, while suppressing irrelevant noise and outdated information, thereby providing more accurate time series feature analysis for fault prediction.

[0045] In this embodiment, the gated loop layer selectively retains or updates the state information of the previous moment through its update gate and reset gate mechanism. The update gate determines the influence weight of the state of the previous moment on the current output, while the reset gate determines whether to ignore the information of the previous moment. This structure enables the gated loop layer to capture the long-term and short-term dependencies in the operation of the medical accelerator, thereby identifying the potential abnormal patterns of the device at different time points.

[0046] Finally, after being processed layer by layer by various layer structures of the fault prediction neural network, the output layer of the fault prediction neural network generates a fault diagnosis result. The fault diagnosis result shows whether the current operating status of the medical accelerator is normal, whether there is a potential risk of failure, or it also includes the identification of abnormal components. For example, the fault diagnosis result can be a classification problem (such as normal, may fail; a certain component is normal, a certain component fails) or a regression problem (such as a fault risk score or the probability of a certain component failing), which depends on the specific design of the fault prediction task and will not be repeated here.

[0047] In this embodiment, the preprocessed second medical accelerator data is input into the trained fault prediction neural network to predict the medical accelerator fault. The use of the gated loop layer ensures the extraction and learning of key features in the time series data, further improves the accuracy and real-time performance of fault prediction, and ensures that the system can provide early warning before the fault occurs.

[0048] like Figure 2 As shown, Figure 2 is a schematic diagram of the structure of the fault prediction neural network. In one embodiment, the fault prediction neural network includes at least a first gated loop layer and a second gated loop layer, wherein the second medical accelerator data is time series data, and step S30 also includes steps S301-S304, which are as follows:

[0049] Step S301: obtaining a first medical accelerator feature vector based on second medical accelerator data.

[0050] In this embodiment, the fault prediction neural network further includes a one-dimensional convolution layer, and the second medical accelerator data is input into the one-dimensional convolution layer to obtain the first medical accelerator feature vector, wherein the first medical accelerator feature vector is a one-dimensional vector.

[0051] In one embodiment, the one-dimensional convolution layer (1D CNN) is used to extract important local features from the preprocessed time series data and compress the high-dimensional time series data into a one-dimensional feature vector. Through the operation of the one-dimensional convolution layer, the complex second medical accelerator data is converted into a representation form that is easier to process by subsequent layer structures.

[0052] Specifically, the second medical accelerator data is input into the one-dimensional convolution layer as a one-dimensional time series signal. The one-dimensional convolution layer locally scans the second medical accelerator data through a sliding window, and extracts local features of fixed length in the second medical accelerator data during the sliding process. The convolution kernel in the one-dimensional convolution layer plays a role in identifying signal patterns in the above process. It can identify potential key patterns in time series data, such as mutations in voltage, current or vibration data. Through the use of layer-by-layer convolution and activation functions (such as ReLU), the one-dimensional convolution layer can gradually extract multi-level feature representations and summarize these representations into a more compact one-dimensional feature vector.

[0053] In this embodiment, by using a one-dimensional convolutional layer to extract features from the second medical accelerator data, the high-dimensional second medical accelerator data is converted into a simplified one-dimensional feature vector, and the key local features are retained. This feature representation can be more efficiently input into the subsequent layer structure, and the first medical accelerator feature vector with time series features is very suitable for being further processed by the subsequent first gated loop layer and the second gated loop layer, especially the processing of the one-dimensional first medical accelerator feature vector along different time paths.

[0054] Step S302: inputting the first medical accelerator feature vector into the first gated recurrent layer and the second gated recurrent layer respectively to obtain a first timing sequence vector and a second timing sequence vector.

[0055] In this embodiment, the first gated training layer and the second gated recurrent layer process the time series data along different time paths.

[0056] In one embodiment, the time series data of the first medical accelerator feature vector is processed through different time paths to ensure that key information during the operation of the device is captured from multiple time direction angles.

[0057] Specifically, when processing the first medical accelerator feature vector, the first gated loop layer processes the first medical accelerator feature vector in the forward time sequence of the first medical accelerator feature vector to obtain a first time sequence vector, wherein the forward time sequence is the sequence from the past to the present.

[0058] In other words, the first gated loop layer selectively retains or forgets the historical information in the time series data, especially the part closely related to the current state, through its internal gating mechanism. In this process, the operation of the update gate and the reset gate can ensure that the useful information of the previous moment is retained and combined with the characteristics of the current moment. This processing method can capture the changes in the medical accelerator from the past to the current operation process and identify the potential failure trend of the medical accelerator. Finally, the first gated loop layer outputs the first time series sequence vector.

[0059] At the same time, when processing the first medical accelerator feature vector, the second gated loop layer processes the first medical accelerator feature vector in the reverse time sequence of the first medical accelerator feature vector to obtain a second time sequence vector, wherein the reverse time sequence is the order from the future to the past.

[0060] In this embodiment, the second gated loop layer uses another time path to process the input timing data. Different from the forward processing of the first gated loop layer, the second gated loop layer processes the timing data in reverse order of time. Through this reverse processing method, the second gated loop layer can extract potential influencing factors from the information of subsequent moments of the operation of the medical accelerator, thereby identifying the relationship between the current moment and future faults. This layer of processing enables the system to rely not only on past data, but also to use future information to conduct a more comprehensive analysis of the current device status. Finally, the second gated loop layer outputs a second timing sequence vector, which represents the reverse timing pattern of the device operation characteristics.

[0061] This step processes and analyzes the first medical accelerator feature vector along different time paths, thereby combining the past operating trends of the medical accelerator and the influencing factors of future data. This multi-time path timing analysis method can more comprehensively reflect the operating status of the medical accelerator and ensure greater accuracy in the fault prediction process. Through this differentiated processing of time paths, the accuracy and reliability of fault diagnosis are improved, and potential faults that may occur in the medical accelerator are ultimately discovered in advance.

[0062] Step S303: Fusing the first time series vector and the second time series vector to obtain a second medical accelerator feature vector.

[0063] In one embodiment, a second medical accelerator feature vector is generated by fusing the first time sequence vector with the second time sequence vector. This fusion process can be performed in a variety of ways, including but not limited to vector concatenation, weighted averaging, or combining through additive functions, so as to make full use of the information in the two time paths. The first time sequence vector represents the operating characteristics of the forward time sequence, while the second time sequence vector represents the operating characteristics of the reverse time sequence. Through this fusion method, the generated second medical accelerator feature vector contains more comprehensively processed equipment operating characteristics, providing a highly comprehensive information basis for subsequent processing layers.

[0064] Step S304: The subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result.

[0065] In one embodiment, the subsequent layers of the fault prediction neural network refer to layer structures located after the first gated loop layer and the second gated loop layer, and these subsequent layer structures are used to receive the second medical accelerator feature vector and process the second medical accelerator feature vector.

[0066] In one embodiment, the subsequent layer structure of the fault prediction neural network includes at least an attention module layer. Further, the second medical accelerator feature vector is input into the attention module layer to obtain a fault diagnosis result.

[0067] In this embodiment, the subsequent layer structure of the fault prediction neural network includes at least one attention module layer, and the attention module layer may include one or more layers, wherein the one or more layers constitute the attention module layer. The attention module layer is used to further process the second medical accelerator feature vector. The attention module layer focuses on the feature information that is critical to fault prediction by calculating the importance weights of different parts in the feature vector, thereby improving the accuracy and robustness of fault diagnosis.

[0068] After the previous processing, the second medical accelerator feature vector already contains information of different time paths. The second medical accelerator feature vector is then input into the attention module layer. The attention module layer weights each part of the input feature vector to identify the part that has the greatest impact on the result in fault prediction. The attention module calculates the contribution of each feature in the second medical accelerator feature vector to the output result at the current moment through a scoring function (such as a dot product or additive function). Subsequently, these contributions are converted into attention weights after normalization by the Softmax function. These weights represent the importance of each feature relative to other features.

[0069] Next, the attention module applies the attention weights to the input feature vector. After each eigenvalue is multiplied by the corresponding weight, the key parts of the feature vector are amplified, while irrelevant or minor features are suppressed. This process ensures that the fault prediction neural network can focus more resources on features that are directly related to fault prediction and ignore those parts that have less impact on fault prediction. In this way, the attention module can not only improve the accuracy of the diagnosis results, but also help the neural network better understand the complex time series data structure, thereby improving the prediction performance.

[0070] Finally, the feature vector processed by the attention module layer is sent to the next layer of the fault prediction neural network for the generation of fault diagnosis results.

[0071] In this embodiment, the second medical accelerator feature vector includes multiple categories of features (such as current data and / or voltage data, internal pressure, internal vacuum, internal temperature and internal vibration), and also includes features after two different time path analyses. Therefore, the information included in the second medical accelerator feature vector is relatively complex and diverse.

[0072] In this embodiment, the prediction capability of the fault prediction neural network is significantly improved by automatically focusing on the important features in the feature vector of the second medical accelerator using the attention module layer with an attention mechanism. Furthermore, the adaptive weighting processing of feature importance is realized through the attention module layer, ensuring that the fault prediction neural network can accurately capture the most relevant fault signals in a multi-parameter and highly complex feature space, thereby improving the accuracy and reliability of fault diagnosis.

[0073] Step S40: training the fault prediction neural network by back propagation based on the fault diagnosis result and the label in the corresponding first medical accelerator data.

[0074] In this embodiment, the first medical accelerator data at least includes data of the medical accelerator when it is normal and data of the medical accelerator within a period of time before a failure.

[0075] Furthermore, the first medical accelerator data also includes data of the medical accelerator when it fails.

[0076] In this embodiment, the training of the fault prediction neural network is completed by the back propagation algorithm, and the training optimization is performed by combining the fault diagnosis result and the label in the first medical accelerator data. Specifically, in this embodiment, the current predicted fault diagnosis result is first obtained by the fault prediction neural network, and it is compared with the known true label (i.e., the label in the first medical accelerator data) to calculate the loss function. Then back propagation is performed.

[0077] The first medical accelerator data in this embodiment includes data of the medical accelerator in a normal state and data in a period of time before the fault. Through these two types of data, the fault prediction neural network can predict the fault. In addition, the two types of data of the first medical accelerator data also provide sufficient comparative information for the fault prediction neural network, so that the fault prediction neural network can distinguish the characteristic differences between the normal state and the potential fault of the medical accelerator. This feature learning process is the key to improving the accuracy of model fault prediction.

[0078] Furthermore, the first medical accelerator data also includes data of the medical accelerator in a fault state. The data of the medical accelerator in a fault state will play a role in the processing of the previous first gated loop layer and the second gated loop layer. The fault data not only enhances the network's ability to capture fault modes, but also provides more information about the system behavior after the fault occurs during the previous reverse timing path processing, helping the fault prediction neural network to more comprehensively understand the time evolution of the fault.

[0079] In addition, in this embodiment, methods such as cross-validation may be used to evaluate the accuracy and generalization ability of the fault prediction neural network to ensure the effectiveness of the fault prediction neural network in practical applications.

[0080] This step combines normal, pre-fault, and fault data to fully train the fault prediction neural network, ensuring that the network can accurately predict the time and nature of the fault based on data in different states. The back propagation process optimizes the network weights through the loss function, continuously improving the prediction performance of the model, so that the fault prediction system can more reliably and effectively identify potential equipment failure risks in advance in practical applications.

[0081] like Figure 4 As shown, the present invention also provides a medical accelerator fault diagnosis method, comprising:

[0082] Obtaining first real-time medical accelerometer data;

[0083] Preprocessing the first real-time medical accelerator data to obtain second real-time medical accelerator data;

[0084] The second real-time medical accelerator data is input into a trained fault prediction neural network that at least includes a gated loop layer to obtain a fault diagnosis result.

[0085] In this embodiment, the fault diagnosis result includes the medical accelerator fault prediction result and the component detection result in the medical accelerator. Further, the fault diagnosis result is sent back to the user. The fault diagnosis result can be further generated into a diagnosis report and sent back to the user.

[0086] Furthermore, in a medical accelerator fault diagnosis method, the fault prediction neural network includes at least a first gated loop layer and a second gated loop layer, wherein the second medical accelerator data is time series data, and the second real-time medical accelerator data is input into a fault prediction neural network including at least a gated loop layer to obtain a fault diagnosis result, including: obtaining a first medical accelerator feature vector based on the second real-time medical accelerator data; inputting the first medical accelerator feature vector into the first gated loop layer and the second gated loop layer respectively to obtain a first time series sequence vector and a second time series sequence vector, wherein the first gated training layer and the second gated loop layer process the time series data along different time paths; fusing the first time series sequence vector and the second time series sequence vector to obtain a second medical accelerator feature vector; and the subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result.

[0087] Furthermore, in the medical accelerator fault diagnosis method, the time path includes a forward timing sequence and a reverse timing sequence, wherein the first gated training layer and the second gated loop layer process the timing data along different time paths, including: wherein the first gated loop layer processes the first medical accelerator feature vector in the forward timing sequence of the first medical accelerator feature vector to obtain a first timing sequence vector when processing the first medical accelerator feature vector; wherein the second gated loop layer processes the first medical accelerator feature vector in the reverse timing sequence of the first medical accelerator feature vector to obtain a second timing sequence vector when processing the first medical accelerator feature vector.

[0088] Furthermore, in the medical accelerator fault diagnosis method, the fault prediction neural network also includes a one-dimensional convolution layer, and based on the second real-time medical accelerator data, a first medical accelerator feature vector is obtained, including: inputting the second real-time medical accelerator data into the one-dimensional convolution layer to obtain the first medical accelerator feature vector, wherein the first medical accelerator feature vector is a one-dimensional vector.

[0089] Furthermore, in the medical accelerator fault diagnosis method, the subsequent layer structure of the fault prediction neural network includes at least an attention module layer, and the subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result, including: inputting the second medical accelerator feature vector into the attention module layer to obtain a fault diagnosis result.

[0090] Furthermore, in the medical accelerator fault diagnosis method, the first real-time medical accelerator data at least includes: current data and / or voltage data, internal pressure, internal vacuum degree, internal temperature and internal vibration.

[0091] Furthermore, in the medical accelerator fault diagnosis method, the first real-time medical accelerator data includes data of multiple categories, and the first real-time medical accelerator data is preprocessed to obtain second real-time medical accelerator data, including: performing denoising and standardization processing on the first real-time medical accelerator data to obtain the second real-time medical accelerator data, wherein the standardization processing is used to eliminate the dimensional differences of data of different categories in the first real-time medical accelerator data.

[0092] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0093] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of each of the above method embodiments when executed by the processor. Among them, the computer program includes a computer program, and the computer program can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electric carrier signals and telecommunication signals.

[0094] Furthermore, the present invention also provides a control device. In an embodiment of a control device according to the present invention, the control device includes a processor and a storage device, the storage device can be configured to store a program for executing the training method of the medical accelerator fault prediction model of the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the training method of the medical accelerator fault prediction model of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0095] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the training method of the medical accelerator fault prediction model of the above method embodiment, and the program can be loaded and run by the processor to implement the above training method of the medical accelerator fault prediction model. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0096] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0097] Those skilled in the art will appreciate that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0098] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A training method for a medical accelerator fault prediction model, characterized in that: include: Obtaining the first medical accelerator data; preprocessing the first medical accelerator data to obtain second medical accelerator data; The second medical accelerator data is input into a fault prediction neural network including at least a gated loop layer to obtain a fault diagnosis result; and the fault prediction neural network is trained based on the fault diagnosis result and a label in the corresponding first medical accelerator data by back propagation.

2. The method for training a medical accelerator fault prediction model according to claim 1, characterized in that: The fault prediction neural network at least includes a first gated loop layer and a second gated loop layer, wherein the second medical accelerator data is time series data, and the second medical accelerator data is input into a fault prediction neural network at least including a gated loop layer to obtain a fault diagnosis result, including: Based on the second medical accelerator data, obtaining a first medical accelerator feature vector; Inputting the first medical accelerator feature vector into the first gated loop layer and the second gated loop layer respectively to obtain a first time series vector and a second time series vector, wherein the first gated loop layer and the second gated loop layer process the time series data along different time paths; fusing the first time series vector and the second time series vector to obtain a second medical accelerator feature vector; The subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result.

3. The method for training a medical accelerator fault prediction model according to claim 2, characterized in that: The time path includes a forward time sequence and a reverse time sequence, wherein the first gated training layer and the second gated recurrent layer process the time series data along different time paths, including: wherein when processing the first medical accelerator feature vector, the first gated loop layer processes the first medical accelerator feature vector in the forward time sequence of the first medical accelerator feature vector to obtain a first time sequence vector; When processing the first medical accelerator feature vector, the second gated loop layer processes the first medical accelerator feature vector in a reverse time sequence of the first medical accelerator feature vector to obtain a second time sequence vector.

4. The method for training a medical accelerator fault prediction model according to any one of claims 1 to 3, characterized in that: The fault prediction neural network further includes a one-dimensional convolution layer, and the first medical accelerator feature vector is obtained based on the second medical accelerator data, including: The second medical accelerator data is input into the one-dimensional convolution layer to obtain the first medical accelerator feature vector, wherein the first medical accelerator feature vector is a one-dimensional vector.

5. The method for training a medical accelerator fault prediction model according to any one of claims 1 to 3, characterized in that: The subsequent layer structure of the fault prediction neural network includes at least an attention module layer, and the subsequent layer structure in the fault prediction neural network processes the second medical accelerator feature vector to obtain a fault diagnosis result, including: The second medical accelerator feature vector is input into the attention module layer to obtain a fault diagnosis result.

6. The method for training a medical accelerator fault prediction model according to claim 1, characterized in that: The first medical accelerator data at least includes: current data and / or voltage data, internal pressure, internal vacuum degree, internal temperature and internal vibration; and / or, the first medical accelerator data includes multiple categories of data, and the first medical accelerator data is preprocessed to obtain the second medical accelerator data, including: The first medical accelerator data is subjected to denoising and standardization processing to obtain second medical accelerator data, wherein the standardization processing is used to eliminate dimensional differences of data of different categories in the first medical accelerator data.

7. The method for training a medical accelerator fault prediction model according to claim 1 or 6, characterized in that: The first medical accelerator data at least includes data of the medical accelerator when it is normal and data of the medical accelerator within a period of time before failure; or, the first medical accelerator data at least includes data of the medical accelerator when it is normal, data of the medical accelerator within a period of time before failure, and data of the medical accelerator when the failure occurs.

8. A method for diagnosing faults of a medical accelerator, characterized in that: include: Obtaining first real-time medical accelerometer data; preprocessing the first real-time medical accelerator data to obtain second real-time medical accelerator data; The second real-time medical accelerator data is input into a trained fault prediction neural network that at least includes a gated loop layer to obtain a fault diagnosis result.

9. A control device, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of computer programs, characterized in that: The computer program is suitable for being loaded and run by the processor to execute the training method of the medical accelerator failure prediction model according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a plurality of computer programs, characterized in that: The computer program is suitable for being loaded and run by a processor to execute the training method of a medical accelerator failure prediction model according to any one of claims 1 to 7.

Images