Radiographic machine room maintenance method and maintenance system

Abstract

The invention relates to the field of radiation machine room maintenance, aims at solving the problems that a traditional maintenance method is low in efficiency and large in careless omissions and an existing maintenance system is insufficient in intelligence, and provides a radiation machine room maintenance method and system. According to the method, multi-source data such as temperature and humidity, equipment vibration and the like are collected through a distributed sensor network. Analyzing the vibration data by using an improved convolutional neural network, and judging whether the equipment part is loosened or not; an improved recurrent neural network is used to process temperature and humidity data, and environmental parameter changes are predicted. A self-adaptive maintenance strategy is automatically generated according to an analysis result, and maintenance personnel can remotely obtain data and tasks. The system comprises a sensor module, a transmission module, a central processing module and a display application module which cooperate with each other. The system overcomes the defects of traditional maintenance, can accurately monitor the equipment and environment of the machine room in real time, intelligently generates a maintenance strategy, improves the maintenance efficiency and accuracy, reduces the cost, and guarantees the efficient and safe operation of the radiation machine room.

Description

Technical Field

[0001] The present invention relates to the field of radiation room maintenance, and in particular to a radiation room maintenance method and a maintenance system. Background Art

[0002] Equipment within radiology rooms, such as X-ray machines and CT scanners, plays a key role in medical diagnosis. However, these devices require demanding operating environments, maintaining specific temperature and humidity levels while preventing radiation leaks. Traditional radiology room maintenance relies on regular manual inspections and recording of equipment operating and environmental parameters, which is inefficient and prone to oversights. Existing maintenance systems lack a high level of intelligence and are unable to accurately monitor and maintain the equipment and environment in real time, making them unable to meet the demands of modern healthcare for efficient and safe operation of radiology rooms. Summary of the Invention

[0003] The object of the present invention is to provide a radiation room maintenance method and maintenance system to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a method for maintaining a radiology room, comprising the following steps: Step 1: Real-time multi-source data collection: Utilizing a distributed sensor network, temperature and humidity sensors and equipment vibration sensors are deployed in the radiation room. These sensors collect environmental parameters and equipment operating parameter data in real time, i.e., multi-source data. Step 2: Data Fusion and Intelligent Analysis: The collected multi-source data is transmitted to the central processing unit. An improved convolutional neural network is used to analyze the equipment vibration data to determine whether any internal components of the equipment are loose. An improved recurrent neural network is used to process the temperature and humidity time series data to predict the changing trends of the computer room environmental parameters. By establishing equipment operation status models and computer room environment models, equipment failure risks and environmental parameter anomalies are automatically identified. Step 3: Adaptive maintenance strategy generation: Based on the data analysis results, the system automatically generates an adaptive maintenance strategy according to preset rules; Step 4: Remote collaborative maintenance: Maintenance personnel obtain real-time data on equipment and environment in the computer room through mobile terminals or remote workstations, and receive maintenance tasks generated by the system in step 3.

[0005] Preferably, the improved convolutional neural network implements the following logic in equipment vibration data analysis: Multi-source data aggregation and integration: pre-process the multi-source data collected in step 1 by performing noise reduction, format unification, and time alignment operations; The improved convolutional neural network is used to extract equipment vibration data from preprocessed multi-source data and divide the preprocessed equipment vibration data into training, validation, and test sets. The improved CNN model is trained using the training set data and the parameters of the improved CNN model are continuously adjusted through the back-propagation algorithm to enable the improved CNN model to extract features related to the status of the internal components of the equipment from the vibration data. Fault diagnosis: When the trained improved CNN model receives real-time equipment vibration data, it can quickly extract and analyze the data features. By comparing the data with the normal equipment vibration features learned in the model and the vibration features under various fault modes, it can determine whether there are loose components in the equipment. If the model output results indicate that the degree of match between the current vibration data and the features under the fault mode exceeds the set threshold, the equipment is judged to have the corresponding fault risk and the results are transmitted to the maintenance strategy generation module.

[0006] Preferably, the specific steps of constructing the improved CNN model are as follows: Determine the characteristics and requirements of equipment vibration data: By obtaining historical vibration data from equipment in the radiology room, clarify its vibration data characteristics under different operating conditions and failure modes. Specifically, collect equipment vibration data covering normal operation, loose components, and wear failure conditions. Then use signal analysis tools to analyze the vibration data in the time domain, frequency domain, and time-frequency domain. Based on these analysis results, determine the vibration features that should be extracted by the improved CNN model. Design a custom convolution kernel: Based on the determined vibration characteristics of the equipment, design a specialized custom convolution kernel. For localized, high-frequency vibration characteristics, use a 3×3 convolution kernel; for widely distributed, low-frequency vibration characteristics, use a 7×7 convolution kernel. Build a basic CNN architecture: The architecture includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer is set according to the format and dimension of the device vibration data. In the convolutional layer, multiple convolutional layers are stacked in sequence. Each convolutional layer uses the custom convolution kernel designed previously to perform convolution operations, extracting different levels of features from the device vibration data through convolution operations. The pooling layer follows the convolution layer and uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer, reducing the amount of data and the computational complexity of the model while retaining the main feature information. The fully connected layer flattens the pooled feature map and integrates and classifies the features through multiple fully connected neurons. Finally, the output layer outputs the corresponding results according to the task requirements of the model. Introducing the attention mechanism: An attention mechanism module is introduced into the improved CNN model, and the attention mechanism module is located between the convolutional layers. The improved CNN model receives the feature map output by the previous convolutional layer, and performs weighted processing on the feature map by calculating the attention weight of each feature channel or spatial position. Specifically, the feature map is input into a subnetwork containing multiple fully connected layers, and the attention weight vector of each position is obtained after calculation. The weight vector reflects the importance of the position feature for the model to judge the operating status of the equipment. Then, the original feature map is multiplied element-by-element by the attention weight vector to obtain the feature map enhanced by the attention mechanism, and then it is input into the next convolutional layer for further processing. Through the attention mechanism module, the improved CNN model automatically focuses on the vibration features that are most valuable for equipment fault diagnosis, thereby improving the accuracy and robustness of the model.

[0007] Preferably, the specific implementation steps for processing the temperature and humidity time series data using the improved recurrent neural network are as follows: Data preprocessing: Continuously collect data from temperature and humidity sensors distributed throughout the radiation room, ensuring that the acquisition frequency matches the actual dynamics of the room's environmental changes. A moving average filtering algorithm is then used to clean the raw time series data, removing outliers caused by sensor failures or transient interference. Empirical mode decomposition is also used to decompose the temperature and humidity time series into multiple intrinsic mode functions (IMFs). By analyzing the frequency characteristics of each IMF component, high-frequency noise components unrelated to normal changes in the room's environment are filtered out, resulting in pure time series data. Build an improved recurrent neural network model: Based on the existing long short-term memory network model, embed a self-attention mechanism module. The self-attention mechanism module enables the long short-term memory network model to automatically pay attention to the degree of correlation between data at different time steps when processing time series data, and predict key time points that have the greatest impact on the results; Multi-scale feature extraction module: A multi-scale feature extraction module is added to the improved recurrent neural network model. The multi-scale feature extraction module consists of multiple one-dimensional convolutional layers with different convolution kernel sizes, which perform convolution operations on time series data respectively; Dynamically adjusted hidden layer structure: The number of neurons and layers in the hidden layer are dynamically adjusted based on the complexity of the input time series data and performance feedback during model training; Loss function design: For temperature and humidity time series data prediction tasks, we use mean absolute percentage error as the loss function to evaluate the relative error between the predicted value and the true value, ensuring the model's prediction accuracy for data of different magnitudes. Optimization algorithm selection and adjustment: An adaptive moment estimation optimization algorithm is used to train and update the parameters of the improved recurrent neural network model. During training, the learning rate is dynamically adjusted. A larger learning rate is set initially to quickly explore the parameter space. When the model loss value on the validation set fluctuates or no longer decreases significantly, the learning rate is gradually reduced to enable more precise parameter adjustments near the optimal solution. Real-time data prediction: After training, the improved recurrent neural network model is deployed to the radiology room maintenance system, which receives the latest data collected by temperature and humidity sensors in real time. The improved recurrent neural network model takes the current and past time series data as input, extracts and calculates features through each layer of the model, and outputs predicted values ​​for temperature and humidity environmental parameters for a period of time in the future. Abnormal warning mechanism: The environmental parameter values ​​predicted by the model are compared with the pre-set normal range. If the predicted value exceeds the normal range, the system will immediately issue an abnormal warning signal. At the same time, the abnormal situation is analyzed and classified based on historical data and model prediction results.

[0008] Preferably, a radiation room maintenance system includes a sensor module, a transmission module, a central processing module and a display application module, and is characterized in that: the sensor module is responsible for collecting vibration data and temperature and humidity data of equipment in the room, wherein the sensor used has self-calibration and self-diagnosis functions, and can automatically detect whether its own working status is normal at regular intervals to ensure the accuracy and reliability of the collected data; the transmission module adopts a hybrid communication method combining wireless and wired to quickly transmit the data collected by the sensor module to the central processing module; the central processing module includes a data storage unit, a data analysis unit and a maintenance strategy generation unit, and the data storage unit adopts distributed database technology to efficiently store and manage the massive amount of collected data; the data analysis unit runs an improved CNN model and an improved recurrent neural network model to deeply mine and analyze the data; the maintenance strategy generation unit generates specific maintenance tasks and instructions based on the analysis results; the display application module provides an operation interface for maintenance personnel and management personnel, including an equipment status monitoring interface, an environmental parameter display interface, a maintenance task management interface and a remote collaboration platform. Through an intuitive visual interface, users can obtain required information and perform corresponding operations.

[0009] Compared with the existing technology, the beneficial effects of the present invention are: efficient and accurate equipment fault monitoring: through the improved convolutional neural network to conduct in-depth analysis of equipment vibration data, it is possible to keenly capture subtle changes in internal equipment components such as whether they are loose. Traditional manual inspections are difficult to detect the hidden dangers of loose components in the early stage, but the present invention uses advanced signal processing and deep learning technology to detect potential failure risks in advance. For example, when equipment components begin to become slightly loose, the model can issue an early warning based on the unique vibration characteristic changes, buying time for timely maintenance, greatly reducing the probability of sudden equipment failure, ensuring stable operation of the equipment, and avoiding delays in medical diagnosis due to equipment failure.

[0010] Accurate prediction and control of environmental parameters: The improved recurrent neural network accurately processes temperature and humidity time series data, effectively predicting the changing trends of computer room environmental parameters. Under traditional maintenance methods, computer room environmental control often lags behind, causing temperature and humidity to exceed the appropriate operating range of the equipment, affecting equipment performance and life. The present invention can predict temperature and humidity changes in advance. When the temperature or humidity is about to deviate from the normal range, the system automatically initiates corresponding control measures, such as adjusting the air conditioning operation mode in advance, turning on dehumidification or humidification equipment, to ensure that the computer room environment is always in the best state for equipment operation, extend the service life of the equipment, and improve the stability of equipment operation.

[0011] Convenient and efficient remote collaborative maintenance mode: Maintenance personnel can obtain computer room data and receive maintenance tasks in real time through mobile terminals or remote workstations, breaking the limitations of time and space. When encountering complex equipment failures, experts in different regions can use the remote collaboration platform to jointly consult based on shared real-time data. In traditional maintenance, experts need to be present on site, which consumes a lot of time and energy. The remote collaboration mode of the present invention allows experts to provide professional guidance in a timely manner, quickly solve problems, improve maintenance efficiency, reduce equipment downtime, and ensure the continuous and stable operation of the radiology room. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0014] Example 1

[0015] See also Figure 1 The present invention provides a technical solution: a radiation room maintenance method, comprising the following steps: Step 1: Real-time multi-source data collection: Utilizing a distributed sensor network, temperature and humidity sensors and equipment vibration sensors are deployed in the radiation room. These sensors collect environmental parameters and equipment operating parameter data in real time, i.e., multi-source data. Step 2: Data Fusion and Intelligent Analysis: The collected multi-source data is transmitted to the central processing unit. An improved convolutional neural network is used to analyze the equipment vibration data to determine whether any internal components of the equipment are loose. An improved recurrent neural network is used to process the temperature and humidity time series data to predict the changing trends of the computer room environmental parameters. By establishing equipment operation status models and computer room environment models, equipment failure risks and environmental parameter anomalies are automatically identified. The improved convolutional neural network is used to implement the following logic in equipment vibration data analysis: Multi-source data aggregation and integration: pre-process the multi-source data collected in step 1 by performing noise reduction, format unification, and time alignment operations; The improved convolutional neural network is used to extract equipment vibration data from preprocessed multi-source data and divide the preprocessed equipment vibration data into training, validation, and test sets. The improved CNN model is trained using the training set data and the parameters of the improved CNN model are continuously adjusted through the back-propagation algorithm to enable the improved CNN model to extract features related to the status of the internal components of the equipment from the vibration data. Fault diagnosis: When the trained improved CNN model receives real-time equipment vibration data, it can quickly extract and analyze the data features. By comparing the data with the normal equipment vibration features learned in the model and the vibration features under various fault modes, it can determine whether there are loose components in the equipment. If the model output results indicate that the degree of match between the current vibration data and the features under the fault mode exceeds the set threshold, the equipment is judged to have the corresponding fault risk and the results are transmitted to the maintenance strategy generation module.

[0016] The specific steps for building the improved CNN model are as follows: Determine the characteristics and requirements of equipment vibration data: By obtaining historical vibration data from equipment in the radiology room, clarify its vibration data characteristics under different operating conditions and failure modes. Specifically, collect equipment vibration data covering normal operation, loose components, and wear failure conditions. Then use signal analysis tools to analyze the vibration data in the time domain, frequency domain, and time-frequency domain. Based on these analysis results, determine the vibration features that should be extracted by the improved CNN model. Design a custom convolution kernel: Based on the determined vibration characteristics of the equipment, design a specialized custom convolution kernel. For localized, high-frequency vibration characteristics, use a 3×3 convolution kernel; for widely distributed, low-frequency vibration characteristics, use a 7×7 convolution kernel. Build a basic CNN architecture: The architecture includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer is set according to the format and dimension of the device vibration data. In the convolutional layer, multiple convolutional layers are stacked in sequence. Each convolutional layer uses the custom convolution kernel designed previously to perform convolution operations, extracting different levels of features from the device vibration data through convolution operations. The pooling layer follows the convolution layer and uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer, reducing the amount of data and the computational complexity of the model while retaining the main feature information. The fully connected layer flattens the pooled feature map and integrates and classifies the features through multiple fully connected neurons. Finally, the output layer outputs the corresponding results according to the task requirements of the model. Introducing the attention mechanism: An attention mechanism module is introduced into the improved CNN model, and the attention mechanism module is located between the convolutional layers. The improved CNN model receives the feature map output by the previous convolutional layer, and performs weighted processing on the feature map by calculating the attention weight of each feature channel or spatial position. Specifically, the feature map is input into a subnetwork containing multiple fully connected layers, and the attention weight vector of each position is obtained after calculation. The weight vector reflects the importance of the position feature for the model to judge the operating status of the equipment. Then, the original feature map is multiplied element-by-element by the attention weight vector to obtain the feature map enhanced by the attention mechanism, and then it is input into the next convolutional layer for further processing. Through the attention mechanism module, the improved CNN model automatically focuses on the vibration features that are most valuable for equipment fault diagnosis, thereby improving the accuracy and robustness of the model.

[0017] The specific implementation steps for the improved recurrent neural network to process temperature and humidity time series data are as follows: Data preprocessing: Continuously collect data from temperature and humidity sensors distributed throughout the radiation room, ensuring that the acquisition frequency matches the actual dynamics of the room's environmental changes. A moving average filtering algorithm is then used to clean the raw time series data, removing outliers caused by sensor failures or transient interference. Empirical mode decomposition is also used to decompose the temperature and humidity time series into multiple intrinsic mode functions (IMFs). By analyzing the frequency characteristics of each IMF component, high-frequency noise components unrelated to normal changes in the room's environment are filtered out, resulting in pure time series data. Build an improved recurrent neural network model: Based on the existing long short-term memory network model, embed a self-attention mechanism module. The self-attention mechanism module enables the long short-term memory network model to automatically pay attention to the degree of correlation between data at different time steps when processing time series data, and predict key time points that have the greatest impact on the results; Multi-scale feature extraction module: A multi-scale feature extraction module is added to the improved recurrent neural network model. The multi-scale feature extraction module consists of multiple one-dimensional convolutional layers with different convolution kernel sizes, which perform convolution operations on time series data respectively; Dynamically adjusted hidden layer structure: The number of neurons and layers in the hidden layer are dynamically adjusted based on the complexity of the input time series data and performance feedback during model training; Loss function design: For temperature and humidity time series data prediction tasks, we use mean absolute percentage error as the loss function to evaluate the relative error between the predicted value and the true value, ensuring the model's prediction accuracy for data of different magnitudes. Optimization algorithm selection and adjustment: An adaptive moment estimation optimization algorithm is used to train and update the parameters of the improved recurrent neural network model. During training, the learning rate is dynamically adjusted. A larger learning rate is set initially to quickly explore the parameter space. When the model loss value on the validation set fluctuates or no longer decreases significantly, the learning rate is gradually reduced to enable more precise parameter adjustments near the optimal solution. Real-time data prediction: After training, the improved recurrent neural network model is deployed to the radiology room maintenance system, which receives the latest data collected by temperature and humidity sensors in real time. The improved recurrent neural network model takes the current and past time series data as input, extracts and calculates features through each layer of the model, and outputs predicted values ​​for temperature and humidity environmental parameters for a period of time in the future. Abnormal warning mechanism: The environmental parameter values ​​predicted by the model are compared with the pre-set normal range. If the predicted value exceeds the normal range, the system will immediately issue an abnormal warning signal. At the same time, the abnormal situation is analyzed and classified by combining historical data and model prediction results. Step 3: Adaptive maintenance strategy generation: Based on the data analysis results, the system automatically generates an adaptive maintenance strategy according to preset rules; Step 4: Remote collaborative maintenance: Maintenance personnel obtain real-time data on equipment and environment in the computer room through mobile terminals or remote workstations, and receive maintenance tasks generated by the system in step 3.

[0018] Example 2

[0019] See also Figure 1, a radiation room maintenance system, including a sensor module, a transmission module, a central processing module and a display application module, wherein the sensor module is responsible for collecting vibration data and temperature and humidity data of equipment in the room, wherein the sensor used has self-calibration and self-diagnosis functions, and can automatically detect whether its own working status is normal regularly to ensure the accuracy and reliability of the collected data; the transmission module adopts a hybrid communication method combining wireless and wired to quickly transmit the data collected by the sensor module to the central processing module; the central processing module includes a data storage unit, a data analysis unit and a maintenance strategy generation unit, and the data storage unit adopts distributed database technology to efficiently store and manage the massive amount of collected data; the data analysis unit runs an improved CNN model and an improved recurrent neural network model to deeply mine and analyze the data; the maintenance strategy generation unit generates specific maintenance tasks and instructions based on the analysis results; the display application module provides an operation interface for maintenance personnel and management personnel, including an equipment status monitoring interface, an environmental parameter display interface, a maintenance task management interface and a remote collaboration platform. Through an intuitive visual interface, users can obtain required information and perform corresponding operations.

[0020] The present invention provides a maintenance method and maintenance system for a radiology room. The method collects multi-source data in real time by deploying temperature and humidity sensors and equipment vibration sensors in the radiology room; uses an improved convolutional neural network to analyze equipment vibration data to determine component loosening problems, and uses an improved recurrent neural network to process temperature and humidity time series data to predict environmental parameter change trends, thereby achieving data fusion and intelligent analysis; automatically generates adaptive maintenance strategies based on the analysis results, and supports maintenance personnel to perform remote collaborative maintenance through mobile terminals or remote workstations. The corresponding maintenance system includes a sensor module, a transmission module, a central processing module, and a display application module, and each module works together to ensure accurate monitoring and maintenance of the equipment and environment in the room. The present invention improves maintenance efficiency and accuracy, can promptly detect potential problems, achieve intelligent and scientific maintenance, effectively reduce maintenance costs, and extend the service life of equipment.

[0021] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for maintaining a radiology room, characterized in that: The following steps are involved: Step 1: Real-time multi-source data collection: Utilizing a distributed sensor network, temperature and humidity sensors and equipment vibration sensors are deployed in the radiation room. These sensors collect environmental parameters and equipment operating parameter data in real time, i.e., multi-source data. Step 2: Data Fusion and Intelligent Analysis: The collected multi-source data is transmitted to the central processing unit. An improved convolutional neural network is used to analyze the equipment vibration data to determine whether any internal components of the equipment are loose. An improved recurrent neural network is used to process the temperature and humidity time series data to predict the changing trends of the computer room environmental parameters. Based on the obtained equipment operating status and computer room environment, equipment failure risks and abnormal environmental parameters are automatically identified. Step 3: Adaptive maintenance strategy generation: Based on the data analysis results, the system automatically generates an adaptive maintenance strategy according to preset rules; Step 4: Remote collaborative maintenance: Maintenance personnel obtain real-time data on equipment and environment in the computer room through mobile terminals or remote workstations, and receive maintenance tasks generated by the system in step 3.

2. A radiation room maintenance method according to claim 1, characterized in that: The improved convolutional neural network is used to implement the following logic in equipment vibration data analysis: Multi-source data aggregation and integration: pre-process the multi-source data collected in step 1 by performing noise reduction, format unification, and time alignment operations; The improved convolutional neural network is used to extract equipment vibration data from preprocessed multi-source data and divide the preprocessed equipment vibration data into training, validation, and test sets. The improved CNN model is trained using the training set data and the parameters of the improved CNN model are continuously adjusted through the back-propagation algorithm to enable the improved CNN model to extract features related to the status of the internal components of the equipment from the vibration data. Fault diagnosis: When the trained improved CNN model receives real-time equipment vibration data, it can quickly extract and analyze the data features. By comparing the data with the normal equipment vibration features learned in the model and the vibration features under various fault modes, it can determine whether there are loose components in the equipment. If the model output results indicate that the degree of match between the current vibration data and the features under the fault mode exceeds the set threshold, the equipment is judged to have the corresponding fault risk and the results are transmitted to the maintenance strategy generation module.

3. A radiation room maintenance method according to claim 2, characterized in that: The specific construction steps of the improved CNN model are as follows: Determine the characteristics and requirements of equipment vibration data: By obtaining historical vibration data from equipment in the radiology room, clarify its vibration data characteristics under different operating conditions and failure modes. Specifically, collect equipment vibration data covering normal operation, loose components, and wear failure conditions. Then use signal analysis tools to analyze the vibration data in the time domain, frequency domain, and time-frequency domain. Based on these analysis results, determine the vibration features that should be extracted by the improved CNN model. Design a custom convolution kernel: Based on the determined vibration characteristics of the equipment, design a specialized custom convolution kernel. For localized, high-frequency vibration characteristics, use a 3×3 convolution kernel; for widely distributed, low-frequency vibration characteristics, use a 7×7 convolution kernel. Build a basic CNN architecture: The architecture includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and an output layer. The input layer is set according to the format and dimension of the device vibration data. In the convolutional layer, multiple convolutional layers are stacked in sequence. Each convolutional layer uses the custom convolution kernel designed previously to perform convolution operations, extracting different levels of features from the device vibration data through convolution operations. The pooling layer follows the convolution layer and uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer, reducing the amount of data and the computational complexity of the model while retaining the main feature information. The fully connected layer flattens the pooled feature map and integrates and classifies the features through multiple fully connected neurons. Finally, the output layer outputs the corresponding results according to the task requirements of the model. Introducing the attention mechanism: An attention mechanism module is introduced into the improved CNN model, and the attention mechanism module is located between the convolutional layers. The improved CNN model receives the feature map output by the previous convolutional layer, and performs weighted processing on the feature map by calculating the attention weight of each feature channel or spatial position. Specifically, the feature map is input into a subnetwork containing multiple fully connected layers, and the attention weight vector of each position is obtained after calculation. The weight vector reflects the importance of the position feature for the model to judge the operating status of the equipment. Then, the original feature map is multiplied element-by-element by the attention weight vector to obtain the feature map enhanced by the attention mechanism, and then it is input into the next convolutional layer for further processing. Through the attention mechanism module, the improved CNN model automatically focuses on the vibration features that are most valuable for equipment fault diagnosis, thereby improving the accuracy and robustness of the model.

4. A radiation room maintenance method according to claim 1, characterized in that: The specific implementation steps for the improved recurrent neural network to process temperature and humidity time series data are as follows: Data preprocessing: Continuously collect data from temperature and humidity sensors distributed throughout the radiation room, ensuring that the acquisition frequency matches the actual dynamics of the room's environmental changes. A moving average filtering algorithm is then used to clean the raw time series data, removing outliers caused by sensor failures or transient interference. Empirical mode decomposition is also used to decompose the temperature and humidity time series into multiple intrinsic mode functions (IMFs). By analyzing the frequency characteristics of each IMF component, high-frequency noise components unrelated to normal changes in the room's environment are filtered out, resulting in pure time series data. Build an improved recurrent neural network model: Based on the existing long short-term memory network model, embed a self-attention mechanism module. The self-attention mechanism module enables the long short-term memory network model to automatically pay attention to the degree of correlation between data at different time steps when processing time series data, and predict key time points that have the greatest impact on the results; Multi-scale feature extraction module: A multi-scale feature extraction module is added to the improved recurrent neural network model. The multi-scale feature extraction module consists of multiple one-dimensional convolutional layers with different convolution kernel sizes, which perform convolution operations on time series data respectively; Dynamically adjusted hidden layer structure: The number of neurons and layers in the hidden layer are dynamically adjusted based on the complexity of the input time series data and performance feedback during model training; Loss function design: For temperature and humidity time series data prediction tasks, we use mean absolute percentage error as the loss function to evaluate the relative error between the predicted value and the true value, ensuring the model's prediction accuracy for data of different magnitudes. Optimization algorithm selection and adjustment: An adaptive moment estimation optimization algorithm is used to train and update the parameters of the improved recurrent neural network model. During training, the learning rate is dynamically adjusted. A larger learning rate is set initially to quickly explore the parameter space. When the model loss value on the validation set fluctuates or no longer decreases significantly, the learning rate is gradually reduced to enable more precise parameter adjustments near the optimal solution. Real-time data prediction: After training, the improved recurrent neural network model is deployed to the radiology room maintenance system, which receives the latest data collected by temperature and humidity sensors in real time; The improved recurrent neural network model takes the current and past time series data as input, extracts and calculates features through each layer of the network in the model, and outputs the predicted values ​​of temperature and humidity environmental parameters in the future. Abnormal warning mechanism: The environmental parameter values ​​predicted by the model are compared with the pre-set normal range. If the predicted value exceeds the normal range, the system will immediately issue an abnormal warning signal. At the same time, the abnormal situation is analyzed and classified based on historical data and model prediction results.

5. A radiology room maintenance system according to any one of claims 1 to 4, comprising a sensor module, a transmission module, a central processing module and a display application module, characterized in that: The sensor module is responsible for collecting vibration data and temperature and humidity data of equipment in the computer room. The sensor has self-calibration and self-diagnosis functions, and regularly and automatically detects whether its working status is normal to ensure the accuracy and reliability of the collected data. The transmission module uses a hybrid communication method combining wireless and wired to quickly transmit the data collected by the sensor module to the central processing module. The central processing module includes a data storage unit, a data analysis unit and a maintenance strategy generation unit. The data storage unit uses distributed database technology to efficiently store and manage the massive amount of collected data. The data analysis unit runs an improved CNN model and an improved recurrent neural network model to deeply mine and analyze the data. The maintenance strategy generation unit generates specific maintenance tasks and instructions based on the analysis results; the display application module provides an operation interface for maintenance personnel and managers, including an equipment status monitoring interface, an environmental parameter display interface, a maintenance task management interface, and a remote collaboration platform. Through the intuitive visual interface, users can obtain the required information and perform corresponding operations.

Images