Primary helium fan fault diagnosis system and method based on deep learning

Abstract

The invention provides a main helium fan fault diagnosis system and method based on deep learning, and belongs to the technical field of main helium fan fault diagnosis, and the method comprises the steps: fusing vibration, current, temperature and other multi-source sensor data, and carrying out the fault mode recognition and root positioning through an attention mechanism convolutional neural network; meanwhile, a health index prediction model and a dynamic early warning threshold value are introduced, and multi-level real-time early warning is achieved; in addition, through online learning and transfer learning technologies, the diagnosis model is endowed with sustainable evolution and rapid deployment capabilities. According to the method, fault diagnosis, trend prediction, early warning decision and adaptive learning are integrated, a new thought is provided for intelligent operation and maintenance of the main helium fan, and the method has important value in the aspect of improving safety, reliability and economical efficiency of a nuclear power unit.

Description

Technical Field

[0001] The present invention relates to the technical field of main helium blower fault diagnosis, and in particular to a main helium blower fault diagnosis system and method based on deep learning. Background Art

[0002] The main helium blower is one of the key equipment in a nuclear power plant. As the main circulating power of the primary loop system of the nuclear reactor, it shoulders the important responsibilities of coolant circulation and heat transfer. Its operating status is directly related to the safety and efficiency of the nuclear power unit. However, due to its complex internal structure and harsh working environment, the main helium blower will inevitably suffer from various mechanical failures and performance degradation problems during long-term operation, such as bearing wear, blade cracking, rotor imbalance, etc. Once these failures occur, they may affect the operating efficiency of the unit at the least, or may cause accidents at the worst, resulting in major safety hazards and economic losses. Therefore, it is very important to carry out status monitoring and fault diagnosis of the main helium blower.

[0003] Traditional main helium blower fault diagnosis mainly relies on manual inspections and regular maintenance, which has many limitations: first, offline detection cannot detect early abnormalities of the equipment in time, which may miss the best time for maintenance; second, disassembly inspection will destroy the integrity of the equipment and introduce new failure risks; third, maintenance decisions are highly dependent on expert experience and lack objective quantitative basis. With the development of condition monitoring technology, a large amount of multi-source sensor data is generated during the operation of the unit, which contains important information about the health status of the equipment. How to efficiently integrate these heterogeneous data and diagnose main helium blower faults in a timely and accurate manner has become a problem that needs to be solved urgently.

[0004] In recent years, artificial intelligence technology has been widely used in the field of mechanical fault diagnosis. Among them, deep learning-based methods have achieved remarkable results in fault identification of key components such as bearings and gears by virtue of their powerful feature extraction and pattern recognition capabilities. However, most of the existing deep learning diagnosis models target a single type of signal, such as vibration signals or current signals, ignoring the correlation and complementarity between different signals. In addition, most models belong to a "black box" structure, lack of interpretability, and are difficult to gain the trust of professionals. At the same time, existing research mainly focuses on post-fault diagnosis after a fault occurs, and pays insufficient attention to fault warning and trend prediction, making it difficult to support proactive preventive maintenance decisions.

[0005] In view of the above problems, it is urgent to develop a new intelligent fault diagnosis method for main helium blowers. This method should make full use of the multi-source heterogeneous monitoring data generated during the operation of the unit to construct comprehensive health indicators; adopt an explainable artificial intelligence model to improve the credibility of the diagnosis results; introduce fault warning and remaining life prediction functions to achieve the transition from passive diagnosis to active prevention; at the same time, this method should have online adaptive capabilities, be able to continuously learn, and adapt to the dynamic changes in equipment status. This has important theoretical significance and practical value for improving the safety, reliability and economy of main helium blowers and even nuclear power units. Summary of the invention

[0006] In view of the above-mentioned existing problems, the present invention realizes real-time monitoring and fault diagnosis of the operating status of the main helium blower. The system collects multi-dimensional sensor data such as vibration, current, temperature, etc. of the main helium blower, cleans, normalizes and extracts time-frequency features of the data, and then uses an interpretable deep learning model to identify fault patterns and diagnose the root causes of faults. At the same time, the system also has a fault warning function, which can realize multi-stage fault trend prediction and warning based on the health index model of the main helium blower and the dynamically optimized warning rules. In addition, the system uses an online adaptive module to continuously optimize the deep learning model through a transfer learning method to adapt to changes in the operating status of the equipment and improve the accuracy and efficiency of fault diagnosis.

[0007] In order to solve the above technical problems, a main helium blower fault diagnosis system based on deep learning is proposed, which is characterized by including a data acquisition module, a data preprocessing module, a fault warning module, a fault diagnosis module and an online adaptive module.

[0008] The data acquisition module is used to collect multi-dimensional sensor data of vibration, current and temperature during the operation of the main helium blower.

[0009] The data preprocessing module is used to clean, normalize and extract time-frequency features from the collected raw data.

[0010] The fault warning module realizes multi-stage fault trend prediction and warning based on the health index model and dynamically optimized warning rules.

[0011] The fault diagnosis module adopts an interpretable deep learning model to perform fault mode recognition and fault root cause diagnosis based on the time-frequency characteristics.

[0012] The online adaptive module uses a transfer learning method to continuously optimize the deep learning model to adapt to changes in the operating status of the device.

[0013] As a preferred solution of the main helium blower fault diagnosis system based on deep learning described in the present invention, the data acquisition module includes a vibration sensor, a current sensor, and a temperature sensor deployed on the system to collect the main helium blower operation data in real time.

[0014] The data preprocessing module includes an outlier detection and removal submodule, a data normalization submodule and a time-frequency feature extraction submodule.

[0015] The outlier detection and removal submodule identifies and removes outliers in the collected data.

[0016] The data normalization submodule normalizes data of different dimensions to the interval [0, 1].

[0017] The time-frequency feature extraction submodule uses Fourier transform and wavelet transform to extract time-frequency domain features.

[0018] As a preferred solution of the main helium blower fault diagnosis system based on deep learning described in the present invention, the fault warning module includes a health index prediction submodule and a warning rule optimization submodule.

[0019] The health index prediction submodule establishes a main helium blower health index prediction model based on the degradation trend analysis of historical data.

[0020] The warning rule optimization submodule utilizes reinforcement learning to dynamically optimize multi-level warning rules to improve the accuracy of warnings.

[0021] The fault diagnosis module includes data fusion by feature-level fusion and decision-level fusion, adopts a convolutional neural network with an attention mechanism as a deep learning model, adaptively weights the importance of fault features, labels training samples with fault modes and root causes, and uses the Adam optimizer for module optimization and sets reasonable hyperparameters.

[0022] The fault diagnosis module also includes the use of streaming data processing technology to continuously analyze sensor data during the operation of the main helium blower. When an abnormal situation occurs, fault diagnosis is immediately performed and the results are fed back, thereby shortening the response time for fault discovery and positioning and reducing equipment downtime.

[0023] As a preferred solution of the main helium blower fault diagnosis system based on deep learning described in the present invention, the online adaptive module includes an online learning submodule and a cross-device migration submodule.

[0024] The online learning submodule regularly uses newly collected data to fine-tune the deep learning model parameters.

[0025] The cross-device migration submodule uses a transfer learning method to migrate the trained model to a new main helium blower device, thereby reducing training costs.

[0026] Another object of the present invention is to provide a main helium blower fault diagnosis method based on deep learning. The method collects multi-dimensional sensor data for real-time monitoring and fault diagnosis, uses a health index model and dynamic early warning rules to achieve fault prediction, uses an interpretable deep learning model to identify fault modes, and continuously optimizes the model through transfer learning to improve the accuracy and efficiency of fault diagnosis.

[0027] As a preferred solution of the main helium blower fault diagnosis method based on deep learning described in the present invention, the outlier detection and removal submodule includes: using a comprehensive outlier detection method based on Z-score and box plot, when the data obeys a normal distribution, the Z-score represents the degree of deviation of the data point from the mean, and the calculation formula is:

[0028]

[0029] Among them, y i is the original value of the ith data point, μ and are the mean and standard deviation of the data respectively.

[0030] Calculate the upper and lower quartiles Q1 and Q3 of the data, as well as the interquartile range IQR = Q3-Q1, and determine the abnormal value as:

[0031]

[0032] When |Z is satisfied i |When it is greater than 3 and meets criteria 1 or 2, the current data point is determined to be an outlier and is removed.

[0033] The data normalization submodule includes using the Min-Max normalization method to linearly transform the data to the [0, 1] interval:

[0034]

[0035] Among them, y min and max are the minimum and maximum values ​​of the data, respectively, i ′ is the normalized value.

[0036] The time-frequency feature extraction submodule includes performing short-time Fourier transform on the vibration signal to extract time-frequency spectrum features:

[0037]

[0038] Among them, x(t) is the time domain signal, w(t) is the window function, τ is the time shift, and f is the frequency parameter.

[0039] The current signal is decomposed by wavelet packets and expanded in both time and frequency directions to extract the features on the time and frequency scales:

[0040] W j,k (t) = 2 j / 2 W(2 j tk)

[0041] Among them, W j,k (t) is the wavelet function, W is the mother wavelet function, j is the scale parameter, k is the translation parameter, and t is the time.

[0042] As a preferred solution of the main helium blower fault diagnosis method based on deep learning described in the present invention, the health index prediction submodule includes a health index prediction model that uses a time series analysis method to analyze the degradation trend according to the historical operation data of the main helium blower and establish a gray prediction model.

[0043] The warning rule optimization submodule includes dynamically optimizing multi-level warning rules using a reinforcement learning method, setting an early warning threshold, a mid-term warning threshold, and a critical warning threshold based on the health index prediction results. If the health index continues to drop to near the mid-term warning threshold, a mid-term warning is issued in time to alert maintenance personnel. If the health index continues to drop further to the critical warning threshold, an emergency warning is issued, a fault diagnosis result is immediately given, and maintenance suggestions are provided in combination with the warning information.

[0044] As a preferred solution of the main helium blower fault diagnosis method based on deep learning described in the present invention, wherein: the deep learning model includes a convolutional neural network using an attention mechanism as a deep learning model, and the convolutional neural network of the attention mechanism is:

[0045] Convolutional layer, let x i represents the input feature map of the i-th convolutional layer, w i and b i Respectively represent the weight and bias of the convolution kernel, and perform convolution operations:

[0046] x i+1 =f(w i *x i +b i )

[0047] Where f(·) is the activation function, and f(x) = max(0, x).

[0048] Attention module generates attention weights to adjust the importance of feature channels:

[0049]

[0050] s=F ex (z,W)=σ(W2δ(W1z))

[0051]

[0052] Among them, F sq (·) is the global average pooling operation, F ex (·) is a two-layer fully connected network, σ(·) is the Sigmoid activation function, δ(·) is the ReLU function, and s is the learned attention weight vector to weight the original feature x.

[0053] The fully connected layer flattens the features extracted by the convolutional layer and maps them to the classification space, outputs the predicted probability of the fault mode and the root cause of the fault, and when the output of the last convolutional layer is x f , the weights and biases of the fully connected layer are W f and b f , the output is:

[0054] y=softmax(W f x f +b f )

[0055] Among them, the softmax function is used to normalize the output into a probability distribution:

[0056]

[0057] Attention mechanism convolutional neural network extracts high-level features through convolutional layers, the attention module adaptively adjusts the importance of features, and the fully connected layer completes the classification decision. It adopts an end-to-end learning method and optimizes the parameters of each layer through the back propagation algorithm to minimize the loss function:

[0058]

[0059] Among them, N is the number of samples, M is the number of categories, and y ij is the true label of sample i belonging to category j, The probability predicted by the network.

[0060] As a preferred solution of the main helium blower fault diagnosis method based on deep learning described in the present invention, the online learning submodule includes, based on the original trained model, using new data for short-term fine-tuning, and adopting incremental learning to regularly use newly collected real-time operation data to fine-tune the parameters of the deep learning diagnosis model.

[0061] The cross-device migration submodule includes fine-tuning the pre-trained model on the new device using only a small amount of labeled data to quickly adapt to the new device environment, and using transfer learning technology to migrate the trained diagnostic model to the new device for use.

[0062] The data fusion includes: fusing multi-source data by combining feature-level fusion and decision-level fusion, performing short-time Fourier transform on the vibration signal to obtain a time-frequency spectrum, decomposing the current signal through wavelet packets, extracting wavelet coefficients of different frequency bands, fusing the vibration signal and the current signal through multi-source data, inputting the attention mechanism convolutional neural network for training and identification, and obtaining fault mode and root cause diagnosis results.

[0063] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of a method described in a main helium blower fault diagnosis based on deep learning are implemented.

[0064] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of a method described in a main helium blower fault diagnosis based on deep learning are implemented.

[0065] Beneficial effects of the present invention: The present invention first collects multi-dimensional sensor data during the operation of the main helium blower, and through data preprocessing and feature extraction, fully mines the fault information contained in different physical quantities, and constructs a comprehensive feature set that fully reflects the health status of the equipment. The system uses a combination of feature-level fusion and decision-level fusion to analyze the equipment status from multiple angles, improves the accuracy and stability of diagnosis, and avoids the limitations of a single data source.

[0066] Second, the attention mechanism is innovatively introduced into the convolutional neural network. The attention module adaptively adjusts the weights of different features, highlights key fault features, and reduces the interference of redundant information. At the same time, the classification output of the model directly corresponds to different fault modes and root causes, and the diagnosis results are clear. The use of an interpretable attention mechanism convolutional neural network can achieve efficient and accurate fault mode recognition and root cause location.

[0067] Third, a health index prediction model for the main helium blower was constructed. By learning the historical degradation trend of the equipment, the future health status of the equipment can be accurately estimated. On this basis, the system uses reinforcement learning to dynamically optimize the multi-level warning thresholds, and sets multiple levels such as early warning, mid-term warning, and critical warning for different health levels. This real-time dynamic warning mechanism can help operation and maintenance personnel to promptly detect abnormal conditions of the equipment, formulate maintenance plans in advance, effectively avoid failure risks, and maximize the safe and economical operation of the unit.

[0068] Fourth, the online adaptive module of the present invention adopts an incremental learning method to regularly fine-tune the model using newly collected data, so that the diagnostic model can continue to learn and evolve and adapt to the dynamic changes in the device status. At the same time, through transfer learning technology, the trained model can be quickly migrated to new devices, greatly reducing the cost of repeated modeling and improving the deployment efficiency and scalability of the system.

[0069] Fifth, it organically combines fault diagnosis, health prediction, and early warning strategy optimization to form a complete set of intelligent operation and maintenance solutions. The system can achieve health management of the equipment throughout its life cycle, empowering various links such as status monitoring, fault diagnosis, maintenance strategy formulation, and spare parts management of the main helium blower, promoting professionals to shift from passive maintenance to active prevention, fundamentally improving the safety, reliability, and economy of power plants, and setting a new benchmark for intelligent operation and maintenance in the industry.

[0070] In summary, the present invention integrates multi-source heterogeneous data and uses explainable artificial intelligence algorithms to build an intelligent operation and maintenance system for main helium blowers that integrates fault diagnosis, trend prediction, early warning decision-making, and adaptive evolution. The system has made breakthrough progress in improving diagnostic accuracy, extending early warning time, and optimizing maintenance decisions, providing a solid technical guarantee for the safe and efficient operation of the main helium blowers, and has significant innovation and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0072] Figure 1 A structural principle block diagram of a main helium blower fault diagnosis system based on deep learning provided for one embodiment of the present invention.

[0073] Figure 2 A data preprocessing flowchart of a main helium blower fault diagnosis system based on deep learning is provided for one embodiment of the present invention.

[0074] Figure 3 A main helium blower health status prediction and early warning flow chart of a main helium blower fault diagnosis system based on deep learning is provided for one embodiment of the present invention.

[0075] Figure 4 A convolutional neural network structure diagram of an attention mechanism for a main helium blower fault diagnosis system based on deep learning provided for an embodiment of the present invention.

[0076] Figure 5 A deep learning-based fault diagnosis flowchart of a main helium blower fault diagnosis system based on deep learning is provided as an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0078] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0079] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an embodiment that is mutually exclusive with other embodiments, either individually or selectively.

[0080] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0081] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0082] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0083] Example 1, reference Figure 1-Figure 5 , which is the first embodiment of the present invention, provides a main helium blower fault diagnosis system based on deep learning, including a data acquisition module, a data preprocessing module, a fault diagnosis module, a fault warning module and an online adaptive module.

[0084] Furthermore, the data acquisition module is equipped with vibration sensors, current sensors, temperature sensors, etc., which are used to collect multi-dimensional operating data of the main helium blower in real time. These sensors can fully sense the changes in various physical quantities of the main helium blower during operation, providing reliable data support for subsequent fault diagnosis.

[0085] The data preprocessing module includes submodules such as outlier detection and removal, data normalization, and time-frequency feature extraction. The specific technical solutions are as follows: Figure 2 As shown; the outlier detection and removal submodule is used to identify and remove outliers in the collected data to ensure the accuracy and reliability of the data; the data normalization submodule normalizes data of different dimensions to the [0, 1] interval to eliminate the impact of the dimension on the analysis; the time-frequency feature extraction submodule uses Fourier transform, wavelet transform and other methods to extract time-frequency domain features from the original data to provide effective feature input for subsequent fault diagnosis.

[0086] In this embodiment, the technical solutions of each submodule in the data preprocessing module are as follows:

[0087] The constant value detection and removal submodule adopts a comprehensive outlier detection method based on Z-score and box plot. When the data follows a normal distribution, the Z-score represents the degree of deviation of the data point from the mean, and the calculation formula is:

[0088]

[0089] Among them, y i is the original value of the ith data point, μ and are the mean and standard deviation of the data respectively.

[0090] Calculate the upper and lower quartiles Q1 and Q3 of the data, as well as the interquartile range IQR = Q3-Q1, and determine the abnormal value as:

[0091]

[0092] When |Z is satisfied i |When it is greater than 3 and meets criteria 1 or 2, the current data point is determined to be an outlier and is removed.

[0093] The data normalization submodule includes using the Min-Max normalization method to linearly transform the data to the [0, 1] interval:

[0094]

[0095] Among them, y min and max are the minimum and maximum values ​​of the data, respectively, i is the normalized value.

[0096] The time-frequency feature extraction submodule includes performing short-time Fourier transform on the vibration signal to extract time-frequency spectrum features:

[0097]

[0098] Among them, x(t) is the time domain signal, w(t) is the window function, τ is the time shift, and f is the frequency parameter.

[0099] The current signal is decomposed by wavelet packets and expanded in both time and frequency directions to extract the features on the time and frequency scales:

[0100] W j,k (t) = 2 j / 2 W(2 j tk)

[0101] Among them, W j,k (t) is the wavelet function, W is the mother wavelet function, j is the scale parameter, k is the translation parameter, t is the time, and the decomposed wavelet coefficients reflect the characteristics of the signal at different time and frequency scales.

[0102] It should be noted that the fault warning module includes a health index prediction submodule and a warning rule optimization submodule. Its workflow is as follows: Figure 3 As shown in the figure, the health index prediction submodule establishes a health index prediction model for the main helium blower based on the degradation trend analysis of historical data, which can predict the future health status of the equipment. The early warning rule optimization submodule uses reinforcement learning methods to dynamically optimize multi-level early warning rules, such as early warning thresholds, mid-term warning thresholds, etc., to improve the accuracy of early warnings.

[0103] In this embodiment, the technical solution of the fault warning module is as follows:

[0104] The health index prediction submodule adopts the time series analysis method to analyze the degradation trend of the main helium blower based on its historical operation data, and establishes a health index prediction model including a grey prediction model and an exponential smoothing model, which can predict the health status of the equipment within a certain period of time in the future; the health index can comprehensively reflect multiple key status parameters of the equipment, such as vibration, current, temperature, etc., and is used to comprehensively evaluate the overall health level of the equipment.

[0105] The warning rule optimization submodule uses reinforcement learning methods to dynamically optimize multi-level warning rules. According to the health index prediction results, it sets early warning thresholds, mid-term warning thresholds, and critical warning thresholds. Through the reward and punishment mechanism, it continuously adjusts these threshold parameters so that the warning rules can adapt to the actual operating conditions of the equipment and improve the accuracy of the warning. If the health index continues to drop to near the mid-term warning threshold, a mid-term warning is issued in time to remind maintenance personnel to pay attention. If it further drops to the critical warning threshold, an emergency warning is issued and maintenance is recommended as soon as possible.

[0106] The health index prediction curve and multi-level warning threshold are intuitively displayed to maintenance personnel in a graphical form. When an abnormal situation is detected, the fault diagnosis result is immediately given, and maintenance suggestions are provided in combination with the warning information. This can help maintenance personnel quickly understand the current health status of the equipment, take targeted maintenance measures, and reduce equipment downtime. In short, the fault warning module makes full use of time series analysis and reinforcement learning technology to achieve the prediction of the health status of the main helium fan and multi-level warning, providing strong support for the reliable operation and maintenance of the equipment.

[0107] It should also be noted that the fault diagnosis module uses a convolutional neural network with an attention mechanism as a deep learning model, such as Figure 4 As shown in the figure, the model can adaptively weight the importance of fault features and improve the interpretability of diagnosis. The training samples are labeled based on expert experience, and the fault mode and root cause labels are given to the training samples. The fault diagnosis model is optimized using the Adam optimizer, and reasonable hyperparameters such as learning rate and batch size are set.

[0108] The attention mechanism convolutional neural network mainly includes the following modules:

[0109] Convolutional layer, let x i represents the input feature map of the i-th convolutional layer, w i and b i Respectively represent the weight and bias of the convolution kernel, and perform convolution operations:

[0110] x i+1 =f(w i *x i +b i )

[0111] Where f(·) is the activation function, and f(x) = max(0, x).

[0112] Attention module generates attention weights to adjust the importance of feature channels:

[0113]

[0114] s=F ex (z,W)=σ(W2δ(W1z))

[0115]

[0116] Among them, F sq (·) is the global average pooling operation, F ex (·) is a two-layer fully connected network, σ(·) is the Sigmoid activation function, δ(·) is the ReLU function, and s is the learned attention weight vector to weight the original feature x.

[0117] The fully connected layer flattens the features extracted by the convolutional layer and maps them to the classification space, outputs the predicted probability of the fault mode and the root cause of the fault, and when the output of the last convolutional layer is x f , the weights and biases of the fully connected layer are W f and b f , the output is:

[0118] y=softmax(W f x f +b f )

[0119] Among them, the softmax function is used to normalize the output into a probability distribution:

[0120]

[0121] Attention mechanism convolutional neural network extracts high-level features through convolutional layers, the attention module adaptively adjusts the importance of features, and the fully connected layer completes the classification decision. It adopts an end-to-end learning method and optimizes the parameters of each layer through the back propagation algorithm to minimize the loss function:

[0122]

[0123] Among them, N is the number of samples, M is the number of categories, and y ij is the true label of sample i belonging to category j, The probability predicted by the network.

[0124] Preferably, the online adaptive module includes an online learning submodule and a cross-device migration submodule. The online learning submodule can regularly use newly collected data to fine-tune the deep learning model so that it can adapt to changes in the operating status of the device. The cross-device migration submodule uses the transfer learning method to migrate the trained model to the new main helium blower device, greatly reducing the training cost.

[0125] In this embodiment, the specific technical solution of the online adaptive module is as follows:

[0126] The online learning submodule adopts an incremental learning approach and regularly uses newly collected real-time operating data to fine-tune the parameters of the deep learning diagnostic model. The specific operation is: based on the original trained model, use the new data for short-term fine-tuning so that the model can adapt to changes in the equipment's operating status in a timely manner. A smaller learning rate can be used during fine-tuning to prevent the model from being completely changed by new data and losing its original diagnostic capabilities. Through periodic online learning, the diagnostic model can continue to maintain the best diagnostic performance.

[0127] The cross-device migration submodule uses transfer learning technology to migrate the diagnostic model trained on a main helium blower device to a new device for use. The specific approach is: on the new device, only a small amount of labeled data is needed to fine-tune the pre-trained model, so that it can quickly adapt to the new device environment. This greatly reduces the modeling cost of the new device and avoids the need to train the model from scratch. Transfer learning can make full use of the knowledge accumulated by the existing diagnostic model to improve the diagnostic accuracy and generalization ability of the new device.

[0128] Through online learning and cross-device migration, the diagnostic model can continuously optimize its own parameters and adapt to the dynamic changes in the equipment's operating status. This ensures that the fault diagnosis system can maintain stable and reliable diagnostic performance for a long time, reducing the decline in diagnostic accuracy due to factors such as equipment aging and environmental changes. At the same time, it also greatly reduces the cost and time of deploying the diagnostic system on new equipment and improves the deployment flexibility of the system. The online adaptive module makes full use of incremental learning and transfer learning technologies, allowing the diagnostic system to continuously optimize itself, achieve rapid response and adaptation to changes in the equipment's operating status, and provide strong support for the reliable operation and maintenance of the main helium blower throughout its life cycle.

[0129] In addition, the system also uses a combination of feature-level fusion and decision-level fusion to fuse multi-source data. The fusion diagnosis process is as follows: Figure 5As shown in the figure, taking the vibration signal as an example, short-time Fourier transform is first performed on it to obtain a time-frequency spectrum; the current signal is decomposed through wavelet packets to extract wavelet coefficients in different frequency bands. The two types of features are then fused at the feature level and input into the attention mechanism convolutional neural network for training and recognition, and finally the fault mode and root cause diagnosis results are given. This diagnostic method integrates multi-source heterogeneous data and has excellent diagnostic performance.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0131] Embodiment 2, the third embodiment of the present invention, is different from the first two embodiments in that:

[0132] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0133] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0134] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0135] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0136] Example 3, reference Figure 1-Figure 5 , which is the fourth embodiment of the present invention, and provides a main helium blower fault diagnosis method based on deep learning, comprising:

[0137] Preferably, the outlier detection and removal submodule includes using a comprehensive outlier detection method based on Z-score and box plot. When the data obeys a normal distribution, the Z-score represents the degree of deviation of the data point from the mean, and the calculation formula is:

[0138]

[0139] Among them, x i is the original value of the ith data point, μ and σ are the mean and standard deviation of the data respectively.

[0140] Calculate the upper and lower quartiles Q1 and Q3 of the data, as well as the interquartile range IQR = Q3-Q1, and determine the abnormal value as:

[0141]

[0142] When |Z is satisfied i |When it is greater than 3 and meets criteria 1 or 2, the current data point is determined to be an outlier and is removed.

[0143] The data normalization submodule includes using the Min-Max normalization method to linearly transform the data to the [0, 1] interval:

[0144]

[0145] Among them, x min and x max are the minimum and maximum values ​​of the data, respectively, and x′ i is the normalized value.

[0146] The time-frequency feature extraction submodule includes performing short-time Fourier transform on the vibration signal to extract time-frequency spectrum features:

[0147]

[0148] Among them, x(t) is the time domain signal, w(t) is the window function, τ is the time shift, and f is the frequency parameter.

[0149] The current signal is decomposed by wavelet packets and expanded in both time and frequency directions to extract the features on the time and frequency scales:

[0150] W j,k (t) = 2 j / 2 W(2 j tk)

[0151] Among them, W(t) is the mother wavelet function, j is the scale parameter, and k is the translation parameter.

[0152] Furthermore, the health index prediction submodule includes a health index prediction model that uses a time series analysis method to analyze the degradation trend based on the historical operating data of the main helium blower and establish a grey prediction model.

[0153] The warning rule optimization submodule includes dynamically optimizing multi-level warning rules using a reinforcement learning method, setting an early warning threshold, a mid-term warning threshold, and a critical warning threshold based on the health index prediction results. If the health index continues to drop to near the mid-term warning threshold, a mid-term warning is issued in time to alert maintenance personnel. If the health index continues to drop further to the critical warning threshold, an emergency warning is issued, a fault diagnosis result is immediately given, and maintenance suggestions are provided in combination with the warning information.

[0154] It should be noted that the deep learning model includes a convolutional neural network using an attention mechanism as the deep learning model, and the convolutional neural network using the attention mechanism is:

[0155] Convolutional layer, let x i represents the input feature map of the i-th convolutional layer, w i and b iRespectively represent the weight and bias of the convolution kernel, and perform convolution operations:

[0156] x i+1 =f(w i *x i +b i )

[0157] Where f(·) is the activation function, and f(x) = max(0, x).

[0158] Attention module generates attention weights to adjust the importance of feature channels:

[0159]

[0160] s=F ex (z,W)=σ(W2δ(W1z))

[0161]

[0162] Among them, F sq (·) is the global average pooling operation, F ex (·) is a two-layer fully connected network, σ(·) is the Sigmoid activation function, δ(·) is the ReLU function, and s is the learned attention weight vector to weight the original feature x.

[0163] The fully connected layer flattens the features extracted by the convolutional layer and maps them to the classification space, outputs the predicted probability of the fault mode and the root cause of the fault, and when the output of the last convolutional layer is x f , the weights and biases of the fully connected layer are W f and b f , the output is:

[0164] y=softmax(W f x f +b f )

[0165] Among them, the softmax function is used to normalize the output into a probability distribution:

[0166]

[0167] Attention mechanism convolutional neural network extracts high-level features through convolutional layers, the attention module adaptively adjusts the importance of features, and the fully connected layer completes the classification decision. It adopts an end-to-end learning method and optimizes the parameters of each layer through the back propagation algorithm to minimize the loss function:

[0168]

[0169] Among them, N is the number of samples, M is the number of categories, and y ij is the true label of sample i belonging to category j, The probability predicted by the network.

[0170] It should also be noted that the online learning submodule includes, based on the original trained model, using new data for short-term fine-tuning, and adopting an incremental learning approach to regularly use newly collected real-time operating data to fine-tune the parameters of the deep learning diagnostic model.

[0171] The cross-device migration submodule includes fine-tuning the pre-trained model on the new device using only a small amount of labeled data to quickly adapt to the new device environment, and using transfer learning technology to migrate the trained diagnostic model to the new device for use.

[0172] The data fusion includes: fusing multi-source data by combining feature-level fusion and decision-level fusion, performing short-time Fourier transform on the vibration signal to obtain a time-frequency spectrum, decomposing the current signal through wavelet packets, extracting wavelet coefficients of different frequency bands, fusing the vibration signal and the current signal through multi-source data, inputting the attention mechanism convolutional neural network for training and identification, and obtaining fault mode and root cause diagnosis results.

[0173] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A main helium blower fault diagnosis system based on deep learning, characterized in that: It includes data acquisition module, data preprocessing module, fault warning module, fault diagnosis module and online adaptive module; The data acquisition module is used to collect multi-dimensional sensor data of vibration, current and temperature during the operation of the main helium blower; The data preprocessing module is used to clean, normalize and extract time-frequency features of the collected raw data; The fault warning module implements multi-stage fault trend prediction and warning based on the health index model and dynamically optimized warning rules; The fault diagnosis module adopts an interpretable deep learning model to perform fault mode recognition and fault root cause diagnosis based on the time-frequency characteristics; The online adaptive module uses a transfer learning method to continuously optimize the deep learning model to adapt to changes in the operating status of the device.

2. A main helium blower fault diagnosis system based on deep learning as claimed in claim 1, characterized in that: The data acquisition module includes a vibration sensor, a current sensor, and a temperature sensor deployed on the system to collect the main helium blower operation data in real time; The data preprocessing module includes an outlier detection and removal submodule, a data normalization submodule and a time-frequency feature extraction submodule; The outlier detection and removal submodule identifies and removes outliers in the collected data; The data normalization submodule normalizes data of different dimensions to the interval [0, 1]; The time-frequency feature extraction submodule uses Fourier transform and wavelet transform to extract time-frequency domain features.

3. A main helium blower fault diagnosis system based on deep learning as claimed in claim 2, characterized in that: The fault warning module includes a health index prediction submodule and a warning rule optimization submodule; The health index prediction submodule establishes a main helium blower health index prediction model based on the degradation trend analysis of historical data; The warning rule optimization submodule uses reinforcement learning to dynamically optimize multi-level warning rules to improve the accuracy of warnings; The fault diagnosis module includes data fusion by feature-level fusion and decision-level fusion, uses a convolutional neural network with an attention mechanism as a deep learning model, adaptively weights the importance of fault features, labels training samples with fault modes and root causes, and uses the Adam optimizer for module optimization and sets reasonable hyperparameters; The fault diagnosis module also includes the use of streaming data processing technology to continuously analyze sensor data during the operation of the main helium blower. When an abnormal situation occurs, fault diagnosis is immediately performed and the results are fed back, thereby shortening the response time for fault discovery and positioning and reducing equipment downtime.

4. A main helium blower fault diagnosis system based on deep learning as claimed in claim 3, characterized in that: The online adaptive module includes an online learning submodule and a cross-device migration submodule; The online learning submodule regularly uses newly collected data to fine-tune the deep learning model parameters; The cross-device migration submodule uses a transfer learning method to migrate the trained model to a new main helium blower device, thereby reducing training costs.

5. A method using a main helium blower fault diagnosis system based on deep learning as claimed in any one of claims 1 to 4, characterized in that: The outlier detection and removal submodule includes a comprehensive outlier detection method based on Z-score and box plot. When the data obeys a normal distribution, the Z-score represents the degree of deviation of the data point from the mean, and the calculation formula is: Among them, y i is the original value of the ith data point, μ and are the mean and standard deviation of the data respectively; Calculate the upper and lower quartiles Q1 and Q3 of the data, as well as the interquartile range IQR = Q3-Q1, and determine the abnormal value as: When |Z is satisfied i |When it is greater than 3 and meets criteria 1 or 2, the current data point is considered an outlier and is removed; The data normalization submodule includes using the Min-Max normalization method to linearly transform the data to the [0, 1] interval: Among them, y min and max are the minimum and maximum values ​​of the data, respectively, i ′ is the normalized value; The time-frequency feature extraction submodule includes performing short-time Fourier transform on the vibration signal to extract time-frequency spectrum features: Among them, x(t) is the time domain signal, w(t) is the window function, τ is the time shift, and f is the frequency parameter; The current signal is decomposed by wavelet packets and expanded in both time and frequency directions to extract the features on the time and frequency scales: W j,k (t)=2 j / 2 W(2 j t-k) Among them, W j,k (t) is the wavelet function, W is the mother wavelet function, j is the scale parameter, k is the translation parameter, and t is the time.

6. A method for diagnosing main helium blower faults based on deep learning as claimed in claim 5, characterized in that: The health index prediction submodule includes a health index prediction model that uses a time series analysis method to analyze the degradation trend based on the historical operation data of the main helium blower and establish a grey prediction model; The warning rule optimization submodule includes dynamically optimizing multi-level warning rules using a reinforcement learning method, setting an early warning threshold, a mid-term warning threshold, and a critical warning threshold based on the health index prediction results. If the health index continues to drop to near the mid-term warning threshold, a mid-term warning is issued in time to alert maintenance personnel. If the health index continues to drop further to the critical warning threshold, an emergency warning is issued, a fault diagnosis result is immediately given, and maintenance suggestions are provided in combination with the warning information.

7. A main helium blower fault diagnosis method based on deep learning as claimed in claim 6, characterized in that: The deep learning model includes a convolutional neural network with an attention mechanism as a deep learning model, and the convolutional neural network with the attention mechanism is: Convolutional layer, let x i represents the input feature map of the i-th convolutional layer, w i and b i Respectively represent the weight and bias of the convolution kernel, and perform convolution operations: x i+1 =f(w i *x i +b i ) Where f(·) is the activation function, f(x) = max(0,x); Attention module generates attention weights to adjust the importance of feature channels: s=F ex (z,W)=σ(W2δ(W1z)) Among them, F sq (·) is the global average pooling operation, F ex (·) is a two-layer fully connected network, σ(·) is the Sigmoid activation function, δ(·) is the ReLU function, and s is the learned attention weight vector to weight the original feature x; The fully connected layer flattens the features extracted by the convolutional layer and maps them to the classification space, outputs the predicted probability of the fault mode and the root cause of the fault, and when the output of the last convolutional layer is x f , the weights and biases of the fully connected layer are W f and b f , the output is: y =softmax(W f x f +b f ) Among them, the softmax function is used to normalize the output into a probability distribution: Attention mechanism convolutional neural network extracts high-level features through convolutional layers, the attention module adaptively adjusts the importance of features, and the fully connected layer completes the classification decision. It adopts an end-to-end learning method and optimizes the parameters of each layer through the back propagation algorithm to minimize the loss function: Among them, N is the number of samples, M is the number of categories, and y ij is the true label of sample i belonging to category j, The probability predicted by the network.

8. A method for diagnosing main helium blower faults based on deep learning as claimed in claim 7, characterized in that: The online learning submodule includes, based on the original trained model, using new data for short-term fine-tuning, using incremental learning to regularly use newly collected real-time operation data to fine-tune the parameters of the deep learning diagnostic model; The cross-device migration submodule includes fine-tuning the pre-trained model on the new device using only a small amount of labeled data to quickly adapt to the new device environment, and using transfer learning technology to migrate the trained diagnostic model to the new device for use; The data fusion includes: fusing multi-source data by combining feature-level fusion and decision-level fusion, performing short-time Fourier transform on the vibration signal to obtain a time-frequency spectrum, decomposing the current signal through wavelet packets, extracting wavelet coefficients of different frequency bands, fusing the vibration signal and the current signal through multi-source data, inputting the attention mechanism convolutional neural network for training and identification, and obtaining fault mode and root cause diagnosis results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 5 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 5 to 8 are implemented.

Images