Centrifugal machine remote monitoring method and system, electronic equipment and storage medium

Abstract

The invention provides a centrifugal machine remote monitoring method and system, electronic equipment and a storage medium, and belongs to the technical field of industrial equipment monitoring, and the method comprises the steps: obtaining multi-dimensional operation state data collected in real time by an edge computing node deployed on a centrifugal machine body, constructing an operation state tensor with time and space correlation characteristics based on the multi-dimensional operation state data; performing multi-scale feature extraction on the running state tensor based on a depth time sequence clustering algorithm to obtain a first feature vector set, and determining a working mode of the centrifugal machine based on the first feature vector set and a feature-working mode mapping table; and constructing a monitoring strategy matrix based on a federated learning framework, activating the corresponding sensor group according to the working mode and the monitoring strategy matrix, and monitoring the centrifugal machine by using the sensor group. According to the centrifugal machine remote monitoring method and system, the electronic equipment and the storage medium, the complex operation state of the centrifugal machine can be accurately monitored and analyzed.

Description

Technical Field

[0001] The present application belongs to the technical field of industrial equipment monitoring, and more specifically, relates to a centrifuge remote monitoring method and system, electronic equipment, and storage medium. Background Art

[0002] In modern industrial production, centrifuges are widely used as key equipment in chemical, pharmaceutical, food and other fields. The stability of the centrifuge's operating state directly affects production efficiency and product quality. The current centrifuge monitoring methods have problems such as incomplete data collection and poor real-time performance, making it difficult to accurately monitor and analyze the complex operating status of the centrifuge.

[0003] Therefore, there is an urgent need for a centrifuge remote monitoring method and system, electronic equipment, and storage medium. Summary of the Invention

[0004] The purpose of this application is to provide a centrifuge remote monitoring method and system, electronic equipment, and storage medium to improve the accurate monitoring and analysis of the complex operating conditions of the centrifuge.

[0005] A first aspect of an embodiment of the present application provides a centrifuge remote monitoring method, comprising: Acquire multi-dimensional operating status data collected in real time by edge computing nodes deployed on the centrifuge body, and construct an operating status tensor with temporal and spatial correlation characteristics based on the multi-dimensional operating status data; Performing multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determining the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table; A monitoring strategy matrix is ​​constructed based on a federated learning framework, a corresponding sensor group is activated according to the working mode and the monitoring strategy matrix, and the centrifuge is monitored using the sensor group.

[0006] A second aspect of the embodiments of the present application provides a centrifuge remote monitoring system, comprising: A data processing module is used to obtain multi-dimensional operating status data collected in real time by edge computing nodes deployed on the centrifuge body, and to construct an operating status tensor with temporal and spatial correlation characteristics based on the multi-dimensional operating status data; a mode determination module, configured to perform multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determine the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table; The monitoring module is used to build a monitoring strategy matrix based on the federated learning framework, activate the corresponding sensor group according to the working mode and the monitoring strategy matrix, and use the sensor group to monitor the centrifuge.

[0007] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned centrifuge remote monitoring method when executing the computer program.

[0008] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned centrifuge remote monitoring method are implemented.

[0009] The beneficial effects of the centrifuge remote monitoring method and system, electronic device, and storage medium provided in the embodiments of the present application are: by obtaining multi-dimensional operating status data collected by edge computing nodes and constructing an operating status tensor, the present application can comprehensively and accurately reflect the operating status of the centrifuge, providing a richer and more accurate data basis for monitoring. Secondly, the deep time series clustering algorithm is used for multi-scale feature extraction, which can deeply mine the key features in the data, accurately determine the centrifuge operating mode in combination with the feature-operating mode mapping table, and improve the cognitive accuracy of the equipment operating status. Finally, a monitoring strategy matrix is ​​constructed based on the federated learning framework, and the corresponding sensor group is activated according to the operating mode to realize the intelligent dynamic allocation of sensor resources, which not only reduces unnecessary resource consumption, but also can perform efficient monitoring for different working conditions, effectively improve monitoring efficiency and accuracy, timely detect potential faults, ensure stable operation of the centrifuge, and reduce maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0011] Figure 1 A flow chart of a centrifuge remote monitoring method provided in one embodiment of the present application; Figure 2 A structural block diagram of a centrifuge remote monitoring system provided in one embodiment of the present application; Figure 3 A schematic block diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0012] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0013] In order to make the purpose, technical solutions and advantages of this application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0014] Please refer to Figure 1 , Figure 1 A flow chart of a centrifuge remote monitoring method provided in one embodiment of the present application, the method comprising: S101: Acquire multi-dimensional operating status data collected in real time by edge computing nodes deployed on the centrifuge body, and construct an operating status tensor with time and space correlation characteristics based on the multi-dimensional operating status data.

[0015] In this embodiment, edge computing nodes deployed within the centrifuge body collect multi-dimensional operational status data in real time, including but not limited to parameters such as the centrifuge's speed, temperature, vibration frequency, torque, pressure, and flow rate. In this embodiment, edge computing nodes can collect data using various sensors located at key locations within the centrifuge to comprehensively and accurately reflect the centrifuge's operating status. Among these, speed sensors monitor the centrifuge's rotational speed in real time, temperature sensors detect heating of key components, vibration frequency sensors capture vibration characteristics during centrifuge operation, torque sensors measure the torque output of the drive motor, and pressure sensors monitor pressure changes within the centrifuge.

[0016] In this embodiment, the collected multi-dimensional operating status data needs to be preprocessed. The preprocessing method includes: performing data cleaning on the multi-dimensional operating status data to remove noise, outliers, and missing values. For example, by setting a reasonable threshold range, speed or temperature data that exceeds the normal range is identified and eliminated; missing values ​​are filled using interpolation or prediction methods based on historical data. The multi-dimensional operating status data is then normalized to unify different types of data into the same numerical range for subsequent analysis and processing.

[0017] In this embodiment, a method for constructing an operating status tensor with temporal and spatial correlation characteristics based on preprocessed multi-dimensional operating status data includes: in the temporal dimension, using the time sequence of the multi-dimensional operating status data to reflect the temporal trend of the centrifuge's operating status; in the spatial dimension, combining the sensor installation locations to reflect the correlation between the operating status of different parts. For example, data collected by different sensors at different time points is organized according to certain rules into a three-dimensional tensor, where one dimension represents time and the other two dimensions represent different sensor locations and data types, respectively.

[0018] S102: Perform multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determine the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table.

[0019] In this embodiment, the deep time series clustering algorithm can automatically mine the feature information at different scales in the operating status tensor, such as short-term fluctuation characteristics and long-term trend characteristics. Therefore, this embodiment can perform multi-layer convolution and pooling operations on the operating status tensor through the deep time series clustering algorithm to extract representative feature vectors and obtain the first feature vector set.

[0020] In this embodiment, a feature-to-operating mode mapping table is generated by analyzing and clustering a large amount of historical operating data. It records the correspondence between different feature vector combinations and centrifuge operating modes. It specifies the feature vector characteristics corresponding to each of the start mode, stop mode, normal operating mode, and fault mode. After matching the extracted first feature vector set with the feature vectors in the mapping table, the current operating mode of the centrifuge can be determined based on the matching results. For example, when the feature vector set exhibits a specific combination of speed, temperature, and vibration frequency, it corresponds to the normal operating mode of the centrifuge; however, when abnormal vibration characteristics and temperature rise occur, it corresponds to a faulty operating mode.

[0021] S103: Construct a monitoring strategy matrix based on the federated learning framework, activate the corresponding sensor group according to the working mode and the monitoring strategy matrix, and use the sensor group to monitor the centrifuge.

[0022] In this embodiment, the federated learning framework allows collaborative learning among multiple participants without sharing original data, thus protecting data privacy and security. This embodiment collects operating data and monitoring experience from different centrifuges, uses a federated learning algorithm to train a universal monitoring strategy model, and converts it into a monitoring strategy matrix, which includes various monitoring strategies for different operating modes, such as monitoring frequency, monitoring parameter range, etc.

[0023] During centrifuge monitoring, the centrifuge will focus on different operating parameters in different operating modes, so the corresponding sensor group needs to be activated for targeted monitoring. For example, in the centrifuge's startup mode, the focus is on monitoring changes in speed, torque, and vibration frequency. In stop mode, the speed, torque, and vibration frequency are also monitored, and the monitoring frequency can be appropriately reduced. In normal operation mode, only speed and temperature need to be monitored regularly. In fault warning mode, real-time monitoring of vibration frequency and torque is required, and the corresponding sensor group is activated at this time. The sensor group transmits the real-time collected data back to the edge computing node, which performs preliminary processing and analysis on the data and uploads the results to the remote monitoring center. The remote monitoring center can further evaluate the operating status of the centrifuge based on the received data, promptly identify potential problems, and take appropriate measures.

[0024] From the above, it can be concluded that this application can comprehensively and accurately reflect the operating status of the centrifuge by obtaining multi-dimensional operating status data collected by edge computing nodes and constructing an operating status tensor, providing a richer and more accurate data basis for monitoring. Secondly, the deep time series clustering algorithm is used for multi-scale feature extraction, which can deeply mine the key features in the data, and accurately determine the working mode of the centrifuge in combination with the feature-working mode mapping table, thereby improving the accuracy of the cognition of the equipment's operating status. Finally, a monitoring strategy matrix is ​​constructed based on the federated learning framework, and the corresponding sensor group is activated according to the working mode to realize the intelligent dynamic allocation of sensor resources, which not only reduces unnecessary resource consumption, but also can perform efficient monitoring for different working conditions, effectively improve monitoring efficiency and accuracy, timely detect potential faults, ensure the stable operation of the centrifuge, and reduce maintenance costs.

[0025] In one embodiment of the present application, an operating state tensor with temporal and spatial correlation characteristics is constructed based on multi-dimensional operating state data, including: The multi-dimensional operating status data is segmented according to the preset time window, and the time domain statistical features of the data in the time window are extracted to generate a time series matrix; the time domain statistical features include: mean, variance and peak value; Construct an adjacency matrix based on the centrifuge structure topology and sensor deployment locations; The time series matrix and the adjacency matrix are fused through the spatiotemporal graph convolutional network to obtain the operating state tensor. The three dimensions of the operating state tensor represent the time step, sensor node and feature channel respectively.

[0026] In this embodiment, the preset time window is set based on the centrifuge's operating characteristics and the frequency of data changes. For example, for a centrifuge with relatively stable operation and slowly changing data, a relatively long time window, such as 10 minutes, can be set; whereas for a centrifuge with frequent parameter changes during operation, a shorter time window, such as 1 minute, should be set. This embodiment uses preset time windows to divide continuous multi-dimensional operating status data into multiple discrete time periods, facilitating subsequent analysis of the data characteristics within each time period.

[0027] In this embodiment, the time domain statistical features of the data in each time window are extracted. These features include mean, variance and peak value. Among them, mean can reflect the average level of the data in the time window and reflect the central tendency of the data. Variance measures the degree of dispersion of the data relative to the mean. The larger the variance, the greater the fluctuation of the data. Peak value refers to the maximum value reached by the data in the time window. For certain key parameters, such as torque peak value, it can reflect the maximum load borne by the centrifuge at a specific moment. Specifically, the mean value takes the speed data as an example. The mean of the speed in each time window is calculated to understand the average speed of the centrifuge in this time period. The variance takes the vibration frequency as an example. The variance of the vibration frequency data in each time window is calculated. If the variance is relatively large, it indicates that the vibration of the centrifuge in this time window is unstable.

[0028] In this embodiment, generating a time series matrix based on the extracted time-domain statistical features involves arranging the mean, variance, and peak values ​​corresponding to each time window into rows or columns of the matrix in a specific order. For example, each row of the matrix can represent a time window, and each column can correspond to a different time-domain statistical feature (mean, variance, peak value). Different columns can also correspond to different types of operating status data (such as speed, temperature, vibration frequency, etc.). In this way, the time series matrix can clearly display the changes in the time-domain statistical features of the multi-dimensional operating status data within different time windows.

[0029] In this embodiment, the centrifuge's structural topology is derived from the physical connections and interactions between its components. Specifically, a centrifuge typically consists of multiple components, including a rotor, motor, bearings, and housing. Various relationships exist between these components, including mechanical transmission, heat conduction, and vibration transfer. For example, the rotor is connected to the motor via a transmission device, and the motor's operation drives the rotor's rotation. The bearings support the rotor's rotation while also bearing the rotor's load and vibration. This embodiment, through analysis of the centrifuge's structural topology, can clearly define the spatial relationships and mutual influences between the various components.

[0030] In practice, sensors are typically installed in key locations on a centrifuge to monitor operational status data in real time. Furthermore, the spatial distance and physical connection between different sensor nodes can affect the correlation between collected data. For example, data collected by sensors installed on the same component exhibits strong correlation, while data collected by sensors installed on different components exhibits weaker correlation. Therefore, when constructing an adjacency matrix, it is necessary to obtain the spatial positional relationships between sensor nodes in order to accurately determine the degree of correlation between different sensor nodes. Therefore, this embodiment determines the spatial correlation between individual sensor nodes by combining the centrifuge's structural topology with the sensor deployment locations.

[0031] In this embodiment, the adjacency matrix is ​​a two-dimensional matrix with the same number of rows and columns as the number of sensor nodes. Each element in the two-dimensional matrix represents the association between two corresponding sensor nodes. If two sensor nodes are physically adjacent or closely connected, the corresponding element in the adjacency matrix is ​​1; otherwise, it is 0. For example, if sensor A and sensor B are mounted on the same component, the association element between sensor A and sensor B is 1; if sensor A and sensor C are mounted on different components and have no direct physical connection, the association element between sensor A and sensor C is 0. The adjacency matrix constructed in this embodiment can clearly represent the spatial association structure between individual sensor nodes.

[0032] In this embodiment, a spatiotemporal graph convolutional network is used to fuse the time series matrix and the adjacency matrix to obtain an operating state tensor with temporal and spatial correlation characteristics. The basic principle of the spatiotemporal graph convolutional network is to process the adjacency matrix through a graph convolution operation to extract the spatial features of the data. At the same time, the time series matrix is ​​processed through a convolution operation to extract the temporal features of the data. Then, the spatial features and the temporal features are fused to obtain comprehensive spatiotemporal features.

[0033] Specifically, the fusion process of the time series matrix and adjacency matrix input into the spatiotemporal graph convolutional network involves the following steps: At the input layer of the spatiotemporal graph convolutional network, the time series matrix is ​​input as node features, and the adjacency matrix is ​​input as graph structural information. At the hidden layer of the spatiotemporal graph convolutional network, the adjacency matrix is ​​convolved with the graph convolution layer to fuse the features of each sensor node with those of its neighboring nodes, thereby extracting the spatial features of the data. Simultaneously, the time series matrix is ​​convolved with the convolutional layer of the spatiotemporal graph convolutional network to extract the temporal features of the data. At the output layer of the network, the spatial and temporal features are fused to produce an operating state tensor. The three dimensions of the operating state tensor represent time steps, sensor nodes, and feature channels, respectively. The time steps correspond to the time windows in the time series matrix, the sensor nodes correspond to the sensor nodes in the adjacency matrix, and the feature channels correspond to the extracted temporal statistical features (mean, variance, and peak).

[0034] In this embodiment, the time series matrix and the adjacency matrix are fused through the spatiotemporal graph convolutional network, and the obtained operating status tensor can comprehensively reflect the temporal and spatial correlation characteristics of the centrifuge operating status.

[0035] In one embodiment of the present application, constructing an operating state tensor with temporal and spatial correlation features based on multi-dimensional operating state data further includes: Obtain the rotation speed of the centrifuge and calculate the time window length based on a preset first window formula; The preset time window is extended and controlled according to the length of the time window to obtain a first time window; The multi-dimensional operating status data is segmented according to the first time window, the time domain statistical features of the data in the first time window are extracted, and a time series matrix is ​​generated.

[0036] In this embodiment, the rotation speed of the centrifuge can obtain accurate rotation speed data through a rotation speed sensor installed on the centrifuge motor shaft; the rotation speed sensor usually adopts electromagnetic induction, photoelectric encoding and other principles to convert the rotation speed of the centrifuge into an electrical signal, and transmits it to the edge computing node through the data acquisition system.

[0037] In this embodiment, the preset first window formula is pre-set based on the operating characteristics of the centrifuge and the data analysis requirements, and the time window length can be calculated based on the speed of the centrifuge. For example, the first window formula can be expressed as:

[0038] in, Indicates the length of the time window, is the harmonic coverage factor, C is the periodic constant related to the centrifuge structure, is the minimum window protection threshold, is the rotation speed of the centrifuge.

[0039] Among them, the period constant C=60×N pole / P, where N pole is the number of motor pole pairs, and P is the reduction ratio of the transmission system. The harmonic coverage factor can be adjusted according to the speed of the centrifuge, with a high value in high-speed mode and a low value in low-speed mode. For example, the value range of the harmonic coverage factor can be [2, 5]. The principle of the first window formula is that when the centrifuge speed is high, its operating state changes rapidly, and a shorter time window is required to capture the details of the data changes; when the centrifuge speed is low, the operating state is relatively stable, and a longer time window can be used to reduce the amount of data processing.

[0040] In this embodiment, if the calculated time window length is less than the preset time window length, the preset time window is shortened; conversely, if the calculated time window length is greater than the preset time window length, the preset time window is extended. For example, if the preset time window is 5 minutes and the calculated time window length is 3 minutes, the preset time window is shortened to 3 minutes, resulting in the first time window. This embodiment utilizes this scalable control method to better adapt the time window length to the real-time operating status of the centrifuge, improving the rationality of data segmentation.

[0041] In one embodiment of the present application, a monitoring strategy matrix is ​​established based on a federated learning framework, including: In the federated learning framework, the monitoring strategy submatrix of each edge computing node of the centrifuge body is used as the local model parameter. The column vectors of the monitoring strategy submatrix correspond to different working modes, and the row vectors represent the activation probability of the sensor group. A federated learning alliance is established between multiple edge nodes through a secure multi-party computing protocol, and a homomorphic encryption algorithm is used to encrypt and transmit local gradient update values. Construct a joint loss function based on the running state tensor, and solve the optimal solution of the joint loss function through the alternating direction multiplication method; Perform tensor splicing on the encrypted policy sub-matrices of each edge node to generate a global monitoring policy matrix; The joint loss function includes the pattern recognition accuracy term, the sensor energy consumption cost term, and the regularization term of the strategy difference between nodes; Based on the residual connection structure, the local model parameters and the global monitoring strategy matrix are weightedly fused to obtain the monitoring strategy matrix.

[0042] In this embodiment, within the federated learning framework, the monitoring strategy submatrix for each edge computing node in the centrifuge is treated as a local model parameter. This monitoring strategy submatrix has a specific structure, with column vectors corresponding to different operating modes, such as startup mode, normal operation mode, stop mode, and fault mode; and row vectors representing the activation probability of sensor groups. Each edge computing node has its own independent monitoring strategy submatrix, depending on its location and the data it collects. For example, an edge computing node located in a critical part of the centrifuge will have its monitoring strategy submatrix specifically focused on setting the activation probability of the sensor groups associated with that location.

[0043] In this embodiment, the present application utilizes a secure multi-party computing protocol to build a federated learning alliance among multiple edge nodes. The secure multi-party computing protocol ensures that each edge node can participate in model training without disclosing its own data privacy. It also enables each edge node to exchange information and update models in a secure environment. The present application utilizes a homomorphic encryption algorithm in federated learning to ensure that the privacy and security of data is protected when each edge node transmits local gradient update values. Even if the data is intercepted during transmission, an attacker cannot obtain sensitive information.

[0044] In this embodiment, the joint loss function contains multiple important terms, namely, the pattern recognition accuracy term, the sensor energy consumption cost term, and the inter-node strategy difference regularization term; the pattern recognition accuracy term is used to measure the accuracy of the model in identifying the centrifuge working mode. By increasing the value of this term, the model can more accurately identify the current working mode of the centrifuge; the sensor energy consumption cost term is the energy consumption of the sensor during the monitoring process. By optimizing this term, the energy consumption of the sensor can be reduced and the energy utilization efficiency can be improved; the inter-node strategy difference regularization term is used to constrain the differences in monitoring strategies between edge nodes to avoid inconsistent strategies. This application uses the alternating direction multiplier method to solve the optimal solution of the joint loss function; this algorithm can effectively balance multiple factors such as pattern recognition accuracy, sensor energy consumption, and inter-node strategy differences.

[0045] In this embodiment, in the federated learning alliance, each edge node sends its own monitoring strategy sub-matrix to the central server or other designated node through encrypted transmission; the central server or other designated node performs tensor splicing operations on the encrypted strategy sub-matrix to obtain a global monitoring strategy matrix; wherein, the global monitoring strategy matrix can comprehensively reflect the operation monitoring requirements of the centrifuge.

[0046] In this embodiment, the present application performs weighted fusion of local model parameters and the global monitoring strategy matrix through a residual connection structure. The monitoring strategy matrix obtained by fusing the local information contained in the local model parameters with the global information contained in the global monitoring strategy matrix can be more accurate.

[0047] In one embodiment of the present application, the centrifuge remote monitoring method further includes: Iteratively optimize the parameters of the fault prediction model to obtain the target fault prediction model; The target monitoring data obtained by the sensor group monitoring the centrifuge is input into the target fault prediction model to obtain the fault diagnosis result of the centrifuge.

[0048] In this embodiment, the fault prediction model can be constructed based on machine learning or deep learning algorithms, such as neural networks, decision trees, and support vector machines. The model's input can be multi-dimensional centrifuge operating status data, and its output is fault prediction results, such as fault type and probability of occurrence. This embodiment iteratively optimizes the parameters of the fault prediction model, which requires historical data as support. This historical data includes various status information of the centrifuge during normal operation and during failures.

[0049] The specific steps for iteratively optimizing the parameters of the fault prediction model are as follows: The first step is to divide the historical data into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust the model parameters during the training process, and the test set is used to evaluate the performance of the final optimized model.

[0050] The second step is to define a suitable loss function to measure the difference between the model prediction results and the actual fault situation; common loss functions include mean square error, cross entropy loss, etc.

[0051] The third step is to select an optimization algorithm (such as a genetic algorithm) to update the parameters of the model so that the value of the loss function continues to decrease.

[0052] In the fourth step, during each iteration, the training set data is fed into the model, the loss function is calculated, and the model parameters are updated using the optimization algorithm. The updated model is then evaluated using the validation set, and the optimization algorithm parameters are adjusted or iterations are stopped based on the evaluation results. After multiple iterations, the target fault prediction model is ultimately obtained.

[0053] The target monitoring data obtained by the sensor group during centrifuge monitoring is input into the target fault prediction model. This target monitoring data is multi-dimensional data collected by the sensor group during real-time monitoring of the centrifuge's operating status, including speed, temperature, vibration frequency, torque, and other data. The sensor group's data collection frequency can be appropriately set based on the centrifuge's operating characteristics and fault prediction requirements to ensure timely capture of changes in the centrifuge's operating status.

[0054] In this embodiment, the target fault prediction model analyzes and calculates the input target monitoring data and outputs centrifuge fault diagnosis results. These results may include information such as whether a fault has occurred, the fault type (e.g., mechanical, electrical, etc.), the probability of the fault occurring, and the severity of the fault. This information is crucial for implementing timely fault repair measures and ensuring the safe and stable operation of the centrifuge. For example, if the model predicts a high probability of mechanical failure in the centrifuge, personnel can schedule maintenance and conduct inspections and repairs in advance to prevent the occurrence and escalation of the fault.

[0055] In summary, this embodiment can accurately diagnose centrifuge faults by iteratively optimizing the parameters of the fault prediction model and inputting real-time monitoring data into the optimized model, providing strong support for remote monitoring and maintenance of the centrifuge.

[0056] In one embodiment of the present application, iteratively optimizing the parameters of the fault prediction model to obtain a target fault prediction model includes: Iteratively optimize the parameters of the fault prediction model to obtain target parameters, which are parameters corresponding to the target fault prediction model; Construct the dimensions of individuals in the genetic algorithm based on the preset time window. The dimensions of individuals in the genetic algorithm include the maximum depth, the minimum number of samples, and the number of features. Based on the multi-scale feature distribution of the running state tensor, the fitness function of the genetic algorithm is determined. Determine the target crossover probability of the genetic algorithm based on the running speed of the genetic algorithm; Determine the target mutation probability of the genetic algorithm based on the accuracy of the fault prediction model on the validation set; The fault prediction model is optimized based on the fitness function, target crossover probability and target mutation probability to obtain the target parameters; The fault prediction model is optimized according to the target parameters to obtain the target fault prediction model.

[0057] In this embodiment, the dimensions of individuals in the genetic algorithm are constructed based on a preset time window; the preset time window has been dynamically determined according to the operating status of the centrifuge in the previous step, and the dimensions of individuals in the genetic algorithm include the maximum depth, the minimum number of samples, and the number of features.

[0058] In this embodiment, the operating state tensor contains temporal and spatial correlation characteristics of the centrifuge's operating state. Its multi-scale feature distribution reflects characteristic information at different scales. The fitness function is used to evaluate the performance of each individual (i.e., a set of model parameters) in the genetic algorithm. In this embodiment, the fitness function is designed to utilize the multi-scale characteristics of the operating state tensor, ensuring that parameter combinations that better match the data characteristics achieve higher fitness values. For example, the fitness function can be defined based on how well the model fits the different scale characteristics of the training data.

[0059] In this embodiment, if the genetic algorithm runs slowly, it means that the diversity of the population will be low. In this case, the crossover probability can be appropriately increased to promote gene exchange between individuals, increase the diversity of the population, and accelerate the convergence of the algorithm. Conversely, if the running speed is too fast, the algorithm will converge to the local optimal solution too early. In this case, the crossover probability can be reduced to reduce unnecessary gene exchange.

[0060] In this example, mutation probability refers to the probability of a gene mutation occurring during the genetic process. When the fault prediction model's accuracy on the validation set is low, it indicates that the model has fallen into a local optimum. In this case, the mutation probability can be appropriately increased, with the introduction of new genes to increase population diversity and help the algorithm escape the local optimum. When the model's accuracy on the validation set is high, the mutation probability can be reduced to prevent excessive mutations from disrupting the already found optimal solution.

[0061] In this embodiment, the fault prediction model is optimized based on a fitness function, a target crossover probability, and a target mutation probability. In each iteration of the genetic algorithm, the fitness value of each individual is evaluated according to the fitness function, and individuals with higher fitness are selected for crossover and mutation operations to generate new individuals. Through continuous iteration, the individuals in the population gradually approach the optimal solution, ultimately obtaining the target parameters. The target parameters are then applied to the fault prediction model, and the model's parameter settings are updated to obtain the target fault prediction model. At this point, the target fault prediction model can better utilize the centrifuge's operating status data for fault prediction tasks, improving the accuracy and reliability of fault prediction.

[0062] In one embodiment of the present application, determining the target mutation probability of the genetic algorithm based on the accuracy of the fault prediction model on the validation set includes: Determine the initial mutation probability of the genetic algorithm; In response to the accuracy of the fault prediction model on the validation set being less than or equal to a first accuracy threshold, adjusting the initial mutation probability according to the first step length until a deviation between the first mutation probability and the initial mutation probability is equal to a first difference, thereby obtaining a target mutation probability; The first mutation probability is the mutation probability obtained after each adjustment when the accuracy is less than or equal to the first accuracy threshold; In response to the accuracy of the fault prediction model on the validation set being greater than or equal to a second accuracy threshold, adjusting the initial mutation probability according to the first step length until the deviation between the second mutation probability and the initial mutation probability is greater than the first difference, thereby obtaining a target mutation probability; The second mutation probability is the mutation probability obtained after each adjustment when the accuracy is greater than or equal to the second accuracy threshold.

[0063] In this embodiment, the initial mutation probability of the genetic algorithm must be determined. This initial mutation probability is the value used by the genetic algorithm at the beginning of the optimization process. This value affects the genetic algorithm's search performance and convergence speed. For example, if the initial mutation probability is set too high, the algorithm will be too random and have difficulty converging to the optimal solution; if it is set too low, the algorithm will be stuck in a local optimum.

[0064] In this embodiment, when the accuracy of the target fault prediction model on the validation set is less than or equal to the first accuracy threshold, it means that the performance of the model is poor and it has fallen into a local optimal solution. It is necessary to add a mutation operation to introduce new genes to improve the diversity of the population.

[0065] At this point, the initial mutation probability needs to be continuously adjusted according to the first step length until the deviation between the first mutation probability and the initial mutation probability equals the first difference. The first step length is a pre-set fixed value that controls the magnitude of each mutation probability adjustment. The first difference is also a pre-set fixed value that limits the range of mutation probability adjustment. When the deviation between the first mutation probability and the initial mutation probability equals the first difference, the first mutation probability is determined as the target mutation probability.

[0066] When the fault prediction model's accuracy on the validation set is greater than or equal to the second accuracy threshold, the model has achieved good performance and no excessive mutation operations are needed to avoid destroying the already found optimal solution. The adjustment process is as follows: Continuously adjust the initial mutation probability according to the first step length until the deviation between the second mutation probability and the initial mutation probability equals the first difference. When this condition is met, the second mutation probability can be used as the target mutation probability.

[0067] This embodiment is suitable for the same sensitivity to mutation probability adjustment by setting the same adjustment step size. Regardless of the level of model accuracy, the mutation probability is adjusted with the same force to balance the exploration and utilization capabilities of the algorithm.

[0068] In summary, this embodiment adjusts the initial mutation probability of the genetic algorithm according to the accuracy of the fault prediction model on the validation set, and can dynamically determine the target mutation probability, so that the genetic algorithm can more effectively optimize the parameters of the fault prediction model under different model performance conditions and improve the fault prediction accuracy of the model.

[0069] Corresponding to the centrifuge remote monitoring method of the above embodiment, Figure 2 This is a structural block diagram of a centrifuge remote monitoring system provided by an embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 2 The centrifuge remote monitoring system 20 includes: a data processing module 21, a mode determination module 22 and a monitoring module 23.

[0070] The data processing module 21 is used to obtain multi-dimensional operating status data collected in real time by the edge computing node deployed on the centrifuge body, and construct an operating status tensor with time and space correlation characteristics based on the multi-dimensional operating status data; a mode determination module 22 for performing multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determining the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table; The monitoring module 23 is used to build a monitoring strategy matrix based on the federated learning framework, activate the corresponding sensor group according to the working mode and the monitoring strategy matrix, and use the sensor group to monitor the centrifuge.

[0071] In one embodiment of the present application, the data processing module 21 is specifically configured to: The multi-dimensional operating status data is segmented according to the preset time window, and the time domain statistical features of the data in the time window are extracted to generate a time series matrix; the time domain statistical features include: mean, variance and peak value; Construct an adjacency matrix based on the centrifuge structure topology and sensor deployment locations; The time series matrix and the adjacency matrix are fused through the spatiotemporal graph convolutional network to obtain the operating state tensor. The three dimensions of the operating state tensor represent the time step, sensor node and feature channel respectively.

[0072] In one embodiment of the present application, the data processing module 21 is specifically configured to: Obtain the rotation speed of the centrifuge and calculate the time window length based on a preset first window formula; The preset time window is extended and controlled according to the length of the time window to obtain a first time window; The multi-dimensional operating status data is segmented according to the first time window, the time domain statistical features of the data in the first time window are extracted, and a time series matrix is ​​generated.

[0073] In one embodiment of the present application, the mode determination module 22 is specifically configured to: In the federated learning framework, the monitoring strategy submatrix of each edge computing node of the centrifuge body is used as the local model parameter. The column vectors of the monitoring strategy submatrix correspond to different working modes, and the row vectors represent the activation probability of the sensor group. A federated learning alliance is established between multiple edge nodes through a secure multi-party computing protocol, and a homomorphic encryption algorithm is used to encrypt and transmit local gradient update values. Construct a joint loss function based on the running state tensor, and solve the optimal solution of the joint loss function through the alternating direction multiplication method; Perform tensor splicing on the encrypted policy sub-matrices of each edge node to generate a global monitoring policy matrix; The joint loss function includes the pattern recognition accuracy term, the sensor energy consumption cost term, and the regularization term of the strategy difference between nodes; Based on the residual connection structure, the local model parameters and the global monitoring strategy matrix are weightedly fused to obtain the monitoring strategy matrix.

[0074] In one embodiment of the present application, the data processing module 22 is further specifically configured to: Iteratively optimize the parameters of the fault prediction model to obtain the target fault prediction model; The target monitoring data obtained by the sensor group monitoring the centrifuge is input into the target fault prediction model to obtain the fault diagnosis result of the centrifuge.

[0075] In one embodiment of the present application, the data processing module 22 is specifically configured to: Iteratively optimize the parameters of the fault prediction model to obtain target parameters, which are parameters corresponding to the target fault prediction model; Construct the dimensions of individuals in the genetic algorithm based on the preset time window. The dimensions of individuals in the genetic algorithm include the maximum depth, the minimum number of samples, and the number of features. Based on the multi-scale feature distribution of the running state tensor, the fitness function of the genetic algorithm is determined. Determine the target crossover probability of the genetic algorithm based on the running speed of the genetic algorithm; Determine the target mutation probability of the genetic algorithm based on the accuracy of the fault prediction model on the validation set; The fault prediction model is optimized based on the fitness function, target crossover probability and target mutation probability to obtain the target parameters; The fault prediction model is optimized according to the target parameters to obtain the target fault prediction model.

[0076] In one embodiment of the present application, the data processing module 22 is specifically configured to: Determine the initial mutation probability of the genetic algorithm; In response to the accuracy of the fault prediction model on the validation set being less than or equal to a first accuracy threshold, adjusting the initial mutation probability according to the first step length until a deviation between the first mutation probability and the initial mutation probability is equal to a first difference, thereby obtaining a target mutation probability; The first mutation probability is the mutation probability obtained after each adjustment when the accuracy is less than or equal to the first accuracy threshold; In response to the accuracy of the fault prediction model on the validation set being greater than or equal to a second accuracy threshold, adjusting the initial mutation probability according to the first step length until the deviation between the second mutation probability and the initial mutation probability is greater than the first difference, thereby obtaining a target mutation probability; The second mutation probability is the mutation probability obtained after each adjustment when the accuracy is greater than or equal to the second accuracy threshold.

[0077] See also Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided in one embodiment of the present application. Figure 3 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned device embodiments, such as Figure 2 The functions of the data processing module 21, the pattern determination module 22 and the monitoring module 23 are shown.

[0078] It should be understood that in the embodiment of the present application, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0079] The input device 302 may include a touchpad, a fingerprint collection sensor (for collecting user fingerprint information and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.

[0080] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store device type information.

[0081] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present application can execute the implementation methods described in the first and second embodiments of the centrifuge remote monitoring method provided in the embodiments of the present application, and can also execute the implementation methods of the electronic device described in the embodiments of the present application, which will not be repeated here.

[0082] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, all or part of the process of the method in the above embodiment is implemented. The computer program can also be used to instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments are implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0083] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the computer-readable storage medium can include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or is about to be output.

[0084] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0085] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or can be an electrical, mechanical or other form of connection.

[0087] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0088] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0089] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A centrifuge remote monitoring method, characterized in that: include: Acquire multi-dimensional operating status data collected in real time by edge computing nodes deployed on the centrifuge body, and construct an operating status tensor with temporal and spatial correlation characteristics based on the multi-dimensional operating status data; Performing multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determining the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table; A monitoring strategy matrix is ​​constructed based on a federated learning framework, a corresponding sensor group is activated according to the working mode and the monitoring strategy matrix, and the centrifuge is monitored using the sensor group.

2. The centrifuge remote monitoring method according to claim 1, wherein: The constructing of an operating status tensor having temporal and spatial correlation characteristics based on the multi-dimensional operating status data includes: Segment the multi-dimensional operating status data according to a preset time window, extract the time domain statistical features of the data in the time window, and generate a time series matrix; the time domain statistical features include: mean, variance, and peak value; Construct an adjacency matrix based on the centrifuge structure topology and sensor deployment locations; The time series matrix and the adjacency matrix are fused through a spatiotemporal graph convolutional network to obtain an operating state tensor, wherein the three dimensions of the operating state tensor represent the time step, the sensor node and the feature channel respectively.

3. The centrifuge remote monitoring method according to claim 2, wherein: Also includes: Obtaining the rotation speed of the centrifuge, and calculating the time window length based on a preset first window formula; Performing expansion and contraction control on the preset time window according to the length of the time window to obtain a first time window; The multi-dimensional operating status data is segmented according to the first time window, the time domain statistical features of the data in the first time window are extracted, and a time series matrix is ​​generated.

4. The centrifuge remote monitoring method according to claim 1, wherein: The monitoring strategy matrix is ​​established based on the federated learning framework, including: In the federated learning framework, the monitoring strategy submatrix of each edge computing node of the centrifuge body is used as a local model parameter. The column vectors of the monitoring strategy submatrix correspond to different working modes, and the row vectors represent the activation probability of the sensor group. A federated learning alliance is established between multiple edge nodes through a secure multi-party computing protocol, and a homomorphic encryption algorithm is used to encrypt and transmit local gradient update values. Constructing a joint loss function according to the running state tensor, and solving the optimal solution of the joint loss function by an alternating direction multiplication method; Perform tensor splicing on the encrypted policy sub-matrices of each edge node to generate a global monitoring policy matrix; The joint loss function includes a pattern recognition accuracy term, a sensor energy consumption cost term, and a regularization term for strategy differences between nodes; Based on the residual connection structure, the local model parameters and the global monitoring strategy matrix are weightedly fused to obtain the monitoring strategy matrix.

5. The centrifuge remote monitoring method according to claim 1, wherein: Also includes: Iteratively optimize the parameters of the fault prediction model to obtain the target fault prediction model; The target monitoring data obtained by the sensor group monitoring the centrifuge is input into a target fault prediction model to obtain a fault diagnosis result of the centrifuge.

6. The centrifuge remote monitoring method according to claim 5, wherein: The iterative optimization of the parameters of the fault prediction model to obtain the target fault prediction model includes: Iteratively optimizing the parameters of the fault prediction model to obtain target parameters, where the target parameters are parameters corresponding to the target fault prediction model; Constructing the dimensions of individuals in the genetic algorithm based on a preset time window, wherein the dimensions of individuals in the genetic algorithm include a maximum depth, a minimum number of samples, and a number of features; Determining the fitness function of the genetic algorithm based on the multi-scale feature distribution of the operating state tensor, determining a target crossover probability of the genetic algorithm based on an operating speed of the genetic algorithm; Determining a target mutation probability of the genetic algorithm based on the accuracy of the fault prediction model on the validation set; Optimizing the fault prediction model based on the fitness function, the target crossover probability, and the target mutation probability to obtain the target parameters; The fault prediction model is optimized according to the target parameters to obtain a target fault prediction model.

7. The centrifuge remote monitoring method according to claim 6, wherein: The accuracy of the fault prediction model on the validation set includes: determining an initial mutation probability of the genetic algorithm; In response to the accuracy of the fault prediction model on the validation set being less than or equal to a first accuracy threshold, adjusting the initial mutation probability according to the first step length until a deviation between the first mutation probability and the initial mutation probability is equal to a first difference, thereby obtaining the target mutation probability; The first mutation probability is the mutation probability obtained after each adjustment when the accuracy is less than or equal to the first accuracy threshold; In response to the accuracy of the fault prediction model on the validation set being greater than or equal to a second accuracy threshold, adjusting the initial mutation probability according to the first step length until the deviation between the second mutation probability and the initial mutation probability is greater than the first difference, thereby obtaining the target mutation probability; The second variation probability is the variation probability obtained after each adjustment when the accuracy is greater than or equal to a second accuracy threshold.

8. A centrifuge remote monitoring system, characterized in that: include: A data processing module is used to obtain multi-dimensional operating status data collected in real time by edge computing nodes deployed on the centrifuge body, and to construct an operating status tensor with temporal and spatial correlation characteristics based on the multi-dimensional operating status data; a mode determination module, configured to perform multi-scale feature extraction on the operating state tensor based on a deep temporal clustering algorithm to obtain a first feature vector set, and determine the operating mode of the centrifuge based on the first feature vector set and a feature-operating mode mapping table; The monitoring module is used to build a monitoring strategy matrix based on the federated learning framework, activate the corresponding sensor group according to the working mode and the monitoring strategy matrix, and use the sensor group to monitor the centrifuge.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images