Intelligent monitoring method and system for fan failure

Abstract

The application relates to the technical field of fault detection, and specifically discloses an intelligent monitoring method and system for fan faults, which comprises the following steps: obtaining a to-be-identified component of a fan, evaluating the to-be-identified component, classifying and grading the to-be-identified component; determining the installation parameters of a sensor based on the classification and grading results of the to-be-identified component; taking the to-be-identified component as a reference, obtaining a sensor group indexed by the name of the to-be-identified component, constructing a data template based on the sensor group, counting the sensor data at each moment, and obtaining a space-time matrix; identifying the space-time matrix, determining the running state of the to-be-identified component, and determining the fan identification result according to the running state of each to-be-identified component. The application takes the to-be-identified component and the sensor in the fan as two groups, and mutually registers them, so that each to-be-identified component corresponds to a more suitable sensor group, the monitoring architecture is optimized as a whole, and the resource utilization rate is improved.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, specifically to an intelligent monitoring method and system for wind turbine faults. Background Technology

[0002] As a crucial power source in industrial production and energy systems, fans are widely used in power, metallurgy, chemical, mining, and ventilation industries. Their operating status directly affects production safety, energy efficiency, and equipment lifespan. Fans typically operate continuously or over long periods. Under complex loads and environmental conditions, key components such as impellers, shafts, bearings, gearboxes, and drive motors are prone to wear, loosening, or structural damage, leading to problems such as increased vibration, noise, and performance degradation.

[0003] With the development of sensing technology and intelligent analysis methods, some wind turbine monitoring systems have begun to introduce multi-source sensor data to comprehensively evaluate the operating status of wind turbines. However, existing multi-source monitoring schemes mostly adopt equal-intensity, full-coverage monitoring strategies, where the sensors at each location in the wind turbine are predetermined. This is not necessarily the optimal solution and results in some resource waste. The technical problem that this invention aims to solve is how to provide a better registration architecture between sensors and monitoring locations to optimize the resource utilization of the entire monitoring process. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent monitoring method and system for wind turbine faults to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An intelligent monitoring method for wind turbine faults, the method comprising: The components to be identified in the wind turbine are acquired, evaluated, and classified and graded. The installation parameters of the sensor are determined based on the classification and grading results of the component to be identified; the installation parameters include the sensing location and sensing type. Based on the component to be identified, a sensor group indexed by the name of the component to be identified is obtained. A data template is constructed based on the sensor group, and the sensing data at each time moment is statistically analyzed to obtain the spatiotemporal matrix. The spatiotemporal matrix is ​​identified to determine the operating status of the components to be identified, and the wind turbine identification result is determined based on the operating status of each component to be identified. For any spatiotemporal matrix, sampling is performed, and the corresponding operating state is used as the true value to evaluate the sampling process. The value score of each sampling process is determined, and the sensor parameter acquisition process is recursively updated based on the value score.

[0006] As a further aspect of the present invention: the steps of acquiring the component to be identified from the wind turbine, evaluating the component to be identified, and classifying and grading the component to be identified include: Obtain the BIM model of the wind turbine and read the component library from the BIM model; The failure frequency and impact of each component in the component library are queried sequentially in the historical record database. The selection score for each component to be identified is determined based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. The components to be identified are classified based on the selected scores, and simultaneously graded according to the average selected scores of each category of components to be identified.

[0007] As a further aspect of the present invention: the step of determining the sensor installation parameters based on the classification and grading results of the component to be identified includes: In the BIM model, the component to be identified is virtually mapped based on the hierarchical results to obtain the node corresponding to the component to be identified; the node contains a level label. In the BIM model, query the installation area for each type of sensor, divide the installation area into grids to obtain grid cells; the grid cells are rectangular with preset dimensions. For any grid cell, the distance between its center and each node is calculated, and the effective value of the network cell is determined based on the distance and the level label of the node; the effective value is used to characterize the data acquisition capability of the sensor. When the effective value reaches the preset effective value threshold, the grid cell is marked, and the center of the marked grid cell is used as the sensing position of the corresponding type of sensor.

[0008] As a further aspect of the present invention: the steps of obtaining a sensor group indexed by the name of the component to be identified, constructing a data template based on the sensor group, and statistically analyzing the sensing data at each time point to obtain the spatiotemporal matrix include: Using the component to be identified as a reference, the connectivity of each sensor is determined based on its distance from each sensor, the number of connecting components of each sensor, and the number of similar sensors already connected to the component to be identified. Sensors whose connectivity reaches a preset connectivity condition are marked, and a sensor group is created; wherein, the connectivity condition is a dynamic condition; The order of each sensor in the sensor group is determined according to the preset sorting rules, and a data template is constructed. Based on the data template, the sensor data at each time point are statistically analyzed to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time.

[0009] As a further aspect of the present invention: the step of identifying the spatiotemporal matrix, determining the operating state of the component to be identified, and determining the wind turbine identification result based on the operating state of each component to be identified includes: The spatiotemporal matrix is ​​traversed and matched based on preset anomaly features, and the matching degree is output. When the matching degree reaches the preset threshold, query and output the running status corresponding to the abnormal features; The operating status of each component to be identified is statistically analyzed to determine the fan identification result.

[0010] As a further aspect of the present invention: the steps of sampling any spatiotemporal matrix, evaluating the sampling process using the corresponding operating state as the true value, determining the value score of each sampling process, and recursively updating the sensor parameter acquisition process based on the value score include: Select the spatiotemporal matrix in sequence and query its running status as the standard running status; The spatiotemporal matrix is ​​randomly sampled to obtain a submatrix, and the submatrix is ​​upsampled to obtain a virtual matrix; the random sampling includes the number of elements and the position of elements, which is represented by a zero-one matrix as the random feature of the random sampling process; The virtual matrix is ​​identified to determine the operating status of the component to be identified, which is then used as the virtual operating status. A value score is generated by comparing the standard operating state and the virtual operating state; the value score is labeled with random features. Select random features based on value scores, query the sensors corresponding to the random features, configure them as the first frequency, and configure other sensors as the second frequency.

[0011] The present invention also provides an intelligent monitoring system for wind turbine faults, the system comprising: The component classification and grading module is used to acquire the components to be identified from the wind turbine, evaluate the components to be identified, and classify and grade the components to be identified. A sensor calibration module is used to determine the sensor's installation parameters based on the classification and grading results of the component to be identified; the installation parameters include the sensing location and sensing type. The spatiotemporal matrix generation module is used to obtain a sensor group indexed by the name of the component to be identified, based on the component to be identified, construct a data template based on the sensor group, and statistically analyze the sensing data at each time point to obtain the spatiotemporal matrix. The identification result output module is used to identify the spatiotemporal matrix, determine the operating status of the components to be identified, and determine the wind turbine identification result based on the operating status of each component to be identified. The sampling adjustment module is used to sample any spatiotemporal matrix, evaluate the sampling process using the corresponding operating state as the true value, determine the value score of each sampling process, and recursively update the sensor parameter acquisition process based on the value score.

[0012] As a further aspect of the present invention: the component classification and grading module includes: The component library reading unit is used to obtain the BIM model of the wind turbine and read the component library from the BIM model. The information query unit is used to sequentially query the failure frequency and failure impact of each component in the component library from the historical record database; A selection scoring unit is used to determine the selection score of each component to be identified based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. The classification and grading execution unit is used to classify the parts to be identified based on the selected score, and simultaneously grade the parts to be identified based on the average selected score of each category.

[0013] As a further aspect of the present invention: the sensor calibration module includes: A virtual mapping unit is used to virtually map the component to be identified in the BIM model based on the hierarchical results, so as to obtain a node corresponding to the component to be identified; the node contains a level label. Mesh segmentation units are used to query the installation area of ​​each type of sensor in the BIM model, and to segment the installation area into mesh units; the mesh units are rectangular and their size is a preset value. An effective value determination unit is used to calculate the distance between the center of any grid cell and each node, and determine the effective value of the network cell based on the distance and the level label of the node; the effective value is used to characterize the data acquisition capability of the sensor. A position generation unit is used to mark the grid cell when the effective value reaches a preset effective value threshold, and to use the center of the marked grid cell as the sensing position of the corresponding type of sensor.

[0014] As a further aspect of the present invention: the spatiotemporal matrix generation module includes: The connectivity determination unit is used to determine the connectivity of each sensor based on the component to be identified, the distance between the component and each sensor, the number of connected components of each sensor, and the number of similar sensors already connected to the component to be identified. A sensor marking unit is used to mark sensors whose connectivity reaches a preset connectivity condition, thereby creating a sensor group; wherein the connectivity condition is a dynamic condition. The module generation unit is used to determine the order of each sensor in the sensor group according to the preset sorting rules and to build a data template. The data statistics unit is used to statistically analyze the sensor data at each time point based on the data template to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time.

[0015] Compared with the prior art, the beneficial effects of the present invention are: the present invention treats the components to be identified and the sensors in the wind turbine as two groups and performs mutual registration, so that each component to be identified corresponds to a more suitable sensor group, thereby optimizing the monitoring architecture as a whole and improving resource utilization. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0017] Figure 1 This is a flowchart of an intelligent monitoring method for wind turbine faults.

[0018] Figure 2 The first sub-flowchart of the intelligent monitoring method for wind turbine faults.

[0019] Figure 3 This is the second sub-flowchart of the intelligent monitoring method for wind turbine faults.

[0020] Figure 4 This is the third sub-flowchart of the intelligent monitoring method for wind turbine faults.

[0021] Figure 5 This is the fourth sub-flowchart of the intelligent monitoring method for wind turbine faults.

[0022] Figure 6 This is the fifth sub-flowchart of the intelligent monitoring method for wind turbine faults.

[0023] Figure 7 This is a block diagram showing the composition of an intelligent monitoring system for wind turbine faults. Detailed Implementation

[0024] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0025] Example 1 Figure 1 The flowchart illustrates an intelligent monitoring method for wind turbine faults. In this embodiment of the invention, an intelligent monitoring method for wind turbine faults includes: Step S100: Obtain the component to be identified from the wind turbine, evaluate the component to be identified, and classify and grade the component to be identified; A wind turbine can be understood as an integrated device. The intelligent monitoring requirement for wind turbines is actually aimed at the various components in the wind turbine, which are called the components to be identified. The components to be identified are acquired, evaluated, and the importance of each component is determined. Then, the components to be identified are classified and graded. Classification and grading are two actions. First, they are classified, and then the overall importance of each type of component is determined, which is represented by a level.

[0026] Step S200: Determine the sensor installation parameters based on the classification and grading results of the component to be identified; the installation parameters include the sensing location and sensing type; After classifying and grading the components to be identified, the installation parameters of the sensors are determined based on the classification and grading results. The installation parameters include the sensing location and sensing type. In simple terms, it is to determine which sensors to install in which locations.

[0027] Step S300: Based on the component to be identified, obtain the sensor group indexed by the name of the component to be identified, construct a data template based on the sensor group, and statistically analyze the sensing data at each time point to obtain the spatiotemporal matrix; After determining the installation parameters of the sensors, the sensors need to detect each component to be identified. This involves which sensors are used to acquire relevant data from which components to be identified. At this point, based on the component to be identified, the sensors that can monitor it are obtained, resulting in a sensor group. A data template is constructed based on the sensor group, and the sensor data at each time moment is statistically analyzed. Each time moment corresponds to an array, and the data at different times correspond to a matrix, called the spatiotemporal matrix.

[0028] Step S400: Identify the spatiotemporal matrix, determine the operating status of the components to be identified, and determine the wind turbine identification result based on the operating status of each component to be identified; The spatiotemporal matrix serves as monitoring data for the components to be identified, reflecting their operational status. By identifying the spatiotemporal matrix, the operating status of the components to be identified is determined. The identification results for each component are statistically analyzed to obtain the overall identification result of the wind turbine, which is called the wind turbine identification result.

[0029] Step S500: Sample any spatiotemporal matrix, evaluate the sampling process using the corresponding operating state as the true value, determine the value score of each sampling process, and recursively update the sensor parameter acquisition process based on the value score. When the process reaches step S400, the intelligent recognition result has already been obtained. Step S400 is the output step. Based on this, the spatiotemporal matrix of any component to be identified is sampled. The already identified operating state is used as the accurate value, that is, the real value mentioned above, to evaluate the sampling process and determine the value score of each sampling process. The value score is used to characterize the quality of the sampling process. After obtaining the value score, the parameter acquisition process of the sensor is recursively updated according to the value score.

[0030] Figure 2 The first sub-flowchart of the intelligent monitoring method for wind turbine faults includes the following steps: acquiring the components to be identified from the wind turbine, evaluating the components to be identified, and classifying and grading the components to be identified: Step S101: Obtain the BIM model of the wind turbine and read the component library from the BIM model; Step S102: In the historical record database, query the failure frequency and failure impact of each component in the component database in turn; Step S103: Determine the selection score of each component to be identified based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. Step S104: Classify the parts to be identified based on the selection score, and simultaneously grade the parts to be identified according to the average selection score of each category.

[0031] The above describes the classification and grading process of the components to be identified. The BIM model of the wind turbine is obtained, and the component library is read from the BIM model. For the technical solution of this invention, this is known data and can be read directly. Then, the failure frequency and failure impact degree of each component in the historical record library are queried sequentially. The failure impact degree is described by the ratio of the wind turbine's actual power to its rated power. The failure impact degree is inversely proportional to the ratio. If the wind turbine is unusable, the actual power is zero, and the failure impact degree is taken as the maximum value. The selection score of each component to be identified is determined based on the failure frequency and failure impact degree. The selection score is directly proportional to both the failure frequency and failure impact degree. Based on the selection score, the components to be identified are classified, and simultaneously, the components to be identified are graded based on the average selection score of each category.

[0032] Figure 3 The second sub-flowchart of the intelligent monitoring method for wind turbine faults includes the step of determining the sensor installation parameters based on the classification and grading results of the component to be identified, which comprises: Step S201: In the BIM model, the component to be identified is virtually mapped based on the hierarchical results to obtain the node corresponding to the component to be identified; the node contains a level label; Step S202: Query the installation area of ​​each type of sensor in the BIM model, divide the installation area into grids to obtain grid cells; the grid cells are rectangular and their size is a preset value; Step S203: For any grid cell, calculate the distance between its center and each node, and determine the effective value of the network cell based on the distance and the level label of the node; the effective value is used to characterize the data acquisition capability of the sensor; Step S204: When the effective value reaches the preset effective value threshold, mark the grid cell and use the center of the marked grid cell as the sensing position of the corresponding type of sensor.

[0033] In one example of the technical solution of this invention, the installation process of the sensor is described. Specifically, it involves determining which type of sensor to install at which location. In the BIM model, a virtual mapping is performed on the component to be identified based on the classification results, resulting in nodes corresponding to the component. These nodes contain level labels, using the classification results of the component to be identified. Then, the installation area for each type of sensor is queried in the BIM model. The installation area is then divided into grids to obtain grid cells. These grid cells are rectangular with preset dimensions. The grid function is a basic function of the BIM model and is not complex; it can be applied directly. For any grid cell, the distance between its center and each node is calculated. Based on the distance and the level labels of the nodes, the effective value of the grid cell is determined. When the effective value reaches a preset effective value threshold, the grid cell is marked, and the center of the marked grid cell is used as the sensing position of the corresponding type of sensor.

[0034] It should be noted that the above content analyzes different types of sensors sequentially. Different types of sensors have different installation areas. By repeating the above process multiple times to analyze each type of sensor, the installation location of different types of sensors can be obtained.

[0035] One feasible calculation scheme for the valid values ​​mentioned above is as follows: In the formula, Indicates the center The effective value at that point, The total number of nodes. For the first At the level of each node, Indicates the first Each node and center distance, This is the preset index.

[0036] The principle behind the calculation is not complicated; it involves calculating the influence of each node on the center and then summing the results. The influence is directly proportional to the level, has an adjustable relationship with distance, and is generally inversely proportional. Not less than one.

[0037] Figure 4 The third sub-flowchart of the intelligent monitoring method for wind turbine faults includes the following steps: obtaining a sensor group indexed by the name of the component to be identified, constructing a data template based on the sensor group, and statistically analyzing the sensor data at each time point to obtain the spatiotemporal matrix. Step S301: Using the component to be identified as a reference, determine the connectivity of each sensor based on its distance from each sensor, the number of connecting components of each sensor, and the number of similar sensors already connected to the component to be identified. Step S302: Mark the sensors whose connectivity reaches the preset connectivity condition and create a sensor group; wherein, the connectivity condition is a dynamic condition; Step S303: Determine the order of each sensor in the sensor group according to the preset sorting rules, and construct a data template; Step S304: Based on the data template, statistically analyze the sensor data at each time point to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time.

[0038] The above describes the process of generating the spatiotemporal matrix. Taking the component to be identified as a reference, the connectivity of each sensor is determined based on its distance from each sensor, the number of connected components of each sensor, and the number of similar sensors already connected to the component to be identified. The connectivity represents the cost-effectiveness of the sensor in identifying the component to be identified. Sensors whose connectivity reaches the preset connectivity condition are marked to obtain the sensor group of the component to be identified. The connectivity condition is a dynamic condition.

[0039] Based on the sensor group, the order of each sensor in the sensor group is determined according to a preset sorting rule (such as the first letter order of the model or the position order), a data template is constructed, and the sensor data at each time moment is statistically analyzed based on the data template to obtain a spatiotemporal matrix. One dimension of the spatiotemporal matrix represents time, and the other dimension represents the sensor type.

[0040] The connectivity mentioned above actually represents the "cost-effectiveness" of a sensor in identifying a certain component, which can also be understood as a value. It is inversely proportional to the distance to the sensor, meaning that the closer the distance, the greater the significance of identification. It is also inversely proportional to the number of components connected to the sensor, meaning that the less workload a sensor has, the greater the significance of identification. Furthermore, it is inversely proportional to the number of similar sensors connected to the component to be identified, meaning that the less duplicate identification, the greater the significance of identification. As for the degree of influence of these three factors, they can be adjusted by the preset weights set by the operator. There are no specific limitations; the technical solution of this invention only limits the existence of the weight parameter.

[0041] Figure 5 The fourth sub-flowchart of the intelligent monitoring method for wind turbine faults includes the following steps: identifying the spatiotemporal matrix, determining the operating status of the components to be identified, and determining the wind turbine identification result based on the operating status of each component to be identified. Step S401: Perform traversal matching on the spatiotemporal matrix based on preset anomaly features and output the matching degree; Step S402: When the matching degree reaches the preset threshold, query and output the running status corresponding to the abnormal feature; Step S403: Statistically analyze the operating status of each component to be identified and determine the fan identification result.

[0042] For the spatiotemporal matrix recognition process, the spatiotemporal matrix is ​​traversed and matched based on preset abnormal features, and the matching degree is output. This process is a convolutional recognition process. There are many open-source recognition methods in the existing technology, which will not be elaborated here. During the matching process, the matching degree is output in real time. When the matching degree reaches a preset threshold, the running status corresponding to the abnormal feature is queried and output. The abnormal feature is equivalent to the convolution kernel in the convolutional recognition model. The running status is the inherent attribute corresponding to each abnormal feature that is preset and can be read directly. The running status of each component to be identified is statistically analyzed to determine the wind turbine recognition result.

[0043] Figure 6 The fifth sub-flowchart of the intelligent monitoring method for wind turbine faults includes the following steps: sampling any spatiotemporal matrix, evaluating the sampling process using the corresponding operating state as the true value, determining the value score of each sampling process, and recursively updating the sensor parameter acquisition process based on the value score. Step S501: Select the spatiotemporal matrix in sequence and query its running status as the standard running status; Step S502: Randomly sample the spatiotemporal matrix to obtain a submatrix, and upsample the submatrix to obtain a virtual matrix; the random sampling includes the number of elements and the position of elements, which is represented by a zero-one matrix as the random feature of the random sampling process; Step S503: Identify the virtual matrix and determine the operating status of the component to be identified, which is then used as the virtual operating status; Step S504: Compare the standard operating state and the virtual operating state to generate a value score; the value score is labeled with random features; Step S505: Select random features based on value scores, query the sensors corresponding to the random features, configure them as the first frequency, and configure the other sensors as the second frequency.

[0044] For the sensor parameter acquisition process, the spatiotemporal matrix is ​​selected sequentially, and the already determined operating states of the spatiotemporal matrix are queried as the standard operating states. Random sampling is then performed on the spatiotemporal matrix to obtain sub-matrices. This process involves randomly selecting a portion of data from the spatiotemporal matrix; the dimension of the resulting matrix remains unchanged, but the data is finite, hence the name sub-matrix. The sub-matrix is ​​then upsampled to obtain a virtual matrix. The virtual matrix has the same dimension as the original spatiotemporal matrix, the difference being that some of the data is simulated (upsampling). The identification process, as in step S400, is applied to the virtual matrix to determine the operating state of the component to be identified, which is then used as the virtual operating state. By comparing the standard operating state and the virtual operating state, a value score is generated. The value score can be expressed as a similarity score, which indicates the quality of the corresponding sampling process. The higher the similarity score, the better the sampling process. The above process is repeated, and a value score is obtained for each random sampling process. Random features are selected based on the value scores, and the corresponding sensors are queried and configured as the first frequency. Other sensors are configured as the second frequency. If the first frequency is greater than the second frequency, it means that the randomly sampled sensor is sufficient to display the working state of the component to be identified, and its frequency should be higher. Other sensors can collect data at a lower frequency, and in extreme cases, no data is collected.

[0045] Example 2 Figure 7 This is a block diagram of the composition structure of an intelligent monitoring system for wind turbine faults. In this embodiment of the invention, an intelligent monitoring system for wind turbine faults, the system 10 includes: The component classification and grading module 11 is used to acquire the components to be identified in the wind turbine, evaluate the components to be identified, and classify and grade the components to be identified. The sensor calibration module 12 is used to determine the sensor installation parameters based on the classification and grading results of the component to be identified; the installation parameters include the sensing position and the sensing type. The spatiotemporal matrix generation module 13 is used to obtain a sensor group indexed by the name of the component to be identified, based on the component to be identified, construct a data template based on the sensor group, and statistically analyze the sensing data at each time to obtain the spatiotemporal matrix. The identification result output module 14 is used to identify the spatiotemporal matrix, determine the operating status of the component to be identified, and determine the wind turbine identification result based on the operating status of each component to be identified. The sampling adjustment module 15 is used to sample any spatiotemporal matrix, evaluate the sampling process using the corresponding operating state as the true value, determine the value score of each sampling process, and recursively update the sensor parameter acquisition process based on the value score.

[0046] Furthermore, the component classification and grading module 11 includes: The component library reading unit is used to obtain the BIM model of the wind turbine and read the component library from the BIM model. The information query unit is used to sequentially query the failure frequency and failure impact of each component in the component library from the historical record database; A selection scoring unit is used to determine the selection score of each component to be identified based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. The classification and grading execution unit is used to classify the parts to be identified based on the selected score, and simultaneously grade the parts to be identified based on the average selected score of each category.

[0047] Specifically, the sensor calibration module 12 includes: A virtual mapping unit is used to virtually map the component to be identified in the BIM model based on the hierarchical results, so as to obtain a node corresponding to the component to be identified; the node contains a level label. Mesh segmentation units are used to query the installation area of ​​each type of sensor in the BIM model, and to segment the installation area into mesh units; the mesh units are rectangular and their size is a preset value. An effective value determination unit is used to calculate the distance between the center of any grid cell and each node, and determine the effective value of the network cell based on the distance and the level label of the node; the effective value is used to characterize the data acquisition capability of the sensor. A position generation unit is used to mark the grid cell when the effective value reaches a preset effective value threshold, and to use the center of the marked grid cell as the sensing position of the corresponding type of sensor.

[0048] Furthermore, the spatiotemporal matrix generation module 13 includes: The connectivity determination unit is used to determine the connectivity of each sensor based on the component to be identified, the distance between the component and each sensor, the number of connected components of each sensor, and the number of similar sensors already connected to the component to be identified. A sensor marking unit is used to mark sensors whose connectivity reaches a preset connectivity condition, thereby creating a sensor group; wherein the connectivity condition is a dynamic condition. The module generation unit is used to determine the order of each sensor in the sensor group according to the preset sorting rules and to build a data template. The data statistics unit is used to statistically analyze the sensor data at each time point based on the data template to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time.

[0049] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent monitoring method for wind turbine faults, characterized in that, The method includes: The process includes acquiring the components of the wind turbine to be identified, evaluating these components, and classifying and grading them; specifically, this includes: Obtain the BIM model of the wind turbine and read the component library from the BIM model; The failure frequency and impact of each component in the component library are queried sequentially in the historical record database. The selection score for each component to be identified is determined based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. The parts to be identified are classified based on the selected scores, and the parts to be identified are graded according to the average selected scores of each category. The sensor's installation parameters are determined based on the classification and grading results of the component to be identified; these installation parameters include sensing location and sensing type; specifically, they include: In the BIM model, the component to be identified is virtually mapped based on the hierarchical results to obtain the node corresponding to the component to be identified; the node contains a level label. In the BIM model, query the installation area for each type of sensor, divide the installation area into grids to obtain grid cells; the grid cells are rectangular with preset dimensions. For any grid cell, calculate the distance from its center to each node, and determine the effective value of the grid cell based on the distance and the node's level label; the effective value is used to characterize the data acquisition capability of the grid cell; the calculation scheme for the effective value is as follows: In the formula, Indicates the center The effective value at that point, The total number of nodes. For the first At the level of each node, Indicates the first Each node and center distance, The preset index; When the effective value reaches the preset effective value threshold, the grid cell is marked, and the center of the marked grid cell is used as the sensing position of the corresponding type of sensor. Using the component to be identified as a reference, a sensor group indexed by the name of the component is obtained. A data template is constructed based on the sensor group, and the sensing data at each time point is statistically analyzed to obtain a spatiotemporal matrix; specifically including: Using the component to be identified as a reference, the connectivity of each sensor is determined based on its distance from each sensor, the number of connecting components of each sensor, and the number of similar sensors already connected to the component to be identified. Sensors whose connectivity reaches a preset connectivity condition are marked, and a sensor group is created; wherein, the connectivity condition is a dynamic condition; The order of each sensor in the sensor group is determined according to the preset sorting rules, and a data template is constructed. Based on the data template, the sensor data at each time moment is statistically analyzed to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time; and the other dimension represents the sensor type. The spatiotemporal matrix is ​​identified to determine the operating status of the components to be identified, and the wind turbine identification result is determined based on the operating status of each component to be identified. For any spatiotemporal matrix, sampling is performed, and the corresponding operating state is used as the true value to evaluate the sampling process, determining the value score of each sampling process. The sensor parameter acquisition process is then recursively updated based on the value score; specifically, this includes: Select the spatiotemporal matrix in sequence, query the already determined running state of the spatiotemporal matrix, and use it as the standard running state; The spatiotemporal matrix is ​​randomly sampled to obtain a submatrix, and the submatrix is ​​upsampled to obtain a virtual matrix; the random sampling includes the number of elements and the position of elements, which is represented by a zero-one matrix as the random feature of the random sampling process; The virtual matrix is ​​identified to determine the operating status of the component to be identified, which is then used as the virtual operating status. A value score is generated by comparing the standard operating state and the virtual operating state; the value score is labeled with random features; the value score adopts similarity. Select random features based on value scores, query the sensors corresponding to the random features, configure them as the first frequency, and configure other sensors as the second frequency.

2. The intelligent monitoring method for wind turbine faults according to claim 1, characterized in that, The steps of identifying the spatiotemporal matrix, determining the operating status of the components to be identified, and determining the wind turbine identification result based on the operating status of each component to be identified include: The spatiotemporal matrix is ​​traversed and matched based on preset anomaly features, and the matching degree is output. When the matching degree reaches the preset threshold, query and output the running status corresponding to the abnormal features; The operating status of each component to be identified is statistically analyzed to determine the fan identification result.

3. An intelligent monitoring system for wind turbine faults, characterized in that, The system includes: The component classification and grading module is used to acquire the components to be identified from the wind turbine, evaluate the components, and classify and grade them; specifically, it includes: The component library reading unit is used to obtain the BIM model of the wind turbine and read the component library from the BIM model. The information query unit is used to sequentially query the failure frequency and failure impact of each component in the component library from the historical record database; A selection scoring unit is used to determine the selection score of each component to be identified based on the fault frequency and the degree of fault impact; the selection score is directly proportional to both the fault frequency and the degree of fault impact. The classification and grading execution unit is used to classify the parts to be identified based on the selection score, and simultaneously grade the parts to be identified according to the average selection score of each category. A sensor calibration module is used to determine the sensor's installation parameters based on the classification and grading results of the component to be identified; the installation parameters include sensing location and sensing type; specifically including: A virtual mapping unit is used to virtually map the component to be identified in the BIM model based on the hierarchical results, so as to obtain a node corresponding to the component to be identified; the node contains a level label. Mesh segmentation units are used to query the installation area of ​​each type of sensor in the BIM model, and to segment the installation area into mesh units; the mesh units are rectangular and their size is a preset value. An effective value determination unit is used to calculate the distance between the center of any grid cell and each node, and determine the effective value of the grid cell based on the distance and the level label of the nodes; the effective value is used to characterize the data acquisition capability of the grid cell; the calculation scheme for the effective value is as follows: In the formula, Indicates the center The effective value at that point, The total number of nodes. For the first At the level of each node, Indicates the first Each node and center distance, The preset index; A position generation unit is used to mark the grid cell when the effective value reaches a preset effective value threshold, and to use the center of the marked grid cell as the sensing position of the corresponding type of sensor. The spatiotemporal matrix generation module is used to obtain a sensor group indexed by the name of the component to be identified, based on the component to be identified; construct a data template based on the sensor group; and statistically analyze the sensing data at each time point to obtain the spatiotemporal matrix; specifically, it includes: The connectivity determination unit is used to determine the connectivity of each sensor based on the component to be identified, the distance between the component and each sensor, the number of connected components of each sensor, and the number of similar sensors already connected to the component to be identified. A sensor marking unit is used to mark sensors whose connectivity reaches a preset connectivity condition, thereby creating a sensor group; wherein the connectivity condition is a dynamic condition. The module generation unit is used to determine the order of each sensor in the sensor group according to the preset sorting rules and to build a data template. The data statistics unit is used to statistically analyze the sensor data at each time point based on the data template to obtain a spatiotemporal matrix; wherein, one dimension of the spatiotemporal matrix represents time; and the other dimension represents the sensor type. The identification result output module is used to identify the spatiotemporal matrix, determine the operating status of the components to be identified, and determine the wind turbine identification result based on the operating status of each component to be identified. The sampling adjustment module is used to sample any spatiotemporal matrix, evaluate the sampling process using the corresponding operating state as the true value, determine the value score of each sampling process, and recursively update the sensor parameter acquisition process based on the value score; specifically including: Select the spatiotemporal matrix in sequence, query the already determined running state of the spatiotemporal matrix, and use it as the standard running state; The spatiotemporal matrix is ​​randomly sampled to obtain a submatrix, and the submatrix is ​​upsampled to obtain a virtual matrix; the random sampling includes the number of elements and the position of elements, which is represented by a zero-one matrix as the random feature of the random sampling process; The virtual matrix is ​​identified to determine the operating status of the component to be identified, which is then used as the virtual operating status. A value score is generated by comparing the standard operating state and the virtual operating state; the value score is labeled with random features; the value score adopts similarity. Select random features based on value scores, query the sensors corresponding to the random features, configure them as the first frequency, and configure other sensors as the second frequency.

Images