A method and system for detecting a flow restrictor ring

Abstract

The application provides a flow limiting ring detection method and system, and relates to the technical field of flow limiting ring detection. The method comprises the following steps: acquiring ultrasonic images of multiple flow limiting rings and flow information of inflow under a working environment of the flow limiting rings; determining a preliminary abnormal flow limiting ring by using an abnormal flow limiting ring model based on the ultrasonic images of the multiple flow limiting rings; determining K test flow rates and multiple simulation flow rates based on the flow information of the inflow under the working environment of the flow limiting rings; acquiring infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flow rates; and determining whether the preliminary abnormal flow limiting ring is qualified based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flow rates and the infrared thermal image videos of the preliminary abnormal flow limiting ring under the multiple simulation flow rates. The method can efficiently, comprehensively and accurately detect whether the flow limiting ring is qualified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flow-restricting ring detection, and particularly relates to a flow-restricting ring detection method and system. BACKGROUND

[0002] As a key component for restricting liquid flow in aviation pipelines, the performance abnormality of the flow-restricting ring directly affects the safe operation of aviation equipment. During long-term service, the flow-restricting ring is prone to damage such as cracks, corrosion pits and deformation due to the influence of medium scouring, corrosion, fatigue load and potential material defects. These damages not only change the preset flow regulating characteristics of the flow-restricting ring, leading to system control inaccuracy and increased energy consumption, but also continuously expand at stress concentration sites, causing sudden failure or even rupture of the component, resulting in serious safety accidents and economic losses. The traditional detection method has significant limitations and relies on local detection of a single type of instrument, which is difficult to cover the overall structure and working scenarios of the flow-restricting ring and cannot find internal hidden damage, subtle size deviations and flow-restricting capacity fluctuations under different flow conditions. At the same time, the traditional method can only test a few fixed flow points and cannot cover various flow conditions under the working environment of the flow-restricting ring, making it difficult to find potential performance problems under different flow conditions, resulting in incomplete identification of problems such as liquid flow resistance abnormalities and reduced flow capacity of the flow-restricting channel. In addition, the traditional detection method is low in efficiency and strong in subjectivity, making it difficult to achieve rapid screening and accurate evaluation of batch flow-restricting rings and unable to meet the stringent requirements of the aviation field for detection reliability and comprehensiveness.

[0003] Therefore, how to efficiently, comprehensively and accurately detect whether the flow-restricting ring is qualified is a problem to be solved at present. SUMMARY

[0004] The technical problem solved by the present application is how to efficiently, comprehensively and accurately detect whether the flow-restricting ring is qualified.

[0005] According to a first aspect, the present application provides a flow-restricting ring detection method, comprising: acquiring ultrasonic images of a plurality of flow-restricting rings and flow information of inflow under a working environment of the flow-restricting ring; determining a preliminary abnormal flow-restricting ring using an abnormal flow-restricting ring model based on the ultrasonic images of the plurality of flow-restricting rings; determining K test flow rates and a plurality of simulation flow rates based on the flow information of the inflow under the working environment of the flow-restricting ring; acquiring infrared thermal image videos of the preliminary abnormal flow-restricting ring under the K test flow rates; generating infrared thermal image videos of the preliminary abnormal flow-restricting ring under the plurality of simulation flow rates based on the infrared thermal image videos of the preliminary abnormal flow-restricting ring under the K test flow rates; and determining whether the preliminary abnormal flow-restricting ring is qualified based on the infrared thermal image videos of the preliminary abnormal flow-restricting ring under the K test flow rates and the infrared thermal image videos of the preliminary abnormal flow-restricting ring under the plurality of simulation flow rates.

[0006] In a possible implementation, the determining whether the preliminary abnormal flow limiting ring is qualified based on the infrared thermal image video of the preliminary abnormal flow limiting ring under the K test flows and the infrared thermal image video of the preliminary abnormal flow limiting ring under the multiple simulation flows includes: constructing an infrared graph, the infrared graph including multiple nodes and edges between the multiple nodes, the multiple nodes including K test flow nodes and multiple simulation flow nodes, each simulation flow node establishing an edge with a K test flow node, the edge between the nodes being a difference between a simulation flow and a test flow, a node feature of a test flow node being the infrared thermal image video of the preliminary abnormal flow limiting ring under the test flow, and a node feature of a simulation flow node being the infrared thermal image video of the preliminary abnormal flow limiting ring under the simulation flow; and determining whether the preliminary abnormal flow limiting ring is qualified based on processing the infrared graph by using a graph neural network.

[0007] In a possible implementation, the determining the K test flows and the multiple simulation flows based on the inflow flow information under the working environment of the flow limiting ring includes: determining a K value based on the inflow flow information under the working environment of the flow limiting ring; obtaining K clusters by using a K-means clustering algorithm based on the inflow flow information under the working environment of the flow limiting ring and the K value; determining the K test flows based on the K clusters, each test flow being selected from each cluster; and determining the multiple simulation flows based on the K clusters and the K test flows.

[0008] In a possible implementation, the generating the infrared thermal image video of the preliminary abnormal flow limiting ring under the multiple simulation flows based on the infrared thermal image video of the preliminary abnormal flow limiting ring under the K test flows includes: generating the infrared thermal image video of the preliminary abnormal flow limiting ring under the multiple simulation flows by using a generative adversarial network based on the infrared thermal image video of the preliminary abnormal flow limiting ring under the K test flows.

[0009] According to a second aspect, the application provides a flow limiting ring detection system, including: an information acquisition module, configured to acquire ultrasonic images of multiple flow limiting rings and inflow flow information under a working environment of the flow limiting ring; a preliminary abnormality determination module, configured to determine a preliminary abnormal flow limiting ring by using an abnormal flow limiting ring model based on the ultrasonic images of the multiple flow limiting rings; a flow determination module, configured to determine K test flows and multiple simulation flows based on the inflow flow information under the working environment of the flow limiting ring; a test video acquisition module, configured to acquire infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows; a simulation video generation module, configured to generate infrared thermal image videos of the preliminary abnormal flow limiting ring under the multiple simulation flows based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows; and a qualification determination module, configured to determine whether the preliminary abnormal flow limiting ring is qualified based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows and the infrared thermal image videos of the preliminary abnormal flow limiting ring under the multiple simulation flows.

[0010] In a possible implementation, the eligibility determining module is further configured to: construct an infrared spectrum, the infrared spectrum comprising a plurality of nodes and edges between the plurality of nodes, the plurality of nodes comprising the K test flow nodes and a plurality of simulation flow nodes, each simulation flow node establishing an edge with each of the K test flow nodes, the edges between the nodes being differences between the simulation flow and the test flow, the node feature of the test flow node being the infrared thermal image video of the preliminary abnormal flow limiting ring under the test flow, and the node feature of the simulation flow node being the infrared thermal image video of the preliminary abnormal flow limiting ring under the simulation flow; and determine whether the preliminary abnormal flow limiting ring is eligible based on processing the infrared spectrum by using a graph neural network.

[0011] In a possible implementation, the flow determining module is further configured to: determine the K value based on the inflow flow information under the flow limiting ring working environment; obtain K clusters by using a K-means clustering algorithm based on the inflow flow information under the flow limiting ring working environment and the K value; determine the K test flows based on the K clusters, each test flow being selected from each cluster; and determine the plurality of simulation flows based on the K clusters and the K test flows.

[0012] In a possible implementation, the simulation video generating module is further configured to: generate the infrared thermal image video of the preliminary abnormal flow limiting ring under the plurality of simulation flows by using a generative adversarial network based on the infrared thermal image video of the preliminary abnormal flow limiting ring under the K test flows.

[0013] According to a third aspect, embodiments of the present application provide an electronic device, comprising: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: obtaining ultrasonic images of a plurality of flow limiting rings and inflow flow information under a flow limiting ring working environment; determining a preliminary abnormal flow limiting ring by using an abnormal flow limiting ring model based on the ultrasonic images of the plurality of flow limiting rings; determining K test flows and a plurality of simulation flows based on the inflow flow information under the flow limiting ring working environment; obtaining infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows; generating infrared thermal image videos of the preliminary abnormal flow limiting ring under the plurality of simulation flows based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows; and determining whether the preliminary abnormal flow limiting ring is eligible based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows and the infrared thermal image videos of the preliminary abnormal flow limiting ring under the plurality of simulation flows.

[0014] According to a fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned method for detecting a current-limiting loop. The method includes: acquiring ultrasonic images of multiple current-limiting loops and inflow information under the operating environment of the current-limiting loops; determining a preliminary abnormal current-limiting loop using an abnormal current-limiting loop model based on the ultrasonic images of the multiple current-limiting loops; determining K test flows and multiple simulated flows based on the inflow information under the operating environment of the current-limiting loops; acquiring infrared thermal images of the preliminary abnormal current-limiting loop under the K test flows; generating infrared thermal images of the preliminary abnormal current-limiting loop under the multiple simulated flows based on the infrared thermal images of the preliminary abnormal current-limiting loop under the K test flows; and determining whether the preliminary abnormal current-limiting loop is qualified based on the infrared thermal images of the preliminary abnormal current-limiting loop under the K test flows and the infrared thermal images of the preliminary abnormal current-limiting loop under the multiple simulated flows.

[0015] This invention provides a method and system for detecting current-limiting loops. The method includes acquiring ultrasonic images of multiple current-limiting loops and inflow information under the operating environment of the current-limiting loops; determining a preliminary abnormal current-limiting loop using an abnormal current-limiting loop model based on the ultrasonic images of the multiple current-limiting loops; determining K test flow rates and multiple simulated flow rates based on the inflow information under the operating environment of the current-limiting loops; acquiring infrared thermal images of the preliminary abnormal current-limiting loop under the K test flow rates; generating infrared thermal images of the preliminary abnormal current-limiting loop under the multiple simulated flow rates based on the infrared thermal images of the preliminary abnormal current-limiting loop under the K test flow rates; and determining whether the preliminary abnormal current-limiting loop is qualified based on the infrared thermal images of the preliminary abnormal current-limiting loop under the K test flow rates and the infrared thermal images of the preliminary abnormal current-limiting loop under the multiple simulated flow rates. This method can efficiently, comprehensively, and accurately detect whether the current-limiting loop is qualified. Attached Figure Description

[0016] Figure 1 A schematic flowchart of a current-limiting ring detection method provided in an embodiment of the present invention;

[0017] Figure 2 A schematic diagram of a current-limiting ring provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of an ultrasonic testing device provided in an embodiment of the present invention;

[0019] Figure 4 This is a flowchart illustrating the process of determining K test traffic flows and multiple simulated traffic flows, provided as an embodiment of the present invention.

[0020] Figure 5 A schematic diagram of a process for determining whether a preliminary abnormal current-limiting loop is qualified, provided for an embodiment of the present invention;

[0021] Figure 6 A schematic diagram of a detection system of a flow restrictor ring is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0022] The application will be further described below in conjunction with the drawings. Like elements in different embodiments are denoted by like reference numerals. In the following embodiments, many details are described in order to provide a more thorough understanding of the present application. However, it will be apparent to one skilled in the art that some features can be omitted, or replaced by other elements, materials, methods, in different cases. In some cases, some operations related to the present application are not shown or described in the specification, in order to avoid the core of the present application being overwhelmed by too much description, and it is not necessary for one skilled in the art to describe these related operations in detail according to the description in the specification and general technical knowledge in the art.

[0023] In the embodiments of the present application, a detection method of a flow restrictor ring is provided, as shown in the figure. Figure 1 The detection method of the flow restrictor ring includes steps S1-S6:

[0024] Step S1: Obtain ultrasonic images of multiple flow restrictor rings and flow information of inflow under the working environment of the flow restrictor rings.

[0025] The flow restrictor ring is a functional component installed in the aviation pipeline and can be used to limit the flow of liquid in the pipeline. The flow restrictor ring can generate a certain resistance to the liquid flowing through by its specific structural design, such as internal passage diameter, throttling structure, etc., so as to control the liquid flow within a reasonable range required by the aviation pipeline operation, so as to ensure the stable operation of the aviation pipeline system, and avoid affecting the normal operation of the aviation equipment due to excessive or insufficient flow. Figure 2 A schematic diagram of a flow restrictor ring is provided for the embodiments of the present application.

[0026] The ultrasonic images of the multiple flow restrictor rings are images obtained by scanning the multiple flow restrictor rings in the aviation pipeline by an ultrasonic detection device. Figure 3 A schematic diagram of an ultrasonic detection device is provided for the embodiments of the present application. The ultrasonic images of the flow restrictor ring can record the internal structure, topographic features of the flow restrictor ring, and acoustic echo distribution of the flow restrictor ring under the running state.

[0027] The ultrasonic images of the flow restrictor ring can be used to reflect the material integrity, geometric deviation, and possible defect distribution of the flow restrictor ring.

[0028] The inflow flow information under the working environment of the flow restriction ring refers to a set of measurement values about fluid flow rate, instantaneous flow and flow variation trend obtained by the pipeline sensor and the flow monitoring device under the working condition of the pipeline where the flow restriction ring is located.

[0029] The inflow flow information under the working environment of the flow restriction ring can represent the input side fluid condition and fluctuation characteristics under the working environment of the flow restriction ring.

[0030] In step S2, a preliminary abnormal flow restriction ring is determined based on the ultrasonic images of the plurality of flow restriction rings using an abnormal flow restriction ring model.

[0031] The abnormal flow restriction ring model is a convolutional neural network model. The input of the abnormal flow restriction ring model is the ultrasonic images of the plurality of flow restriction rings, and the output of the abnormal flow restriction ring model is the preliminary abnormal flow restriction ring.

[0032] The convolutional neural network model includes a convolutional neural network (CNN), which is a kind of feedforward neural network with convolution operation as the core and capable of processing two-dimensional or multi-dimensional grid data. The convolutional neural network can extract local spatial features and construct hierarchical abstract representations layer by layer through multiple convolution layers, activation layers and down-sampling layers. The weights of the convolution kernel can be optimized through supervised learning to enable the model to distinguish between normal and abnormal image patterns. The convolutional neural network has good performance and generalization ability in tasks such as edge detection, texture recognition and defect positioning.

[0033] The preliminary abnormal flow restriction ring refers to a set of flow restriction rings that may have structural abnormalities, damages or abnormal working states determined by the abnormal flow restriction ring model after analyzing the ultrasonic images of the plurality of flow restriction rings.

[0034] The ultrasonic images of the plurality of flow restriction rings can provide spatially distributed acoustic response information for the convolutional neural network model. These acoustic responses can reflect the geometric characteristics and possible defect traces of the flow restriction ring through image pixel intensity and texture patterns.

[0035] The convolutional neural network can preprocess the ultrasonic images of the flow restriction ring, including normalization, necessary cropping and window function operation to reduce noise and standardize scale. Then the convolutional neural network extracts local acoustic texture and edge features in the ultrasonic images of the flow restriction ring through multiple convolution layers, and then realizes the aggregation of spatial information and reduces the dimension through pooling and down-sampling operations. The convolutional neural network can form a high-level semantic representation of abnormal patterns through stacked convolution and nonlinear activation in a deeper network. The model can map the extracted features to an abnormal probability distribution through multiple fully connected layers, and on this basis, it can give an abnormal score and a classification label to each flow restriction ring sample. Finally, the model can output the abnormal probability of each flow restriction ring sample in the inference stage and filter out the preliminary abnormal flow restriction ring according to the set threshold.

[0036] In some embodiments, determining the preliminary abnormal flow-restricting ring using the abnormal flow-restricting ring model based on the ultrasonic images of the plurality of flow-restricting rings includes steps S21-S23:

[0037] In step S21, internal structural integrity information, surface state information, key dimension accuracy information, and flow-restricting channel defect occlusion ratio information of each flow-restricting ring are determined based on the ultrasonic images of the plurality of flow-restricting rings.

[0038] In some embodiments, the internal structural integrity information, the surface state information, the key dimension accuracy information, and the flow-restricting channel defect occlusion ratio information of each flow-restricting ring can be determined using a convolutional neural network.

[0039] The internal structural integrity information is the overall structural state information of the flow-restricting ring determined by the convolutional neural network. The internal structural integrity information includes whether the flow-restricting ring has defects such as cracks, holes, and structural fractures, as well as quantitative descriptions of the distribution range and severity of the defects.

[0040] The surface state information is the appearance feature information of the inner and outer surfaces of the flow-restricting ring determined by the convolutional neural network based on the ultrasonic images of the flow-restricting ring. The surface state information includes the degree of surface wear, the distribution of corrosion traces, the flatness, the number and size of scratches or dents, and the surface deformation of the flow-restricting ring.

[0041] The key dimension accuracy information is the deviation information of the actual value of the core functional dimension of the flow-restricting ring from the design standard size determined by the convolutional neural network. The key dimension accuracy information includes the actual value and deviation of the inner diameter, the width and deviation of the flow-restricting channel, the thickness and deviation of the end face, and the diameter and deviation of the sealing surface.

[0042] The flow-restricting channel defect occlusion ratio information of each flow-restricting ring is the visual coverage range of defects such as cracks, holes, and inclusions in the flow-restricting channel of each flow-restricting ring determined by the convolutional neural network based on the ultrasonic images of the plurality of flow-restricting rings.

[0043] The convolutional neural network has progressive extraction capability of the ultrasonic image features of the flow restrictor through multi-layer convolution operation. After the input layer of the convolutional neural network receives the original pixel information of the ultrasonic image of the flow restrictor, the shallow convolution layer can capture basic features such as edge gray level mutation, so as to identify the profile boundary of the flow channel of the flow restrictor, surface wear texture change and internal defect shape profile. The middle convolution layer can abstractly combine these basic features, and the middle convolution layer can distinguish different defects such as cracks and holes, quantify the distribution range and shielding proportion of the defects in the flow channel, and extract the geometric features of the key size such as the inner diameter of the flow channel. The deep convolution layer can comprehensively judge the internal structure integrity state and the overall features of the surface state by combining global and local features. After a large number of labeled samples are trained, the convolutional neural network can learn the mapping relationship between the ultrasonic image features of the flow restrictor and the structure parameters of the flow restrictor, so as to realize the accurate conversion of the ultrasonic image of the flow restrictor to specific information.

[0044] In step S22, the liquid flow resistance prediction value, the flow channel flow capacity evaluation grade, the flow restriction structure integrity risk grade, and the flow restriction function effective guarantee coefficient of each flow restrictor are determined based on the internal structure integrity information, the surface state information, the key size precision information, and the flow channel defect shielding proportion information of each flow restrictor.

[0045] In some embodiments, a deep neural network can be used to determine the liquid flow resistance prediction value, the flow channel flow capacity evaluation grade, the flow restriction structure integrity risk grade, and the flow restriction function effective guarantee coefficient of each flow restrictor.

[0046] The liquid flow resistance prediction value is the expected resistance value of the liquid flowing through the flow restrictor output by the deep neural network.

[0047] The flow channel flow capacity evaluation grade is the ability grade of the flow channel to allow the liquid to pass, which is output by the deep neural network. The higher the flow channel flow capacity evaluation grade, the stronger the potential of the flow channel of the flow restrictor, and the better the smoothness of the liquid flowing through.

[0048] The flow restriction structure integrity risk grade is the risk grade of the flow restrictor to maintain the stability of the structure, which is output by the deep neural network. The higher the flow restriction structure integrity risk grade, the higher the severity of the structural defects such as cracks and interlayer peeling of the flow restrictor, and the greater the probability of structural failure under working conditions.

[0049] The effective guarantee coefficient of the flow limiting function is a coefficient output by the deep neural network for quantifying the comprehensive guarantee capability of the flow limiting ring in realizing the design of the flow limiting function. The effective guarantee coefficient of the flow limiting function ranges from 0 to 1. The closer the value is to 1, the better the flow capacity, structural stability and fluid flow adaptability of the flow limiting ring, and the more stable the flow limiting ring can achieve the design flow limiting requirement. The closer the value is to 0, the weaker the comprehensive flow limiting guarantee capability of the flow limiting ring, and the greater the probability of flow limiting failure.

[0050] The deep neural network is good at deep fusion and reasoning of multi-dimensional features. The model can convert the internal structural integrity information, surface state information, key size precision information and flow limiting channel defect shielding proportion information of each flow limiting ring into a standardized feature vector. Then, through the collaborative operation of multiple neurons, the model can accurately capture the internal correlation between the information, such as the comprehensive influence of the quantified flow limiting channel defect shielding proportion and the key size precision information on the liquid flow resistance. The model can determine the flow limiting structural integrity risk level in combination with the internal structural integrity information and the surface state information of the flow limiting ring.

[0051] In step S23, a preliminary abnormal flow limiting ring is determined based on the liquid flow resistance prediction value of each flow limiting ring, the flow limiting channel flow capacity evaluation level, the flow limiting structural integrity risk level and the effective guarantee coefficient of the flow limiting function.

[0052] In some embodiments, a deep neural network can be used to determine the preliminary abnormal flow limiting ring.

[0053] The deep neural network can convert the liquid flow resistance prediction value, the flow limiting channel flow capacity evaluation level, the flow limiting structural integrity risk level and the effective guarantee coefficient of the flow limiting function of each flow limiting ring into high-dimensional features through multiple layers of nonlinear transformation, and accurately capture the synergistic influence and abnormal threshold boundary between the parameters. The deep neural network can refer to the parameter distribution law of normal and abnormal flow limiting rings learned in training to quantify the deviation of each parameter from the standard range, and give higher weight to key parameters such as the effective guarantee coefficient of the flow limiting function. The model can accurately distinguish between normal and abnormal states by comprehensively evaluating the rationality of the liquid flow resistance prediction value, the compliance of the flow capacity level and the level of the structural integrity risk level, and finally determine the determination result of the preliminary abnormal flow limiting ring.

[0054] In step S3, K test flows and multiple simulation flows are determined based on the inflow flow information under the working environment of the flow limiting ring.

[0055] In some embodiments, Figure 4 A flowchart for determining K test flows and multiple simulation flows is provided for the embodiments of the present application. The determination of K test flows and multiple simulation flows includes steps S31-S34.

[0056] Step S31, determining the K value based on the inflow traffic information under the flow limiting ring working environment.

[0057] In some embodiments, the K value is determined using a complexity evaluation model based on the inflow traffic information under the flow limiting ring working environment. The complexity evaluation model is a Transformer model. The input of the complexity evaluation model is the inflow traffic information under the flow limiting ring working environment, and the output of the complexity evaluation model is the K value.

[0058] The Transformer model is a sequence modeling structure based on self-attention mechanism. The Transformer model includes an encoder and a decoder, and each part is stacked by multiple identical layers. The role of the encoder is to perform representation learning on the input sequence, which includes self-attention mechanism and feed-forward neural network. The decoder additionally introduces a multi-head attention mechanism on the basis of the encoder. The decoder can be used to decode the encoder output and generate target sequences.

[0059] The K value is a cluster number parameter output by the complexity evaluation model for clustering processing. The K value represents the number of representative test traffic categories obtained by complexity evaluation from the inflow traffic information. The K value can be used to guide the K-means clustering algorithm to divide the cluster number of traffic samples.

[0060] The inflow traffic information under the flow limiting ring working environment includes the amplitude, periodic variation and mutation event of the fluid under different operating conditions. Therefore, the model can calculate the internal distribution complexity and typical mode number of the data by encoding these time series signals, and then infer the cluster number K value suitable for representative test traffic selection.

[0061] The transformer model can arrange the inflow flow information under the flow-limiting ring working environment in time sequence into sequence data and as the input sequence of the model. The self-attention mechanism in the encoder can calculate the association weight of each time step flow data with all other time step data, thereby capturing the dependence of the flow data in the time dimension, such as the association between flow peaks and troughs, the connection characteristics of stable and fluctuating sections, etc. The decoder part combines the multi-head attention mechanism to deeply analyze the features output by the encoder and mine the distribution rules and complexity of the flow data. The model can simultaneously observe all data points in the entire time sequence and calculate the mutual influence and association strength between the flow values of any two time points. For example, the model can identify that the flow values in a period of time have fluctuations but the overall trend is stable, thereby forming a pattern; at the same time, it can also identify that the flow has periodic sharp peaks in another period of time, thereby forming another completely different pattern. By modeling the global dependence of the entire sequence, the transformer model can quantitatively analyze the differences between these different patterns and the consistency of each internal pattern, thereby identifying how many essentially different flow behavior patterns exist in the data in total, and finally the model can determine the number of all essentially different flow behavior patterns identified as the K value.

[0062] In step S32, K clusters are obtained based on the inflow flow information under the flow-limiting ring working environment and the K value using a K-means clustering algorithm.

[0063] The K-means clustering algorithm is a divisive clustering method based on the distance between samples. The K-means clustering algorithm can initially divide the samples with K clusters and update the cluster centers through iteration to minimize the within-cluster sum of squares, so that the data points in each cluster have high similarity, while the data points in different clusters have low similarity.

[0064] The K clusters are K groups of flow data with similar characteristics obtained by clustering the inflow flow information under the flow-limiting ring working environment using the K-means clustering algorithm. The flow data in each cluster has consistency in terms of change trend, value range, fluctuation characteristics, etc., and can represent a typical flow working condition, while the data points between different clusters show obvious differences.

[0065] The inflow flow information under the flow-limiting ring working environment contains a large number of instantaneous flow measurement values, and these data points are distributed in the value space. The K value can provide the number of target clusters for the K-means clustering algorithm, and the K value can limit the number of different working states that need to be divided from the flow information.

[0066] In some embodiments, when clustering the inflow traffic information under the throttling ring working environment using the K-means clustering algorithm, K data points can be randomly selected from the traffic data as initial cluster centers. Then, the Euclidean distance of each traffic data point to each initial cluster center is calculated, and each data point is assigned to the corresponding cluster according to the nearest distance principle to form the initial K clusters. Next, for each cluster, the mean of all traffic data points in the cluster is calculated, and the mean is used as a new cluster center to replace the original initial cluster center. Then, the distance of each data point to the new cluster center is recalculated, and the cluster assignment is performed again to update the members of each cluster. The cluster center updating and data point assignment process is repeated until the change in the cluster center is less than a preset threshold or the maximum number of iterations is reached. At this time, the clustering is completed, and the K clusters obtained by K-means clustering are the final results. The traffic data in each cluster has high similarity and can represent different types of traffic conditions.

[0067] In step S33, K test traffics are determined based on the K clusters, each test traffic being selected from each cluster.

[0068] In some embodiments, the K test traffics can be determined using a test traffic model. The test traffic model is a Transformer model. The input of the test traffic model is the K clusters, and the output of the test traffic model is the K test traffics.

[0069] The K test traffics are representative traffic values selected from the K clusters by a test traffic determination model. Each test traffic is selected from each cluster, and each cluster corresponds to a test traffic.

[0070] The K test traffics can reflect the traffic characteristics of the corresponding clusters and cover the typical working condition space of the inflow traffic.

[0071] Each cluster in the K clusters contains a group of traffic data with similar characteristics. The data in these clusters can represent specific traffic conditions, and the model can select the traffic that best reflects the characteristics of each cluster as a test traffic from each cluster.

[0072] The encoder of the transformer model can analyze the distribution characteristics of the traffic data in each cluster through the self-attention mechanism, and can capture key information such as the concentration trend, peak value, and valley value of the data, thereby identifying the most representative range of traffic values in the cluster. The multi-head attention mechanism can simultaneously focus on different dimensional features of the traffic data in the cluster, such as the value size, change rate, and duration, and can comprehensively evaluate the representativeness of each traffic data in the cluster. The decoder can sort and filter the traffic data in each cluster based on the feature information output by the encoder, and can preferentially select traffic values that are near the cluster center, have a high frequency of occurrence, and can cover the main characteristics in the cluster, and use them as the test traffic corresponding to the cluster.

[0073] In step S34, a plurality of simulation traffics are determined based on the K clusters and the K test traffics.

[0074] In some embodiments, a simulation traffic model can be used to determine the plurality of simulation traffics. The simulation traffic model is a deep neural network model. The input of the simulation traffic model is the K clusters and the K test traffics, and the output of the simulation traffic model is the plurality of simulation traffics.

[0075] The deep neural network model includes a deep neural network (DNN). The deep neural network is a neural network model composed of multiple hidden layers. Through the non-linear transformation of multiple layers of neurons, the deep neural network can learn the complex mapping relationship in the data. The deep neural network has strong feature learning and fitting capabilities, and can handle high-dimensional data and complex function mapping problems.

[0076] The plurality of simulation traffics is a set of traffic values that numerically extend and supplement the K test traffics, generated by the simulation traffic model based on the K clusters and the K test traffics.

[0077] The plurality of simulation traffics can cover the range of traffic changes that may occur in each cluster but are not directly represented by the test traffic, to achieve more comprehensive detection. The plurality of simulation traffics and the test traffic complement each other, and can comprehensively reflect various traffic scenarios that the flow limiting ring may face.

[0078] The deep neural network can utilize the data distribution information of the K clusters to learn the intrinsic law of each traffic pattern. The model can take the K test traffics as the starting point or center of generation, and refer to the information such as the boundary, density distribution and variation trend of the cluster where the test traffic is located. For example, for a certain cluster, the model analyzes whether the data points are concentrated around the test traffic or uniformly distributed in a wider interval. The hidden layer of the deep neural network can learn the nonlinear generation relationship from the "test traffic" to "other traffics in the cluster". When generating multiple simulation traffics, the model will interpolate and extrapolate based on the learned distribution law on the basis of the K test traffics, to create new traffic values that belong to the traffic pattern represented by the cluster and have certain differences from the test traffic. For example, in a cluster representing traffic fluctuating from 100 units / sec to 120 units / sec, if the test traffic is 110 units / sec, the model may generate 105, 115 and other simulation traffics to more finely detect the performance of the flow limiting ring in this working interval.

[0079] Step S4, acquiring an infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics.

[0080] The infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics is a video acquired by an infrared thermal imager when the preliminary abnormal flow limiting ring is under the action of K test traffics respectively. The infrared thermal image video records the dynamic process of the temperature field of the surface of the flow limiting ring changing with time.

[0081] The infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics can reflect the heat distribution caused by friction, throttling and other effects when the liquid flows through the flow limiting ring.

[0082] Step S5, generating an infrared thermal image video of the preliminary abnormal flow limiting ring under multiple simulation traffics based on the infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics.

[0083] In some embodiments, a generative adversarial network can be used to generate an infrared thermal image video of the preliminary abnormal flow limiting ring under multiple simulation traffics based on the infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics. The input of the generative adversarial network is the infrared thermal image video of the preliminary abnormal flow limiting ring under K test traffics, and the output of the generative adversarial network is the infrared thermal image video of the preliminary abnormal flow limiting ring under multiple simulation traffics.

[0084] A generative adversarial network (GAN) consists of two competing neural networks, a generator and a discriminator. The generator's task is to learn the distribution of real data and generate new, fake data that resembles the real data. The discriminator's task is to accurately determine whether the input data is from the real dataset or generated by the generator. Both are trained through an adversarial game process, and the discriminator eventually drives the generator to produce highly realistic data.

[0085] The infrared thermal image video of the preliminary abnormal restrictor under multiple simulation flow rates is generated by the generative adversarial network, which can simulate the dynamic process of the surface temperature field of the restrictor over time under multiple simulation flow rates.

[0086] The infrared thermal image video of the preliminary abnormal restrictor under K test flow rates records the thermal characteristics of the restrictor under different flow rates, which contain the correlation between flow rate and temperature change, providing learning samples for the generative adversarial network, so that the generative adversarial network can generate infrared thermal image videos under simulation flow rates based on these samples.

[0087] The generative adversarial network can decompose the infrared thermal image video of the preliminary abnormal restrictor under K test flow rates into image sequences and perform standardization processing. The generator receives random noise signals and image sequence features under test flow rates, and then learns the temperature distribution pattern, change trend, and other features of the infrared thermal image under test flow rates through transpose convolution, Batch Normalization, etc., and generates infrared thermal image frames under simulation flow rates similar to the features of the real video. The discriminator receives real infrared thermal image frames under test flow rates and the infrared thermal image frames generated by the generator under simulation flow rates, and then extracts features through convolution layers to determine the authenticity of the input frames, and then feeds back the discrimination results to the generator. The generator can adjust the network parameters according to the feedback of the discriminator to optimize the authenticity of the generated frames, and the discriminator can continuously improve the ability to distinguish real frames and generated frames. After multiple rounds of adversarial training, the generator can generate infrared thermal image frames under simulation flow rates that are highly consistent with the features of real infrared thermal image videos, and then combine these frames in order to obtain the infrared thermal image video of the preliminary abnormal restrictor under multiple simulation flow rates.

[0088] Step S6, determining whether the preliminary abnormal restrictor is qualified based on the infrared thermal image video of the preliminary abnormal restrictor under K test flow rates and the infrared thermal image video of the preliminary abnormal restrictor under multiple simulation flow rates.

[0089] In some embodiments, Figure 5A flowchart for determining whether the preliminary abnormal current limiting ring is qualified is provided for the embodiment of the present application, and the determination of whether the preliminary abnormal current limiting ring is qualified comprises steps S61-S62:

[0090] In step S61, an infrared spectrum is constructed, the infrared spectrum comprises a plurality of nodes and edges between the plurality of nodes, the plurality of nodes comprise K test flow nodes and a plurality of simulation flow nodes, each simulation flow node establishes an edge with the K test flow nodes respectively, the edges between the nodes are the differences between the simulation flow and the test flow, the node characteristics of the test flow nodes are the infrared thermal image videos of the preliminary abnormal current limiting ring under the test flow, and the node characteristics of the simulation flow nodes are the infrared thermal image videos of the preliminary abnormal current limiting ring under the simulation flow.

[0091] The spectrum is a kind of structured data composed of nodes (vertices) and edges (edges), which is used to represent the relationship between nodes. The infrared spectrum is a kind of spectrum data comprising a plurality of nodes and edges between the nodes, and the infrared spectrum can integrate the infrared thermal image video information and the flow difference relationship of the preliminary abnormal current limiting ring under the test flow and the simulation flow.

[0092] The K test flow nodes are nodes in the infrared spectrum for storing the characteristic information corresponding to the K test flows, each test flow node corresponds to a test flow, and the node characteristics of the test flow node are the infrared thermal image videos of the preliminary abnormal current limiting ring under the test flow.

[0093] The plurality of simulation flow nodes are nodes in the infrared spectrum for storing the characteristic information corresponding to the plurality of simulation flows, each simulation flow node corresponds to a simulation flow, and the node characteristics of the simulation flow node are the infrared thermal image videos of the preliminary abnormal current limiting ring under the simulation flow.

[0094] The edges between the nodes connect the simulation flow nodes and the test flow nodes in the infrared spectrum, and the numerical values represented by the edges are the differences between the corresponding simulation flow and the test flow. The edges of the infrared spectrum can quantify the difference degree between different flows.

[0095] In step S62, the infrared spectrum is processed based on a graph neural network to determine whether the preliminary abnormal current limiting ring is qualified.

[0096] The graph neural network (GNN) is a kind of deep learning model capable of processing spectrum, and the graph neural network can learn and reason by using the topological structure and node characteristics of the spectrum. The graph neural network updates its own node characteristics by aggregating the feature information of the neighbor nodes, so as to capture the association relationship between the nodes. The input of the graph neural network is the infrared spectrum, and the output of the graph neural network is whether the preliminary abnormal current limiting ring is qualified.

[0097] The pass or fail of the preliminary abnormal flow limiting ring is determined by analyzing the infrared spectrum by the graph neural network, and is a judgment result about whether the preliminary abnormal flow limiting ring can meet the working requirements of the aviation pipeline flow limiting. Pass means that the flow limiting ring can stably play the flow limiting function under the working conditions of the aviation pipeline and avoid the influence of abnormal flow on the operation of aviation equipment. Fail means that the flow limiting performance or structural stability of the flow limiting ring cannot meet the working requirements of the aviation pipeline.

[0098] By constructing the infrared spectrum, the internal correlation network between the thermal responses of the flow limiting ring under different flow conditions can be clearly reflected. This correlation information is crucial for accurate qualification determination, because defects of the flow limiting ring may not be exposed at a single flow point, but in the abnormality of the thermal response law when the flow changes. The infrared thermal image video of the preliminary abnormal flow limiting ring under the test flow and the infrared thermal image video of the preliminary abnormal flow limiting ring under multiple simulated flows are taken as node features, and the difference between the simulated flow and the test flow is taken as an edge. This can organically integrate the real physical test data and the generated simulation prediction data in the same structure. This helps the model to more comprehensively understand the consistency and continuity of the thermodynamic behavior of the flow limiting ring in the entire working range. The graph neural network can effectively learn the complex heat conduction relationship between nodes in the infrared spectrum based on flow changes, so as to more accurately mine whether the overall thermal response mode of the flow limiting ring conforms to the physical law. Compared with analyzing the video at each flow point in isolation, the graph neural network has stronger ability to capture the global pattern and dynamic evolution law of heat distribution when processing infrared spectrum containing physical correlation, so that a more reliable qualification judgment can be made.

[0099] The graph neural network can propagate and aggregate information on the infrared spectrum. For a simulated flow node on the infrared spectrum, the model can aggregate the feature information of K test flow nodes connected to it along the edge, and consider the weight of the edge in the aggregation process. The smaller the flow difference of the test flow node, the greater the influence of the node feature information on the current simulated flow node. Based on this way, the feature of the simulated flow node can be supplemented and calibrated by the real feature information of the connected test flow nodes. Through multiple rounds of message passing, the feature of each node can be fused with the relevant information of its neighbor nodes, thereby forming a node embedding vector that can reflect the global topology relationship and local features. The graph neural network can perform global pooling processing on the embedding vectors of all nodes to obtain the global feature vector of the entire infrared spectrum, and the vector integrates the performance characteristics of the flow limiting ring under all test flows and simulated flow conditions. Finally, the model can input the global feature vector into the classification layer, and then output the probability value of the preliminary abnormal flow limiting ring passing through the activation function. When the probability value exceeds the preset threshold, it is determined that the preliminary abnormal flow limiting ring is qualified, otherwise it is determined as unqualified.

[0100] based on the same inventive concept, Figure 6 A detection system of a flow limiting ring is provided for an embodiment of the present application, and the detection system of the flow limiting ring comprises:

[0101] An information acquisition module 71 is configured to acquire ultrasonic images of multiple flow limiting rings and flow information of inflow under a working environment of the flow limiting rings.

[0102] A preliminary anomaly determination module 72 is configured to determine a preliminary abnormal flow limiting ring by using an abnormal flow limiting ring model based on the ultrasonic images of the multiple flow limiting rings.

[0103] A flow determination module 73 is configured to determine K test flows and multiple simulation flows based on the flow information of the inflow under the working environment of the flow limiting rings.

[0104] A test video acquisition module 74 is configured to acquire infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows.

[0105] A simulation video generation module 75 is configured to generate infrared thermal image videos of the preliminary abnormal flow limiting ring under the multiple simulation flows based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows.

[0106] A qualification determination module 76 is configured to determine whether the preliminary abnormal flow limiting ring is qualified based on the infrared thermal image videos of the preliminary abnormal flow limiting ring under the K test flows and the infrared thermal image videos of the preliminary abnormal flow limiting ring under the multiple simulation flows.

[0107] It should be noted that, in order to simplify the expressions disclosed in the present specification and to help the understanding of one or more embodiments of the present application, in the foregoing description of the embodiments of the present specification, various features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the present specification are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the disclosed single embodiments.

[0108] Finally, it should be understood that the embodiments described in the present specification are only used to illustrate the principles of the embodiments of the present specification. Other variations can also belong to the scope of the present specification. Therefore, as an example but not limitation, alternative configurations of the embodiments of the present specification can be considered consistent with the teachings of the present specification. Accordingly, the embodiments of the present specification are not limited to the embodiments explicitly introduced and described in the present specification.

Claims

1. A method for detecting a current-limiting ring, characterized in that, include: Acquire ultrasonic images of multiple flow-limiting rings and inflow information under the working environment of the flow-limiting rings; Based on the ultrasonic images of the multiple current-limiting loops, an abnormal current-limiting loop model is used to determine the preliminary abnormal current-limiting loops. This determination includes: Based on the ultrasonic images of the multiple flow-limiting rings, the internal structural integrity information, surface condition information, key dimension accuracy information, and defect occlusion ratio information of each flow-limiting ring are determined. Based on the internal structural integrity information, surface condition information, key dimension accuracy information, and the proportion of defects obstructing the flow restriction channel of each flow restriction ring, the predicted value of liquid flow resistance, the flow capacity assessment level of the flow restriction channel, the risk level of the integrity of the flow restriction structure, and the effective guarantee coefficient of the flow restriction function of each flow restriction ring are determined. Based on the predicted liquid flow resistance of each flow-limiting ring, the assessment level of the flow capacity of the flow-limiting channel, the risk level of the integrity of the flow-limiting structure, and the effective guarantee coefficient of the flow-limiting function, the preliminary abnormal flow-limiting rings are determined. Based on the inflow information under the current limiting ring operating environment, K test flows and multiple simulated flows are determined. This determination includes: The K value is determined based on the inflow flow information under the current limiting ring working environment. The determination of the K value based on the inflow flow information under the current limiting ring working environment includes: using a complexity evaluation model to determine the K value based on the inflow flow information under the current limiting ring working environment, wherein the complexity evaluation model is the Transformer model. Based on the inflow information under the working environment of the flow limiting ring, the K value is used to obtain K clusters using the K-means clustering algorithm; K test traffic flows are determined based on the K clusters, and each test traffic flow is selected from each cluster; Multiple simulated traffic flows are determined based on the K clusters and the K test traffic flows. The determination of multiple simulated traffic flows based on the K clusters and the K test traffic flows includes: using a simulated traffic flow model to determine multiple simulated traffic flows, wherein the simulated traffic flow model is a deep neural network model. Obtain infrared thermal images of the initial abnormal current limiting loop under K test flow rates; Based on the infrared thermal images of the initial abnormal current limiting loop under K test flow rates, infrared thermal images of the initial abnormal current limiting loop under multiple simulated flow rates are generated. This generation of infrared thermal images of the initial abnormal current limiting loop under multiple simulated flow rates based on the infrared thermal images of the initial abnormal current limiting loop under K test flow rates includes: Based on the infrared thermal images of the initial abnormal current limiting loop under K test flows, a generative adversarial network is used to generate infrared thermal images of the initial abnormal current limiting loop under multiple simulated flows. The determination of whether the preliminary abnormal current limiting ring is qualified is based on the infrared thermal images of the preliminary abnormal current limiting ring under K test flow rates and the infrared thermal images of the preliminary abnormal current limiting ring under multiple simulated flow rates. This determination includes: An infrared spectrum is constructed, which includes multiple nodes and edges between the nodes. The multiple nodes include K test flow nodes and multiple simulated flow nodes. Each simulated flow node establishes an edge with each of the K test flow nodes. The edge between the nodes is the difference between the simulated flow and the test flow. The node feature of the test flow node is the infrared thermal image video of the preliminary abnormal current limiting loop under the test flow. The node feature of the simulated flow node is the infrared thermal image video of the preliminary abnormal current limiting loop under the simulated flow. The infrared spectrum is processed using a graph neural network to determine whether the preliminary abnormal current-limiting loop is qualified.

2. A detection system for a current-limiting ring, characterized in that, include: The information acquisition module is used to acquire ultrasonic images of multiple flow-limiting rings and inflow information under the working environment of the flow-limiting rings. The preliminary anomaly determination module is used to determine preliminary abnormal current-limiting loops based on the ultrasonic images of the multiple current-limiting loops using an abnormal current-limiting loop model. The determination of preliminary abnormal current-limiting loops based on the ultrasonic images of the multiple current-limiting loops using the abnormal current-limiting loop model includes: Based on the ultrasonic images of the multiple flow-limiting rings, the internal structural integrity information, surface condition information, key dimension accuracy information, and defect occlusion ratio information of each flow-limiting ring are determined. Based on the internal structural integrity information, surface condition information, key dimension accuracy information, and the proportion of defects obstructing the flow restriction channel of each flow restriction ring, the predicted value of liquid flow resistance, the flow capacity assessment level of the flow restriction channel, the risk level of the integrity of the flow restriction structure, and the effective guarantee coefficient of the flow restriction function of each flow restriction ring are determined. Based on the predicted liquid flow resistance of each flow-limiting ring, the assessment level of the flow capacity of the flow-limiting channel, the risk level of the integrity of the flow-limiting structure, and the effective guarantee coefficient of the flow-limiting function, the preliminary abnormal flow-limiting rings are determined. The flow determination module is used to determine K test flow rates and multiple simulated flow rates based on the inflow flow information under the operating environment of the current limiting loop. The flow determination module is also used for: The K value is determined based on the inflow flow information under the current limiting ring working environment. The determination of the K value based on the inflow flow information under the current limiting ring working environment includes: using a complexity evaluation model to determine the K value based on the inflow flow information under the current limiting ring working environment, wherein the complexity evaluation model is the Transformer model. Based on the inflow information under the working environment of the flow limiting ring, the K value is used to obtain K clusters using the K-means clustering algorithm; K test traffic flows are determined based on the K clusters, and each test traffic flow is selected from each cluster; Multiple simulated traffic flows are determined based on the K clusters and the K test traffic flows. The determination of multiple simulated traffic flows based on the K clusters and the K test traffic flows includes: using a simulated traffic flow model to determine multiple simulated traffic flows, wherein the simulated traffic flow model is a deep neural network model. The test video acquisition module is used to acquire infrared thermal images of the initial abnormal current limiting loop under K test flows; The simulation video generation module is used to generate infrared thermal images of the initial abnormal current limiting loop under multiple simulated flow rates based on the infrared thermal images of the initial abnormal current limiting loop under K test flow rates. The simulation video generation module is also used for: Based on the infrared thermal images of the initial abnormal current limiting loop under K test flows, a generative adversarial network is used to generate infrared thermal images of the initial abnormal current limiting loop under multiple simulated flows. The pass / fail determination module is used to determine whether the preliminary abnormal current limiting loop is qualified based on the infrared thermal images of the preliminary abnormal current limiting loop under K test flow rates and the infrared thermal images of the preliminary abnormal current limiting loop under multiple simulated flow rates. The pass / fail determination module is also used for: An infrared spectrum is constructed, which includes multiple nodes and edges between the nodes. The multiple nodes include K test flow nodes and multiple simulated flow nodes. Each simulated flow node establishes an edge with each of the K test flow nodes. The edge between the nodes is the difference between the simulated flow and the test flow. The node feature of the test flow node is the infrared thermal image video of the preliminary abnormal current limiting loop under the test flow. The node feature of the simulated flow node is the infrared thermal image video of the preliminary abnormal current limiting loop under the simulated flow. The infrared spectrum is processed using a graph neural network to determine whether the preliminary abnormal current-limiting loop is qualified.

3. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the current-limiting loop detection method as described in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the current-limiting loop detection method as described in claim 1.

Images