An RFID-based defect early warning method, electronic device, computer readable storage medium and program product

Abstract

The application relates to the field of defect early warning, and provides a defect early warning method based on RFID, an electronic device, a computer readable storage medium and a program product. The method comprises the following steps: acquiring an internal pressure value of a gas cylinder and temperature values and strain values of a plurality of detection points on the surface of the gas cylinder; obtaining temperature compensation strain and a first coefficient based on the temperature values and the strain values; constructing a graph structure based on the temperature values, the strain values, the temperature compensation strain and the first coefficient; inputting the graph structure into a deep learning model to obtain first coupling features, and obtaining temperature corrected strain values based on the first coupling features; and when the temperature corrected strain values exceed a preset strain threshold value and the pressure value exceeds a preset pressure threshold value, it is determined that the area has a defect risk. The application can improve the problem of low compensation accuracy of strain caused by temperature on strain caused by defects in gas cylinder fatigue testing, and improve the early warning accuracy of gas cylinder production defects.

Description

Technical Field

[0001] This invention relates to the field of object defect early warning technology, and more specifically, to an RFID-based defect early warning method, electronic device, computer-readable storage medium, and program product. Background Technology

[0002] With the rapid development of the hydrogen energy industry, fiber-wound composite gas cylinders with plastic liner are core components of high-pressure hydrogen storage systems. Fatigue performance is one of the most critical safety performance indicators for composite gas cylinders. Before leaving the factory, gas cylinders must undergo tens of thousands of repeated filling and discharging fatigue tests to verify their reliability within their designed service life. Because fatigue testing with gas as the filler is extremely dangerous, anti-wear hydraulic oil is often used as the filler in actual production. During fatigue testing, the gas cylinder needs to repeatedly cycle between high and normal pressure. During repeated pressurization and depressurization, the surface temperature of the cylinder fluctuates due to fluid pressure circulation and the thermo-coupling effect of the cylinder body. This thermal strain caused by temperature changes is superimposed on the actual strain caused by cylinder fatigue damage, making traditional strain measurement methods unable to effectively distinguish between normal thermal strain and abnormal defect strain.

[0003] In existing technologies, strain gauge compensation and linear temperature coefficient compensation methods are commonly used to eliminate the influence of temperature. However, these methods only consider the temperature-strain relationship at a single measuring point, neglecting the spatial continuity and correlation of the strain field of the gas cylinder as a continuous elastic body. They cannot accurately distinguish between thermal strain caused by temperature changes and abnormal strain caused by local defects, nor can they adapt to differences in temperature or strain characteristics at different locations within the gas cylinder due to structural variations. This results in temperature compensation being unable to cover all locations within the gas cylinder. Furthermore, existing methods typically require manual pre-setting of compensation parameters and recalibration based on different gas cylinder specifications and test conditions. This method suffers from poor adaptability and low compensation accuracy. Consequently, strain data in fatigue testing is biased, making it difficult to develop effective early warning measures and accurately identify manufacturing defects in the gas cylinder. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide an RFID-based defect early warning method, electronic device, computer-readable storage medium and program product, which can improve the low accuracy of strain value compensation for defect-induced strain caused by temperature-induced strain in gas cylinder fatigue testing, and improve the early warning accuracy of gas cylinder production defects.

[0005] To achieve the above technical objectives, the technical solution adopted in this application is as follows:

[0006] In a first aspect, embodiments of this application provide an RFID-based defect early warning method, comprising:

[0007] The pressure value, which characterizes the internal pressure of the gas cylinder, and the temperature and strain values ​​at several detection points on the surface of the gas cylinder are obtained. The temperature and strain values ​​are obtained based on RFID tags with sensing functions arranged in an orthogonal matrix on the surface of the gas cylinder. The pressure value is obtained based on a pressure sensor arranged on the connecting pipe at the gas cylinder mouth.

[0008] Based on the temperature and strain values, the temperature-compensated strain and the first coefficient are obtained. The temperature-compensated strain is used to characterize the strain value of the gas cylinder after excluding the global thermal expansion strain, and the first coefficient is used to characterize the degree of strain change caused by the local unit temperature change.

[0009] Based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected by each RFID tag, a graph structure is constructed. The nodes of the graph structure are the positions mapped to each RFID tag. The node features include the temperature value, strain value, temperature-compensated strain, and the first coefficient. The edges of the graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features include the arc distance between the RFID tags and a distance attenuation factor. The distance attenuation factor is obtained based on the Pearson coefficient of the strain value and is used to characterize the strain transmission attenuation degree of the strain value corresponding to the RFID tag in space.

[0010] The graph structure is input into a deep learning model. Based on the graph structure, the deep learning model obtains a first coupling feature and a temperature-corrected strain value. The first coupling feature is used to characterize the coupling change characteristics of temperature and strain values ​​in spatial relationship. The temperature-corrected strain value is used to characterize the strain of the gas cylinder caused by its own defects.

[0011] When the temperature-corrected strain value of any RFID tag area exceeds a preset strain threshold and the pressure value exceeds a preset pressure threshold, the warning result is determined to indicate that there is a defect risk in that area.

[0012] According to the first aspect, obtaining the temperature-compensated strain and the first coefficient based on the temperature value and the strain value includes:

[0013] Based on the difference between the strain value, temperature value and reference temperature, the temperature compensation strain is obtained through the linear expansion coefficient of the gas cylinder. The reference temperature includes the average surface temperature of the gas cylinder when it is at room temperature and unloaded.

[0014] The first coefficient is obtained based on the temperature value and average temperature within a preset sliding time window, and the strain value and average strain within the preset sliding time window.

[0015] According to the first aspect, the construction of the graph structure based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected by each of the RFID tags includes:

[0016] The temperature value, strain value, temperature-compensated strain, and first coefficient of the node are concatenated to form the initial node features;

[0017] Based on the initial node features, the initial node features are standardized using the mean and standard deviation of the initial node features to obtain the node features.

[0018] According to the first aspect, the construction of the graph structure based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected by each of the RFID tags further includes:

[0019] When the arc distance between any two RFID tags is less than or equal to a preset value, it is determined that there is an adjacency relationship between the two RFID tags. The edges of the graph structure represent the adjacency relationship between the RFID tags. The arc distance is calculated based on the circumferential and axial positions of the RFID tags.

[0020] Based on the arc distance between two adjacent RFID tags, the distance attenuation factor is obtained through the Pearson coefficient of the strain values ​​corresponding to the two RFID tags;

[0021] The edge feature is obtained by concatenating the distance attenuation factor with the arc distance.

[0022] According to the first aspect, the deep learning model includes a graph attention network, wherein inputting the graph structure into the deep learning model, and the deep learning model obtaining a first coupling feature based on the graph structure, includes:

[0023] The graph structure is input into the graph attention network, and the graph attention layer of the graph attention network maps all the node features in the graph structure to a high-dimensional feature space to obtain high-dimensional node features. The dimension of the high-dimensional node features is greater than that of the node features.

[0024] Based on the high-dimensional node features and the distance decay factor in the edge features, the attention score of the node to its neighboring nodes is obtained, wherein the neighboring nodes include all other nodes that have the adjacency relationship with the node.

[0025] The attention scores are normalized to obtain the attention weights of the node to its neighboring nodes.

[0026] Based on the attention weights, the high-dimensional node features of the node and its neighboring nodes are weighted and aggregated to obtain the first coupling feature of the node. The first coupling feature is used to enable the deep learning model to distinguish between strain caused by temperature or strain caused by defects.

[0027] According to the first aspect, obtaining the temperature-corrected strain value based on the first coupling feature includes:

[0028] The first coupling feature is input to the fully connected layer, which processes the first coupling feature into the temperature-corrected strain value. The fully connected layer has the same dimension as the first coupling feature and outputs the first coupling feature as a single-dimensional scalar value.

[0029] According to the first aspect, before obtaining the pressure value, temperature value, and strain value, the method further includes:

[0030] Obtain multiple historical temperature values ​​and historical strain values, as well as historical true temperature corrected strain values ​​corresponding to each of the historical temperature values ​​and historical strain values;

[0031] Based on the historical temperature and historical strain values, the historical temperature compensation strain and the historical first coefficient are obtained.

[0032] Based on the historical temperature values, historical strain values, historical temperature-compensated strain, and historical first coefficient, a historical graph structure is constructed.

[0033] The nodes of the historical graph structure represent the positions mapped to each RFID tag. The node features of the historical graph structure include the historical temperature value, historical strain value, historical temperature-compensated strain, and historical first coefficient. The edges of the historical graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features of the historical graph structure include the arc distance between the RFID tags and the historical distance attenuation factor.

[0034] The historical graph structure and the historical true temperature corrected strain values ​​are integrated into training data;

[0035] A deep learning model is trained and initialized based on the training data, so that after the graph structure is input into the deep learning model, the deep learning model outputs the temperature-corrected strain value.

[0036] Secondly, embodiments of this application also provide an electronic device, the electronic device including a processor and a memory coupled to each other, the memory storing a computer program, and when the computer program is executed by the processor, causing the electronic device to perform the method as described in the first aspect.

[0037] Thirdly, embodiments of this application also provide a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when run on a computer, causes the computer to perform the method described in the first aspect.

[0038] Fourthly, embodiments of this application also provide a program product, characterized in that it includes a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0039] The invention employing the above technical solution has the following advantages: By deploying RFID tags with sensing capabilities on the surface of the gas cylinder, the surface temperature and strain values ​​are collected simultaneously, avoiding the errors caused by the separate collection of temperature and strain values ​​in existing technologies, and improving the timing matching accuracy of temperature and strain values. Based on the temperature and strain values, a temperature-compensated strain and a first coefficient are obtained, and node features and edge features are constructed using these to form a graph structure. Through the spatial aggregation capability of the deep learning model, the features of each node are fused with the features of surrounding neighboring nodes, generating a first coupled feature that simultaneously contains local information and spatial correlation information. This allows the deep learning model to distinguish between temperature-induced strain and defect-induced strain. The deep learning model processes the first coupled feature into a temperature-corrected strain, thereby obtaining an accurate temperature-corrected true defect strain value, improving the accuracy of safety warnings. Attached Figure Description

[0040] This application can be further illustrated by the non-limiting embodiments given in the accompanying drawings. It should be understood that the following drawings only illustrate some embodiments of this application and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained from these drawings without any inventive effort.

[0041] Figure 1 A flowchart of an RFID-based defect early warning method provided in an embodiment of this application.

[0042] Figure 2 This is a schematic diagram of a graph attention network architecture provided in an embodiment of this application. Detailed Implementation

[0043] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts are referred to by the same reference numerals in the drawings or description. Implementations not shown or described in the drawings are forms known to those skilled in the art. In the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0044] Please refer to Figure 1This application provides an RFID-based defect early warning method, which can be applied to electronic devices and whose steps can be executed or implemented by the electronic devices. The electronic devices can be, but are not limited to, personal computers, smartphones, and other electronic devices. The RFID-based defect early warning method may include the following steps:

[0045] S10 acquires the pressure value characterizing the internal pressure of the gas cylinder, as well as the temperature and strain values ​​of several detection points on the surface of the gas cylinder. The temperature and strain values ​​are obtained based on RFID tags with sensing functions arranged in an orthogonal matrix on the surface of the gas cylinder, and the pressure value is obtained based on the pressure sensor arranged on the gas cylinder opening connecting pipe.

[0046] Based on the temperature and strain values, S20 obtains the temperature compensation strain and the first coefficient. The temperature compensation strain is used to characterize the strain value of the gas cylinder after excluding the global thermal expansion strain, and the first coefficient is used to characterize the degree of strain change caused by the local unit temperature change.

[0047] S30 constructs a graph structure based on the temperature and strain values, temperature compensation strain, and a first coefficient collected by each RFID tag. The nodes of the graph structure are the positions mapped to each RFID tag. The node features include the temperature value, strain value, temperature compensation strain, and the first coefficient. The edges of the graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features include the arc distance between the RFID tags and a distance attenuation factor. The distance attenuation factor is obtained based on the Pearson coefficient of the strain value and is used to characterize the strain transmission attenuation degree of the strain value corresponding to the RFID tag in space.

[0048] S40 inputs the graph structure into the deep learning model. The deep learning model obtains a first coupling feature based on the graph structure and obtains a temperature-corrected strain value based on the first coupling feature. The first coupling feature is used to characterize the coupling change characteristics of temperature value and strain value in spatial relationship. The temperature-corrected strain value is used to characterize the strain generated by the gas cylinder due to its own defects.

[0049] S50 When the temperature-corrected strain value of any RFID tag area exceeds a preset strain threshold and the pressure value exceeds a preset pressure threshold, the warning result is determined to indicate that there is a defect risk in that area.

[0050] In the above implementation, RFID tags are arranged in an orthogonal matrix on the surface of the gas cylinder. The surface temperature and strain values ​​at the RFID tag locations are simultaneously collected, and the overall internal pressure of the gas cylinder is obtained through a pressure sensor. Temperature compensation strain and a first coefficient are calculated from the temperature and strain values ​​and used as derived features. Before using the deep learning model, a graph structure needs to be constructed from the input values ​​to meet the data structure requirements of the graph attention layer in the deep learning model. The location mapped to each RFID tag is used as a node in the graph structure, and corresponding node features are constructed. Nodes with adjacency are determined based on the arc distance between nodes less than or equal to a preset value. Edges of the graph structure are constructed between these adjacency nodes, and the edges include combinations of adjacency nodes. Corresponding edge features are constructed. The graph structure containing all nodes, edges, node features, and edge features is input into the deep learning model. The graph attention layer in the deep learning model calculates attention weights based on the node features and the distance attenuation factor in the edge features, which characterizes the strain transmission attenuation of the strain value corresponding to the RFID tag in space. The first coupling feature is obtained through weighted aggregation, and the final temperature-corrected strain value is obtained based on the first coupling feature. The temperature-corrected strain value is the true strain value caused by the defect after excluding temperature strain. The early warning result is obtained based on the temperature-corrected strain value, pressure value, preset strain threshold, and preset pressure threshold.

[0051] In S10, RFID (Radio Frequency Identification) tags are placed on the surface of the gas cylinder under test, and an RFID reader is used to read and write data onto the tags. Temperature changes on the cylinder surface cause a change in resistance in the temperature sensing unit within the RFID tag, resulting in the emission of a continuous radio frequency signal. The RFID reader processes this radio frequency signal to obtain the temperature value at the location of the RFID tag. Similarly, the RFID reader emits a swept-frequency signal; when strain changes occur, the RFID tag generates a backscattered signal. The RFID reader obtains the strain value by comparing the frequency offset between the swept-frequency signal and the backscattered signal. The RFID tags are arranged on the surface of the cylinder using an orthogonal array or other dense arrangement method. The spacing between each RFID tag is determined based on the signal acquisition requirements of the actual RFID reader and the specific filling and discharging process requirements during fatigue testing. In this embodiment, the RFID tags used are passive RFID tags with integrated sensing units. The pressure value is obtained through a pressure sensor placed on the high-pressure connecting pipe section between the gas cylinder opening and the main filling and discharging pipeline of the fatigue testing platform. Because the fluid pressure is the same everywhere in the connected container under steady-state or quasi-steady-state flow conditions, the inside of the gas cylinder and the connecting pipeline form a completely connected closed system, and the pressure inside the pipeline is consistent with the pressure inside the gas cylinder.

[0052] In S20, obtaining the temperature-compensated strain and the first coefficient based on the temperature and strain values ​​specifically includes:

[0053] S21 obtains the temperature compensation strain based on the difference between the strain value, temperature value and reference temperature, and the linear expansion coefficient of the gas cylinder. The reference temperature includes the average surface temperature of the gas cylinder when it is at room temperature and unloaded.

[0054] S22 obtains the first coefficient based on the temperature value and average temperature within a preset sliding time window, and the strain value and average strain within the preset sliding time window.

[0055] In S21, calculate the temperature-compensated strain R:

[0056]

[0057] in, This represents the temperature compensation strain of the i-th RFID tag at time t. This represents the strain value of the i-th RFID tag at time t. This represents the coefficient of thermal expansion of the gas cylinder material; in this embodiment, it refers to the change in length of the gas cylinder material per unit temperature change. This represents the temperature value of the i-th RFID tag at time t. This represents the reference temperature, which is the average surface temperature of the gas cylinder when it is at room temperature and unloaded. The output is a dimensionless value. The value represents strain, which is also dimensionless. The calculated temperature-compensated strain R is also dimensionless.

[0058] At time t, a temperature compensation strain calculation operation is performed on all RFID tags to obtain the temperature compensation strain R(t) of all RFID tags at time t.

[0059] Temperature-compensated strain R is a commonly used strain correction value based on the material's coefficient of thermal expansion in existing technologies. However, current technologies use R for single-point compensation, resulting in low compensation accuracy and an inability to identify early defects. Since plastic-lined fiber-wound composite gas cylinders are anisotropic and non-uniform structures, and temperature-compensated strain R is calculated using a uniform coefficient of thermal expansion, it cannot be adapted to all locations within the cylinder. Furthermore, gas cylinders are continuous elastic bodies, and their strain field exhibits spatial continuity. Temperature-induced strain is smoothly and continuously distributed in space, while defect-induced strain creates a strain gradient between the affected area and its surroundings. Therefore, it is impossible to use information from surrounding RFID tags to correct measurement errors. Using temperature-compensated strain R as a nodal feature initially filters out large-amplitude uniform thermal strain, accelerating model convergence. This allows the model to focus on remaining local defect strain components, improving defect identification sensitivity and reducing training data requirements.

[0060] In S22, the first coefficient K is calculated:

[0061]

[0062] in, Let represent the first coefficient corresponding to the i-th RFID tag at time t, n represent the sampling time within the time window, n=t-T+1, and T represent the length of the time window in the preset sliding time window; This represents the temperature value of the i-th RFID tag at time n. This represents the average temperature value within a preset sliding time window. This represents the strain value of the i-th RFID tag at time n. This represents the average strain value within the time window. This represents the minimum value, and is used when calculating the first coefficient. To prevent the denominator from being 0, it can be 10 in this embodiment. -6 .

[0063] The preset sliding time window is a continuous sliding time window of fixed length, where the time window length is T. When calculating the first coefficient at time t, the time window ends at time t and includes the temperature and strain values ​​at time t and the preceding T consecutive sampling times. As the sampling time progresses from t to t+1, the time window synchronously slides forward by one sampling step, updating to include the data at time t+1 and the preceding T consecutive sampling times, thus realizing the dynamic calculation of the first coefficient.

[0064] The first coefficient K characterizes the actual strain change caused by a unit temperature change at the location corresponding to an RFID tag within a short time window near the current moment. The first coefficient K is input as a node feature into the deep learning model, establishing a physical precondition: the strain change should be approximately equal to the product of the first coefficient and the temperature change; any strain change deviating from this relationship may be caused by a defect. Existing technologies use a uniform coefficient of linear expansion to calculate temperature-compensated strain, ignoring the differences in thermal strain at different locations. The first coefficient K, however, can adaptively adjust the thermal strain compensation amount based on the differences in thermal strain at different locations on the gas cylinder surface. When a local defect exists, the material stiffness decreases, and the strain change does not completely follow the temperature change, causing the first coefficient K to fluctuate significantly. This allows the deep learning model to identify strain that may be caused by a defect at that location.

[0065] In S30, the construction of the graph structure based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected by each RFID tag includes:

[0066] The temperature value, strain value, temperature-compensated strain, and first coefficient of the node are concatenated to form the initial node features; based on the initial node features, the initial node features are standardized using the mean and standard deviation of the initial node features to obtain the node features.

[0067] In S30, for node i corresponding to the i-th RFID tag, the initial node feature X at time t is... i (t) is:

[0068]

[0069] in, This represents the node feature vector of the i-th node at time t. Let represent the temperature value of the i-th node at time t. This represents the strain value of the i-th node at time t. This represents the temperature compensation strain at the i-th node at time t. Let represent the first coefficient corresponding to the i-th node at time t.

[0070] To eliminate the dimensional differences between different physical quantities in the node features, the node features are standardized using Z-score:

[0071]

[0072] in, This represents the k-th feature of the i-th node in the standardized node features. This represents the k-th feature of the i-th node in the initial node features. Let represent the mean of the k-th feature. This represents the standard deviation of the k-th feature. This represents the minimum value when standardizing features in node features. To prevent the denominator from being 0, it can be 10 in this embodiment. -6 , where k represents the dimension number of the node feature. For each feature in the node features, the feature is standardized by using its mean and standard deviation, thereby achieving the standardization of the node features.

[0073] In S30, the construction of the graph structure based on the temperature and strain values, temperature-compensated strain, and the first coefficient collected by each RFID tag further includes:

[0074] S31 When the arc distance between any two RFID tags is less than or equal to a preset value, it is determined that there is an adjacency relationship between the two RFID tags. The edges of the graph structure represent the adjacency relationship between the RFID tags. The arc distance is calculated based on the circumferential and axial positions of the RFID tags.

[0075] S32, based on the arc distance between two adjacent RFID tags, obtains the distance attenuation factor through the Pearson coefficient of the strain values ​​corresponding to the two RFID tags;

[0076] S33 concatenates the distance attenuation factor with the arc distance to obtain the edge feature.

[0077] In S31, let p be the spatial position of the i-th RFID tag on the surface of the gas cylinder. i :

[0078]

[0079] in, This indicates the angular position of the i-th RFID tag along the circumference of the gas cylinder. This indicates the position of the i-th RFID tag along the axial direction of the gas cylinder;

[0080] Because gas cylinders have a cylindrical structure with their ends touching in the circumference, the angle difference between two RFID tags... It can be represented as:

[0081]

[0082] in, This represents the minimum angular difference between the i-th RFID tag and the j-th RFID tag in the circumferential direction. This represents the circumferential angular position of the i-th RFID tag. This represents the circumferential angular position of the j-th RFID tag. The angle difference was calculated. Calculate the supplementary angles of the circle and output the results. The dimension remains plane angle.

[0083] Let the radius of the cylinder's circumference be r, calculate the arc distance d on the cylinder's surface:

[0084]

[0085] in, This represents the arc distance between the i-th RFID tag and the j-th RFID tag. This represents the minimum angular difference between the i-th RFID tag and the j-th RFID tag in the circumferential direction. This indicates the position of the i-th RFID tag along the axial direction of the gas cylinder. This indicates the position of the j-th RFID tag along the axial direction of the gas cylinder. The arc length in the circumferential direction is calculated. The axial distance is calculated. The cylinder surface of the gas cylinder is extended and treated as a plane. The Pythagorean theorem is used to calculate the distance between two points, thus obtaining the arc distance between points i and j. .

[0086] When the arc distance between any two RFID tags is less than or equal to a preset value d0, it is determined that there is an adjacency relationship between the two RFID tags, that is, there is an adjacency relationship between the i-th RFID tag and the j-th RFID tag, with the i-th RFID tag as the center and the j-th RFID tag as the neighboring RFID tag:

[0087]

[0088] in, Let j represent the set of neighboring RFID tags of the i-th RFID tag, and j represent the neighboring RFID tag number. d0 represents the arc distance between the i-th RFID tag and the j-th RFID tag, and d0 represents the preset value.

[0089] In S32, the distance attenuation factor B is calculated. ij :

[0090]

[0091] in, Let M represent the covariance between the strain values ​​of the i-th RFID tag and the j-th RFID tag, and let M represent the total number of sampling times. This represents the strain value of the i-th RFID tag at time t. This represents the average strain value of the i-th RFID tag at all times. This represents the strain value of the j-th RFID tag at time t. This represents the average strain value of the j-th RFID tag at all times.

[0092]

[0093] in, Let M represent the standard deviation of the strain value of the i-th RFID tag, and M represent the total number of sampling times. This represents the strain value of the i-th RFID tag at time t. This represents the average strain value of the i-th RFID tag at all times.

[0094]

[0095] in, The Pearson coefficient, representing the strain values ​​of the i-th and j-th RFID tags, characterizes the correlation between the two strain values. This represents the covariance between the strain values ​​of the i-th RFID tag and the j-th RFID tag. and Let represent the standard deviations of the strain values ​​of the i-th and j-th RFID tags, respectively;

[0096]

[0097] in, This represents the distance attenuation factor between the i-th RFID tag and the j-th RFID tag. The Pearson coefficient represents the strain values ​​of the i-th and j-th RFID tags, and exp() represents the exponential function. d represents the arc distance between the i-th RFID tag and the j-th RFID tag. max This indicates the maximum distance among all RFID tags on the surface of the gas cylinder.

[0098] In S33, for the i-th RFID tag and the j-th RFID tag, the edge feature F at time t is... ij (t) is:

[0099]

[0100] in, This represents the arc distance between the i-th RFID tag and the j-th RFID tag. This represents the distance attenuation factor between the i-th RFID tag and the j-th RFID tag.

[0101] Defects on the surface of gas cylinders typically have a localized effect; that is, a defect may affect the strain state of nearby RFID tag locations without significantly affecting the strain of RFID tag locations at greater distances. The arc distance d of the gas cylinder surface is used. ij Using ≤d0 as the basis for establishing the edge of the graph structure allows the graph structure to focus on expressing the local spatial relationships between RFID tags. This enables the graph attention network to prioritize learning the coupling relationship between temperature and strain between adjacent regions, improving the model's ability to identify local defects. Simultaneously, it reduces the impact of strain noise and irrelevant strain changes at distant RFID tag locations on the prediction of the true strain at the current RFID tag, improving the stability of temperature correction. Secondly, local defects typically cause the temperature compensation strain or strain value of neighboring RFID tags to exhibit local continuous changes, specifically: the strain is greatest in the defect center region and gradually decreases in the surrounding region. The arc distance d on the gas cylinder surface is used... ij Using ≤d0 as the basis for establishing the graph structure enables the model to capture the local expansion trend of defects on the surface of the gas cylinder and helps to locate the defect area.

[0102] In S40, the graph structure includes the nodes, edges, node features, and edge features corresponding to all RFID tags:

[0103]

[0104] Where G(t) represents the graph structure at time t, V represents the set of nodes, each node corresponding to an RFID tag, E represents the set of edges, representing the connection relationships between adjacent RFID tags, X(t) represents the node feature matrix at time t, which is obtained by concatenating the standardized node features of all nodes in order of their corresponding RFID tag IDs, and F(t) represents the edge feature matrix at time t, which is obtained by concatenating the edge features of all nodes in order of their corresponding RFID tag IDs. Taking the node feature matrix as an example:

[0105]

[0106] in, Let n represent the node characteristics of the nth node at time t, where n is the total number of nodes.

[0107] Similarly, the composition of the edge feature matrix F(t) will not be elaborated further.

[0108] In S40, the deep learning model includes a graph attention network. The step of inputting the graph structure into the deep learning model, and the deep learning model obtaining a first coupled feature based on the graph structure, includes:

[0109] S41 The graph structure is input into the graph attention network. The graph attention layer of the graph attention network maps all the node features in the graph structure to a high-dimensional feature space to obtain high-dimensional node features. The dimension of the high-dimensional node features is greater than that of the node features.

[0110] S42 obtains the attention score of the node to its neighboring nodes based on the high-dimensional node features and the distance decay factor in the edge features, wherein the neighboring nodes include all other nodes that have the adjacency relationship with the node.

[0111] S43 normalizes the attention score to obtain the attention weight of the node to its neighboring nodes;

[0112] Based on the attention weight, S44 performs weighted aggregation of the high-dimensional node features of the node and its neighboring nodes to obtain the first coupling feature of the node. The first coupling feature is used to enable the deep learning model to distinguish between strain caused by temperature or strain caused by defects.

[0113] Please refer to Figure 2 In S41, the graph structure is input to the graph input layer of the graph attention network, and the graph input layer uses the node feature matrix in the graph structure as the initial input to the graph attention network.

[0114]

[0115] in, Let represent the initial node feature matrix of the node set at time t in the 0th layer of the graph attention network.

[0116] The initial node feature matrix is ​​input into the graph attention layer. The graph attention layer performs a linear transformation on the node features of all nodes, mapping them to a unified high-dimensional feature space, thus obtaining high-dimensional node features:

[0117]

[0118] in, This represents the high-dimensional node features of the i-th node in the attention layer of the l-th graph after linear mapping. This represents the trainable weight matrix of the attention layer in the l-th graph. Let represent the initial node feature matrix of the node set at time t in the 0th layer of the graph attention network, and l represent the layer number in the graph attention layer.

[0119] In S42, based on the high-dimensional node features and the distance decay factor in the edge features, the attention score of the node to its neighboring nodes is obtained:

[0120]

[0121] in, Let represent the attention score of the i-th node to the j-th node. Let represent the transpose of the trainable single-head attention parameter vector of layer l, ‖ denotes vector concatenation, and LeakyReLU() represents the linear unit activation function with leakage correction. This represents the high-dimensional node features of the i-th node. This represents the high-dimensional node features of the j-th node. This represents the distance decay factor in the edge features between the i-th node and the j-th node.

[0122] In another possible embodiment, the attention score of a node to its neighboring nodes is:

[0123]

[0124] The difference between this embodiment and the previous embodiment is that the previous embodiment lacks the distance decay factor in the edge features between the i-th node and the j-th node for attention calculation. It calculates attention solely based on the feature similarity between the two nodes, without considering the spatial physical laws of the gas cylinder as a continuous elastic body. The deep learning model needs to learn from scratch the importance of neighboring node j to the current node i. The model will assign higher weights to nodes that are farther away, leading to a significant deviation between the model output and reality. In the application scenario of this application, this will render temperature compensation meaningless, misjudging normal thermal strain as a defect.

[0125] In this embodiment, the attention score is used to determine the importance of the strain value of neighboring node j to the strain value of the current node i. In this embodiment, when d ij Increase, that is, when the arc distance between the neighboring node and the current node increases, the distance decay factor B between them increases. ij A decrease indicates a weaker correlation between the strain values ​​of the two nodes. When the strain value of the i-th node is transmitted to the j-th node, the strain value is smaller, and conversely, the neighboring nodes have a smaller influence on the strain value of the current node. This characterizes the physical property that the strain field on the gas cylinder surface gradually decays with increasing spatial distance. (The last part, "B," appears to be a typo and can be left as is.) ij Involving attention score calculations makes the neighborhood relationships learned by the graph attention network more consistent with the actual spatial impact of local defects on gas cylinders. Simultaneously, it prioritizes neighboring nodes that are spatially closer and physically more connected to the current node, improving the accuracy of local first coupling feature extraction. Secondly, since local defects on the gas cylinder surface primarily affect the area near the defect, it prioritizes distorting the strain values ​​corresponding to RFID tags in the defect-prone area, rather than uniformly affecting the entire gas cylinder. Therefore, a distance attenuation factor B is used. ijCalculate the attention score, which makes it easier to allocate attention to local abnormal strain values ​​and enhances the model's ability to identify abnormal strains caused by local defects.

[0126] In S43, the attention score is normalized to obtain the attention weight of the node to its neighboring nodes:

[0127]

[0128]

[0129]

[0130] in, This represents the normalized attention weight of the i-th node to the j-th node. N represents the attention score of the i-th node to the j-th node, k represents the index of the current node i in the node set, and N represents the attention score of the i-th node to the j-th node. i Represents a set of nodes. This represents the set of neighboring nodes.

[0131] In S44, based on the attention weight, the high-dimensional node features of the node and its neighboring nodes are weighted and aggregated to obtain the first coupling feature:

[0132]

[0133] in ReLU() represents the node feature matrix output by the i-th node in the (l+1)-th layer, and ReLU() represents the activation function. This represents the normalized attention weight of the i-th node to the j-th node. This represents the high-dimensional node features of the j-th node. This represents the trainable bias term of the l-th layer.

[0134] This output serves as the input to the next graph attention layer. In this embodiment, the graph attention network has L layers, and the final output is the first coupling feature of the i-th node. .

[0135] The first coupling feature characterizes the temperature and strain of the current node and its neighboring nodes, as well as their spatial relationship and the resulting strain transmission correlation. In existing technologies, temperature or strain sensors can only characterize the state of a single sampling point. The first coupling feature, however, integrates information from both the current and neighboring nodes, avoiding isolated judgments based solely on the temperature or strain of a single sampling point, thus aligning with the actual physical state of the gas cylinder and the spatial characteristics of strain transmission. Secondly, during fatigue testing, overall temperature changes in the gas cylinder lead to similar strain changes in adjacent RFID tags, but local defects can cause an abnormal deviation of a node relative to its neighbors. The first coupling feature compares the temperature and strain states of the current node and its neighbors, helping the model distinguish between temperature strain caused by overall thermal expansion and contraction and abnormal strain caused by local defects. The first coupling feature improves the accuracy of deep learning models in predicting strain caused by defects, increasing the confidence level of the final warning result.

[0136] In S40, obtaining the temperature-corrected strain value based on the first coupling feature includes: inputting the first coupling feature to the fully connected layer, wherein the fully connected layer processes the first coupling feature into the temperature-corrected strain value.

[0137]

[0138] in, This represents the temperature-corrected strain value at the i-th node. This represents a trainable weight vector. This indicates a trainable bias term.

[0139] In S50, the temperature-corrected strain value of each node is traversed. When the temperature-corrected strain value of any node exceeds a preset strain threshold, it indicates that the strain value at the location of the RFID tag corresponding to that node, after excluding temperature-induced strain, exceeds the strain threshold, suggesting a potential defect. Furthermore, if the overall internal pressure of the gas cylinder exceeds a preset pressure threshold, the warning result confirms a defect risk at that location. It should be noted that using a preset pressure threshold as one of the warning conditions is to avoid false alarms triggered by strain fluctuations under low-pressure, non-hazardous conditions, ensuring that warnings only occur when the internal pressure of the gas cylinder is sufficient to cause defect expansion.

[0140] In actual fatigue testing of gas cylinders, since the anti-wear hydraulic oil used as the filler in the cylinder fatigue test is almost incompressible, the synchronization between pressure and strain is good. Preset pressure thresholds and preset strain thresholds are used simultaneously as early warning conditions to adapt to actual fatigue testing conditions. High pressure inside the gas cylinder is the direct cause of defect propagation and eventual rupture. Therefore, at higher pressure levels, manufacturing defects in the cylinder, such as cracks in metal fittings, will produce significant strain responses and strain concentration. For flexible defects in the cylinder, such as micro-cracks in the plastic liner, they will crack under high pressure with increasing fatigue cycles, and the strain will abnormally disappear or weaken after depressurization. Therefore, using both preset pressure and preset strain thresholds as early warning conditions improves the accuracy of the warning. In one possible implementation, when the location of an RFID tag reaches the early warning condition, a defect risk is determined, and an early warning can be issued and the warning location marked. Based on the experimental data at the time of the warning, the experimenters can further determine the cause of the cylinder defect and adjust the production process. Simultaneously, when the warning occurs, the cylinder is still in a non-severe rupture state, and the experimenters do not need to work under the dangerous condition of excessively high pressure and high strain, ensuring the safety of the experimenters.

[0141] In S40, the deep learning model mentioned can be trained in the following way:

[0142] Multiple historical temperature values ​​and historical strain values ​​are obtained, along with a corresponding historical true temperature-corrected strain value for each historical temperature value and historical strain value. Based on the historical temperature values ​​and historical strain values, historical temperature-compensated strain and a historical first coefficient are obtained. A historical graph structure is constructed based on the historical temperature values, historical strain values, historical temperature-compensated strain, and historical first coefficient. The nodes of the historical graph structure represent the positions mapped to each RFID tag. The node features of the historical graph structure include the historical temperature value, historical strain value, historical temperature-compensated strain, and historical first coefficient. The edges of the historical graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features of the historical graph structure include the arc distance between the RFID tags and a historical distance attenuation factor. The historical graph structure and the historical true temperature-corrected strain values ​​are integrated into training data. An initialized deep learning model is trained based on the training data so that after the graph structure is input into the deep learning model, the deep learning model outputs the temperature-corrected strain value. In this embodiment, the historical true temperature-corrected strain value can be obtained through strain experiments under isothermal conditions.

[0143] The loss function during training can be expressed as:

[0144]

[0145] This represents the total loss value;

[0146] Represents the predicted value of a deep learning model Corresponding historical true temperature corrected strain value The mean square error;

[0147] A training dataset is constructed using historical graph structures as input features and historical temperature-corrected strain values ​​as supervisory signals. After initializing the deep learning model, a loss function is used for training, enabling the model to learn the relationship between temperature and strain while calculating the actual strain value excluding temperature-induced strain. After training, in the testing phase, only the graph structure needs to be input, and the model can directly output the temperature-corrected strain value. In this embodiment, the collected training dataset is divided into training and testing sets in an 8:2 ratio, with 80% of the data used for model training and the remaining 20% ​​used for testing.

[0148] This application provides an electronic device that may include a processing module and a memory. The memory stores a computer program, which, when executed by the processor, enables the electronic device to perform the corresponding steps in the RFID-based defect early warning method described above.

[0149] In this embodiment, the processor can be an integrated circuit chip with signal processing capabilities. For example, the processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0150] The memory can be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, etc.

[0151] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic device described above can be referred to the corresponding steps in the aforementioned method, and will not be elaborated further here.

[0152] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to execute the RFID-based defect warning method as described in the above embodiments.

[0153] Computer-readable storage media may be magnetic disks, optical disks, read-only memory, random access memory, flash memory, USB flash drives, hard disks, or solid-state drives, etc., and may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implement the methods shown in the above embodiments.

[0154] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the RFID-based defect early warning method described above. The computer program product may exist in a computer-readable storage medium in forms including, but not limited to, source files, executable files, and installation package files.

[0155] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, electronic device, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0156] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0157] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A defect early warning method based on RFID, characterized in that, The method includes: The pressure value, which characterizes the internal pressure of the gas cylinder, and the temperature and strain values ​​at several detection points on the surface of the gas cylinder are obtained. The temperature and strain values ​​are obtained based on RFID tags with sensing functions arranged in an orthogonal matrix on the surface of the gas cylinder. The pressure value is obtained based on a pressure sensor arranged on the connecting pipe at the gas cylinder mouth. Based on the temperature and strain values, the temperature-compensated strain and the first coefficient are obtained. The temperature-compensated strain is used to characterize the strain value of the gas cylinder after excluding the global thermal expansion strain, and the first coefficient is used to characterize the degree of strain change caused by the local unit temperature change. Based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected by each RFID tag, a graph structure is constructed. The nodes of the graph structure are the positions mapped to each RFID tag. The node features include the temperature value, strain value, temperature-compensated strain, and the first coefficient. The edges of the graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features include the arc distance between the RFID tags and a distance attenuation factor. The distance attenuation factor is obtained based on the Pearson coefficient of the strain value and is used to characterize the strain transmission attenuation degree of the strain value corresponding to the RFID tag in space. The graph structure is input into a deep learning model. Based on the graph structure, the deep learning model obtains a first coupling feature and a temperature-corrected strain value. The first coupling feature is used to characterize the coupling change characteristics of temperature and strain values ​​in spatial relationship. The temperature-corrected strain value is used to characterize the strain of the gas cylinder caused by its own defects. When the temperature-corrected strain value of any RFID tag area exceeds a preset strain threshold and the pressure value exceeds a preset pressure threshold, the warning result is determined to indicate that there is a defect risk in that area.

2. The method according to claim 1, characterized in that, The process of obtaining the temperature-compensated strain and the first coefficient based on the temperature and strain values ​​includes: Based on the difference between the strain value, temperature value and reference temperature, the temperature compensation strain is obtained through the linear expansion coefficient of the gas cylinder. The reference temperature includes the average surface temperature of the gas cylinder when it is at room temperature and unloaded. The first coefficient is obtained based on the temperature value and average temperature within a preset sliding time window, and the strain value and average strain within the preset sliding time window.

3. The method according to claim 1, characterized in that, The construction of the graph structure based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected from each RFID tag includes: The temperature value, strain value, temperature-compensated strain, and first coefficient of the node are concatenated to form the initial node features; Based on the initial node features, the initial node features are standardized using the mean and standard deviation of the initial node features to obtain the node features.

4. The method according to claim 3, characterized in that, The step of constructing a graph structure based on the temperature and strain values, temperature-compensated strain, and a first coefficient collected from each RFID tag also includes: When the arc distance between any two RFID tags is less than or equal to a preset value, it is determined that there is an adjacency relationship between the two RFID tags. The edges of the graph structure represent the adjacency relationship between the RFID tags. The arc distance is calculated based on the circumferential and axial positions of the RFID tags. Based on the arc distance between two adjacent RFID tags, the distance attenuation factor is obtained through the Pearson coefficient of the strain values ​​corresponding to the two RFID tags; The edge feature is obtained by concatenating the distance attenuation factor with the arc distance.

5. The method according to claim 1, characterized in that, The deep learning model includes a graph attention network. The graph structure is input into the deep learning model, and the deep learning model obtains a first coupling feature based on the graph structure, including: The graph structure is input into the graph attention network, and the graph attention layer of the graph attention network maps all the node features in the graph structure to a high-dimensional feature space to obtain high-dimensional node features. The dimension of the high-dimensional node features is greater than that of the node features. Based on the high-dimensional node features and the distance decay factor in the edge features, the attention score of the node to its neighboring nodes is obtained, wherein the neighboring nodes include all other nodes that have the adjacency relationship with the node. The attention scores are normalized to obtain the attention weights of the node to its neighboring nodes. Based on the attention weights, the high-dimensional node features of the node and its neighboring nodes are weighted and aggregated to obtain the first coupling feature of the node. The first coupling feature is used to enable the deep learning model to distinguish between strain caused by temperature or strain caused by defects.

6. The method according to claim 1, characterized in that, The process of obtaining the temperature-corrected strain value based on the first coupling feature includes: The first coupling feature is input to the fully connected layer, which processes the first coupling feature into the temperature-corrected strain value. The fully connected layer has the same dimension as the first coupling feature and outputs the first coupling feature as a single-dimensional scalar value.

7. The method according to claim 1, characterized in that, Before obtaining the pressure, temperature, and strain values, the method further includes: Obtain multiple historical temperature values ​​and historical strain values, as well as historical true temperature corrected strain values ​​corresponding to each of the historical temperature values ​​and historical strain values; Based on the historical temperature and historical strain values, the historical temperature compensation strain and the historical first coefficient are obtained. Based on the historical temperature values, historical strain values, historical temperature-compensated strain, and historical first coefficient, a historical graph structure is constructed. The nodes of the historical graph structure represent the positions mapped to each RFID tag. The node features of the historical graph structure include the historical temperature value, historical strain value, historical temperature-compensated strain, and historical first coefficient. The edges of the historical graph structure represent the adjacency relationship between any two RFID tags whose arc distance on the gas cylinder surface is less than or equal to a preset value. The edge features of the historical graph structure include the arc distance between the RFID tags and the historical distance attenuation factor. The historical graph structure and the historical true temperature corrected strain values ​​are integrated into training data; A deep learning model is trained and initialized based on the training data, so that after the graph structure is input into the deep learning model, the deep learning model outputs the temperature-corrected strain value.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory coupled together, the memory storing a computer program that, when executed by the processor, causes the electronic device to perform the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 7.

10. A program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images