Reliability real-time monitoring and failure early warning method and device of connection system and vehicle

Abstract

The invention discloses a reliability real-time monitoring and failure early warning method and device for a connection system, a medium and equipment. The method comprises the steps that the initial axial load of a target connection piece is obtained; acquiring a current time-frequency graph of the target connecting piece; determining the current axial load of the target connecting piece according to the current time-frequency diagram; determining an axial load attenuation rate of the current axial load relative to the initial axial load; and determining the reliability of the target connecting piece according to the size of the axial load attenuation rate. According to the invention, real-time monitoring of the reliability of the target connector can be realized according to the current time-frequency diagram.

Description

Technical Field

[0001] This application relates to the field of connector technology, and in particular to a method, apparatus, and vehicle for real-time reliability monitoring and failure early warning of connection systems. Background Technology

[0002] Connectors, including bolts, nuts, or rivets, are mainly used for mechanical fastening between various components and systems, such as engine cylinder heads, crankshaft connecting rod mechanisms, flywheel mechanisms, and power battery housing assemblies, thereby ensuring the normal operation of the system. However, under long-term operation or vibration environments, the reliability of connectors may decrease. When the decrease rate is too large, problems such as connection failure may occur, leading to serious malfunctions.

[0003] Currently, special bolts or dedicated equipment are typically used to measure the axial load of connectors. However, this method is costly and time-consuming, and is generally used for timely or post-failure measurements, making it impossible to monitor the reliability of connectors in real time. Furthermore, it is mainly used for measuring the axial load of the assembly and cannot be used for measuring the residual axial load of the connector, thus making it impossible to quantitatively judge the reliability of the connector. In addition, when the installation location of the connector is narrow and concealed, testing is difficult or direct measurement is not possible. Summary of the Invention

[0004] The embodiments of this application provide a method, apparatus, vehicle, medium, and equipment for real-time reliability monitoring and failure early warning of a connection system, which solves the technical problem that existing connection reliability measurement methods cannot achieve real-time monitoring of connection reliability.

[0005] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0006] According to a first aspect of this application, a method for real-time reliability monitoring and failure early warning of a connection system is provided, the connection system including at least one connector for fastening connection, the method comprising: Obtain the initial axial load on the target connector; Obtain the current time-frequency map of the target connector, wherein the current time-frequency map is the spectrum corresponding to the detection signal of the current length of the target connector; The current axial load of the target connector is determined based on the current time-frequency diagram; Determine the axial load attenuation rate of the current axial load relative to the initial axial load; The reliability of the target connector is determined based on the magnitude of the axial load attenuation rate, and the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

[0007] In some embodiments of this application, based on the foregoing scheme, determining the current axial load of the target connector according to the current time-frequency diagram includes: The current time-frequency map is input into the first prediction model, and the current axial load of the target connector is output. The first prediction model is obtained by training a deep learning network through multiple first data samples. The first data samples include: the first time-frequency map and the first axial load label corresponding to the first time-frequency map. The first time-frequency map is the time-frequency map of the target time when the connector is fastened. The target time is after the initial time.

[0008] In some embodiments of this application, based on the foregoing scheme, obtaining the initial axial load of the target connector includes: Obtain the initial time-frequency diagram of the target connector; The initial time-frequency map is input into the second prediction model, which outputs the initial axial load of the target connector. The second prediction model is obtained by training a deep learning network with multiple second data samples. The second data samples include: the second time-frequency map and the second axial load label corresponding to the second time-frequency map. The second time-frequency map is the time-frequency map of the initial moment when the connector is fastened.

[0009] In some embodiments of this application, based on the foregoing scheme, the first prediction model or the second prediction model includes a feature extraction unit, a difference calculation unit, and a metric learning unit. The training method of the first prediction model or the second prediction model is as follows: A training dataset is obtained, comprising a support set, a query set, and a baseline time-frequency plot. The support set and query set include multiple data samples, each including a time-frequency plot and corresponding axial load labels. The baseline time-frequency plot is the time-frequency plot when the axial load of the connector is zero. When training the first prediction model, the time-frequency plot is the first time-frequency plot and the axial load label is the first axial load label. When training the second prediction model, the time-frequency plot is the second time-frequency plot and the axial load label is the second axial load label. The training dataset is input into the feature extraction unit to obtain a baseline feature vector and at least one target feature vector. The target feature vector is the feature vector of each time-frequency graph in the support set or the query set, and the baseline feature vector is the feature vector of the baseline time-frequency graph. For each target feature vector, the difference calculation unit calculates the target vector difference between the target feature vector and the reference feature vector; The first prediction model or the second prediction model is trained by inputting multiple target vector differences and corresponding axial load labels into the metric learning unit and learning the mapping relationship between the vector differences and the axial load magnitude.

[0010] In some embodiments of this application, based on the foregoing scheme, the step of inputting the training dataset into the feature extraction unit to obtain the target feature vector and the baseline feature vector includes: Based on the time-frequency diagram, a first mode time-frequency diagram, a second mode time-frequency diagram, and a third mode time-frequency diagram are obtained. The first mode time-frequency diagram is the direct wave time-frequency diagram corresponding to the time-frequency diagram. The second mode time-frequency diagram is the resonant wave time-frequency diagram corresponding to the time-frequency diagram. The third mode time-frequency diagram is the low-frequency wave time-frequency diagram corresponding to the time-frequency diagram. The first modal time-frequency graph is input into the feature extraction unit, and the first modal feature vector is output. The second modal time-frequency graph is input into the feature extraction unit, and the second modal feature vector is output. The third modal time-frequency graph is input into the feature extraction unit, and the third modal feature vector is output. The reference time-frequency graph is input into the feature extraction unit, and the reference feature vector is output. The target feature vector is obtained by weighted summation of the first modality feature vector, the second modality feature vector, and the third modality feature vector.

[0011] In some embodiments of this application, based on the foregoing scheme, calculating the target vector difference between the target feature vector and the reference feature vector through the difference calculation unit includes: Obtain the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector; The target vector difference is obtained by weighted fusion of the absolute difference, the relative difference, and the phase difference.

[0012] In some embodiments of this application, based on the foregoing scheme, the weighted fusion of the absolute difference, the relative difference, and the phase difference to obtain the target vector difference includes: The first weighting coefficients for the absolute difference, the relative difference, and the phase difference are determined using a multilayer perceptron model. Normalize the three first weighting coefficients to obtain second weighting coefficients for the absolute difference, the relative difference, and the phase difference; The target vector difference is obtained by weighting and summing the absolute difference, the relative difference, and the phase difference based on the three second weighting coefficients.

[0013] In some embodiments of this application, based on the foregoing scheme, determining the reliability of the target connector according to the magnitude of the axial load attenuation rate includes: Based on a preset mapping relationship, a target reliability threshold corresponding to the target connector is determined, and the target reliability threshold includes a first threshold and a second threshold. If the axial load attenuation rate is less than or equal to the first threshold, the connector is in a safe state. If the axial load attenuation rate is greater than the first threshold and less than the second threshold, the connector is in a warning state. If the axial load attenuation rate is greater than or equal to the second threshold, the connector is in an emergency alarm state.

[0014] According to a second aspect of this application, a real-time reliability monitoring and failure early warning device for a connection system is provided, the connection system including at least one connector for fastening the connection, the device comprising: The first acquisition unit acquires the initial axial load of the target connector; The second acquisition unit acquires the current time-frequency map of the target connector, wherein the current time-frequency map is a spectrum corresponding to the detection signal of the current length of the target connector; The first determining unit determines the current axial load of the target connector based on the current time-frequency diagram; The second determining unit determines the axial load attenuation rate of the current axial load relative to the initial axial load; The third determining unit determines the reliability of the target connector based on the magnitude of the axial load attenuation rate, wherein the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

[0015] In some embodiments of this application, based on the foregoing scheme, the first determining unit is configured as follows: The first output unit inputs the current time-frequency map into the first prediction model and outputs the current axial load of the target connector. The first prediction model is obtained by training a deep learning network with multiple first data samples. The first data samples include: a first time-frequency map and a first axial load label corresponding to the first time-frequency map. The first time-frequency map is the time-frequency map of the target time when the connector is fastened. The target time is after the initial time.

[0016] In some embodiments of this application, based on the foregoing scheme, the first acquisition unit is configured as follows: The second acquisition unit acquires the initial time-frequency diagram of the target connector; The second output unit inputs the initial time-frequency map into the second prediction model and outputs the initial axial load of the target connector. The second prediction model is obtained by training a deep learning network with multiple second data samples. The second data samples include: the second time-frequency map and the second axial load label corresponding to the second time-frequency map. The second time-frequency map is the time-frequency map of the initial moment when the connector is fastened.

[0017] In some embodiments of this application, based on the foregoing scheme, the first prediction model or the second prediction model includes a feature extraction unit, a difference calculation unit, and a metric learning unit. The apparatus further includes a first training unit, which is used to train the first prediction model or the second prediction model. The first training unit is configured as follows: The third acquisition unit acquires a training dataset, which includes a support set, a query set, and a baseline time-frequency map. The support set and query set include multiple data samples, and each data sample includes a time-frequency map and an axial load label corresponding to the time-frequency map. The baseline time-frequency map is the time-frequency map when the axial load of the connector is zero. When training the first prediction model, the time-frequency map is the first time-frequency map and the axial load label is the first axial load label. When training the second prediction model, the time-frequency map is the second time-frequency map and the axial load label is the second axial load label. The first obtaining unit inputs the training dataset into the feature extraction unit to obtain a baseline feature vector and at least one target feature vector. The target feature vector is the feature vector of each time-frequency graph in the support set or the query set, and the baseline feature vector is the feature vector of the baseline time-frequency graph. The first calculation unit calculates the target vector difference between the target feature vector and the reference feature vector for each target feature vector through the difference calculation unit; The second training unit inputs multiple target vector differences and corresponding axial load labels into the metric learning unit, and trains the first prediction model or the second prediction model by learning the mapping relationship between the vector differences and the axial load magnitude.

[0018] In some embodiments of this application, based on the foregoing scheme, the first obtaining unit is configured as follows: The fourth determining unit obtains a first mode time-frequency diagram, a second mode time-frequency diagram, and a third mode time-frequency diagram based on the time-frequency diagram. The first mode time-frequency diagram is the direct wave time-frequency diagram corresponding to the time-frequency diagram, the second mode time-frequency diagram is the resonant wave time-frequency diagram corresponding to the time-frequency diagram, and the third mode time-frequency diagram is the low-frequency wave time-frequency diagram corresponding to the time-frequency diagram. The third output unit inputs the first modal time-frequency graph into the feature extraction unit and outputs the first modal feature vector; inputs the second modal time-frequency graph into the feature extraction unit and outputs the second modal feature vector; inputs the third modal time-frequency graph into the feature extraction unit and outputs the third modal feature vector; and inputs the reference time-frequency graph into the feature extraction unit and outputs the reference feature vector. The second obtaining unit performs a weighted summation of the first modal feature vector, the second modal feature vector, and the third modal feature vector to obtain the target feature vector.

[0019] In some embodiments of this application, based on the foregoing scheme, the first calculation unit includes a second calculation unit, which is used to calculate the target vector difference between the target feature vector and the reference feature vector through the difference calculation unit. The second calculation unit is configured as follows: The fourth acquisition unit acquires the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector; The third obtaining unit performs weighted fusion of the absolute difference, the relative difference, and the phase difference to obtain the target vector difference.

[0020] In some embodiments of this application, based on the foregoing scheme, the third obtaining unit is configured as follows: The fifth determining unit determines the first weighting coefficients of the absolute difference, the relative difference, and the phase difference using a multilayer perceptron model; The fourth unit normalizes the three first weighting coefficients to obtain second weighting coefficients for the absolute difference, the relative difference, and the phase difference. The fifth obtaining unit calculates the target vector difference by weighting and summing the absolute difference, the relative difference, and the phase difference based on the three second weighting coefficients.

[0021] In some embodiments of this application, based on the foregoing scheme, the third determining unit is configured as follows: The sixth determining unit determines a target reliability threshold corresponding to the target connector based on a preset mapping relationship. The target reliability threshold includes a first threshold and a second threshold. If the axial load attenuation rate is less than or equal to the first threshold, the connector is in a safe state. The second judgment unit determines that if the axial load attenuation rate is greater than the first threshold and less than the second threshold, the connector is in a warning state. If the axial load attenuation rate is greater than or equal to the second threshold, the connector is in an emergency alarm state.

[0022] According to a third aspect of this application, a vehicle is provided, including a plurality of coupling systems, each coupling system including at least one of the connectors configured to securely connect the subsystems, characterized in that it further includes: a processor for performing the method described in any embodiment of the first aspect of this application.

[0023] According to a fourth aspect of this application, a computer-readable storage medium is provided that stores a computer program thereon, the computer program including executable instructions that, when executed by a processor, implement the method described in any embodiment of the first aspect of this application.

[0024] According to a fifth aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing executable instructions of the processors, which, when executed by the one or more processors, cause the one or more processors to implement the method described in any embodiment of the first aspect of this application.

[0025] The beneficial effects of this application are as follows: The current time-frequency diagram corresponds to the current length of the target connector. The current length of the target connector can be determined based on the current time-frequency diagram, thereby determining the axial load attenuation rate of the current axial load relative to the initial axial load. The reliability of the target connector can be determined by the axial load attenuation rate. In other words, the reliability of the target connector can be monitored in real time based on the current time-frequency diagram.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 A flowchart of a method for real-time reliability monitoring and failure early warning of a connection system according to an embodiment of this application is shown; Figure 2 A front view of the vehicle's flywheel mechanism is shown; Figure 3 A side view of the vehicle's flywheel mechanism is shown; Figure 4 A schematic diagram of the vehicle's crank-connecting rod mechanism is shown; Figure 5A schematic diagram of the vehicle's body profile and liquid cooling plate is shown; Figure 6 A schematic diagram of the vehicle's engine cylinder head is shown; Figure 7 A block diagram of the real-time reliability monitoring and failure early warning device for the connection system in an embodiment of this application is shown; Figure 8 A schematic diagram of a computer-readable storage medium in an embodiment of this application is shown; Figure 9 A schematic diagram of the system structure of an electronic device in an embodiment of this application is shown. Detailed Implementation

[0028] Figure 1 A flowchart of the real-time reliability monitoring and failure early warning method for the connection system in an embodiment of this application is shown. See [link to relevant documentation]. Figure 1 A method for real-time reliability monitoring and failure early warning of a connection system is provided. The connection system includes at least one connector for fastening the connection. The method includes steps S1 to S5, which are described in detail below: In step S1, the initial axial load of the target connector is obtained. The initial axial load is the axial load of the target connector in its initial connection state. For example, when the target connector is a bolt, the initial axial load is the axial load when the bolt is just tightened.

[0029] In step S2, the current time-frequency diagram of the target connector is obtained. The current time-frequency diagram is a spectrum corresponding to the detection signal of the current length of the target connector. The current time-frequency diagram is the time-frequency diagram of the target connector in its current connection state, and the time-frequency diagram is a spectrum of frequency relative to time.

[0030] In step S3, the current axial load of the target connector is determined based on the current time-frequency diagram. The current axial load is the axial load of the target connector in its current connection state. For example, if the target connector is a bolt, the current axial load is the axial load when the bolt is tightened one revolution.

[0031] In step S4, the axial load attenuation rate of the current axial load relative to the initial axial load is determined.

[0032] In some embodiments, determining the axial load attenuation rate of the current axial load relative to the initial axial load includes: obtaining the axial load difference between the initial axial load and the current axial load, and using the ratio of the axial load difference to the initial axial load as the axial load attenuation rate.

[0033] In some implementations, the formula for calculating the axial load attenuation rate is as follows: , The axial load attenuation rate, This is the initial axial load. This represents the current axial load.

[0034] In step S5, the reliability of the target connector is determined based on the magnitude of the axial load attenuation rate, and the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

[0035] In some implementations, determining the current axial load of the target connector based on the current time-frequency map includes: inputting the current time-frequency map into a first prediction model and outputting the current axial load of the target connector. The first prediction model is obtained by training a deep learning network with multiple first data samples. The first data samples include: a first time-frequency map and a first axial load label corresponding to the first time-frequency map. The first time-frequency map is a time-frequency map of the target time when the connector is fastened, and the target time is after the initial time.

[0036] In some embodiments, obtaining the initial axial load of the target connector includes: obtaining an initial time-frequency map of the target connector; inputting the initial time-frequency map into a second prediction model and outputting the initial axial load of the target connector, wherein the second prediction model is obtained by training a deep learning network with multiple second data samples, and the second data samples include: a second time-frequency map and a second axial load label corresponding to the second time-frequency map, wherein the second time-frequency map is the time-frequency map of the initial moment when the connector is fastened.

[0037] In some implementations, the first prediction model or the second prediction model includes a feature extraction unit, a difference calculation unit, and a metric learning unit. The training method for the first prediction model or the second prediction model is as follows: A training dataset is obtained, comprising a support set, a query set, and a baseline time-frequency map. The support set and query set include multiple data samples, each data sample including a time-frequency map and corresponding axial load labels. The baseline time-frequency map is the time-frequency map when the axial load of the connector is zero. When training the first prediction model, the time-frequency map is the first time-frequency map and the axial load label is the first axial load label. When training the second prediction model, the time-frequency map... The second time-frequency map is defined as follows: the axial load label is defined as the second axial load label; the training dataset is input into the feature extraction unit to obtain a baseline feature vector and at least one target feature vector, wherein the target feature vector is the feature vector of each time-frequency map in the support set or the query set, and the baseline feature vector is the feature vector of the baseline time-frequency map; for each target feature vector, the difference calculation unit calculates the target vector difference between the target feature vector and the baseline feature vector; multiple target vector differences and corresponding axial load labels are input into the metric learning unit, and the first prediction model or the second prediction model is trained by learning the mapping relationship between the vector difference and the axial load magnitude.

[0038] In some implementations, the metric learning unit optimizes the feature space through a contrastive loss function, such that similar feature vectors are closer together in the space, while feature vectors of different categories are farther apart.

[0039] In some implementations, the contrastive loss function is calculated using the following formula: , For axial force label, To set boundaries, The difference between the first target vectors. The difference between the second target vectors. The Euclidean distance is the difference between the two target vectors.

[0040] In some implementations, the metric learning unit includes a classifier. The step of training the first prediction model or the second prediction model by learning the mapping relationship between vector differences and axial load magnitude includes: the metric learning unit aggregating multiple target vector differences to obtain a global feature vector; and inputting the global feature vector into the classifier for classification prediction to obtain the mapping relationship between vector differences and axial load magnitude. Common aggregation methods include max pooling, average pooling, or weighted summation, with the aggregation formula as follows: The classifier is typically a fully connected layer that outputs class probabilities. The loss function in this application is the cross-entropy loss function, as shown in the following formula: , For the number of categories, For real labels, This represents the predicted class probability. Common optimizers such as the Adam optimizer can be used to optimize the model's parameters.

[0041] It should be noted that the time-frequency graphs in the training dataset can be processed with a uniform size, such as H*M. The first prediction model can be understood as a deep learning model for the reliability decay of the connector connection, and the second deep learning model can be understood as a deep learning model for the reliability of the connector connection.

[0042] In some embodiments, the first time-frequency diagram, the second time-frequency diagram, and the reference time-frequency diagram are obtained through the following steps: obtaining a preset upper limit value of the axial load of the calibration connector; with the axial load of the calibration connector at zero, measuring the calibration connector by ultrasonic waves to obtain a reference waveform diagram, and converting the reference waveform diagram into the reference time-frequency diagram; with the axial load of the calibration connector greater than or equal to the preset upper limit value of the axial load, measuring the calibration connector by ultrasonic waves to obtain a waveform diagram before the experiment, and converting the waveform diagram before the experiment into the second time-frequency diagram; and performing ultrasonic testing on the calibration connector... After reliability testing of the calibrated connector, ultrasonic measurement is performed to obtain a waveform diagram. This waveform diagram is then converted into a second time-frequency diagram. The calibrated connector is of the same type and operating environment as the target connector. The ultrasonic measurement of the calibrated connector includes at least one ultrasonic transceiver process. This process involves: emitting ultrasonic waves from a first end to a second end of the calibrated connector; the ultrasonic waves propagate from inside the calibrated connector to the second end; the waves are reflected by the second end of the calibrated connector; and the reflected ultrasonic waves are received by the calibrated connector. An ultrasonic sensor can be used to measure the calibrated connector. This ultrasonic sensor can be a piezoelectric sensor with an ultrasonic self-transmitting and self-receiving mode.

[0043] For example, Figure 2 A front view of the vehicle's flywheel mechanism is shown. Figure 3 A side view of the vehicle's flywheel mechanism is shown. Figure 2 and Figure 3 In the diagram, 1 represents the flywheel mechanism, 2 represents the first bolt, and 3 represents the mounting position of the ultrasonic sensor. (See attached diagram.) Figure 2 and Figure 3 When the connecting part is the first bolt 2 of the flywheel mechanism 1, the ultrasonic sensor is located at Figure 2 The front end face of the first bolt 2, that is Figure 3 The left end face of the first bolt 2 in the middle, Figure 4 A schematic diagram of the vehicle's crank-connecting rod mechanism is shown. Figure 4 In the diagram, 1 represents the flywheel mechanism, 4 represents the crank-connecting rod mechanism, 5 represents the ultrasonic sensor mounting position, and 6 represents the second bolt. (See also...) Figure 4 When the connecting part is the second bolt 6 of the crank-connecting rod mechanism 4, the ultrasonic sensor is located at Figure 4 The lower end face of the second bolt 6 in the middle, Figure 5 A schematic diagram of the vehicle's body profile and liquid cooling plate is shown. Figure 5 In the diagram, 7 represents the installation location of the ultrasonic sensor, 8 represents the housing, 9 represents the cold-rolled plate, and 10 represents the sealing rivet nut or pop rivet. (See also...) Figure 5 When the connecting component is a sealing rivet nut or blind rivet 10 connecting the housing 8 and the cold plate 9, the sealing rivet nut or blind rivet 10 can fix the liquid cooling plate and ensure the airtightness of the housing, thereby guaranteeing the safety performance of the power battery. The ultrasonic sensor is located at... Figure 5 The front end face, Figure 6 A schematic diagram of the vehicle's engine cylinder head is shown. Figure 6 In the diagram, 11 represents the cylinder head, 12 represents the cylinder block, 13 represents the cylinder head bolts, and 14 represents the ultrasonic sensor mounting location. (See attached diagram.) Figure 6 When the connecting component is cylinder head bolt 13, the ultrasonic sensor is located at... Figure 6 The front face.

[0044] It should be noted that the reference waveform or the reference time-frequency diagram can be used to determine the reference length of the calibration connector. The fact that the calibration connector and the target connector are of the same type indicates that their connection axial load, material yield strength, yield point connection axial load, and friction coefficient are the same. The fact that the calibration connector and the target connector have the same working environment indicates that their end face parallelism, surface roughness, material properties, mechanical properties, clamping length, and surface flatness are the same, thereby simulating the real object as much as possible. In order to simulate the real object as much as possible, the calibration connector is usually modified, such as grinding the end of the calibration connector to make the end face parallelism of the calibration connector 0.02 and the surface roughness Ra 0.8-1.6mm. The test gasket and nut are also processed to make the material properties, mechanical properties, clamping length, and surface flatness the same as the target connector.

[0045] In some embodiments, the method further includes: obtaining a first axial load on the calibration connector when the axial load of the calibration connector is greater than or equal to the preset upper limit value of the axial load, and using the first axial load as a first axial load label on the first time-frequency graph; after performing a reliability test on the calibration connector, obtaining a second axial load on the calibration connector, and using the second axial load as a second axial load label on the second time-frequency graph. The first axial load and the second axial load of the calibration connector can be obtained using existing methods.

[0046] In some embodiments, obtaining the first axial load of the calibration connector includes: obtaining a first length and a reference length of the calibration connector, using the difference between the first length and the reference length as a first deformation, and determining the first axial load corresponding to the first deformation based on a target mapping relationship.

[0047] In some embodiments, obtaining the second axial load of the calibrated connector includes: obtaining a second length and a reference length of the calibrated connector; using the difference between the second length and the reference length as a second deformation; and determining the second axial load corresponding to the second deformation based on a target mapping relationship. The second axial load can be understood as a residual axial load.

[0048] In some implementations, the target mapping relationship is determined through the following steps: during the tightening of the calibration connector, multiple sets of tightening data for the calibration connector are acquired, each set of tightening data including axial load and deformation; based on the multiple sets of tightening data, a target mapping relationship between the axial load and deformation is determined. The target mapping relationship can be an axial load-connector deformation signal relationship curve, or it can be formed into a calibration file and stored in the device, thereby determining the first axial load through the first length, i.e., the length before the reliability test.

[0049] It should be noted that during the tightening of the calibration connector, the axial load of the calibration connector changes from zero to a value greater than the preset upper limit of the axial load. The preset upper limit of the axial load can be understood as the upper limit of the designed axial load range. The calibration connector can be tightened by a tightening machine.

[0050] In some implementations, the reliability test is to conduct various tests on an engine bench under different operating conditions, such as idle speed at 1000 rpm, partial load at 1500 rpm to 3500 rpm, rated speed at 5500 rpm, and external characteristic conditions.

[0051] In some implementations, inputting the training dataset into the feature extraction unit to obtain the target feature vector and the reference feature vector includes: inputting the time-frequency map into the feature extraction unit and outputting the target feature vector, and inputting the reference time-frequency map into the feature extraction unit and outputting the reference feature vector.

[0052] In some implementations, the step of inputting the time-frequency image into the feature extraction unit and outputting the target feature vector includes: performing multi-scale feature fusion calculation on the time-frequency image to obtain fused features, performing dimensionality reduction processing on the fused features, and outputting the target feature vector.

[0053] For example, the time-frequency graph is X (X∈R) 4 / H×4 / W×Cin (where Cin is the number of input channels), and the feature map Fa is obtained by processing through the first residual block. The feature map Fa is processed by the second residual block to obtain Fb. The feature map Fb is processed by the third residual block to obtain Fc. The bilinear interpolation upsampling (UpSample) operation is used to adjust Fb by a factor of 2 and Fc by a factor of 4 to the same size as Fa, and then concatenate them along the channel dimension to obtain the final multi-scale fused feature F. final , in, , For Ffinal Dimensionality reduction is performed to obtain the target feature vector.

[0054] In some implementations, inputting the training dataset into the feature extraction unit to obtain the target feature vector and the reference feature vector includes: obtaining a first mode time-frequency map, a second mode time-frequency map, and a third mode time-frequency map based on the time-frequency map, wherein the first mode time-frequency map is the straight wave time-frequency map corresponding to the time-frequency map, the second mode time-frequency map is the resonance wave time-frequency map corresponding to the time-frequency map, and the third mode time-frequency map is the low-frequency wave time-frequency map corresponding to the time-frequency map; inputting the first mode time-frequency map into the feature extraction unit to output the first mode feature vector, inputting the second mode time-frequency map into the feature extraction unit to output the second mode feature vector, inputting the third mode time-frequency map into the feature extraction unit to output the third mode feature vector, and inputting the reference time-frequency map into the feature extraction unit to output the reference feature vector; and weighting and summing the first mode feature vector, the second mode feature vector, and the third mode feature vector to obtain the target feature vector.

[0055] In some implementations, the target feature vector is calculated using the following formula: , For the target feature vector, The first mode feature vector, This is the second modality feature vector. This is the third mode feature vector. The weighting coefficients for fusion can be set according to the actual situation, and this application does not impose any restrictions.

[0056] In some embodiments, obtaining the first mode time-frequency diagram, the second mode time-frequency diagram, and the third mode time-frequency diagram based on the time-frequency diagram includes: acquiring the time-domain waveform diagram corresponding to the time-frequency diagram; determining the target arrival time of the straight wave based on the time-domain waveform diagram; extracting the waveform from the starting time to the first time from the time-domain waveform diagram, performing an 8-12MHz zero-phase FIR bandpass filter on the extracted waveform, and converting the filtered waveform into a first mode waveform diagram, wherein the first time is after the target time; extracting the waveform from the second time to the third time from the time-domain waveform diagram, and performing a 3-7MHz Chebyshev I filter on the extracted waveform. A bandpass filter is used to convert the filtered waveform into a second-mode time-frequency diagram. The second time point is located between the target time point and the first time point, and the third time point is located after the first time point. The waveform from the fourth time point to the fifth time point is extracted from the time-domain waveform diagram. The extracted waveform is subjected to a 0.5-2MHz Butterworth low-pass filter to obtain the maximum amplitude value in the filtered waveform. In the filtered waveform, the waveform with an amplitude of 0.5 times the maximum amplitude value is identified, and the wave packet with the longest duration and highest energy is retained. The wave packet is converted into a third-mode time-frequency diagram, and the fourth time point is located after the third time point.

[0057] For example, the target time is t A The first time point is t A +1.0μs, the second time is t A +0.5μs, the third time is t A +15μs, the fourth time is t A +20μs, the fifth time point is 50μs.

[0058] In some implementations, determining the target arrival time of the straight wave based on the time-domain waveform diagram includes: performing an 8-12MHz bandpass filter on the time-domain waveform diagram, calculating the Hilbert envelope of the filtered waveform, obtaining the maximum value of the Hilbert envelope, and taking the first peak time point exceeding the target threshold as the target time, wherein the target threshold is 0.3 times the maximum value of the Hilbert envelope.

[0059] In some implementations, calculating the target vector difference between the target feature vector and the reference feature vector using the difference calculation unit includes: obtaining the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector; and performing weighted fusion on the absolute difference, the relative difference, and the phase difference to obtain the target vector difference.

[0060] In some implementations, obtaining the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector includes: taking the absolute value of the difference between the target feature vector and the reference feature vector as the absolute difference; taking the ratio of the absolute difference to the absolute value of the target feature vector as the relative difference; obtaining the vector product of the target feature vector and the reference feature vector, and the product of the magnitude of the target feature vector and the magnitude of the reference feature vector; obtaining the target ratio of the vector product to the magnitude product, determining the target angle corresponding to the target ratio using an inverse cosine function, and taking the ratio of the target angle to π as the phase difference.

[0061] In some implementations, the weighted fusion of the absolute difference, the relative difference, and the phase difference to obtain the target vector difference includes: determining first weight coefficients for the absolute difference, the relative difference, and the phase difference using a multilayer perceptron model; normalizing the three first weight coefficients to obtain second weight coefficients for the absolute difference, the relative difference, and the phase difference; and weighted summing the absolute difference, the relative difference, and the phase difference based on the three second weight coefficients to obtain the target vector difference.

[0062] In some implementations, the first weighting coefficient is calculated using the following formula: , The first weighting coefficient for the absolute difference. The first weighting coefficient for the relative difference. This is the first weighting coefficient for the phase difference.

[0063] In some implementations, the second weighting coefficient is calculated using the following formula: , The second weighting coefficient for the absolute difference. The second weighting coefficient for the relative difference. This is the second weighting coefficient for the phase difference.

[0064] In some implementations, the formula for calculating the target vector difference is as follows: , The second weighting coefficient for the absolute difference. The second weighting coefficient for the relative difference. The second weighting coefficient for the phase difference. For element-wise multiplication, The absolute difference The relative difference This represents the phase difference.

[0065] In some implementations, determining the reliability of the target connector based on the magnitude of the axial load attenuation rate includes: determining a target reliability threshold corresponding to the target connector based on a preset mapping relationship, wherein the target reliability threshold includes a first threshold and a second threshold; if the axial load attenuation rate is less than or equal to the first threshold, the connector is in a safe state; if the axial load attenuation rate is greater than the first threshold and less than the second threshold, the connector is in a warning state; if the axial load attenuation rate is greater than or equal to the second threshold, the connector is in an emergency alarm state.

[0066] It is understandable that during long-term operation, the connector and the connected parts are continuously subjected to alternating loads, which may cause creep thinning. Alternatively, the microscopic unevenness of the connector head, nut, or the surface of the connected parts may be gradually flattened under the axial load, resulting in a decrease in the elongation of the connector. Temperature changes may also cause the connector and the connected parts to have different coefficients of thermal expansion, leading to changes in the axial load. All of these factors can cause the axial load of the connector to decrease to varying degrees. According to the design requirements of the connector, the residual axial load attenuation rate of the connector after the reliability test must meet certain standards to ensure the reliability and stability of the system connection.

[0067] For example, taking the engine crankshaft connecting rod mechanism as an example, the key connecting components are the main bearing bolts and connecting rod big end bolts. The main bearing bolts are primarily used to fasten the main bearing cap, lower main bearing shell, crankshaft, upper main bearing shell, and cylinder block, ensuring that the main bearing cap does not loosen during engine operation and facilitating the creation of an appropriate and uniform lubricating oil film clearance between the main bearing bore and the crankshaft main bearing diameter, preventing wear of the main bearing shell and thus ensuring the reliability and durability of the crankshaft during high-speed operation. The connecting rod bolts are primarily used to fasten the connecting rod big end, lower connecting rod shell, crankshaft crankpin, upper connecting rod shell, and connecting rod body, providing support and positioning for the crankshaft crankpin. During engine operation, if the axial load on the connecting rod bolts decreases to a certain level, insufficient fastening force at the connecting rod big end can lead to loosening, causing connection failure, or even serious malfunctions such as engine damage. Therefore, when the connection reliability decay rate... If the connection rate is ≤30%, it is considered a safe state, and the connection system can continue to operate normally. However, the early warning system will still periodically monitor the connection reliability to ensure continued security. When 30% < If the connection reliability is less than 40%, it is considered a warning state. The warning system will activate a connection reliability fault warning, indicating the need for real-time monitoring and possible maintenance of connection performance to prevent further deterioration of connection reliability. If the connection failure rate is ≥40%, it is considered an emergency alarm state. The system will immediately activate the connection performance failure alarm and take effective protective measures.

[0068] In some embodiments, when the connecting component is used in a flywheel mechanism, the key connecting component is the flywheel bolt, mainly used for the fastening connection between the flywheel and the rear end face of the crankshaft, ensuring a stable output of engine torque. In actual operation, microscopic unevenness such as roughness on the flywheel surface may cause compression of the microscopic area of ​​the flywheel bolt after assembly, leading to bolt loosening. Furthermore, because the flywheel bolt is subjected to significant alternating impact loads, creep occurs between the flywheel bolt and flywheel materials during long-term operation, further attenuating the axial load on the flywheel bolt, causing connection failure, or even serious malfunctions such as engine damage. Therefore, the first threshold is 20%, and the second threshold is 25%.

[0069] In some embodiments, the connecting component is used for the engine cylinder head, with the cylinder head bolts being the key connecting component. The cylinder head bolts are primarily used for the fastening connection of the engine cylinder head, cylinder head gasket, and cylinder block. Tightening them generates a uniform and appropriate axial tightening force on the cylinder head gasket, sealing the high-temperature, high-pressure gas in the cylinder, while also sealing the coolant and lubricating oil. In actual operation, on the one hand, the cylinder head bolts may loosen due to the compression of the cylinder head gasket corrugations, resulting in a decrease in axial load. On the other hand, after prolonged engine operation, the cylinder head gasket's corrugated compression-rebound capability decreases, leading to permanent deformation, which also causes a decrease in the axial load of the cylinder head bolts. When the axial force of the cylinder head bolts decreases to a certain extent, it will lead to sealing failure, resulting in the engine's "three leaks" problem, affecting engine reliability and even causing engine damage. Therefore, the first threshold is 30%, and the second threshold is 40%.

[0070] In some embodiments, when the connecting component is used for a torsional vibration damper, the key connecting component is the torsional vibration damper fixing bolt, mainly due to the fastening between the torsional vibration damper at the front of the engine, the crankshaft signal disc, the timing chain drive wheel, the oil pump drive wheel, and the crankshaft. On the one hand, because the torsional vibration damper fixing bolt connection system involves multiple parts and mating surfaces, in actual operation, the micro-uneven areas of each mating surface are compressed after assembly due to roughness, flatness, and other micro-unevennesses, leading to axial load attenuation of the torsional vibration damper fixing bolt. On the other hand, because the torsional vibration damper fixing bolt is subjected to a large alternating impact load, creep occurs between the torsional vibration damper fixing bolt and the connected parts during long-term operation, resulting in axial load attenuation, thereby causing connection failure or even engine damage. Therefore, the first threshold is 10%, and the second threshold is 20%.

[0071] It should be noted that critical connectors can be understood as connectors that are of primary concern. In some cases, we can focus only on the reliability of critical connectors, thereby reducing the computational workload of reliability assessment.

[0072] In some implementations, the output terminal of the second predictive model is used to directly issue warnings to on-site personnel. By constructing a closed loop of "data preparation-analysis-early warning-decision", the traditional "on-time maintenance" or "post-failure maintenance" mode is transformed into a real-time monitoring "predictive maintenance" mode, which significantly improves the operational safety of major equipment, reduces unplanned downtime, and lowers maintenance costs. It is a deep application of industrial internet and intelligent operation and maintenance in the field of key mechanical structures, and has extremely high economic value and strategic significance.

[0073] This application proposes a dual-intelligent model architecture (a first prediction model and a second prediction model) to accurately decouple the physical characteristics of the initial and attenuation states of the connector. The dual models avoid the prediction bias of a single model for mixed inputs, improve prediction accuracy, monitor the reliability of the connection system in real time, and issue early warnings before failure occurs. It also classifies failure risk states into categories, including safety, early warning, and emergency alarm, thereby achieving predictive maintenance and avoiding catastrophic accidents and economic losses. Furthermore, a difference calculation unit is proposed, which innovatively integrates the absolute difference, relative difference, and phase difference between feature vectors. A comprehensive difference is generated through dynamic weighting, and the comprehensive difference is input into the metric learning module, improving the robustness of signal change detection of the connector.

[0074] Figure 7 A block diagram of a real-time reliability monitoring and failure early warning device for a connection system according to an embodiment of this application is shown. The connection system includes at least one connector for fastening the connection. See [link to relevant documentation]. Figure 7 According to a second aspect of this application, a real-time reliability monitoring and failure early warning device 100 for a connection system is provided, the device comprising: The first acquisition unit 101 acquires the initial axial load of the target connector; The second acquisition unit 102 acquires the current time-frequency map of the target connector, wherein the current time-frequency map is a spectrum corresponding to the detection signal of the current length of the target connector; The first determining unit 103 determines the current axial load of the target connector based on the current time-frequency diagram; The second determining unit 104 determines the axial load attenuation rate of the current axial load relative to the initial axial load; The third determining unit 105 determines the reliability of the target connector based on the magnitude of the axial load attenuation rate, wherein the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

[0075] In some implementations, the first determining unit is configured as: a first output unit, which inputs the current time-frequency map into a first prediction model and outputs the current axial load of the target connector. The first prediction model is obtained by training a deep learning network with multiple first data samples. The first data samples include: a first time-frequency map and a first axial load label corresponding to the first time-frequency map. The first time-frequency map is the time-frequency map of the target time when the connector is fastened. The target time is after the initial time.

[0076] In some embodiments, the first acquisition unit is configured as follows: a second acquisition unit, which acquires the initial time-frequency map of the target connector; and a second output unit, which inputs the initial time-frequency map into a second prediction model and outputs the initial axial load of the target connector. The second prediction model is obtained by training a deep learning network with multiple second data samples. The second data samples include: the second time-frequency map and the second axial load label corresponding to the second time-frequency map. The second time-frequency map is the time-frequency map of the initial moment when the connector is fastened.

[0077] In some embodiments, the first prediction model or the second prediction model includes a feature extraction unit, a difference calculation unit, and a metric learning unit. The device further includes a first training unit, which is used to train the first prediction model or the second prediction model. The first training unit is configured as follows: a third acquisition unit, which acquires a training dataset, the training dataset including a support set, a query set, and a baseline time-frequency map. The support set and the query set include multiple data samples, the data samples including a time-frequency map and an axial load label corresponding to the time-frequency map, the baseline time-frequency map being the time-frequency map when the axial load of the connector is zero. When training the first prediction model, the time-frequency map is the first time-frequency map and the axial load label is the first axial load label. When training the second prediction model... When the prediction model is trained, the time-frequency map is the second time-frequency map and the axial load label is the second axial load label; the first obtaining unit inputs the training dataset into the feature extraction unit to obtain a baseline feature vector and at least one target feature vector, wherein the target feature vector is the feature vector of each time-frequency map in the support set or the query set, and the baseline feature vector is the feature vector of the baseline time-frequency map; the first calculation unit calculates the target vector difference between the target feature vector and the baseline feature vector for each target feature vector through the difference calculation unit; the second training unit inputs multiple target vector differences and corresponding axial load labels into the metric learning unit, and trains the first prediction model or the second prediction model by learning the mapping relationship between the vector difference and the axial load magnitude.

[0078] In some embodiments, the first obtaining unit is configured as follows: a fourth determining unit, which obtains a first modal time-frequency map, a second modal time-frequency map, and a third modal time-frequency map based on the time-frequency map, wherein the first modal time-frequency map is a straight wave time-frequency map corresponding to the time-frequency map, the second modal time-frequency map is a resonant wave time-frequency map corresponding to the time-frequency map, and the third modal time-frequency map is a low-frequency wave time-frequency map corresponding to the time-frequency map; a third output unit, which inputs the first modal time-frequency map into the feature extraction unit and outputs a first modal feature vector, inputs the second modal time-frequency map into the feature extraction unit and outputs a second modal feature vector, inputs the third modal time-frequency map into the feature extraction unit and outputs a third modal feature vector, and inputs the reference time-frequency map into the feature extraction unit and outputs a reference feature vector; and a second obtaining unit, which performs a weighted summation of the first modal feature vector, the second modal feature vector, and the third modal feature vector to obtain the target feature vector.

[0079] In some embodiments, the first calculation unit includes a second calculation unit, which is used to calculate the target vector difference between the target feature vector and the reference feature vector through the difference calculation unit. The second calculation unit is configured as follows: a fourth acquisition unit, which acquires the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector; and a third obtaining unit, which performs weighted fusion on the absolute difference, the relative difference, and the phase difference to obtain the target vector difference.

[0080] In some implementations, the third obtaining unit is configured as follows: a fifth determining unit, which determines first weighting coefficients for the absolute difference, the relative difference, and the phase difference using a multilayer perceptron model; a fourth obtaining unit, which normalizes the three first weighting coefficients to obtain second weighting coefficients for the absolute difference, the relative difference, and the phase difference; and a fifth obtaining unit, which performs a weighted summation of the absolute difference, the relative difference, and the phase difference based on the three second weighting coefficients to obtain the target vector difference.

[0081] In some implementations, the third determining unit is configured as follows: a sixth determining unit, which determines a target reliability threshold corresponding to the target connector based on a preset mapping relationship, the target reliability threshold including a first threshold and a second threshold; a first judging unit, which determines that if the axial load attenuation rate is less than or equal to the first threshold, the connector is in a safe state; and a second judging unit, which determines that if the axial load attenuation rate is greater than the first threshold and less than the second threshold, the connector is in a warning state. If the axial load attenuation rate is greater than or equal to the second threshold, the connector is in an emergency alarm state.

[0082] According to a third aspect of this application, a vehicle is provided, including a plurality of coupling systems, each coupling system including at least one of the connectors configured to securely connect the subsystems, characterized in that it further includes: a processor for performing the method described in any embodiment of the first aspect of this application.

[0083] Based on the same inventive concept, as a fourth aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program including executable instructions that, when executed by a processor, implement the method described in any embodiment of the first aspect of this application.

[0084] In some possible implementations, various aspects of this application may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this application.

[0085] refer to Figure 8 As shown, a program product 200 for implementing the above-described method according to an embodiment of this application is described. The program product 200 of this application can be in the form of a portable compact disc read-only memory (CD-ROM), containing program code, and can run on terminal devices such as personal computers. Its readable medium is divided into two categories, which can be used individually or in combination. The readable storage medium can be a tangible medium such as an electrical, magnetic, or optical medium, specifically including electrical connections with wires, portable discs, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, CD-ROM, optical / magnetic storage devices, etc. The readable signal medium is a propagation signal carrying program code, such as electromagnetic signals in baseband or carrier waves, optical signals, etc., which can be used to send and propagate the program. The program code can be transmitted wirelessly, wired, optically, or via RF, and supports programming languages ​​such as Java, C++, and C. It can be executed in user devices, remote devices, or a combination of both. In remote scenarios, remote computing devices can connect to user devices via local area networks (LANs), wide area networks (WANs), or the Internet.

[0086] See Figure 9 This application also provides an electronic device 300 for implementing the aforementioned method. This device is merely an example and does not limit the function or scope of use of this application. Its core structure and characteristics are as follows: Core components: These exist in the form of a general-purpose computing device, including at least one processing unit 310, at least one storage unit 320, and a bus 330 connecting the components. The program code stored in the storage unit 320 can be executed by the processing unit 310 to complete the aforementioned method steps.

[0087] Storage unit composition: includes volatile storage units (such as random access memory unit RAM321, cache storage unit 322), read-only memory unit ROM323, and also includes a program / utility 324 containing program module 325. Program module 325 covers operating system, application program, other program modules and program data.

[0088] Bus and Communication: Bus 330 can adopt various structures such as storage unit bus and peripheral bus; the device can communicate with external devices 400 (such as keyboard, Bluetooth device) through input / output (I / O) interface 350, and connect to LAN, WAN, Internet and other networks through network adapter 360.

[0089] Expandability: It can be used in conjunction with microcode, device drivers, redundant processing units, and other hardware and software modules not shown. The above descriptions are merely embodiments of this application and are not intended to limit the 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 the claims of this application.

[0090] The above description is merely an embodiment of this application and is not intended to limit 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 the claims of this application.

Claims

1. A method for real-time reliability monitoring and failure early warning of a connection system, wherein the connection system includes at least one connector for fastening the connection, characterized in that, The method includes: Obtain the initial axial load on the target connector; Obtain the current time-frequency map of the target connector, wherein the current time-frequency map is the spectrum corresponding to the detection signal of the current length of the target connector; The current axial load of the target connector is determined based on the current time-frequency diagram; Determine the axial load attenuation rate of the current axial load relative to the initial axial load; The reliability of the target connector is determined based on the magnitude of the axial load attenuation rate, and the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

2. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 1, characterized in that, Determining the current axial load of the target connector based on the current time-frequency diagram includes: The current time-frequency map is input into the first prediction model, and the current axial load of the target connector is output. The first prediction model is obtained by training a deep learning network through multiple first data samples. The first data samples include: the first time-frequency map and the first axial load label corresponding to the first time-frequency map. The first time-frequency map is the time-frequency map of the target time when the connector is fastened. The target time is after the initial time.

3. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 2, characterized in that, The process of obtaining the initial axial load of the target connector includes: Obtain the initial time-frequency diagram of the target connector; The initial time-frequency map is input into the second prediction model, which outputs the initial axial load of the target connector. The second prediction model is obtained by training a deep learning network with multiple second data samples. The second data samples include: the second time-frequency map and the second axial load label corresponding to the second time-frequency map. The second time-frequency map is the time-frequency map of the initial moment when the connector is fastened.

4. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 3, characterized in that, The first prediction model or the second prediction model includes a feature extraction unit, a difference calculation unit, and a metric learning unit. The training method for the first prediction model or the second prediction model is as follows: A training dataset is obtained, comprising a support set, a query set, and a baseline time-frequency plot. The support set and query set include multiple data samples, each including a time-frequency plot and corresponding axial load labels. The baseline time-frequency plot is the time-frequency plot when the axial load of the connector is zero. When training the first prediction model, the time-frequency plot is the first time-frequency plot and the axial load label is the first axial load label. When training the second prediction model, the time-frequency plot is the second time-frequency plot and the axial load label is the second axial load label. The training dataset is input into the feature extraction unit to obtain a baseline feature vector and at least one target feature vector. The target feature vector is the feature vector of each time-frequency graph in the support set or the query set, and the baseline feature vector is the feature vector of the baseline time-frequency graph. For each target feature vector, the difference calculation unit calculates the target vector difference between the target feature vector and the reference feature vector; The first prediction model or the second prediction model is trained by inputting multiple target vector differences and corresponding axial load labels into the metric learning unit and learning the mapping relationship between the vector differences and the axial load magnitude.

5. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 4, characterized in that, The step of inputting the training dataset into the feature extraction unit to obtain the target feature vector and the baseline feature vector includes: Based on the time-frequency diagram, a first mode time-frequency diagram, a second mode time-frequency diagram, and a third mode time-frequency diagram are obtained. The first mode time-frequency diagram is the direct wave time-frequency diagram corresponding to the time-frequency diagram. The second mode time-frequency diagram is the resonant wave time-frequency diagram corresponding to the time-frequency diagram. The third mode time-frequency diagram is the low-frequency wave time-frequency diagram corresponding to the time-frequency diagram. The first modal time-frequency graph is input into the feature extraction unit, and the first modal feature vector is output. The second modal time-frequency graph is input into the feature extraction unit, and the second modal feature vector is output. The third modal time-frequency graph is input into the feature extraction unit, and the third modal feature vector is output. The reference time-frequency graph is input into the feature extraction unit, and the reference feature vector is output. The target feature vector is obtained by weighted summation of the first modality feature vector, the second modality feature vector, and the third modality feature vector.

6. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 4, characterized in that, The step of calculating the target vector difference between the target feature vector and the reference feature vector through the difference calculation unit includes: Obtain the absolute difference, relative difference, and phase difference between the target feature vector and the reference feature vector; The target vector difference is obtained by weighted fusion of the absolute difference, the relative difference, and the phase difference.

7. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 6, characterized in that, The step of weighted fusing the absolute difference, the relative difference, and the phase difference to obtain the target vector difference includes: The first weighting coefficients for the absolute difference, the relative difference, and the phase difference are determined using a multilayer perceptron model. Normalize the three first weighting coefficients to obtain second weighting coefficients for the absolute difference, the relative difference, and the phase difference; The target vector difference is obtained by weighting and summing the absolute difference, the relative difference, and the phase difference based on the three second weighting coefficients.

8. The method for real-time reliability monitoring and failure early warning of a connection system according to claim 1, characterized in that, Determining the reliability of the target connector based on the magnitude of the axial load attenuation rate includes: Based on a preset mapping relationship, a target reliability threshold corresponding to the target connector is determined, and the target reliability threshold includes a first threshold and a second threshold. If the axial load attenuation rate is less than or equal to the first threshold, the connector is in a safe state. If the axial load attenuation rate is greater than the first threshold and less than the second threshold, the connector is in a warning state. If the axial load attenuation rate is greater than or equal to the second threshold, the connector is in an emergency alarm state.

9. A real-time reliability monitoring and failure early warning device for a connection system, wherein the connection system includes at least one connector for fastening the connection, characterized in that, The device includes: The first acquisition unit acquires the initial axial load of the target connector; The second acquisition unit acquires the current time-frequency map of the target connector, wherein the current time-frequency map is a spectrum corresponding to the detection signal of the current length of the target connector; The first determining unit determines the current axial load of the target connector based on the current time-frequency diagram; The second determining unit determines the axial load attenuation rate of the current axial load relative to the initial axial load; The third determining unit determines the reliability of the target connector based on the magnitude of the axial load attenuation rate, wherein the reliability of the target connector is negatively correlated with the magnitude of the axial load attenuation rate.

10. A vehicle comprising a plurality of coupling systems, the coupling system including at least one of the connectors configured to securely connect the subsystems, characterized in that, Also includes: A processor for performing the method according to any one of claims 1-8.

Images