A pressure vessel pressure resistance detection device

Abstract

The application relates to the technical field of pressure container detection, and discloses a pressure container pressure resistance detection device, wherein the pressure container pressure resistance detection device comprises a detection frame, a container support saddle, a pressurizing assembly, a strain data acquisition assembly and a processing unit. The processing unit comprises a coordinate transformation and anisotropy index calculation module, an anisotropy anomaly detection and classification module, a directionality space clustering module, a normalization and cross-detection tracking module, a group expansion basic model training module, an individualized Bayesian updating module, a forward extrapolation and early warning module and a report generation module, which are used for performing debonding defect identification and expansion trend prediction on a composite material pressure container.

Description

Technical Field

[0001] This invention relates to the field of pressure vessel testing technology, and more specifically, to a pressure vessel pressure resistance testing device. Background Technology

[0002] In industrial settings such as hydrogen refueling stations and chemical plants, multiple composite material pressure vessels manufactured in the same batch need to undergo periodic pressure tests to assess their structural safety. During the pressure test, a triaxial strain rosette array is typically deployed on the outer surface of the vessel, and multi-directional strain time-series data is collected under stepped pressure loading. By analyzing strain field anomalies, deterioration and defect areas inside the vessel can be identified.

[0003] Existing pressure testing devices perform amplitude comparison or unidirectional nonlinear fitting on the collected strain data to determine weak areas, and independently perform trend analysis on the test results of a single vessel.

[0004] However, the aforementioned existing technologies have the following drawbacks. First, the strain change rate along the fiber direction and perpendicular to the fiber direction in the debonding region exhibits differentiated nonlinear characteristics during pressurization. Anomaly detection methods based on strain amplitude or slope in a single direction cannot characterize this anisotropic response mode. Furthermore, the strain anomaly amplitude caused by debonding overlaps with the normal interlayer stress redistribution phenomenon in amplitude range, leading to a high false alarm rate in weak area identification. Second, the debonding region exhibits an elliptical evolution characteristic that preferentially expands along the fiber direction, and its boundary parameters constitute multidimensional structured information. However, existing devices only output scalar degradation indicators, which cannot characterize this directional expansion characteristic. Third, a single vessel typically has only three to five periodic inspection records, and the data points are insufficient to independently fit a debonding expansion prediction model. Existing devices fail to utilize group data from the same batch to assist in individual debonding expansion prediction. Summary of the Invention

[0005] This invention provides a pressure vessel pressure resistance testing device, which solves the technical problems of high false alarm rate in interlayer debonding identification, difficulty in characterizing debonding expansion morphology, and insufficient debonding prediction data in related technologies.

[0006] This invention discloses a pressure vessel pressure resistance testing device, comprising a testing frame, a pair of vessel support saddles, a pressurization assembly, a strain data acquisition assembly, and a processing unit. The testing frame serves as the base for the entire device. The vessel support saddles are fixedly connected to the testing frame and are spaced apart along the axial direction of the pressure vessel under test. The pressurization assembly, mounted on the testing frame, includes a booster pump, pressure pipeline, proportional pressure regulating valve, and pressure sensor, used to apply a stepped pressure increase to the inner cavity of the pressure vessel under test. The strain data acquisition assembly includes a triaxial strain rosette array adhered to the outer surface of the pressure vessel under test, a signal conditioning module, and a data acquisition unit. The processing unit is data-connected to the data acquisition unit and includes a coordinate transformation and anisotropy index calculation module, an anisotropy anomaly detection and classification module, a directional spatial clustering module, a normalization and cross-detection tracking module, a population expansion basic model training module, and a personalized Bayesian update module.

[0007] Furthermore, the processing unit also includes a forward extrapolation and early warning module. This module extrapolates the cumulative number of cycles corresponding to the next detection cycle on the posterior personalized extension trajectory, predicts the predicted values ​​and uncertainty ranges of the major axis length, minor axis length, and severity of each debonding region, and compares the upper bound of the predicted elliptical area uncertainty with the maximum allowable debonding area threshold of the layup. For debonding regions where the upper bound of the elliptical area uncertainty exceeds the maximum allowable debonding area threshold of the layup, an accelerated detection early warning mark is generated.

[0008] Furthermore, the processing unit is also connected to the proportional pressure regulating valve spool via a control signal connection. A closed-loop pressure control circuit is formed between the processing unit and the pressure sensor via a data acquisition unit. The processing unit outputs a stepped pressure increase command to the proportional pressure regulating valve spool, which adjusts the pipeline pressure. The pressure sensor collects the real-time pipeline pressure value and feeds it back to the processing unit via the data acquisition unit. The processing unit adjusts the control signal based on the deviation between the feedback pressure value and the target pressure value.

[0009] Furthermore, the proportional pressure regulating valve includes a proportional pressure regulating valve body and a proportional pressure regulating valve core. The proportional pressure regulating valve body is fixed to the pressure pipeline via a flange connection. The proportional pressure regulating valve core is installed in the inner cavity of the proportional pressure regulating valve body, forming a sliding pair between the valve core and the valve body. The proportional pressure regulating valve core reciprocates linearly along the inner cavity of the proportional pressure regulating valve body to adjust the flow cross-section. The displacement of the proportional pressure regulating valve core is controlled by an electrical signal. An O-ring seal is used between the proportional pressure regulating valve core and the proportional pressure regulating valve body.

[0010] Furthermore, a safety valve is installed on the pressure pipeline. The safety valve is located between the proportional pressure regulating valve and the pressure vessel under test. The safety valve's set pressure is higher than the highest pressure level required for the test; when the pipeline pressure exceeds the set pressure, the safety valve automatically opens to release pressure. The pressure sensor is fixed to the pressure pipeline via a threaded connection, located between the proportional pressure regulating valve and the pressure vessel under test. The booster pump is an electro-hydraulic pump station, which is fixedly installed on the testing frame.

[0011] Furthermore, the anisotropic anomaly detection and classification module calculates the theoretical anisotropic ratio R0 based on laminate theory and the layup parameters at the measurement point location. When the anisotropic strain ratio R of a measurement point deviates from R0 by more than the adaptive deviation threshold, the measurement point is marked as an anisotropic anomaly response point, where R0 is the theoretically predicted anisotropic strain ratio and R is the measured anisotropic strain ratio. The adaptive deviation threshold is determined based on the statistical distribution of R values ​​of normal measurement points in the same batch under the same pressure level. For measurement points marked as anisotropic anomaly response points, the fiber direction slope deviation rate, perpendicular fiber direction slope deviation rate, anisotropic ratio deviation, and in-plane shear strain fluctuation characteristics are extracted to form a multidimensional anisotropic anomaly feature vector. This multidimensional anisotropic anomaly feature vector is input into the classification network pre-deployed in the processing unit, and the defect category probability and debonding severity score are output.

[0012] Furthermore, in the directional spatial clustering module, the clustering distance metric is calculated by dividing the distance component along the fiber direction by the fiber direction scaling factor sf, and the distance component perpendicular to the fiber direction by the perpendicular fiber direction scaling factor st, where sf is greater than st, sf is the fiber direction scaling factor, and st is the perpendicular fiber direction scaling factor. After clustering, an elliptical boundary is fitted to each debonded region cluster. The parameters of the elliptical boundary include the center coordinates, major axis length a, minor axis length b, and principal axis direction. And the comprehensive severity score S, where a is the length of the major axis of the ellipse and b is the length of the minor axis of the ellipse. Let S be the direction angle of the principal axis of the ellipse, and S be the overall severity score of the debonding region cluster.

[0013] Furthermore, in the normalization and cross-detection tracking module, the major axis length *a* and the minor axis length *b* are divided by the container perimeter *C* to generate the normalized major axis length *a* = *a* / C and the normalized minor axis length *b* = *b* / C, where *C* is the container perimeter, *a* is the normalized major axis length, and *b* is the normalized minor axis length. (Main axis direction) Subtract the local fiber entanglement angle θlocal corresponding to the debonding center to generate the normalized principal axis direction. = −θlocal, where θlocal is the local fiber entanglement angle at the debonding center. The normalized principal axis direction is used. The method for establishing cross-detection tracking links is as follows: when the distance between the center coordinates of the debonding region in the current detection and the center coordinates of the debonding region in the previous detection is less than the matching distance threshold, the two are associated as the same debonding region.

[0014] Furthermore, the group expansion basic model training module gathers cross-detection tracking link data for all debonding regions in a subset of containers with more than the minimum sequence length threshold. The group debonding expansion basic model is trained using the initial normalized debonding parameters and the container's operating condition parameters as condition variables. The group debonding expansion basic model outputs normalized major axis growth rate Δa, normalized minor axis growth rate Δb, and severity growth rate ΔS as multidimensional outputs. Here, Δa represents the change in normalized major axis length during adjacent detection weeks, Δb represents the change in normalized minor axis length during adjacent detection weeks, and ΔS represents the change in comprehensive severity score during adjacent detection weeks. The personalized Bayesian update module calculates the operating condition similarity between the target container and each container in the group based on the target container's current normalized debonding feature vector and container operating condition parameters. It uses operating condition similarity weighted to retrieve the debonding expansion trajectories of neighboring containers to generate prior expansion trajectories. A multidimensional Bayesian update is performed on the prior expansion trajectories using the joint likelihood function of Δa, Δb, and ΔS, generating the posterior personalized expansion trajectories and their multidimensional uncertainty intervals for each debonding region. The training of the group decoupling extended base model adopts an online incremental update method. Each time new container detection data is added, the incremental cross-detection tracking link data is integrated into the existing model parameters to complete the update.

[0015] Furthermore, the processing unit also includes a report generation module. This module summarizes the elliptical boundary parameters, defect categories, severity scores, post-hoc personalized extension trajectories, multi-dimensional parameter predictions, uncertainty ranges, and warning markers for each debonding region of each vessel, as well as the overall distribution reference information of the debonding status of the same batch of vessels provided by the group debonding extension basic model, and outputs pressure vessel pressure resistance test evaluation report data. The vessel support saddle is slidably installed on the test frame along the axial direction of the pressure vessel to be inspected, and is fixed to the test frame by locking bolts after sliding to the target position. The spacing of the triaxial strain gauge array along the axial and circumferential directions in the vessel end cap region is smaller than the spacing in the cylindrical section region of the vessel.

[0016] This invention provides a pressure vessel pressure resistance testing device that solves the technical problems of existing devices, such as high false alarm rates in weak area identification due to single-direction strain amplitude analysis, inability to characterize the directional expansion features of debonding regions by only outputting scalar degradation indicators, and insufficient data from a single vessel to independently fit a debonding expansion prediction model. The invention achieves the following technical effects: By transforming strain data to a fiber coordinate system and calculating anisotropic strain ratio indicators, it can distinguish between anisotropic stiffness degradation caused by debonding and uniform strain changes caused by normal interlaminar stress redistribution, reducing the false alarm rate in weak area identification; by directional spatial clustering and fitting elliptical boundaries, it generates multidimensional structured boundary parameters including major axis length, minor axis length, and principal axis direction, enabling a complete characterization of the morphology and directional expansion information of the debonding region; by collecting normalized debonding evolution data from multiple vessels in the same batch to train a group debonding expansion basic model, and using multidimensional Bayesian updates to generate posterior personalized expansion trajectories, it achieves prediction of the debonding region expansion trend under the condition of limited single vessel testing records. Attached Figure Description

[0017] Figure 1 This is a front view of a pressure vessel pressure resistance testing device according to the present invention; Figure 2 This is a longitudinal sectional view of a pressure vessel pressure resistance testing device according to the present invention; Figure 3 This is a cross-sectional view of the pressure vessel pressure resistance testing device of the present invention along the vessel support saddle. Figure 4 This is a cross-sectional view of the proportional pressure regulating valve in a pressure vessel pressure resistance testing device of the present invention. Figure 5 This is a side view of the pressure vessel to be tested in a pressure vessel pressure resistance testing device of the present invention; Figure 6 This is a top view of a pressure vessel pressure resistance testing device according to the present invention.

[0018] In the diagram: Detection frame-1, container support saddle-2, pressure vessel to be inspected-3, electric hydraulic pump station-4, pressure pipeline-5, proportional pressure regulating valve body-6, proportional pressure regulating valve core-7, pressure sensor-8, triaxial strain gauge-9, signal conditioning module-10, data acquisition unit-11, processing unit-12, safety valve-13. Detailed Implementation

[0019] In industrial settings such as hydrogen refueling stations and chemical plants, multiple composite material pressure vessels manufactured in the same batch need to undergo periodic pressure tests to assess their structural safety. During these tests, a triaxial strain rosette array (9-array) is typically deployed on the vessel's outer surface. Multi-directional strain time-series data is collected under stepped pressure loading, and anomalies in the strain field are analyzed to identify deterioration defect areas within the vessel. The primary form of deterioration in composite material pressure vessels is interlaminar debonding. Due to the directional constraints of the fiber winding structure, the strain response and propagation morphology of the debonded areas exhibit significant anisotropic characteristics.

[0020] Existing pressure testing devices perform amplitude comparison or unidirectional nonlinear fitting on the collected strain data to determine weak areas, and independently perform trend analysis on the test results of a single vessel. This has the following technical defects.

[0021] First, the strain change rate along the fiber direction and perpendicular to the fiber direction in the interlaminar debonding region exhibits differentiated nonlinear characteristics during the pressurization process. Anomaly detection methods based on strain amplitude or slope in a single direction cannot characterize this anisotropic response mode. Furthermore, the abnormal strain amplitude caused by debonding overlaps with the normal interlaminar stress redistribution phenomenon in amplitude range, resulting in a high false alarm rate for weak area identification.

[0022] Second, the debonding region exhibits an elliptical evolution characteristic that preferentially expands along the fiber direction. Its boundary parameters include the major axis, minor axis, and principal axis directions, constituting multidimensional structured information. However, existing devices only output scalar degradation indicators and cannot characterize this directional expansion characteristic.

[0023] Third, a single container typically has only three to five periodic inspection records, and the data points are insufficient to independently fit a debonding propagation prediction model. Although multiple containers in the same batch share similar degradation characteristics, existing devices have failed to utilize batch group data to assist in the prediction of individual debonding propagation.

[0024] Product Structure of this Embodiment: According to an embodiment of this embodiment, a pressure vessel pressure resistance testing device is provided, which is used to perform pressure resistance testing on composite material pressure vessels of the same batch and evaluate the current state and expansion trend of debonding defects. A pressure vessel pressure resistance testing device includes at least a testing frame 1, a vessel support saddle 2, a pressurization assembly, a strain data acquisition assembly, and a processing unit 12.

[0025] The testing frame 1 serves as the supporting base for the entire device, used to fix and connect all testing components and support the pressure vessel 3 to be tested. The pressure vessel 3 to be tested is a cylindrical shell, with its axis placed horizontally along the vessel axis.

[0026] The container support saddles 2 are a pair, fixedly connected to the testing frame 1, and arranged at intervals along the container axis. Each container support saddle 2 has a saddle arc surface, which is in close contact with the outer wall arc surface of the pressure vessel 3 to be tested, constraining the displacement of the pressure vessel 3 in the radial horizontal direction and the height direction, so that the pressure vessel 3 to be tested is stably positioned on the testing frame 1.

[0027] In some embodiments, the container support saddle 2 is an adjustable spacing support saddle, which is slidably installed on the detection frame 1 along the container axis. After sliding to the target position, it is fixed to the detection frame 1 by locking bolts to adapt to pressure vessels of different lengths.

[0028] The pressurization assembly is used to apply a stepped pressure increase to the inner cavity of the pressure vessel 3 under test, and includes at least a booster pump, a pressure line 5, a proportional pressure regulating valve, and a pressure sensor 8. The booster pump is fixedly mounted on the testing frame 1, with its inlet connected to a liquid source and its outlet connected to one end of the pressure line 5. The pressure line 5, fixed to the testing frame 1, transmits liquid pressure. The proportional pressure regulating valve is installed on the pressure line 5 and includes a proportional pressure regulating valve body 6 and a proportional pressure regulating valve core 7. The proportional pressure regulating valve body 6 is fixed to the pressure line 5 via a flange connection. The proportional pressure regulating valve core 7 is installed within the inner cavity of the proportional pressure regulating valve body 6, forming a sliding pair with the valve body 6. The proportional pressure regulating valve core 7 reciprocates linearly along the inner cavity of the proportional pressure regulating valve body 6 to adjust the flow cross-section, thereby controlling the pressure level of the pipeline. Pressure sensor 8 is fixed to pressure line 5 via a threaded connection, located between the proportional pressure regulating valve and the pressure vessel 3 under test, and is used to collect real-time pressure values ​​within the line. The other end of pressure line 5 is connected to the inlet of pressure vessel 3 under test, applying liquid pressure to the vessel's interior. The force transmission path is as follows: liquid pressure is output from the booster pump to pressure line 5, adjusted by the proportional pressure regulating valve, and then transmitted along pressure line 5 to the inlet of pressure vessel 3 under test. This pressure enters the vessel's interior and applies internal pressure to the vessel wall, causing strain on the vessel wall under the influence of this internal pressure.

[0029] In some embodiments, the booster pump is an electro-hydraulic pump station 4, which provides a stable hydraulic output flow.

[0030] In some embodiments, the proportional pressure regulating valve is an electro-proportional pressure reducing valve. The displacement of the proportional pressure regulating valve core 7 of the electro-proportional pressure reducing valve is controlled by an electrical signal. The pressure in the pressure line 5 is continuously adjustable by adjusting the opening of the proportional pressure regulating valve core 7. An O-ring seal is used between the proportional pressure regulating valve core 7 and the proportional pressure regulating valve body 6.

[0031] In some embodiments, the inlet of the pressure vessel 3 to be inspected is connected to the pressure pipeline 5 via a quick connector, which facilitates the quick replacement of different vessels.

[0032] Furthermore, in order to prevent the pipeline pressure from exceeding the safe range during the pressurization process, a safety valve 13 is also installed on the pressure pipeline 5. The safety valve 13 is located between the proportional pressure regulating valve and the pressure vessel 3 to be tested. The set pressure of the safety valve 13 is higher than the highest pressure level required for testing. When the pipeline pressure exceeds the set pressure, the safety valve 13 automatically opens to release pressure.

[0033] The strain data acquisition component is used to acquire multi-directional strain time-series data of the outer wall of the pressure vessel 3 under test, and includes at least a triaxial strain gauge array 9, a signal conditioning module 10, and a data acquisition unit 11. The triaxial strain gauge array 9 is bonded and fixed to the outer surface of the pressure vessel 3 under test, and is uniformly distributed along the vessel's axial direction and circumference. Each triaxial strain gauge 9 contains three strain-sensitive grids arranged in different directions, used to simultaneously acquire strain signals from three directions at that measuring point. The signal conditioning module 10 is fixedly installed on the detection frame 1 and connected to the triaxial strain gauge array 9 via electrical connection, used to amplify, filter, and bridge the original strain electrical signal. The data acquisition unit 11 is fixedly installed on the detection frame 1 and connected to the signal conditioning module 10 via electrical connection, used to convert the conditioned analog signal into a digital signal and perform synchronous sampling. The data acquisition unit 11 is also connected to the pressure sensor 8 via electrical connection to synchronously acquire the real-time pressure value within the pipeline.

[0034] Furthermore, in order to improve the spatial resolution of debonding identification, the triaxial strain flower 9 array is densely deployed in the end cap region of the container, with the deployment spacing along the container axis and circumferential direction being smaller than the deployment spacing in the cylindrical section region of the container.

[0035] The processing unit 12 is fixedly installed on the detection frame 1 and connected to the data acquisition unit 11 via a data connection to receive digitized strain timing data and real-time pressure values. The processing unit 12 is also connected to the proportional pressure regulating valve core 7 via a control signal connection, outputting a stepped pressure increase control signal to control the displacement of the proportional pressure regulating valve core 7. A pressure closed-loop control circuit is formed between the processing unit 12 and the pressure sensor 8 via the data acquisition unit 11: the processing unit 12 outputs a stepped pressure increase command to the proportional pressure regulating valve core 7, the proportional pressure regulating valve core 7 adjusts the pipeline pressure, the pressure sensor 8 collects the real-time pipeline pressure value and feeds it back to the processing unit 12 via the data acquisition unit 11, and the processing unit 12 adjusts the control signal according to the deviation between the feedback pressure value and the target pressure value.

[0036] The processing unit 12 contains the following functional modules, which sequentially perform all data processing and model calculations.

[0037] The coordinate transformation and anisotropic index calculation module is used to receive triaxial strain time series data and the fiber winding angle at the corresponding measuring point position. It performs a rotation transformation on the strain data from the sensor coordinate system to the fiber coordinate system to obtain time series data of strain components along the fiber direction and strain components perpendicular to the fiber direction. It performs online piecewise linear fitting on the strain components in the two directions respectively, calculates the strain-pressure slope value of each, and calculates the ratio of the slope values ​​in the two directions to generate the anisotropic strain ratio index and its change trajectory with increasing pressure.

[0038] The anisotropic anomaly detection and classification module is used to perform abrupt change detection on the change trajectory of the anisotropic strain ratio index. The measurement points that deviate from the theoretical anisotropic ratio value by more than the adaptive deviation threshold are marked as anisotropic anomaly response points. The multidimensional anisotropic anomaly feature vector of the measurement point is extracted and input into the pre-deployed classification network to output the defect category probability and debonding severity score.

[0039] The directional spatial clustering module is used to perform directional spatial clustering on the measurement points identified as interlayer debonding. The clustering distance metric uses different scale factors along the fiber direction and perpendicular to the fiber direction to match the elliptical expansion characteristics of the debonding region, generating debonding region clusters and their elliptical boundary parameters.

[0040] The normalization and cross-detection tracking module is used to perform scale normalization and orientation normalization on the debonding ellipse parameters identified in each detection, generate a normalized debonding feature vector, and perform spatial position matching on the debonding regions between adjacent detection cycles to establish cross-detection tracking links for the same debonding region.

[0041] The Population Expansion Base Model Training Module is used to aggregate cross-detection tracking link data of a subset of containers from multiple containers in the same batch that have exceeded the minimum sequence length threshold, and train the Population Decoupling Expansion Base Model. The Population Decoupling Expansion Base Model outputs normalized major axis growth rate, normalized minor axis growth rate, and severity growth rate in multiple dimensions.

[0042] The personalized Bayesian update module is used to retrieve the debonding extension trajectories of neighboring containers from the group debonding extension base model with working condition similarity weighting for each identified debonding region in the target container to generate a prior extension trajectory. Then, it performs multidimensional Bayesian update using the limited detection data of the target container to generate the posterior personalized extension trajectory and its multidimensional uncertainty interval for each debonding region.

[0043] The forward extrapolation and early warning module is used to extrapolate the cumulative number of cycles corresponding to the next detection cycle on the posterior personalized extension trajectory, predict the predicted values ​​and uncertainty range of the major axis length, minor axis length and severity of each debonding region, and compare the upper bound of the predicted elliptical area uncertainty with the maximum allowable debonding area threshold of the layup, and generate accelerated detection early warning marks for debonding regions that exceed the threshold.

[0044] The report generation module is used to summarize the elliptical boundary parameters, defect categories, severity scores, posterior extension trajectories, multidimensional parameter predictions, uncertainty ranges, early warning markers, and overall distribution reference information of the debonding status of containers in the same batch provided by the group debonding extension basic model for each debonding area of ​​each container, and output pressure vessel pressure resistance test evaluation report data.

[0045] The following steps are performed when a pressure vessel pressure resistance testing device is in operation.

[0046] Processing unit 12 acquires multi-directional strain time-series data collected by a triaxial strain gauge array 9 during the stepped pressurization process of each pressure test for all M composite material pressure vessels in the same batch, where M is the total number of vessels in the same batch. Processing unit 12 also acquires the fiber winding angle distribution map, laminate layup parameters, real-time pressure values ​​for each pressure level, service condition parameters for each vessel, and the cumulative number of charge-discharge cycles for each test interval for each vessel. The stepped pressurization process is achieved by processing unit 12 driving the proportional pressure regulating valve core 7 to adjust the pipeline pressure step by step through a pressure closed-loop control loop.

[0047] The coordinate transformation and anisotropic index calculation module performs coordinate transformation on the time-series strain data in the three directions at each of the nine measuring points of the triaxial strain gauge. Specifically, for any measuring point, based on the fiber winding angle θ corresponding to that measuring point location, the strain data is rotated from the sensor coordinate system to the fiber coordinate system to obtain the time-series data of the fiber-direction strain component εf and the perpendicular-fiber-direction strain component εt, where εf represents the fiber-direction strain value and εt represents the perpendicular-fiber-direction strain value. Then, online piecewise linear fitting is performed on εf and εt respectively to calculate their respective strain-to-pressure slope values ​​kf and kt within the current pressure range, where kf is the slope of the fiber-direction strain with respect to pressure, and kt is the slope of the perpendicular-fiber-direction strain with respect to pressure. The ratio of the two slope values ​​is calculated to generate the anisotropic strain ratio index R = kf / kt, where R represents the anisotropic strain ratio. The trajectory of R as pressure increases is monitored.

[0048] It should be noted that the slope ratio of the two orthogonal directions was calculated after transforming the strain data to the fiber coordinate system, utilizing the physical characteristic of the differential strain response in the debonding region along the fiber direction and perpendicular to the fiber direction. The R-value in the normal region remains stable at a level consistent with the theoretical prediction of the laminate under various pressure levels. However, due to the anisotropic characteristics of local stiffness degradation in the debonding region, the R-value deviates with increasing pressure. This deviation pattern differs from the unidirectional amplitude change caused by normal interlaminar stress redistribution.

[0049] The anisotropy anomaly detection and classification module performs abrupt change detection on the trajectory of the anisotropic strain ratio index R. For any measuring point, the theoretical anisotropy ratio R0 is calculated based on laminate theory and the layup parameters at that measuring point location, where R0 is the theoretically predicted anisotropic strain ratio. When the R value at a measuring point deviates from R0 by more than an adaptive deviation threshold, the measuring point is marked as an anisotropic anomaly response point. The adaptive deviation threshold is determined based on the statistical distribution of R values ​​at the same pressure level for normal measuring points in the same batch.

[0050] For measurement points marked as anisotropic anomaly response points, their fiber direction slope deviation rate, perpendicular fiber direction slope deviation rate, anisotropy ratio deviation, and in-plane shear strain fluctuation characteristics are extracted to form a multidimensional anisotropic anomaly feature vector. The multidimensional anisotropic anomaly feature vector is input into the classification network pre-deployed in the processing unit 12, and the defect category probability and debonding severity score are output.

[0051] It should be noted that the multidimensional anisotropic anomaly feature vector integrates the slope deviation information of the fiber direction and the perpendicular fiber direction as well as the shear strain fluctuation information. Compared with the discrimination of strain amplitude in a single direction, it can distinguish between the anisotropic stiffness degradation caused by debonding and the uniform strain change caused by normal interlaminar stress redistribution, thereby reducing the false alarm rate of weak area identification.

[0052] The directional spatial clustering module performs directional spatial clustering on the measurement points identified as interlayer debonding, based on their spatial distribution and debonding severity score. Specifically, in the clustering distance metric, the distance component along the fiber direction is divided by the fiber direction scaling factor sf, and the distance component perpendicular to the fiber direction is divided by the perpendicular fiber direction scaling factor st, where sf is greater than st, to match the elliptical feature of the debonding region preferentially expanding along the fiber direction. After clustering, an elliptical boundary is fitted for each debonding region cluster, generating elliptical boundary parameters, including center coordinates, major axis length a, minor axis length b, and principal axis direction. And the comprehensive severity score S, where a is the length of the major axis of the ellipse and b is the length of the minor axis of the ellipse. Let S be the direction angle of the principal axis of the ellipse, and S be the overall severity score of the debonding region cluster.

[0053] The normalization and cross-detection tracking module normalizes the debonding ellipse parameters identified in each pressure test of each vessel. The major axis length *a* and minor axis length *b* are divided by the vessel circumference *C* for scale normalization, generating normalized major axis length *a* = *a* / *C* and normalized minor axis length *b* = *b* / *C*, where *C* is the vessel circumference, *a* is the normalized major axis length, and *b* is the normalized minor axis length. (Major axis direction) Subtract the local fiber entanglement angle θlocal corresponding to the debonding center position and normalize the direction to generate the normalized principal axis direction. = −θlocal, where θlocal is the local fiber entanglement angle at the debonding center. This represents the normalized principal axis direction. This generates a normalized debonding feature vector that eliminates differences in container scale and local winding angles.

[0054] Subsequently, spatial location matching is performed on the debonding regions between adjacent detection cycles. When the distance between the center coordinates of the debonding region in the current detection and the center coordinates of the debonding region in the previous detection is less than the matching distance threshold, the two are associated as the same debonding region, and a cross-detection tracking link is established.

[0055] It should be noted that scale normalization eliminates the influence of different container sizes on the absolute values ​​of debonding boundary parameters, and orientation normalization eliminates the systematic deviation of the principal axis orientation caused by different local fiber winding angles in the debonding areas of different containers. This allows the debonding evolution data from different containers in the same batch to be directly compared and aggregated in a unified normalized space.

[0056] The swarm expansion basic model training module gathers cross-detection tracking link data for all debonding regions in a subset of containers where the number of detections exceeds the minimum sequence length threshold. The swarm debonding expansion basic model is trained using the initial normalized debonding parameters and the container's operational parameters as conditional variables. The swarm debonding expansion basic model outputs three dimensions: normalized major axis growth rate Δa, normalized minor axis growth rate Δb, and severity growth rate ΔS. Here, Δa represents the change in normalized major axis length between adjacent detection weeks, Δb represents the change in normalized minor axis length between adjacent detection weeks, and ΔS represents the change in the overall severity score between adjacent detection weeks. The swarm debonding expansion basic model learns the swarm baseline expansion rate and covariance structure of each dimension for debonding regions within the same batch of containers.

[0057] The personalized Bayesian update module calculates the similarity of the operating conditions of each identified debonding region in the target container with that of each container in the group, based on its current normalized debonding feature vector and the container's operating condition parameters. It then retrieves the debonding extension trajectories of neighboring containers from the group's debonding extension baseline model using a weighted average of operating condition similarity, generating prior extension trajectories. Subsequently, using the normalized debonding feature vectors from the limited existing detection data of the target container, it performs a multidimensional Bayesian update on the prior extension trajectories using the joint likelihood function of the changes in parameters Δa, Δb, and ΔS across all dimensions, generating the posterior personalized extension trajectories of each debonding region and their multidimensional uncertainty intervals.

[0058] It should be noted that the swarm decoupling expansion baseline model aggregates decoupling evolution data from multiple containers in the same batch, providing swarm baseline information for decoupling expansion. For target containers with only three to five detection records, the prior expansion trajectory generated by weighting based on operational similarity compensates for the lack of individual data. The multidimensional Bayesian update uses the actual measured data of the target container to correct the prior expansion trajectory, so that the posterior personalized expansion trajectory simultaneously reflects the swarm baseline characteristics and the actual evolutionary trend of the individual.

[0059] The forward extrapolation and early warning module extrapolates the cumulative number of cycles corresponding to the next detection cycle along the posterior personalized extension trajectory, predicting the predicted values ​​and uncertainty ranges of the major axis length, minor axis length, and severity of each debonding region. The upper bound of the predicted elliptical area uncertainty is compared with the maximum allowable debonding area threshold for the layup. For debonding regions where the upper bound of the elliptical area uncertainty exceeds the maximum allowable debonding area threshold, an accelerated detection early warning marker is generated.

[0060] The report generation module summarizes the elliptical boundary parameters, defect categories, severity scores, post-hoc personalized extension trajectories, multi-dimensional parameter predictions, uncertainty ranges, and warning markers for each debonding area of ​​each vessel, as well as the overall distribution reference information of the debonding status of the same batch of vessels provided by the group debonding extension basic model, and outputs pressure vessel pressure resistance test and evaluation report data.

[0061] In some embodiments, the training of the swarm decoupling extended base model adopts an online incremental update method. Each time new container detection data is added, the swarm extended base model training module integrates the incremental cross-detection tracking link data into the existing model parameters to complete the update of the swarm decoupling extended base model, without having to re-perform batch training on all historical data.

[0062] The technical effect of this embodiment: The pressure vessel pressure resistance testing device of this embodiment transforms the strain data collected by the triaxial strain rosette array 9 to the fiber coordinate system through the coordinate transformation and anisotropic index calculation module of the processing unit 12, and calculates the ratio of strain to pressure slope in the fiber direction and perpendicular fiber direction to generate anisotropic strain ratio index. It utilizes the physical characteristics of the differential strain response changes in the interlaminar debonding region in two orthogonal directions. Therefore, the anisotropic anomaly detection and classification module can identify debonding anomalies based on the deviation pattern of the anisotropic strain ratio index with increasing pressure, and distinguish between the anisotropic stiffness degradation caused by debonding and the uniform strain change caused by normal interlaminar stress redistribution. This overcomes the problem that the single-direction strain amplitude or slope analysis cannot characterize the anisotropic response mode and overlaps with the normal interlaminar stress redistribution phenomenon in amplitude range, and reduces the false alarm rate of weak area judgment.

[0063] Because the directional spatial clustering module performs directional clustering of the debonding measurement points along the fiber direction and perpendicular to the fiber direction using different scale factors, and fits the clustering results to an elliptical boundary to generate multidimensional structured boundary parameters including center coordinates, major axis length, minor axis length, principal axis direction and comprehensive severity score, it overcomes the problem that existing devices only output scalar degradation indicators and cannot characterize the elliptical directional characteristics of the debonding region that preferentially expands along the fiber direction, thus enabling a complete characterization of the morphology and directional expansion information of the debonding region.

[0064] Because the normalization and cross-detection tracking modules perform scale and orientation normalization on the debonding ellipse parameters, eliminating the systematic bias of scale and local entanglement angle differences between different containers on the aggregation of group data, the group expansion basic model training module can effectively aggregate normalized debonding evolution data from multiple containers in the same batch to train the group debonding expansion basic model. Furthermore, because the personalized Bayesian update module retrieves the prior expansion trajectory from the group debonding expansion basic model using working condition similarity weighting and performs multidimensional Bayesian update to generate the posterior personalized expansion trajectory using a small amount of measured data from the target container, it overcomes the problem that data points are insufficient to independently fit the debonding expansion prediction model when a single container has only three to five detection records, thus achieving reliable prediction of anisotropic expansion of the debonding region.

[0065] As can be seen, the pressure vessel pressure resistance testing device of this embodiment provides a complete evaluation capability for pressure vessel pressure resistance testing, from the identification of the current debonding state to the prediction of the debonding expansion in the next cycle.

[0066] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A pressure vessel pressure resistance testing device, characterized in that, include: The test frame (1) includes a pair of container support saddles (2) fixedly connected to the test frame (1) and spaced apart along the axis of the pressure vessel (3) to be tested; a pressurization assembly installed on the test frame (1), including a booster pump, a pressure pipeline (5), a proportional pressure regulating valve and a pressure sensor (8), used to apply a stepped pressure increase to the inner cavity of the pressure vessel (3) to be tested; a strain data acquisition assembly including a triaxial strain flower (9) array pasted on the outer surface of the pressure vessel (3) to be tested, a signal conditioning module (10) and a data acquisition unit (11); and a processing unit (12) connected to the data acquisition unit (11). The processing unit (12) includes: a coordinate transformation and anisotropic index calculation module, which transforms the strain data from the sensor coordinate system to the fiber coordinate system and calculates the ratio of the strain to pressure slope in the fiber direction and the direction perpendicular to the fiber direction to generate anisotropic strain ratio index; and an anisotropic anomaly detection and classification module, which performs abrupt change detection on the change trajectory of the anisotropic strain ratio index and outputs a debonding severity score. The directional spatial clustering module uses different scale factors to cluster debonding measurement points along the fiber direction and perpendicular to the fiber direction, generating elliptical boundary parameters; The normalization and cross-detection tracking module normalizes the parameters of the debonded ellipse and establishes cross-detection tracking links. The group expansion basic model training module gathers tracking link data from multiple containers in the same batch to train the group decoupling expansion basic model; the personalized Bayesian update module retrieves the prior expansion trajectory from the group decoupling expansion basic model and uses the target container detection data to perform Bayesian update to generate the posterior personalized expansion trajectory.

2. The pressure vessel pressure resistance testing device according to claim 1, characterized in that, The processing unit (12) further includes a forward extrapolation and early warning module. The forward extrapolation and early warning module extrapolates the cumulative number of cycles corresponding to the next detection cycle on the posterior personalized extension trajectory, predicts the predicted values ​​and uncertainty ranges of the major axis length, minor axis length and severity of each debonding region, and compares the upper bound of the predicted elliptical area uncertainty with the maximum allowable debonding area threshold of the layup. For debonding regions where the upper bound of the elliptical area uncertainty exceeds the maximum allowable debonding area threshold of the layup, an accelerated detection early warning mark is generated.

3. The pressure vessel pressure resistance testing device according to claim 1, characterized in that, The processing unit (12) is simultaneously connected to the proportional pressure regulating valve core (7) of the proportional pressure regulating valve via a control signal. The processing unit (12) and the pressure sensor (8) form a pressure closed-loop control circuit through the data acquisition unit (11). The processing unit (12) outputs a step-by-step pressure increase command to the proportional pressure regulating valve core (7), which regulates the pipeline pressure. The pressure sensor (8) collects the real-time pipeline pressure value and feeds it back to the processing unit (12) via the data acquisition unit (11). The processing unit (12) adjusts the control signal according to the deviation between the feedback pressure value and the target pressure value.

4. The pressure vessel pressure resistance testing device according to claim 3, characterized in that, The proportional pressure regulating valve includes a proportional pressure regulating valve body (6) and a proportional pressure regulating valve core (7). The proportional pressure regulating valve body (6) is fixed to the pressure pipeline (5) by a flange connection. The proportional pressure regulating valve core (7) is installed in the inner cavity of the proportional pressure regulating valve body (6). The proportional pressure regulating valve core (7) and the proportional pressure regulating valve body (6) form a sliding pair. The proportional pressure regulating valve core (7) moves linearly back and forth along the inner cavity of the proportional pressure regulating valve body (6) to adjust the flow cross section. The displacement of the proportional pressure regulating valve core (7) is controlled by an electrical signal. The proportional pressure regulating valve core (7) and the proportional pressure regulating valve body (6) are sealed with an O-ring.

5. The pressure vessel pressure resistance testing device according to claim 4, characterized in that, A safety valve (13) is also installed on the pressure pipeline (5). The safety valve (13) is located between the proportional pressure regulating valve and the pressure vessel (3) to be tested. The set pressure of the safety valve (13) is higher than the highest pressure level required for testing. When the pipeline pressure exceeds the set pressure, the safety valve (13) automatically opens to release pressure. The pressure sensor (8) is fixed to the pressure pipeline (5) by a threaded connection and is located between the proportional pressure regulating valve and the pressure vessel (3) to be tested. The booster pump is an electric hydraulic pump station (4), which is fixedly installed on the testing frame (1).

6. The pressure vessel pressure resistance testing device according to claim 1, characterized in that, The anisotropic anomaly detection and classification module calculates the theoretical anisotropic ratio R0 based on the laminate theory and the layup parameters of the measuring point location. When the anisotropic strain ratio index R of the measuring point deviates from R0 by more than the adaptive deviation threshold, the measuring point is marked as an anisotropic anomaly response point. R0 is the theoretically predicted anisotropic strain ratio, and R is the measured anisotropic strain ratio. The adaptive deviation threshold is determined based on the statistical distribution of R values ​​of normal measuring points in the same batch under the same pressure level. For the measuring point marked as anisotropic anomaly response point, the fiber direction slope deviation rate, the vertical fiber direction slope deviation rate, the anisotropic ratio deviation degree, and the in-plane shear strain fluctuation characteristics are extracted to form a multidimensional anisotropic anomaly feature vector. The multidimensional anisotropic anomaly feature vector is input into the classification network pre-deployed in the processing unit (12) to output the defect category probability and debonding severity score.

7. The pressure vessel pressure resistance testing device according to claim 1, characterized in that, In the directional spatial clustering module, the clustering distance metric is calculated by dividing the distance component along the fiber direction by the fiber direction scaling factor sf, and the distance component perpendicular to the fiber direction by the perpendicular fiber direction scaling factor st, where sf is greater than st, sf is the fiber direction scaling factor, and st is the perpendicular fiber direction scaling factor. After clustering, an elliptical boundary is fitted to each debonding region cluster. The parameters of the elliptical boundary include the center coordinates, major axis length a, minor axis length b, and principal axis direction. And the comprehensive severity score S, where a is the length of the major axis of the ellipse and b is the length of the minor axis of the ellipse. Let S be the direction angle of the principal axis of the ellipse, and S be the overall severity score of the debonding region cluster.

8. The pressure vessel pressure resistance testing device according to claim 7, characterized in that, In the normalization and cross-detection tracking module, the major axis length *a* and the minor axis length *b* are divided by the container perimeter *C* to generate the normalized major axis length *a* = *a* / C and the normalized minor axis length *b* = *b* / C, where *C* is the container perimeter, *a* is the normalized major axis length, and *b* is the normalized minor axis length; (Main axis direction) Subtract the local fiber entanglement angle θlocal corresponding to the debonding center to generate the normalized principal axis direction. = −θlocal, where θlocal is the local fiber entanglement angle at the debonding center. The normalized principal axis direction; the method for establishing the cross-detection tracking link is as follows: when the distance between the center coordinates of the debonding region in the current detection and the center coordinates of the debonding region in the previous detection is less than the matching distance threshold, the two are associated as the same debonding region.

9. A pressure vessel pressure resistance testing device according to claim 8, characterized in that, The group expansion basic model training module gathers cross-detection tracking link data of all debonding regions in the container subset where the number of detections exceeds the minimum sequence length threshold. The group debonding expansion basic model is trained with the initial normalized debonding parameters and the operating parameters of the container as condition variables. The group debonding expansion basic model outputs normalized major axis growth rate Δa, normalized minor axis growth rate Δb, and severity growth rate ΔS as multidimensional outputs, where Δa is the change in normalized major axis length during adjacent detection weeks, Δb is the change in normalized minor axis length during adjacent detection weeks, and ΔS is the change in comprehensive severity score during adjacent detection weeks. The personalized Bayesian update module calculates the similarity of the target container's current normalized decoupling feature vector and container operating parameters with the operating conditions of each container in the group. It uses the operating condition similarity to weight the decoupling extension trajectory of the nearest neighboring container to generate a prior extension trajectory. The prior extension trajectory is then updated using a joint likelihood function of Δa, Δb, and ΔS to generate a posterior personalized extension trajectory and its multidimensional uncertainty interval for each decoupling region. The training of the group decoupling extension base model adopts an online incremental update method. Each time new container detection data is added, the incremental cross-detection tracking link data is integrated into the existing model parameters to complete the update.

10. A pressure vessel pressure resistance testing device according to claim 2, characterized in that, The processing unit (12) also includes a report generation module, which summarizes the elliptical boundary parameters, defect categories, severity scores, post-hoc personalized extension trajectories, multi-dimensional parameter prediction values, uncertainty ranges and warning markers of each debonding area of ​​each container, as well as the overall distribution reference information of the debonding status of the same batch of containers provided by the group debonding extension basic model, and outputs pressure vessel pressure resistance test evaluation report data; the container support saddle (2) is slidably installed on the test frame (1) along the axial direction of the pressure vessel (3) to be inspected, and is fixed to the test frame (1) by locking bolts after sliding to the target position; the arrangement spacing of the triaxial strain rose (9) array in the end cap area of ​​the container along the axial direction and circumferential direction is smaller than the arrangement spacing in the cylindrical section area of ​​the container.

Images