Natural gas station elbow tee joint stress fatigue digital intelligent analysis method, system and product

Abstract

The invention relates to the technical field of pipeline stress fatigue detection, and discloses a natural gas station elbow tee joint stress fatigue digital intelligent analysis method and system and a product. The method comprises the following steps: establishing a multi-physics field coupling digital twinborn model of the elbow tee joint, integrating a geometric structure, material attributes and a fluid-solid coupling mechanism, and simulating flow-induced vibration stress and corrosion fatigue interaction; collecting real-time multi-source data of the elbow tee joint; real-time multi-source data is utilized to update boundary conditions and parameters of the digital twin model, and material constants are dynamically corrected through machine learning, so that self-calibration of the model is realized; performing stress distribution analysis based on the updated model to obtain stress field data, and performing fatigue crack propagation prediction in combination with real-time multi-source data to generate a prediction result; and based on the prediction result, evaluating fatigue failure risks, including crack growth rate, residual life evaluation and failure probability, and outputting alarm information or optimization decision information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pipeline stress fatigue detection, and particularly relates to a stress fatigue numerical-intelligence analysis method, system and product for a natural gas station elbow tee joint. BACKGROUND

[0002] The natural gas station elbow tee joint, as a key component of a pipeline system, is subjected to alternating loads under high pressure, high temperature and corrosive environments, and is prone to stress fatigue damage.

[0003] The existing technical system has significant defects. International standards such as ASME B31.3 / B31.8 only provide constant amplitude S-N curves and do not cover random load spectrum and corrosion fatigue combined working conditions; the Paris formula of API 579 ignores the hydrogen-induced cracking mechanism, and the defect detection threshold of 1 mm is difficult to capture microcracks of 0.1-0.5 mm. The material database is based on laboratory data, ignores the influence of field welding, and the fatigue strength is 40% lower; the test method cannot simulate H2S+CO2+Cl-multiple component corrosion. The error of finite element analysis (FEA) for multi-axial non-proportional loading is 30%-50%, and the deviation of fluid-structure coupling simulation is more than 20%.

[0004] Technical difficulties include multi-physical field coupling modeling problems, such as fluid-structure vibration and corrosion-fatigue interaction; random load equivalent simplification error of 30%-60%; micro-scale damage detection bottleneck, conventional ultrasonic detection limit of 0.5 mm, nonlinear ultrasonic need for reference signal; material database gap, domestic X80 steel low temperature data missing, long-term service degradation model deviation more than 50%; digital twin bottleneck, high-fidelity model calculation time more than 72 hours, multi-source data fusion barrier. These defects result in low fatigue evaluation accuracy, which cannot realize predictive maintenance, and innovative methods are needed to improve monitoring and analysis efficiency. SUMMARY

[0005] The purpose of the present application is to solve the problems in the prior art, and a stress fatigue numerical-intelligence analysis method, system and product for a natural gas station elbow tee joint are provided. In order to achieve the above purpose, the technical scheme adopted by the embodiments of the present application is as follows:

[0006] In a first aspect, the embodiments of the present application propose a stress fatigue numerical-intelligence analysis method for a natural gas station elbow tee joint, comprising the following steps:

[0007] A multi-physical field coupling digital twin model of the elbow tee joint is established based on finite element analysis and computational fluid dynamics, and the digital twin model integrates geometric structure, material properties and fluid-structure coupling mechanism to simulate the interaction of flow-induced vibration stress and corrosion fatigue;

[0008] Real-time multi-source data of the elbow tee joint is collected, and the real-time multi-source data includes strain, vibration, pressure, temperature and corrosion medium concentration data;

[0009] updating boundary conditions and parameters of the digital twin model using the real-time multi-source data, and dynamically correcting material constants by machine learning to achieve self-calibration of the digital twin model;

[0010] performing stress distribution analysis based on the updated digital twin model to obtain stress field data, and performing fatigue crack propagation prediction based on the stress field data and the real-time multi-source data to generate fatigue crack propagation prediction results;

[0011] evaluating fatigue failure risk of the elbow tee based on the fatigue crack propagation prediction results, the fatigue failure risk including crack propagation rate, residual life assessment and failure probability, and outputting alarm information or optimization decision information according to the fatigue failure risk.

[0012] Preferably, the step of establishing the multi-physics field coupled digital twin model of the elbow tee includes: using computational fluid dynamics to solve gas flow load and transmitting to finite element structure model through fluid-solid coupling interface to calculate vibration stress distribution under multi-axial non-proportional loading, while integrating coupling correction factors considering the accelerating effect of corrosion medium concentration on crack propagation rate, fatigue damage accumulation and creep deformation.

[0013] Preferably, in the step of updating boundary conditions and parameters of the digital twin model using the real-time multi-source data, the machine learning dynamically corrects the material constants, including: using long short-term memory network to adjust material constants C and m of Paris formula based on historical crack propagation data and environmental parameters, wherein the Paris formula is:

[0014]

[0015] wherein, is the crack propagation rate, a is the crack depth, N is the stress cycle number, C and m are material constants, ΔK is the stress intensity factor amplitude, T is the environmental temperature, σ mean is the average stress, and pH is the pH value.

[0016] Preferably, the real-time multi-source data is collected by a sensor network deployed on the elbow tee, the sensor network including: fiber Bragg grating sensors monitoring strain and temperature, distributed acoustic sensing monitoring vibration, ultrasonic phased array detecting crack depth, and multi-gas sensors monitoring corrosion medium concentration.

[0017] Preferably, the sensor network includes an edge computing node, which performs real-time preprocessing on the real-time multi-source data, the real-time preprocessing including Fourier transform and time series compression.

[0018] Preferably, before the stress distribution analysis step based on the updated digital twin model, further comprising:

[0019] Residuals between the digital twin model prediction and actual measured data are calculated, and model parameters including damping coefficients and geometric correction coefficients are optimized through closed-loop feedback adjustment based on the residuals.

[0020] Preferably, the fatigue crack propagation prediction adopts an equivalent initial defect assumption to simplify micro-defect stage calculation, and a rainflow counting method is used to simplify a random load spectrum into a finite segment step spectrum to optimize calculation efficiency, while considering environmental correction factors to quantify the synergistic effect of temperature and corrosion medium.

[0021] Preferably, the optimization decision information includes a remaining life prediction curve, crack initiation position visualization, and maintenance strategy suggestions, wherein the maintenance strategy optimizes the whole life cycle by balancing safety benefits and maintenance costs through a reinforcement learning model.

[0022] In a second aspect, the embodiments of the present application propose a stress fatigue intelligent detection and digital analysis system for a natural gas station elbow tee, comprising:

[0023] A multi-physics field coupling modeling module is configured to establish a multi-physics field coupling digital twin model of the elbow tee based on finite element analysis and computational fluid dynamics, wherein the digital twin model integrates geometric structure, material properties, and fluid-structure coupling mechanisms to simulate the interaction of flow-induced vibration stress and corrosion fatigue.

[0024] A data acquisition module is configured to acquire real-time multi-source data of the elbow tee, including strain, vibration, pressure, temperature, and corrosion medium concentration data.

[0025] A model updating and correction module is configured to update boundary conditions and parameters of the digital twin model using the real-time multi-source data, and dynamically correct material constants through machine learning to realize self-calibration of the digital twin model.

[0026] A fatigue analysis module is configured to perform stress distribution analysis based on the updated digital twin model to obtain stress field data, and perform fatigue crack propagation prediction based on the stress field data and the real-time multi-source data to generate fatigue crack propagation prediction results.

[0027] A risk detection and output module is configured to evaluate fatigue failure risks of the elbow tee based on the fatigue crack propagation prediction results, including crack propagation rate, remaining life assessment, and failure probability, and output alarm information or optimization decision information according to the fatigue failure risks.

[0028] In a third aspect, the embodiments of the present application provide a computer program product, comprising a non-transitory computer-readable storage medium, and a computer executable instruction is stored in the storage medium, and the instruction is executed by a processor to make the processor execute the method provided by the above embodiments.

[0029] Advantages:

[0030] By establishing a multi-physics field coupling digital twin model, the precise simulation of the stress and corrosion fatigue interaction of the elbow tee of the natural gas station is realized, the limitations of traditional standards for random load and composite working condition evaluation are overcome, and the accuracy and reliability of stress fatigue analysis are improved; real-time multi-source data is collected and model parameters are dynamically corrected using machine learning, ensuring the self-calibration capability of the digital twin model, making the simulation results more consistent with the actual operating environment, and reducing the prediction deviation caused by material degradation and environmental factors; based on the stress distribution analysis and fatigue crack propagation prediction of the updated model, a comprehensive fatigue failure risk assessment is generated, including crack propagation rate, remaining life and failure probability, effectively supporting predictive maintenance and reducing the risk of sudden accidents; finally, alarm information or optimization decision information is output, realizing the whole life cycle management, helping to optimize the maintenance strategy, prolonging the service life of the equipment, and improving the operation safety and economic benefits of the natural gas station. The method integrates advanced sensing, artificial intelligence and digital twin technology, solves the detection bottleneck and data fusion problem in the prior art, and promotes the intelligent transformation of the industry. BRIEF DESCRIPTION OF DRAWINGS

[0031] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not considered a limitation of the present application. Furthermore, like reference numerals are used to denote identical components throughout the drawings. In the drawings:

[0032] Figure 1 A flowchart of a stress fatigue numerical analysis method for an elbow tee of a natural gas station is provided for an embodiment of the present application;

[0033] Figure 2 A hybrid modeling flowchart is provided for an embodiment of the present application;

[0034] Figure 3 A correction mechanism working principle is provided for an embodiment of the present application;

[0035] Figure 4 A simplified strategy selection logic flowchart is provided for an embodiment of the present application;

[0036] Figure 5 An implementation flowchart of cloud digital twin updating is provided for an embodiment of the present application;

[0037] Figure 6 This is a flowchart of the real-time execution process of the STM32H7 chip provided in an embodiment of the present invention;

[0038] Figure 7 This is a flowchart of an intelligent work order system provided in an embodiment of the present invention;

[0039] Figure 8 This is a schematic diagram of a "3-layer 5-library" technical framework provided in an embodiment of the present invention;

[0040] Figure 9 This is a schematic diagram of the structure of a stress fatigue intelligent analysis system for a natural gas station bend tee provided in an embodiment of the present invention. Detailed Implementation

[0041] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0042] For the first aspect, see [link / reference] Figure 1 As shown in the figure, this invention proposes a numerical analysis method for stress fatigue of natural gas station bend tees, which is applied to a numerical analysis system for stress fatigue of natural gas station bend tees. The numerical analysis system can be executed by, but is not limited to, computer equipment with certain computing resources, such as personal computers (PCs, smartphones, personal digital assistants, or platform servers). This embodiment provides a numerical analysis method for stress fatigue of natural gas station bend tees. This method addresses the stress fatigue problem of natural gas station bend tees under high pressure, high temperature, and corrosive environments. By integrating finite element analysis (FEA), computational fluid dynamics (CFD), and artificial intelligence technologies, it achieves real-time monitoring and predictive analysis of fatigue damage. The specific steps are as follows:

[0043] Step S1: Based on finite element analysis and computational fluid dynamics, a multi-physics coupled digital twin model of the bend tee is established. The digital twin model integrates geometric structure, material properties, and fluid-structure interaction mechanism to simulate the interaction between flow-induced vibration stress and corrosion fatigue.

[0044] In natural gas stations, elbow tees serve as critical components that bear complex loads induced by high-pressure natural gas flow, including flow-induced vibration and corrosion fatigue interaction. Traditional methods rely on empirical design and regular maintenance, which cannot meet the challenges of high load and long-term operation. This method realizes accurate simulation of these interactions by establishing a multi-physical field coupling digital twin model. The model is based on finite element analysis (FEA) and computational fluid dynamics (CFD), integrating geometric structures (such as three-dimensional CAD models, including internal surface flow channels and external structural features), material properties (such as the elastic modulus, Poisson's ratio, and fatigue curve of domestic X80 steel), and fluid-structure coupling mechanisms.

[0045] Specifically, the process of establishing the model includes model preparation and initialization, fluid-structure coupling calculation, and output and verification. First, geometric modeling and meshing are performed: high-fidelity three-dimensional geometric models are established using CAD / CAE software (such as SolidWorks or ANSYS DesignModeler), hybrid meshing techniques (such as hexahedral + tetrahedral mesh) are used to divide the calculation domain, and the number of meshes is controlled at the million level to balance accuracy and computational resources. Boundary condition setting includes gas flow input / output boundaries such as pressure inlet (10-15 MPa fluctuation range), velocity outlet or wall boundary (no-slip condition), input fluid parameters (such as natural gas composition CH4 92%, H2S concentration 50 ppm), flow rate (3-10 m / s), and temperature, etc. based on SCADA system historical data.

[0046] The fluid dynamics equation is solved using the Navier-Stokes equation set (continuity equation and momentum equation), and the standard K-ε turbulence model or large eddy simulation (LES) is used to capture the turbulence effect. Run in transient mode to simulate fluid disturbance caused by random pressure fluctuations (such as 8 times per day cycle). Fluid-structure coupling calculation is achieved through software interface (such as the bidirectional coupling module of ANSYS Fluent and Mechanical): extract fluid load (such as fluctuating pressure distribution, velocity field) as FEA input, use conjugate heat transfer and structural dynamics method to map fluid load to structure mesh. Structural response analysis includes modal analysis and transient dynamics simulation, calculating the deformation, stress, and vibration frequency (20-500 Hz range) of the pipe wall under fluid load. Focus on solving the problem of multi-axial non-proportional loading (tension-torsion-bending combined load), using the critical plane method (such as Findley criterion) to evaluate fatigue damage.

[0047] The iterative coupling and convergence use a sequential coupling method: CFD calculation → output fluid load → FEA calculation of structural displacement → update CFD mesh → repeat iteration until convergence (convergence criteria such as maximum displacement change <1%). The output stress result extracts the von Mises stress field, identifies high stress areas (such as three-way welds), includes time domain and frequency domain data, and is used for fatigue analysis (such as rainflow counting input). Precision verification is through laboratory tests (such as fan simulation air circulation) or comparison with measured vibration data (error controlled within 10%). The model realizes efficient coupling through the OpenFOAM multi-field solver, reducing calculation time by 40% compared to the traditional ANSYS method, while quantifying the synergistic effect of chemical factors such as H2S environment on vibration (such as crack propagation rate increasing by 3-5 times).

[0048] Step S2, collecting real-time multi-source data of the elbow three-way pipe, the real-time multi-source data including strain, vibration, pressure, temperature and corrosion medium concentration data.

[0049] This step is realized by deploying an intelligent sensor network using an "end-edge-cloud" collaborative architecture. The perception layer (end) deploys sensors: fiber Bragg grating (FBG) measures local strain / temperature change (accuracy ±1με); distributed acoustic sensing (DAS) monitors pipeline vibration and leakage (spatial resolution 1m); phased array ultrasonic testing (PAUT) detects crack depth; multi-gas sensor monitors H2S, CO2 and other corrosion medium concentrations. Data collection is compatible with Modbus / OPC UA protocols, supports LoRaWAN / NB-IoT transmission, and is suitable for field environments. The edge computing layer (edge) uses STM32H7 chips or NVIDIA Jetson AGX for local preprocessing such as fast Fourier transform (FFT) and filtering, extracting feature values (such as spectral peak, root mean square value). Time series compression algorithm (TSC) is used to reduce bandwidth by 80%, solving the problem of heterogeneous multi-source data (SCADA 1Hz vs. fiber 10kHz), and ensuring spatial consistency through timestamp synchronization. The cloud platform (cloud) stores and preliminarily analyzes data, realizing big data fusion. The collection frequency is adjusted according to parameters: strain and temperature @100Hz, vibration @10kHz, pressure and corrosion medium @1min. This step ensures real-time data and supports full life cycle monitoring.

[0050] Step S3, updating the boundary conditions and parameters of the digital twin model using the real-time multi-source data, and dynamically correcting the material constants through machine learning to realize self-calibration of the digital twin model.

[0051] The specific process includes data collection and preprocessing, model calibration and updating, and closed-loop feedback and output. First, real-time data sources are deployed: fiber optic sensors (FBG / DAS) monitor strain, temperature, and vibration at key locations; wireless accelerometers cover the entire field; H2S and temperature sensors input environmental information. Edge preprocessing performs FFT and filtering. Then, in the cloud, compare the measured data with the model predicted values, and calculate the residual error (such as vibration frequency spectrum deviation exceeding 5%). Use LSTM neural networks or physical information models to dynamically update key parameters: modify the C and m coefficients of the Paris formula (based on the influence of H2S concentration and temperature), FEA damping coefficients, or geometric correction coefficients Y. The training strategy uses transfer learning to fine-tune local models from the ASME case library, and uses federated learning to aggregate multi-site data; input at least 3 years of historical data to train the error model. Optimization algorithms such as Bayesian optimization or gradient descent minimize residual errors, enabling real-time adjustments (edge layer 50ms level, cloud level hours). Reinforcement learning optimizes maintenance strategies, quantifying the benefits of pressure fluctuation control. Finally, the calibrated parameters are fed back to the digital twin system, triggering FEA+CFD recalculation, updating the remaining life prediction and crack location (accuracy ±0.05mm). The results are visualized through AR / VR interfaces to support decision-making. Continuous iteration establishes a blockchain record to store calibration history, ensuring data integrity. This step combines edge intelligence and cloud AI, reducing latency to seconds, improving detection rates, and reducing unplanned downtime.

[0052] Step S4, performing stress distribution analysis based on the updated digital twin model to obtain stress field data, and performing fatigue crack propagation prediction based on the stress field data and the real-time multi-source data to generate fatigue crack propagation prediction results.

[0053] This step uses a hybrid modeling system (Hybrid Physics-Informed ML) with a nonlinear physical model as the main trunk and a neural network for auxiliary correction. The core equation is an extended version of the Paris formula:

[0054]

[0055] where, is the crack propagation rate, a is the crack depth, N is the number of stress cycles, C and m are material constants, ΔK is the stress intensity factor amplitude, T is the environmental temperature, σ mean is the average stress, and pH is the pH value.

[0056] Stress distribution analysis derives the von Mises stress field from the updated model, identifying high stress areas. Crack propagation prediction inputs the rainflow-counted stress spectrum + material S-N curve correction coefficient, and outputs the failure probability curve (Weibull distribution). Parameter correction considers temperature T, pH value pH, average stress σ meanThe coupling effect of the crack growth parameters C, m. The LSTM network adjusts the C, m parameters in real-time (based on historical crack growth data).

[0057] Exponential term (ΔK) m Leading to nonlinearity (power-law growth for m = 3.2)

[0058] Environmental correction factor f(T,σ mean , pH) contains exponential function e Q / RT and power function pH 1.7 ; Machine learning dynamic correction: adjust C, m parameters in real-time through LSTM network (based on historical crack growth data);

[0059] Inputs: temperature T, vibration frequency spectrum, corrosion medium concentration; outputs: corrected material constants C*, m*. Hybrid modeling (Physics-Informed ML), as shown in Figure 2 The physical layer uses the Paris formula to obtain the basic crack growth prediction, and the data layer obtains the parameters C* and m* based on the LSTM dynamic correction, and the hybrid output Complete mathematical model expression Target output: crack growth rate and remaining life N remaining

[0060] Step-by-step representation: stress intensity factor amplitude (geometric nonlinearity):

[0061]

[0062] Y is the geometric correction coefficient, see API 579 Appendix F, Δσ is the stress amplitude, a is the crack depth, and the alternating load intensity index. Among them, the multi-physical field coupling correction:

[0063]

[0064] Where, τ max is the maximum shear stress (FEA calculation); σ eq is the equivalent stress (Mises criterion); Q is the activation energy (typical value of carbon steel 80 kJ / mol); R: gas constant (8.314 J / (mol·K)); ΔK is the basic term; is the multi-axial correction; f(T, pH) is the environmental correction.

[0065] Machine learning correction term:

[0066]

[0067] A rms is the effective value of vibration acceleration (g); is the H2S concentration (ppm); historical Historical fatigue crack growth rate, the change in crack length per cycle of loading.

[0068] Final crack growth equation:

[0069]

[0070] Integral form of remaining life:

[0071]

[0072] The model correction method includes:

[0073]

[0074] where C * is the dynamic correction, ΔK eff is the multi-field coupling, m * is the dynamic correction.

[0075] Table 1. Data sources for model correction:

[0076]

[0077] Dynamic correction parameters (LSTM network output):

[0078]

[0079] Multi-field coupling stress intensity factor:

[0080]

[0081] 1) Environmental correction factor, quantifying the synergistic effect of corrosion medium (H2S / CO2 / Cl-) and temperature:

[0082]

[0083] Q is the activation energy (typical value for carbon steel 80 kJ / mol); R is the gas constant (8.314 J / (mol·K));

[0084] 2) Multi-axial stress correction, addressing the problem of tensile-torsional-bending combined load:

[0085] Multi-axial correction:

[0086] τ max is the maximum shear stress (FEA calculation), σ eq is the equivalent stress (Mises criterion)

[0087] 3) Vibration stress additional term (activated when A rms > 0.3g)

[0088]

[0089] where Δσ vib is the calculated vibrational stress, A rms is the root mean square (RMS) value of the vibrational acceleration (g), and f is the dominant frequency (Hz). The correction mechanism works as shown in Figure 3 The correction trigger conditions are: sudden load, pressure fluctuation > 15% of the design value; environmental shock, temperature drop > 10℃ / h or H2S concentration doubled; crack acceleration, ultrasonic detection of weekly growth > 20%. S-N curve substitution method (simplified in the crack initiation stage), simplified content: when the crack size < 0.1mm (micro-defect stage), skip the fracture mechanics analysis, directly use the Basquin equation to calculate the fatigue life. Steps:

[0090] Confirm that there are no macro-defects (cracks < 0.1mm) by ultrasonic / eddy current detection; extract the stress amplitude Δσ (real-time monitored by FBG sensor); call the material-based S-N curve (Basquin equation):

[0091]

[0092] The original fatigue life formula is:

[0093] where C0, m0 are determined by GB / T 3075 test (e.g. X80 steel: C0 = 1.2 x 1012, m0 = 3.0). Applicable conditions: uniform stress field, non-corrosive environment, stable load spectrum. Basis: micro-crack propagation is dominated by material microstructure, and the fracture mechanics model is not sensitive to this scale (error < 5%).

[0094] Equivalent initial defect assumption, simplified content: equivalent material micro-unevenness (such as welding residual stress, inclusions) to initial crack size a0. Steps: defect assessment process based on API 579 Appendix 1; map equivalent a0 according to material type (domestic X70 steel: a0 = 0.05mm, imported X70: a0 = 0.03mm); directly as the lower limit of Paris formula integration, skip the initiation stage simulation. Basis: statistical distribution of micro-defects (Baosteel 2023 test: 95% of the test samples have crack sizes of 0.03-0.08mm).

[0095] Optimization of calculation amount

[0096] (1) 3D FEA dimension reduction simplification, simplification content: 3D entity model is reduced to 2D axisymmetric model. Steps: extract the symmetry axis of tee structure (such as branch pipe-main pipe axis); cut along the axis to establish a 2D cross-section model; map 3D boundary conditions (pressure, constraint); output cross-section stress field, extrapolate 3D stress according to API 579 formula:

[0097] σ 3D = 1.15σ 2D · κ

[0098] Wherein, κ is the geometric correction factor; K = 0.92 is used for 90° elbow tee.

[0099] Error control: compared with 1 billion grid 3D model, the maximum stress deviation is <15%, and the calculation time is reduced from 72 hours to 2 hours.

[0100] (2) CFD quasi-steady fluid-structure coupling simplification

[0101] Simplification content: transient fluid simulation → quasi-steady time-averaged flow field. Steps:

[0102] Obtain pressure fluctuation statistical characteristics (mean P avg , amplitude ΔP) through SCADA;

[0103] Use P avg as steady-state CFD input to calculate the basic flow field;

[0104] Superimpose the empirical formula of pulsating stress:

[0105]

[0106] Wherein, A rms is fitted from DAS vibration monitoring data. Applicable conditions: Reynolds number Re <1e5 (typical flow velocity of natural gas station 3-10 m / s, Re = 5e4 ~ 2e5), error <8%.

[0107] (3) Load spectrum segment number simplification

[0108] Simplification content: full random load spectrum → 5-segment simplified spectrum, steps:

[0109] Rainflow counting method is used to count stress amplitude distribution;

[0110] The highest amplitude Δσ max , the lowest amplitude Δσ min , and the mean value are retained;

[0111] According to the principle of equivalent cumulative damage, the intermediate amplitudes are combined into 2 characteristic blocks;

[0112] Table 2. Generate 5-segment step spectrum (example):

[0113]

[0114]

[0115] Error: Compared to full spectrum (>100 segments), cumulative damage error <20%, calculation speed improved 50 times.

[0116] Simplified strategy includes S-N curve instead of micro-defect stage, equivalent initial defect assumption (a0=0.05mm for X70 steel), three-dimensional FEA reduced to two-dimensional axisymmetric model (error <15%), CFD quasi-steady state simplification (error <8%), load spectrum simplified to 5-step spectrum (damage error <20%), which realizes sub-millimeter crack prediction.

[0117] Step S5, based on the fatigue crack propagation prediction results, evaluate the fatigue failure risk of the elbow tee, including crack propagation rate, residual life assessment and failure probability, and output alarm information or optimization decision information according to the fatigue failure risk.

[0118] Please refer to Figure 4 Simplified strategy selection logic flow chart.

[0119] Empirical formula derivation:

[0120] 1. Corrosion fatigue life formula (based on NACE TM0177 data regression):

[0121]

[0122] Where, N air Air environment life; Hydrogen sulfide concentration (ppm).

[0123] 2. Vibration-fatigue coupling formula (field measured data fitting):

[0124]

[0125] Where, A rms Vibration acceleration effective value (g), f is the dominant frequency (Hz)

[0126] Table 3. Performance requirement evaluation:

[0127]

[0128] Table 4. Symbol system comparison table:

[0129]

[0130] Please refer to Figure 5Implementation flowchart for updating the cloud digital twin.

[0131] Case background, project name: West-East Gas Transmission Project, some booster station outlet tee pipe operating parameters: design pressure: 10 MPa, actual fluctuation range: 6-9.5 MPa (8 times per day cycle); medium composition: CH4 (92%) + H2S (50 ppm) + CO2 (2%); temperature: -20℃-50℃ (annual temperature difference); material: L485M (X70) steel pipe, weld meets API 1104 standard Find problems: ultrasonic testing found 0.8mm deep surface cracks at the branch pipe weld.

[0132] Table 5. Input parameters and data, basic parameter table:

[0133]

[0134]

[0135] Step-by-step calculation process:

[0136] 1. Stress intensity factor amplitude calculation:

[0137]

[0138] 2. Crack propagation rate calculation (after correction):

[0139]

[0140] 3. Residual life estimation:

[0141]

[0142] Consider 8 times per day cycle:

[0143]

[0144] Table 6. Comparison and verification of calculation results:

[0145] Method Remaining life prediction Error source analysis Paris formula (standard) 9.2 years H2S corrosion and temperature fluctuations are not considered Modified model (this case) 6.3 years Including environmental factors and multi-axial stress correction Actual failure time 5.8 years Unforeseen additional stress of soil settlement

[0146] Error analysis: The model predicted failure 0.5 years earlier than the actual failure (conservatism 8.6%), which meets the engineering safety margin requirements. The main error comes from the unmodeled soil load (post-FEA shows that the additional stress is about 22 MPa). This case realizes the following through a multi-parameter coupled model: an improvement in prediction accuracy of 42% compared to the traditional method (compared to the standard Paris formula); a clear contribution of H2S corrosion accounting for 35% of the loss of life; and a quantification of the benefits of pressure fluctuation control (136 days of life extension per 1 MPa reduction). This step realizes full life cycle optimization, reduces maintenance costs, and reduces unplanned downtime. The method of this embodiment solves the problems of standard lag, experimental limitations, and evaluation bottlenecks in the prior art, improves the accuracy and efficiency of fatigue management, and is suitable for high-load, long-period running natural gas stations.

[0147] In specific implementation, multi-physical field coupling modeling: fluid-structure coupling vibration (CFD-FSI) and corrosion-fatigue synergistic effect are integrated into a unified model to quantify the environmental-mechanical interaction through coupled variables. A soil-pipe dynamic interaction module is introduced to solve the problem of additional stress caused by foundation settlement ignored by traditional models (such as the 22 MPa error not modeled in the case). The nonlinear interaction between high-pressure gas flow induced vibration (mechanical load), corrosion medium (chemical load), and fatigue damage (material response) is quantified, and the establishment of the computational model covers the application. The final model formula is the result of multi-physical field coupling. A multi-field solver based on Open FOAM is developed, and the calculation efficiency is improved by 40% (compared to traditional sequential solution).

[0148] Dynamic life prediction based on machine learning, hybrid physics-informed ML modeling is proposed: physical layer: Paris formula calculates basic crack propagation; data layer: LSTM network real-time corrects model parameters (such as dynamically adjusting C, m); training data comes from the digital twin history library of 20 stations (including 12,000 failure cases). Technical breakthrough: realizes sub-millimeter crack prediction (error

[0149] ±0.05mm), 6 months earlier than traditional ultrasonic detection. A special algorithm library "FatigueNet" is developed to support TensorFlow and ANSYS joint simulation.

[0150] Intelligent edge computing architecture, design hierarchical edge computing nodes: first-level node (pipe end): STM32H7 chip performs fast Fourier transform (FFT), completes vibration spectrum analysis within 10 ms. Second-level node (field station): NVIDIA Jetson AGX runs lightweight ML model, updates residual life in real time. Adopt federated learning, local training of each field station data, cloud aggregation of global model. Wide occupation is reduced by 70% (only feature parameters are transmitted instead of original waveforms). Through time series compression algorithm (TSC), storage demand is reduced by 80%.

[0151] Table 7. Digital twin full life cycle management, build five-dimensional twin model (geometry-physical-behavior-rules-service):

[0152]

[0153] Integrate SCADA, GIS and detection robot data flow; support VR / AR interactive operation and maintenance decision; develop blockchain storage system to ensure that detection data cannot be tampered with (comply with GB / T 38644-2020).

[0154] Full life cycle management process:

[0155] Stage 1: design and manufacturing (BIM collaboration); lightweight design based on topology optimization (15% weight reduction and 22% stress reduction); welding digital twin: real-time monitoring of heat-affected zone temperature field, prediction of residual stress (error <10%).

[0156] Stage 2: deployment and monitoring (IoT interconnection);

[0157] Table 8. Sensory network deployment:

[0158]

[0159]

[0160] Edge computing, see Figure 6 , for STM32H7 chip real-time execution flowchart:

[0161] Stage 3: operation and optimization (AI driven); dynamic life prediction closed loop:

[0162] 1) Data fusion: 6.3 years

[0163] SCADA pressure fluctuation + fiber strain → generate rainflow count matrix; ultrasonic crack depth + H2S concentration → input extended Paris formula:

[0164]

[0165] 2) Model Update: LSTM network retrain every 6 hours (input: latest sensor data + historical failure case library); output: remaining life N remaining and 95% confidence interval.

[0166] 3) Decision Optimization: Reinforcement learning model trade-off economic benefits (loss of production vs. maintenance cost):

[0167]

[0168] max: maximize objective function, optimize decision to achieve highest benefit.

[0169] Cumulative sum from current time (t=0) to end of life (t=T), considering long-term benefits over the entire life cycle. γ: discount factor, time value quantification: recent benefits are weighted higher, the smaller γ means more emphasis on current benefits. R safe : safe operation benefits, including: normal production benefits, accident avoidance benefits, equipment life extension value. C repair : maintenance cost, including: spare parts cost, labor cost, loss of production, opportunity cost.

[0170] Phase 4: Maintenance and Upgrade (blockchain notarization); please refer to Figure 7 for the intelligent work order system flowchart.

[0171] Simulate high-risk scenarios (such as leaking under pressure), feed failure analysis data to material library (such as S-N curve correction of X80 steel in H2S environment); realize 3D visualization of crack propagation, positioning accuracy ±2mm (ISO 9712 certification). Through reinforcement learning to optimize maintenance strategy, maintenance cost reduced by 45%.

[0172] Table 9. Technical comparison and advantages:

[0173] Traditional technology Innovation points of this program Benefit improvement Single field FEA analysis Multi-physical field coupling modeling Prediction accuracy ↑ 40% Manual periodic detection Edge intelligence real-time diagnosis Early warning time ↑ 300% Homogeneous material tee Gradient functional material design Life cycle cost ↓ 28% Empirical maintenance Digital twin optimization decision Unplanned downtime ↓ 90%

[0174] Preferably, the step of establishing the multi-physical field coupling digital twin model of the elbow tee includes: using computational fluid dynamics to solve the gas flow load and transmitting it to the finite element structure model through the fluid-solid coupling interface to calculate the vibration stress distribution under multi-axial non-proportional loading, while considering the coupling correction factors of the acceleration effect of the corrosion medium concentration on the crack propagation rate, fatigue damage accumulation and creep deformation. Specifically, the CFD solution uses the Navier-Stokes equation to capture the turbulent flow effect caused by high-pressure gas flow (10-15 MPa), and extracts the pulsating pressure as the FEA input. Load transfer is achieved through the fluid-solid coupling interface (such as the two-way module of ANSYS), FEA calculates the vibration stress distribution, and Findley criterion is used to handle multi-axial loading (tension-torsion-bending). The coupling correction factors include corrosion acceleration (H2S increases the propagation rate by 3-5 times, which is modified by the Paris formula C and m), fatigue accumulation (Miner rule is extended to nonlinear accumulation), and creep deformation (Norton-Bailey model is integrated, considering the creep strain at a temperature of 60-80°C). This preferred mode solves the corrosion-fatigue-creep multi-mechanism coupling challenge, and the data gap such as da / dN-ΔK test in the pH=3-5 range is supplemented by historical data, which improves the simulation accuracy of the model under complex working conditions.

[0175] Preferably, in the step of updating the boundary conditions and parameters of the digital twin model using the real-time multi-source data, the machine learning dynamically corrects the material constants, which includes: using a long short-term memory network to adjust the material constants C and m of the Paris formula based on historical crack propagation data and environmental parameters, where the Paris formula is an extended version:

[0176]

[0177] where, is the crack propagation rate, a is the crack depth, N is the number of stress cycles, C and m are material constants, ΔK is the stress intensity factor amplitude, T is the environmental temperature, σ mean is the average stress, and pH is the pH value. The specific process is: the LSTM network inputs historical crack propagation data (at least 3 years of records, including da / dN measured values) and environmental parameters (such as temperature T, H2S concentration, pH value), and outputs the corrected C and m. Training uses transfer learning, fine-tunes the pre-trained model from the ASME case library to the site-specific model, and aggregates multi-site data using federated learning. Correction mechanism: calculate the residual error minimize the error by gradient descent. Example: when the H2S concentration is 50 ppm, C increases by 20%, and m is adjusted to 3.5. This preferred mode solves the conservatism of the Paris formula to random loads and environmental effects, reducing the cumulative error by 30%-60%.

[0178] Preferably, the real-time multi-source data is collected by a sensor network deployed on the elbow tee, which includes fiber bragg grating sensors monitoring strain and temperature, distributed acoustic sensing monitoring vibration, phased array ultrasonic testing detecting crack depth, and multi-gas sensors monitoring corrosion medium concentration. Specifically, the FBG fiber array is fixed on the tee weld surface to monitor micro-strain (±1με) and temperature (accuracy 0.1℃); the DAS fiber is laid along the pipeline to capture vibration signals (resolution 1m); the PAUT ultrasonic probe is embedded in the high stress area to detect 0.1mm cracks (full matrix capture FMC + AI image recognition); and the multi-gas sensor (such as electrochemical type) is placed on the inner wall of the medium flow channel to monitor H2S (0.1ppm accuracy) and CO2. The network supports wireless transmission (LoRaWAN), battery life 5 years+, explosion-proof level Ex dⅡC T6, and adapts to -40℃~120℃ environment. This preferred mode overcomes the detection bottleneck and improves the micro-crack detection rate by 40%.

[0179] Preferably, the sensor network includes an edge computing node that performs real-time preprocessing on the real-time multi-source data, including Fourier transform and time series compression. Specifically, the first-level edge node (STM32H7 chip) performs FFT analysis of vibration spectrum (extracts peak value, root mean square), and the second-level node (NVIDIA Jetson AGX) runs a lightweight ML model. The time series compression (TSC) algorithm reduces the data volume by 80% and filters out noise. The preprocessing delay is <50ms, supporting timestamp synchronization of multi-source data. This preferred mode reduces cloud load by 70% and realizes real-time early warning.

[0180] Preferably, before the stress distribution analysis step based on the updated digital twin model, it further includes calculating the residual error between the digital twin model prediction value and the actual measured data, and adjusting and optimizing the model parameters including damping coefficient and geometric correction coefficient based on the residual error through closed-loop feedback. Specifically, compare the FEA predicted stress / vibration with the measured data to calculate the residual error. The closed-loop feedback uses Bayesian optimization to adjust the damping coefficient (adjust when the measured deviation is more than 20%) and the geometric correction Y (based on API 579). Iterate until the residual error is <1%. This preferred mode ensures model self-calibration and improves accuracy.

[0181] Preferably, the fatigue crack propagation prediction adopts an equivalent initial defect assumption to simplify micro-defect stage calculation, and simplifies random load spectrum into a finite segment step spectrum through rainflow counting method to optimize calculation efficiency, while considering environmental correction factor to quantify the synergistic effect of temperature and corrosion medium. Specifically: equivalent a0=0.05mm(X70 steel), skip initiation stage; rainflow counting is simplified into 5 segment spectrum (peak value 85MPa accounts for 5%, etc.), error <20%; environmental factor quantifies synergy (such as temperature rise, crack rate increases 3 times). The preferred way to improve the calculation efficiency by 50 times.

[0182] Preferably, the optimization decision information includes residual life prediction curve, crack initiation position visualization and maintenance strategy suggestion, wherein the maintenance strategy optimizes the whole life cycle by weighing safety benefits and maintenance costs through a reinforcement learning model. Specifically: the residual life curve is based on Weibull distribution; the crack position is 3D visualized (±2mm); the reinforcement learning model weighs economic benefits (loss of production vs maintenance cost):

[0183]

[0184] max: maximize the objective function, optimize the decision to achieve the highest benefit.

[0185] Cumulative summation from the current time (t=0) to the end of the life cycle (t=T), considering the long-term benefits in the whole life cycle.

[0186] γ: discount factor, time value quantification: recent benefits are weighted more, the smaller γ means more emphasis on current benefits.

[0187] R safe : safe operation benefits, including: normal production benefits, accident avoidance benefits, and equipment life extension value.

[0188] C repair : maintenance cost, including: spare parts cost, labor cost, downtime loss, and opportunity cost. Optimize the maintenance cycle. This preferred way reduces the cost by 45% and reduces downtime by 90%.

[0189] Second aspect, please refer to Figure 9 , the embodiment provides a stress fatigue intelligent detection and digital analysis system for a natural gas station elbow tee joint. The system adopts an "end-edge-cloud" collaborative architecture and a "3-layer 5-library" technical framework as shown in Figure 8 , to realize full life cycle fatigue management of the elbow tee joint. The system includes a multi-physical field coupling modeling module, a data acquisition module, a model updating and correction module, a fatigue analysis module, and a risk detection and output module. The embodiment has been described in detail in the foregoing embodiments.

[0190] A multi-physics coupling modeling module for establishing a multi-physics coupling digital twin model of the elbow tee based on finite element analysis and computational fluid dynamics, the digital twin model integrating geometric structure, material properties, and fluid-structure coupling mechanisms to simulate the interaction of flow-induced vibration stress and corrosion fatigue. Specifically, this module uses CAD / CAE software to establish a high-fidelity three-dimensional geometric model, including weld details and flow characteristics, and uses mixed mesh division to calculate the domain. Boundary conditions are set based on SCADA historical data, such as pressure 10-15 MPa fluctuations and natural gas composition (CH4 92%, H2S 50 ppm). CFD solves the Navier-Stokes equation, using the K-ε turbulence model to capture fluid disturbances, and transmits the pulsating pressure to the FEA structure model through a fluid-structure coupling interface (such as the ANSYS two-way module). FEA calculates the vibration stress under multi-axial non-proportional loading (20-500 Hz), and uses the Findley criterion to evaluate damage. The model integrates corrosion correction factors (such as H2S accelerating crack propagation by 3-5 times), fatigue accumulation (Miner's rule extension), and creep deformation (Norton-Bailey model) to solve the multi-mechanism coupling problem. Output stress field data to support OpenFOAM efficient solution, reducing calculation time by 40%.

[0191] A data acquisition module for acquiring real-time multi-source data of the elbow tee, including strain, vibration, pressure, temperature, and corrosion medium concentration data. This module deploys a sensor network: fiber Bragg grating (FBG) monitors strain (±1με) and temperature; distributed acoustic sensing (DAS) captures vibration (resolution 1m); phased array ultrasonic testing (PAUT) detects crack depth (0.1mm level, FMC+AI imaging); multi-gas sensor monitors H2S / CO2 concentration (0.1ppm precision). Supports LoRaWAN / NB-IoT transmission, compatible with Modbus / OPC UA protocols. Edge nodes (STM32H7 chips) perform preprocessing such as FFT spectrum analysis and TSC compression, reducing bandwidth by 80%. Acquisition frequency: strain / temperature @100Hz, vibration @10kHz, corrosion medium @1min, ensuring real-time data and heterogeneous fusion.

[0192] A model updating and correction module is configured to update boundary conditions and parameters of the digital twin model by using the real-time multi-source data, and dynamically correct material constants by machine learning to realize self-calibration of the digital twin model. The module calculates the deviation (e.g., vibration deviation exceeding 5%) between the predicted value and the measured data, and adjusts Paris formula C and m based on historical crack data and environmental parameters (e.g., T and pH) by using an LSTM network. Training strategies include transfer learning (ASME library fine-tuning) and federated learning (multi-site aggregation). An optimization algorithm (e.g., Bayesian) adjusts the damping coefficient and the geometric Y factor, and a closed-loop feedback triggers FEA / CFD recalculation. Calibration frequency: edge 50 ms, cloud hourly, supporting blockchain notarization, ensuring model accuracy improvement, error < 1%.

[0193] A fatigue analysis module is configured to perform stress distribution analysis based on the updated digital twin model to obtain stress field data, and perform fatigue crack propagation prediction based on the stress field data and the real-time multi-source data to generate fatigue crack propagation prediction results. The module derives the von Mises stress field from the model and identifies high stress areas (e.g., welds).

[0194] A risk detection and output module is configured to evaluate the fatigue failure risk of the elbow tee based on the fatigue crack propagation prediction results, including crack propagation rate, remaining life assessment and failure probability, and output alarm information or optimization decision information according to the fatigue failure risk. The module quantifies the risk: crack rate da / dN, remaining life (Weibull distribution) and failure probability (considering H2S contribution of 35%). Alarm thresholds such as crack growth > 20% trigger a 6-month early warning, and crack location (±2mm) is visualized through AR / VR. Optimization decisions include life curve, maintenance recommendations, and trade-offs between benefits and costs. Full life cycle optimization reduces maintenance costs by 45% and downtime by 90%. The system of the embodiment integrates IoT, AI and digital twin, solves the existing detection bottleneck, improves accuracy by 40%, and is suitable for stations such as the West-East Gas Transmission Project.

[0195] In a third aspect, the embodiment of the present application provides a computer program product, including a non-transitory computer readable storage medium, and the storage medium stores computer executable instructions. When the instructions are executed by a processor, the processor executes the method provided by the above-mentioned embodiment.

[0196] The present application also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned wellhead device and station process detection and evaluation method provided by the present application. The computer program product should be understood as a software product that mainly realizes the solution of the present application through a computer program, such as a program product integrated in the cloud or a software library.

[0197] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", "one implementation", "a preferred implementation" or "some examples" etc. means that the particular feature, structure, material or characteristic being described in connection with the embodiment or example is included in at least one embodiment or example of the application. The illustrative appearances of the above-mentioned terms in various places in the specification are not necessarily intended to refer to the same embodiment or example. Moreover, the particular features, structures, materials or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0198] Although embodiments of the present application have been shown and described, it would be appreciated by those skilled in the art that changes, modifications, alternatives and variations to these embodiments could be made without departing from the principles and spirit of the application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for stress fatigue numerical analysis of a natural gas station elbow tee, characterized in that, The method comprises the following steps: establishing a multi-physics coupled digital twin model of the elbow tee based on finite element analysis and computational fluid dynamics, the digital twin model integrating geometric structure, material properties, and fluid-structure coupling mechanisms to simulate the interaction of flow-induced vibration stress and corrosion fatigue; collecting real-time multi-source data of the elbow tee, including strain, vibration, pressure, temperature, and corrosion medium concentration data; updating the boundary conditions and parameters of the digital twin model using the real-time multi-source data, and dynamically correcting material constants through machine learning to achieve self-calibration of the digital twin model; based on the updated digital twin model, performing stress distribution analysis to obtain stress field data, and based on the stress field data and the real-time multi-source data, predicting fatigue crack propagation to generate fatigue crack propagation prediction results; based on the fatigue crack propagation prediction results, evaluating the fatigue failure risk of the elbow tee, including crack propagation rate, remaining life assessment, and failure probability, and outputting alarm information or optimization decision information according to the fatigue failure risk.

2. The method of claim 1, wherein, The step of establishing the multi-physics coupled digital twin model of the elbow tee includes: using computational fluid dynamics to solve the gas flow load and passing it to the finite element structure model through the fluid-structure coupling interface, calculating the vibration stress distribution under multi-axial non-proportional loading, while integrating the coupling correction factors considering the acceleration effect of corrosion medium concentration on crack propagation rate, fatigue damage accumulation, and creep deformation.

3. The method of claim 2, wherein, In the step of updating the boundary conditions and parameters of the digital twin model using the real-time multi-source data, the machine learning dynamically corrects the material constants, including: using a long short-term memory network to adjust the material constants C and m of the Paris formula based on historical crack propagation data and environmental parameters, where the Paris formula is: wherein, is the crack growth rate, a is the crack depth, N is the number of stress cycles, C and m are material constants, ΔK is the stress intensity factor amplitude, T is the environmental temperature, σ mean is the average stress, and pH is the pH value.

4. The method of claim 3, wherein, The real-time multi-source data is collected by a sensor network deployed on the elbow tee, which includes: fiber Bragg grating sensors to monitor strain and temperature, distributed acoustic sensing to monitor vibration, ultrasonic phased array to detect crack depth, and multi-gas sensors to monitor corrosion medium concentration.

5. The method of claim 4, wherein, The sensor network includes an edge computing node that performs real-time preprocessing on the real-time multi-source data, including Fourier transform and time series compression.

6. The method of claim 5, wherein, Before the step of performing stress distribution analysis based on the updated digital twin model, it also includes: calculating the residual error between the predicted value of the digital twin model and the actual measured data, and adjusting the model parameters including damping coefficient and geometric correction coefficient through closed-loop feedback based on the residual error.

7. The method of claim 6, wherein, The fatigue crack propagation prediction adopts the equivalent initial defect assumption to simplify the micro-defect stage calculation, and uses the rainflow counting method to simplify the random load spectrum into a finite segment step spectrum to optimize the calculation efficiency, while considering the environmental correction factor to quantify the synergistic effect of temperature and corrosion medium.

8. The method of claim 7, wherein, The optimization decision information includes residual life prediction curve, crack initiation position visualization, and maintenance strategy suggestion, where the maintenance strategy is optimized through a reinforcement learning model to balance safety benefits and maintenance costs throughout the life cycle.

9. A stress fatigue intelligent detection and digital analysis system for a natural gas station elbow tee, characterized in that, The method comprises the following steps: a multi-physics coupling modeling module configured to establish a multi-physics coupling digital twin model of the elbow tee based on finite element analysis and computational fluid dynamics, the digital twin model integrating geometric structure, material properties, and fluid-structure coupling mechanisms to simulate the interaction of flow-induced vibration stress and corrosion fatigue; a data acquisition module configured to acquire real-time multi-source data of the elbow tee, the real-time multi-source data including strain, vibration, pressure, temperature, and corrosion medium concentration data; a model updating and correction module configured to update boundary conditions and parameters of the digital twin model using the real-time multi-source data, and dynamically correct material constants through machine learning to achieve self-calibration of the digital twin model; a fatigue analysis module configured to perform stress distribution analysis based on the updated digital twin model to obtain stress field data, and perform fatigue crack propagation prediction based on the stress field data and the real-time multi-source data, to generate fatigue crack propagation prediction results; a risk detection and output module configured to evaluate fatigue failure risks of the elbow tee based on the fatigue crack propagation prediction results, the fatigue failure risks including crack propagation rate, remaining life assessment, and failure probability, and output alarm information or optimization decision information according to the fatigue failure risks.

10. A computer program product, characterised in that, A non-transitory computer readable storage medium, having stored therein computer executable instructions, which when executed by a processor, cause the processor to perform the method of any one of claims 1 to 8.

Images