Method and system for modeling burst disc lifetime based on material degradation

Abstract

The application discloses a burst disc life modeling method and system based on material degradation, and belongs to the technical field of computer-aided engineering and pressure vessel safety. The method comprises the following steps: establishing an initial constitutive model of an in-service burst disc and a material fingerprint library; constructing a real-time sensing field of multi-source service environment parameters and burst disc state parameters; establishing a material property degradation model coupled by a chemical corrosion thinning sub-model, a high-temperature microstructure softening sub-model and a fatigue damage accumulation sub-model; constructing a dynamic mapping equation of the material microscopic degradation quantity and the macroscopic burst pressure deviation; realizing dynamic prediction and reliability evaluation of the remaining life by using a particle filtering algorithm; and adopting an end-cloud collaborative architecture. The application solves the technical problem that the prediction accuracy is limited due to the disconnection between macroscopic monitoring and microscopic mechanism, and realizes quantitative mapping and dynamic sensing of the microscopic degradation and the macroscopic failure.

Description

Technical Field

[0001] This invention belongs to the field of computer-aided engineering and pressure vessel safety technology, specifically relating to a method and system for modeling the life of rupture discs based on material degradation. Background Technology

[0002] In chemical production, nuclear energy utilization, and aerospace propulsion systems involving high-temperature, high-pressure, flammable, explosive, or highly corrosive media, rupture discs, as a critical non-re-closed overpressure relief safety device, serve as the last line of defense for ensuring the structural integrity and operational safety of the entire system. When the pressure inside the protected system rises rapidly due to abnormal operating conditions and reaches a preset threshold, the rupture disc ruptures instantaneously through pre-designed structural instability or strength failure, thereby establishing a low-resistance relief channel. This allows the system pressure to be released rapidly, preventing catastrophic equipment explosions or radioactive material leaks caused by overpressure.

[0003] As industrial technology advances towards larger scale, integration, and extreme applications, the actual service conditions of rupture discs are becoming increasingly complex, often involving the coupled effects of cyclical alternating loads, high-low temperature cycles, and corrosion from chemically active media. During this long-term service, the metals or composite materials used to manufacture the rupture discs undergo irreversible performance evolution. When the inherent properties of the materials experience this irreversible degradation, the actual burst pressure of the rupture disc will gradually deviate from the initial design calibration value, potentially leading to serious problems such as premature bursting or delayed failure. This not only results in unplanned shutdowns and production reductions but also poses a profound threat to the intrinsic safety of industry.

[0004] Chinese patent CN120470906A discloses a method for predicting the life of a rupture disc based on intelligent sensing. This method uses a multimodal intelligent sensor array to collect pressure, temperature, strain, and acoustic emission signals during the disc's operation in real time. After extracting features using a deep learning model, it calculates the damage amount and predicts the remaining life based on fatigue damage accumulation theory. However, in complex service environments, material failure is often not solely caused by external load cycles, but is dominated by factors such as grain boundary corrosion and stress corrosion cracking induced by the environmental medium. Methods based on external monitoring are insufficient to effectively characterize the inherent embrittlement or softening of materials due to changes in their physicochemical properties. This results in a theoretically insurmountable upper limit to the accuracy of the model when predicting failure behavior driven by the evolution of material properties.

[0005] The invention disclosed in CN115165582B is a method for assessing the remaining life of rupture membranes in large-diameter vacuum systems of nuclear power plants. It obtains the remaining life by performing metallic and non-metallic material condition testing on rupture membrane fragments, combined with fatigue testing. While this method can access the microscopic state of the material, it faces significant challenges in engineering practice due to insufficient real-time performance and a lack of dynamic mapping. Because the material degradation of rupture membranes is a dynamic process that evolves nonlinearly over time, offline, fragmented sampling inspections cannot capture the evolutionary trajectory of performance degradation during service in real time.

[0006] Existing technologies for predicting the lifespan of rupture discs generally suffer from systemic defects such as insufficient depth of mechanistic modeling, decoupling of external monitoring and internal evolution, and lack of dynamic quantitative correlation. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for modeling the life of rupture discs based on material degradation, which solves the technical problems in existing rupture disc life prediction technologies, such as the disconnect between macroscopic monitoring and microscopic mechanisms, the inability to quantitatively and dynamically map the relationship between material degradation and burst pressure offset in real time, resulting in limited prediction accuracy and lack of dynamic monitoring.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: The fragment lifetime modeling method based on material degradation includes the following steps: Step 1: Establish the initial constitutive model and material fingerprint library of the in-service rupture disc. By conducting material physicochemical analysis on standard specimens from the same batch as the in-service rupture disc, obtain the microstructure parameters and mechanical constitutive parameters in the initial state. Step 2: Construct a real-time sensing field of multi-source service environment parameters and rupture disc state parameters. Deploy sensor arrays at the rupture disc installation site and its upstream and downstream sides to collect environmental stress data and rupture disc state monitoring data in real time during the rupture disc's service process. Step 3: Establish a material property degradation model based on multi-physics coupling. The degradation model is composed of a chemical corrosion thinning sub-model, a high-temperature microstructure softening sub-model, and a fatigue damage accumulation sub-model through coupling terms. It is used to calculate the real-time damage variables of the rupture disc material during service. Step 4: Construct a dynamic mapping equation between the material's microscopic degradation and the macroscopic burst pressure offset. Based on the geometric topology of the rupture disc and its mechanical instability criteria, use the wall thickness reduction, microstructure softening parameters, and fatigue damage factors obtained in Step 3 as inputs to establish a real-time burst pressure prediction model to characterize the evolution trajectory of macroscopic burst capability. Step 5: Achieve dynamic prediction and reliability assessment of the remaining life of the rupture disc. Set the failure criteria for the rupture disc, use the particle filter algorithm, take the degradation model established in Step 3 as the state transition equation, take the rupture disc state monitoring data collected in Step 2 as the observation value, update the life distribution function in real time, and calculate the remaining life based on the updated state variables using the dynamic mapping equation in Step 4.

[0009] Furthermore, in step one, the microstructure parameters include grain size grade, phase composition ratio, average precipitate particle size, initial surface roughness, grain boundary characteristic parameters based on scanning electron microscope image recognition, and initial orientation distribution function obtained by electron backscatter diffraction technology; the mechanical constitutive parameters include initial yield strength, initial tensile strength, work hardening index, and elastic modulus.

[0010] Furthermore, in step two, the environmental stress data includes the instantaneous pressure of the fluid medium, the real-time temperature of the rupture disc surface, the conductivity of the medium, the pH value, and the stress cycle frequency; the rupture disc condition monitoring data includes the remaining wall thickness monitored in real time by an ultrasonic thickness sensor, the acoustic emission event count collected by an acoustic emission sensor, or the local strain measured by a strain gauge.

[0011] Furthermore, the chemical corrosion thinning sub-model described in step three calculates the real-time thinning amount of the effective wall thickness of the rupture disc using Faraday's law of electrolysis. The formula for calculating the real-time thinning amount is as follows: , Where M is the molar mass of the rupture disc material. The self-corrosion current density is corrected for real-time temperature and pH, z is the reaction charge number, F is the Faraday constant, and ρ is the material density.

[0012] Furthermore, the self-corrosion current density is obtained in real time by an electrochemical monitoring probe. The linear polarization resistance method is used to apply a small potential scan to the working electrode and measure the polarization resistance. The self-corrosion current density is equal to the Stern-Geary constant divided by the polarization resistance. The Stern-Geary constant is calibrated by electrochemical tests in the material fingerprint library.

[0013] Furthermore, for the rupture disc with a pre-grooved structure, the chemical corrosion thinning sub-model also introduces a local stress concentration correction factor, and the relationship between the thinning rate at the pre-grooved area and the thinning rate in the flat area is defined as follows: ; in For local stress at the bottom of the trench, Where is the nominal stress, and β and m are constants related to the corrosive activity of the medium.

[0014] Furthermore, the high-temperature tissue softening sub-model described in step three uses the Arrhenius equation to describe the decay of strength parameters with the thermal activation process, and the formula for calculating real-time tensile strength is: ; in Let A be the initial tensile strength, t be the material-specific frequency factor, Q be the time, and R be the activation energy for microstructure evolution. This is the real-time temperature.

[0015] Furthermore, when calculating the strength decay, the coarsening rate of the precipitated phase is calculated using the Lifshitz-Slyozov-Wagner theory, and the frequency factor A is corrected accordingly. If the service environment is a low-temperature environment, the system automatically switches to the low-temperature embrittlement monitoring mode. By calculating the material's ductile-brittle transition temperature shift and the impact energy evolution process, the nonlinear correction index of the kinetic mapping equation in step four is corrected.

[0016] Furthermore, in step three, the fatigue damage accumulation sub-model uses the rainflow counting method to process the real-time acquired pressure waveform, decomposing the variable amplitude load into an equivalent cyclic load. Combining this with the stress-life curves in the material fingerprint database, the fatigue damage factor is calculated using the Miner linear cumulative damage criterion. The formula for calculating the fatigue damage factor is as follows: ; in For the count of the i-th stress cycle, This represents the number of failure cycles obtained from the stress-life curve at the corresponding stress amplitude.

[0017] Furthermore, considering the amplifying effect of the corrosive environment on the stress concentration factor, the stress amplitude is corrected by a dynamic stress concentration factor. The dynamic stress concentration factor is updated in real time with the surface roughness evolution value calculated by the chemical corrosion thinning sub-model, and its functional relationship is calibrated by a pre-existing corrosion fatigue test.

[0018] Furthermore, the dynamic mapping equation described in step four is specifically set as follows: ; Where K is the structural shape coefficient. This is the initial wall thickness. To reduce the amount of material thinned in real time, α is the instantaneous radius of curvature that evolves with plastic deformation, and α is the nonlinear correction exponent for the contribution of material damage to strength.

[0019] Furthermore, the evolution of the instantaneous radius of curvature is updated by coupling geometric relationships through the Norton creep constitutive equation. First, the equivalent creep strain rate is calculated by the Norton equation, and the cumulative creep strain is obtained by integration. For a positive arch-shaped rupture disc, the arch height is updated in real time based on the geometric relationship and the approximate relationship between creep strain and arch height increment, and then the instantaneous radius of curvature is calculated.

[0020] When making remaining lifetime predictions, it is necessary to combine the evolution of the radius of curvature with the assumption of future pressure: within each prediction step, the current stress is calculated based on the assumed future pressure (steady-state extrapolation or stochastic process simulation), and the creep strain increment is calculated by substituting it into the Norton equation, thereby updating the radius of curvature and ensuring the closure of the state recursion.

[0021] Furthermore, step five specifically includes: The failure criterion is set as follows: the real-time burst pressure is lower than the minimum allowable working pressure of the system, or the deviation between the actual burst pressure and the initial calibration pressure exceeds a preset threshold. During the execution of the particle filtering algorithm, a particle swarm containing the effective thickness of the material, tensile strength and cumulative damage factor is initialized. The degradation model established in step three is used as the state transition equation, and the rupture fragment state monitoring data collected in step two is used as the observation value to construct the observation equation. The particle weights are adjusted according to the observation values, and the residual resampling method is used to overcome the particle scarcity problem. Calculate remaining lifetime ,in The predicted value is based on the posterior state estimate. The system sets a preset safety threshold and outputs a predicted remaining lifetime value with confidence intervals and a failure probability distribution curve. When predicting the remaining lifetime, the particles need to be pushed forward along the time axis, which requires knowing the service environment parameters (pressure, temperature, pH, etc.) at future moments. This invention uses one of the following two methods to predict future operating conditions: (1) Steady-state extrapolation: It is assumed that the environmental parameters at future times will remain unchanged from the current measurement values; (2) Stochastic process simulation: Based on historical service data, establish an autoregressive model (such as ARIMA) or Markov chain of environmental parameters, generate multiple possible future operating conditions through the Monte Carlo method, predict the life of each path separately, and finally obtain the probability distribution of the remaining life.

[0022] Furthermore, the management module continuously corrects the material fingerprint database through a deep learning neural network, taking service history data, material composition ratios, and macroscopic failure cases as inputs, and outputting correction coefficients in the kinetic equations.

[0023] In addition, the present invention also discloses a fragment lifetime modeling system based on material degradation for implementing the method described above, the system comprising: The underlying sensing module is located at the installation site of the rupture disc and is used to collect multi-physics field environmental stress signals and rupture disc status monitoring signals in real time during the service process of the rupture disc. The edge computing module is connected to the underlying sensing module and is used to process the collected signals locally in real time and run the material degradation mechanism kernel to perform initial life assessment and early warning. The cloud management module communicates with the edge computing module and is used to deploy the digital twin engine and the full lifecycle database to achieve global adaptive optimization and visual management of the model.

[0024] Furthermore, the underlying sensing module integrates an absolute pressure thin-film pressure sensor, a platinum resistance thermometer, an electrochemical monitoring probe based on a three-electrode system, and an ultrasonic thickness sensor. The working electrode and auxiliary electrode of the electrochemical monitoring probe are both made of the same metal material as the in-service rupture disc; the signal conditioning unit includes a high-impedance instrumentation amplifier array and a low-pass filter.

[0025] Furthermore, the edge computing module includes a microprocessor, a synchronous sampling controller, and an industrial Ethernet interface; The microprocessor has a built-in organic kernel, which encapsulates a chemical corrosion thinning sub-model, a high-temperature tissue softening sub-model, a fatigue damage accumulation sub-model, and a kinetic mapping equation. The microprocessor adopts a dual-core parallel processing mode, in which the first core is responsible for the fast Fourier transform analysis of the high-speed pressure waveform to extract the load spectrum features, and the second core is responsible for the numerical solution of the physical field coupling equation and the iterative update of the particle filter algorithm.

[0026] Furthermore, the digital twin engine in the cloud management module trains a proxy model based on the finite element analysis results, and corrects the model parameters in the edge computing module through residual feedback between the physical mechanism model and the data-driven model; the cloud management module also includes a visual early warning interface, which is used to display the probability distribution curve and confidence interval of the remaining life of the rupture disc in real time, and trigger a multi-level alarm mechanism according to the preset alarm threshold to automatically generate a maintenance recommendation form.

[0027] Furthermore, the construction of the real-time sensing field in step two also includes online quality assessment and adaptive calibration of the sensor data. The online quality assessment and adaptive calibration includes the following sub-steps: Sub-step 2a: Based on sensor redundancy configuration and physical correlation analysis, construct a sensor fault diagnosis model to detect and identify faulty sensors in real time; Sub-step 2b: For sensors diagnosed as faulty or with abnormal data, use the spatiotemporal correlation of multi-source heterogeneous sensor data to generate alternative data through a data reconstruction model to ensure the continuity of observation data; Sub-step 2c: To address the zero-point drift and sensitivity decay caused by long-term sensor operation, the sensor is automatically calibrated periodically using a built-in reference source or cross-calibration method, and the calibration parameters are updated to the edge computing module.

[0028] Furthermore, the sensor fault diagnosis model is constructed based on extended Kalman filtering or Bayesian networks. It utilizes the measurement redundancy of multiple similar sensors and the physical constraints between different types of sensors to calculate the residual sequence and set a dynamic threshold. When the residual exceeds the threshold, it is determined to be a sensor fault. The data reconstruction model adopts a prediction algorithm based on principal component analysis or long short-term memory neural networks. It uses historical data and data from currently healthy sensors to estimate the true value of the faulty sensor online.

[0029] Furthermore, the built-in reference source calibration includes: A sealed standard pressure chamber is set at the rupture disc mounting flange. The standard pressure source is connected regularly to calibrate the pressure sensor on site, and the deviation before and after calibration is recorded and the measured value is corrected. For electrochemical probes, periodic calibration is performed using an automated standard solution injection method to correct the Stern-Geary constant B and the correction function. .

[0030] Compared with the prior art, the present invention has the following beneficial effects: Existing technologies are limited to collecting external macroscopic parameters such as pressure and temperature, relying on data-driven models for fatigue accumulation calculations. Their fundamental flaw lies in simplifying the complex failure process to a response to external loads, failing to address the failure mechanisms caused by changes in the material's intrinsic properties. This invention thoroughly decomposes the failure process of a rupture disc to the most basic unit of materials science. By constructing a material property degradation model coupled with three sub-models—chemical corrosion thinning, high-temperature microstructure softening, and fatigue damage accumulation—this invention achieves real-time quantitative characterization of the evolution of the material's microstructure (such as dislocation movement, grain boundary corrosion, and precipitate coarsening). This transforms lifetime prediction from a correlational fitting of phenomena to a precise reproduction of the failure causal chain, solving the theoretical upper limit of prediction accuracy limitations caused by the lack of underlying mechanisms in traditional methods.

[0031] Existing technologies employ offline sampling inspection, which is essentially a discontinuous evaluation and cannot capture the nonlinear evolution trajectory of material performance degradation. This invention establishes a dynamic mapping equation incorporating key variables such as structural shape coefficients, instantaneous radius of curvature, and nonlinear correction exponents. This directly transforms microscale wall thickness reduction, strength attenuation, and fatigue damage factors into macroscale real-time burst pressure predictions through rigorous mechanical constitutive relations. Crucially, this mapping equation is not a static formula; rather, it dynamically corrects the geometric topology of the burst fragment (such as the instantaneous radius of curvature) by introducing the Norton creep constitutive equation and utilizes a particle filtering algorithm to continuously update the state vector and estimate the posterior probability. This enables the system to output a remaining lifetime probability distribution with confidence intervals in real time, upgrading traditional "static calibration" safety devices into intelligent terminals with "dynamic sensing" capabilities, providing a measurable and verifiable technical means for intrinsic industrial safety.

[0032] This invention incorporates a multiphysics coupling mechanism kernel within the edge computing module, responsible for localized rapid computation and initial early warning of high-frequency, real-time data, ensuring the real-time performance and reliability of predictions. The cloud management module deploys a digital twin engine and a full-lifecycle database, using deep learning neural networks to mine massive amounts of data from different service locations and reverse-correct key parameters in the mechanism model (such as the frequency factor and activation energy in the Arrhenius equation). The model can absorb experience from actual failure cases, achieving a leap from a "fixed model" to an "adaptive evolution model," significantly improving the universality and predictive confidence of the technical solution under different material systems and service conditions. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0034] Figure 1 This is an overall flowchart of the method described in this invention.

[0035] Figure 2 This is a simplified flowchart of the core method for modeling the lifespan of a rupture disc in this invention.

[0036] Figure 3 This describes the system-end-cloud collaborative workflow described in this invention.

[0037] Figure 4 This is one of the system operation interfaces described in this invention.

[0038] Figure 5This is the second system operation interface described in this invention.

[0039] Figure 6 This is the third type of system operation interface described in this invention.

[0040] Figure 7 This is the fourth type of system operation interface described in this invention. Detailed Implementation

[0041] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0042] The following is in conjunction with the appendix Figures 1-7 The embodiments of the present invention will be described in detail below.

[0043] This invention discloses a method and system for modeling the life of rupture discs based on material degradation. In specific engineering implementations, for the safety protection requirements of typical high-pressure hydrogenation reactors in the petrochemical industry, austenitic stainless steel conforming to GB / T 316L standard is selected as the main material of the rupture disc.

[0044] The first phase involves establishing an initial constitutive model and material fingerprint database for in-service rupture fragments.

[0045] Three standard specimens were randomly selected from the production batch for microstructure analysis. Grain boundary morphology was observed using a scanning electron microscope at an accelerating voltage of 20 kV, and the average grain size was determined to be 8.5 grade, with austenite content greater than 99.5% in the phase composition. The initial orientation distribution function of the material was obtained using electron backscatter diffraction technology, and an initial texture library was established.

[0046] In the mechanical performance testing phase, a universal testing machine equipped with a high and low temperature environment chamber was used to perform tensile tests at four temperature nodes: 25℃, 150℃, 300℃, and 450℃.

[0047] The initial yield strength is extracted by fitting the stress-strain curve of the engineering project. The initial tensile strength is 210 MPa. The initial work hardening index is 520 MPa. The value is 0.32. These multi-scale parameters together constitute the "digital identity" of this batch of materials, which is stored in the HDF5 format database of the material fingerprint library to provide an original benchmark for subsequent degradation assessment.

[0048] In the stage of constructing a real-time sensing field for multi-source service environment parameters, the bottom-level sensing module is installed at the vent flange of the reactor. This module integrates an absolute pressure thin-film pressure sensor with a range set from 0 to 10 MPa, a nonlinearity error controlled within 0.05% FS, and the ability to capture high-frequency pressure pulses. The temperature sensor uses a four-wire Pt1000 platinum resistance thermometer, wrapped with a Hastelloy C276 corrosion-resistant shielding layer, with a response time of less than 0.5 s. To monitor chemical corrosion, the electrochemical monitoring probe adopts a three-electrode system, with both the working electrode and auxiliary electrode made of 316L stainless steel, the same material as the rupture disc, and a saturated calomel electrode as the reference electrode.

[0049] The signal conditioning unit includes a high-impedance instrumentation amplifier array and a fourth-order low-pass filter with a cutoff frequency of 2kHz, ensuring that the common-mode rejection ratio (CMRR) of the original weak signal is greater than 110dB before being converted into a digital signal. Additionally, an ultrasonic thickness sensor (such as a piezoelectric ceramic wafer with a center frequency of 5MHz) is installed on or near the rupture disc for real-time monitoring of the remaining wall thickness. During sampling, a synchronous sampling controller ensures that the pressure signal is acquired at a frequency of 1.2kHz, temperature and electrochemical parameters at a frequency of 1Hz, and ultrasonic thickness data at a frequency of 0.1Hz.

[0050] A material property degradation model based on multiphysics coupling was established, and a chemical corrosion thinning sub-model was implemented to read the pH and conductivity of the medium in real time during operation. A microprocessor-controlled electrochemical interface circuit periodically (e.g., once per hour) applied a linear potential scan (±10 mV) to the working electrode, recording the current response and calculating the polarization resistance. And then based on Calculate the instantaneous self-corrosion current density, where The values ​​are obtained from the built-in correction table based on the real-time temperature and pH value (the correction table is pre-calibrated through accelerated aging tests).

[0051] Using Faraday's law of electrolysis, the effective wall thickness reduction is accumulated within each integration cycle (set to 1 hour). If the initial wall thickness of the rupture disc is... The thickness is 0.5mm, and after 1000 hours of operation, the calculated cumulative thinning amount may reach 0.015mm.

[0052] Meanwhile, the high-temperature tissue softening sub-model dynamically corrects the tensile strength using the Arrhenius equation for continuous operation at 450℃. (Frequency factor) and activation energy The microprocessor was pre-calibrated based on high-temperature tensile test data from a material fingerprint database. It not only considered the instantaneous effects of temperature but also calculated the weakening of the matrix strength due to the coarsening of the precipitated phase through the time-cumulative effect.

[0053] The microprocessor in the edge computing module executes a rainflow counting method on the collected pressure time sequence, decomposing the complex pressure fluctuations into a series of cycles with specific mean and amplitude, and converting them into corresponding stress amplitudes using a pre-calibrated stress conversion coefficient in the finite element method. By combining the SN curves in the material fingerprint database, the fatigue damage factor is calculated using the Miner linear cumulative damage criterion. ,in This represents the number of failure cycles at the corresponding stress amplitude. The damage increment generated by each pressure cycle is... Damage factors accumulate. The amplification effect of the corrosive environment on stress concentration is achieved through the dynamic stress concentration factor. accomplish, Surface roughness calculated using the chemical corrosion thinning sub-model Related, for example , The material sensitivity coefficient is determined by corrosion fatigue testing.

[0054] For a positively arched rupture disc, its stress state macroscopically manifests as a compression process on a thin spherical shell. Real-time burst pressure. In the calculation formula, the structural shape factor The initial determination is based on the arch height, thickness, and span ratio of the rupture disc.

[0055] Due to prolonged service under high temperature and pressure, rupture discs will undergo irreversible creep deformation.

[0056] In practical applications, the system integrates the Norton creep constitutive equation: ,in The creep strain rate, , , This is the material's creep constant. The microprocessor determines the creep constant based on the current stress. (by pressure) (and geometric estimation) and temperature The creep strain increment is calculated in real time and accumulated to obtain the creep strain. .

[0057] For a positively arched rupture fragment, assuming its spherical cap shape and initial radius of curvature... Arch height ,caliber satisfy Creep deformation leads to an increase in camber. Based on the assumption of constant volume and the simplification of thin-shell theory, it can be approximated as follows: .

[0058] Updated arch height Substituting into the above formula yields the current radius of curvature. .

[0059] For more complex structures, pre-established finite element simulations can be used. and The mapping table is updated in real time via lookup. It is important to note that in the remaining lifetime prediction stage (step five), in order to advance the evolution of the radius of curvature, assumptions about future pressures are necessary. This method uses steady-state extrapolation or stochastic process simulation to provide a sequence of future pressures, and then calculates the creep strain and radius of curvature at each future time point, achieving recursive prediction of the dynamic mapping equation.

[0060] For example, when the radius of curvature increases from the initial 50 mm to 52 mm due to creep, even if the tensile strength and wall thickness remain unchanged, the theoretical burst pressure will change significantly.

[0061] Dynamic prediction and reliability assessment of the remaining life of the rupture disc are achieved through a particle filter algorithm. During execution, the microprocessor first initializes 1000 particles, each representing a potential degradation trajectory state space (including wall thickness, strength, damage factor, etc.).

[0062] During the state transition phase, the established physical mechanism model is used to predict the state at the next moment. When new observational data (such as ultrasonic thickness measurements) are obtained... After inputting the data, the likelihood of each particle (i.e., the degree of matching between the observed and predicted values) is calculated, and the particle weights are adjusted according to the distribution of observation errors.

[0063] After residual resampling, the distribution of the particle cloud represents the optimal probability estimate of the current state of the rupture disc. Monte Carlo prediction is used to advance the particle cloud along the time axis until the failure criterion is met. During this advancement, assumptions about future environmental parameters are required. This embodiment uses steady-state extrapolation, assuming that parameters such as pressure, temperature, and pH remain unchanged from their current measurements. The failure criterion in this embodiment is strictly set as: predicted rupture pressure. The pressure drops to 5.2 MPa (initial calibrated burst pressure is 5.5 MPa, with a lower tolerance of -5%). Output the predicted remaining lifespan and its 95% confidence interval.

[0064] The cloud management module deploys a digital twin engine for rupture discs. It continuously calibrates the edge model using failure data from similar operating conditions stored globally in a full lifecycle database. The cloud management module employs convolutional neural networks to identify real-time image features of the rupture disc surface (if video monitoring is available) or indirectly derived corrosion morphology parameters, outputting correction coefficients for the Arrhenius equation. This cloud-edge collaborative architecture ensures high robustness of the lifecycle prediction model even in the event of sensor temperature drift or drastic environmental changes.

[0065] To facilitate a better understanding of the present invention by those skilled in the art, the present invention will be further described below with reference to specific embodiments.

[0066] Example 1: The object of treatment was a 316L positive arch rupture disc in a hydrocracking unit of a chemical plant.

[0067] calibrated burst pressure MPa, design wall thickness mm. Temperature fluctuation range 380℃-420℃, medium contains hydrogen sulfide and chloride ions, pressure accompanied by small amplitude pulsation of 0.5Hz.

[0068] The material degradation-based modeling system described in this invention has the following interface: Figures 4-7 As shown.

[0069] At 4000 hours of service, an abnormally high chloride ion concentration was detected, and the thinning rate calculated by the chemical corrosion thinning sub-model was reduced. mm / h increased to mm / h. Meanwhile, the high-temperature tissue softening sub-model detected a rapid decrease in material strength due to a 5% over-temperature.

[0070] System Decision: At hour 4200, the visual warning interface displays the remaining lifespan. With only 15 days remaining (95% confidence interval), immediate replacement is recommended.

[0071] Actual results: During the shutdown inspection at the 4210th hour, the measured wall thickness of the rupture disc was 0.542 mm (the model prediction value was 0.545 mm). The actual burst pressure measured by the offline burst test was 5.72 MPa, which is close to the safety threshold.

[0072] Comparative Example 1: The experimental subjects were from the same batch and under the same operating conditions as in Example 1. A traditional periodic replacement strategy (set to 6 months based on experience) and an overpressure alarm based on a pressure gauge were used. Due to the lack of microscopic degradation monitoring, the system did not issue any abnormal signals during its service life. At 4300 hours, the rupture disc prematurely and unplannedly detonated at 5.65 MPa (94% below its calibration value), causing an abnormal shutdown of the entire hydrogenation production line and resulting in significant economic losses.

[0073] Table 1 below summarizes the key performance indicators of Embodiment 1 and Comparative Example 1 in rupture disc management: Table 1: Comparison of performance data for predicting the lifespan of rupture discs;

[0074] Furthermore, as the hardware support for the system architecture of this invention, the microprocessor of the edge computing module adopts a dual-core parallel processing mode. Core A is specifically responsible for the real-time analysis of high-speed pressure waveforms, performing spectral analysis on the pressure fluctuation signal by calling the built-in FFT library functions. If abnormal high-frequency pulsations between 10Hz and 50Hz are detected (usually indicating internal leakage of upstream valves or fluid-induced vibration), Core A will immediately increase the weighting coefficient of the fatigue damage factor. Core B is responsible for the numerical solution of the physical field coupling equations, which uses an improved Euler method for time-domain step calculation. To prevent the numerical solution from diverging in the nonlinear region, an adaptive step size adjuster is set inside the microprocessor, which automatically shortens the calculation step size from 1 hour to 1 minute when the material strain rate exceeds the critical threshold.

[0075] In the design of the electrochemical monitoring probe for the underlying sensing module, to cope with the extreme physical environment inside the autoclave, the probe's insulation encapsulation adopts polyetheretherketone (PEEK) and ceramic sealing technology to ensure no media leakage occurs under 30MPa pressure. The sensor leads are led out through a specially designed high-temperature shielded cable, and a surge protector is connected to the front end of the signal conditioning unit to shield against electromagnetic interference generated when large motors start up on site.

[0076] For rupture discs with pre-grooved structures, this embodiment defines the local stress concentration correction factor in detail in step three. The thinning rate at the pre-grooved area... The effect of stress level is extremely significant.

[0077] Due to the multiaxial stress state at the bottom of the tank, its electrochemical activity is higher than that of the flat area. The system pre-calculates the stress concentration factor under different pressure loads using finite element analysis. When calculating the thinning amount, the model will... By introducing an exponential term, the model can accurately capture the physical process of premature failure of the pre-etched groove due to local preferential corrosion.

[0078] On the visual early warning interface of the cloud management module, the system not only presents the current health status score, but also dynamically generates a cloud map of the burst pressure degradation trajectory.

[0079] The cloud map consists of thousands of gray possible paths and a red mean path, with the failure boundary ( The path is marked as a horizontal red line on the horizontal axis. When the red path crosses the red line, the system automatically triggers a maintenance command. In addition, the interface integrates a geographic information system-based map of blasting fragment assets, facilitating centralized monitoring of safety devices in factories worldwide by large, cross-regional enterprise groups.

[0080] Furthermore, to optimize the efficiency of the particle filter algorithm and prevent the "particle scarcity" phenomenon, the microprocessor introduces an adaptive noise correction term during the resampling stage. When the residual between the observed and predicted values ​​exceeds the threshold for three consecutive cycles, the algorithm automatically increases the process noise variance in the state transition equation, forcing the particle swarm to diffuse into a wider space, thereby enhancing the system's ability to capture sudden operating conditions (such as instantaneous overheating caused by runaway reaction).

[0081] In the mapping equation, for the instantaneous radius of curvature The dynamic correction takes into account the anisotropic degradation of the rupture disc material. Due to the microstructure differences between the vertices and edges of the rupture disc caused by the cold forming process, the cloud-based digital twin engine assigns different degradation weights to different regions of the rupture disc based on the initial texture library data. In solving the macroscopic mechanical equations, a piecewise integration method is used, dividing the entire spherical surface of the rupture disc into 128 micro-elements. Degradation calculations are performed on each micro-element separately before assembling the structural stiffness matrix. Although this increases the computational load, the high-performance DSP support of the edge computing module still meets real-time requirements, and reduces the absolute error of the rupture pressure prediction from the traditional 15% to less than 3%.

[0082] This invention also provides an environmental uncertainty compensation algorithm for handling random drift in sensor signals. Specifically, after acquiring pressure data in step two, outliers are first removed using first-order differential detection, and then an adaptive Kalman filter is used to compensate for the pressure zero-point shift caused by temperature effects. This compensation process is based on a built-in pressure-temperature linkage experimental curve, ensuring that every pressure value input into the degradation model is a true, effective value after temperature correction.

[0083] For rupture discs operating in cryogenic environments (such as liquid nitrogen and liquid oxygen storage tanks), this invention automatically switches to a cryogenic embrittlement monitoring mode based on a high-temperature softening sub-model. In this mode, the model focuses on monitoring the shift of the material's ductile-brittle transition temperature with cumulative radiation (in the case of a nuclear power plant environment) or service time. By calculating the evolution of the material's impact energy, the nonlinear correction exponent in the mapping equation is corrected. This can prevent premature breakage failure caused by the material becoming brittle at low temperatures.

[0084] The cloud management module also integrates a global optimization logic based on federated learning. Edge computing modules distributed across different sites periodically upload encrypted local model parameters (rather than raw sensitive data). The cloud engine aggregates these parameters, identifies general degradation characteristics for specific media (such as high-concentration chlor-alkali), and distributes the optimized global weights to each service location. This mechanism enables newly installed rupture disc systems to quickly achieve high-precision life prediction capabilities even in the absence of historical service data, thanks to "collective intelligence."

[0085] Example 2: This example is a further optimization based on Example 1; in this example, in the fatigue damage accumulation sub-model of step three, the dynamic stress concentration factor is... The update depends not only on surface roughness Furthermore, factors influencing the morphology of corrosion pits were introduced.

[0086] The equivalent depth of corrosion pits can be estimated in real time by monitoring with online electrochemical impedance spectroscopy (EIS) and combining it with a pre-established corrosion morphology evolution model or electrochemical noise analysis. and equivalent width Alternatively, the growth pattern of corrosion pits calibrated offline can be used to estimate the current pit depth and width based on the accumulated corrosion charge, and then a correction relationship can be established by combining this with the stress distribution from the finite element pre-analysis. ; in, and The material sensitivity coefficient (dimensionless) is calibrated through corrosion fatigue testing. Surface roughness (unit: μm); Equivalent depth of corrosion pit (unit: μm); Equivalent width of corrosion pit (unit: μm); The activation energy (unit: J / mol) reflects the effect of temperature on the interaction between corrosion and fatigue. The gas constant is 8.314 J / (mol·K). The temperature is in real-time (in K). This formula reflects the coupled effect of corrosion pit morphology and temperature on fatigue stress concentration.

[0087] Frequency factor in the high-temperature tissue softening sub-model It is no longer a constant, but varies with the thickness of the corrosion product layer. Dynamic correction: ; in, Initial frequency factor (unit: s) - ¹); The coupling coefficient (dimensionless) was calibrated through a high-temperature corrosion test. The thickness of the corrosion product layer (unit: μm) is obtained by electrochemical impedance spectroscopy or online eddy current thickness measurement. The reference thickness (unit: μm) is set to 1 μm for normalization. The presence of the corrosion product layer alters heat conduction and interfacial reactions, thereby affecting the microstructure coarsening rate.

[0088] In the instantaneous radius of curvature update in step four, the stress increase caused by corrosion thinning, which accelerates creep deformation, is considered. This is solved iteratively. Based on current pressure Calculate nominal stress based on geometric dimensions (Unit: MPa); Based on the amount of thinning due to chemical corrosion Update the effective wall thickness and recalculate the local stress. (Unit: MPa) ; Will Substitute into Norton's creep equation to calculate the creep strain rate ,in , , Let be the material creep constant, and the cumulative creep strain is obtained by integration. This leads to the updating of the arch height and radius of curvature.

[0089] The aforementioned coupling mechanism ensures that the model can accurately reflect the material degradation process under multiple field interactions.

[0090] In the particle filter algorithm, the state transition equation simultaneously contains coupling terms from three sub-models, namely the state vector of each particle. The evolution depends on the current values ​​of corrosion, softening, and fatigue, and takes into account the cross-effects.

[0091] In addition to ultrasonic thickness measurement, the observation equation can also incorporate observed values ​​from electrochemical impedance spectroscopy (such as polarization resistance). As an auxiliary observation, it further improves the accuracy of state estimation. To reduce the polarization resistance... Incorporating particle filter observation updates requires establishing their functional relationship with state variables. This embodiment employs an empirical calibration method: establishing the polarization resistance through offline electrochemical testing. With corrosion rate relation ,in It is the Stern-Geary constant; while the corrosion rate With wall thickness reduction The derivative is directly related to it, that is Thus, it can As an indirect observation of the wall thickness reduction rate, an observation equation is constructed. ,in The wall thickness variation rate, To observe noise.

[0092] To verify the superiority of the coupling model of this invention, the following comparative examples are set: Comparative Example A: Based solely on the fatigue damage model, the load is monitored using only a pressure sensor, and the life is predicted based on the Miner linear accumulation criterion, without considering corrosion and high-temperature softening.

[0093] Comparative Example B: Based solely on the corrosion thinning model, the wall thickness reduction is calculated solely through electrochemical monitoring. Assuming that the material strength remains unchanged, the remaining wall thickness is substituted into the static formula to calculate the burst pressure, without considering fatigue and high-temperature softening.

[0094] Comparative Example C: The simple superposition model was run independently for corrosion, high-temperature softening, and fatigue sub-models, and then the three damage factors were simply multiplied together (i.e., To correct the burst pressure, without considering coupling interaction terms.

[0095] Comparative Example D: The deterministic model without particle filtering adopts the same coupling mechanism model as Example 2, but uses deterministic numerical integration instead of particle filtering updates, and directly calculates the remaining lifetime based solely on initial parameters and real-time sensor data.

[0096] Accelerated life tests were conducted on 316L austenitic stainless steel arch-shaped rupture discs (initial wall thickness 0.6 mm, rated burst pressure 6.0 MPa) from the same batch under simulated hydrogenation reactor conditions.

[0097] Operating parameters: Temperature fluctuation 400±20℃, pressure fluctuation amplitude ±0.3 MPa, frequency 0.5 Hz, medium containing 500ppm Cl - And 100 ppm H2S.

[0098] Online predictions were performed using the methods of Example 2 and Comparative Example AC, and the results were compared with those of actual destructive tests. The prediction results are shown in Table 2.

[0099] Table 2 Comparison of prediction results from different models:

[0100] Table 2 shows that: Comparative Example A, considering only fatigue damage, significantly overestimates the remaining life (error +38.4%) because it ignores the strength degradation caused by corrosion and high-temperature softening. Comparative Example B, considering only corrosion thinning, still has a large error (+13.8%) because it does not account for the accumulation of fatigue damage. Comparative Example C simply superimposes various damage factors, and because it ignores the coupling enhancement effect, its predicted life is conservative (error -14.1%), which may lead to unnecessary premature replacement. Comparative Example D uses a coupled model but without particle filtering; although the average error is small, it lacks uncertainty quantification and has poor adaptability to sudden operating conditions.

[0101] The coupled model in Example 2, combined with particle filtering, exhibits the smallest prediction error (-2.7%) and the highest accuracy in predicting burst pressure (+1.2%), demonstrating the synergistic advantages of multi-field coupling and dynamic updating. This example, through quantitative comparison, confirms the significant superiority of the proposed coupled modeling and particle filtering method in predicting burst fragment life, effectively addressing the issues of missing mechanisms and insufficient accuracy in existing technologies.

[0102] Example 3: This example is a further optimization based on Example 1 and Example 2. The core of this example is the multiphysics coupled material property degradation model described in step three. This model is not a simple superposition of three sub-models, but achieves interaction through coupling terms.

[0103] In practical implementation: Chemical corrosion-fatigue coupling term: Dynamic stress concentration factor in the fatigue damage accumulation sub-model. The update depends not only on surface roughness Furthermore, factors influencing the morphology of corrosion pits were introduced, and the depth of corrosion pits obtained through electrochemical monitoring was also considered. and width The stress amplitude in fatigue damage calculations is corrected in real time using a temperature correction term in the form of Arrhenius. This coupling term reflects the physical nature of corrosion-accelerated fatigue.

[0104] High-temperature softening-corrosion coupling term: frequency factor in the high-temperature microstructure softening sub-model. With the thickness of the corrosion product layer Dynamic correction: The presence of corrosion product layers alters the thermal conductivity and interfacial reactivity of the material surface, thereby affecting the grain coarsening rate and the microstructure softening process.

[0105] Corrosion-creep coupling term: at instantaneous radius of curvature In the update, the coupling relationship between stress, wall thickness and creep is solved iteratively. Corrosion thinning leads to a reduction in effective wall thickness, which in turn increases local stress. The increased stress accelerates creep deformation, which changes the geometric configuration and further affects the stress distribution.

[0106] The mathematical expressions of the above coupling mechanisms are all written into the state transition equations as explicit formulas to ensure that the model has clear physical interpretability.

[0107] In step five of this embodiment, a particle filter algorithm is used for real-time state estimation and lifetime prediction, as specifically implemented as follows: State vector definition: Defines the state vector of the system at time t. The state vector is ,in For effective wall thickness, For real-time tensile strength, It is a fatigue damage factor.

[0108] State transition equation: The multiphysics coupling degradation model established in step three (including the above coupling terms) is used as the state transition equation, i.e. ,in For a moment The input vector (including pressure, temperature, pH value, conductivity, etc.) This represents process noise, characterizing model uncertainty.

[0109] Observation equation: Based on the fragmentation condition monitoring data collected in step two (such as ultrasonic thickness measurement) Electrochemical polarization resistance Using (etc.) as observed values, construct observation equations. ,in To measure the noise, it follows a zero-mean Gaussian distribution, and its covariance matrix is ​​determined by the sensor accuracy calibration.

[0110] Particle filter recursion: Initialization: from prior distribution Extraction Particles The initial weights are all .

[0111] Prediction step: Based on the state transition equation, perform a one-step prediction for each particle to obtain... .

[0112] Update step: When new observations are obtained Then, the likelihood of each particle is calculated. and update the weights .

[0113] Resampling: Residual resampling is performed on the particles according to their weights to obtain a set of particles with equal weights. To overcome the problem of particle scarcity.

[0114] State estimation: Calculate the posterior state estimate And estimate the posterior covariance.

[0115] Based on the posterior particle set at the current moment Using the dynamic mapping equation from step four, the state vector of each particle is advanced to predict the explosion pressure at future moments. When moving forward, environmental parameters (pressure, temperature, pH, etc.) for future moments need to be provided.

[0116] This embodiment employs the steady-state extrapolation assumption, meaning that future parameters remain unchanged from the current measured values. Alternatively, a stochastic process can be used to simulate and generate multiple future operating path scenarios, predicting each path separately and then merging the results. When the predicted blast pressure of a certain particle first falls below the safety threshold... At that time, record the remaining lifetime of the particle. The final remaining lifetime prediction is the statistical distribution of the remaining lifetime of all particles, and can output the mean, median, and 95% confidence interval.

[0117] To further highlight the inventiveness of this embodiment, the following comparative examples are provided: Comparative Example E: Uncoupled mechanistic model and particle filter, using the same particle filter framework as Example 3, but the degenerate model in step three is a simple superposition of three sub-models (i.e., without considering coupling terms), used to verify the contribution of coupling terms to accuracy.

[0118] Comparative Example F: Coupled Mechanism Model and Deterministic Update. The same multiphysics coupled degradation model as in Example 3 is used, but instead of particle filtering, deterministic numerical integration (such as the Euler method) is used to directly calculate the state evolution to verify the contribution of particle filtering to uncertainty quantification.

[0119] The experimental conditions were the same as in Example 2, and the prediction results are shown in Table 3.

[0120] Table 3 Comparison of prediction results from different models:

[0121] Comparative Example D (compared to Example 3) still has a slightly higher prediction error due to the lack of coupling terms, indicating the contribution of coupling terms to accuracy. Comparative Example F has a mean prediction accuracy comparable to Example 3, but it cannot output confidence intervals and has poor adaptability to sudden operating conditions. Example 3 integrates a multiphysics coupling mechanism model and a particle filter algorithm, achieving the highest prediction accuracy and outputting prediction results with confidence intervals, providing the most comprehensive information for pre-maintenance decision-making.

[0122] This embodiment addresses the disconnect between macroscopic monitoring and microscopic mechanisms. By constructing a multiphysics-coupled degradation model, it maps microscopic material degradation parameters to macroscopic burst pressure offsets in real time, achieving a leap from phenomenological monitoring to essential characterization. It also solves the problem of offline assessments being unable to provide real-time dynamic monitoring. Through the collaboration of the underlying sensing module and edge computing unit, it achieves real-time tracking and dynamic updating of the performance degradation of the burst disc during its service life. Using a particle filtering algorithm, it outputs a remaining lifetime probability distribution with confidence intervals, providing a scientific basis for pre-maintenance decisions and avoiding the blindness of traditional deterministic predictions. Through strong multiphysics-coupled modeling, it realistically recreates the material degradation process under the interaction of corrosion, fatigue, and creep, reducing the prediction error by an order of magnitude compared to existing technologies.

[0123] Example 4: Based on Examples 1, 2, and 3, this example further addresses the potential faults (such as zero drift, sensitivity decay, and sudden failure) and data loss issues that industrial field sensors may experience during long-term service. It proposes an online quality assessment and adaptive calibration method for multi-source heterogeneous sensor data and deeply integrates it with a particle filter algorithm to improve the system's robustness and prediction accuracy in harsh environments.

[0124] In practical implementation: The underlying perception module adopts a heterogeneous redundancy design, with the following specific configuration: Pressure measurement: Install an absolute pressure diaphragm pressure sensor (model: PX409-5.0KG5V, range 0~10 MPa, accuracy 0.05% FS) on both the upstream and downstream sides of the rupture disc to form dual redundancy; in addition, reserve a standard pressure calibration interface on the flange body for connecting an external standard pressure source.

[0125] Temperature measurement: Three four-wire Pt1000 platinum resistance thermometers are arranged on the surface of the rupture disc in a 120° circle. The average value is taken as the final temperature value. The local temperature gradient and sensor status are monitored by the deviation between the three sensors.

[0126] Electrochemical monitoring: A dual three-electrode probe system is used, namely two independent working electrodes, auxiliary electrodes and reference electrodes, which are packaged in the same probe and serve as backups for each other; one set is used for normal measurement, and the other set is switched periodically for online calibration.

[0127] Wall thickness measurement: Two independent ultrasonic thickness sensors (center frequency 5 MHz) are installed at the center and edge of the rupture disc, respectively, for mutual verification.

[0128] Auxiliary monitoring: Add an acceleration sensor (to measure vibration) and a conductivity sensor to establish a correlation model with the main parameters.

[0129] Based on the aforementioned redundant configuration, the correlation relationships between various physical quantities are defined. For example, pressure P satisfies the gas law or empirical formula with respect to temperature T and flow rate Q; the electrochemical corrosion current density i_corr exhibits an Arrhenius-type relationship with temperature T, pH value, and conductivity κ. These relationships constitute the physical constraints for subsequent fault diagnosis.

[0130] This embodiment employs an extended Kalman filter (EKF) to fuse data from multiple similar sensors and utilizes innovation sequences for fault detection. Taking a pressure sensor as an example, consider two pressure sensors simultaneously measuring the same pressure P, with measured values ​​z1,k and z2,k, and measurement noise variances R1 and R2, respectively. The system state equation is: ; Where Q is the process noise variance. The observation equation is: ; The EKF recursive process (simplified here to be linear Kalman filtering) yields the state estimate. and its covariance And calculate the innovation vector The new information covariance is Define fault detection statistics: ; It follows a chi-square distribution with 2 degrees of freedom. The significance level is set to α = 0.05, corresponding to a threshold χ²(2, 0.05) = 5.991. If... If so, it is considered that there is a sensor malfunction.

[0131] Furthermore, by analyzing the contribution of each component in the information, the faulty sensor can be located: if the first component in the information is much larger than the second component, then sensor 1 is determined to be faulty; otherwise, sensor 2 is determined to be faulty; if both are large, then the system state may change abruptly or both may be faulty.

[0132] For fault detection between different types of sensors, residuals are constructed using physical constraints. For example, pressure P satisfies the following relationship with temperature T and flow rate Q under steady state: The residuals can be established by simplifying the model, and the chi-square test can be used to judge them.

[0133] When a sensor is diagnosed as faulty or has missing data, its true value needs to be reconstructed using healthy sensors and historical data. This embodiment uses a Long Short-Term Memory (LSTM) neural network for time-series prediction. Taking a pressure sensor fault as an example, the training data comes from historical normal periods and data from other healthy sensors under the same operating conditions. Input features include: healthy sensor data within a time window (the previous 60 seconds) (such as data from another pressure sensor, temperature, flow rate, vibration, etc.), and the output is the pressure estimate for the current moment.

[0134] LSTM model construction: The network structure consists of an input layer (feature dimension D=5), two LSTM hidden layers (64 neurons per layer), and a fully connected output layer (output dimension 1).

[0135] The Adam optimizer, mean squared error loss function, and 200 training epochs were used, with early stopping to prevent overfitting. After training, during real-time operation, whenever pressure sensor 1 malfunctions, the health sensor data sequence of the previous 60 seconds is input into the LSTM to obtain the reconstructed pressure value. And estimate the reconstruction uncertainty. (Derived through test set error statistics). This reconstructed value and its variance will replace the original measurement value and be input into the subsequent particle filter observation equation.

[0136] In practical applications, this embodiment designs two calibration mechanisms to address sensor drift caused by long-term service: Built-in standard pressure source calibration: A sealed, small standard pressure chamber (approximately 50 mL) is installed at the rupture disc mounting flange and connected to a high-pressure nitrogen cylinder via a miniature solenoid valve. Every 30 days (or when the system detects a drift trend), the valve connecting the rupture disc to the process piping is automatically closed, the standard chamber valve is opened, and a known standard pressure (e.g., 5.00 MPa) is applied to the pressure sensor for 10 seconds. The sensor measurements are recorded, the deviation ΔP is calculated, and the sensor's zero-point offset and gain coefficient are updated. After calibration, the process connection is restored. This process does not affect normal production.

[0137] Automatic calibration of electrochemical probes: A miniature reservoir containing a standard solution of known concentration with the same composition as the service medium (e.g., containing 500 ppm Cl) is placed near the probe. - (Buffer solution with pH=7). Every 15 days, a standard solution is injected into the probe measurement chamber using a micro-peristaltic pump to replace the process medium, and linear polarization resistance is measured to obtain the polarization resistance R_p. The actual value of the Stern-Gearty constant B is then calculated. The correction factor is calculated by comparing the B value before calibration. This is used to update subsequent corrosion current calculations. After calibration, the probe is rinsed with process medium to restore the measurement.

[0138] All calibration parameters (zero point, gain, B-value correction factor) are stored in the non-volatile memory of the edge computing module and applied to subsequent measurements in real time.

[0139] This embodiment uses the output of the aforementioned data quality assessment and calibration module as the observation input for the particle filter, and considers observation uncertainties. The state vector of the particle filter remains... The state transition equation is the same as in Example 3 (including multiphysics coupling terms). The observation equation has two cases: When all sensors are functioning normally, the observation vector Observation noise covariance matrix It is a synthesis of sensor accuracy and calibration uncertainty.

[0140] When a sensor (such as ultrasonic thickness gauge) malfunctions and is reconstructed, the observed value... Replace with reconstructed values The observation noise variance was adjusted to The sum of (reconstruction uncertainty) and sensor reference noise.

[0141] If multiple sensors fail simultaneously, a weighted fusion strategy is adopted, prioritizing the use of reconstructed values. If reconstruction is not possible, the weight of the observation is temporarily reduced, or the observation is even ignored (relying solely on state transitions).

[0142] In particle filtering algorithms, the observation likelihood function is calculated based on the probability distribution of observed values; therefore, accurate quantification of observation noise is crucial. This embodiment uses online calibration and fault diagnosis to update the observation noise covariance matrix in real time, enabling the particle filter to adaptively adjust its confidence level in the observations and avoid filter divergence caused by sensor malfunctions.

[0143] To verify the effectiveness of the method in this embodiment, a six-month field test was conducted on a hydrocracking unit in a chemical plant, comparing three different approaches: Option A: The method described in Example 3, sensorless calibration and data reconstruction.

[0144] Option B: The method described in Example 4 includes sensor fault diagnosis, data reconstruction, and adaptive calibration.

[0145] Option C: The traditional periodic replacement strategy, used as a benchmark.

[0146] During the test, sensor failure scenarios were artificially simulated (disconnecting a pressure sensor signal on day 30, applying drift to the ultrasonic thickness gauge on day 60, and causing the electrochemical probe to fail on day 90) to observe the system response and lifetime prediction accuracy. The final actual failure time of the rupture disc was day 135. The prediction results are shown in Table 4.

[0147] Table 4 Comparison of prediction performance of different schemes under sensor failure scenarios;

[0148] As shown in Table 4, Scheme A exhibits a significant decrease in prediction accuracy during sensor failures, even leading to false alarms and unplanned downtime. In contrast, Scheme B, through fault diagnosis, data reconstruction, and adaptive calibration, maintains consistently high prediction accuracy without false alarms or missed alarms, verifying the robustness and reliability of the method in this embodiment. This addresses the industry pain point of poor sensor reliability leading to prediction failure in harsh environments. Through redundant design, fault diagnosis, and data reconstruction, it ensures that even with partial sensor failure, the system can still output reliable lifetime prediction results, significantly improving the availability of industrial safety systems. It achieves self-calibration of sensors during long-term service. The built-in reference source and automatic calibration mechanism effectively suppress zero-point drift and sensitivity decay, maintaining measurement accuracy at its initial level over the long term and avoiding system errors introduced by sensor aging. Deep integration with the particle filter algorithm improves state estimation accuracy. By online evaluation of observation noise and reconstruction uncertainty, the particle filter can adaptively adjust observation weights, avoiding filter divergence caused by erroneous observations and making remaining lifetime prediction more accurate.

[0149] Field tests show that this solution reduces unplanned downtime caused by sensor problems to zero, creating significant economic and safety benefits for enterprises.

[0150] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for modeling the lifespan of a rupture disc based on material degradation, characterized in that, Includes the following steps: Step 1: Establish the initial constitutive model and material fingerprint library of the in-service rupture disc. By conducting material physicochemical analysis on standard specimens from the same batch as the in-service rupture disc, obtain the microstructure parameters and mechanical constitutive parameters in the initial state. Step 2: Construct a real-time sensing field for multi-source service environment parameters and rupture disc status parameters. Deploy sensor arrays at the rupture disc installation location and its upstream and downstream sides to collect environmental stress data and rupture disc status monitoring data in real time during the rupture disc's service process. Step 3: Establish a material property degradation model based on multi-physics coupling. The degradation model is composed of a chemical corrosion thinning sub-model, a high-temperature microstructure softening sub-model, and a fatigue damage accumulation sub-model through coupling terms. It is used to calculate the real-time damage variables of the rupture disc material during service. Step 4: Construct a dynamic mapping equation between the material's microscopic degradation and the macroscopic burst pressure offset. Based on the geometric topology of the rupture disc and its mechanical instability criteria, use the wall thickness reduction, microstructure softening parameters, and fatigue damage factors obtained in Step 3 as inputs to establish a real-time burst pressure prediction model to characterize the evolution trajectory of macroscopic burst capability. Step 5: Achieve dynamic prediction and reliability assessment of the remaining life of the rupture disc. Set the failure criteria for the rupture disc, use the particle filter algorithm, take the degradation model established in Step 3 as the state transition equation, take the rupture disc state monitoring data collected in Step 2 as the observation value, update the life distribution function in real time, and calculate the remaining life based on the updated state variables using the dynamic mapping equation in Step 4.

2. The method for modeling the lifespan of a rupture disc based on material degradation according to claim 1, characterized in that, In step one, the microstructure parameters include grain size grade, phase composition ratio, average precipitate particle size, initial surface roughness, grain boundary characteristic parameters based on scanning electron microscope image recognition, and initial orientation distribution function obtained by electron backscatter diffraction technology; the mechanical constitutive parameters include initial yield strength, initial tensile strength, work hardening index, and elastic modulus.

3. The method for modeling the lifespan of a rupture disc based on material degradation according to claim 1, characterized in that, In step two, the environmental stress data includes the instantaneous pressure of the fluid medium, the real-time temperature of the rupture disc surface, the conductivity of the medium, the pH value, and the stress cycle frequency; the rupture disc condition monitoring data includes the remaining wall thickness monitored in real time by an ultrasonic thickness sensor, the acoustic emission event count collected by an acoustic emission sensor, or the local strain measured by a strain gauge.

4. The method for modeling the life of a rupture disc based on material degradation according to claim 1, characterized in that, The chemical corrosion thinning sub-model described in step three calculates the real-time thinning amount of the effective wall thickness of the rupture disc using Faraday's law of electrolysis. The formula for calculating the real-time thinning amount is as follows: ， Where M is the molar mass of the rupture disc material. The self-corrosion current density is corrected for real-time temperature and pH, z is the reaction charge number, F is the Faraday constant, and ρ is the material density.

5. The method for modeling the life of a fragmentation disc based on material degradation according to claim 4, characterized in that, The self-corrosion current density is obtained in real time by an electrochemical monitoring probe. The linear polarization resistance method is used to apply a small potential scan to the working electrode and measure the polarization resistance. The self-corrosion current density is equal to the Stern-Geary constant divided by the polarization resistance. The Stern-Geary constant is calibrated by electrochemical tests in the material fingerprint library.

6. The method for modeling the lifespan of a fragmentation disc based on material degradation according to claim 4, characterized in that, For rupture discs with pre-grooved structures, the chemical corrosion thinning sub-model also introduces a local stress concentration correction factor. The relationship between the thinning rate at the pre-grooved area and the thinning rate in the flat area is defined as follows: ； in For local stress at the bottom of the trench, Where β is the nominal stress, and m are constants related to the corrosive activity of the medium. The thinning rate at the pre-cut groove, The thinning rate of the flattened area.

7. The method for modeling the life of a rupture disc based on material degradation according to claim 1, characterized in that, The high-temperature tissue softening sub-model described in step three uses the Arrhenius equation to describe the decay of strength parameters with the thermal activation process. The formula for calculating real-time tensile strength is: ； in Let A be the initial tensile strength, t be the material-specific frequency factor, Q be the time, and R be the activation energy for microstructure evolution. This is the real-time temperature.

8. The method for modeling the life of a fragmentation disc based on material degradation according to claim 7, characterized in that, When calculating the strength decay, the coarsening rate of the precipitated phase is calculated using the Lifshitz-Slyozov-Wagner theory, and the frequency factor A is corrected accordingly. If the service environment is a low-temperature environment, the system automatically switches to the low-temperature embrittlement monitoring mode. The nonlinear correction index of the kinetic mapping equation in step four is corrected by calculating the material's ductile-brittle transition temperature shift and the impact energy evolution process.

9. A fragmentation lifetime modeling system based on material degradation for implementing the method of any one of claims 1 to 8, characterized in that, The system includes: The underlying sensing module is located at the installation site of the rupture disc and is used to collect multi-physics field environmental stress signals and rupture disc status monitoring signals in real time during the service process of the rupture disc. The edge computing module is connected to the underlying sensing module and is used to process the collected signals locally in real time and run the material degradation mechanism kernel to perform initial life assessment and early warning. The cloud management module communicates with the edge computing module and is used to deploy the digital twin engine and the full lifecycle database to achieve global adaptive optimization and visual management of the model.

Images