Stainless steel high-temperature fatigue prediction method based on noise adaptive particle filtering

Abstract

The invention discloses a stainless steel high-temperature fatigue prediction method based on noise adaptive particle filtering, and relates to the technical field of metal material high-temperature fatigue nondestructive testing and prediction. According to the method, a temperature, stress, scanning distance and local stress position coupled dislocation-magnetic induction intensity model is constructed, and a standardized magnetic induction signal is collected as an observation reference; particle swarm optimization (PSO) is utilized to optimize particle filtering to screen initial particles, iterative prediction is carried out in combination with a noise adaptive particle filtering algorithm, the optimal estimated value of the magnetic induction intensity of each time step is obtained, and finally stainless steel high-temperature fatigue magnetic induction signal acquisition and prediction are realized. Within the range from normal temperature to 900 DEG C, the method has the advantages that the prediction error of the stainless steel high-temperature fatigue is greatly reduced compared with the traditional particle filtering, the magnetic induction intensity sudden increase and sudden drop signals can be accurately tracked, and reliable technical support is provided for the early fatigue detection, trend prediction and engineering early warning of the stainless steel volute.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing of high-temperature fatigue in metallic materials, and specifically to a method for predicting high-temperature fatigue of stainless steel based on noise adaptive particle filtering. Background Technology

[0002] Driven by global energy transition and carbon reduction strategies, the automotive industry has an increasingly urgent need for more efficient powertrains and cleaner emission systems. Turbocharging technology has become one of the core technologies due to its significant fuel-saving effects and power-boosting capabilities. The turbocharger turbine housing operates in a high-temperature alternating stress environment of 300℃-900℃ for extended periods. The choice of material directly determines the service life and safety of the turbine housing. Stainless steel, as a typical representative of austenitic heat-resistant stainless steel, has become the core material for the turbine housing due to its excellent high-temperature resistance and mechanical stability.

[0003] However, during service, high temperatures in stainless steel turbochargers can cause grain boundary weakening and dislocation proliferation in their microstructure. Combined with alternating loads, this easily leads to accumulated fatigue damage, ultimately resulting in crack initiation and propagation. Simultaneously, temperature gradients, differences in thermal expansion coefficients, and structural constraints further exacerbate thermal stress, accelerating turbocharger failure. Therefore, early detection and accurate prediction of the high-temperature fatigue process in stainless steel are crucial for ensuring the safe operation of automotive turbochargers.

[0004] Existing technologies, such as ultrasonic testing, eddy current testing, radiographic testing, and electromagnetic particle testing, have significant limitations: ultrasonic testing requires high-quality acoustic properties of the material and cannot be adapted to surface defects; eddy current testing is greatly affected by temperature, with accuracy dropping sharply at high temperatures; radiographic testing carries radiation risks and cannot achieve real-time monitoring; and electromagnetic particle testing introduces external magnetic fields, interfering with the magnetic properties of stainless steel and causing distorted test results. All of these technologies struggle to capture early microscopic damage within the material (such as dislocation changes and minor stress concentrations), easily missing the optimal maintenance window.

[0005] In recent years, deep learning and particle filtering techniques have been attempted for metal fatigue prediction: Deep learning can uncover hidden relationships in data (such as predicting the fatigue life of welded components based on convolutional neural networks), but it is difficult to cope with noise interference in high-temperature environments when used alone; Traditional particle filtering can achieve dynamic signal tracking, but it has defects such as particle initialization distribution deviating from the true state, concentration of importance weights (insufficient effective particles), and sensitivity to noise. In the stage of sudden change of magnetic induction signal in high-temperature fatigue of stainless steel (such as the sudden drop in magnetic induction intensity at the moment of fracture), the prediction error of traditional particle filtering increases significantly and cannot meet engineering requirements.

[0006] In summary, existing technologies cannot simultaneously solve the three core problems of "poor adaptability to high-temperature environments", "weak noise interference suppression" and "low accuracy of early damage prediction". There is an urgent need for a high-temperature fatigue prediction method for stainless steel that can couple multiple physical parameters, dynamically optimize noise, and adapt to a wide temperature range. Summary of the Invention

[0007] The purpose of this invention is to provide a high-temperature fatigue prediction method for stainless steel based on noise adaptive particle filtering, which accurately captures early microscopic damage in stainless steel (such as dislocation multiplication and micro-stress concentration). The core parameters of the particle filter are optimized using a particle swarm optimization (PSO) algorithm to solve the problems of initial distribution deviation and insufficient effective particles in PSO-optimized particle filtering, thus improving signal tracking stability. A noise adaptive particle filtering algorithm is introduced to dynamically optimize noise parameters and suppress non-Gaussian noise (such as electromagnetic interference in the furnace and sensor thermal drift) under high-temperature conditions. Accurate prediction is achieved over a wide temperature range from room temperature to 900℃, especially suitable for scenarios with a sharp increase (accumulation of fatigue damage) and a sudden decrease (instantaneous fracture) in magnetic induction intensity, providing a scientific basis for vortex shell maintenance decisions.

[0008] This invention provides a method for predicting high-temperature fatigue of stainless steel based on noise adaptive particle filtering, comprising the following steps: S1: Construct a dislocation-magnetic induction intensity model that couples multiple physical parameters, including temperature, stress, scanning distance, and local stress location; S2: Collect the magnetic induction signal of stainless steel in the high temperature fatigue test and obtain the standardized magnetic induction signal; S3: Using the standardized magnetic induction signal as the observation benchmark, a particle swarm is constructed based on the dislocation-magnetic induction intensity model, with each particle corresponding to a set of initial state parameters for particle filtering; using the "minimization of prediction error" between the predicted particle state value and the standardized magnetic induction signal as the fitness function, the particle position and velocity are iteratively updated through the particle swarm optimization algorithm to select an optimized initial particle set that closely matches the fluctuation range of the real magnetic induction signal. S4: Enter the time step loop and iteratively predict the optimized initial particle set input noise adaptive particle filter algorithm; within each time step, perform online noise parameter adaptation, particle state propagation, weight update and resampling operations in sequence, and finally output the optimal estimate of the magnetic induction intensity at the current time step; S5: Integrates the optimal estimates of magnetic induction intensity for all time steps to form a complete prediction curve, and outputs the trend of magnetic induction signal change and fatigue damage prediction results for high-temperature fatigue of stainless steel.

[0009] Furthermore, both the noise-adaptive particle filtering algorithm and the particle swarm optimization algorithm must conform to the physical constraints of the dislocation-magnetic induction intensity model.

[0010] Furthermore, the inertia weight of the particle filter in the particle swarm optimization algorithm adopts a linear decreasing strategy.

[0011] Furthermore, the iterative prediction step includes: Online noise parameter estimation: The optimized initial particle set state estimate is input into the pre-trained MLP noise adaptive module, which outputs the optimal process noise covariance and observation noise covariance for this time step, thus obtaining the optimal noise parameters; Particle state prediction: Based on the optimal noise parameters and combined with the particle filter state transition equation, the state of each particle is dynamically propagated to obtain the predicted state of the particle at time step k. Particle weight update: Calculate the likelihood between the predicted state of each particle and the normalized magnetic induction signal observation corresponding to time step k, and update and normalize the weight of each particle accordingly. Effective particle count calculation and state estimation: Calculate the effective particle count to obtain the rear particles, and perform a weighted average on the rear particles to obtain the optimal estimate of the magnetic induction intensity at time step k.

[0012] Furthermore, the MLP noise adaptive module has the following characteristics: The MLP noise adaptive module has one neuron in the input layer and one neuron in the output layer, and the hidden layer uses the ReLU activation function. The MLP noise adaptive module uses the magnetic induction signal state estimate output by the particle swarm optimization algorithm as input features and the process noise covariance matrix and measurement noise covariance matrix as output labels. It constructs a training dataset through Z-score normalization and completes the training. The trained MLP noise adaptive module receives the optimized initial particle set state estimate, predicts the optimal noise parameters in real time, and dynamically adjusts the process noise covariance and observation noise covariance of the particle filter after inverse normalization and safe truncation. In the online prediction stage, the MLP noise adaptive module performs inverse normalization and safe truncation on the output results.

[0013] Furthermore, the method for calculating the effective particle count is as follows: if the effective particle count is less than 0.6 × the total particle count, then system resampling is performed to replace particles with too small a weight; if the effective particle count is greater than 0.6 × the total particle count, then the current particle distribution is retained; and the subsequent particles are obtained.

[0014] Furthermore, the time step loop stops under the following conditions: if the current time step k is the preset total time step, then the loop terminates; otherwise, the value of k is incremented by 1, and the time step loop continues to iterate.

[0015] A prediction device for a high-temperature fatigue prediction method for stainless steel based on noise adaptive particle filtering includes: an industrial control computer and a high-precision magnetic sensor. The plug of the high-precision magnetic sensor is electrically connected to the interface of the industrial control computer. The high-precision magnetic sensor is used to acquire data on the change of magnetic induction intensity signal on the stainless steel surface. The industrial control computer is used to process the magnetic induction intensity data measured by the high-precision magnetic sensor, predict the fracture process based on the change of magnetic induction intensity data, and finally output the same type of magnetic induction intensity prediction data.

[0016] The beneficial effects of this invention are: The established dislocation-magnetic induction intensity model can better reflect the influence of temperature and stress on magnetic signals. By introducing scanning distance and local stress parameters, the magnetic signal changes of stainless steel materials under different states can be analyzed more accurately, providing theoretical support for high-temperature fatigue damage detection. PSO optimizes the initial parameters and importance weights of particles, laying the core foundation for noise-adaptive particle filtering. PSO guides the particle swarm through "individual extrema" and "global extrema," solving the problems of initial distribution deviation and insufficient effective particles in PSO particle filtering, and improving the basic tracking ability of magnetic induction signal fluctuations (especially slowly changing signals in the fatigue accumulation stage). The noise-adaptive particle filter algorithm significantly improves the prediction accuracy of magnetic induction signals during high-temperature fatigue. Comparative experiments show that the noise-adaptive particle filter algorithm dynamically optimizes noise parameters through the MLP noise adaptation module, and can still stably track signal changes during critical stages such as sharp increases in magnetic induction intensity (e.g., during the fatigue damage accumulation stage) and sudden decreases (e.g., at the moment of fracture). Attached Figure Description

[0017] Figure 1 This is a flowchart of fatigue prediction using noise adaptive particle filtering according to an embodiment of the present invention; Figure 2 A comparison chart of the prediction effects when the magnetic induction intensity of stainless steel under fatigue load increases sharply according to an embodiment of the present invention; Figure 3 A comparison chart showing the prediction effect of the sudden drop in magnetic induction intensity at fracture of stainless steel under fatigue load according to an embodiment of the present invention; Figure 4 This is an example of the noise adaptive particle filter prediction effect of stainless steel at 600℃ in an embodiment of the present invention. Figure 5 This is an example of the noise adaptive particle filter prediction effect of stainless steel at 300℃ in an embodiment of the present invention. Figure 6 This is an example of the noise adaptive particle filter prediction effect of stainless steel at room temperature in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] Traditional prediction methods using crack length as a state parameter are inadequate for stainless steel during high-temperature fatigue, as the magnetic flux density signal exhibits nonlinear fluctuations with temperature and stress, and no significant plastic deformation occurs before fracture. This invention addresses this issue by establishing a dislocation-magnetic flux density model coupled with multiple physical parameters, PSO-optimized particle filtering, and noise-adaptive particle filtering to achieve accurate prediction of the high-temperature fatigue process in stainless steel.

[0020] like Figure 1 As shown, the specific steps of the high-temperature fatigue prediction method for stainless steel based on noise adaptive particle filtering provided by this invention are as follows: S1: Construct a dislocation-magnetic induction intensity model that couples multiple physical parameters, including temperature, stress, scanning distance, and local stress location; S2: Collect the magnetic induction signal of stainless steel in the high temperature fatigue test and obtain the standardized magnetic induction signal; S3: Using the standardized magnetic induction signal as the observation benchmark, a particle swarm is constructed based on the dislocation-magnetic induction intensity model, with each particle corresponding to a set of initial state parameters for particle filtering; using the "minimization of prediction error" between the predicted particle state value and the standardized magnetic induction signal as the fitness function, the particle position and velocity are iteratively updated through the particle swarm optimization algorithm to select an optimized initial particle set that closely matches the fluctuation range of the real magnetic induction signal. S4: Enter the time step loop and iteratively predict the optimized initial particle set input noise adaptive particle filter algorithm; within each time step, perform online noise parameter adaptation, particle state propagation, weight update and resampling operations in sequence, and finally output the optimal estimate of the magnetic induction intensity at the current time step; S5: Integrates the optimal estimates of magnetic induction intensity for all time steps to form a complete prediction curve, and outputs the trend of magnetic induction signal change and fatigue damage prediction results for high-temperature fatigue of stainless steel.

[0021] The dislocation-magnetic flux density model clarifies the quantitative correlation between "microscopic dislocation evolution and macroscopic magnetic signal change," providing physical constraints for noise-adaptive particle filtering and particle swarm optimization particle filtering algorithms. The specific construction process is as follows: Effective magnetic field and pinning field construction: Total effective magnetic field inside the material Determined by the external magnetic field, interdomain interaction, and dislocation pinning effect, the expression is: ; in, It is an external magnetic field. It is the mean field term, representing the interaction between magnetic domains. It is magnetization. It's a nailing site.

[0022] nailing field The strength of the dislocation is directly proportional to the density of pinning defects. The higher the value, the stronger the pinning effect on the domain walls. We can establish the following relationship: ; in, It is a pinning coefficient related to material properties. This relationship shows that dislocation density is a key microscopic parameter affecting the magnetization behavior of materials.

[0023] Dislocation density under steady-state conditions Through stress and temperature To describe it. We will use the classic dislocation density evolution equation: ; in, It is the yield strength. For creep activation energy, It is the gas constant. and These are the power-law exponents that control the dislocation annihilation rate and stress sensitivity, respectively. It is a material constant.

[0024] Stress not only alters magnetic properties by affecting dislocation density, but also directly influences magnetization through magnetoelastic effects. (Applied stress) A magnetoelastic energy will be generated. Under simple tensile stress, it can be expressed as: ; in, It is the saturation magnetostriction coefficient. It is the angle between the magnetization direction and the stress direction.

[0025] Based on thermodynamic equilibrium, stress equivalent magnetic field for: ; in, The permeability of free space, This represents the temperature-dependent saturation magnetization.

[0026] Introducing the Curie temperature of stainless steel The optimized temperature correction model shows that the saturation magnetization varies with temperature as follows: ; in, Reference temperature Saturation magnetization at the following values It is the Curie temperature of stainless steel. It is a temperature correction factor.

[0027] The stress distribution in the stress concentration region is described using a Gaussian function, expressed as: ; in It is a spatial location variable. It is the center of stress concentration. It is the basic stress. It is the peak stress. It is the degree of spatial expansion of the stress concentration peak (which controls the range of stress concentration).

[0028] Considering the distance between the sensor probe and the surface of the stainless steel material during the scanning process The scanning distance affects the signal amplitude, so an attenuation factor is introduced. ,in This is the attenuation coefficient. Integrating these factors into the model yields the final magnetic flux density model that considers spatial distribution: ; This model incorporates the distance between the sensor probe and the surface of the stainless steel material. ,stress and temperature magnetic induction intensity The impact.

[0029] The PSO particle filter inertial weight (ω) employs a linear decreasing strategy to iteratively update particle position and velocity, selecting an optimized initial particle set that closely approximates the fluctuation range of the real magnetic induction signal; the real-time particle position (ω) ) and speed ( It can be expressed by the following formula: ; ; in, Inertial weight (controls the search range). , It is a learning factor (which guides particles toward individual and global optimality). For the individual optimal particle The optimal particle is represented by i, where i is the particle index and k is the time step. An optimized initial particle set is selected through iterative filtering to ensure that the initial particle distribution more closely approximates the fluctuation range of the actual magnetic induction signal, thereby reducing initial bias.

[0030] The iterative prediction steps of the noise adaptive particle filter algorithm are as follows: Online noise parameter estimation: The state estimate of the optimized initial particle set is input into the pre-trained MLP noise adaptive module, which outputs the optimal process noise covariance and observation noise covariance at this time step, thus obtaining the optimal noise parameters; Particle state prediction: Based on the optimal noise parameters and combined with the particle filter state transition equation, the state of each particle is dynamically propagated to obtain the predicted state of the particle at time step k. Particle weight update: Calculate the likelihood between the predicted state of each particle and the normalized magnetic induction signal observation corresponding to time step k, and update and normalize the weight of each particle accordingly. Effective particle count calculation and state estimation: Calculate the effective particle count. If the effective particle count is less than 0.6 × total particle count, perform system resampling and replace particles with too small a weight. If the effective particle count is greater than 0.6 × total particle count, retain the current particle distribution. Obtain the subsequent particles and perform a weighted average on the subsequent particles to obtain the optimal estimate of the magnetic induction intensity at time step k.

[0031] The time step loop stops when the current time step k is the preset total time step (set according to the total duration of the fatigue test), the loop terminates; otherwise, the value of k is incremented by 1, and the time step loop continues to iterate.

[0032] The MLP noise adaptive module has one neuron in both its input and output layers, and the hidden layers use the ReLU activation function. During data preparation, the module uses the magnetic induction signal state estimate (reflecting the predicted state of the current signal) output by the PSO particle filter as input features, and uses the process noise covariance matrix and measurement noise covariance matrix as output labels. A training dataset is constructed using Z-score normalization, and training is completed. The trained MLP noise adaptive module receives the current state estimate, predicts the optimal noise parameters in real time, and dynamically adjusts the process noise covariance and observation noise covariance of the particle filter after inverse normalization and safe truncation. In the online prediction phase, the MLP noise adaptive module performs inverse normalization and safe truncation on the output results.

[0033] A prediction device for a high-temperature fatigue prediction method for stainless steel based on noise adaptive particle filtering includes: an industrial control computer and a high-precision magnetic sensor. The plug of the high-precision magnetic sensor is electrically connected to the interface of the industrial control computer. The high-precision magnetic sensor is used to acquire data on the change of magnetic induction intensity signal on the stainless steel surface. The industrial control computer is used to process the magnetic induction intensity data measured by the high-precision magnetic sensor, predict the fracture process based on the change of magnetic induction intensity data, and finally output the same type of magnetic induction intensity prediction data.

[0034] The high-precision magnetic sensor is cylindrical in shape, and the sensor probe surface is circular.

[0035] In practice, the stainless steel sample (total length 190mm, parallel test area length 60mm, thickness 3mm) is first surface-treated to remove machining marks. Then, a high-precision weak magnetic sensor is fixed above the sample surface, ensuring a stable scanning distance. The sensor is connected to the data acquisition module and the industrial control computer, and the sample is mounted in the high-temperature fatigue testing system. Next, a cyclic fatigue load is applied to the stainless steel, and the industrial control computer begins monitoring the changes in the magnetic induction intensity. Simultaneously, the magnetic induction intensity data monitored by the industrial control computer is continuously imported into the established noise adaptive particle filter model for stainless steel. The industrial control computer runs the noise adaptive particle filter algorithm in real time and outputs a predicted magnetic induction intensity curve.

[0036] The parameters for the high-temperature fatigue test include: stress ratio R=0.1, cyclic waveform is a triangular wave, cyclic frequency is 15Hz, temperature gradient is room temperature, 300℃, 600℃, 900℃, and each temperature is held for 30min to ensure uniform sample temperature; the scanning distance d of the weak magnetic field detection system is set to 160mm, and the stress concentration position x is set to 80mm.

[0037] Example

[0038] Based on the dislocation-magnetic flux density model, the variation of magnetic flux density with stress, temperature, and scanning distance was simulated in MATLAB. The reference temperature was set to T0 = 0°C, the temperature gradients were 100°C, 200°C, 300°C, and 400°C, the stress was σ = 500 MPa (simulating stress concentration), the scanning distance was d = 160 mm, and the stress concentration location was x = 80 mm.

[0039] Simulation results show that the magnetic induction intensity has a peak at x=80mm (corresponding to the stress concentration area), and the peak amplitude decreases with increasing temperature (the peak amplitude at 100°C is 12.3% higher than that at 400°C), which verifies the model's ability to locate local stress concentration and the attenuation effect of temperature on the magnetic signal.

[0040] Figure 2 - Figure 3To conduct high-temperature fatigue tests on stainless steel samples at 900°C, the prediction effects of noise-adaptive particle filtering and the original particle filtering were compared. To further verify the reliability of the proposed noise-adaptive particle filtering method for predicting the high-temperature fatigue failure process of 309 stainless steel, the fatigue failure process of 309 stainless steel at different temperatures was predicted. Figure 4 - Figure 6 The prediction results show that the noise adaptive particle filtering method can accurately predict the fatigue failure trend of 309 stainless steel at different temperatures (room temperature, 300℃, 600℃), and has good adaptability and reliability in predicting the fatigue failure process of 309 stainless steel.

[0041] Experimental results show that during the stage of rapid increase in magnetic flux density, the maximum prediction error of the noise adaptive particle filter algorithm is reduced by 42.6% compared with the traditional particle filter algorithm; during the stage of rapid decrease in magnetic flux density at the moment of fracture, its prediction fluctuation amplitude is reduced by 58.3%.

[0042] Table 1 shows the analysis of prediction result error evaluation indicators. The comparison of error evaluation indicators (mean absolute error, root mean square error, and mean relative error) shows that at 900°C, the mean absolute error of the noise adaptive particle filter algorithm is 125.32 (compared to 303.44 for the traditional particle filter algorithm), the root mean square error is 1896.38 (compared to 3156.98 for the traditional particle filter algorithm), and the mean relative error is 0.0055% (compared to 0.015% for the traditional particle filter algorithm). All of these are significantly better than the traditional particle filter algorithm, verifying the stability and accuracy of this method in high-temperature environments and during periods of sudden changes in magnetic signals.

[0043] Table 1. Analysis of Prediction Result Error Evaluation Indicators

[0044] Based on the experimental data, it can be seen that the method proposed in this invention can accurately capture the dynamic changes in magnetic induction intensity during the high-temperature fatigue process of stainless steel: When no significant damage occurs, the magnetic signal exhibits periodic fluctuations with fatigue load; During the damage accumulation phase, the signal amplitude shows an upward trend; Before the fracture, the signal dropped sharply.

[0045] By using a noise-adaptive particle filtering algorithm to track the process in real time, early detection and warning of high-temperature fatigue damage in stainless steel can be achieved, providing technical support for the service safety of turbocharger volutes.

[0046] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A stainless steel high-temperature fatigue prediction method based on noise-adaptive particle filtering, characterized by, The method comprises the following steps: S1: constructing a dislocation-magnetic induction strength model coupled with multiple physical parameters of temperature, stress, scanning distance and local stress position; S2: collecting magnetic induction signals of stainless steel in a high-temperature fatigue test to obtain standardized magnetic induction signals; S3: taking the standardized magnetic induction signals as an observation reference, constructing a particle swarm based on the dislocation-magnetic induction strength model, each particle corresponding to a set of particle filtering initial state parameters; taking "prediction error minimization" between particle state prediction values and the standardized magnetic induction signals as a fitness function, iteratively updating particle positions and velocities through a particle swarm optimization algorithm, and screening an optimized initial particle set close to a real magnetic induction signal fluctuation range; S4: entering a time step cycle, inputting the optimized initial particle set into a noise adaptive particle filtering algorithm for iterative prediction; in each time step, sequentially performing noise parameter online adaptation, particle state propagation, weight updating and resampling operations, and finally outputting a magnetic induction strength optimal estimation value at the current time step; S5: integrating the magnetic induction strength optimal estimation values of all time steps to form a complete prediction curve, and outputting a magnetic induction signal change trend and a fatigue damage prediction result of the stainless steel high-temperature fatigue.

2. The stainless steel high-temperature fatigue life prediction method based on noise adaptive particle filter according to claim 1, characterized in that, The noise adaptive particle filtering algorithm and the particle swarm optimization algorithm both conform to the physical constraints of the dislocation-magnetic induction strength model.

3. The stainless steel high-temperature fatigue life prediction method based on noise adaptive particle filter according to claim 1, characterized in that, The inertia weight of the particle filtering of the particle swarm optimization algorithm adopts a linear decreasing strategy.

4. The stainless steel high-temperature fatigue life prediction method based on noise adaptive particle filter according to claim 1, characterized in that, The step of iterative prediction comprises: Noise parameter online estimation: inputting the state estimation value of the optimized initial particle set into a pre-trained MLP noise adaptive module to output the optimal process noise covariance and observation noise covariance at this time step, and obtaining optimal noise parameters; Particle state prediction: based on the optimal noise parameters, dynamically propagating the state of each particle according to a particle filtering state transition equation to obtain a particle prediction state at time step k; Particle weight updating: calculating the likelihood between each particle prediction state and the standardized magnetic induction signal observation value corresponding to time step k to update and normalize the weight of each particle; Resampling and state estimation: calculating the effective particle number to obtain a rear particle, and performing weighted averaging on the rear particle to obtain the magnetic induction strength optimal estimation value at time step k.

5. The stainless steel high-temperature fatigue life prediction method based on noise adaptive particle filter according to claim 4, characterized in that, The MLP noise adaptive module has the following characteristics: The MLP noise adaptive module has 1 neuron in the input layer, 1 neuron in the output layer, and uses a ReLU activation function in the hidden layer; The MLP noise adaptive module uses the magnetic induction signal state estimation value output by the particle swarm optimization algorithm as an input feature, and uses the process noise covariance matrix and the measurement noise covariance matrix as output labels, constructs a training data set through Z-score standardization, and completes training; The trained MLP noise adaptive module receives the state estimation value of the optimized initial particle set, predicts the optimal noise parameters in real time, and dynamically adjusts the process noise covariance and the observation noise covariance of the particle filtering after inverse normalization and safety truncation processing; In the online prediction stage, the MLP noise adaptive module performs inverse normalization and safety truncation on the output results.

6. The stainless steel high-temperature fatigue life prediction method based on noise adaptive particle filter according to claim 4, characterized in that, The method for calculating the effective particle number is: if the effective particle number is less than 0.6 times the total particle number, system resampling is performed to replace particles with too small weights; if the effective particle number is greater than 0.6 times the total particle number, the current particle distribution is kept; and rear particles are obtained.

7. The noise-adaptive particle filter based stainless steel high-temperature fatigue life prediction method of claim 1, wherein, The condition for stopping the time step loop is: if the current time step k is the preset total time step, the loop is terminated; if not, the value of k is increased by 1, and the time step loop is returned to continue iteration.

8. A prediction device of a stainless steel high-temperature fatigue prediction method based on noise-adaptive particle filtering, comprising: The industrial computer and the high-precision magnetic sensor, the plug of the high-precision magnetic sensor is electrically connected with the interface of the industrial computer, the high-precision magnetic sensor is used for acquiring the magnetic induction intensity signal change data of the stainless steel surface, the industrial computer is used for processing the magnetic induction intensity data measured by the high-precision magnetic sensor, and the fracture process is predicted according to the change of the magnetic induction intensity data, and finally the same type of magnetic induction intensity prediction data is output.

Images