Ship lock bottom pivot mushroom head wear real-time prediction and reliability evaluation method

Abstract

The invention belongs to the technical field of water transportation engineering, and discloses a ship lock bottom pivot wear real-time prediction and reliability evaluation method, which comprises the following steps: establishing a finite element simulation model of a gate leaf and a bottom pivot of a ship lock miter gate, and obtaining bottom pivot wear rule data; a Bayesian neural network agent model is constructed based on finite element simulation data, and efficient prediction of the abrasion loss of the bottom pivot is achieved; and a Gamma process-based bottom pivot performance degradation evaluation model is established in combination with a proxy model prediction result, model parameters are corrected in real time by adopting a sequence Bayesian method, and the bottom pivot wear reliability and residual life are dynamically updated. The method can effectively solve the problems that the ship lock bottom pivot is hidden underwater, online monitoring is difficult, and a traditional reliability evaluation method is insufficient, bottom pivot abrasion prediction performance and reliability evaluation credibility are improved, and intelligent guarantee is provided for safe and efficient operation of the ship lock.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy engineering technology, and in particular relates to the operation, maintenance and safety assessment technology of the bottom pivot of a lock gate in water transport engineering. Specifically, it discloses a method for real-time prediction and reliability assessment of wear of the mushroom head of the bottom pivot of a lock. Background Technology

[0002] As a key facility in waterway engineering, the safe and reliable operation of locks plays a vital role in ensuring smooth navigation. However, during operation, locks are prone to various malfunctions due to factors such as wear, fatigue, and corrosion of structural components, leading to prolonged navigation disruptions and significant economic losses. Therefore, in-depth research on the reliability of lock structures is of great practical significance for improving operational safety and reducing maintenance costs. Among the many lock structures, the bottom pivot, as the core operating component and important support of the miter gate, is subjected to a complex underwater environment of intermittent low-speed heavy loads and silt pollution. Its special working conditions make it highly susceptible to wear between friction pairs, making it difficult to form a stable fluid lubricating oil film. Severe wear can cause the gate shaft to tilt and even lead to gate instability, posing a serious threat to operational safety. In addition, the special underwater environment in which the bottom pivot is located makes it difficult to monitor and assess its wear status in real time, resulting in extremely difficult and costly maintenance work after failure. Although existing lock gate design specifications mostly adopt the traditional allowable stress method, this method fails to fully consider the impact of component type, design parameters, and time variability during service on structural reliability.

[0003] Currently, research on the wear problem of the mushroom head of the bottom pivot of a lock mainly focuses on the design level, such as structural optimization, lubrication improvement, and material matching of friction pairs. However, there is relatively little research on the real-time prediction and safety and reliability assessment of the mushroom head wear state of lock bottom pivots that are already in operation. In the field of gate reliability research, although there are results for single components such as the main beam, there is still a lack of research specifically on the reliability of the mushroom head wear failure of the lock bottom pivot. In addition, traditional reliability assessment methods are insufficient in utilizing historical degradation data, and face problems of large amount of repetitive calculation and low computational efficiency when dynamically updating, making it difficult to meet real-time requirements. In summary, the current research on the wear problem of the mushroom head of the lock bottom pivot lacks effective online monitoring and prediction methods, and the reliability assessment methods are still imperfect and the computational efficiency needs to be improved. There is an urgent need for a new method that integrates multidisciplinary theories and data-driven technology to break through the bottleneck in the research on the reliability of the mushroom head wear of the bottom pivot of the lock miter gate.

[0004] CN108318363A discloses a detection device and method for detecting wear of the mushroom head of the bottom pivot. The device acquires surface data and reconstructs the spherical surface by contacting the surface of the mushroom head with a mechanical probe, and then compares the differences in the spherical surface before and after wear to evaluate the wear characteristics. This method can perform wear detection without disassembling the equipment. However, this method is essentially an offline detection technology that relies on manual operation and measurement on site. It cannot achieve continuous and real-time monitoring of the wear state of the bottom pivot and is difficult to capture the dynamic changes in the wear process. In addition, it acquires data through contact measurement, which is relatively cumbersome, and the measurement accuracy is affected by the underwater environment (such as silt and attached substances).

[0005] CN111859732A discloses an automatic monitoring method for the damage degree of lock gates and their supporting moving parts. This method involves deploying sensors and acquiring vibration response parameters in real time, then using a vibration parameter response relationship model to indirectly monitor underwater concealed parts, including the bottom pivot. While this method achieves automatic monitoring of underwater concealed parts, it relies on inferring damage indirectly through vibration response, making it difficult to directly and quantitatively determine the wear amount of the bottom pivot's mushroom head. The assessment results are uncertain, and the mapping relationship between vibration response and wear state is complex, requiring a precise model with high generalization ability. Furthermore, none of the above methods integrate dynamic monitoring of wear state with prediction of the long-term reliability and remaining life of structural components, failing to provide forward-looking guidance for lock maintenance decisions. Therefore, existing technologies have significant shortcomings in real-time, quantitative prediction of bottom pivot wear and its dynamic integration with reliability assessment.

[0006] Existing technologies for monitoring and assessing the wear and reliability of the mushroom head of a lock bottom pivot suffer from problems such as offline measurement, inability to monitor in real time, inability to directly quantify wear, and lack of dynamic integration with reliability assessment. This invention addresses these problems by proposing a real-time prediction and reliability assessment method for the wear of the mushroom head of a lock bottom pivot that integrates finite element simulation, Bayesian neural networks, Gamma stochastic processes, and sequential Bayesian methods. This method aims to solve the problems of difficulty in predicting the wear state of the mushroom head of an already operating lock bottom pivot in real time, the imperfect reliability assessment method, and the need to improve computational efficiency. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: establish a finite element simulation model of the friction pair between the gate leaf and the bottom pivot of the lock, consider the dynamic simulation of frictional contact, simulate the dynamic wear of the bottom pivot mushroom head during the opening and closing process of the gate, and obtain the wear amount of the bottom pivot mushroom head and the corresponding structural dynamic response simulation data under different working conditions. Using the acquired finite element simulation data, a Bayesian neural network surrogate model, namely the mushroom head wear state surrogate model, is constructed and trained based on the Bayesian neural network prediction algorithm. The mushroom head wear state surrogate model is used to efficiently predict the wear amount of the bottom pivot mushroom head based on the dynamic response simulation data of the structure. A mushroom head performance degradation assessment model is established based on Gamma stochastic process theory to describe the long-term stochastic process of wear evolution over time, and historical data is used to initialize and estimate the model parameters. By combining the structural dynamic response data monitored in real time on site, the current wear amount is output through the mushroom head wear state proxy model. Then, the wear amount is used as new observation data, and the parameters of the mushroom head performance degradation assessment model are corrected in real time using the sequential Bayesian method, so as to dynamically update the reliability and remaining life prediction results of the mushroom head wear failure.

[0010] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the bottom pivot mushroom head of the lock as described in this invention, the finite element simulation model uses the Lagrange contact algorithm to simulate the friction of the mushroom head, and calculates the wear amount of the bottom pivot mushroom head using the Archard wear model, which is expressed as follows: in, For wear volume, The wear coefficient is a dimensionless coefficient. For normal load, This is the relative sliding distance. The value represents the material hardness.

[0011] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, the input features of the Bayesian neural network surrogate model are selected from a combination of one or more physical quantities such as stress, strain, vibration response, acoustic emission signal, motor current of the drive mechanism, and water level of the gate, and the output feature is the cumulative wear of the mushroom head.

[0012] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, the Bayesian neural network surrogate model is trained and predicted using a deterministic approximation method of variational inference algorithm. The problem of solving the latent parameter posterior is transformed into a variational optimization problem of maximizing the evidence lower bound ELBO. A better posterior distribution solution is found through iteration, achieving rapid prediction of the bottom pivot wear state. The ELBO formula is: in, Let ELBO be the lower bound of evidence, which is the objective function to be maximized in variational inference. For variational approximation distribution Seeking expectations, The finite element simulation data that serves as input features to the surrogate model, where Z represents latent variables, i.e., the weights and biases of the Bayesian neural network model parameters. For approximating the true posterior distribution The variational approximation distribution, Finite element simulation data given the latent variable Z The likelihood function, Let Z be the prior distribution of the latent variable. For variational approximation distribution With prior distribution of latent variables The Kullback-Leibler divergence between them.

[0013] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, the cumulative wear amount of the mushroom head performance degradation assessment model... The probability density function is: in, In time The cumulative wear amount is The probability density, In time The cumulative wear over time, It is the shape parameter of the Gamma distribution, a function of time, representing the degree of cumulative wear. Let be the scale parameter of the Gamma distribution, reflecting the average magnitude of wear, and e be the base of the natural logarithm. It is the Gamma function.

[0014] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, the initial parameters of the mushroom head performance degradation assessment model are... The estimation is performed using the maximum likelihood estimation method, the objective function of which is: in, Let be the likelihood function. Represents the parameters of the Gamma stochastic process model, i.e. and , Historical observation data used for parameter estimation In time Observed cumulative wear For given parameters Cumulative wear during the Gamma stochastic process The probability density function.

[0015] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, the sequential Bayesian method modifies the parameters of the performance degradation assessment model in real time based on Bayes' theorem. Iterative updates are performed, and their mathematical expression is as follows: in, In order to observe real-time monitoring data After that, parameters The posterior probability distribution, For given parameters In the case of the latest monitoring data The likelihood function, Let be the posterior probability distribution of the previous time step, and be the updated prior probability distribution. Indicates proportionality. These are the parameters for the Gamma stochastic process model.

[0016] As a preferred embodiment of the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention, after dynamically updating the reliability and remaining life prediction results of the mushroom head wear failure, the method further includes formulating and implementing a set of predictive maintenance strategies for the mushroom head, specifically including the following steps: Calculate the cumulative distribution function of remaining lifetime: Based on the real-time updated mushroom head performance degradation assessment model, a predefined wear failure threshold is set. Calculate at the current time After that, remaining lifespan cumulative distribution function This function represents the cumulative wear of the bottom pivot over future time. The failure threshold is reached or exceeded for the first time within the period. The probability of; Establish the objective function for predictive maintenance strategy: Establish a function that minimizes the long-term expected total cost. This function is used to assess when future maintenance is needed. The economic efficiency of preventative maintenance takes the following forms: in, The cost of performing a single preventative maintenance, The cost of repairing after a failure (usually) ), For calculation in time The probability of failure occurring before, and Then it is in time The probability that no failure has occurred before; Determining the optimal timing for preventative maintenance: This involves solving an optimization problem to determine the optimal timing for preventative maintenance. That is, to find the objective function defined in the previous step. Maintenance time point to obtain minimum value : The optimal timing for preventative maintenance It provides a window for optimal maintenance decisions based on cost-benefit analysis.

[0017] As a preferred embodiment of the real-time prediction and reliability assessment method for wear of the mushroom head of the lock bottom pivot as described in this invention, the steps further include formulating and implementing a maintenance strategy, wherein: Based on the calculated optimal timing for preventative maintenance Based on the current running time and status of the bottom pivot, maintenance suggestions are generated; When the actual running time is close to When the system triggers a maintenance alarm, it guides relevant personnel to perform preventative maintenance and feeds the maintenance data back to the historical database after maintenance is completed, for use in the next round of model initialization and updates, forming a closed-loop management process for continuous optimization.

[0018] The beneficial effects of this invention are as follows: This invention establishes a reliable wear-response prior correlation through finite element frictional contact dynamics, constructs a Bayesian neural network surrogate model to achieve probabilistic and rapid estimation of wear amount, obtains a reliability / RUL analytical framework directly related to the failure threshold using Gamma stochastic processes, and achieves continuous adaptive correction for field data using sequential Bayesian methods. This forms a real-time, probabilistic, interpretable, and deployable reliability assessment and life prediction system for the bottom pivot of a herringbone door; it enables online visualization and end-to-end quantitative assessment of wear state, incorporates uncertainty into decision-making, supports maintenance resource optimization and precise planning of downtime windows, and significantly reduces the dependence on offline disassembly and inspection and large-scale simulation. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the real-time prediction and reliability assessment method for the wear of the mushroom head of the lock bottom pivot as described in this invention. Figure 2 This is a schematic diagram of the structure of the Bayesian neural network surrogate model shown in this invention; Figure 3 This is a schematic diagram illustrating the online parameter correction process for the performance degradation model based on the sequence Bayesian method as shown in this invention. Figure 4 This is a schematic diagram of the output interface of the intelligent system shown in this invention. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] This embodiment takes the bottom pivot of the miter gate of an inland waterway lock as an example to illustrate in detail the complete implementation process of a method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot.

[0024] Reference Figure 1 The implementation steps of the method of the present invention are as follows: (1) Establish a finite element model of the lock and obtain simulation data of wear mechanism. In the finite element software ANSYS, a three-dimensional refined finite element model including the miter gate leaf, top pivot, and bottom pivot is established. The material properties, contact parameters, boundary conditions, etc. of the model are set according to the design drawings and material manual. To ensure that the model can truly reflect the mechanical properties of the actual structure, the stress calculation results of the main beam, web, etc. of the finite element model can be compared with the stress monitoring data at the same location on site to cover the opening and closing operation conditions of the miter gate of the lock.

[0025] In the three-dimensional finite element model, by modifying the geometry of the contact surface of the bottom pivot mushroom head, 11 different wear depth levels from healthy to near failure are systematically simulated: 0mm (healthy), 2mm, 4mm, 6mm, 8mm, 10mm, 12mm, 14mm, 16mm, 18mm, and 20mm (near failure). The simulation method adopts transient dynamic analysis. Under each wear level, the upstream water level is applied with loads of 3m, 5m, and 7m respectively to perform transient dynamic simulation of the entire gate opening and closing process. During the simulation, the wear amount in the contact area is calculated and accumulated in real time using the Archard wear model by embedding a user subroutine. The Archard wear model is represented as: in, For wear volume, The wear coefficient is a dimensionless coefficient. For normal load, This is the relative sliding distance. The value represents the material hardness.

[0026] Measuring points are set at key locations such as the door leaf panel, main beam, and bottom pivot base in the simulation model. Time-history data such as vibration acceleration and von Mises stress at these locations during the opening and closing process are extracted. To convert the simulation-obtained time-history data into a format suitable for machine learning, feature engineering is required. The feature extraction method is time-domain statistical analysis. For the extracted vibration acceleration and von Mises stress time-history data x at each measuring point, the root mean square (RMSE), peak value, and kurtosis are calculated. The RMSE reflects the energy magnitude of the signal, and the calculation formula is as follows: The peak value reflects the maximum impact level of the signal, and the calculation formula is: Kurtosis reflects the impulsiveness or outlier distribution of a signal, and is calculated using the following formula: in, Let be the sampled value of the signal at time point i, and N be the total number of sampling points. This represents the signal mean.

[0027] The calculation results of all simulation conditions (11 wear levels × 3 water level conditions) are summarized into a structured feature dataset, which is expressed in the form of a... The numerical matrix, where The total number of samples is 11×3, and the last column is the output target. The total dimension D of the input features is specifically 19. For example, assuming there are 3 key measuring points (door leaf panel, main beam, and bottom pivot base), and 3 statistical features (RMSE, Peak, Kurtosis) of vibration acceleration and von Mises stress are extracted for each measuring point, then each measuring point contributes 6 features, for a total of 18 features, plus 1 feature for the water level condition, i.e. ; For example, the input features are represented as a D-dimensional feature vector. ,For example The output target is expressed as a scalar value. , representing the final cumulative wear depth obtained from the Arcard model simulation under this working condition.

[0028] Finally, the input feature sample dataset is compiled, with each sample containing a set of input features (such as the root mean square value of the vibration signal, peak value, kurtosis, and water level) and the corresponding output target (cumulative wear depth).

[0029] (2) Construct a wear state proxy model Reference Figure 2 A Bayesian neural network was selected as the surrogate model. A Bayesian neural network with two hidden layers and 64 neurons in each layer was constructed. The number of nodes in the input layer corresponds to the number of dynamic response features extracted in step (1) (e.g., 19). The output layer is a single node, representing the cumulative wear amount.

[0030] The dataset generated in step (1) is divided into a training set and a test set in a ratio of 8:2. The Bayesian neural network is trained and predicted using the deterministic approximation method of the variational inference algorithm. The problem of solving the latent parameter posterior is transformed into a variational optimization problem of maximizing the evidence lower bound ELBO. A better posterior distribution solution is found through iteration, so as to realize the rapid prediction of the wear state of the bottom pivot. For example, the specific training method is as follows: initialize the prior distribution of the latent variable Z. The distribution is Gaussian N(0,1); a variational approximation distribution is chosen. The mean field is Gaussian distributed; iterative optimization is performed using the Adam optimizer (learning rate 0.001), and the expectation term of ELBO is estimated in each iteration through Monte Carlo sampling (sampling times S=50). And calculate the Kullback-Leibler divergence term. Train for 100 epochs until ELBO converges (change < 1e-4); the ELBO formula is: in, Let ELBO be the lower bound of evidence, which is the objective function to be maximized in variational inference. For variational approximation distribution Seeking expectations, The finite element simulation data that serves as input features to the surrogate model, where Z represents latent variables, i.e., the weights and biases of the Bayesian neural network model parameters. For approximating the true posterior distribution The variational approximation distribution, Finite element simulation data given the latent variable Z The likelihood function, Let Z be the prior distribution of the latent variable. For variational approximation distribution With prior distribution of latent variables The Kullback-Leibler divergence between them.

[0031] After training, the surrogate model is expressed as a parameterized probabilistic model: given input features The output is the posterior predicted distribution of cumulative wear. , through from Sample to generate multiple predicted values And calculate its mean. and variance As a point estimate and uncertainty quantification, this surrogate model has the ability to quickly provide a probabilistic estimate of wear based on input vibration, stress and other data.

[0032] As an example, the testing method is as follows: input the input features of the test set into the trained Bayesian neural network, obtain the probability distribution of the prediction results through multiple forward propagations, and calculate the prediction mean as the point estimate.

[0033] (3) Establish a performance degradation assessment model Gamma stochastic process was selected to measure wear depth To model, its probability density function for: in, In time The cumulative wear amount is The probability density, In time The cumulative wear over time, It is the shape parameter of the Gamma distribution, a function of time, representing the degree of cumulative wear. Let be the scale parameter of the Gamma distribution, reflecting the average magnitude of wear, and e be the base of the natural logarithm. It is the Gamma function.

[0034] It should be noted that, To initialize the model, prior judgments need to be constructed using historical observation data. Specifically, this involves reviewing the lock's maintenance log and extracting reliable measurement records from previous major overhauls or special inspections regarding the wear of the replenished mushroom head on the bottom pivot. For example, this embodiment collected two data points: a wear depth of 5mm measured during the 2015 major overhaul (5 years after commissioning) and a wear depth of 12mm measured during the 2020 major overhaul (10 years after commissioning). Using these two historical observation data points, the initial estimates of the model parameters are obtained through maximum likelihood estimation (MLE). .

[0035] Furthermore, The calculation process is as follows: First, convert cumulative wear into incremental wear ( Year, ; Year, ); Assumption ,parameter Construct the log-likelihood function: in, It is a wear rate parameter, reflecting the rate at which the degree of wear increases per unit time; It is a scale parameter that controls the scale of wear distribution; It is the duration of the i-th time interval. It is the wear increment within the i-th time period. It is the Gamma function.

[0036] Maximize the function using a numerical optimization algorithm (such as L-BFGS-B) under constraints. , To maximize the above log-likelihood function, set an initial guess value (e.g.) Substitute incremental data Iteratively calculate the gradient and update the parameter values, repeating until convergence (e.g., relative change < 1 × 10⁻). 6 Finally, the parameter estimation results are obtained: ,thereby ,this This constitutes a priori judgment on the long-term wear trend of the bottom pivot mushroom head.

[0037] (4) Online correction and dynamic prediction During actual operation of the lock, vibration signals are collected in real time by accelerometers installed on the bottom pivot base. At time k (e.g., after 2000 cycles), the system automatically collects vibration data for a complete opening and closing cycle, i.e., the entire operation process from the fully closed state of the miter gate, through the opening process, maintaining the open state, the closing process, and finally returning to the fully closed state. The total duration is approximately 10-15 minutes, of which the dynamic opening and closing process takes about 3 minutes. These 3 minutes are the critical period for data acquisition. The system automatically processes the collected 3-minute vibration time history data. The process involves extracting statistical features identical to those in step (1), including root mean square value, peak value, and kurtosis. These features are then combined with the current water level conditions to form a feature vector. .

[0038] Input this vector into the Bayesian neural network surrogate model trained in step (2) to obtain a probabilistic estimate of the current cumulative wear amount, specifically: input the feature vector Normalize the data according to the standardized parameters used during training; then use the trained variational posterior distribution. Sample the network weights M times (e.g., M=100) to obtain network weight samples. For each weighted sample Calculate the corresponding predicted value ; Calculate the mean of the predicted distribution and variance ,For example .

[0039] Reference Figure 3 The wear estimate obtained in the previous step As the latest observational evidence in the sequence Bayesian update formula Using the particle filter algorithm, the parameters of the Gamma process are... The iterative update is performed, and the specific iterative update method is as follows: (a) Initialization / Prediction: From the posterior distribution of the parameters at the previous time step (k-1) N particles (parameter samples) are generated by sampling. For example, if N=1000, since wear is a slowly changing process, the parameters remain constant during the prediction step, i.e. ; (b) Weight calculation: based on the latest observations and its variance Calculate the importance weight for each particle. The observed values ​​are generated by the model parameters represented by the particle. The likelihood is directly proportional to the probability; assuming the observation error follows a Gaussian distribution, the likelihood function is: in, It is composed of particles Predicted average wear; (c) Resampling: Based on the calculated weights, the particle set is resampled; particles with higher weights are more likely to be copied, while particles with lower weights may be discarded. After resampling, a new set of particles with equal weights is obtained. This particle set approximately represents the updated posterior distribution of the parameters. ,like Figure 3 As shown, this process corrects the original predicted trajectory (dashed line) to a new predicted trajectory (solid line) that is closer to the current actual state. Assuming wear failure threshold It is 20mm, based on the corrected parameters. posterior distribution Recalculate the reliability of the bottom pivot at the current moment. The formula is: Predict the probability distribution of remaining useful life (RUL), where RUL is the first time the wear increment reaches the remaining wear amount. The time required for each updated particle Increment at any future time Internal wear increment It follows a Gamma distribution with shape parameters as follows: The scale parameter is The cumulative distribution function (CDF) of RUL is: The final RUL probability density function (PDF) can be obtained by taking the Monte Carlo average of the RUL distributions calculated for all N particles.

[0040] Reference Figure 4 The intelligent monitoring platform's visual interface displays the updated results in real time, such as... Figure 4 As shown, the current reliability is 99.6%, and the probability density function of RUL shows that its expected value is 3.8 years. Since the 5th percentile of RUL (e.g., 3.1 years) is less than the preset next major overhaul cycle (4 years), the system automatically triggers a yellow warning and pushes a report to the operation and maintenance management personnel, suggesting that they conduct a key inspection of the bottom pivot during the next maintenance window, providing decision support for them to formulate a scientific predictive maintenance strategy.

[0041] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for real-time prediction and reliability assessment of wear on the mushroom head of a lock bottom pivot, characterized in that, include: A finite element simulation model of the friction pair between the gate leaf and the bottom pivot of the lock was established. The dynamic simulation of frictional contact was considered to simulate the dynamic wear of the bottom pivot mushroom head during the opening and closing process of the gate, so as to obtain the wear amount of the mushroom head and the corresponding structural dynamic response simulation data under different working conditions. Using the acquired finite element simulation data, a Bayesian neural network surrogate model, namely the mushroom head wear state surrogate model, is constructed and trained based on the Bayesian neural network prediction algorithm. The mushroom head wear state surrogate model is used to efficiently predict the wear amount of the mushroom head based on the dynamic response simulation data of the structure. A mushroom head performance degradation assessment model is established based on Gamma stochastic process theory to describe the long-term stochastic process of wear evolution over time, and historical data is used to initialize and estimate the model parameters. By combining the structural dynamic response data monitored in real time on site, the current wear amount is output through the mushroom head wear state proxy model. Then, the wear amount is used as new observation data, and the parameters of the mushroom head performance degradation assessment model are corrected in real time using the sequential Bayesian method, so as to dynamically update the reliability and remaining life prediction results of the mushroom head wear failure.

2. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 1, characterized in that, The finite element simulation model uses the Lagrange contact algorithm to simulate the friction of the mushroom head, and calculates the wear of the bottom pivot mushroom head using the Archard wear model, which is expressed as follows: ； in, For wear volume, The wear coefficient is a dimensionless coefficient. For normal load, This is the relative sliding distance. The value represents the material hardness.

3. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 1 or 2, characterized in that, The input features of the Bayesian neural network surrogate model are selected from one or more physical quantities such as stress, strain, vibration response, acoustic emission signal, motor current of the drive mechanism, and water level of the gate, and the output feature is the cumulative wear of the mushroom head.

4. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 3, characterized in that, The Bayesian neural network surrogate model is trained and predicted using a deterministic approximation method of variational inference algorithm. Solving the latent parameter posterior problem is transformed into a variational optimization problem of maximizing the evidence lower bound ELBO. Through iteration, a better posterior distribution solution is found, enabling rapid prediction of the wear state of the bottom pivot. The ELBO formula is: ； in, Let ELBO be the lower bound of evidence, which is the objective function to be maximized in variational inference. For variational approximation distribution Seeking expectations, The finite element simulation data that serves as input features to the surrogate model, where Z represents latent variables, i.e., the weights and biases of the Bayesian neural network model parameters. For approximating the true posterior distribution The variational approximation distribution, Finite element simulation data given the latent variable Z The likelihood function, Let Z be the prior distribution of the latent variable. For variational approximation distribution With prior distribution of latent variables The Kullback-Leibler divergence between them.

5. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 1, characterized in that, The cumulative wear amount of the mushroom head performance degradation assessment model The probability density function is: ； in, In time The cumulative wear amount is The probability density, In time The cumulative wear over time, It is the shape parameter of the Gamma distribution, a function of time, representing the degree of cumulative wear. Let be the scale parameter of the Gamma distribution, reflecting the average magnitude of wear, and e be the base of the natural logarithm. It is the Gamma function.

6. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 1 or 5, characterized in that, The initial parameters of the mushroom head performance degradation assessment model The estimation is performed using the maximum likelihood estimation method, the objective function of which is: ； in, Let be the likelihood function. Represents the parameters of the Gamma stochastic process model, i.e. and , Historical observation data used for parameter estimation In time Observed cumulative wear For given parameters Cumulative wear during Gamma stochastic process The probability density function.

7. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 6, characterized in that, The sequential Bayesian method uses Bayes' theorem to continuously adjust the parameters of the performance degradation evaluation model. Iterative updates are performed, and their mathematical expression is as follows: ； in, In order to observe real-time monitoring data After that, parameters The posterior probability distribution, For given parameters In the case of the latest monitoring data The likelihood function, Let be the posterior probability distribution of the previous time step, and be the updated prior probability distribution. Indicates proportionality. These are the parameters for the Gamma stochastic process model.

8. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 1, characterized in that, After dynamically updating the reliability and remaining life prediction results of the mushroom head wear failure, the system also includes developing and implementing a predictive maintenance strategy for the mushroom head, specifically including the following steps: Calculate the cumulative distribution function of remaining lifetime: Based on the real-time updated mushroom head performance degradation assessment model, a predefined wear failure threshold is set. Calculate at the current time After that, remaining lifespan cumulative distribution function This function represents the cumulative wear of the bottom pivot over future time. The failure threshold is reached or exceeded for the first time within the period. The probability of; Establish the objective function for predictive maintenance strategy: Establish a function that minimizes the long-term expected total cost. This function is used to assess when future maintenance is needed. The economic efficiency of preventative maintenance takes the following specific forms: ； in, The cost of performing a single preventative maintenance, The cost of repairing after a failure (usually) ), For calculation in time The probability of failure occurring before, and Then it is in time The probability that no failure has occurred before; Determining the optimal timing for preventative maintenance: This involves solving an optimization problem to determine the optimal timing for preventative maintenance. That is, to find the objective function defined in the previous step. Maintenance time point to obtain minimum value : ； The optimal timing for preventative maintenance It provides a window for optimal maintenance decisions based on cost-benefit analysis.

9. The method for real-time prediction and reliability assessment of wear of the mushroom head of the lock bottom pivot according to claim 8, characterized in that, The steps also include developing and implementing a maintenance strategy, wherein: Based on the calculated optimal timing for preventative maintenance Based on the current running time and status of the bottom pivot, maintenance suggestions are generated; When the actual running time is close to When the system triggers a maintenance alarm, it guides relevant personnel to perform preventative maintenance and feeds the maintenance data back to the historical database after maintenance is completed, for use in the next round of model initialization and updates, forming a closed-loop management process for continuous optimization.

Images