A dynamic load adaptive reliability optimization method and system for a marine propulsion device

Abstract

This invention discloses a dynamic load adaptive reliability optimization method and system for ship propulsion systems. The method includes: collecting real-time operating data of the ship propulsion system; constructing a dynamic reliability assessment model of the propulsion system based on an improved Weibull proportional hazards function and a nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); calculating the dynamic load spectrum correction coefficient κ based on the collected real-time operating data; constructing a multi-objective optimization function based on λ(t) and κ; verifying the optimization results through a digital twin model; and writing the final optimized parameter set into the propulsion system control system. This invention achieves synergistic optimization of the reliability and efficiency of ship propulsion systems, and has broad application prospects and significant engineering value.

Description

Technical Field

[0001] This invention belongs to the field of marine mechanical engineering, and in particular relates to a dynamic load adaptive reliability optimization method and system for ship propulsion devices. Background Technology

[0002] With the rapid development of the global shipping industry, the reliability, efficiency, and environmental friendliness of ship propulsion systems have become key factors restricting their performance. As the core power source for ship operation, the performance of the ship propulsion system directly affects the ship's navigation safety, economy, and environmental friendliness. However, due to the complex and variable operating environment of ships, propulsion systems face various failure risks during long-term operation, such as shaft fatigue, bearing wear, lubricating oil leakage, vibration, and noise. Therefore, how to improve the reliability and efficiency of ship propulsion systems through optimized control strategies and modeling methods has become an important research direction in the field of marine engineering.

[0003] Currently, reliability assessment of ship propulsion systems typically relies on traditional methods such as fault tree analysis (FTA) and finite element analysis (FEM). However, these methods have limitations in handling nonlinear degradation processes, multi-source data fusion, and dynamic environmental adaptability. For example, traditional reliability assessment models are often based on static parameters, making it difficult to reflect the coupling effects of dynamic variables such as shaft torque, bearing temperature, and vibration spectrum during actual operation. Furthermore, existing optimization methods mostly employ static optimization algorithms (such as NSGA-II), and their convergence speed and solution uniformity still need improvement when facing high-dimensional, nonlinear, and multi-objective optimization problems. Meanwhile, the application of reinforcement learning in ship propulsion systems is still in the exploratory stage; how to combine it with physical models to achieve online adaptive control remains a pressing issue.

[0004] In recent years, with the development of artificial intelligence, big data, and digital twin technologies, optimization methods for ship propulsion systems have gradually moved towards intelligence and dynamism. For example, deep learning-based vibration spectrum analysis, reinforcement learning-based online control strategies, and real-time monitoring systems based on multi-source sensor data fusion have achieved preliminary results in some studies. However, a systematic and integrated adaptive reliability optimization method for dynamic loads in ship propulsion systems is still lacking, capable of simultaneously addressing key aspects such as failure risk prediction, dynamic load correction, multi-objective optimization solutions, and online adaptive adjustment. Therefore, it is necessary to propose an adaptive reliability optimization method for dynamic loads in ship propulsion systems based on multi-source data-driven approaches and integrating physical models and intelligent algorithms to improve its adaptability and reliability under complex operating conditions. Summary of the Invention

[0005] To address the shortcomings of the existing technology, this invention provides a dynamic load adaptive reliability optimization method for ship propulsion devices, comprising the following steps: Step S101: Collect real-time operating data of the ship's propulsion system; Step S103: Construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); Step S105: Calculate the dynamic load spectrum correction coefficient κ based on the real-time running data collected in step S101; Step S107: Based on λ(t) from step S103 and κ from step S105, construct a multi-objective optimization function; Step S109: Verify the optimization results using a digital twin model, and write the final optimized parameter set into the propulsion device control system.

[0006] The real-time operating data includes propeller speed ω, shaft torque τ, and bearing temperature T. b Lubricating oil pressure P l Vibration acceleration spectrum A(f), propulsion efficiency η p and the turbulence intensity ε in the wake field w .

[0007] The instantaneous failure risk rate λ(t) mentioned in step S103 is calculated using the following formula: Where β is the shape parameter, η is the scale parameter, γ is the degenerate coupling coefficient, σ is the Gaussian kernel bandwidth, and D(u) is the coefficient based on shaft torque τ and bearing temperature T. b The nonlinear degradation quantity is expressed as follows: α1, α2, and α3 are degradation weighting coefficients.

[0008] The weighting coefficients α1, α2, and α3 of the nonlinear degradation quantity D(u) are solved using an improved gray wolf optimization algorithm, specifically including: (a) Establish a fitness function with the shaft system historical fault data as input, the expression of which is the sum of mean square error and L1 regularization term; (b) Introduce Tent chaotic mapping to initialize the population and enhance global search capabilities; (c) The exploration-exploitation balance of the Grey Wolf algorithm is adjusted by a dynamic weighting mechanism, and the weight update formula includes the hyperbolic tangent function.

[0009] Specifically, step S105 includes: An adaptive Kalman filter and a long short-term memory (LSTM) network are used as a joint algorithm to perform mode decomposition based on the vibration acceleration spectrum A(f) acquired in step S101, and extract the first k natural frequencies f. k and damping ratio ζ k And combined with the turbulence intensity ε of the wake field w Calculate the dynamic load spectrum correction factor κ.

[0010] The expression for the dynamic load spectrum correction coefficient κ is as follows: .

[0011] In step S107, the NSGA-II algorithm is used to solve for the optimal control parameter set {ω,τ,P}. l }

[0012] In step S109, the digital twin model uses a coupled CFD-FEM method to simulate the stress-strain field of the propulsion device under a κ-corrected load spectrum, and outputs the fatigue life N of the corresponding components. f .

[0013] The method further includes: if N f Below the preset threshold N th This triggers an online adjustment strategy based on reinforcement learning, dynamically updating the control parameter set in step S107.

[0014] This invention also proposes a dynamic load adaptive reliability optimization system for ship propulsion devices, comprising: The data acquisition module is used to collect real-time operating data of the ship's propulsion system. The dynamic reliability assessment module is used to construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); The dynamic load coefficient acquisition module is used to calculate the dynamic load spectrum correction coefficient κ based on the collected real-time running data. The multi-objective optimization module is used to construct a multi-objective optimization function based on the instantaneous failure risk rate λ(t) and the dynamic load spectrum correction coefficient κ; The verification module is used to verify the optimization results through a digital twin model and write the final optimized parameter set into the propulsion device control system.

[0015] Compared with the prior art, the present invention has the following advantages: Real-time monitoring driven by multi-source data. This method acquires real-time operational data of the ship's propulsion system using a multi-source sensor array (such as propeller speed sensors, shaft torque sensors, bearing temperature sensors, and vibration acceleration sensors). Combined with anti-aliasing filtering and synchronous compressed wavelet transform denoising techniques, the reliability and consistency of the data are improved. This approach effectively captures the dynamic response of the propulsion system under different operating conditions, providing high-quality data support for subsequent reliability assessment and optimization.

[0016] A dynamic reliability assessment model was constructed. An improved Weibull proportional hazards function was used to fuse the model with a nonlinear degradation process, incorporating nonlinear degradation parameters of shaft torque and bearing temperature. An improved gray wolf optimization algorithm was then employed to solve for the degradation weight coefficients. This model accurately reflects the failure risk rate of the propulsion system under different operating conditions, providing a theoretical basis for subsequent multi-objective optimization.

[0017] The introduction of a multi-objective optimization algorithm. Based on the NSGA-II algorithm, this method combines simulated binary crossover and polynomial mutation operators, and introduces crowding calculation based on Mahalanobis distance to improve the uniformity of the Pareto front distribution. This method can achieve a balance between minimizing the failure risk rate and maximizing propulsion efficiency, outputting an optimal set of control parameters.

[0018] Validation and feedback of the digital twin model. The stress-strain field of the propulsion system under a modified load spectrum is simulated using a coupled CFD-FEM method, outputting the fatigue life of key components. If the fatigue life falls below a preset threshold, an online adjustment strategy based on reinforcement learning is triggered to dynamically update the control parameter set, achieving closed-loop optimization.

[0019] Online adaptive control using reinforcement learning. A reinforcement learning strategy is constructed using the DDPG algorithm. The reward function design comprehensively considers key indicators such as failure risk rate, propulsion efficiency, and lubricating oil pressure to achieve dynamic adaptive control of the propulsion system. This method can effectively cope with uncertainties under complex operating conditions and improve the robustness of the system.

[0020] Multiphysics collaborative optimization. By employing a multiphysics sensor array, a dynamic compensation algorithm for guide wheel angle, and an energy efficiency optimization decision model, combined with deep learning algorithms, a mapping relationship between the propeller wake field and guide wheel parameters is established, achieving a dynamic response within 0.1 seconds. This method can significantly reduce propulsion energy consumption and improve propulsion efficiency, demonstrating good economic benefits and environmental value.

[0021] Model interpretability and generalizability. A parameter interpretation mechanism based on the physical model is introduced to enhance the interpretability and generalizability of the model, ensuring the applicability of the optimization results under different ship types and operating environments.

[0022] Experimental verification and standard drafting. Through shipboard testing and digital twin model verification, a draft standard for accelerated reliability testing of marine transmission devices was developed, providing technical support for the development and reliability verification of ship power systems in my country. Attached Figure Description

[0023] The above and other objects, features, and advantages of exemplary embodiments of the present disclosure will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the present disclosure are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating a dynamic load adaptive reliability optimization method for a ship propulsion device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0025] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0026] It should be understood that although the terms first, second, third, etc., may be used to describe... in the embodiments of the present invention, these... should not be limited to these terms. These terms are only used to distinguish... For example, first... may also be referred to as second... without departing from the scope of the embodiments of the present invention, and similarly, second... may also be referred to as first...

[0027] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0028] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0029] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0030] The optional embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] Example 1 like Figure 1 As shown, this invention discloses a dynamic load adaptive reliability optimization evaluation method for ship propulsion devices, comprising the following steps: Step S101: Collect real-time operating data of the ship's propulsion system; Step S103: Construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); Step S105: Calculate the dynamic load spectrum correction coefficient κ based on the real-time running data collected in step S101; Step S107: Based on λ(t) from step S103 and κ from step S105, construct a multi-objective optimization function; Step S109: Verify the optimization results using a digital twin model, and write the final optimized parameter set into the propulsion device control system.

[0032] Example 2 The present invention proposes a dynamic load adaptive reliability optimization method for ship propulsion devices, comprising the following steps: Step S101: Collect real-time operating data of the ship's propulsion system; Step S103: Construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); Step S105: Calculate the dynamic load spectrum correction coefficient κ based on the real-time running data collected in step S101; Step S107: Based on λ(t) from step S103 and κ from step S105, construct a multi-objective optimization function; Step S109: Verify the optimization results using a digital twin model, and write the final optimized parameter set into the propulsion device control system.

[0033] The real-time operating data includes propeller speed ω, shaft torque τ, and bearing temperature T. b Lubricating oil pressure P l Vibration acceleration spectrum A(f), propulsion efficiency η p and the turbulence intensity ε in the wake field w .

[0034] The instantaneous failure risk rate λ(t) mentioned in step S103 is calculated using the following formula: Where β is the shape parameter, η is the scale parameter, γ is the degenerate coupling coefficient, σ is the Gaussian kernel bandwidth, and D(u) is the coefficient based on shaft torque τ and bearing temperature T. b The nonlinear degradation quantity is expressed as follows: α1, α2, and α3 are degradation weighting coefficients.

[0035] The weighting coefficients α1, α2, and α3 of the nonlinear degradation quantity D(u) are solved using an improved gray wolf optimization algorithm, specifically including: (a) Establish a fitness function with the shaft system historical fault data as input, which is expressed as the sum of mean square error and L1 regularization term.

[0036] In this invention, an improved Grey Wolf optimization algorithm is used, and the fitness function is an important tool for evaluating the merits of candidate solutions. According to the problem description, the fitness function is expressed as the sum of the mean squared error (MSE) and the L1 regularization term. The MSE measures the difference between predicted and actual values, while the L1 regularization term helps prevent overfitting and introduces sparsity. This combination not only effectively evaluates the model's fit but also controls the model's complexity to some extent, thereby improving the model's generalization ability.

[0037] (b) Introduce Tent chaotic mapping to initialize the population and enhance global search capabilities.

[0038] In traditional gray wolf optimization algorithms, the initial population is typically generated randomly, which can lead to uneven population distribution and affect the algorithm's global search capability. To address this issue, the improved gray wolf optimization algorithm of this invention introduces the Tent chaotic map to initialize the population. The Tent chaotic map possesses the characteristics of randomness, regularity, and ergodicity, enabling the generation of an initial population that is uniformly distributed in the search space, thereby improving the algorithm's global search capability. Furthermore, the introduction of the Tent chaotic map also enhances population diversity, preventing the algorithm from prematurely getting trapped in local optima.

[0039] (c) The exploration-exploitation balance of the Grey Wolf algorithm is adjusted by a dynamic weighting mechanism, and the weight update formula includes the hyperbolic tangent function.

[0040] In the Grey Wolf Optimization Algorithm, the balance between exploration and exploitation is a key factor affecting algorithm performance. To better balance this process, the improved Grey Wolf Optimization Algorithm introduces a dynamic weight mechanism. This mechanism adjusts the weights of individual grey wolves, causing the algorithm to favor global search in the early stages and local search in the later stages, thereby improving the algorithm's convergence speed and accuracy. Specifically, the weight update formula includes a hyperbolic tangent function (tanh), which dynamically adjusts the weights based on the current iteration number, thus achieving a smooth transition between exploration and exploitation.

[0041] The improved Weibull proportional hazards function also introduces a covariate Z(t) to represent the lubricating oil pressure P. l The time-varying influence takes the form of P c This is the critical pressure threshold.

[0042] In traditional proportional hazards models, covariates are typically fixed values ​​or variables that change slowly over time. However, in practical engineering applications, many key parameters (such as lubricating oil pressure) change dynamically over time. Therefore, it is necessary to introduce a time-dependent covariate Z(t). For example, lubricating oil pressure P... l This risk may gradually decrease as the equipment operates for longer periods, thus affecting the equipment's failure risk.

[0043] In this embodiment of the invention, Z(t) is defined as: Among them, P l This is the current lubricating oil pressure, P. c It is the critical pressure threshold. When P l Approaching P c When the time elapses, the value of Z(t) increases, indicating that the influence of lubricating oil pressure on equipment lifespan is enhanced. This form of Z(t) can capture the nonlinear effect of lubricating oil pressure changing with time.

[0044] In proportional hazards models, time-dependent covariates can be introduced in several ways. A common approach is to incorporate the interaction between the covariate and time into the model. For example, the Cox proportional hazards model includes time-dependent covariates.

[0045] In this embodiment of the invention, Z(t) is itself a time-dependent covariate because it directly depends on the lubricating oil pressure P at the current time t. l Therefore, Z(t) can be directly used as a covariate X. i A portion of it is incorporated into the exponential term of the model.

[0046] Based on the above discussion, the improved Weibull proportional hazards model can be expressed as: in, These are covariate components, including lubricating oil pressure P. l The time-varying effects of lubricating oil pressure are introduced. By incorporating this, the model can more accurately reflect the impact of lubricating oil pressure on equipment life, thereby improving the accuracy of predictions.

[0047] The Z(t) introduced in this invention can also be used to predict the remaining lifespan of equipment. By estimating model parameters using historical monitoring data and verifying the feasibility of the model using simulation examples, the accuracy of remaining lifespan prediction can be improved.

[0048] Specifically, step S105 includes: An adaptive Kalman filter and a long short-term memory (LSTM) network are used as a joint algorithm to perform mode decomposition based on the vibration acceleration spectrum A(f) acquired in step S101, and extract the first k natural frequencies f. k and damping ratio ζ k And combined with the turbulence intensity ε of the wake field w Calculate the dynamic load spectrum correction factor κ.

[0049] The input layer of the LSTM network contains a vibration spectrum sequence processed by a time sliding window. The window length is determined by the mutual information entropy and the permutation entropy. The output layer uses the Softmax function to classify the critical states of the vibration modes.

[0050] LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) specifically designed for processing sequential data, such as time series and text. In this invention, the input layer receives a vibration spectrum sequence processed by a time sliding window. This means the input data is a vibration signal processed by a time window, with each window representing the vibration spectrum characteristics over a period of time. This processing method helps to capture local features and time dependencies in the vibration signal.

[0051] Sliding window. A sliding window is a commonly used time series processing method that extracts features from different time periods by moving a window across the time series. In this invention, the window length is jointly determined by mutual information entropy and permutation entropy. Mutual information entropy measures the correlation of information between different time points, while permutation entropy assesses the complexity and disorder of the time series. By using these two entropy values ​​together, the window length can be selected more accurately to retain sufficient information and avoid overfitting.

[0052] Input Dimension. The input to an LSTM is typically a three-dimensional structure, namely [batch size, time steps, attribute_dimension], where time steps is the time step size and attribute_dimension is the feature dimension of each time step. In this invention, the input features are vibration spectrum sequences, therefore attribute_dimension corresponds to multiple frequency components of the spectrum.

[0053] The core of LSTM lies in its gating mechanism, including forget gates, input gates, and output gates. These gating units control the flow of information through sigmoid and tanh functions, thereby enabling the modeling of long-term dependencies.

[0054] The forgetting gate determines which information needs to be forgotten in order to maintain the stability of memory in long sequences.

[0055] The input gate determines which new information needs to be added to memory.

[0056] The output gate determines which information to retrieve from memory and pass to the next level.

[0057] The hidden state h of LSTM t and cell state c t It is key to its processing of sequence data. Hidden state h t The output at the current time step is calculated using an activation function (such as tanh), while the cell state c... t The information is then updated through a gating mechanism to retain or discard it.

[0058] After the LSTM processes the input sequence, the output layer is typically a fully connected layer, whose output is classified using the Softmax function. The Softmax function maps the input to a probability distribution such that the output value for each class is between 0 and 1, and the sum of the probabilities of all classes is 1. This allows the model to output the probability of each class, thus enabling multi-class classification tasks.

[0059] Output dimension, if the number of nodes in the output layer is num classesEach node corresponds to a category. In this invention, the output layer is used to classify the critical states of vibration modes, therefore num classes It equals the number of vibration modes.

[0060] The connection method allows LSTM output to be either two-dimensional or three-dimensional, depending on whether a complete sequence is returned. If `return` is set... sequence = TRUE, then the output is three-dimensional, i.e., [seq len batch, hidden size If set to FALSE, the output is two-dimensional, i.e., [batch, hidden]. size In classification tasks, the return option is usually chosen. sequence = FALSE, because only the output of the last time step is needed as the classification criterion.

[0061] In vibration modal analysis, LSTM networks can effectively process time series data, capture long-term dependencies in vibration signals, and improve the robustness and accuracy of the model by determining the window length through sliding windows and entropy values.

[0062] The input layer of the LSTM network processes the vibration spectrum sequence through a time-sliding window. The window length is jointly determined by mutual information entropy and permutation entropy to retain key information and avoid overfitting. The gating mechanism of LSTM enables it to effectively capture long-term dependencies and transmit information through hidden states and cell states. Finally, the output layer uses a softmax function for classification, outputting the critical state probabilities of the vibration modes.

[0063] The vibration acceleration spectrum A(f) is acquired using a combined denoising method of anti-aliasing filtering and synchronous compressed wavelet transform, and its cutoff frequency is set according to three times the maximum rotational speed of the propulsion device.

[0064] In the acquisition of the vibration acceleration spectrum A(f), a combined denoising method of anti-aliasing filtering and synchronous compressed wavelet transform was adopted, with the cutoff frequency set according to three times the maximum rotational speed of the propulsion device. The purpose of this method is to ensure that high-frequency noise and aliasing effects can be effectively suppressed during the acquisition process, thereby improving the accuracy of spectrum analysis.

[0065] Anti-aliasing filtering is a preprocessing step performed before signal sampling to remove signal components above the Nyquist frequency (half the sampling frequency) to prevent frequency aliasing. Frequency aliasing occurs when signal components above half the sampling frequency are incorrectly mapped to a lower frequency range, leading to spectral distortion. Anti-aliasing filters are typically low-pass filters with a cutoff frequency of... It should be set to the highest frequency of the useful signal to ensure that signal components above that frequency are filtered out.

[0066] The Synchronized Compressed Wavelet Transform (SCWT) is a nonlinear transform method that effectively compresses the spectral energy of a signal, thereby improving its signal-to-noise ratio. In the spectral analysis of vibration signals, SCWT can be used to remove high-frequency noise while preserving useful low-frequency signal components. This method is particularly suitable for processing non-stationary signals, such as vibration signals, because it can capture the instantaneous frequency characteristics of the signal.

[0067] In this invention, the cutoff frequency of the anti-aliasing filter is set according to three times the maximum rotational speed of the propulsion device. This means that if the maximum rotational speed of the propulsion device is N, then the cutoff frequency f of the anti-aliasing filter is... c It should be set to 3N. This setting is to ensure that all possible high-frequency noise components (such as vibrations caused by the propulsion device) are filtered out during the sampling process, thereby avoiding aliasing.

[0068] The combined use of anti-aliasing filtering and synchronous compressed wavelet transform combines the advantages of both methods. Anti-aliasing filtering removes high-frequency noise before sampling, while synchronous compressed wavelet transform further optimizes the signal's spectral characteristics after sampling. This combined approach not only effectively suppresses aliasing effects but also improves the accuracy and reliability of spectral analysis.

[0069] The expression for the dynamic load spectrum correction coefficient κ is as follows: .

[0070] In step S107, the NSGA-II algorithm is used to solve for the optimal control parameter set {ω,τ,P}. l }

[0071] The NSGA-II algorithm employs simulated binary crossover as its crossover operator, polynomial mutation as its mutation operator, and introduces crowding calculation based on Mahalanobis distance to improve the uniformity of the Pareto front distribution.

[0072] Simulated binary crossover (SBX) is a commonly used crossover operator for generating new individuals. Its core idea is to control the similarity between offspring and parents using a distribution index. Specifically, SBX uses a random variable to determine the crossover point between two parent individuals, thus generating two offspring individuals.

[0073] Multinomial mutation (PM) is a commonly used mutation operator used to introduce random perturbations into a population to enhance the exploratory capabilities of an algorithm. Its basic idea is to control the degree of mutation using a distribution index.

[0074] Crowding density calculation measures the density of individuals in the target space, ensuring a uniform distribution of solutions on the Pareto front. Traditional crowding density calculation methods are typically based on Euclidean distance, but to improve computational efficiency and accuracy, NSGA-II introduced a crowding density calculation method based on Mahalanobis distance. Mahalanobis distance considers the correlation between different targets and can more accurately reflect the differences between individuals.

[0075] By introducing simulated binary crossover and polynomial mutation, NSGA-II can improve the algorithm's convergence speed and search capability while maintaining population diversity. Furthermore, the crowding calculation method based on Mahalanobis distance can more effectively maintain the uniformity of solution distribution on the Pareto front, thereby improving the overall performance of the algorithm.

[0076] In step S109, the digital twin model uses a coupled CFD-FEM method to simulate the stress-strain field of the propulsion device under a κ-corrected load spectrum, and outputs the fatigue life N of the corresponding components. f .

[0077] In stress-strain field simulation of propulsion devices, commonly used numerical methods include computational fluid dynamics (CFD) and finite element analysis (FEM). CFD is used to simulate fluid flow, while FEM is used to simulate the mechanical response of the structure. To more accurately simulate the behavior of propulsion devices under complex operating conditions, a CFD-FEM coupled method, namely the fluid-structure interaction (FSI) model, is usually adopted.

[0078] The CFD section uses CFD simulation to measure the fluid flow around the propulsion device and calculates the pressure distribution and velocity changes acting on the structure.

[0079] The FEM part uses the fluid load provided by CFD to calculate the stress, strain and displacement of the structure.

[0080] The coupling mechanism allows CFD and FEM to exchange data through an interface, enabling the interaction between the fluid and the structure. For example, the pressure and velocity of the fluid affect the deformation of the structure, and the deformation of the structure, in turn, changes the flow state of the fluid.

[0081] This coupling method can more realistically reflect the dynamic behavior of the propulsion device in actual operation, especially in high-precision simulations, where it can capture nonlinear effects such as aeroelastic vibration and structural deflection.

[0082] In fatigue life prediction of propulsion systems, the load spectrum is a crucial parameter describing the load history experienced by the equipment during operation. The κ-corrected load spectrum is a method used to correct the traditional load spectrum, aiming to more accurately reflect load variations during actual operation.

[0083] The principle of κ correction: κ correction is typically used to adjust the statistical characteristics of the load spectrum to better reflect actual operating conditions. For example, by introducing a κ factor, the mean, variance, or kurtosis of the load can be adjusted, thereby more accurately simulating the load distribution of the equipment under different operating conditions.

[0084] In digital twin models, the κ-corrected load spectrum can be used as an input parameter to simulate the stress-strain response of the propulsion device under different operating conditions. By introducing κ correction, the accuracy of the load spectrum can be improved, thereby enhancing the reliability of fatigue life prediction.

[0085] In digital twin models, fatigue life prediction is typically based on the following steps: Data acquisition and modeling involve collecting real-time operational data of the propulsion device, such as pressure, temperature, and vibration, through sensors. This data is then input into a digital twin model to update the parameters of the virtual model.

[0086] CFD-FEM coupled simulation utilizes the CFD-FEM coupling method to simulate the stress-strain field of a propulsion device under a κ-corrected load spectrum. Fluid loads are calculated using CFD, and structural response is calculated using FEM, thereby obtaining the stress history of key components.

[0087] Fatigue life calculation is based on the Rainflow counting method and Wöhler curves. The simulated stress history is analyzed to calculate the fatigue life of the material. The Rainflow counting method is used to count the number of stress cycles and stress amplitude, while the Wöhler curves are used to evaluate the fatigue life under different stress amplitudes.

[0088] Model updates and predictions are achieved by comparing actual operating data with simulation results using Bayesian algorithms or other machine learning methods, thereby updating the parameters of the digital twin model and improving the accuracy of predictions.

[0089] The fatigue life N f The calculation adopts Miner's linear cumulative damage theory and combines λ(t) for Bayesian probability correction.

[0090] Miner's theory is a classic linear fatigue cumulative damage theory. Its core idea is that under different stress levels, the fatigue damage a material experiences is a linear sum of the damage at each stress level. When the total cumulative damage reaches 1, fatigue failure occurs. This theory is applicable to multi-condition fatigue analysis; that is, under different load conditions, the fatigue life of a material can be predicted by accumulating the damage at each stage.

[0091] In practical engineering, fatigue life N f Calculations are typically based on the material's SN curve (stress-life curve), which is the fatigue life N of the material under a specific stress amplitude. f The relationship with stress amplitude. For example, for metallic materials, fatigue life. It can be represented as: Where C and m are fatigue characteristic parameters of the material.

[0092] Calculation steps: Based on the load spectrum, the complex load is decomposed into cyclic loads at multiple stress levels; For each stress level, calculate its corresponding fatigue life. ; Calculate the damage proportion at each stress level ; Add up all the damage proportions to determine if the failure threshold D=1 has been reached.

[0093] Traditional Miner's theory assumes that all load effects are independent and that damage is linearly cumulative. However, in practical engineering, load effects often have a time dependence, such as the duration and frequency of the load, and environmental factors, which can affect the fatigue life of materials.

[0094] The role of the instantaneous failure risk rate λ(t) is introduced. λ(t) is a load effect coefficient used to correct the linear damage accumulation assumption in Miner's theory. It considers the impact of load history (such as time, frequency, and environmental conditions) on fatigue life. For example, the instantaneous failure risk rate λ(t) can reflect the cumulative effect of the load at different time points, thus more accurately predicting the fatigue life of the material.

[0095] The revised formula, with the introduction of the instantaneous failure risk rate λ(t), yields the following linear damage accumulation model: ,in, It is the load effect coefficient under the i-th stress level. When At that point, the formula degenerates into standard Miner's theory.

[0096] The Bayesian method is a method that updates probability distributions based on prior knowledge and observed data. In fatigue life prediction, the Bayesian method can be used to update the fatigue life distribution of materials, thereby improving prediction accuracy. For example, by introducing a prior distribution and a likelihood function, the fatigue life N of the material can be estimated. f The posterior distribution is then calculated, and its expected value and confidence interval are determined.

[0097] The method further includes: if N f Below the preset threshold N th This triggers an online adjustment strategy based on reinforcement learning, dynamically updating the control parameter set in step S107.

[0098] After triggering the online policy adjustment in reinforcement learning, the system dynamically updates the control parameter set based on the output of the reinforcement learning algorithm. This process typically includes the following steps: State observation involves the system collecting the current environmental state and control parameters in real time, which serve as input for reinforcement learning algorithms.

[0099] Policy evaluation: Based on the current state, the reinforcement learning algorithm evaluates the performance of the current policy and calculates the expected reward.

[0100] Policy update: Based on the evaluation results, the algorithm adjusts the policy parameters and generates new control actions.

[0101] Execution and feedback: The system executes new control actions and feeds the results back to the reinforcement learning algorithm, forming a closed-loop control.

[0102] The method further includes: writing the final optimized parameter set into the propulsion device control system, and monitoring λ(t) and η in real time. p If the deviation rate exceeds 5%, return to step S101.

[0103] Example 3 This invention also proposes a dynamic load adaptive reliability optimization system for ship propulsion devices, comprising: The data acquisition module is used to collect real-time operating data of the ship's propulsion system. The dynamic reliability assessment module is used to construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); The dynamic load coefficient acquisition module is used to calculate the dynamic load spectrum correction coefficient κ based on the collected real-time running data. The multi-objective optimization module is used to construct a multi-objective optimization function based on the instantaneous failure risk rate λ(t) and the dynamic load spectrum correction coefficient κ; The verification module is used to verify the optimization results through a digital twin model and write the final optimized parameter set into the propulsion device control system.

[0104] The system described above in this invention aims to achieve synergistic optimization of the reliability and efficiency of ship propulsion devices through multi-source data acquisition, dynamic reliability assessment, dynamic load correction, multi-objective optimization, and digital twin verification. The system consists of five core modules: a data acquisition module, a dynamic reliability assessment module, a dynamic load coefficient acquisition module, a multi-objective optimization module, and a verification module. These modules exchange data and transmit signaling through standardized interfaces to ensure efficient system operation and closed-loop control.

[0105] The data acquisition module is responsible for collecting real-time operating data of the ship's propulsion system, including propeller speed ω, shaft torque τ, and bearing temperature T. b Lubricating oil pressure P l Vibration acceleration spectrum A(f), propulsion efficiency η p and the turbulence intensity ε in the wake field w These data are collected using a multi-source sensor array (such as propeller speed sensor, shaft torque sensor, bearing temperature sensor, vibration acceleration sensor, etc.), and the reliability and consistency of the data are improved by a joint denoising method of anti-aliasing filtering and synchronous compressed wavelet transform.

[0106] Data acquisition: Operating parameters are collected in real time through various sensors deployed on the ship's propulsion system.

[0107] Data preprocessing: Anti-aliasing filtering and synchronous compressed wavelet transform techniques are used to denoise the acquired vibration acceleration spectrum A(f) to improve data quality.

[0108] Data storage: The processed data is stored in a real-time database for subsequent modules to access.

[0109] Signaling transmission: The acquisition module sends the processed data to the dynamic reliability assessment module and the dynamic load coefficient acquisition module via the local area network to ensure the real-time performance and accuracy of the data.

[0110] The dynamic reliability assessment module, based on a fusion model of the improved Weibull proportional hazards function and the nonlinear degradation process, constructs a dynamic reliability assessment model for the propulsion device and calculates the instantaneous failure risk rate λ(t). The model incorporates shaft torque τ and bearing temperature T. b The nonlinear degradation factor D(u) is obtained, and the degradation weight coefficients α1, α2, and α3 are solved using the improved Grey Wolf optimization algorithm.

[0111] Operational mechanism: Model construction: An improved Weibull proportional risk function is used to fuse the nonlinear degradation process into the model.

[0112] Parameter optimization: The degradation weight coefficients are solved by an improved gray wolf optimization algorithm to minimize the sum of the mean square error and L1 regularization term of the shaft system historical fault data.

[0113] Receive real-time operational data from the acquisition module. Send the calculated instantaneous failure risk rate λ(t) to the multi-objective optimization module and the verification module.

[0114] The dynamic load coefficient acquisition module calculates the dynamic load spectrum correction coefficient κ based on the collected real-time operating data. Modal decomposition of the vibration acceleration spectrum A(f) is performed using an LSTM network to extract the first k natural frequencies f. k and damping ratio ζ k And combined with the turbulence intensity ε of the wake field w Calculate the dynamic load spectrum correction factor κ.

[0115] The vibration acceleration spectrum A(f) is decomposed using an LSTM network to extract the first k natural frequencies f. k and damping ratio ζ k Combined with the wake field turbulence intensity ε w The dynamic load spectrum correction coefficient κ is calculated, and its expression is: Receive real-time running data from the acquisition module. Send the calculated dynamic load spectrum correction coefficient κ to the multi-objective optimization module and the verification module.

[0116] The multi-objective optimization module uses the instantaneous failure risk rate λ(t) and the dynamic load spectrum correction coefficient κ to construct a multi-objective optimization function. The objective is to minimize the failure risk rate and maximize propulsion efficiency. The NSGA-II algorithm is used to solve for the optimal control parameter set {ω, τ, P}. l }

[0117] Objective function construction: Constructing a multi-objective optimization function.

[0118] The NSGA-II algorithm is used to solve for the optimal control parameter set {ω, τ, P_l}. Furthermore, a crowding calculation based on Mahalanobis distance is introduced to improve the uniformity of the Pareto front distribution.

[0119] λ(t) and κ are received from the dynamic reliability assessment module and the dynamic load factor acquisition module. The calculated optimal control parameter set {ω, τ, P} is then used. l The data is then sent to the verification module and control system.

[0120] The verification module validates the optimization results using a digital twin model and employs a coupled CFD-FEM method to simulate the stress-strain field of the propulsion device under a κ-corrected load spectrum, outputting the fatigue life N of key components. f If N f Below the preset threshold N th This triggers an online adjustment strategy based on reinforcement learning, dynamically updating the control parameter set.

[0121] Digital twin modeling: The coupled CFD-FEM method is used to simulate the stress-strain field of the propulsion device under the κ-corrected load spectrum, and the fatigue life N of key components is output. f .

[0122] Online adjustment: If N f Below the preset threshold N th This triggers an online adjustment strategy based on reinforcement learning, dynamically updating the control parameter set.

[0123] Receive the optimal control parameter set {ω, τ, P} from the multi-objective optimization module. l The calculation results are sent to the control system, and in N... f When the threshold is lowered, a reinforcement learning strategy is triggered.

[0124] The control system is integrated, and the final optimized parameter set is written into the propulsion device control system, and λ(t) and η are monitored in real time. p If the deviation rate exceeds 5%, the process will return to the acquisition module and restart the optimization process.

[0125] Parameter writing: Write the optimized set of control parameters into the propulsion device control system.

[0126] Real-time monitoring: Real-time monitoring of λ(t) and η p If the deviation rate exceeds 5%, the process will return to the acquisition module and restart the optimization process.

[0127] The optimal control parameter set is received from the multi-objective optimization module. The optimized parameter set is written into the control system, and a re-optimization process is triggered when the deviation rate exceeds a threshold.

[0128] Example 4 This disclosure provides a non-volatile computer storage medium storing computer-executable instructions that can perform the steps described in the above embodiments.

[0129] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0130] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0131] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (AN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0133] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0134] The preferred embodiments of the present invention have been described above to make the spirit of the present invention clearer and easier to understand, and are not intended to limit the present invention. All modifications, substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope summarized by the appended claims.

Claims

1. A dynamic load adaptive reliability optimization method for a ship propulsion device, characterized in that, Includes the following steps: Step S101: Collect real-time operating data of the ship's propulsion system; Step S103: Construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); Step S105: Calculate the dynamic load spectrum correction coefficient κ based on the real-time running data collected in step S101; Step S107: Based on λ(t) from step S103 and κ from step S105, construct a multi-objective optimization function; Step S109: Verify the optimization results using a digital twin model, and write the final optimized parameter set into the propulsion device control system.

2. The method as described in claim 1, characterized in that, The real-time operating data includes propeller speed ω, shaft torque τ, and bearing temperature T. b Lubricating oil pressure P l Vibration acceleration spectrum A(f), propulsion efficiency η p and the turbulence intensity ε in the wake field w .

3. The method as described in claim 2, characterized in that, The instantaneous failure risk rate λ(t) mentioned in step S103 is calculated using the following formula: Where β is the shape parameter, η is the scale parameter, γ is the degenerate coupling coefficient, σ is the Gaussian kernel bandwidth, and D(u) is the coefficient based on shaft torque τ and bearing temperature T. b The nonlinear degradation quantity is expressed as follows: α1, α2, and α3 are degradation weighting coefficients.

4. The method as described in claim 3, characterized in that, The weighting coefficients α1, α2, and α3 of the nonlinear degradation quantity D(u) are solved using an improved gray wolf optimization algorithm, specifically including: (a) Establish a fitness function with the shaft system historical fault data as input, the expression of which is the sum of mean square error and L1 regularization term; (b) Introduce Tent chaotic mapping to initialize the population and enhance global search capabilities; (c) The exploration-exploitation balance of the Grey Wolf algorithm is adjusted by a dynamic weighting mechanism, and the weight update formula includes the hyperbolic tangent function.

5. The method as described in claim 2, characterized in that, Step S105 specifically includes: An adaptive Kalman filter and a long short-term memory (LSTM) network are used as a joint algorithm to perform mode decomposition based on the vibration acceleration spectrum A(f) acquired in step S101, and extract the first k natural frequencies f. k and damping ratio ζ k And combined with the wake field turbulence intensity ε w Calculate the dynamic load spectrum correction factor κ.

6. The method as described in claim 5, characterized in that, The expression for the dynamic load spectrum correction coefficient κ is: 。 7. The method as described in claim 1, characterized in that, In step S107, the NSGA-II algorithm is used to solve for the optimal control parameter set {ω,τ,P}. l } 8. The method as described in claim 1, characterized in that, In step S109, the digital twin model uses a coupled CFD-FEM method to simulate the stress-strain field of the propulsion device under a κ-corrected load spectrum, and outputs the fatigue life N of the corresponding components. f .

9. The method as described in claim 8, characterized in that, The method further includes: if N f Below the preset threshold N th This triggers an online adjustment strategy based on reinforcement learning, dynamically updating the control parameter set in step S107.

10. A dynamic load adaptive reliability optimization system for a ship propulsion device, comprising: The data acquisition module is used to collect real-time operating data of the ship's propulsion system. The dynamic reliability assessment module is used to construct a dynamic reliability assessment model for the propulsion device based on the improved Weibull proportional risk function and nonlinear degradation process to obtain the instantaneous failure risk rate λ(t); The dynamic load coefficient acquisition module is used to calculate the dynamic load spectrum correction coefficient κ based on the collected real-time running data. The multi-objective optimization module is used to construct a multi-objective optimization function based on the instantaneous failure risk rate λ(t) and the dynamic load spectrum correction coefficient κ; The verification module is used to verify the optimization results through a digital twin model and write the final optimized parameter set into the propulsion device control system.

Images