Method and system for reliability simulation of ship transmission shaft based on deep reinforcement learning

Abstract

The application discloses a reliability simulation method and system for a ship transmission shaft based on deep reinforcement learning, and belongs to the field of ship transmission shaft design. The application organically combines a deep reinforcement learning algorithm with fault features and reliability characteristic quantities, substitutes reliability characteristic quantities of transmission shaft parts into a fault tree for quantitative analysis, calculates state transition probabilities of each node, uses a Markov chain model to summarize state transition rules of each node fault mode to establish a system fault state transition probability formula; a fault mode analysis establishes a fault state space; a probability correlation degree between a bottom event and a top event of the fault tree qualitative analysis constitutes an action space; a reward function weight factor of a Markov decision model is generated through a mathematical relationship of reliability characteristic quantities of each level unit, a reliability simulation event table column is solved and generated, and transmission shaft reliability change conditions can be timely reflected, so that the reliability comprehensive prediction precision of the transmission shaft is greatly improved, and the reliability and availability of the transmission shaft during operation are ensured.

Description

Technical Field

[0001] This invention belongs to the field of ship drive shaft design, and more specifically, relates to a reliability simulation method and system for ship drive shafts based on deep reinforcement learning. Background Technology

[0002] With the rapid development of advanced technologies such as computer-aided manufacturing and intelligent manufacturing, shipbuilding technology and processes have advanced rapidly. Electronic, digital, and intelligent equipment is increasingly being applied to ships, and the performance of ship propulsion equipment needs to meet correspondingly higher requirements. As a complex, multi-tasking, and repairable system in terms of structure, function, and performance, the ship's driveshaft requires careful consideration. Determining the relationship between product design, structure, process, and failure, establishing a reliability simulation analysis model, identifying the mechanisms or causes of failure, and maximizing the detection of defects caused by unreliable factors are crucial for ensuring more complete system reliability information. This is not only an important component of the equipment's tactical and technical specifications but also serves as the basis for the contractor's equipment development, production, and testing. Therefore, to improve the combat effectiveness and support capabilities of ship propulsion system driveshafts, accurate and effective reliability modeling and analysis should be conducted simultaneously during the design and development process.

[0003] The function of a ship's driveshaft is to transmit the power generated by the engine to the propeller shaft via an intermediate shaft. The propeller shaft drives the propeller to generate thrust, propelling the ship forward. During operation, the driveshaft system itself generates three types of vibration: torsional vibration, longitudinal vibration, and gyroscopic vibration. These vibrations not only severely affect the operation of the driveshaft system but also cause abnormal wear and fatigue damage, leading to underwater low-frequency and multi-frequency strong line spectrum radiation noise. This results in a sharp weakening of the ship's acoustic stealth performance during operation, thus generating failure events of varying degrees. On the other hand, external environmental factors, such as the back impact force exerted on the ship's driveshaft system by the ship's own weapon systems during operation, and the impact loads from underwater contact and non-contact explosions, all pose significant threats to the reliability of the ship's driveshaft, leading to failure events. Therefore, it is crucial to analyze the physicochemical processes, design defects, manufacturing defects, and other process factors that directly cause failures or lead to performance degradation and further development into failures, and to qualitatively analyze and summarize the failure mechanisms of ship driveshafts. By analyzing the failure mechanism, the failure mode and failure impact can be accurately described, providing support for the periodic characteristics of failures in the reliability modeling and analysis of ship drive shafts.

[0004] The transmission shaft system of a ship's propulsion system is a complex mechanism system characterized by its complex structure, wide power density range, complex and harsh working environment, variable operating conditions, and long power transmission path. Numerous random and fuzzy factors influence the coordinated motion of the units. Vibrations generated by the movement of the units themselves, transmitted impacts between different systems, vibrations from drastic changes in the external environment, and frictional wear caused by the relative motion between mating units all have varying degrees of impact on the reliability of the transmission shaft system, leading to frequent failures, conflicts over resource allocation, and other negative events that affect the ship's combat effectiveness and support capabilities. Establishing a complete and reliable simulation model of the ship's propulsion system transmission shaft reliability is of great significance for improving the system's operational reliability and for researching simulation modeling techniques for the reliability of the entire transmission system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a reliability simulation method and system for ship drive shafts based on deep reinforcement learning. The aim is to establish a reliability analysis and evaluation model for ship drive shafts, improve the mission success rate and combat readiness of equipment involving drive shafts, and solve the problem that the reliability of drive shafts cannot be analyzed and evaluated in a timely and accurate manner during design and use.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a reliability simulation method for ship drive shafts based on deep reinforcement learning, the method comprising:

[0007] S1. Determine the component set of the ship's drive shaft, the failure mode set of each component, and the reliability characteristics;

[0008] S2. Use FMEA to analyze the reliability characteristics of each component of the ship's drive shaft, and determine the risk, task, and time priority coefficients of each component's failure mode;

[0009] S3. Construct a fault tree for the transmission shaft failure mode with the ship's transmission shaft failure as the top event, and determine all minimum cut sets of the top event of the fault tree. The minimum cut set is the set of all possible bottom events that cause the top event of the fault tree to occur, with one bottom event arbitrarily removed.

[0010] S4. The fault modes of each node in the fault tree constitute the state space; the changes of each base event from the fault state to the normal state in all minimal cut sets constitute the action space; the Markov chain model is used to summarize the state change rules of each node's fault mode and determine the system fault state transition probability formula; the three priority coefficients of each component's fault mode form a weighting factor, which together with maintenance resources and support resources constitute a reward function, resulting in a reinforcement learning reliability simulation model for the ship's drive shaft.

[0011] Preferably, the failure mode set and reliability characteristics of each component are obtained in the following manner:

[0012] (1) Dynamic simulation was performed using transmission loads to obtain the failure modes of each component;

[0013] (2) Input different loads and determine the maximum tensile stress and maximum shear stress of each component under each failure mode according to the four major strength theories of mechanics of materials.

[0014] (3) Based on the component's own ability to resist failure and the stress borne by each component, calculate the component's reliability characteristic quantities [R(t),λ(t),θ]:

[0015] R(t) = β*t

[0016]

[0017]

[0018] Where R(t) is the reliability at time t, λ(t) is the failure rate at time t, θ is the average lifetime, and β is the reliability coefficient, which is determined by the component's ability to resist failure and the stress borne by each component.

[0019] Preferably, the formula for calculating the reliability coefficient is:

[0020]

[0021] Where, μ r μ s These represent the component's inherent resistance to failure and the mathematical expectation of the stress borne by each component, respectively; σ r σ s These are the component's ability to resist failure and the standard deviation that each component can withstand, respectively.

[0022] Preferably, step S2 includes:

[0023] S21. Construct an FMEA worksheet based on the reliability characteristics of each component;

[0024] S22. Based on the failure modes and FEMA worksheet, determine the severity level and priority detection difficulty level of each component's failure mode;

[0025] S23. Calculate the risk priority coefficient RPN and task priority coefficient P for each component's failure mode. m and time priority coefficient P t :

[0026] RPN = P * S * D

[0027] Pm =R(t)*θ / λ(t)

[0028] P t = k*θ*λ(t)

[0029] Where O is the probability of the fault mode occurring, S is the severity level, D is the priority detection difficulty level, O,S,D∈(1,10); k is the weight parameter, with a value less than 0.01.

[0030] Preferably, the method of using a Markov chain model and the state change probabilities of each node's failure mode to determine the system failure state transition probability is as follows:

[0031]

[0032] Among them, I gi (t) represents the probability importance of the i-th unit, F s (t) represents the system failure probability, F i (t) represents the failure probability of the i-th unit.

[0033] Preferably, the reward function is:

[0034] R = (P m +P t )·M r (t)+RPN·G r (t)

[0035] Among them, M r (t) represents the amount of maintenance resources, G r (t) represents the guaranteed resource quantity, RPN is the risk priority coefficient, and P m P represents the task priority coefficient. t This represents the time priority coefficient.

[0036] Preferably, the component set of the ship's drive shaft includes a thrust shaft, a stern shaft, a propeller shaft, and an intermediate shaft.

[0037] To achieve the above objectives, in a second aspect, the present invention provides a reliability simulation system for ship drive shafts based on deep reinforcement learning, comprising: a processor and a memory; the memory for storing computer execution instructions; and the processor for executing the computer execution instructions, such that the method described in the first aspect is executed.

[0038] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art:

[0039] This invention proposes a reliability simulation method and system for ship driveshafts based on deep reinforcement learning. It organically combines deep reinforcement learning algorithms with fault features and reliability features. By establishing a fault state space, action space, fault state transition probabilities, and a reward function, it guides the optimal event list for driveshaft reliability simulation modeling and analysis. Specifically, reliability features are introduced into a feature tree for quantitative analysis, calculating the probability of state change at each node. A Markov chain model and the fault mode states of each node are used to establish the system fault state transition probability formula. Based on this, a fault state space is established through fault analysis; the action space is constructed by qualitatively analyzing the probability correlation between bottom and top events using a fault tree; and the reward function weight factors of the Markov decision model are generated through the mathematical relationships of reliability features at each level, solving to generate a reliability simulation event list. This reliably and accurately generates a driveshaft reliability simulation model, which is then analyzed and evaluated. This model can promptly reflect changes in driveshaft reliability, significantly improving the overall reliability prediction accuracy of the driveshaft, achieving an accuracy rate of over 90%, and ensuring the reliability and availability of the driveshaft during operation. Attached Figure Description

[0040] Figure 1 This is a flowchart of a reliability simulation method for ship drive shafts based on deep reinforcement learning provided by the present invention.

[0041] Figure 2 This is a schematic diagram of a fault tree provided in an embodiment of the present invention. Detailed Implementation

[0042] 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 and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0043] Driveshafts are a crucial component of the propulsion shaft system in ship propulsion, primarily consisting of thrust shafts, stern shafts, propeller shafts, and intermediate shafts. Addressing the need for highly accurate reliability simulation analysis of driveshafts during ship operation, this invention analyzes and summarizes the causes and mechanisms of driveshaft failures, analyzes the impact of failure modes, establishes a failure state space and action space, qualitatively and quantitatively analyzes unit reliability characteristics, and establishes calculation models for relevant transition probabilities and reward functions. This results in a systematic driveshaft reliability simulation modeling and analysis method based on deep reinforcement learning algorithms.

[0044] This invention addresses the challenges of heavy loads, high torque, multi-stage power density, and complex operating conditions in ship transmission bearings, necessitating simultaneous reliability modeling and analysis during ship development. It conducts research on transmission shaft reliability simulation modeling and analysis, organically combining advanced deep reinforcement learning algorithms with fault characteristics and reliability features. A method for establishing and analyzing transmission shaft reliability simulation models is proposed, guided by a fault state space, execution action space, fault state transition probabilities, and a reward function to establish an optimal event list. The study investigates potential fault modes and their mechanisms during the transmission shaft's functional execution. A fault tree diagram is constructed using a series of event symbols, logic gate symbols, and transition symbols to describe the logical causal relationships between various system events. This method also introduces... Quantitative analysis is performed by substituting reliability features into the feature tree to calculate the probability of state change at each node. A Markov chain model and the fault mode states of each node are used to establish the system fault state transition probability formula. Based on this, a fault state space is established through fault analysis, and the action space is constructed by qualitatively analyzing the probability correlation between the bottom event and the top event using the fault tree. The reward function weight factor of the Markov decision model is generated through the mathematical relationship of reliability features of each level unit, and the reliability simulation event table is generated by solving. Thus, a reliable and accurate drive shaft reliability simulation model is generated, and its simulation model is analyzed and evaluated. It can reflect the changes in drive shaft reliability in a timely manner, greatly improve the comprehensive prediction accuracy of drive shaft reliability, and ensure the reliability and availability of drive shaft during operation.

[0045] like Figure 1 As shown, this invention provides a reliability simulation method for ship drive shafts based on deep reinforcement learning, the method comprising:

[0046] S1. Determine the component set of the ship's drive shaft, the failure mode set of each component, and the reliability characteristics;

[0047] S2. Use FMEA to analyze the reliability characteristics of each component of the ship's drive shaft, and determine the risk, task, and time priority coefficients of each component's failure mode;

[0048] S3. Construct a fault tree for the transmission shaft failure mode with the ship's transmission shaft failure as the top event, and determine all minimum cut sets of the top event of the fault tree. The minimum cut set is the set of all possible bottom events that cause the top event of the fault tree to occur, with one bottom event arbitrarily removed.

[0049] S4. The fault modes of each node in the fault tree constitute the state space; the changes of each base event from the fault state to the normal state in all minimal cut sets constitute the action space; the Markov chain model is used to summarize the state change rules of each node's fault mode and determine the system fault state transition probability formula; the three priority coefficients of each component's fault mode form a weighting factor, which together with maintenance resources and support resources constitute a reward function, resulting in a reinforcement learning reliability simulation model for the ship's drive shaft.

[0050] The input data for step S1 consists of transmission loads of different magnitudes and three-dimensional solid data of the transmission shaft; the output data consists of reliability characteristics of each component of the transmission shaft. The intermediate processing specifically includes:

[0051] S11. Use transmission load to perform dynamic simulation to obtain the failure modes of each component.

[0052] Torsional vibration can lead to failure modes such as deformation, fatigue failure, and torsional breakage in drive shafts. A finite element model of the dynamic response of a drive shaft unit under fault conditions is established. The fault mechanism is studied through simulation analysis, and the fault modes caused by the fault mechanism are summarized.

[0053] S12. Based on the different input load magnitudes, determine the magnitudes of normal stress and shear stress borne by the parts under each fault mode.

[0054] Based on the dimensions and materials of the 3D model, its performance parameters such as elastic modulus and yield strength can be determined directly by referring to national standard tables. By inputting different loads, the stress magnitude that the shaft bears when it undergoes torsion or bending can be calculated based on the different loads applied to the shaft and the power and rotational speed transmitted by the shaft. According to the four strength theory formulas, the maximum tensile stress and maximum shear stress of the shaft under various failure modes can be calculated.

[0055] Based on the dynamic model established in the ANSYS environment, transmission loads are applied, and the role and influence of different faults in the model's dynamic response are analyzed. The failure distribution types of each unit are obtained, and the estimated values ​​of reliability characteristic quantities are obtained from the preset parameter values ​​of the distribution.

[0056] S13. Calculate the reliability characteristics of each component of the drive shaft based on the physical failure indicators under each failure mode.

[0057] In the reliability design of mechanical parts, all external forces that cause failure are called stress, and the part's ability to resist failure is called strength. By calculating the probability that the strength is higher than the stress, the reliability of the part can be obtained.

[0058] The formula for calculating reliability is:

[0059] R(t) = β·t

[0060]

[0061] Where, μ r μ s These are the mathematical expectations of strength and stress, respectively; σ r σ s α and β are the standard deviations of strength and stress, respectively; β is the reliability coefficient.

[0062] Failure rate formula:

[0063] λ(t)=(1-R(t))′ / R(t).

[0064] Where ′ represents the first derivative.

[0065] Average lifespan formula:

[0066]

[0067] The input data for step S2 consists of the reliability characteristics of each component of the drive shaft; the output data includes the risk priority coefficient, task priority coefficient, and time priority coefficient for each component's failure mode. The intermediate processing specifically includes:

[0068] S21. Construct an FMEA worksheet using the reliability characteristics of each component, including reliability, mean life, and failure rate.

[0069] Create an FMEA worksheet, record and number the number and name of components, define the equipment system being analyzed, explain the expected performance, system limitations, and fault criteria at the agreed level, and determine all fault modes of the system's subordinate units.

[0070] S22. Based on the FMEA worksheet and the failure modes of each component, determine the severity level and priority detection difficulty level of each component's failure mode.

[0071] S23. Calculate the risk priority coefficient (RPN) for each component failure mode based on failure rate, severity category, and priority detection difficulty level; calculate the task priority coefficient (P) for each component failure mode based on reliability, mean life, and failure rate among reliability characteristics. m and time priority coefficient P t .

[0072] Risk priority coefficient calculation formula:

[0073] RPN = (P)(S)(D)

[0074] Where P is the probability of the failure mode occurring, S is the severity level, and D is the priority detection difficulty level. P, S, D ∈ (1, 10). The selection of these three values ​​can be calculated by comparing with international standard tables.

[0075] Task priority coefficient calculation formula:

[0076] P m =R(t)·θ / λ(t)

[0077] Formula for calculating time priority coefficient:

[0078] P t =kθλ(t)

[0079] Where k is a weighting parameter, and its value is generally less than 0.01.

[0080] The input data for step S3 are fault event symbols, logic gate symbols, and fault mode transition symbols for drive shaft components; the output data is a fault mode fault tree of the drive shaft with drive shaft fault as the top event.

[0081] The steps for building a fault tree are as follows: 1) Write down the top event. The top event (final failure mode) is represented as the least desirable failure event. The symbol for the top event is the first row of the fault tree; 2) Represent the failure causes that lead to the top event as the second row of the fault tree using the corresponding symbols, and connect it to the top event using appropriate logic gates; 3) Represent the failure causes that lead to the failure events in the second row as the third row, and connect it to the second row using fault gates; 4) Index the failure causes layer by layer to trace all the bottom events that caused the system failure. This forms an inverted fault tree with the top event as the root, intermediate events as nodes, and bottom events as leaves.

[0082] Taking the intermediate shaft, a component of the drive shaft, as an example, the generated fault tree is as follows: Figure 2 As shown. Here, + represents an OR gate: outputs when the input exists; o represents an AND gate: outputs only when both inputs exist; events inside the circle are basic events, representing events where further subdivision is not possible; events inside the rectangle include the top event and subdivisible events.

[0083] The input data for step S4 are the fault tree and the reliability characteristics of each component of the drive shaft; the output data are the minimum cut set and first-order cut set representing the cause of the fault, the importance of each node unit of the fault tree, and the reliability characteristics of the drive shaft system.

[0084] Fault tree analysis reveals how failure modes of individual components propagate and transfer to the failure modes of the driveshaft system. Therefore, qualitative analysis of the fault tree reveals the causes and combinations of causes for driveshaft failure as the top event. All minimal cut sets represent all possible combinations of base events leading to the top event in the fault tree. If at least one base event in each minimal cut set never occurs, then the top event never occurs; this base event is a first-order minimal cut set. To determine which base event combination has a lower probability of occurrence, quantitative fault tree analysis is introduced to assess unit importance. Higher unit importance corresponds to a greater probability of the top event occurring.

[0085] Formula for calculating the importance of unit probability:

[0086]

[0087] Among them, I gi (t) represents the probability importance of the i-th unit, F s (t) represents the failure probability of the system, F i (t) represents the failure probability of the i-th unit. The failure probability of the i-th unit is the failure rate of the i-th unit. Since the system follows an exponential or normal distribution, the system failure rate can be calculated from the unit failure rates. After the system failure rate is calculated, the system reliability and mean lifetime can be obtained by applying the relationship between the failure rate formula, the reliability formula, and the mean lifetime formula.

[0088] Based on the principle of minimum cut set importance determination, high-frequency low-frequency events and their corresponding minimum cut set order are determined. Minimum cut sets can guide the priority of system fault diagnosis and maintenance. By calculating the reliability characteristics of each minimum cut set component, the membership low-frequency event with the highest probability of occurrence is determined. The first-order minimum cut set of each minimum cut set in the reliability-eliminating system is used as the action space of the Markov decision model.

[0089] The input data for step S5 is all the outputs of the first four steps, and the output data is a reinforcement learning reliability simulation model with clearly defined four elements.

[0090] The input of a reliability simulation model is characterized by a Markov decision process (MDP) quadruple [S,A,P,R] using deep reinforcement learning.

[0091] Step i. Determining the state space S. Each cell in the fault tree represents a fault mode, which forms the system's state space.

[0092] Step ii. Determining the action space A. Each basic event in all minimal cut sets changes from a fault state to a normal state.

[0093] Step iii. Determining the transition probability P. This is expressed as the rate of change of the system failure probability caused by the change in the failure rate of the i-th component (unit).

[0094]

[0095] Among them, I gi (t) represents the probability importance of the i-th unit, F s (t) represents the failure probability of the system, F i (t) represents the failure probability of the i-th unit.

[0096] Step iv. Determining the reward function R.

[0097] The weighting factors, composed of the risk priority coefficient, task priority coefficient, and time priority coefficient for unit failures, together with maintenance resources and support resources, constitute the reward function, as shown in the following formula:

[0098] R = (P m +P t )·M r (t)+RPN·G r (t)

[0099] Among them, M r (t) represents the amount of maintenance resources, G r (t) is to ensure the amount of resources.

[0100] The final Markov decision model solution yields the optimal fault event processing sequence list. The simulation clock employs a tiered fixed-step time-lapse mechanism, meaning it advances using a simulation clock with fixed time intervals and uses a tiered configuration for multiple cyclical advancements to record fault state transitions. Because the Average Logistics Delay Time (MLDT) is absolutely different for different fault events, the state transitions will inevitably differ in each loop. After n loops, all fault state event lists will appear, and the iterative update time values ​​of the event lists will be recorded.

[0101] The simulation is again driven by the combined action of the system simulation global clock TIME (i.e., the time range of the tiered fixed step) and the time cell (the time value of the event table iterative update) of the flag event's own clock. Here, the flag event is the elimination of the first-order minimal cut set, i.e., the generation of the maintenance event. Through the records of the time cell, the reliability characteristics of the unit fault state and the system reliability characteristics can be calculated in reverse. This process can provide data support for reliability analysis and evaluation.

[0102] The parameters of the Deep Reinforcement Learning (DQN) algorithm were determined: experience pool size D, discount factor γ, replay batch τ, number of rounds Θ, round duration T, and learning rate C. The DQN physics engine (MDP environment) was written using the openAI / gym platform, the stable-baselines tool was called, and the relevant data was recorded using tensorboardX.

[0103] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A reliability simulation method for ship drive shafts based on deep reinforcement learning, characterized in that, The method includes: S1. Determine the component set of the ship's drive shaft, the failure mode set of each component, and the reliability characteristics; S2. Use FMEA to analyze the reliability characteristics of each component of the ship's drive shaft, and determine the risk, task, and time priority coefficients of each component's failure mode; S3. Construct a fault tree for the transmission shaft failure mode with the ship's transmission shaft failure as the top event, and determine all minimum cut sets of the top event of the fault tree. The minimum cut set is the set of all possible bottom events that cause the top event of the fault tree to occur, with one bottom event arbitrarily removed. S4. The fault modes of each node in the fault tree constitute the state space; the changes of each base event of all minimal cut sets from the fault state to the normal state constitute the action space; the Markov chain model is used to summarize the state change rules of each node's fault mode and determine the system fault state transition probability formula; the three priority coefficients of each component's fault mode form a weighting factor, which together with maintenance resources and support resources constitute a reward function to obtain the reinforcement learning reliability simulation model of the ship's drive shaft. The failure mode set and reliability characteristics of each component are obtained in the following way: (1) Dynamic simulation was performed using transmission loads to obtain the failure modes of each component; (2) Input different loads and determine the tensile and shear stresses borne by each component under each failure mode according to the four major strength theories of mechanics of materials; (3) Calculate the reliability characteristics of the components based on their ability to resist failure and the stress they bear. : in, Reliability at any moment, The efficiency of momentary failure, For average lifespan, The reliability coefficient is determined by the component's ability to resist failure and the stress borne by each component. The formula for calculating the reliability coefficient is: in, These are the component's ability to resist failure and the expected value of the stress borne by each component, respectively. These are the component's ability to resist failure and the standard deviation of the stress borne by each component, respectively. Step S2 includes: S21. Construct an FMEA worksheet based on the reliability characteristics of each component; S22. Based on the failure modes and FEMA worksheet, determine the severity level and priority detection difficulty level of each component's failure mode; S23. Calculate the risk priority coefficient of each component's failure mode. Task priority coefficient and time priority coefficient : in, This represents the probability of a failure mode occurring. Severity level, Prioritize detection based on difficulty level. ; This is a weighting parameter, with a value less than 0.01; The steps for building a fault tree are as follows: 1) Write down the top event, which represents the least desirable fault event. The symbol for the top event is the first row of the fault tree; 2) The fault causes that lead to the top event are represented as the second row of the fault tree using the corresponding symbols, and connected to the top event using appropriate logic gates; 3) The fault causes that lead to the fault events in the second row are represented as the third row, and connected to the second row using fault gates; 4) According to the cause of the fault, index layer by layer to trace all the bottom events that caused the system fault. This forms an inverted fault tree with the top event as the root, intermediate events as nodes, and bottom events as leaves.

2. The method as described in claim 1, characterized in that, The method utilizes a Markov chain model and the state change probabilities of each node's failure mode to determine the system's failure state transition probabilities. in, For the first The probability importance of each unit. The probability of system failure. For the first The failure probability of each unit.

3. The method as described in claim 1, characterized in that, The reward function is: in, For maintenance resources, To ensure sufficient resources.

4. The method according to any one of claims 1 to 3, characterized in that, The components of the ship's drive shaft include the thrust shaft, stern shaft, propeller shaft, and intermediate shaft.

5. A reliability simulation system for ship drive shafts based on deep reinforcement learning, characterized in that, include: Processor and memory; The memory is used to store computer-executed instructions; The processor is configured to execute the computer execution instructions, such that the method described in any one of claims 1 to 4 is executed.

Images