Lightweight i-beam reliability analysis system and method

Abstract

The present application relates to the technical field of structural health monitoring and artificial intelligence decision-making, in particular to a light I-beam reliability analysis system and method thereof, comprising: acquiring attack frequency, attack amplitude and conventional operation stress collected by a multi-modal sensor; calculating basic stress; solving resonance amplification factor; calculating total effective stress; calculating structural integrity entropy, entropy growth rate and entropy growth acceleration; predicting entropy increase collapse critical time window; inputting a deep reinforcement learning model; solving and outputting closed-loop correction control action; if the closed-loop correction control action is to maintain the status quo, then the structural natural frequency and structural damping ratio remain unchanged; if the closed-loop correction control action is preventive structural reinforcement, then the structural natural frequency is increased; if the closed-loop correction control action is to deploy active damper, then the structural damping ratio is increased; the present application constructs a self-consistent feedback closed loop from intelligent decision-making to physical execution, ensuring the physical realizability and convergence of the system anti-fragile strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of structural health monitoring and artificial intelligence decision-making, in particular to a light I-beam reliability analysis system and method. BACKGROUND

[0002] In the field of structural engineering, the long-term reliability of key components such as light I-beams is crucial; traditional reliability analysis methods that rely on static models have significant limitations when faced with external covert resonance attacks; such attacks can continuously apply low-energy excitation, leading to accelerated accumulation of structural damage, and ultimately causing catastrophic destruction;

[0003] Existing technologies focus on post-damage detection and lack forward-looking risk quantification models and real-time adaptive decision-making capabilities for dynamic high-risk threats such as covert resonance; current countermeasures rely heavily on human intervention, with slow response times and a lack of closed-loop control from threat identification, cost trade-offs to physical corrections, making it difficult to effectively suppress damage; therefore, how to provide a reliability analysis method that can predict failure risks under resonance attacks and autonomously decide to sacrifice short-term costs for long-term structural survival is a technical problem that needs to be solved in the field. SUMMARY

[0004] To solve the above technical problems, the present application provides a light I-beam reliability analysis system and method, specifically, the technical solution of the present application is:

[0005] A light I-beam reliability analysis method, comprising:

[0006] S1, obtaining attack frequency, attack amplitude and regular operation stress collected by a multi-modal sensor, and structure inherent frequency and structure damping ratio determined by a structure finite element model;

[0007] S2, calculating the basic stress based on the attack amplitude and the preset machine-sound transmission coefficient;

[0008] S3, combining the attack frequency, structure inherent frequency and structure damping ratio to solve the resonance amplification factor;

[0009] S4, calculating the total effective stress based on the resonance amplification factor, the basic stress and the regular operation stress;

[0010] S5, calculating the structure integrity entropy, entropy growth rate and entropy growth acceleration based on the total effective stress and the preset material parameters;

[0011] S6, combining the structure integrity entropy, entropy growth rate, entropy growth acceleration and the preset collapse critical point to predict the entropy increase collapse critical time window;

[0012] S7, input the structural integrity entropy, entropy growth rate, entropy growth acceleration and entropy growth collapse critical time window into the deep reinforcement learning model;

[0013] S8, the deep reinforcement learning model solves and outputs the closed-loop correction control action based on the self-defined reward function;

[0014] S9, in response to the closed-loop correction control action:

[0015] If the closed-loop correction control action is to maintain the status quo, the natural frequency of the structure and the structural damping ratio remain unchanged;

[0016] If the closed-loop correction control action is preventive structural reinforcement, the natural frequency of the structure is increased;

[0017] If the closed-loop correction control action is to deploy active dampers, the structural damping ratio is increased.

[0018] Optionally, the multi-modal sensor includes: an acoustic sensor, a strain gauge and an acceleration sensor.

[0019] Optionally, the material parameters include: initial entropy value, material degradation coefficient, material reference stress and damage accumulation index.

[0020] Optionally, S4 calculates the total effective stress, including:

[0021] Based on the resonance amplification factor and the basic stress, the additional vibration stress is calculated;

[0022] The additional vibration stress is superimposed with the conventional operation stress to obtain the total effective stress.

[0023] Optionally, S5 calculates the structural integrity entropy, including:

[0024] Based on the total effective stress, the growth rate of entropy is determined;

[0025] Wherein, the growth rate of entropy and the total effective stress are in exponential relationship, and the index is the damage accumulation index.

[0026] Optionally, the self-defined reward function includes:

[0027] A risk penalty term for punishing the entropy growth acceleration;

[0028] A cost penalty term for punishing the short-term cost of the closed-loop correction control action;

[0029] And a terminal state function term for defining collapse penalty and survival reward.

[0030] Optionally, S9 responds to the feedback loop formed, including:

[0031] The increase of the natural frequency of the structure or the structural damping ratio leads to the decrease of S3 resonance amplification factor;

[0032] The resonance amplification factor decreases, resulting in a decrease in the total effective stress S4;

[0033] The total effective stress decreases, resulting in suppression of the entropy growth rate and the entropy growth acceleration of S5.

[0034] A light girder reliability analysis system comprises:

[0035] A data acquisition module is configured to acquire attack frequency and attack amplitude collected by a multi-modal sensor, and regular operation stress, as well as structure inherent frequency and structure damping ratio determined by a structure finite element model;

[0036] A stress calculation unit is configured to calculate basic stress based on the attack amplitude and a preset machine-sound transmission coefficient;

[0037] A factor calculation unit is configured to calculate a resonance amplification factor in combination with the attack frequency, the structure inherent frequency and the structure damping ratio;

[0038] A total stress calculation unit is configured to calculate total effective stress based on the resonance amplification factor, the basic stress and the regular operation stress;

[0039] An entropy calculation unit is configured to calculate structure integrity entropy, an entropy growth rate and an entropy growth acceleration based on the total effective stress and preset material parameters;

[0040] A time prediction unit is configured to predict an entropy increase collapse critical time window in combination with the structure integrity entropy, the entropy growth rate, the entropy growth acceleration and a preset collapse critical point;

[0041] A decision input unit is configured to input the structure integrity entropy, the entropy growth rate, the entropy growth acceleration and the entropy increase collapse critical time window into a deep reinforcement learning model;

[0042] A decision output unit is configured to calculate and output a closed-loop correction control action based on a self-defined reward function by the deep reinforcement learning model;

[0043] A closed-loop correction unit is configured to respond to the closed-loop correction control action;

[0044] If the closed-loop correction control action is to maintain the status quo, the closed-loop correction unit keeps the structure inherent frequency and the structure damping ratio unchanged;

[0045] If the closed-loop correction control action is preventive structure reinforcement, the closed-loop correction unit increases the structure inherent frequency;

[0046] If the closed-loop correction control action is to deploy an active damper, the closed-loop correction unit increases the structure damping ratio.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. The present application accurately captures the frequency and amplitude of covert resonance attacks through multi-modal sensors such as acoustic, strain gauges, etc., and separates them from regular operating stresses; it uses a stress solving path that amplifies first and then superimposes, calculates additional vibration stress based on resonance amplification factor, and then superimposes with regular stress, this design clearly decouples the contribution of different stress sources, avoids the risk of underestimation caused by model confusion, and provides a high-fidelity data foundation for accurate and reliable analysis;

[0049] 2. The present application introduces structural integrity entropy as a damage quantification index, which is calculated based on calibratable material parameters and has clear physical meaning; more importantly, this method establishes a nonlinear exponential relationship between entropy growth rate and total effective stress, which can greatly amplify the contribution of high stress events such as resonance attacks to damage accumulation, this high sensitivity damage model is more in line with the physical reality of material fatigue, ensuring the realization of strong and non-delayed risk assessment;

[0050] 3. The present application predicts the critical time window of structure collapse based on structural integrity entropy, entropy growth rate and its acceleration; and inputs these dynamic risk indicators into a deep reinforcement learning model, which uses a self-defined reward function centered on punishing entropy growth acceleration, making it highly predictive of risks, and can calculate the optimal control strategy that sacrifices short-term cost for long-term survival, realizing the intelligent upgrade of decision-making from passive response to active defense;

[0051] 4. The present application constructs a self-consistent feedback loop from intelligent decision-making to physical execution; the control actions output by the deep reinforcement learning model, such as preventive reinforcement or deployment of dampers, directly act on the physical structure to increase its natural frequency or damping ratio, these changes in physical parameters will suppress the resonance amplification factor and thus reduce the total effective stress and suppress the growth of entropy; this negative feedback path couples abstract AI decisions with specific physical parameters, ensuring the physical realizability and convergence of the system's anti-fragile strategy. BRIEF DESCRIPTION OF DRAWINGS

[0052] The present application will be further explained in conjunction with the accompanying drawings and embodiments:

[0053] Figure 1 is a flowchart of the method of the present application;

[0054] Figure 2 is a structural diagram of the system of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail in conjunction with specific embodiments.

[0056] Example 1:

[0057] Referring to Figure 1 A light H-beam reliability analysis method, comprising:

[0058] S1, acquiring attack frequency, attack amplitude and regular operation stress collected by a multi-modal sensor, and structure inherent frequency and structure damping ratio determined by a structure finite element model;

[0059] S2, calculating basic stress based on attack amplitude and preset machine-sound transmission coefficient;

[0060] S3, calculating resonance amplification factor in combination with attack frequency, structure inherent frequency and structure damping ratio;

[0061] S4, calculating total effective stress based on resonance amplification factor, basic stress and regular operation stress;

[0062] S5, calculating structure integrity entropy, entropy growth rate and entropy growth acceleration based on total effective stress and preset material parameters;

[0063] S6, predicting entropy growth collapse critical time window in combination with structure integrity entropy, entropy growth rate, entropy growth acceleration and preset collapse critical point;

[0064] S7, inputting structure integrity entropy, entropy growth rate, entropy growth acceleration and entropy growth collapse critical time window into a deep reinforcement learning model;

[0065] S8, calculating and outputting closed-loop correction control action by the deep reinforcement learning model based on a self-defined reward function;

[0066] S9, responding to the closed-loop correction control action:

[0067] if the closed-loop correction control action is to maintain the status quo, keeping the structure inherent frequency and the structure damping ratio unchanged;

[0068] if the closed-loop correction control action is preventive structure reinforcement, increasing the structure inherent frequency;

[0069] if the closed-loop correction control action is to deploy active damper, increasing the structure damping ratio.

[0070] The embodiment provides a light H-beam reliability analysis method, aiming to solve the problem of how to sacrifice short-term optimal operation efficiency to obtain long-term structure survival through an anti-fragile AI decision strategy when the structure faces hidden resonance attack; the method contains complete steps of S1 to S9, forming an adaptive closed-loop control system;

[0071] S1, acquiring attack frequency, attack amplitude and regular operation stress collected by a multi-modal sensor, and structure inherent frequency and structure damping ratio determined by a structure finite element model;

[0072] In this embodiment, this step is the data acquisition stage; multi-modal sensors, whose purpose is to collect data from different dimensions, in this embodiment, can specifically include: acoustic sensors, strain gauges, and acceleration sensors; attack frequency , refers to the frequency of external infrasound wave attack, which acts as the core input of resonance analysis, identified and collected by acoustic sensors; attack amplitude , refers to the amplitude of external infrasound wave attack, which acts to quantify the energy of the attack, identified and collected by acoustic sensors; regular operation stress , refers to the stress that the structure bears in the normal operation state, which acts as the baseline for stress superposition, monitored by strain gauges and acceleration sensors deployed on the I-beam; structure natural frequency and structure damping ratio , refer to the physical properties of the structure itself, which acts to determine whether the structure is prone to resonance, calibrated and determined based on the structure finite element model FEM established by laser scanning;

[0073] S2, based on the attack amplitude and the preset machine-sound transfer coefficient, calculate the basic stress;

[0074] In this embodiment, the purpose of this step is to convert the collected attack amplitude into stress units on the structure; machine-sound transfer coefficient , refers to a conversion coefficient, which acts to convert acoustic or displacement amplitude into basic stress ; To ensure implementability, the calibration of this coefficient , the value of can be determined by applying a set of known calibration amplitudes representing the calibration working condition in the FEM model, and measuring the corresponding calibration basic stress , and then fitting by, for example, least squares method;

[0075] To convert the attack amplitude into the basic stress, this embodiment introduces the basic stress , which is calculated as follows:

[0076] ;

[0077] Where, : basic stress, dimensionless, calculated by this step; it represents the basic stress applied to the structure by the attack source in the non-resonance case;

[0078] : machine-sound transfer coefficient, dimensionless, for example, or Calibration by FEM model, as mentioned above;

[0079] : attack amplitude, dimensionless, for example or , collected in S1;

[0080] The calculation logic of this step is to ensure that the subsequent stress analysis has the correct dimension;

[0081] S3, combined with attack frequency, structure inherent frequency and structure damping ratio, to solve the resonance amplification factor;

[0082] In this embodiment, the purpose of this step is to quantify the amplification effect of the resonance attack; the resonance amplification factor , refers to a dimensionless parameter, which is used to describe the linear amplification multiple of stress or amplitude caused by resonance when the external excitation frequency approaches the inherent frequency , which is derived from the forced vibration theory in classical vibration mechanics, and its calculation method is as follows:

[0083] ;

[0084] wherein, : resonance amplification factor, dimensionless, calculated by this step;

[0085] : attack frequency, dimensionless, Hz or rad / s, collected in S1;

[0086] : structure inherent frequency, dimensionless, Hz or rad / s, FEM model calibration in S1;

[0087] : structure damping ratio, dimensionless, FEM model calibration in S1;

[0088] The calculation logic of this step is that when approaches , value will increase sharply, so as to significantly amplify the attack stress in S4;

[0089] S4, based on the resonance amplification factor, the basic stress, the regular operation stress, to calculate the total effective stress;

[0090] In this embodiment, the purpose of this step is to calculate the composite stress actually borne by the structure at present; in one embodiment, based on the resonance amplification factor calculated by S3 and the basic stress calculated by S2, the additional vibration stress is calculated, and the calculation formula is:

[0091] ;

[0092] wherein, : additional vibration stress, dimension of Pa, calculated by this step;

[0093] : resonance amplification factor, dimension of dimensionless, calculated by S3;

[0094] : base stress, dimension of Pa, calculated by S2;

[0095] Subsequently, the additional vibration stress is superimposed with the conventional operation stress obtained by S1 to obtain the total effective stress , and the calculation formula is:

[0096] ;

[0097] wherein, : total effective stress, dimension of Pa, calculated by this step;

[0098] : conventional operation stress, dimension of Pa, obtained by S1;

[0099] : additional vibration stress, dimension of Pa, calculated by this step;

[0100] The output of this step is the core index for driving the accumulation of structural damage in S5;

[0101] S5, based on the total effective stress and the preset material parameters, calculates the structural integrity entropy, entropy growth rate and entropy growth acceleration;

[0102] In this embodiment, the purpose of this step is to quantify the cumulative damage of the structure and its development trend; the structural integrity entropy is a quantitative index, which is used to reflect the degree of cumulative damage of the structure, and its source is the fatigue damage accumulation model in material science and self-defined improvement;

[0103] Material parameters refer to preset parameters required for calculating entropy, which in this embodiment specifically include: initial entropy value , whose source is the preset initial damage value, for example, 0; material degradation coefficient ; material reference stress , whose source is the material inherent attribute, dimension of Pa; and damage accumulation index ;

[0104] To ensure the feasibility, parameters and are obtained by calibrating fatigue experiments on the same material; specifically, a plurality of different constant experimental stresses are applied to a material sample, and the corresponding experimental failure times are measured; based on these experimental data points , parameters , dimension 1 / Time and , dimensionless, are determined by data fitting; to calculate the integrity entropy at time t, the present embodiment introduces the calculation formula:

[0105] ;

[0106] wherein, : integrity entropy at time t, dimensionless, calculated by the present step;

[0107] : initial entropy value, dimensionless, preset parameter;

[0108] : material degradation coefficient, dimension 1 / Time (such as 1 / s), preset parameter;

[0109] : total effective stress at time t, dimension Pa, calculated by S4;

[0110] : material reference stress, dimension Pa, preset parameter;

[0111] : damage accumulation index, dimensionless, preset parameter;

[0112] Entropy growth rate , refers to the growth rate of entropy, which reflects the current rate of damage accumulation; its calculation formula is:

[0113] ;

[0114] This formula shows that the growth of entropy is in exponential relationship with the total effective stress ;

[0115] Entropy growth acceleration , refers to the change trend of the entropy growth rate, which reflects the acceleration of damage speed, which is the first order derivative of time ; in digital implementation, it can be calculated based on the current total effective stress total effective stress of the last time step to calculate and and then obtain wherein is the sampling time interval;

[0116] The output of this step current damage, damage speed, damage acceleration are the core inputs of S6 and S7;

[0117] S6, in combination with the structural integrity entropy, entropy growth rate, entropy growth acceleration and the preset collapse critical point, predicts the entropy increase collapse critical time window;

[0118] In this embodiment, the purpose of this step is to predict how much time is left before the structure completely fails; the collapse critical point refers to a preset threshold, which is used to represent the complete failure of the structure, such as the 10 / 10 failure boundary, which is preset according to the industry design specification or statistical analysis of a large amount of experimental data; the entropy increase collapse critical time window refers to the predicted remaining time;

[0119] to solve In this embodiment, a second-order Taylor expansion based on physical kinematics is used, which assumes that the entropy growth acceleration is constant in a short period of time, so that the entropy value after the future time satisfies:

[0120] ;

[0121] wherein : collapse critical point, dimensionless, preset parameter;

[0122] : current entropy value, dimensionless, calculated by S5;

[0123] : entropy growth rate, dimensionless, calculated by S5;

[0124] : entropy growth acceleration, dimensionless, calculated by S5;

[0125] By solving this quadratic equation about , the entropy increase collapse critical time window is obtained:

[0126] ;

[0127] The output of this step As a key risk indicator, input into the AI decision model of S7;

[0128] S7, the structural integrity entropy, entropy growth rate, entropy growth acceleration and entropy growth collapse critical time window are input into the deep reinforcement learning model;

[0129] In this embodiment, this step is the data preparation stage of AI decision; the deep reinforcement learning DRL model refers to a trained agent, whose purpose is to learn the structural conservatism decision logic;

[0130] In this step, the state space is defined, which refers to the input vector of the DRL agent for decision-making; in this embodiment, the state space contains at least ;

[0131] Each component in the vector, i.e. the integrity entropy at time t, calculated by S5; i.e. the entropy growth rate at time t, calculated by S5; i.e. the entropy growth acceleration at time t, calculated by S5; i.e. the critical time window at time t, calculated by S6;

[0132] S8, the deep reinforcement learning model is based on the self-defined reward function to solve and output the closed-loop correction control action;

[0133] In this embodiment, this step is the core of AI decision; the action space is defined, which refers to the strategy that the agent can choose; in this embodiment, contains: maintaining the status quo, pursuing efficiency, preventive structural reinforcement, deploying high-energy active dampers;

[0134] The goal of the DRL model is to maximize the cumulative value of the self-defined reward function ; The purpose is to guide AI to learn to make a trade-off between short-term cost and long-term survival, which is calculated as follows:

[0135] ;

[0136] The reward function in this embodiment includes three components; the cost penalty term: ; this term represents the short-term cost or resource waste of performing sacrifice action or ; is the action preset cost, quantified by the estimated energy consumption, material cost, or operational disruption loss of different corrective actions; is a weight coefficient with dimension of inverse cost, e.g. 1 / yuan or 1 / Joule, to convert the cost term into a dimensionless reward value; risk penalty term: ; this term is the core of anti-fragile decision making, which directly penalizes the growth acceleration of entropy ; is derived from the calculation of S5, is a weight coefficient with dimension of to ensure this term is dimensionless; terminal state function term: ; this term defines the objective: when , a huge penalty is given; when and , a reward is given;

[0137] When the concern for future risk, the penalty of exceeds the concern for current cost , the DRL model will automatically abandon and choose ;

[0138] This step outputs the closed-loop corrective control action ;

[0139] S9, in response to the closed-loop corrective control action;

[0140] In this embodiment, this step is the physical execution phase of decision making, and constitutes the feedback loop;

[0141] If the closed-loop corrective control action is maintain status quo: the system keeps the natural frequency of the structure in S1 and the structural damping ratio unchanged;

[0142] If the closed-loop corrective control action is preventive structural reinforcement: the system executes the plan, e.g. increasing the stiffness of the target I-beam , the physical result of this action is to cause the natural frequency of the structure in S1 to increase;

[0143] If the closed-loop corrective control action is deploy active dampers: the system starts the active mass damper AMD, the physical result of this action is to cause the structural damping ratio in S1 to increase;

[0144] This step forms a key technical closed loop: the action of S9 leads to or increase; increased or is fed back to S3 as new input in the next calculation cycle, according to the formula of S3 , leading to a resonance amplification factor decrease; based on this, the decreased input S4, according to the formula of S4 , leads to a decrease in total effective stress ; The decrease in is further fed back to S5, according to the formula of S5 , leading to a suppression of the entropy growth rate and the entropy growth acceleration , , as new state input S7, the DRL model is evaluated in S8 to decrease, obtaining less punishment, thus strengthening the correctness of its sacrifice decision;

[0145] The embodiment provides a light H-beam reliability analysis method through the complete closed loop of S1 to S9; the method realizes that when facing a hidden resonance attack, the system can adaptively and prospectively sacrifice short-term operation efficiency, such as executing , at the cost of long-term survival of the structure, to solve the problem that traditional methods are difficult to deal with such high-risk attacks.

[0146] Embodiment 2:

[0147] The multi-modal sensor comprises: an acoustic sensor, a strain gauge, and an acceleration sensor;

[0148] The embodiment limits the specific composition of the multi-modal sensor; the purpose is to ensure that S1 can obtain all heterogeneous data required by subsequent steps; specifically, the acoustic sensor is specially used to identify and monitor the characteristics of external attack sources, and in the embodiment, is specifically used to collect attack frequency and attack amplitude ; at the same time, the strain gauge and the acceleration sensor are specially used to monitor the regular operation state of the structure itself, and in the embodiment, are specifically used to collect regular operation stress ;

[0149] Through this specific combination of multimodal sensors, the present invention is able to simultaneously and accurately capture external attack characteristics. , and the structure's own response This design avoids the lack of a data source, facilitating stress superposition in subsequent S4 steps. Entropy modeling in S5 provides a complete and reliable data foundation, which is a prerequisite for achieving accurate prediction and antifragile decision-making.

[0150] Example 3:

[0151] Material parameters include: initial entropy value, material degradation coefficient, material reference stress, and damage accumulation index.

[0152] This embodiment calculates the structural integrity entropy. The required material parameters were specifically defined; specifically, the initial entropy value. This refers to the damage baseline of the structure at the start of monitoring, a preset initial value, for example, 0 for a brand new structure; material degradation coefficient. and damage accumulation index This refers to an empirical coefficient reflecting the rate of damage accumulation in a material under stress, obtained through fatigue testing on the same material; material reference stress. It refers to a reference stress value used to normalize the stress, an inherent property of the material, such as the material's yield strength;

[0153] By explicitly defining this set of material parameters ( This invention relates to the structural integrity entropy model in S5. This provides clear, calibrable physical input; which makes entropy... The calculations are no longer abstract estimates, but quantitative analyses with clear physical meaning and reproducibility, greatly improving the S5 entropy model and S6 time window prediction. The accuracy.

[0154] Example 4:

[0155] S4 calculates the total effective stress, including:

[0156] Calculate the additional vibration stress based on the resonance amplification factor and the foundation stress;

[0157] The total effective stress is obtained by superimposing the additional vibration stress with the conventional operating stress.

[0158] This embodiment specifies the internal calculation logic; based on the resonance amplification factor calculated by S3. The foundation stress calculated by S2 ,pass Calculate additional vibration stress ; the purpose of this step is to modify the basic stress in the non-resonance case , through the resonance amplification factor to the amplified vibration stress actually caused by the resonance attack ; then, by , the conventional operation stress obtained by S1 is superimposed on the above , to obtain the total effective stress ; the purpose of this step is to synthesize the composite stress that the structure bears under the composite working condition;

[0159] Through this specific calculation path of first amplification and then superposition, the present application clearly decouples the contributions of the conventional operation stress and the resonance attack additional stress to the total stress in S4; this high-resolution stress modeling method ensures that the entropy increase calculation in S5 can accurately reflect the exponential damage caused by the resonance attack, and prevents the risk of underestimation due to the confusion of resonance effects with conventional stress.

[0160] Example 5:

[0161] S5 calculates the structural integrity entropy, including:

[0162] Based on the total effective stress, determine the growth rate of entropy;

[0163] wherein the growth rate of entropy is in exponential relationship with the total effective stress, and the index is the damage accumulation index.

[0164] This embodiment specifically defines the core mathematical relationship of S5 step of calculating the structural integrity entropy ; the core of S5 is to calculate , and the accumulation rate of is the growth rate of entropy ;

[0165] This embodiment clearly states that the growth rate and the total effective stress calculated by S4 follow an exponential relationship, and the specific form of the relationship is:

[0166] ;

[0167] wherein the index of the exponential relationship is the damage accumulation index ;

[0168] By comparing the entropy growth rate and the total effective stress By establishing a nonlinear exponential relationship, this invention significantly amplifies the contribution of high-stress events to structural damage; this design better reflects the physical reality of material fatigue damage, ensuring that the S5 entropy model has extremely high sensitivity to risk; when an attack occurs, and It will grow rapidly, thus providing an early warning for S6. And S8's decision, punishment It provides a strong, non-dull risk signal.

[0169] Example 6:

[0170] Custom reward functions include:

[0171] A risk penalty term used to punish the acceleration of entropy growth;

[0172] Cost penalty items used to penalize short-term costs of closed-loop corrective control actions;

[0173] And terminal state function terms used to define crash penalties and survival rewards.

[0174] This embodiment describes the custom reward function used in the S8 step DRL model. The internal composition of [the document] was specifically defined; The aim is to guide AI to make an antifragile trade-off between short-term costs and long-term survival; the function specifically includes a risk penalty term, which is embodied in this embodiment as Its purpose is to punish the acceleration of risk, namely the acceleration of entropy growth calculated by S5. ;when When the risk increases dramatically, it indicates that the risk is rapidly spiraling out of control. This provides a strong negative reward, forcing the AI ​​to immediately focus on future risks rather than current costs. These are weighting coefficients used to adjust the severity of the penalty; the function also includes a cost penalty term, which is manifested in this embodiment as... Its purpose is to punish the short-term costs of sacrificial actions; It is to perform an action The pre-set costs, such as energy consumption or operational losses, These are weighting coefficients; the function also includes a terminal state function term, which is embodied in this embodiment as... If the structure collapses in the future Then a huge crash penalty will be imposed. If the system reaches its endpoint And the structure survived successfully. Then a survival reward will be given. ;

[0175] By combining these three elements, this invention constructs a system with acceleration reward function that takes the core penalty metric as the reward; compared to traditional models that only look at the current state or velocity , the reward function of the present invention is more forward-looking; it rewards actions that can effectively suppress the entropy growth rate , even if these actions incur short-term costs , thus implementing a structural conservative decision-making logic at the algorithm level that sacrifices short-term efficiency for long-term survival.

[0176] Embodiment 7:

[0177] The feedback loop formed in response to S9 includes:

[0178] An increase in the structural natural frequency or structural damping ratio, leading to a decrease in the resonance amplification factor of S3;

[0179] A decrease in the resonance amplification factor, leading to a decrease in the total effective stress of S4;

[0180] A decrease in the total effective stress, leading to a suppression of the entropy growth rate and entropy growth acceleration of S5.

[0181] This embodiment specifically defines the internal causal chain of the physical feedback loop formed by the S9 step: when the decision of S8 is executed in S9, leading to an increase in the structural natural frequency or structural damping ratio , these changed physical parameters or will be fed back as new inputs to the S3 step; according to the formula of S3 ( ), or an increase in the structural natural frequency or structural damping ratio will inevitably lead to a decrease in the resonance amplification factor; this decrease in the resonance amplification factor will be fed back to the S4 step; according to the formula of S4, a decrease in the resonance amplification factor will inevitably lead to a decrease in the total effective stress ; accordingly, the decrease in the total effective stress in S4 will be fed back to the S5 step; according to the formula of S5, a decrease in the total effective stress will inevitably lead to a suppression of the entropy growth rate and entropy growth acceleration ;

[0182] By clearly defining this negative feedback path of S9→S3→S4→S5, the present invention ensures that the decisions of the DRL agent can produce predictable and convergent control effects on the physical reality of the system; it will abstract AI decisions closely coupled with specific physical parameters (P , ) and core risk indicators (R , ), a self-consistent and convergent closed-loop control system is constructed, ensuring the physical realizability of the anti-fragile strategy, rather than an open-loop, ineffective decision.

[0183] Embodiment 8:

[0184] Please refer to Figure 2 A light H-beam reliability analysis system comprises:

[0185] A data acquisition module is configured to acquire attack frequency, attack amplitude and regular operation stress collected by a multi-modal sensor, and structure inherent frequency and structure damping ratio determined by a structure finite element model;

[0186] A stress calculation unit is configured to calculate basic stress based on the attack amplitude and a preset machine-sound transfer coefficient;

[0187] A factor calculation unit is configured to calculate a resonance amplification factor in combination with the attack frequency, the structure inherent frequency and the structure damping ratio;

[0188] A total stress calculation unit is configured to calculate total effective stress based on the resonance amplification factor, the basic stress and the regular operation stress;

[0189] An entropy calculation unit is configured to calculate structure integrity entropy, entropy growth rate and entropy growth acceleration based on the total effective stress and preset material parameters;

[0190] A time prediction unit is configured to predict an entropy increase collapse critical time window in combination with the structure integrity entropy, the entropy growth rate, the entropy growth acceleration and a preset collapse critical point;

[0191] A decision input unit is configured to input the structure integrity entropy, the entropy growth rate, the entropy growth acceleration and the entropy increase collapse critical time window into a deep reinforcement learning model;

[0192] A decision output unit is configured to calculate and output a closed-loop correction control action based on a self-defined reward function by the deep reinforcement learning model;

[0193] A closed-loop correction unit is configured to respond to the closed-loop correction control action;

[0194] If the closed-loop correction control action is to maintain the status quo, the closed-loop correction unit keeps the structure inherent frequency and the structure damping ratio unchanged;

[0195] If the closed-loop correction control action is preventive structure reinforcement, the closed-loop correction unit increases the structure inherent frequency;

[0196] If the closed-loop correction control action is to deploy an active damper, the closed-loop correction unit increases the structural damping ratio.

[0197] The embodiment provides a light I-beam reliability analysis system, which is a physical execution carrier of the method; in the embodiment, the system comprises: a data acquisition module for corresponding to S1 function, which is used for acquiring attack frequency , attack amplitude and conventional operation stress acquired by a multi-modal sensor and structural inherent frequency and structural damping ratio determined by a structural finite element model; a stress solving unit connected to the data acquisition module, which is used for corresponding to S2 function, and is used for calculating basic stress based on and preset machine-sound transfer coefficient ; a factor solving unit connected to the data acquisition module, which is used for corresponding to S3 function, and is used for solving resonance amplification factor in combination with , and ; a total stress calculation unit connected to the data acquisition module, the stress solving unit and the factor solving unit, which is used for corresponding to S4 function, and is used for calculating total effective stress based on , and , and the unit can further comprise a sub-unit for realizing logic; an entropy calculation unit connected to the total stress calculation unit, which is used for corresponding to S5 function, and is used for calculating structural integrity entropy , entropy growth rate and entropy growth acceleration based on and preset material parameters, and the unit internally comprises calculation logic for realizing exponential relationship; a time prediction unit connected to the entropy calculation unit, which is used for corresponding to S6 function, and is used for predicting entropy increase collapse critical time window in combination with , , and preset collapse critical point ; a decision input unit connected to the entropy calculation unit and the time prediction unit, which is used for corresponding to S7 function, and is used for inputting , , and as state space to a deep reinforcement learning model; a decision output unit connected to the decision input unit, which is used for corresponding to S8 function, and is used for enabling the deep reinforcement learning model to solve and output a closed-loop correction control action , or ; and a closed-loop correction unit connected to the decision output unit, for corresponding to the S9 function, responding to the closed-loop correction control action; wherein, if the action is , the closed-loop correction unit keeps and unchanged; if the action is , the closed-loop correction unit increases ; if the action is , the closed-loop correction unit increases ;

[0198] The system of the embodiment, through the functional coupling and data flow of the above-mentioned modules, provides an end-to-end solution from multi-modal data acquisition to anti-fragile AI decision-making, and then to closed-loop physical control, constituting a light-girder reliability analysis system with complete functions and logically self-consistent.

[0199] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for reliability analysis of a lightweight I-beam, characterized by, Comprising: S1, acquiring attack frequency, attack amplitude and regular operation stress collected by multi-modal sensors, and structure inherent frequency and structure damping ratio determined by structure finite element model; S2, calculating basic stress based on attack amplitude and preset machine-sound transfer coefficient; S3, calculating resonance amplification factor combined with attack frequency, structure inherent frequency and structure damping ratio; S4, calculating total effective stress based on resonance amplification factor, basic stress and regular operation stress; S5, calculating structure integrity entropy, entropy growth rate and entropy growth acceleration based on total effective stress and preset material parameters; S6, predicting entropy increase collapse critical time window combined with structure integrity entropy, entropy growth rate, entropy growth acceleration and preset collapse critical point; S7, inputting structure integrity entropy, entropy growth rate, entropy growth acceleration and entropy increase collapse critical time window into deep reinforcement learning model; S8, deep reinforcement learning model calculating and outputting closed-loop correction control action based on self-defined reward function; S9, in response to the closed-loop correction control action: If the closed-loop correction control action is to maintain the status quo, the structure inherent frequency and the structure damping ratio remain unchanged; If the closed-loop correction control action is preventive structure reinforcement, the structure inherent frequency is increased; If the closed-loop correction control action is to deploy active damper, the structure damping ratio is increased.

2. The method of reliability analysis of a lightweight I-beam according to claim 1, wherein, The multi-modal sensor comprises an acoustic sensor, a strain gauge and an acceleration sensor.

3. The method of claim 1, wherein, The material parameters include initial entropy value, material degradation coefficient, material reference stress and damage accumulation index.

4. The method of claim 1, wherein, S4 calculates the total effective stress, comprising: calculating additional vibration stress based on resonance amplification factor and basic stress; superimposing the additional vibration stress and the regular operation stress to obtain the total effective stress.

5. The method of claim 3, wherein the method comprises: S5 calculates the structure integrity entropy, comprising: determining the growth rate of entropy based on the total effective stress; wherein the growth rate of entropy is in exponential relationship with the total effective stress, and the index is the damage accumulation index.

6. The method of reliability analysis of a lightweight I-beam as claimed in claim 1, wherein, The self-defined reward function comprises: a risk penalty term for punishing the entropy growth acceleration; a cost penalty term for punishing the short-term cost of the closed-loop correction control action; and a terminal state function term for defining collapse penalty and survival reward.

7. The method of reliability analysis of a lightweight I-beam as defined in claim 1, wherein, S9 response to the feedback loop formed, comprising: the increase of the structure inherent frequency or the structure damping ratio, resulting in the decrease of the resonance amplification factor in S3; the decrease of the resonance amplification factor, resulting in the decrease of the total effective stress in S4; the decrease of the total effective stress, resulting in the suppression of the entropy growth rate and the entropy growth acceleration in S5.

8. A light I-beam reliability analysis system based on the light I-beam reliability analysis method according to any one of claims 1-7, characterized in that, Comprising: a data acquisition module for acquiring attack frequency, attack amplitude and regular operation stress collected by multi-modal sensors, and structure inherent frequency and structure damping ratio determined by structure finite element model; a stress calculation unit for calculating basic stress based on attack amplitude and preset machine-sound transfer coefficient; a factor calculation unit for calculating resonance amplification factor combined with attack frequency, structure inherent frequency and structure damping ratio; a total stress calculation unit for calculating total effective stress based on resonance amplification factor, basic stress and regular operation stress; an entropy calculation unit for calculating structure integrity entropy, entropy growth rate and entropy growth acceleration based on total effective stress and preset material parameters; a time prediction unit configured to predict an entropy growth collapse critical time window in combination with the structural integrity entropy, the entropy growth rate, the entropy growth acceleration, and the preset collapse critical point; a decision input unit configured to input the structural integrity entropy, the entropy growth rate, the entropy growth acceleration, and the entropy growth collapse critical time window into the deep reinforcement learning model; a decision output unit configured to calculate and output a closed-loop correction control action based on a self-defined reward function by the deep reinforcement learning model; a closed-loop correction unit configured to respond to the closed-loop correction control action; wherein, if the closed-loop correction control action is to maintain the status quo, the closed-loop correction unit keeps the structural natural frequency and the structural damping ratio unchanged; if the closed-loop correction control action is to preventively reinforce the structure, the closed-loop correction unit increases the structural natural frequency; if the closed-loop correction control action is to deploy an active damper, the closed-loop correction unit increases the structural damping ratio.

Images