Airborne building factory structure health monitoring method and system thereof

Abstract

The present application relates to the technical field of structural health monitoring, in particular to an air building factory structural health monitoring method and system, in the present application, the node stress difference and displacement gradient comparison is established based on the load change rate and vibration offset, the stress state of the structure is formed into a time sequence stress response matrix, the stress dependence rate is calculated and introduced into a dynamic Bayesian network to establish the node time sequence relationship, the confidence level updating process is gradually iterated and smoothed by correcting the cycle load difference and the reaction force difference, the confidence distribution change rate is screened to select the nodes that have fallen below the threshold, the curve difference and the shape similarity are compared, and the long short-term memory network is used to predict the curve trend, so that the degradation trend can be revealed in advance and unified as a clustering trajectory, avoiding scattered results caused by abnormal judgment, and the degradation boundary recognition has high sensitivity and interval determination ability through the extraction of the curve slope, inflection point and span, combined with the rate comparison and cycle attenuation length fitting.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of structural health monitoring, in particular to a structural health monitoring method and system for an air building factory. BACKGROUND

[0002] The technical field of structural health monitoring aims to realize comprehensive evaluation and dynamic management of safety, stability and service life of engineering structures and mechanical systems through real-time detection, data analysis and state evaluation means, identify structural performance degradation, detect potential damage, predict failure risk, and provide quantitative basis for operation and maintenance and safety decision-making, serving bearing systems such as building structures, bridges, tunnels, large mechanical equipment and aerospace structures, and focusing on the mechanical response and health change law of structures under the influence of stress, vibration, temperature and environment.

[0003] The structural health monitoring method for an air building factory aims to realize quantitative evaluation and abnormal early warning of the overall structural health state of the air building factory by establishing a systematic structural state perception and data analysis mechanism, real-time acquisition of stress state, vibration response and displacement change of the building factory during construction, judgment of whether the structure is in a safe range, prevention of structural damage and safety accidents caused by local component fatigue, connection failure and uneven stress, continuous tracking of the mechanical stability of the building factory structure, accurate identification of potential fatigue risk, optimization of maintenance cycle and improvement of the overall operation safety level of the construction equipment.

[0004] The prior art mainly relies on real-time detection and static data analysis of a single index in structural health monitoring, lacks modeling of the dynamic coupling relationship between stress state, vibration response and displacement change, resulting in local judgment of state evaluation, which is difficult to reflect the global evolution trend of the structure in time sequence, and the prior art uses a fixed threshold triggering method, ignores the dynamic differences of periodic load changes and reaction force distribution, causes fatigue accumulation and potential degradation to be unable to be captured in the early stage, causes early warning lag, when the node displacement and vibration do not reach the preset threshold, the system determines that the state is safe, but the node stress distribution has been uneven, forming potential hidden dangers, the life prediction adopts experience regression and linear extrapolation, lacks extraction of curve change form and modeling of periodic attenuation law, which easily leads to large deviation between the life calculation result and the actual value, and weakens the scientificity of the maintenance strategy. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art, and a structural health monitoring method and system for an air building factory are provided.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a structural health monitoring method for an air building factory, comprising the following steps:

[0007] S1: Based on the main bearing arm, lifting guide rail, cantilever frame load rate of change and vibration offset, calculate the node stress difference and displacement gradient, compare the load increase direction and the force distribution offset, form the time sequence stress response matrix, and generate the structure stress dynamic sequence set;

[0008] S2: Based on the structure stress dynamic sequence set, calculate the guide rail node and hydraulic support arm stress dependence rate, establish the node time sequence dependence relationship by dynamic Bayesian network, compare the period load difference and the force change difference value, modify the node confidence level, update the state matrix and smooth the confidence sequence, and generate the structure health confidence distribution set;

[0009] S3: Based on the structure health confidence distribution set, compare the confidence change rate of the main bearing arm and the rotary support shaft, select the descending super threshold node, calculate the curve difference and compare the shape similarity, predict the confidence curve trend by using long short-term memory network, and divide the degradation cluster to generate the unified curve, and generate the structure degradation trajectory set;

[0010] S4: Based on the structure degradation trajectory set, extract the guide rail segment and support node curve slope, inflection point and span, compare the interval rate to judge the degradation boundary, fit the linear change rate and the period decay length, and form the degradation stage distribution set;

[0011] S5: Based on the degradation stage distribution set, calculate the bearing arm and rotary frame stage confidence descending rate and period length ratio, select the super threshold stage to extract the termination time point, substitute into the confidence expression to solve the decay time, and obtain the structure residual life index.

[0012] As a further scheme of the application, the structure stress dynamic sequence set includes load change sequence, stress distribution sequence and displacement gradient sequence, the structure health confidence distribution set includes node confidence value sequence, state transition matrix and smoothed confidence curve, the structure degradation trajectory set includes degradation curve sequence, degradation node set and confidence change rate curve, the degradation stage distribution set includes stage boundary position set, degradation rate set and period decay interval set, and the structure residual life index includes residual life value, life decay equation parameter and time distribution range.

[0013] As a further scheme of the application, the specific steps for generating the structure stress dynamic sequence set are:

[0014] Based on the main bearing arm, lifting guide rail, cantilever frame load rate of change and vibration offset, calculate the node stress value and displacement difference, judge the stress direction by the moment balance calculation between nodes, and accumulate the displacement change in continuous period to form the time sequence distribution table, and generate the node stress difference matrix;

[0015] Based on the node stress difference matrix, proportional conversion is performed on each node load change and counter force distribution value, linear superposition and interval grouping are performed according to the monitoring time axis, a continuous mapping matrix of the node stress difference in the time dimension is formed, and is defined as a time sequence stress response matrix;

[0016] Based on the time sequence stress response matrix, the stress change rate and displacement gradient distribution of each node in the continuous monitoring period are calculated, and a stress-displacement response corresponding table is established with the time sequence as the index, the overall stress response sequence relationship is formed, and a structure stress dynamic sequence set is generated.

[0017] As a further scheme of the application, the specific steps for generating the structure health confidence distribution set are:

[0018] Based on the structure stress dynamic sequence set, the stress and load change values of the guide rail nodes and the hydraulic support arms in the continuous monitoring period are collected, the stress difference between nodes is calculated and compared with the period average value to determine the dependent direction, and difference and ratio conversion are performed in time intervals, the time sequence conditional dependence relationship and state transmission structure between nodes are established by using a dynamic Bayesian network, and a node stress dependence rate set is generated;

[0019] Based on the node stress dependence rate set, the period load difference and the counter force change difference value are compared, the state change amplitude is determined by synchronous calculation of the node force value difference and the counter force deviation, and the linear correction and proportional reduction of the node confidence value deviating from the limit value are performed, and a node confidence correction matrix is generated;

[0020] Based on the node confidence correction matrix, the node confidence values are arranged in time sequence and interval smoothing processing is performed, the confidence curve balance adjustment is completed by confidence value averaging and extreme value removal in the sliding section, and the continuous mapping relationship between periods in the sequence is established, and a structure health confidence distribution set is generated;

[0021] The node confidence level, state matrix and confidence sequence, in the structure stress dynamic sequence set generated in S1, each node corresponds to a group of time sequences containing load change value, stress difference value and displacement gradient, the stress stability index of the node in the current monitoring period is obtained by standardizing the ratio calculation of the stress difference and the counter force difference at each time point in the sequence, the index is defined as the node confidence level, the confidence level values of all nodes are arranged according to the time index and a two-dimensional matrix is constructed, the rows of the matrix correspond to the node numbers, the columns correspond to the monitoring time periods, and the units in the matrix are the confidence level values, the matrix is defined as the state matrix, in the time continuous monitoring process, the confidence level change sequence of each node in the adjacent monitoring period is arranged in time sequence to form the confidence sequence.

[0022] As a further scheme of the present application, the dynamic Bayesian network first takes the monitoring data of the guide rail node, the hydraulic support arm and the main bearing arm as the input source, constructs a node set according to the stress, load and vibration change values in the continuous monitoring period, establishes the time sequence dependent connection between the node states by taking the stress difference and load ratio of each node between the previous period and the current period as the conditional variables, forms the state transition chain structure through the calculation and time index sorting of the conditional probability of adjacent nodes, corrects the conditional probability distribution according to the updated observation value in each monitoring period, so that the nodes remain dynamically associated in the time sequence, obtains the time sequence probability structure containing the node dependent relationship and state transmission direction, and generates the node stress dependent rate set.

[0023] As a further scheme of the present application, the specific steps for generating the structure degradation trajectory set are:

[0024] Based on the structure health confidence distribution set, the confidence value sequence of the main bearing arm and the rotary support shaft is extracted, the change rate is determined by calculating the confidence difference between adjacent time points, and the direction judgment and amplitude normalization are performed according to the period index to form the time sequence confidence difference record and generate the confidence change difference value set.

[0025] Based on the confidence change difference value set, the confidence decline rate threshold node is screened, the degradation trend is determined by rearranging the confidence sequence and calculating the average change rate of adjacent sections, and the fluctuation interval range and stable period length are measured in each sequence to generate the degradation node sequence set.

[0026] Based on the degradation node sequence set, the difference operation is performed on the node confidence curve and the morphological offset value is calculated, the nodes with similar morphology are aggregated by comparing the distance between curves, the long short-term memory network is used for time-dependent prediction and sequence fitting of the confidence curve, and the time synchronization balance calculation is performed in the group to form the continuous degradation trajectory sequence, and the structure degradation trajectory set is generated.

[0027] As a further scheme of the present application, the long short-term memory network takes the confidence value time sequence of each node in the degradation node sequence set as the input, constructs the confidence change rate, fluctuation amplitude and period index value of each node into a sequence vector according to the time sequence, and takes the confidence difference of adjacent time periods as the time step input unit, accumulates the influence of historical confidence change on the current node state through the state transmission between the network in the front and rear time steps, and performs state memory and output mapping operation between each time step, then performs continuous time expansion and numerical regression on the output sequence to obtain the degradation change sequence of the corresponding node, and then synchronously arranges and integrates the degradation sequences of each node according to the time index to generate the structure degradation trajectory set.

[0028] As a further scheme of the present application, the specific steps for generating the degradation stage distribution set are:

[0029] Based on the structure degradation trajectory set, the slope change rate of the guide rail segment and the support node curve is calculated, the degradation rate distribution is judged by comparing the curve shape difference in adjacent periods, and the curve turning points and trend paragraphs are located in the time interval to generate the degradation boundary interval set;

[0030] Based on the degradation boundary interval set, the rate change amount of each interval and the period attenuation value are segmented and counted, the average attenuation period length is obtained by linear segment difference value operation, and the interval attenuation ratio is sequentially arranged to form a stage division table, and a degradation stage distribution set is generated.

[0031] As a further scheme of the present application, the specific steps for generating the structure residual life index are:

[0032] Based on the degradation stage distribution set, the stage confidence decline rate and the period length ratio of the bearing arm and the rotary frame are calculated, the degradation stage boundary is identified by rate threshold determination, and the termination time of the period node is sequentially arranged to generate a life critical stage set;

[0033] Based on the life critical stage set, the termination time and the confidence decline rate are time product calculated to form a life attenuation amount table, and the cumulative attenuation amount and the period distribution parameter are multiplied and added to generate a structure residual life index.

[0034] An aerial building factory structure health monitoring system for executing the above-mentioned aerial building factory structure health monitoring method, the system comprising:

[0035] Load response acquisition module: based on the main bearing arm, the lifting guide rail, the cantilever frame load change rate and the vibration offset, the node stress difference and the displacement gradient are calculated, the direction is judged by comparing the reaction force distribution and the load increment to form the node stress matrix, and the structure stress dynamic sequence set is generated;

[0036] Time sequence dependence calculation module: based on the structure stress dynamic sequence set, the dependence rate of the guide rail node and the hydraulic support arm is calculated, the period stress difference and the reaction force difference ratio are compared, the node time sequence dependence relationship is established by using dynamic Bayesian network, the state matrix is updated and the confidence sequence is smoothed, and a structure health confidence distribution set is generated;

[0037] Confidence evolution inference module: based on the structure health confidence distribution set, the confidence change rate is calculated and the decline nodes are screened, the confidence sequence is rearranged to calculate the change interval, the long short-term memory network is used to predict the confidence curve change trend, a continuous degradation sequence is formed, and a structure degradation trajectory set is generated.

[0038] Degradation stage identification module: based on the structure degradation trajectory set, the curve slope and the inflection point are extracted, the degradation boundary is judged by comparing the interval rate, and the attenuation period is fitted to generate a degradation stage distribution set.

[0039] The life quantitative evaluation module: based on the degradation stage distribution set, the confidence decline rate and the period length ratio are calculated, the termination time point is screened, and the attenuation time is calculated by substituting into the expression, and the structural remaining life index is obtained.

[0040] Compared with the prior art, the advantages and positive effects of the present application are that:

[0041] In the present application, the node stress difference and displacement gradient comparison are established based on the load change rate and vibration offset, the stress state of the structure is formed into a time sequence stress response matrix, the continuity and dynamics of the stress information on the time axis are ensured, the stress dependence rate is calculated and introduced into the dynamic Bayesian network to build the node time sequence relationship, the confidence level updating process is gradually iterated and smoothed by correcting the period load difference and the reaction force difference, and a stable and reliable confidence distribution set is formed;

[0042] In the present application, the confidence distribution change rate is screened to select the decline threshold node, the curve difference and the shape similarity ratio are compared, and the long short-term memory network is used to predict the curve trend, so that the degradation trend can be revealed in advance and unified as a clustering trajectory, the dispersion result caused by abnormal judgment is avoided, and the degradation boundary recognition has high sensitivity and interval determination ability through the extraction of the curve slope, the inflection point and the span, combined with the rate comparison and the period decay length fitting.

[0043] In the present application, the confidence decline rate and the period length ratio are calculated, the termination time point is extracted and substituted into the confidence expression to calculate the attenuation time, the quantitative prediction of the remaining life is realized, the foresight of abnormal identification, the aggregation of degradation trajectory and the accuracy of life prediction are improved, and dynamic, continuous and quantifiable results are provided for structural safety evaluation under complex operating conditions. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The figure is a schematic diagram of the working process of the present application;

[0045] Figure 2 The figure is a system flowchart of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0047] Example 1

[0048] Please refer to Figure 1 The present application provides a technical scheme: an air building factory structural health monitoring method, comprising the following steps:

[0049] S1: Calculate the node stress difference and displacement gradient based on the load change rate and vibration offset of the main bearing arm, lifting guide rail and cantilever frame, compare the load increase direction and reaction force distribution offset, form the time sequence stress response matrix, and generate the structure stress dynamic sequence set;

[0050] S2: Based on the structure stress dynamic sequence set, calculate the stress dependence rate of the guide rail node and hydraulic support arm, establish the node time sequence dependence relationship using dynamic Bayesian network, compare the period load difference and reaction force change difference, correct the node confidence level, update the state matrix and smooth the confidence sequence, and generate the structure health confidence distribution set;

[0051] S3: Based on the structure health confidence distribution set, compare the confidence change rate of the main bearing arm and rotary support shaft, select the descending super-threshold node, calculate the curve difference and compare the shape similarity, use the long short-term memory network to predict the confidence curve trend, and divide the degradation cluster to generate the unified curve, and generate the structure degradation trajectory set;

[0052] S4: Based on the structure degradation trajectory set, extract the curve slope, inflection point and span of the guide rail segment and support node, compare the interval rate to determine the degradation boundary, fit the linear change rate and periodic decay length, and form the degradation stage distribution set;

[0053] S5: Based on the degradation stage distribution set, calculate the bearing arm and rotary frame stage confidence decline rate and period length ratio, select the super-threshold stage to extract the termination time point, substitute into the confidence expression to solve the decay time, and obtain the structure remaining life index.

[0054] The structure stress dynamic sequence set includes load change sequence, stress distribution sequence and displacement gradient sequence, the structure health confidence distribution set includes node confidence value sequence, state transition matrix and smoothed confidence curve, the structure degradation trajectory set includes degradation curve sequence, degradation node set and confidence change rate curve, the degradation stage distribution set includes stage boundary position set, degradation rate set and periodic decay interval set, and the structure remaining life index includes remaining life value, life decay equation parameter and time distribution range.

[0055] The specific steps for generating the structure stress dynamic sequence set are:

[0056] Based on the load change rate and vibration offset of the main bearing arm, lifting guide rail and cantilever frame, calculate the node stress value and displacement difference, judge the stress direction through the moment balance calculation between nodes, and accumulate the displacement change in continuous time period to form the time sequence distribution table, and generate the node stress difference matrix;

[0057] Based on the node force difference matrix, the load change and counter force distribution value of each node are proportionally converted, and linear superposition and interval grouping are performed according to the monitoring time axis to form a continuous mapping matrix of the node force difference in the time dimension, and it is defined as a time series stress response matrix;

[0058] Based on the time series stress response matrix, the force change rate and displacement gradient distribution of each node in the continuous monitoring period are calculated, and a stress-displacement response correspondence table is established with time series as index to form the overall force response sequence relationship and generate the structure force dynamic sequence set;

[0059] Based on the load change rate and vibration offset of the main bearing arm, lifting guide rail and cantilever frame, the node force is numerically solved by using the finite element analysis method, specifically: a three-dimensional model of the overall structure is established, the main bearing arm, lifting guide rail and cantilever frame are divided into equidistant eight-node cubic elements, the element size is set to 5mm, the material property parameters are set including elastic modulus MPa, Poisson's ratio 0.3, density 7850 kg / m3, three translational degrees of freedom are constrained for the fixed end node, the load change rate input value is 500 N / s, the vibration offset of the loaded node is recorded in millimeters, the node force value is calculated and the node difference value is calculated in the time step of sampling once every 0.05 seconds, the force direction is determined according to the moment balance formula between nodes, the node time series distribution table is formed by accumulating the displacement change amount of each time step, and the node force difference matrix is arranged in rows and columns according to the node number to generate the node force difference matrix;

[0060] Based on the node force difference matrix, the load change value and counter force distribution value of each node are proportionally converted, linear superposition and interval grouping calculation are performed according to the monitoring time axis, the node force difference data is input into the calculation module in time sequence, the time step is set to 0.1 seconds, the force change of each node in adjacent time steps is linearly accumulated, the load change is in units of Newton per second, the counter force distribution value is in units of Newton, and the node force change ratio is recorded in the form of ratio. Then, according to the time interval length of 5 seconds, the section is divided, the ratio value in the same section is averaged, the interval force sequence is formed, and the force difference of adjacent intervals is superimposed again to obtain the continuous distribution matrix of the node force difference with time. The matrix reflects the dynamic mapping relationship of the node force response in the time dimension, which is defined as a time series stress response matrix;

[0061] Based on the time series stress response matrix, the force change rate and displacement gradient of each node in the continuous monitoring period are calculated, and a stress-displacement response correspondence table is established with time series as index to form the overall force response sequence relationship, the sequence relationship is used to describe the force evolution characteristics of the structure under multiple nodes and multiple periods, and finally a structure force dynamic sequence set is generated as the input basis for subsequent health confidence distribution construction.

[0062] The specific steps for generating the structural health confidence distribution set are as follows:

[0063] Based on the structural stress dynamic sequence set, the stress and load change values of the guide rail nodes and the hydraulic support arms in the continuous monitoring period are collected, the force difference between nodes is calculated and compared with the period average value to determine the dependent direction, and the difference and ratio conversion are performed in time intervals, the dynamic Bayesian network is used to establish the time sequence conditional dependence relationship and state transmission structure between nodes, and the node stress dependence rate set is generated;

[0064] Based on the node stress dependence rate set, the numerical comparison of the period load difference and the reaction force change difference value is performed, the state change amplitude is determined by the synchronous calculation of the node force value difference and the reaction force deviation, and the linear correction and proportional reduction of the node confidence value deviating from the limit value are performed, and the node confidence correction matrix is generated;

[0065] Based on the node confidence correction matrix, the node confidence values are arranged in time sequence and interval smoothing processing is performed, the confidence curve balance adjustment is completed by the confidence value averaging and extreme value removal in the sliding section, and the inter-period continuous mapping relationship is established in the sequence, and the structural health confidence distribution set is generated;

[0066] The node confidence level, state matrix and confidence sequence, in the structural stress dynamic sequence set generated in S1, each node corresponds to a group of time sequences containing load change value, stress difference value and displacement gradient, the force stability degree index of the node in the current monitoring period is obtained by standardizing the ratio calculation of the force difference and reaction force difference at each time point in the sequence, the index is defined as the node confidence level, the confidence level values of all nodes are arranged according to time index and a two-dimensional matrix is constructed, the rows of the matrix correspond to the node numbers, the columns correspond to the monitoring time periods, and the units in the matrix are the confidence level values, the matrix is defined as the state matrix, in the time continuous monitoring process, the confidence level change sequence of each node in the adjacent monitoring period is arranged in time sequence to form the confidence sequence;

[0067] Based on the structural stress dynamic sequence set, the stress and load change values of the guide rail node and the hydraulic support arm in the continuous monitoring period are collected, the dynamic Bayesian network is used to establish the timing conditional dependence relationship and state transmission structure, the preset period length is 600 seconds and the sampling frequency is 100 Hz, the preset time interval length is 2 seconds and the step is 1 second, the stress difference between nodes is calculated and compared with the period average value to determine the dependence direction, the difference is subtracted by the adjacent time step sequence and the endpoint mirror extension method, the proportional conversion uses the current difference value divided by the interval mean and records the sign bit, the discretization uses the three-state division scheme, the stress low interval is 0 to 80 MPa, the medium interval is 80 to 160 MPa, and the high interval is 160 to 240 MPa, the load low interval is 0 to 2000 Newton, the medium interval is 2000 to 4000 Newton, and the high interval is 4000 to 6000 Newton, the network structure sets two state nodes of guide rail node stress and load and two state nodes of hydraulic support arm stress and load in each time slice, and sets the cross-time slice self-loop and the edge from the dependent node to the dependent node in the dependence direction, the conditional probability table initialization uses Dirichlet prior parameters of 1, 1 and 1, the state transition matrix initialization uses self-maintaining probability of 0.7, forward transition probability of 0.2 and reverse transition probability of 0.1, the training uses expectation maximization algorithm with iteration number of 50 and convergence threshold of 1e-4, the missing segment processing uses the combination strategy of forward filling and interval mean interpolation, the proportion threshold is 10 percent, the output of each node in each time slice corresponding to the dependence direction and the conditional probability value is combined into the dependence rate sequence, and the node stress dependence rate set is generated;

[0068] Based on the node stress dependence rate set, the numerical comparison of the period load difference and the reaction force change difference value is carried out, the synchronous timestamp alignment method and the linear interpolation method are used to fill the gap, the state change amplitude calculation uses the weighted summation scheme with weight load difference of 0.6 and weight reaction force difference of 0.4, the deviation limit value setting uses the median absolute deviation multiplied by three rules and calculated independently according to the node, after time alignment, the node force value difference and reaction force deviation of each node in each time slice are calculated synchronously and the sign bit is recorded, the confidence value range is limited to 0 to 1 with rounding precision of four decimal places, the linear correction is performed on the confidence value of the deviation limit value node, the correction coefficient k is 0.2, the deviation amount is multiplied by the coefficient to perform the deduction processing and the sign is kept consistent, the corrected confidence value is further executed by the proportional reduction, the reduction coefficient is 0.85, which is applied independently according to the node and the boundary clipping is completed in each time slice, the matrix layout uses time as row, node as column and confidence as unit value, the index uses time stamp in ascending order and node number in ascending order, and the node confidence correction matrix is generated.

[0069] Based on the node confidence correction matrix, the node confidence value is arranged in time sequence and interval smoothing processing is performed. The smoothing adopts a sliding average window length of 15 sample steps and a weight of 5 samples. The extreme value processing adopts a Hampel filter window length of 15 samples, a threshold coefficient of 3, and a replacement strategy of the window median. The smoothing sequence adopts the order of first extreme value elimination and then sliding average, and is executed independently according to the node. The cross-cycle mapping adopts a segmented linear interpolation method to insert 20 connection samples at the adjacent cycle boundary with a sampling rate of 10 Hz and the boundary values at both ends as the end points. The sequence alignment adopts a uniform time axis length equal to the total number of samples in each cycle, and performs linear filling on the missing section. The discrete distribution construction adopts a box number of 100 and a box width of 0.01, and the count is normalized to the range of 0 to 1. The output generates a time-aligned sequence and distribution direct vector and a cycle label triplet set according to the node, and generates a structural health confidence distribution set.

[0070] The dynamic Bayesian network first takes the monitoring data of the guide rail node, the hydraulic support arm and the main bearing arm as the input source, constructs a node set according to the stress, load and vibration change values in the continuous monitoring period, and takes the stress difference and load ratio of each node between the previous cycle and the current cycle as the conditional variable to establish the time sequence dependence connection between the node states. Through the calculation and time index sorting of the conditional probability of adjacent nodes, a state transition chain structure is formed. In each monitoring period, the conditional probability distribution is corrected according to the updated observation value, so that the nodes are dynamically associated in time sequence, and a time sequence probability structure containing node dependence relationship and state transmission direction is obtained, and a node stress dependence rate set is generated.

[0071] The dynamic Bayesian network is according to the formula:

[0072]

[0073] Wherein: represents the stress value of the air building factory guide rail node at time t, represents the stress value of the air building factory hydraulic support arm at time t, represents the average stress difference of the guide rail node and the hydraulic support arm in the monitoring period, represents the hydraulic difference change amount of the guide rail node and the hydraulic support arm at time t, represents the temperature gradient difference of the guide rail node and the hydraulic support arm at time t, represents the friction coefficient correction term between the guide rail node and the hydraulic support arm, represents the weight coefficient of the stress difference, represents the weight coefficient of the hydraulic difference, represents the weight coefficient of the temperature gradient difference, represents the weight coefficient of the friction correction term, This represents the improved stress dependence rate calculation value of the aerial building factory at time t.

[0074] Execution process: First, collect stress data sequences of the guide rail nodes and hydraulic support arms during continuous monitoring cycles. and The structural stress state at different time points is recorded in real time using a high-precision strain sensing unit. Then, the stress difference between nodes at each time point is calculated, and the average stress difference within the monitoring period is obtained. It is used to establish a benchmark scale for stress changes and to extract real-time pressure change data of the hydraulic system to calculate the hydraulic differential. The system pressure normalization module corrects the stress level to reflect the load adjustment effect of the support arm, while collecting nodal surface temperature data to calculate the temperature gradient difference. The stress offset effect is corrected using the thermal expansion coefficient of the structural material to compensate for stress errors under high-temperature conditions. Then, the friction coefficient correction term between nodes is calculated using a friction monitoring device. This is used to characterize the nonlinear effect of the contact resistance between the guide rail and the support arm on force transmission. The stress difference, hydraulic difference, temperature difference, and friction correction term are multiplied by weighting coefficients respectively. , , , Furthermore, by using historical monitoring data and employing the minimum mean square error method to determine the weight combination, it can dynamically balance multiple physical influencing factors when calculating the improved stress dependence rate between nodes. This reflects the structural health status between the guide rail nodes and hydraulic support arms in the complex working environment of the aerial building factory, enabling quantitative diagnosis of structural load transfer characteristics and potential abnormal trends.

[0075] The specific steps for generating a set of structural degradation trajectories are as follows:

[0076] Based on the structural health confidence distribution set, the confidence value sequence of the main bearing arm and the slewing support shaft is extracted. The rate of change is determined by calculating the confidence difference between adjacent time points. The direction judgment and amplitude normalization are performed according to the period index to form a time series confidence difference record and generate a confidence change difference set.

[0077] Based on the confidence change difference set, nodes with confidence decrease rate exceeding the threshold are screened, and the degradation trend is determined by rearranging their confidence sequences and calculating the average change rate of adjacent segments. The fluctuation range and stable period length are measured in each sequence to generate a set of degradation node sequences.

[0078] Based on the set of degenerate node sequences, the difference operation is performed on the node confidence curve and the morphological offset value is calculated, the distance between curves is compared to aggregate the node set with similar morphology, the long short-term memory network is used for time-dependent prediction and sequence fitting of the confidence curve, and time synchronization balance calculation is performed within the group to form a continuous degeneration trajectory sequence, and a structure degeneration trajectory set is generated;

[0079] Based on the set of structural health confidence distribution, the confidence value sequence of the main bearing arm and the rotary support shaft is extracted, the time series difference algorithm is used to calculate the confidence change rate, the time sampling interval is set to 0.1 seconds, the pre-defined confidence value range is 0 to 1, the difference order is 1, the end points are filled with mirror image, each end is supplemented with 5 sample points, the instantaneous change is obtained by calculating the confidence value difference between adjacent time points, and the change rate array is formed by taking the absolute value of the change and taking the mean value. Direction judgment is performed by sign function to identify positive and negative directions, amplitude normalization is used to normalize the interval to 0 to 1, direction marking and amplitude remapping are performed for each time slice according to the cycle index, the confidence change rate time sequence is output, the confidence difference sequence is recorded in time order, and the confidence change difference set is generated;

[0080] Based on the set of confidence change difference values, the threshold screening and moving average method are used to determine the degenerate nodes, the confidence drop rate threshold is set to 0.05, the sliding window length is 10 sample points, and the step length is 5 sample points. The window average is performed on each node confidence change sequence, the average drop rate in the window is extracted and compared with the threshold, the super-threshold node index is marked, the super-threshold node confidence sequence is rearranged and aligned in time order, the average change rate is calculated for adjacent segments, the interval length is set to 3 seconds, the time span is increased by a fixed step, the variance of each sequence is calculated and the fluctuation interval range is recorded, the period length is calculated by the first peak distance of the autocorrelation function and is rounded to the stable period length. The degenerate node index and its corresponding average change rate and period length are recorded, and the degenerate node sequence set is generated;

[0081] Based on the set of degenerate node sequence, the first-order difference and morphological distance calculation method is performed on the node confidence curve, the difference step is 1, the end point is filled by forward copying, the morphological offset value calculation adopts Euclidean distance and is normalized in time dimension, the distance matrix is calculated for all node curves and the clustering operation is performed, the clustering method is hierarchical clustering algorithm, the shortest distance is taken as the linkage method, the number of clusters is set to 5, the morphologically similar nodes are divided into the same set, then the long short-term memory network is used for time-dependent prediction and sequence fitting, the network structure includes input layer dimension 1, hidden layer 2 layers, each layer unit number 64, output layer dimension 1, the activation function adopts hyperbolic tangent function, the optimizer selects Adam, the learning rate is 0.001, the batch size is 32, the training rounds are 100, the loss function adopts mean square error, the time synchronization balance calculation is performed on the confidence curve in each clustering group, the balance strategy is interpolation to the uniform time axis length 1000 sample points, the continuous degradation trajectory sequence is generated by stacking according to the time index, and the structural degradation trajectory set is generated.

[0082] The long short-term memory network takes the time sequence of the confidence value of each node in the set of degenerate node sequences as input, constructs the confidence change rate, fluctuation amplitude and period index value of each node into a sequence vector according to time sequence, and takes the confidence difference of adjacent time periods as a time step input unit. Through the state transmission between the network in front and back time steps, the influence of historical confidence change on the current node state is accumulated and expressed. State memory and output mapping operations are performed between each time step. Then, the output sequence is continuously expanded in time and regressed in value to obtain the degradation change sequence of the corresponding node. The degradation sequences of each node are arranged synchronously according to the time index and integrated into trajectories to generate the structural degradation trajectory set.

[0083] The long short-term memory network is according to the formula:

[0084]

[0085] Wherein: represents the improved hidden state output of the air building factory at time t, represents the activation function, represents the input weight matrix, represents the input feature vector of the degenerate node confidence curve at time t, represents the state transmission weight matrix, represents the hidden state vector at the previous time t-1, represents the bias term, represents the morphological skewness weight coefficient, represents the morphological skewness index, represents the curvature energy item weight coefficient, represents the curvature energy density, represents the aggregation condensation item weight coefficient, Indicates the aggregation coefficient. This represents the weighting coefficient of the time synchronization item. This indicates the time synchronization balance offset;

[0086] Execution process: First, confidence curve sequences of each guide rail node and hydraulic support arm are collected. The rate of change of node confidence values ​​over time is calculated through continuous monitoring cycles to obtain morphological offset. The difference results are then concatenated with the morphological features to form the input feature vector. The input is then fed into the input layer and processed by the input weight matrix. Mapping to the hidden state space, the temporal variation trend of node degradation characteristics is characterized, and the hidden state of the previous time step is also mapped. via state transfer weight matrix Passing the process to the current time step to maintain the time dependency of the degradation process and adding a bias term To eliminate systematic prediction bias, a morphological skewness index is introduced. The curvature energy density is then calculated by normalizing the third central moment of the confidence curve to measure its asymmetry. The intensity of curve fluctuations is represented by the second-order difference squared mean, while the aggregation coefficient is calculated by the reciprocal of the normalized distance between the nodal curve and the within-group Fréchet mean curve. To reflect the intra-group concentration of the curve, the time synchronization balance offset is then obtained using a local least squares alignment method. To correct the time misalignment between the curves, the four innovation parameters are multiplied by their corresponding weights. The sum of the weighted terms and the principal terms is then input into the activation function. The activation function performs a non-linear mapping on all terms to output the improved hidden state at the current time step. By iterating continuously over time steps to form a degradation trajectory sequence, the resulting degradation trajectory set of the "sky-building factory" can be used to identify structural performance degradation trends and potential degradation risks, and to achieve dynamic prediction and reliability assessment of the overall structural health status.

[0087] The specific steps for generating the degradation stage distribution set are as follows:

[0088] Based on the set of structural degradation trajectories, the slope change rate of the guide rail segment and support node curves is calculated. By comparing the differences in curve shape within adjacent periods, the degradation rate distribution is determined, and the curve inflection points and trend segments are located within the time interval to generate a set of degradation boundary intervals.

[0089] Based on the set of degradation boundary intervals, the rate change and period decay value of each interval are statistically analyzed. The average decay period length is obtained by linear segment difference calculation. The interval decay ratios are arranged in order to form a stage division table, generating a set of degradation stage distribution.

[0090] Based on the structure degradation trajectory set, the segmented linear regression algorithm is used to calculate the curve slope rate of the guide rail segment and the support node. The input data is the time sequence of the node degradation trajectory, the time sampling interval is set to 0.1 seconds, the segmented interval length is set to 50 sample points, the linear regression process is independently executed according to each time period, the time average and the confidence value average of each interval sample are calculated, the offset of each sample point relative to the time average is calculated, and the product sum and square sum of the offset are accumulated, then the slope value is obtained by dividing the two, the intercept is obtained by subtracting the slope multiplied by the time average from the confidence value average, the slope sequence is formed by arranging the slope of each interval in turn, the slope change rate sequence is obtained by dividing the slope difference of adjacent intervals by the interval time interval, the unit is megapascal per second, the curve patterns of adjacent cycles are compared, the dynamic time warping algorithm is used to perform pattern matching, the Euclidean distance is set as the distance measurement method, the constraint step is 2 sampling points, and the time matching error tolerance is 0.05 seconds. The shortest distance path is recorded by performing column-by-column scanning on the shape difference matrix, and the local minimum value region in the path is marked as a high similarity interval. The turning points are detected in the time sequence by using the second derivative sign change detection, the turning point detection threshold is set to 1.2 times the average change rate of the derivative, the time period between consecutive turning points is divided into trend intervals, the trend direction is determined according to the positive and negative of the slope and is recorded in time sequence, and the degradation boundary interval set is generated.

[0091] Based on the degradation boundary interval set, the linear segment difference value and cycle calculation algorithm is used to perform segment statistics on the rate change and cycle attenuation value of each interval. The input is the start and end time of the degradation boundary interval, the slope sequence and the confidence value sequence. First, the slope difference between the first and last of each interval is calculated according to the time sequence to record the rate change, with the unit of megapascal per second. Then, the average change of the confidence value of each interval is calculated according to the time interval length to obtain the cycle attenuation value. The cycle division length is set to 10 seconds. The average ratio of the rate change and the cycle attenuation value in each cycle is recorded. The difference between the rate change and the cycle attenuation of each interval is linearly superimposed and averaged to obtain the average attenuation cycle length, with the unit of seconds. The average attenuation length of all intervals is arranged in time sequence, and the interval attenuation ratio sequence is generated. The attenuation ratio is defined as the proportion of the attenuation length of a single interval to the total cycle length, with the value ranging from 0 to 1. Finally, the attenuation ratios of all intervals are arranged in order to form a stage index list. The stage numbers are sequentially ordered and saved to a two-dimensional array structure to generate the degradation stage distribution set.

[0092] The specific steps of generating the structure residual life index are as follows:

[0093] Based on the degradation stage distribution set, the stage confidence drop rate and cycle length ratio of the bearing arm and the rotary frame are calculated. The degradation stage boundary is identified by the rate threshold value judgment, and the termination time of the cycle node is arranged in order to generate the life critical stage set.

[0094] Based on the life critical stage set, the termination time is time multiplied with the confidence decline rate to form the life attenuation table, and the cumulative attenuation is multiplied and added with the cycle distribution parameter to generate the structure residual life index;

[0095] Based on the degradation stage distribution set, the threshold segmentation algorithm is used to calculate the confidence decline rate and the cycle length ratio of the bearing arm and the rotary frame stage, the input data includes the confidence value sequence and the time index sequence, the confidence decline rate is calculated by dividing the confidence difference at the beginning and end of the stage by the corresponding cycle time length, the time resolution is set to 0.1 seconds, the result unit is confidence value per second, the decline rate of each stage is calculated and recorded, and the cycle length is calculated and recorded as the cycle length column according to the stage start and end time difference, then the rate ratio sequence is generated by the rate and cycle length ratio operation, the threshold is set to 1.3 times the average value of the decline rate, the threshold is determined by comparing each item of the rate ratio sequence, the determination rule is that greater than the threshold is marked as a degradation boundary point, and less than and equal to the threshold is marked as a stable stage, the time index of the marked boundary point is extracted to obtain the cycle node termination time, the termination time is arranged in time sequence, numbered and output as a time sequence, and the life critical stage set is generated;

[0096] Based on the life critical stage set, the termination time is time multiplied with the confidence decline rate to form the life attenuation table, and the cumulative attenuation is multiplied and added with the cycle distribution parameter to generate the structure residual life index.

[0097] Please refer to Figure 2 , the air building factory structure health monitoring system is used to execute the above-mentioned air building factory structure health monitoring method, the system comprises:

[0098] Load response acquisition module: based on the load change rate and vibration offset of the main bearing arm, lifting guide rail and cantilever frame, the node force difference and displacement gradient are calculated, the direction is determined by comparing the reaction force distribution and load increment, the node force matrix is formed, and the structure stress dynamic sequence set is generated;

[0099] The time sequence dependent calculation module calculates the guide rail node and the hydraulic support arm dependent rate based on the structural stress dynamic sequence set, compares the period stress difference and the reaction force difference ratio, establishes the node time sequence dependent relationship by using a dynamic Bayesian network, updates the state matrix and smooths the confidence sequence, and generates a structural health confidence distribution set;

[0100] The confidence evolution inference module calculates the confidence change rate and selects the descending nodes based on the structural health confidence distribution set, rearranges the confidence sequence to calculate the change interval, predicts the confidence curve change trend by using a long short-term memory network, forms a continuous degradation sequence, and generates a structural degradation trajectory set;

[0101] The degradation phase recognition module extracts the curve slope and the inflection point based on the structural degradation trajectory set, compares the interval rate to judge the degradation boundary and fits the decay period, and generates a degradation phase distribution set.

[0102] The life quantitative evaluation module calculates the confidence decline rate and the period length ratio based on the degradation phase distribution set, selects the termination time point, and substitutes it into the expression to calculate the decay length, and obtains a structural remaining life index.

[0103] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any skilled person in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. An aerial building factory structure health monitoring method, characterized by, The method comprises the following steps: S1: Based on the main bearing arm, lifting guide rail, cantilever frame load rate of change and vibration offset, calculate the node stress difference and displacement gradient, compare the load increase direction and the force distribution offset, form the time sequence stress response matrix, and generate the structure stress dynamic sequence set; S2: Based on the structure stress dynamic sequence set, calculate the guide rail node and hydraulic support arm stress dependence rate, establish the node time sequence dependence relationship by using dynamic Bayesian network, compare the period load difference and the force change difference, modify the node confidence level, update the state matrix and smooth the confidence sequence, generate the structure health confidence distribution set; S3: Based on the structure health confidence distribution set, compare the confidence change rate of the main bearing arm and the rotary support shaft, screen the descending super-threshold node, calculate the curve difference and compare the shape similarity, predict the confidence curve trend by using long short-term memory network, and divide the degradation cluster to generate the unified curve, generate the structure degradation trajectory set; S4: Based on the structure degradation trajectory set, extract the guide rail segment and support node curve slope, inflection point and span, compare the interval rate to judge the degradation boundary, fit the linear change rate and the period decay length, and form the degradation stage distribution set; S5: Based on the degradation stage distribution set, calculate the bearing arm and rotary frame stage confidence descending rate and period length ratio, screen the super-threshold stage to extract the end time point, substitute into the confidence expression to solve the decay time, and obtain the structure remaining life index.

2. The air fabricated factory structure health monitoring method of claim 1, wherein, The structure stress dynamic sequence set comprises a load change sequence, a stress distribution sequence and a displacement gradient sequence, the structure health confidence distribution set comprises a node confidence value sequence, a state transition matrix and a smoothed confidence curve, the structure degradation trajectory set comprises a degradation curve sequence, a degradation node set and a confidence change rate curve, the degradation stage distribution set comprises a stage boundary position set, a degradation rate set and a period decay interval set, and the structure remaining life index comprises a remaining life value, a life decay equation parameter and a time distribution range.

3. The air fabricated factory structure health monitoring method of claim 1, wherein, The specific steps for generating the structure stress dynamic sequence set are: Based on the main bearing arm, lifting guide rail and cantilever frame load rate of change and vibration offset, calculate the node stress value and displacement difference, judge the stress direction through the moment balance calculation between nodes, and accumulate the displacement change in a continuous time period to form a time sequence distribution table, and generate a node stress difference matrix; Based on the node stress difference matrix, perform proportional conversion on the load change and reaction force distribution value of each node, perform linear superposition and interval grouping according to the monitoring time axis, form a continuous mapping matrix of the node stress difference in the time dimension, and define it as a time sequence stress response matrix; Based on the time sequence stress response matrix, calculate the stress change rate and displacement gradient distribution of each node in the continuous monitoring period, and establish a stress-displacement response correspondence table with time sequence as the index, form the overall stress response sequence relationship, and generate the structure stress dynamic sequence set.

4. The air fabricated factory structure health monitoring method of claim 1, wherein, The specific steps for generating the structure health confidence distribution set are: Based on the structure stress dynamic sequence set, the stress and load change values of the guide rail node and the hydraulic support arm in the continuous monitoring period are collected, the force difference between nodes is calculated and compared with the period average value to determine the dependent direction, and the difference and ratio conversion are performed in time interval units, the time sequence conditional dependence relationship and state transmission structure between nodes are established by using dynamic Bayesian network, and the node stress dependence rate set is generated; Based on the node stress dependence rate set, the numerical comparison of the period load difference and the reaction force change difference value is performed, the state change amplitude is determined by synchronous calculation of the node force value difference and the reaction force deviation, and the linear correction and proportional reduction of the node confidence value deviating from the limit value are performed, and the node confidence correction matrix is generated; Based on the node confidence correction matrix, the node confidence values are arranged in time sequence and interval smoothing processing is performed, the confidence curve balance adjustment is completed by confidence value averaging and extreme value removal in sliding section, and the continuous mapping relationship between periods in the sequence is established, and the structure health confidence distribution set is generated.

5. The air fabricated factory structure health monitoring method of claim 4, wherein, The dynamic Bayesian network first takes the monitoring data of the guide rail node, the hydraulic support arm and the main bearing arm as the input source, constructs the node set according to the stress, load and vibration change values in the continuous monitoring period, and establishes the time sequence dependent connection between the node states by taking the stress difference and load ratio of each node between the previous period and the current period as the conditional variables, forms the state transition chain structure by calculating and time index sorting the conditional probability of adjacent nodes, and corrects the conditional probability distribution according to the updated observation value in each monitoring period, so that the nodes remain dynamically associated in time sequence, obtains the time sequence probability structure containing the node dependence relationship and state transmission direction, and generates the node stress dependence rate set.

6. The air fabricated factory structure health monitoring method of claim 1, wherein, The specific steps of generating the structure degradation trajectory set are: Based on the structure health confidence distribution set, the confidence value sequence of the main bearing arm and the slewing support shaft is extracted, the confidence difference between adjacent time points is calculated to determine the change rate, and the direction judgment and amplitude normalization are performed according to the period index to form the time sequence confidence difference record, and the confidence change difference value set is generated; Based on the confidence change difference value set, the confidence drop rate threshold node is screened, the confidence sequence is rearranged, the average change rate of adjacent sections is calculated to determine the degradation trend, and the fluctuation interval range and stable period length are measured in each sequence to generate the degradation node sequence set; Based on the degradation node sequence set, the difference operation is performed on the node confidence curve and the form deviation value is calculated, the nodes with similar forms are aggregated by comparing the distance between curves, the long short-term memory network is used for time-dependent prediction and sequence fitting of the confidence curve, and the continuous degradation trajectory sequence is formed by performing time synchronization balance calculation in the group to generate the structure degradation trajectory set.

7. The air fabricated factory structure health monitoring method of claim 6, wherein, The long short-term memory network takes the time sequence of confidence values of each node in the degradation node sequence set as input, constructs the confidence change rate, fluctuation amplitude and cycle index value of each node into a sequence vector in chronological order, takes the confidence difference between adjacent time periods as a time step input unit, accumulates the influence of historical confidence changes on the current node state through state transmission between the front and rear time steps, performs state memory and output mapping operations between time steps, then performs continuous time expansion and numerical regression on the output sequence to obtain the degradation change sequence of the corresponding node, and then arranges and integrates the degradation sequences of each node according to the time index to generate a structure degradation trajectory set.

8. The method of claim 1, wherein, The specific steps for generating the degradation stage distribution set are: Based on the structure degradation trajectory set, the slope change rate of the guide rail segment and the support node curve is calculated, the degradation rate distribution is judged by comparing the curve shape difference in adjacent periods, and the curve turning points and trend paragraphs are located in the time interval to generate a degradation boundary interval set; Based on the degradation boundary interval set, the rate change amount and the period attenuation value of each interval are statistically analyzed, the average attenuation period length is obtained by linear segment difference value operation, and the interval attenuation ratio is sequentially arranged to form a stage division table to generate a degradation stage distribution set.

9. The air fabricated factory structure health monitoring method of claim 1, wherein, The specific steps for generating the structure remaining life index are: Based on the degradation stage distribution set, the confidence decline rate and cycle length ratio of the bearing arm and the rotary frame stage are calculated, the degradation stage boundary is identified by rate threshold determination, and the termination time of the cycle node is sequentially arranged to generate a life critical stage set; Based on the life critical stage set, the termination time and confidence decline rate are calculated by time product calculation to form a life attenuation amount table, and the accumulated attenuation amount and cycle distribution parameters are multiplied and added to generate a structure remaining life index.

10. An aerial building factory structure health monitoring system, characterized by, The system comprises: A load response collection module: based on the load change rate and vibration offset of the main bearing arm, lifting guide rail and cantilever frame, the node force difference and displacement gradient are calculated, the direction is judged by comparing the reaction force distribution and load increment to form a node force matrix, and a structure stress dynamic sequence set is generated; A time sequence dependent calculation module: based on the structure stress dynamic sequence set, the dependence rate of the guide rail node and the hydraulic support arm is calculated, the period stress difference and reaction force difference ratio are compared, the node time sequence dependence relationship is established by using a dynamic Bayesian network, the state matrix is updated and the confidence sequence is smoothed to generate a structure health confidence distribution set; A confidence evolution inference module: based on the structure health confidence distribution set, the confidence change rate is calculated and the declining nodes are screened, the confidence sequence is rearranged to calculate the change interval, the long short-term memory network is used to predict the confidence curve change trend to form a continuous degradation sequence, and a structure degradation trajectory set is generated; A degradation stage identification module: based on the structure degradation trajectory set, the curve slope and inflection point are extracted, the degradation boundary is judged by comparing the interval rate and the attenuation period is fitted to generate a degradation stage distribution set; A life quantitative evaluation module: based on the degradation stage distribution set, the confidence decline rate and cycle length ratio are calculated, the termination time point is screened and substituted into the expression to calculate the attenuation time, and a structure remaining life index is obtained.

Images