Health state prediction method and device, equipment and storage medium

Through the deep learning technology based on neural network and dynamic Bayesian network, the key features of the health status of the embedded bolt are extracted, and the problem of low accuracy and reliability of traditional methods in noise data is solved, achieving a more efficient health status predict.

CN119066357APending Publication Date: 2024-12-03FASHIDA (DALIAN) IND GRP CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411079740.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional embedded bolt health status prediction methods are difficult to effectively distinguish between real state information and interfering signals when facing noise data, resulting in a decrease in the accuracy and reliability of the prediction results.

Method used

Using deep learning technology based on neural networks, by constructing an embedded bolt state prediction model, estimating the health status transition probability matrix and observed probability matrix, the key features affecting the health status of embedded bolts are automatically extracted, and a dynamic Bayesian network is constructed to improve the Prediction accuracy and robustness.

Benefits of technology

Improve the accuracy and reliability of predicting health status of embedded bolts, and can capture the complex condition dependencies between variables, and update the condition probability distribution as time series data evolves, reflecting the changes in embedded bolt state over time more accurately. law.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066357A_ABST
    Figure CN119066357A_ABST
Patent Text Reader

Abstract

The invention relates to the field of embedded bolt detection, in particular to a health state prediction method and device, equipment and a storage medium. The health state prediction method comprises the steps of estimating parameter values of an embedded bolt state prediction model according to the pre-constructed embedded bolt state prediction model; the parameter values comprise a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is used for establishing a transition relation between health states of the embedded bolt in a preset time sequence; the observation value probability matrix is used for establishing a corresponding relationship between different observation values of the embedded bolt and each health state; and determining each health state of the embedded bolt in the time sequence according to the parameter values. According to the method, the prediction accuracy and robustness are effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of embedded bolt detection, and in particular to a method, device, equipment and storage medium for predicting the health status of embedded bolts. Background Art

[0002] In the field of health status prediction of embedded bolts, traditional methods often focus on directly analyzing the raw signals collected from sensors, trying to identify possible defects or signs of damage to the embedded bolts. However, these methods often have a significant performance hit when faced with a large amount of noisy data caused by factors such as sensor failure and environmental interference. In particular, when the data is highly chaotic in the frequency domain, it is difficult for traditional methods to effectively distinguish between real status information and interference signals, resulting in reduced accuracy and reliability of the prediction results.

[0003] Based on this, the present invention is applicable to prediction models based on neural networks, especially deep learning technology, to improve the prediction accuracy of the health status of embedded bolts. In particular, by using the powerful nonlinear fitting ability and pattern recognition ability of neural networks, deep learning is performed on massive monitoring data, so as to automatically extract the key features that affect the health status of embedded bolts, even if these features are hidden in complex and chaotic data, thereby improving the accuracy and reliability of the prediction results. Summary of the invention

[0004] The embodiments of the present invention provide a health status prediction method, device, equipment and storage medium, which are used to at least improve the accuracy and reliability of health status prediction of embedded bolts.

[0005] In one aspect, the present invention provides a health status prediction method, the health status prediction method comprising:

[0006] According to a pre-built embedded bolt state prediction model, the parameter values ​​of the embedded bolt state prediction model are estimated; the parameter values ​​include a health state transition probability matrix and an observation value probability matrix of the embedded bolts; the health state transition probability matrix is ​​used to establish a transition relationship between various health states of the embedded bolts in a preset time series; the observation value probability matrix is ​​used to establish a corresponding relationship between different observation values ​​of the embedded bolts and various health states;

[0007] According to the parameter value, each health state of the embedded bolt in the time series is determined.

[0008] Optionally, estimating parameter values ​​of the embedded bolt state prediction model includes:

[0009] According to the health status S of the embedded bolts in the time series i Transfer to healthy state S jThe total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or

[0010] According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

[0011] Optionally, the observed value includes a pull-out force value and a displacement value; the parameter value also includes an initial health state probability of the embedded bolt; and the health state prediction method further includes:

[0012] Update the initial health status probability of the embedded bolts:

[0013] π i =γ1(i);

[0014] Initial health state probability π i Indicates that the embedded bolt is in a healthy state S at the initial time t = 1 i γ1(i) indicates the probability of observing the observation sequence at the initial time t = 1 and being in the healthy state S i probability;

[0015] The health state transition probability matrix of the embedded bolt is updated using the following formula: ij :

[0016]

[0017] In the formula, a ij The embedded bolt changes from the healthy state S in the time series i Transfer to healthy state S j probability; Indicates the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times; t (i, j) indicates that the embedded bolt is in the healthy state S at time t in the time series i And at time t+1 it is in the healthy state S j probability; Indicates that the embedded bolts are in a healthy state S i The total number of times; γ t (i) indicates that the embedded bolt is in the healthy state S at time t iThe probability of; 1,2,3,...,t,...,T is a time series, and the time between any two moments in the time series is a time step;

[0018] The observed value probability b of the observed value probability matrix of the embedded bolts is updated using the following formula: j (k):

[0019]

[0020] Where b j (k) indicates that in the health state S j Lower observed value O t is the probability of k; Indicates the health status S of the embedded bolts in the time series j The total number of observations under k; δ(O t =k) ​​has a value of 1 or 0; Indicates that the embedded bolts are in a healthy state S j The total number of times; γ t (j) indicates that the embedded bolt is in the healthy state S at time t j probability.

[0021] Optionally, the health status prediction method further includes:

[0022] Construct the initial health state probability π of the embedded bolt as shown in the following formula;

[0023] π i =P(S1=S i );

[0024] Construct the health state transition probability matrix A of the embedded bolts as shown in the following formula:

[0025] A={a ij}, a ij =P(S t+1 =S j |S t =S i );

[0026] Construct the observation value probability matrix B of the embedded bolts as shown in the following formula:

[0027] B = {b jk}, b jk =P(O t =O k |S t =S j );

[0028] According to the initial health state probability π of the embedded bolts, the health state transition probability matrix A of the embedded bolts and the observed value probability matrix B of the embedded bolts, the embedded bolt state prediction model λ=(π, A, B) is constructed; where P() represents probability calculation; O represents observed value;

[0029] According to the embedded bolt state prediction model, determine γ t (i) and ξ t (i,j).

[0030] Alternatively, γ is determined using the following formula: t (i)

[0031] and / or

[0032] Use the following formula to determine ξ t (i, j):

[0033]

[0034] in,

[0035] α t (i) = P(O1, O2, ..., O t , S t =S i |λ);

[0036]

[0037] α1(i)=π i b i (O1);

[0038] β t (i) = P(O t+1 ,O t+2 , ..., O T |S t =S i ,λ);

[0039]

[0040] β T (i) = 1;

[0041] O1,O2,...,O t ,...,O t Represents a sequence of observations in a time series.

[0042] Optionally, determining the health status of the embedded bolt at each moment of the time series according to the parameter value includes:

[0043] The following formula is used to determine the initial time t = 1 with a healthy state Si The probability of the maximum probability path at the beginning is δ1(i):

[0044] δ1(i)=π i b i (O1);

[0045] The following formula is used to determine the state S at time t j The probability of the maximum probability path ending in t (j):

[0046]

[0047] The final state q of the maximum probability path P* of the observation sequence is determined by the following formula: t *:

[0048]

[0049] Starting from the final moment T, use the path pointer to trace back to the initial moment t=1 to obtain the health status sequence of the embedded bolts in the time series;

[0050]

[0051] Path pointer ψ t ( ) Record the health status S of the embedded bolts at time t j The previous health state of the maximum probability path, q t * It is the maximum possible health state of the embedded bolt at time t.

[0052] In another aspect, the present invention provides a health status prediction device, the health status prediction device comprising:

[0053] A training module is used to estimate the parameter values ​​of the embedded bolt state prediction model according to a pre-constructed embedded bolt state prediction model; the parameter values ​​include a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish the transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish the corresponding relationship between different observation values ​​of the embedded bolt and various health states;

[0054] The prediction module is used to determine the health states of the embedded bolts in the time series according to the parameter values.

[0055] Optionally, in the process of estimating the parameter value of the embedded bolt state prediction model, the training module is specifically used to estimate the health state S of the embedded bolt in the time series. i Transfer to healthy state S jThe total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or

[0056] According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

[0057] On the other hand, the present invention provides an electronic device for health status prediction, the electronic device comprising: a memory and a processor, the memory storing a health status prediction program, and when the health status prediction program is executed by the processor, the steps of the health status prediction method as described in any one of the above items are implemented.

[0058] On the other hand, the present invention provides a computer-readable storage medium having a health status prediction program stored thereon, and the health status prediction program can be executed by at least one processor to implement the steps of any of the above health status prediction methods.

[0059] The present invention estimates parameter values ​​of the embedded bolt state prediction model based on a pre-constructed embedded bolt state prediction model; by selecting parameter values, it includes a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish a transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish a corresponding relationship between different observation values ​​of the embedded bolt and various health states; thereby, the various health states of the embedded bolt in the time series can be determined according to the parameter values, breaking through the limitations of traditional static models in the health state assessment of embedded bolts, not only being able to capture the complex conditional dependency relationship between variables, but also being able to continuously update the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing law of the state of the embedded bolt over time, and being particularly suitable for dynamic Bayesian network implementation.

[0060] By constructing multiple layers of restricted Boltzmann machines and stacking them to form a dynamic Bayesian network, we first learn high-level abstract feature representations in the pre-training stage, and then combine the specific physical knowledge and monitoring data of embedded bolts in the fine-tuning stage to optimize the network weights to adapt to the task of predicting the health status of embedded bolts. This method not only makes full use of the time correlation and state evolution trend in historical monitoring data, but also effectively integrates prior knowledge with real-time monitoring information, improving the accuracy and robustness of predictions.

[0061] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0063] Figure 1 A flowchart of a health status prediction method provided by an embodiment of the present invention;

[0064] Figure 2 A block diagram of a health status prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0066] In the subsequent description, the suffixes such as "module", "component" or "unit" used to represent elements are used only to facilitate the description of the present invention and have no specific meanings. Therefore, "module", "component" or "unit" can be used in a mixed manner. Expressions such as "first" and "second" are only used to distinguish technical terms.

[0067] Method Embodiment

[0068] The embodiment of the present invention provides a health status prediction method, such as Figure 1 As shown, the health status prediction method includes:

[0069] S101, estimating parameter values ​​of the embedded bolt state prediction model according to a pre-constructed embedded bolt state prediction model; the parameter values ​​include a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish a transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish a corresponding relationship between different observation values ​​of the embedded bolt and various health states;

[0070] S102: determining each health state of the embedded bolt in the time series according to the parameter value.

[0071] In some optional implementations, the health status can be set to at least three according to actual needs, for example, at least three health statuses are defined, namely S1: healthy; S2: slightly damaged; S3: seriously damaged.

[0072] In some optional implementations, the time series is determined by a selected time step based on actual measurement accuracy. For example, 1, 2, 3, ..., t, ..., T is a time series, and the duration between any two moments in the time series is a time step.

[0073] In some optional implementations, the observed value represents the value of the measured parameter. In the actual detection process, the measured parameter can be determined according to the actual situation. In some embodiments, a dynamic Bayesian network can be used to implement it. The pull-out force value and displacement value are selected through the observed value, emphasizing the in-depth analysis of the pull-out force-displacement curve characteristics. These characteristics can sensitively reflect the stress state and potential damage of the embedded bolts, providing rich input information for the dynamic Bayesian network. By integrating these features in the dynamic Bayesian network, the model can more finely distinguish data patterns under different health states, and can provide reliable early warning signals even in the face of subtle changes in minor damage or early degeneration.

[0074] The embodiment of the present invention estimates the parameter values ​​of the embedded bolt state prediction model based on a pre-constructed embedded bolt state prediction model; by selecting parameter values, it includes a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish the transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish the corresponding relationship between different observation values ​​of the embedded bolt and various health states; thereby, the various health states of the embedded bolt in the time series can be determined according to the parameter values, breaking through the limitations of traditional static models in the assessment of the health state of embedded bolts, not only capturing the complex conditional dependencies between variables, but also continuously updating the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing law of the state of the embedded bolt over time, and is particularly suitable for dynamic Bayesian network implementation.

[0075] By constructing multiple layers of restricted Boltzmann machines and stacking them to form a dynamic Bayesian network, we first learn high-level abstract feature representations in the pre-training stage, and then combine the specific physical knowledge and monitoring data of embedded bolts in the fine-tuning stage to optimize the network weights to adapt to the task of predicting the health status of embedded bolts. This method not only makes full use of the time correlation and state evolution trend in historical monitoring data, but also effectively integrates prior knowledge with real-time monitoring information, improving the accuracy and robustness of predictions.

[0076] In some embodiments, estimating the parameter value of the embedded bolt state prediction model includes:

[0077] According to the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or

[0078] According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

[0079] The a achieved by this embodiment ij and b j (k) Update: When faced with complex and ever-changing operating environments and uncertainties in time series data, it can not only capture the complex conditional dependencies between variables, but also continuously update the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing pattern of the embedded bolt status over time.

[0080] In some embodiments, the health status prediction method further includes:

[0081] Update the initial health status probability of the embedded bolts:

[0082] π i =γ1(i);

[0083] Initial health state probability π i Indicates that the embedded bolt is in a healthy state S at the initial time t = 1 i γ1(i) indicates the probability of observing the observation sequence at the initial time t = 1 and being in the healthy state S i probability;

[0084] The health state transition probability matrix of the embedded bolt is updated using the following formula: ij :

[0085]

[0086] In the formula, a ij The embedded bolt changes from the healthy state S in the time seriesi Transfer to healthy state S j The probability of Indicates the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times; t (i, j) indicates that the embedded bolt is in the healthy state S at time t in the time series i And at time t+1 it is in the healthy state S j The probability of Indicates that the embedded bolts are in a healthy state S i The total number of times; γ t (i) indicates that the embedded bolt is in the healthy state S at time t i The probability of; 1, 2, 3, ..., t, ..., T is a time series, and the time between any two moments in the time series is a time step;

[0087] The observed value probability b of the observed value probability matrix of the embedded bolts is updated using the following formula: j (k):

[0088]

[0089] Where b j (k) indicates that in the health state S j Lower observed value O t is the probability of k; Indicates the health status S of the embedded bolts in the time series j The total number of observations under k; δ(O t =k) ​​has a value of 1 or 0; Indicates that the embedded bolts are in a healthy state S j The total number of times; γ t (j) indicates that the embedded bolt is in the healthy state S at time t j probability.

[0090] The embodiment of the present invention realizes efficient and accurate prediction of the health status of embedded bolts through advanced feature extraction, time series modeling and probabilistic reasoning, provides a scientific basis for maintenance decisions, helps to take measures in advance to prevent major safety accidents and ensure the safety and stability of the structure.

[0091] Optionally, the health status prediction method further includes:

[0092] Construct the initial health state probability π of the embedded bolt as shown in the following formula;

[0093] π i =P(S1=S i );

[0094] Construct the health state transition probability matrix A of the embedded bolts as shown in the following formula:

[0095] A={a ij}, a ij =P(S t+1 =S j |S t =S i );

[0096] Construct the observation value probability matrix B of the embedded bolts as shown in the following formula:

[0097] B = {b jk}, b jk =P(O t =O k |S t =S j );

[0098] According to the initial health state probability π of the embedded bolts, the health state transition probability matrix A of the embedded bolts and the observed value probability matrix B of the embedded bolts, the embedded bolt state prediction model λ = (π, A, B) is constructed; where P ( ) represents probability calculation; O represents observed value;

[0099] According to the embedded bolt state prediction model, determine γ t (i) and ξ t (i,j).

[0100] Among them, γ is determined by the following formula t (i)

[0101] and / or

[0102] Use the following formula to determine ξ t (i, j):

[0103]

[0104] in,

[0105] α t (i) = P(O1, O2, ..., O t ,S t =S i λ);

[0106]

[0107] α1(i)=π i b i (O1);

[0108] β t (i) = P(O t+1 , Ot+2 , ..., O T |S t =S i ,λ);

[0109]

[0110] β T (i) = 1;

[0111] O1, O2, ..., O t , ..., O t Represents a sequence of observations in a time series.

[0112] In some embodiments, determining the health status of the embedded bolt at each moment of the time series according to the parameter value includes:

[0113] The following formula is used to determine the initial time t = 1 with a healthy state S i The probability of the maximum probability path at the beginning is δ1(i):

[0114] δ1(i)=π i b i (O1);

[0115] The following formula is used to determine the state S at time t j The probability of the maximum probability path ending in t (j):

[0116]

[0117] The maximum probability path P of the observation sequence is determined by the following formula: * The final state q t * :

[0118]

[0119] Starting from the final moment T, use the path pointer to trace back to the initial moment t=1 to obtain the health status sequence of the embedded bolts in the time series;

[0120]

[0121] Path pointer ψ t ( ) Record the health status S of the embedded bolts at time t j The previous health state of the maximum probability path, q t * is the maximum possible health state of the embedded bolt at time t.

[0122] The embodiment of the present invention can continuously update the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing pattern of the state of the embedded bolts over time.

[0123] The following provides a specific embodiment based on the above embodiments, which provides a health status prediction method, aiming to break through the limitations of traditional static models in the health status assessment of embedded bolts, especially when faced with complex and changing operating environments and uncertainties in time series data. As a powerful probabilistic graphical model, the dynamic Bayesian network can not only capture the complex conditional dependencies between variables, but also continuously update the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing law of the embedded bolt status over time.

[0124] Compared with the traditional method, this embodiment constructs a multi-layer restricted Boltzmann machine and stacks it to form a dynamic Bayesian network. First, it learns high-level abstract feature representation in the pre-training stage, and then combines the specific physical knowledge and monitoring data of the embedded bolts in the fine-tuning stage to optimize the network weights to adapt to the embedded bolt health status prediction task. This method not only makes full use of the time correlation and state evolution trend in the historical monitoring data, but also effectively integrates prior knowledge and real-time monitoring information, improving the accuracy and robustness of the prediction.

[0125] In particular, this embodiment emphasizes the in-depth analysis of the pull-out force-displacement curve characteristics, which can sensitively reflect the stress state and potential damage of the embedded bolts, providing rich input information for the dynamic Bayesian network. By integrating these features in the dynamic Bayesian network, the model can more finely distinguish data patterns under different health states, and provide reliable early warning signals even in the face of subtle changes in minor damage or early degradation.

[0126] This method of predicting the health status of embedded bolts based on dynamic Bayesian networks achieves efficient and accurate prediction of the health status of embedded bolts through advanced feature extraction, time series modeling and probabilistic reasoning, providing a scientific basis for maintenance decisions, helping to take measures in advance to prevent major safety accidents and ensure the safety and stability of the structure.

[0127] The specific embodiment provides a health status prediction method, which is mainly divided into building a pre-embedded bolt status prediction model, estimating model parameters and predicting the pre-embedded bolts.

[0128] 1. Building the Model

[0129] The expression of the embedded bolt state prediction model can be described as λ = (π, A, B), where λ represents the parameters π, A and B required by the model, π represents the initial health state probability of the embedded bolt, A represents the health state transition probability matrix of the embedded bolt, and B represents the observed value probability matrix of the embedded bolt. The specific parameters of the model are as follows:

[0130] 1) Health status set S of embedded bolts.

[0131] Function: Use discrete states to describe different health conditions of embedded bolts. In specific implementation, multiple health states can be defined according to actual needs, such as the following three health states. Of course, the embodiment of the present invention is not limited to the following three health states, and four, five, etc. health states can also be defined.

[0132] Principle: The status of the embedded bolt system is hidden (not directly observable), but these conditions affect the observable pull-out force and displacement. By defining states such as healthy (not damaged), slightly damaged, and severely damaged, the health of the embedded bolts can be analyzed.

[0133] Definition: S1: healthy, corresponding to the first health state; S2: slightly damaged, corresponding to the second health state; S3: severely damaged, corresponding to the third health state. That is, the health condition of the first health state is better than that of the second health state, and the health condition of the second health state is better than that of the third health state.

[0134] 2) Embedded bolt observation value set O.

[0135] Function: Provide actual measurement data that can be used for analysis, and use these data to infer the health status of embedded bolts.

[0136] Principle: The observation value of embedded bolts (abbreviated as observation value) is the data such as pull-out force and displacement that can be directly measured on embedded bolts. These observation values ​​are affected by the health status of embedded bolts.

[0137] Assume that the observed values ​​include two characteristic parameters: pull-out force and displacement: O1: pull-out force; O2: displacement.

[0138] 3) Initial state probability π of embedded bolts.

[0139] Function: Initialize the model and describe the health status distribution of embedded bolts at the initial moment.

[0140] Indicates the probability that the embedded bolt is in a certain healthy state at the initial time t=1:

[0141] π i =P(S1=S i ),1≤i≤3

[0142] 4) Health state transition probability matrix A of embedded bolts.

[0143] Function: Capture the pattern of changes in the health status of embedded bolts through dynamic changes in time series.

[0144] Principle: Describes the probability of the embedded bolt transferring from one healthy state to another healthy state. This transfer is Markov, that is, it only depends on the current state.

[0145] a ij From a healthy state S i Transfer to another health state S j Probability:

[0146] A={a ij}, a ij =P(S t+1 =S j |S t =S i ), 1≤i, j≤3

[0147] t represents a specific moment in the time series.

[0148] 5) Observation value probability matrix B of embedded bolts.

[0149] Function: Establish the relationship between the observed value and the hidden health state.

[0150] b jk Indicates that in a certain health state S j A certain observation value (pulling force-displacement characteristic value) O is observed k Probability:

[0151] B = {b jk}, b jk =P(O t =O k |S t =S j ), 1≤j≤3, 1≤k≤2

[0152] 2. Estimating Model Parameters

[0153] In order to train the model, the above parameters π, A, and B need to be estimated. A common approach is to use an expectation maximization algorithm, the Baum-Welch algorithm, to iteratively update the model parameters by maximizing the probability of the observed values, so that the model can better fit the given observed data. The algorithm is mainly divided into the following steps: forward algorithm, backward algorithm, expectation step (E step), and maximization step (M step). The forward algorithm and the backward algorithm are very similar, the only difference is the different local states selected. The key to these two algorithms is to define the forward variables and the backward variables, which are specifically stated as follows:

[0154] 1) Forward algorithm

[0155] The forward algorithm calculates the observed values ​​O1, O2, ..., O in the time series 1, 2, ..., t. t And the health status of the embedded bolts is S i The probability of t (i)

[0156] α t (i) = P(O1, O2, ..., O t ,S t =S i |λ)

[0157] Recursion yields:

[0158]

[0159] Set the initial conditions:

[0160] α1(i)=π i b i (O1)

[0161] 2) Backward Algorithm

[0162] The backward algorithm calculates the embedded bolt status S in the time series t+1, t+2, ..., T i And the observed value O t+1 , O t+2 , …, O π The probability of t (i)

[0163] β t (i) = P(O t+1 , O t+2 , ..., O T |S t =S i ,λ)

[0164] Recursion yields:

[0165]

[0166] Set the initial conditions:

[0167] β T (i)=1

[0168] 3) Expected calculation

[0169] Using the forward probability and backward probability, the hidden health state of the embedded bolt at time t can be calculated as S i Probability:

[0170]

[0171] In addition, the hidden health state of the embedded bolts at time t can be calculated as S i , and at time t+1 the health state is S j Probability:

[0172]

[0173] 4) Parameter update

[0174] Initial health status probability update:

[0175] π i =γ1(i)

[0176] π i Indicates that the embedded bolt is in a healthy state S at the initial time (time t = 1) i Use γ1(i) to estimate the probability of being in a healthy state S at the initial moment. i The probability of γ1(i) is that the observation value O1 is observed at time t=1 and the state is S i Updating the initial state probability can reflect the impact of the observed data and make the model more accurately represent the state distribution of the system at the beginning.

[0177] Health state transition probability update:

[0178]

[0179] a ij Indicates that the embedded bolts are from state S i Transfer to healthy state S j The probability of ξ t (i,j) indicates that the health status of the embedded bolt at time t is S i And at time t+1 the health status is S j The probability of . The numerator From the health state S i Transfer to healthy state S j The total number of times, the denominator The health status is S j The total number of times, γ t (i) represents the probability that the embedded bolt is in the healthy state Si at time t. j The proportion of a is used to update ij Updating the state transition probability can reflect the actual frequency of state transitions in the observed data, allowing the model to more accurately represent the transition relationship between states.

[0180] Observation probability update:

[0181]

[0182] b j (k) indicates that in the health state S j Lower observed value O t is the probability of k. t =k) ​​is an indicator function, when O t = k, the value is 1, otherwise it is 0. Is in a healthy state S j The total number of observations under k, the denominator The health status is S j The total number of times, γ t (j) indicates that the embedded bolt is in the healthy state S at time t j By calculating the probability of j Update b by observing k proportion j (k). Updating the observed value probability can reflect the actual distribution of the observed values ​​under each state in the observed data, so that the model can more accurately represent the relationship between the state and the observed value.

[0183] 3. The model predicts the embedded bolts through the following steps:

[0184] 1) Initialization

[0185] The initialization step combines the observation value O1 to calculate the probability that the embedded bolt is in each state at the initial time t=1.

[0186] δ1(i)=π i b i (O1)

[0187] δ1(i) is the state of health S at the initial time t = 1 i The probability of the most probable path to start with.

[0188] 2) Recursion

[0189] The recursive step calculates the number of cells in each health state S at each time step (i.e., the duration between each moment in the time series). j The path probability is updated by maximizing the probability path of the previous step and combining the state transition and the current observation.

[0190]

[0191] δ t (j) is in state S at time t j The probability of the maximum probability path that ends.

[0192] 3) Termination

[0193] The termination step finds the final state of the path with the highest probability that ends at time T.

[0194]

[0195] P * is the probability of the maximum probability path of the entire observation sequence.

[0196] 4) Path backtracking

[0197] The path backtracking step starts from the final time T and uses the path pointer to backtrack to the initial time t=1, thereby obtaining the most likely health state sequence of the entire observation value.

[0198]

[0199] ψ t () is the path pointer, recording the health status S at time t j The previous healthy state of the maximum probability path. t * is the most likely health state at time t, indicating the most likely health state of the bolt we found at each time step.

[0200] The model was trained using the monthly inspection data of 200 embedded bolts with a design value of 200KN for two years after being put into use for five years, and the model's prediction results on the health status of embedded bolts were verified using two months of sample data. The obtained health status recognition is as follows:

[0201]

[0202] From the data in the table, it can be seen that as the model training time increases, the model's prediction accuracy of the health status of embedded bolts gradually increases, and finally stabilizes at around 97%.

[0203] Therefore, the method proposed in this specific embodiment can be effectively applied to the field of health status prediction of embedded bolts, and has high accuracy and credibility.

[0204] Device Embodiment

[0205] The embodiment of the present invention provides a health status prediction device, such as Figure 2 As shown, the health status prediction device includes:

[0206] The training module 20 is used to estimate the parameter values ​​of the embedded bolt state prediction model according to the pre-built embedded bolt state prediction model; the parameter values ​​include the health state transition probability matrix and the observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish the transition relationship between the various health states of the embedded bolt in the preset time series; the observation value probability matrix is ​​used to establish the corresponding relationship between different observation values ​​of the embedded bolt and each health state;

[0207] The prediction module 40 is used to determine the health states of the embedded bolts in the time series according to the parameter values.

[0208] In some optional implementations, the health status can be set to at least three according to actual needs, for example, at least three health statuses are defined, namely S1: healthy; S2: slightly damaged; S3: seriously damaged.

[0209] In some optional implementations, the time series is determined by a selected time step based on actual measurement accuracy. For example, 1, 2, 3, ..., t, ..., T is a time series, and the duration between any two moments in the time series is a time step.

[0210] In some optional implementations, the observed value represents the value of the measured parameter. In the actual detection process, the measured parameter can be determined according to the actual situation. In some embodiments, a dynamic Bayesian network can be used to implement it. The pull-out force value and displacement value are selected through the observed value, emphasizing the in-depth analysis of the pull-out force-displacement curve characteristics. These characteristics can sensitively reflect the stress state and potential damage of the embedded bolts, providing rich input information for the dynamic Bayesian network. By integrating these features in the dynamic Bayesian network, the model can more finely distinguish data patterns under different health states, and can provide reliable early warning signals even in the face of subtle changes in minor damage or early degeneration.

[0211] The embodiment of the present invention estimates the parameter values ​​of the embedded bolt state prediction model based on a pre-constructed embedded bolt state prediction model; by selecting parameter values, it includes a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish the transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish the corresponding relationship between different observation values ​​of the embedded bolt and various health states; thereby, the various health states of the embedded bolt in the time series can be determined according to the parameter values, breaking through the limitations of traditional static models in the assessment of the health state of embedded bolts, not only capturing the complex conditional dependencies between variables, but also continuously updating the conditional probability distribution in the network as the time series data evolves, thereby more accurately reflecting the changing law of the state of the embedded bolt over time, and is particularly suitable for dynamic Bayesian network implementation.

[0212] By constructing multiple layers of restricted Boltzmann machines and stacking them to form a dynamic Bayesian network, we first learn high-level abstract feature representations in the pre-training stage, and then combine the specific physical knowledge and monitoring data of embedded bolts in the fine-tuning stage to optimize the network weights to adapt to the task of predicting the health status of embedded bolts. This method not only makes full use of the time correlation and state evolution trend in historical monitoring data, but also effectively integrates prior knowledge with real-time monitoring information, improving the accuracy and robustness of predictions.

[0213] Optionally, in the process of estimating the parameter value of the embedded bolt state prediction model, the training module is specifically used to estimate the health state S of the embedded bolt in the time series. i Transfer to healthy state S j The total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or

[0214] According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

[0215] Optionally, the observed values ​​include a pull-out force value and a displacement value; the parameter value also includes an initial health state probability of the embedded bolt; and the training module is further used for:

[0216] Update the initial health status probability of the embedded bolts:

[0217] π i =γ1(i);

[0218] Initial health state probability π i Indicates that the embedded bolt is in a healthy state S at the initial time t = 1 i γ1(i) indicates the probability of observing the observation sequence at the initial time t = 1 and being in the healthy state S i probability;

[0219] The health state transition probability matrix of the embedded bolt is updated using the following formula: ij :

[0220]

[0221] In the formula, a ijThe embedded bolt changes from the healthy state S in the time series i Transfer to healthy state S j probability; Indicates the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times; t (i, j) indicates that the embedded bolt is in the healthy state S at time t in the time series i And at time t+1 it is in the healthy state S j probability; Indicates that the embedded bolts are in a healthy state S i The total number of times; γ t (i) indicates that the embedded bolt is in the healthy state S at time t i The probability of; 1,2,3,...,t,...,T is a time series, and the time between any two moments in the time series is a time step;

[0222] The observed value probability b of the observed value probability matrix of the embedded bolts is updated using the following formula: j (k):

[0223]

[0224] Where b j (k) indicates that in the health state S j Lower observed value O t is the probability of k; Indicates the health status S of the embedded bolts in the time series j The total number of observations under k; δ(O t =k) ​​has a value of 1 or 0; Indicates that the embedded bolts are in a healthy state S j The total number of times; γ t (j) indicates that the embedded bolt is in the healthy state S at time t j probability.

[0225] Optionally, the health status prediction device further includes:

[0226] A building module is used to build the initial health state probability π of the embedded bolt as shown in the following formula;

[0227] π i =P(S1=S i );

[0228] Construct the health state transition probability matrix A of the embedded bolts as shown in the following formula:

[0229] A={a ij}, a ij =P(St+1 =S j |S t =S i );

[0230] Construct the observation value probability matrix B of the embedded bolts as shown in the following formula:

[0231] B = {b jk}, b jk =P(O t =O k |S t =S j );

[0232] According to the initial health state probability π of the embedded bolts, the health state transition probability matrix A of the embedded bolts and the observed value probability matrix B of the embedded bolts, the embedded bolt state prediction model λ=(π, A, B) is constructed; where P() represents probability calculation; O represents observed value;

[0233] According to the embedded bolt state prediction model, determine γ t (i) and ξ t (i,j).

[0234] Among them, γ is determined by the following formula t (i)

[0235] and / or

[0236] Use the following formula to determine ξ t (i, j):

[0237]

[0238] in,

[0239] α t (i) = P(O1, O2, ..., O t , S t =S i |λ);

[0240]

[0241] α1(i)=π i b i (O1);

[0242] β t (i) = P(O t+1 ,O t+2 , ..., O T |S t =S i ,λ);

[0243]

[0244] β T (i) = 1;

[0245] O1,O2,...,O t ,...,O t Represents a sequence of observations in a time series.

[0246] Optionally, the prediction module is specifically used to: determine the health state S at the initial time t=1 by using the following formula i The probability of the maximum probability path at the beginning is δ1(i):

[0247] δ1(i)=π i b i (O1);

[0248] The following formula is used to determine the state S at time t j The probability of the maximum probability path ending in t (j):

[0249]

[0250] The maximum probability path P of the observation sequence is determined by the following formula: * The final state q t * :

[0251]

[0252] Starting from the final moment T, use the path pointer to trace back to the initial moment t=1 to obtain the health status sequence of the embedded bolts in the time series;

[0253]

[0254] Path pointer ψ t ( ) Record the health status S of the embedded bolts at time t j The previous health state of the maximum probability path, q t * It is the maximum possible health state of the embedded bolt at time t.

[0255] Device Embodiment

[0256] An embodiment of the present invention provides an electronic device for health status prediction, the electronic device comprising: a memory and a processor, the memory storing a health status prediction program, and when the health status prediction program is executed by the processor, the steps of the health status prediction method described in any one of the method embodiments are implemented.

[0257] Storage Medium Embodiments

[0258] An embodiment of the present invention provides a computer-readable storage medium, on which a health status prediction program is stored. The health status prediction program can be executed by at least one processor to implement the steps of the health status prediction method as described in any one of the method embodiments.

[0259] In the specific implementation process, the device embodiments, equipment embodiments, and storage medium embodiments may refer to the method embodiments and have corresponding technical effects.

[0260] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0261] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0262] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A health status prediction method, characterized in that: The health status prediction method comprises: According to a pre-built embedded bolt state prediction model, the parameter values ​​of the embedded bolt state prediction model are estimated; the parameter values ​​include a health state transition probability matrix and an observation value probability matrix of the embedded bolts; the health state transition probability matrix is ​​used to establish a transition relationship between various health states of the embedded bolts in a preset time series; the observation value probability matrix is ​​used to establish a corresponding relationship between different observation values ​​of the embedded bolts and various health states; According to the parameter value, each health state of the embedded bolt in the time series is determined.

2. The method according to claim 1, characterized in that The estimating parameter values ​​of the embedded bolt state prediction model includes: According to the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

3. The method according to claim 2, characterized in that The observed values ​​include a pull-out force value and a displacement value; the parameter value also includes an initial health state probability of the embedded bolt; the health state prediction method also includes: Update the initial health status probability of the embedded bolts: p i =γ1(i); Initial health state probability π i Indicates that the embedded bolt is in a healthy state S at the initial time t = 1 i γ1(i) indicates the probability of observing the observation sequence at the initial time t = 1 and being in the healthy state S i probability; The health state transition probability matrix of the embedded bolt is updated using the following formula: ij : In the formula, a ij The embedded bolt changes from the healthy state S in the time series i Transfer to healthy state S j probability; Indicates the health status S of the embedded bolts in the time series i Transfer to healthy state S j The total number of times; t (i, j) indicates that the embedded bolt is in the healthy state S at time t in the time series i And at time t+1 it is in the healthy state S j probability; Indicates that the embedded bolts are in a healthy state S i The total number of times; γ t (i) indicates that the embedded bolt is in the healthy state S at time t i The probability of; 1, 2, 3, ..., t, ..., T is a time series, and the time between any two moments in the time series is a time step; The observed value probability b of the observed value probability matrix of the embedded bolts is updated using the following formula: j (k): Where b j (k) indicates that in the health state S j Lower observed value O t is the probability of k; Indicates the health status S of the embedded bolts in the time series j The total number of observations under k; δ(O t =k) ​​has a value of 1 or 0; Indicates that the embedded bolts are in a healthy state S j The total number of times; γ t (j) indicates that the embedded bolt is in the healthy state S at time t j probability.

4. The method according to claim 3, characterized in that The health status prediction method further includes: Construct the initial health state probability π of the embedded bolt as shown in the following formula; π i =P(S1=S i ); Construct the health state transition probability matrix A of the embedded bolts as shown in the following formula: A={a ij },a ij =P(S t+1 =S j |S t =S i ); Construct the observation value probability matrix B of the embedded bolts as shown in the following formula: B={b jk },b jk =P(O t =O k |S t =S j ); According to the initial health state probability π of the embedded bolts, the health state transition probability matrix A of the embedded bolts and the observed value probability matrix B of the embedded bolts, the embedded bolt state prediction model λ=(π,A,B) is constructed; where P( ) represents probability calculation; O represents observed value; According to the embedded bolt state prediction model, determine γ t (i) and ξ t (i,j).

5. The method according to claim 4, characterized in that Determine γ using the following formula t (i) and / or Use the following formula to determine ξ t (i,j): in, α t (i)=P(O1,O2,...,O t ,S t =S i |λ); α1(i)=π i b i (O1); β t (i)=P(O t+1 ,O t+2 ,...,O T |S t =S i ,λ); β T (i)=1; O1, O2, ..., O t , ..., O t Represents a sequence of observations in a time series.

6. The method according to any one of claims 1 to 5, characterized in that: Determining the health status of the embedded bolt at each moment of the time series according to the parameter value includes: The following formula is used to determine the initial time t = 1 with a healthy state S i The probability of the maximum probability path at the beginning is δ1(i): δ1(i)=π i b i (O1); The following formula is used to determine the state S at time t j The probability of the maximum probability path ending in t (j): The maximum probability path P of the observation sequence is determined by the following formula: * The final state q t * : Starting from the final moment T, use the path pointer to trace back to the initial moment t=1 to obtain the health status sequence of the embedded bolts in the time series; Path pointer ψ t ( ) Record the health status S of the embedded bolts at time t j The previous health state of the maximum probability path, q t * It is the maximum possible health state of the embedded bolt at time t.

7. A health status prediction device, characterized in that: The health status prediction device comprises: A training module is used to estimate the parameter values ​​of the embedded bolt state prediction model according to a pre-constructed embedded bolt state prediction model; the parameter values ​​include a health state transition probability matrix and an observation value probability matrix of the embedded bolt; the health state transition probability matrix is ​​used to establish the transition relationship between various health states of the embedded bolt in a preset time series; the observation value probability matrix is ​​used to establish the corresponding relationship between different observation values ​​of the embedded bolt and various health states; The prediction module is used to determine the health states of the embedded bolts in the time series according to the parameter values.

8. The device according to claim 7, characterized in that The training module is used to estimate the parameter value of the embedded bolt state prediction model according to the health state S of the embedded bolt in the time series. i Transfer to healthy state S j The total number of times and embedded bolts are in healthy state S i The ratio of the total number of times to update the health state transition probability matrix of the embedded bolts is ij ; and / or According to the time series, the embedded bolts are in the healthy state S j The total number of observations under k and the embedded bolts are in the healthy state S j The proportion of the total number of times, update the observation probability matrix of the embedded bolt observation probability b j (k); wherein i and j represent the number of pre-set health states, where i≠j.

9. An electronic device for health status prediction, characterized in that: The electronic device comprises: a memory and a processor, wherein the memory stores a health status prediction program, and when the health status prediction program is executed by the processor, the steps of the health status prediction method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a health status prediction program, which can be executed by at least one processor to implement the steps of the health status prediction method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Welding quality state prediction method and device based on statistical model, equipment and medium

    CN119857959A

  • Power transmission line bolt state real-time monitoring method and system based on intelligent monitoring

    CN120890362A

  • A real-time monitoring method and system for bolt state of a power transmission line based on intelligent monitoring

    CN120890362B