A method for bolt group loosening continuous monitoring based on dynamic knowledge base

Abstract

The application discloses a bolt group loosening continuous monitoring method based on a dynamic knowledge base, and the method comprises the following steps: collecting bolt washer elastic force and nut internal thread pressure signals under different working conditions by using an active piezoelectric sensing monitoring device; pre-processing multi-channel force wave signals corresponding to different measurement positions of the bolt; linearly reconstructing the pre-processed force wave signals by using an automatic coding mechanism; capturing the correlation between the bolt washer elastic force and the pressure on the thread by using a joint proximal alternating optimization method, so as to obtain a reconstructed sparse vector of different types of bolts; obtaining a common dynamic knowledge base by using experience knowledge sharing and information complementary optimization, and then accurately judging whether the bolt group connection is stable or not. The bolt washer elastic force and the pressure on the thread can be easily fused into multi-channel force wave signals, and the connection stability of the bolt group composed of different bolts can be continuously diagnosed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of threaded coupling health monitoring, and particularly relates to a bolt group loosening continuous monitoring method based on a dynamic knowledge base. BACKGROUND

[0002] Bolt connection is simple in structure, easy to process, convenient to disassemble and assemble, conducive to maintenance, and widely used, and has become the mainstream connection mode of steel structure. In a complex environment, the bolt connection is prone to looseness due to long-term exposure of the steel structure and influence of alternating load, bolt quality and the like. For a bolt group installed in a large machine, loosening and damage of a small number of bolts can also make the stress of each bolt unable to be completely uniform, and the bolt component can be prematurely fatigued and broken, causing engineering accidents and personnel casualties. If the bolt group connection failure cannot be monitored in time and repaired, not only will the machine lose control and cause significant economic losses, but also personnel casualties can be caused due to the machine tipping over.

[0003] At present, the traditional bolt group loosening monitoring mainly relies on manual detection, which is time-consuming and laborious, cannot realize real-time monitoring, and cannot make an overall judgment on the stress change of the bolt group according to a loosening point. Although the existing innovative intelligent monitoring method can realize wide-range monitoring and rapid identification of structural damage, it is still limited to stress measurement of a single position, the data is too one-sided, and the method does not have the ability to start from the whole and continuously learn. Sometimes, a problem in a single bolt connection will not cause the overall working failure of the bolt group. Therefore, in order to solve the above problems, the present application provides a bolt group loosening continuous monitoring method based on a dynamic knowledge base, which realizes experience knowledge sharing and information complementation of force measurement of two positions of a bolt on the basis of wide-area and real-time monitoring, and can optimize a continuous learning model when the number of bolt layouts changes, so as to realize long-term accurate monitoring of the fastening state of the bolt group. SUMMARY

[0004] Therefore, the present application provides a bolt group loosening continuous monitoring method based on a dynamic knowledge base, which utilizes a double-channel measurement scheme and a dynamic knowledge base learning strategy of information complementation between channels to not only improve the sensing accuracy but also realize continuous learning and real-time effective monitoring of the connection state of the bolt group.

[0005] The present application provides a bolt group loosening continuous monitoring method based on a dynamic knowledge base, which comprises the following steps:

[0006] S1, collecting bolt washer elastic force and pressure on a second nut thread by using an active piezoelectric sensing monitoring device, and acquiring force wave signals of different bolt connection states of a double-channel steel structure bolt group;

[0007] S2, pre-processing the collected signals;

[0008] S3, a continuous learning model based on a dynamic knowledge base is built, through the learning model, common characteristics of the gasket elasticity and the pressure on the thread of the loose bolt group are captured, and potential mutual relationship between the two is explored, so that the bolt loosening condition is accurately judged; when the bolt category in the bolt group changes, the dynamic knowledge base and the bolt sparse vector alternately optimized strategy can effectively learn the bolt loosening condition.

[0009] S4, the continuous learning model based on the dynamic knowledge base is optimized and trained.

[0010] S5, the piezoelectric sensing signal of the current bolt group is acquired, and input into the continuous learning model based on the dynamic knowledge base to judge the bolt group loosening degree.

[0011] On the basis of the above technical scheme, preferably, S1 specifically comprises the following steps:

[0012] S101, a strain gauge is attached to one side of the bolt gasket to measure the elastic force generated by the gasket being pressed between the first nut and the connecting piece;

[0013] S102, the second nut thread hole structure of the bolt is increased in the corresponding standard tooth base circle diameter, so as to leave a part of thread space at the thread connection between the screw rod and the second nut, and the strain sensor is embedded in the second nut thread in a spiral shape to fill the reserved space;

[0014] S103, the strain sensor adopts a piezoelectric strain sensor, and the stress wave signals collected are in a wireless transmission form.

[0015] On the basis of the above technical scheme, preferably, S2 specifically comprises the following steps: intercepting the stress wave signals of the same type of bolt and calculating the first order difference signals for concat operation; the signals are subjected to wavelet denoising and normalization processing; and the redundant stress wave signals are removed.

[0016] On the basis of the above technical scheme, preferably, the continuous learning model based on the dynamic knowledge base in S3 comprises: the continuous learning system faces a series of continuous identification tasks of different categories of bolts Wherein T is the total number of bolt categories. For the tth category of bolt, Wherein, The n t th bolt data sample represented by the mth dm-dimensional stress wave feature, The label corresponding to the tth category of bolt is represented by M. M is the number of channels, and M=2 (i.e. the channel of gasket elasticity and the pressure on the thread). It is a linear mapping for the mth channel of the tth different category of bolt, and the mapping Can be represented as The bolt tightness classifier is represented. When a new tth category bolt is added, the model needs to predict the t-1th different category bolt and the current tth category bolt, so as to achieve uninterrupted and continuous learning of different category bolts.

[0017] Based on the above technical solutions, preferably, the specific optimization steps of S3 are as follows:

[0018] S201, in order to explore the relationship in the channel between different bolts, a matrix decomposition technology is used to construct a shared specific channel dynamic knowledge base for each channel. For the tth category bolt, the mth channel is expressed as Wherein, L m is a specific channel dynamic knowledge base, represents the latent representation of the mth channel of the tth category bolt; based on the automatic encoder mechanism, the specific dynamic knowledge base L m of the mth channel can be optimized in the following way:

[0019]

[0020] Wherein, represents the logical loss function; represents the dm-dimensional feature of the ith data sample of the mth channel; Encoder: Decoder:

[0021] S202, a channel-invariant space based on sparse constraint is established, and the complementary information across channels is fully utilized. The space reconstructs the latent sparse representation for each bolt through two collaborative components, i.e. the latent representation of the tth category bolt is Wherein: S t =P t +Q t . P t aims to explore the shared knowledge between different channels through row sparse constraint, while Q t identifies the contribution of different channels to the connection state of the tth category bolt through column sparse constraint. The latent representation of the tth category bolt can be obtained by optimizing the following objective:

[0022]

[0023] Wherein, λ1>0 and λ2>0 are balance parameters; ||Pt|| 1，∞By capturing the shared atoms between different dynamic knowledge bases, different channel-specific knowledge bases are encouraged to be more consistent in sharing complementary information across channels;||(Q t ) T || 1，∞ The importance of different channels to the tth class bolt connection state is quantified. When the mth column of Q t is greater than zero, the knowledge base L m of the mth channel plays a more important role, and vice versa.

[0024] S203, the machine learning model contains the overall optimization goal of channel and cross-channel correlation. The formal expression is as follows:

[0025]

[0026] S t =P t +Q t , t∈T

[0027] Wherein, is the balance parameter; and respectively represent the tth class bolt washer spring force and the pressure data of the thread; the channel consistency regularizer aims to ensure the semantic consistency between different channel dynamic knowledge bases, adjust the heterogeneous distribution gap between different channels, and encourage samples to share different channel-specific knowledge bases.

[0028] On the basis of the above technical scheme, preferably, the specific optimization steps of S4 are as follows:

[0029] S301, using Taylor expansion to approximate the continuous learning model based on dynamic knowledge base

[0030]

[0031] Wherein, represents the Hessian matrix of around ; and respectively represent the mth column of p t and q t . When the tth class bolt data is input, the mth channel and

[0032]

[0033]

[0034] The t-th category bolt is calculated by the equation and Then, L m and{P t , Q t}are iteratively updated by an alternating optimization strategy.

[0035] S302, fix L m , update{P t , Q t}:

[0036]

[0037]

[0038] where, denotes the first-order gradient of g(U l-1 ); the step size parameter ξ l is appropriately determined by the backtracking rule; λ1||P|| 1，∞ , λ2||Q T || 1，∞ are positive regularization terms. l reaches the set maximum number of iterations, and the optimal solution{P t , Q t}is obtained when the convergence condition is satisfied.

[0039] S303, fix{P t , Q t}, update L m :

[0040] The column vectorization of L m is updated by (R -1 ) m V m (i.e., vec(L m )=(R -1 ) m V m ), where vec(·) represents the column vectorization operation, R m , V m denote the statistical records of the m-th channel. In order to store the previous knowledge of each channel, R m and V m can be updated by the following strategy:

[0041]

[0042]

[0043]

[0044] where, R represents the Croncke product. m and V m The number increases with the number of bolt types.

[0045] Based on the above technical solutions, the preferred method for determining the degree of looseness of the bolt group in S5 (see...) Figure 4 )as follows:

[0046] The relationship between the channels of the dual-channel measurement of each type of bolt is obtained through a continuous learning model, and their importance to the stability of the bolt group is quantified as their respective weights. The measured data of the elastic force of the washers of each type of bolt and the pressure on the thread of the second nut are processed by a single-chip microcomputer using a weighted average algorithm, and then the two are multiplied by their respective coefficients to obtain the final result. The final result is compared with a preset threshold. If it exceeds the threshold, the bolt group is determined to be unstable; otherwise, monitoring continues.

[0047] Advantages

[0048] (1) The present invention provides a method for continuous monitoring of bolt group loosening based on a dynamic knowledge base. It employs a bolt structure with washers, relying on friction to generate resistance torque to prevent fastener loosening. Strain gauges are placed on the washers to measure elastic force. The major diameter of the nut thread hole is designed to be larger than the corresponding standard thread major diameter, reserving space so that the strain sensor can be tightly attached to the inner wall of the thread to measure pressure. Dual measurement is used to more accurately determine the overall connection status of the bolt group.

[0049] (2) The present invention provides a method for continuous monitoring of bolt group loosening based on dynamic knowledge base. In order to explore the relationship between the elastic force of the washer between bolts and the pressure on the thread, specific knowledge bases are constructed for both to capture the shared experience knowledge between different bolts. This process is combined with an automatic encoder mechanism to ensure the consistency of the characteristic space mapping of the force measured at two parts of different bolts, and provides the possibility for continuous learning of the bolt group anti-loosening monitoring system.

[0050] (3) This invention provides a method for continuous monitoring of bolt group loosening based on a dynamic knowledge base. It constructs a channel invariant space based on sparse constraints, acquires shared supplementary knowledge between washer elasticity and thread pressure to explore their relationship, and quantifies their importance to the stability of the bolt group. A channel consistency regularizer is used to mitigate the heterogeneous differences in the distribution of washer elasticity and thread pressure characteristics, minimizing semantic discrepancies between different knowledge bases. This method not only improves the accuracy of judging the connection status of the bolt group but also avoids integrating the features of washer elasticity and thread pressure into a high-dimensional feature space, saving computational costs and storage.

[0051] (4) The bolt group loosening continuous monitoring method based on the dynamic knowledge base is different from the deep learning method, a small amount of bolt measurement samples can be used to train an optimal model to achieve the effect of monitoring the bolt group state, and different types of bolts can be dynamically updated to continuously learn and detect the bolt tightness. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0053] Figure 1 The flow chart of the bolt group loosening continuous monitoring method based on the dynamic knowledge base.

[0054] Figure 2 The assembly drawing of the bolt group loosening continuous monitoring method based on the dynamic knowledge base.

[0055] Figure 3 The continuous learning model flow chart of the bolt group loosening continuous monitoring method based on the dynamic knowledge base.

[0056] Figure 4 The bolt group connection state determination flow chart of the bolt group loosening continuous monitoring method based on the dynamic knowledge base.

[0057] Figure 5 The schematic diagram of the elevator standard joint bolt group of the bolt group loosening continuous monitoring method based on the dynamic knowledge base.

[0058] In the figure: 1 - gasket, 2 - second nut, 3 - strain gauge, 4 - thread, 5 - screw rod, 6 - first nut, 7 - strain sensor, 8 - M24 type bolt structure, 9 - M16 type bolt structure, 10 - standard joint structure. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described below in conjunction with the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0060] In recent years, mechanical equipment overturning caused by bolt loosening has caused frequent incidents of personnel casualties. In particular, during the operation and maintenance of the construction hoist, in order to avoid the disconnection of the guide rail causing the equipment to overturn, the monitoring of the connection state of the standard joint bolt group is crucial. The present application will be described in conjunction with the standard joint bolt group of the construction hoist, which includes 6 M16 type bolts and 4 M24 type bolts (see Figure 5 ). Figure 1 The bolt group loosening continuous monitoring method based on a dynamic knowledge base provided by the present application includes the following processes.

[0061] S1, the active piezoelectric sensing monitoring device is used to collect the elastic force of the bolt washer 1 and the pressure on the second nut 2 thread 4, and the force wave signals of different bolt connection states of the double-channel steel structure bolt group are obtained;

[0062] In this embodiment, the active piezoelectric sensing monitoring device structure specifically includes the following:

[0063] S101, a strain gauge 3 is attached to one side of the washer 1 of the standard joint M24 type and M16 type bolt, so as to measure the elastic force generated by the washer 1 being pressed between the first nut 6 and the standard joint bracket;

[0064] S102, the second nut 2 thread hole structure of the standard joint M24 type and M16 type bolt is increased in its corresponding standard tooth bottom circle diameter, so as to leave a part of the thread space at the thread connection between the screw rod 5 and the second nut 2, and the strain sensor 7 is embedded in the second nut 2 thread 4 in a spiral shape to fill the reserved space;

[0065] S103, the strain sensors all adopt piezoelectric strain sensors, and the stress wave signals collected all adopt wireless transmission form.

[0066] S2, the collected signals are preprocessed;

[0067] In this embodiment, the stress wave signals of the same type of bolt are intercepted and the first order difference signals are calculated for concat operation; the signals are subjected to wavelet denoising and normalization processing; and the redundant stress wave signals are removed.

[0068] S3, a continuous learning model based on a dynamic knowledge base is built, the common characteristics of the elastic force of the 4 M24 type bolt washers 1 and the pressure on the threads 4 of the standard joint when loosening are captured through the learning model, and the potential mutual relationship between the elastic force of the washer 1 and the pressure on the thread 4 is explored, and the same is true for the 6 M16 type bolts, so as to accurately judge the bolt loosening condition; when the type of the bolt in the bolt group changes, the shared dynamic knowledge base and the alternative optimization strategy of the sparse vector of the bolt can effectively learn the bolt loosening condition.

[0069] In this embodiment, the continuous machine learning model comprises: dynamic knowledge bases of the pressure on the bolt thread 4 and the elasticity of the washer 1, wherein the dynamic knowledge base A contains shared information of the elasticity of the washer 1 of the M16 and M24 bolts, and the dynamic knowledge base B contains shared information of the pressure on the thread 4 of the M16 and M24 bolts; an autoencoder mechanism to reduce the intra-channel mapping divergence between the semantic and feature spaces; a channel-invariant space based on sparse constraints to capture the correlation between the elasticity of the washer 1 and the pressure on the thread 4 when the bolt is connected; and a channel consistency regularizer to ensure semantic consistency across channels.

[0070] S4, optimizing and training the continuous learning model based on the dynamic knowledge base;

[0071] In this embodiment, the method for training and verifying the continuous machine learning model is as follows: determining performance evaluation indicators to prove the superiority of the model; performing extensive comparative experiments on the model and some representative models, and performing empirical comparison using three benchmark data sets; randomly selecting half of each evaluation data set for training and the other half for evaluation; and all results are the average of 5 random runs.

[0072] S5, acquiring piezoelectric sensing signals of the current bolt group and inputting them into the continuous learning model based on the dynamic knowledge base to determine the looseness of the bolt group (see Figure 4 ).

[0073] In this embodiment, the intra-channel and inter-channel relationships of the double-channel measurement of the M24 and M16 bolts are obtained through the continuous learning model, and the importance of the relationships to the stability of the bolt group is quantified as respective weights; the measured data of the elasticity of the washer 1 and the pressure on the thread 4 of the second nut 2 of the M24 and M16 bolts are processed through a single-chip microcomputer using a weighted average algorithm, and then multiplied by respective coefficients to obtain a final result; the final result is compared with a pre-set threshold value, and if the final result exceeds the threshold value, it is determined that the bolt group is unstable, otherwise the monitoring continues.

[0074] On this basis, the embodiment provides a specific calculation process for the continuous machine learning model to realize intra-channel knowledge sharing and cross-channel information complementation, and the flow is shown in detail in Figure 2 .

[0075] S201, in order to explore the intra-channel relationship between the standard section M24 and M16 bolts, a shared specific channel dynamic knowledge base is constructed for each channel using matrix decomposition technology. For the t(th) (t≤2) type bolt (T=1 is set as the M16 bolt and T=2 is set as the M24 bolt), the m(th) (m≤2) channel is expressed as (M=1 is set as the elasticity of the washer 1 and M=2 is set as the pressure on the thread 4). Wherein, Lm is a specific channel dynamic knowledge base, Represents the latent characterization of the m-th channel of the t-th type of bolt; based on the autoencoder mechanism, a specific dynamic knowledge base L for the m-th type of channel. m The following methods can be used for optimization:

[0076]

[0077] in, Represents the logistic loss function; Represents the dm-dimensional feature of the i-th data sample in the m-th channel; Encoder: Decoder:

[0078] S202. Establish a channel-invariant space based on sparse constraints to fully utilize complementary information across channels. This space reconstructs the latent sparse representation for each bolt using two cooperative components, i.e., the latent representation of the t-th bolt type is... Wherein: S t =P t +Q t P t The aim is to explore shared knowledge between different channels through row sparsity constraints, while Q t The contribution of different channels to the connection state of the t-th type of bolt is identified by column sparsity constraints. The latent representation of the t-th type of bolt can be obtained by optimizing the following objective:

[0079]

[0080] Where λ1>0 and λ2>0 are equilibrium parameters; ||Pt||1 ，∞通 By capturing shared atoms between different dynamic knowledge bases, it encourages greater consistency among channel-specific knowledge bases regarding cross-channel shared complementary information; ||(Q t ) T || 1，∞ The importance of different channels to the connection state of the t-th type of bolt is quantified. When Q t When the m-th column is greater than zero, the knowledge base L of the m-th channel m It plays a more important role, and vice versa.

[0081] S203. The overall optimization objective of the continuous machine learning model, which includes intra-channel and cross-channel correlations, is formally stated as follows:

[0082]

[0083] S t =P t +Q t ,t∈T

[0084] in, is a balance parameter; and respectively represent the elastic force of the tth type of bolt washer 1 and the pressure data of the screw thread 4; channel consistency regularizer It aims to ensure the semantic consistency between the dynamic knowledge bases of different channels, adjust the heterogeneous distribution gap between different channels, and encourage sample sharing of specific knowledge bases of different channels.

[0085] On this basis, in order to optimize the overall objective formula in S203, when the standard bolt group increases bolts of different types, the knowledge base can dynamically update the prediction of the two types of bolts (M16 type and M24 type) previously learned and the third type of bolt, achieving continuous learning and detecting the effect of bolt fastening. The embodiment provides a specific optimization calculation process.

[0086] S301, using Taylor expansion to approximate the continuous learning model based on dynamic knowledge base

[0087]

[0088] wherein, represents the Hessian matrix of around ; and respectively represent the mth column of p t and q t . When the third type of bolt data is input, the mth channel can be easily calculated by the following equation and

[0089]

[0090]

[0091] The mth channel of the third type of bolt can be calculated by the following equation and After that, the L m and {P t , Q t} in the equation are iteratively updated by the alternating optimization strategy.

[0092] S302, fix L m , update {P t , Q t}:

[0093]

[0094]

[0095] wherein, Represents g(U) l-1 The first-order gradient of ξ; step size parameter ξ l Determined appropriately by the backtracking rules; λ1||P|| 1，∞ ,λ2||Q T || 1，∞ This is a regularization term. When the maximum set number of iterations is reached and the convergence condition is met, the optimal solution {P} is obtained. t Q t}

[0096] S303, Fixed {P t Q t}, update L m :

[0097] Through (R) m ) -1 V m (i.e., vec(L) m )=(R m ) -1 V m To update L m The column vectorization, where vec(·) represents the column vectorization operation, R m V m This represents the statistical record of the m-th channel. To store the previous knowledge for each channel, R... m and V m The following strategies can be used to update:

[0098]

[0099]

[0100]

[0101] in, R represents the Croncke product. m and V m The number increases with the size of the bolt.

[0102] The specific embodiments of the present invention have been described in detail above, but they are merely examples, and the present invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent changes and modifications made without departing from the spirit and scope of the present invention are covered within the scope of the present invention.

Claims

1. A method for bolt cluster loosening continuous monitoring based on dynamic knowledge base, characterized in that: It comprises the following steps: S1, collecting the elastic force of the bolt washer (1) and the pressure on the thread (4) of the second nut (2) by using an active piezoelectric sensing monitoring device, and obtaining the force wave signals of different bolt connection states of the double-channel steel structure bolt group; S2, preprocessing the collected signals; S3, building a continuous learning model based on a dynamic knowledge base, capturing the common characteristics of the elastic force of the bolt washer (1) and the pressure on the thread (4) of the bolt group when loosening, and exploring the potential mutual relationship between the two to accurately determine the bolt loosening condition; when the bolt category in the bolt group changes, the dynamic knowledge base and the bolt sparse vector alternately optimized strategy can effectively learn the bolt loosening condition; S4, optimizing and training the continuous learning model based on the dynamic knowledge base; S5, obtaining the piezoelectric sensing signals of the current bolt group and inputting them into the continuous learning model based on the dynamic knowledge base to determine the bolt group loosening degree; The continuous machine learning model in S3 comprises: Suppose the continuous learning system faces a series of continuous tasks of identifying different categories of bolt states wherein is the total number of bolt categories, for the th category of bolt, there is ; wherein, represents the th bolt data sample represented by the th dimensional stress wave feature, represents the label corresponding to the th category of bolt, is the number of channels, and i.e. the channel of the pressure received by the washer (1) and the thread (4), is a linear mapping for the th channel of the th different category of bolt, mapping can be represented as , represents the corresponding bolt tightness classifier, when a th category of bolt is added, the model needs to predict the th different category of bolt learned before and the current th category of bolt, achieving uninterrupted and continuous learning of different categories of bolt tightness. S201. To explore the intra-channel relationships between different bolts, matrix decomposition technology is used to construct a shared dynamic knowledge base for each channel; for the first... The first category of bolts, the... Each channel Described as , ;in, For a specific channel dynamic knowledge base, Indicates the first The first category of bolts The potential representation of the channel; based on the autoencoder mechanism, for the ... Specific dynamic knowledge base of each channel The following methods can be used for optimization: wherein, representing a logical loss function; denotes the channel's th data sample's dimensional feature; encoder: ; Decoder: ; S202. Establish a channel-invariant space based on sparse constraints to fully utilize complementary information across channels; this space reconstructs the potential sparse representation for each bolt through two cooperative components, namely the first... The potential representation of each category of bolts is as follows: ( ),in: The aim is to explore shared knowledge between different channels through row sparse constraints, and Identifying different channels for the first [channel] using column sparse constraints. The contribution of each category of bolted connection condition; The potential representation of each category of bolts can be obtained by optimizing the following objectives: in, and It is a balance parameter; By capturing shared atoms between different dynamic knowledge bases, it encourages different channel-specific knowledge bases to be more consistent in their sharing of complementary information across channels; Quantified the effect of different channels on the first The importance of each category of bolt connection status; when The When the column is greater than zero, the first Channel Knowledge Base It plays a more important role, and vice versa; S203, the continuous machine learning model contains the overall optimization target of channel and cross-channel correlation, which is formally expressed as follows: in, It is a balance parameter; and They represent the first Data on the elasticity and thread pressure of each category of bolt washer (1) and thread (4); channel consistency regularizer The aim is to ensure semantic consistency between dynamic knowledge bases of different channels, while adjusting for heterogeneous distribution gaps between different channels and encouraging samples to share specific knowledge bases of different channels.

2. The dynamic knowledge base based continuous monitoring method of bolt cluster loosening according to claim 1, characterized in that: The active piezoelectric sensing monitoring device in S1 specifically comprises the following: S101, a strain gauge (3) is attached to one side of the bolt washer (1) to measure the elastic force of the washer (1) when it is squeezed between the first nut (6) and the connecting piece; S102, the thread hole structure of the second nut (2) of the bolt is increased in diameter to leave a part of the thread space at the thread connection between the screw rod (5) and the second nut (2), and the strain sensor (7) is embedded in the thread (4) of the second nut (2) in a spiral shape to fill the reserved space; S103, the strain sensor adopts a piezoelectric strain sensor, and the stress wave signals collected are transmitted in a wireless manner. The specific implementation steps of S2 are as follows:

3. The dynamic knowledge base based continuous monitoring method of bolt cluster loosening as claimed in claim 1 wherein: The stress wave signals of the same type of bolt are intercepted and the first order difference signals are calculated for concat operation; the signals are wavelet denoised and normalized; and the redundant stress wave signals are removed. The specific optimization steps for S4 are as follows:

4. The dynamic knowledge base based continuous monitoring method of bolt cluster loosening as claimed in claim 1 wherein: S301, approximate the continuous learning model based on the dynamic knowledge base by using Taylor expansion: The method for determining the bolt group loosening degree in S5 is as follows: wherein, denotes the Hessian matrix of and represent the first and columns of When the first channel's and are easily calculated by the following equations:​​​ The first and the second and the third After that, the first and the second are iteratively updated by an alternating optimization strategy. S302, fix , update : wherein ; denotes a first order gradient; a step size parameter is suitably determined by a backtracking rule; , is a regularization term; a set maximum number of iterations is reached, when a convergence condition is fulfilled, an optimal solution is obtained ; S303, fix , update : by updating the column vectorization of ) where represents the column vectorization operation, , , denotes the statistical record of the th channel; to store the previous knowledge of each channel, and can be updated by the following strategies: wherein, represents the Kronig product; and increases with increasing bolt class.

5. The dynamic knowledge base based continuous monitoring method of bolt cluster loosening as claimed in claim 1 wherein: Through the continuous learning model, the relationship between the channels and the channels of the double-channel measurement of each category of bolt is obtained, and the importance of the relationship to the stability of the bolt group is quantified as the respective weight; the measured data of the elastic force of the bolt washer (1) and the pressure on the thread (4) of the second nut (2) are processed by a single-chip microcomputer through a weighted average algorithm, and then multiplied by the respective coefficients to obtain the final result; the final result is compared with the pre-set threshold value, if it exceeds the threshold value, it is determined that the bolt group is not stable, otherwise it continues to be monitored. ​

Images