CN112668526A - Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing - Google Patents
Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing Download PDFInfo
- Publication number
- CN112668526A CN112668526A CN202011641905.9A CN202011641905A CN112668526A CN 112668526 A CN112668526 A CN 112668526A CN 202011641905 A CN202011641905 A CN 202011641905A CN 112668526 A CN112668526 A CN 112668526A
- Authority
- CN
- China
- Prior art keywords
- loosening
- bolt
- neural network
- bolt group
- piezoelectric
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention provides a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing, and provides a method for acquiring piezoelectric original signals of different bolt loosening conditions of a steel structure bolt group by using a sensor array based on the characteristics of a piezoelectric sensor, and constructing a one-dimensional bolt group loosening index vector based on the energy value of three-layer wavelet decomposition to be used as the representation of the different bolt loosening conditions of the steel structure bolt group. Because the convolutional neural network has the characteristics of capability of retaining and extracting features in data, self-learning and the like, the method establishes a mathematical model of different bolt loosening positions of the loosening index and the bolt group based on the Lenet5-CBAM deep learning model, and the model can directly judge the loosening position of the steel structure bolt group according to the loosening index obtained by actually measuring the piezoelectric signal.
Description
Technical Field
The invention relates to the field of bolted steel structure health monitoring, in particular to a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing.
Background
The bolt connection has the advantages of easy construction, detachability, fatigue resistance and the like, and is a main connection mode of a steel structure at present. However, due to fatigue, corrosion and other factors, the stress state of the bolt connection part of the steel structure is very easy to change, and corresponding engineering practices also prove that the stress state of surrounding bolts can be changed when a single bolt in the bolt group is damaged, and if the bolt loose part is not found in time and replaced or repaired, the structural performance can be greatly influenced.
The method not only influences the normal service of the whole structure to a certain extent, but also even causes catastrophic accidents such as structural damage and the like, and causes great economic loss and casualties, thereby having important engineering significance for real-time quantitative monitoring of the bolt connection state.
The traditional piezoelectric active sensing method cannot realize the loosening positioning of the steel structure bolt group, so that in order to solve the problems, the invention provides a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing, the deep learning and piezoelectric active sensing method is applied to the monitoring of the loosening of the steel structure bolt group bolts, and the online monitoring of the loosening position of the steel structure bolt group bolts can be realized.
Disclosure of Invention
In view of the above, the invention provides a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing, and the method can be used for quantitatively monitoring the tightness degree of steel structure bolt connection in real time by acquiring the change of structural nonlinear response signals based on the piezoelectric active sensing method and combining the capability of deep learning to process big data.
The technical scheme of the invention is realized as follows: the invention provides a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing, which comprises the following steps of:
s1, building a piezoelectric active sensing system, monitoring the bolt loosening degree of the bolt group in a piezoelectric active sensing mode, and acquiring piezoelectric signals of different bolt loosening of the steel structure bolt group;
s2, calculating a wavelet decomposition energy value of the piezoelectric signal based on a wavelet packet decomposition method, and forming a one-dimensional loosening index vector by the wavelet energy value calculated by the sensor array;
s3, building a one-dimensional convolutional neural network, and building a relation between an elasticity index vector of the bolt group and a bolt loosening position of the bolt group through the one-dimensional convolutional neural network;
s4, training and verifying a one-dimensional convolutional neural network;
and S5, acquiring a piezoelectric sensing signal of the current bolt, inputting the piezoelectric sensing signal into the one-dimensional convolutional neural network, and acquiring the specific position of the bolt group loosening.
On the basis of the above technical solution, preferably, S2 specifically includes the following steps:
s101, calculating the decomposition of three layers of wavelet packets into 8 sub-signals based on the acquired piezoelectric sensing signals;
s102, calculating wavelet packet decomposition signal energy components of the piezoelectric sensing signals and wavelet packet decomposition sub-signal energy corresponding to each frequency band;
s103, calculating a one-dimensional loosening index vector of the piezoelectric sensor array signal according to the wavelet packet decomposition sub-signal energy;
s104, standardizing the one-dimensional loosening index vector data according to a standardized function; the normalization function is:
wherein E represents a one-dimensional loosening index vector; mu.sEAn average value representing the energy of the wavelet packet decomposition sub-signal; sigmaERepresenting the standard deviation of the wavelet packet decomposition sub-signal energy.
Based on the above technical solution, preferably, the one-dimensional convolutional neural network in S3 includes: the system comprises an input layer, a convolutional neural network Lenet5, a channel attention module, a spatial attention module and an output layer which are connected in sequence.
Based on the above technical solution, preferably, the convolutional neural network Lenet5 includes a convolutional layer 1, a ReLU layer, a maximum pooling layer, a convolutional layer 2, a ReLU layer, and a pooling layer, which are connected in sequence.
On the basis of the above technical solution, preferably, the calculation process of the channel attention module includes the following steps:
s201, a channel attention mechanism extracts high-level features from features extracted by two convolutional layers and a pooling layer of a convolutional neural network Lenet5 through maximum pooling and average pooling respectively;
s202, after the output elements are subjected to addition operation through a shared multilayer sensing machine, the final channel attention weight is generated through sigmoid function activation operation;
and S203, performing point multiplication operation on the channel attention weight and the original input feature to generate the input feature required by the spatial attention mechanism.
On the basis of the above technical solution, preferably, the channel attention module is expressed as:
On the basis of the above technical solution, preferably, the calculation process of the spatial attention module includes the following steps:
s301, taking the feature value output by the channel attention module as a feature map input by the space attention module, and performing global maximum pooling and average maximum pooling on the basis of the channel;
s302, concat operation is carried out on the two channels of the space attention module and the channel attention module, dimension reduction is carried out through convolution operation to obtain a feature map of only one channel, space attention weight is generated through a sigmoid function, and finally point multiplication operation is carried out on the weight and the feature map output by the channel attention module to generate final features.
On the basis of the above technical solution, preferably, the spatial attention module is expressed as:
wherein M issRepresenting spatial attention weight, f1*7Represents 1 x 7 of the convolution layer,
representing the feature values of the spatial attention module after global average pooling;
representing the feature values of the spatial attention module after global max pooling.
On the basis of the technical scheme, preferably, the neurons of the output layer and the output layer of the spatial attention module are in full connection, each neuron corresponds to the working condition of bolt looseness of the steel structure bolt group, and a softmax function is adopted as an activation function of the output layer;
the activation function is:
in the formula, n and k each represent the number of each neuron, anRepresents the output of the kth neuron of the output layer, K represents the number of output nodes, Softmax (a)n) And (3) representing the output obtained after the k-th neuron is activated by the function, and finally outputting a vector of 1 x 9 by the Lenet5-CBAM depth network, wherein each value can be regarded as the confidence probability of the bolt looseness corresponding to the steel structure bolt group, and the tightness state corresponding to the maximum probability is taken as the result of fault diagnosis.
On the basis of the above technical solution, preferably, S4 specifically includes the following steps:
s401, repeatedly acquiring piezoelectric sensing signals under different tightening conditions of the bolt, obtaining a large amount of data of a training and verification model, and calculating according to the step S2 to obtain standardized multichannel wavelet packet component energy;
s402, dividing the component energy of the standardized multi-channel wavelet packet into a training sample and a test sample according to a ratio of 8:2, inputting the training sample as a one-dimensional convolution neural network, and outputting a corresponding bolt loosening position label as an expectation output of a double-attention convolution neural network model;
during first training, the connection weight of each neuron of the one-dimensional convolutional neural network is set in a random initialization mode, the connection weight of the one-dimensional convolutional neural network is updated by adopting an Adam gradient descent algorithm, and finally the trained connection weight is stored to obtain a steel structure bolt group bolt loosening positioning model;
s403, inputting the test sample into the one-dimensional convolutional neural network, testing the generalization performance of the model, and if the one-dimensional convolutional neural network meets the expected requirement, storing the one-dimensional convolutional neural network; otherwise, the one-dimensional convolutional neural network is adjusted, and the weight of the one-dimensional convolutional neural network is updated.
Compared with the prior art, the bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing has the following beneficial effects:
(1) the method based on piezoelectric active sensing is sensitive to monitoring of the loosening of the steel structure bolt, and the loosening state of the bolt can be represented to a certain extent based on the energy decomposed by the wavelet packet;
(2) based on the characteristics of the piezoelectric sensor, a sensor array is provided to obtain piezoelectric original signals of different bolt loosening conditions of a steel structure bolt group, and a one-dimensional bolt group loosening index vector is constructed based on energy values of three-layer wavelet decomposition and is used as the representation of different bolt loosening conditions of the steel structure bolt group. Because the convolutional neural network has the characteristics of capability of retaining and extracting features in data, self-learning and the like, a mathematical model of different bolt loosening positions of a loosening index and a bolt group is established based on a Lenet5-CBAM deep learning model, and the model can directly judge the loosening position of the steel structure bolt group according to the loosening index obtained by actually measuring a piezoelectric signal;
(3) the method applies deep learning and piezoelectric active sensing methods to quantitative monitoring and identification of the tightness degree of the steel structure bolt connection, has clear principle, low cost, simple and easy operation and accurate identification, and can realize long-term online monitoring and identification of the tightness degree of the steel structure bolt, thereby providing scientific basis for safety evaluation and service life prediction of steel bolt connection nodes and reducing potential safety hazard and operation risk caused by bolt connection looseness;
(4) a channel space double attention mechanism is introduced into a convolutional neural network, a portable high-precision identification network is constructed, and the accuracy of bolt tightness identification is higher; the recognition effect of the convolutional neural network is improved by adding a spatial channel attention mechanism, and the robustness of the model is improved while the recognition capability of the model is improved.
(5) A monitoring scheme for monitoring the tightness degree of the bolt through multiple channels is provided, according to the phenomenon that gravitational wave energy is lost when the bolt looses, wavelet packet component energy is used as a damage index, and the strong characteristic extraction and memory capability of a convolutional neural network are combined to realize quantitative monitoring of the tightness degree of the bolt.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing according to the present invention;
FIG. 2 is a flow chart of Lenet5-CBAM model training in the present invention;
FIG. 3 is a diagram of a one-dimensional convolutional neural network structure in accordance with the present invention;
FIG. 4 is a schematic view of a channel space attention mechanism module according to the present invention;
FIG. 5 is a block diagram of a channel attention module and a spatial attention module according to the present invention;
FIG. 6 is a graph of accuracy curves and loss functions for the model training process and the validation process of the present invention;
FIG. 7 is a diagram showing the effect of the network model in identifying the torque of the steel structure bolt and a confusion matrix diagram.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
In recent years, methods for researching bolt tightness monitoring mainly include: a detection method based on structural vibration characteristics, a detection method based on piezoelectric impedance, and a detection method based on a percussion signal. However, the existing methods can only detect whether the bolt is loose, and the loose position cannot be determined, so to solve the above problem, as shown in fig. 1, the present embodiment provides a bolt group loosening positioning and monitoring method based on deep learning and piezoelectric active sensing, which includes the following steps:
s1, building a piezoelectric active sensing system, monitoring the bolt loosening degree of the bolt group in a piezoelectric active sensing mode, and acquiring piezoelectric signals of different bolt loosening of the steel structure bolt group;
the piezoelectric sensor adopted in the embodiment can be piezoelectric ceramic, and the piezoelectric ceramic has the characteristics of quick response, wide frequency range, easiness in cutting, low price and the like, and is widely applied to the technical field of engineering structure monitoring. Preferably, the piezoelectric sensor comprises three piezoelectric ceramics, which are respectively installed at the left, upper and right positions of the bolt connection part of the lower steel plate, wherein the installation position is not limited.
S2, calculating a wavelet decomposition energy value of the piezoelectric signal based on a wavelet packet decomposition method, and forming a one-dimensional loosening index vector by the wavelet energy value calculated by the sensor array;
the method specifically comprises the following steps:
s101, calculating the decomposition of three layers of wavelet packets into 8 sub-signals based on the acquired piezoelectric sensing signals;
marking the collected piezoelectric sensing signal as XjJ represents the jth piezoelectric sensing signal;
the sub-signal components may be represented as: xj＝[Xj,1,Xj,2…Xj,m](j-1, 2, …,8), where m represents the number of signal points, Xj,mRepresenting the mth sub-signal component of the jth piezoelectric sensing signal.
S102, calculating wavelet packet decomposition signal energy components of the piezoelectric sensing signals and wavelet packet decomposition sub-signal energy corresponding to each frequency band;
according to the definition of signal energy, the wavelet packet decomposition sub-signal energy can be expressed as: ej＝Xj,1 2+Xj,2 2+…+Xj,m 2(j＝1,2,…,8)。
S103, calculating a one-dimensional loosening index vector of the piezoelectric sensor array signal according to the wavelet packet decomposition sub-signal energy;
in this embodiment, because three sensors are used, the wavelet base is db2, and the index used by the one-dimensional convolutional neural network is a combined value of wavelet energy components of signals of the three sensors, that is, a value including 24 wavelet energy values, so as to comprehensively represent the tightness degree of the bolt more comprehensively. Therefore, the final one-dimensional loosening index vector can be expressed as: e ═ E1,…E8,E9,…E16,E17,…E24]。
And S104, standardizing the one-dimensional loosening index vector data according to a standardization function. The normalization function is:
wherein E represents a one-dimensional loosening index vector; mu.sEAn average value representing the energy of the wavelet packet decomposition sub-signal; sigmaERepresenting the standard deviation of the wavelet packet decomposition sub-signal energy.
S3, building a one-dimensional convolutional neural network, and building a relation between an elasticity index vector of the bolt group and a bolt loosening position of the bolt group through the one-dimensional convolutional neural network;
in this embodiment, as shown in fig. 3, the one-dimensional convolutional neural network includes an input layer, a convolutional neural network Lenet5, a channel attention module, a spatial attention module, and an output layer, which are connected in sequence; each constituent component is described in detail below:
in this embodiment, the convolutional neural network Lenet5 includes a convolutional layer 1, a ReLU layer, a maximum pooling layer, a convolutional layer 2, a ReLU layer, and a pooling layer, where the convolutional kernel size of the convolutional layer 1 is 1 × 3, the convolutional kernel size of the convolutional layer 2 is 1 × 5, the moving step length is 1, and the pooling layer adopts maximum pooling. The convolutional neural network Lenet5 is simple in structure; the number of parameters is small, the training speed is high, and the like, and a plurality of data sets exceed the traditional benchmark network model.
And the channel attention module and the space attention module are used for obtaining the characteristic that the bolt group is loosened.
Output layer, its neuron number K is 9, with the output layer of space attention module with the form of full connection, every neuron corresponds the not hard up operating mode of corresponding steel construction bolt crowd bolt to adopt the softmax function as the activation function of this output layer, this activation function is:
in the formula, n and k each represent the number of each neuron, anRepresents the output of the kth neuron of the output layer, K represents the number of output nodes, Softmax (a)n) And (3) representing the output obtained after the k-th neuron is activated by the function, and finally outputting a vector of 1 x 9 by the Lenet5-CBAM depth network, wherein each value can be regarded as the confidence probability of the bolt looseness corresponding to the steel structure bolt group, and the tightness state corresponding to the maximum probability is taken as the result of fault diagnosis.
S4, training and verifying a one-dimensional convolutional neural network; as shown in fig. 2, the method specifically includes the following steps:
s401, repeatedly acquiring piezoelectric sensing signals under different tightening conditions of the bolt, obtaining a large amount of data of a training and verification model, and calculating according to the step S2 to obtain standardized multichannel wavelet packet component energy;
in this embodiment, each sampling is 500 samples, a wavelet decomposition energy vector of an original signal is obtained based on wavelet packet decomposition, a one-dimensional loosening index vector is constructed based on a piezoelectric sensor array, a standard formula is adopted to convert an index into a numerical value with a mean value of 0 and a variance of 1, and the iterative training speed of a neural network is accelerated.
S402, dividing the component energy of the standardized multi-channel wavelet packet into a training sample and a test sample according to a ratio of 8:2, inputting the training sample as a one-dimensional convolution neural network, and outputting a corresponding bolt loosening position label as an expectation output of a double-attention convolution neural network model;
during the first training, the connection weight of each neuron of the one-dimensional convolutional neural network is set in a random initialization mode, the connection weight of the one-dimensional convolutional neural network is updated by adopting an Adam gradient descent algorithm, the initial learning rate is set to be 0.0001, the number of training sample cycles is 400, and the model is trained according to the steps of S1-S3 to obtain the model for positioning the loosening of the steel structure bolt group bolts.
And (4) inputting the one-dimensional loosening index vector obtained in the step (S401) into a deep learning model, wherein the model can output corresponding 1 x 9 output vectors and prediction confidence probabilities of 9 loosening positions corresponding to the steel structure bolt group, and the bolt loosening position corresponding to the maximum value is the actual working condition of the loosening position of the bolt group. Therefore, the loosening position of the steel structure bolt group can be determined, and a scientific basis is provided for replacement or repair of engineering management personnel.
S403, inputting the test sample into the one-dimensional convolutional neural network, testing the generalization performance of the model, and if the one-dimensional convolutional neural network meets the expected requirement, storing the one-dimensional convolutional neural network; otherwise, the one-dimensional convolutional neural network is adjusted, and the weight of the one-dimensional convolutional neural network is updated.
And S5, acquiring a piezoelectric sensing signal of the current bolt, inputting the piezoelectric sensing signal into the one-dimensional convolutional neural network, and acquiring the specific position of the bolt group loosening.
The beneficial effect of this embodiment does: the method based on piezoelectric active sensing is sensitive to monitoring of loosening of the steel structure bolt, and the loosening state of the bolt can be represented to a certain extent based on energy decomposed by a wavelet packet.
The invention provides a method for acquiring the piezoelectric original signals of different bolt loosening conditions of a steel structure bolt group by using a sensor array based on the characteristics of a piezoelectric sensor, and constructs a one-dimensional bolt group loosening index vector based on the energy value of three-layer wavelet decomposition as the representation of the different bolt loosening conditions of the steel structure bolt group. Because the convolutional neural network has the characteristics of capability of retaining and extracting features in data, self-learning and the like, a mathematical model of different bolt loosening positions of a loosening index and a bolt group is established based on a Lenet5-CBAM deep learning model, and the model can directly judge the loosening position of the steel structure bolt group according to the loosening index obtained by actually measuring a piezoelectric signal;
the method applies deep learning and piezoelectric active sensing methods to quantitative monitoring and identification of the tightness degree of the steel structure bolt connection, has clear principle, low cost, simple and easy operation and accurate identification, and can realize long-term online monitoring and identification of the tightness degree of the steel structure bolt, thereby providing scientific basis for safety evaluation and service life prediction of steel bolt connection nodes and reducing potential safety hazard and operation risk caused by bolt connection looseness;
a channel space double attention mechanism is introduced into a convolutional neural network, a portable high-precision identification network is constructed, and the accuracy of bolt tightness identification is higher;
a monitoring scheme for monitoring the tightness degree of the bolt through multiple channels is provided, according to the phenomenon that gravitational wave energy is lost when the bolt looses, wavelet packet component energy is used as a damage index, and the strong characteristic extraction and memory capability of a convolutional neural network are combined to realize quantitative monitoring of the tightness degree of the bolt.
Example 2
On the basis of embodiment 1, the present embodiment provides a specific calculation process of the channel attention module and the spatial attention module. The specific calculation process of the channel attention module and the spatial attention module is shown in fig. 5.
Wherein, the calculation process of the channel attention module comprises the following steps:
s201, a channel attention mechanism extracts high-level features from the features extracted by two convolution layers and a pooling layer of Lenet5 through maximum pooling and average pooling respectively;
s202, after the output elements are subjected to addition operation through a shared multilayer sensing machine, the final channel attention weight is generated through sigmoid function activation operation;
and S203, performing point multiplication operation on the channel attention weight and the original input feature to generate the input feature required by the spatial attention mechanism. Finally, the channel attention mechanism can be expressed as:
In addition, the calculation process of the spatial attention module includes the following steps:
s301, taking the feature value output by the channel attention module as a feature map input by the space attention module, and performing global maximum pooling and average maximum pooling on the basis of the channel;
s302, concat operation is carried out on the two channels of the space attention module and the channel attention module, dimension reduction is carried out through convolution operation to obtain a feature map of only one channel, space attention weight is generated through a sigmoid function, and finally point multiplication operation is carried out on the weight and the feature map output by the channel attention module to generate final features. Finally, the spatial attention mechanism can be expressed as:
The beneficial effect of this embodiment does: in the embodiment, the recognition effect of the convolutional neural network is improved by adding the channel attention module and the spatial attention module. Because the attention mechanism can enable the model to pay more effective attention to the characteristics which are helpful to the recognition effect, and the characteristics which are not helpful to the recognition effect of the model, the mechanism can automatically give a smaller weight, and the robustness of the model is improved while the recognition capability of the model is improved.
Example 3
On the basis of embodiment 2, in order to verify the effectiveness of embodiment 2, in this embodiment, each working condition is repeatedly acquired 500 times, a loosening index vector is calculated based on embodiment 2, a training and verification set is divided according to the 8:2 principle, a one-dimensional convolutional neural network is trained based on the training method provided by fig. 2, and final model parameters are determined according to the accuracy and loss values of the training set and the verification set. In order to finally detect the performance of the model, 10 groups of piezoelectric signals are collected again under each working condition to obtain a loosening index vector, the loosening index vector is input into the constructed one-dimensional convolution neural network, and the effectiveness of the invention is detected according to the actual measurement effect. The training, validation, and test data set conditions are shown in table 1:
TABLE 1
The training set and the verification set are respectively input into the one-dimensional convolutional neural network, as shown in fig. 6, through 400 iterations, the recognition accuracy of the one-dimensional convolutional neural network of embodiment 2 on the verification set can reach 100%, the loss value is finally 0.025, and the model recognition under the parameters is better in accurate reading. As shown in fig. 7, the actual test result is represented in a confusion matrix form, except that the misdiagnosis rate of the bolt No. 4 and the bolt No. 6 is 10%, the bolt loosening judgment under other working conditions is 100% accurately, and the total loosening detection accuracy of the model is 98%. .
In order to further verify the effectiveness of the introduced spatial channel attention module, the identification effects of a common convolutional neural network and a one-dimensional convolutional neural network are compared, as shown in table 2, the accuracy of the model provided by the invention is higher than that of the models, and the effectiveness of the model provided by the invention is further verified.
TABLE 2
In this embodiment, two convolutional neural network models of Lenet5 have the advantages of simple structure, small number of parameters, high training speed, and the like, and a plurality of data sets exceed the traditional benchmark network model.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. The bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing is characterized by comprising the following steps of: the method comprises the following steps:
s1, building a piezoelectric active sensing system, monitoring the bolt loosening degree of the bolt group in a piezoelectric active sensing mode, and acquiring piezoelectric signals of different bolt loosening of the steel structure bolt group;
s2, calculating a wavelet decomposition energy value of the piezoelectric signal based on a wavelet packet decomposition method, and forming a one-dimensional loosening index vector by the wavelet energy value calculated by the sensor array;
s3, building a one-dimensional convolutional neural network, and building a relation between an elasticity index vector of the bolt group and a bolt loosening position of the bolt group through the one-dimensional convolutional neural network;
s4, training and verifying a one-dimensional convolutional neural network;
and S5, acquiring a piezoelectric sensing signal of the current bolt, inputting the piezoelectric sensing signal into the one-dimensional convolutional neural network, and acquiring the specific position of the bolt group loosening.
2. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 1, wherein: the step S2 specifically includes the following steps:
s101, calculating the decomposition of three layers of wavelet packets into 8 sub-signals based on the acquired piezoelectric sensing signals;
s102, calculating wavelet packet decomposition signal energy components of the piezoelectric sensing signals and wavelet packet decomposition sub-signal energy corresponding to each frequency band;
s103, calculating a one-dimensional loosening index vector of the piezoelectric sensor array signal according to the wavelet packet decomposition sub-signal energy;
s104, standardizing the one-dimensional loosening index vector data according to a standardized function; the normalization function is:
wherein E represents a one-dimensional loosening index vector; mu.sEAn average value representing the energy of the wavelet packet decomposition sub-signal; sigmaERepresenting the standard deviation of the wavelet packet decomposition sub-signal energy.
3. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 1, wherein: the one-dimensional convolutional neural network in S3 includes: the system comprises an input layer, a convolutional neural network Lenet5, a channel attention module, a spatial attention module and an output layer which are connected in sequence.
4. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 3, wherein: the convolutional neural network Lenet5 comprises a convolutional layer 1, a ReLU layer, a maximum pooling layer, a convolutional layer 2, a ReLU layer and a pooling layer which are connected in sequence.
5. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 3, wherein: the calculation process of the channel attention module comprises the following steps:
s201, a channel attention mechanism extracts high-level features from features extracted by two convolutional layers and a pooling layer of a convolutional neural network Lenet5 through maximum pooling and average pooling respectively;
s202, after the output elements are subjected to addition operation through a shared multilayer sensing machine, the final channel attention weight is generated through sigmoid function activation operation;
and S203, performing point multiplication operation on the channel attention weight and the original input feature to generate the input feature required by the spatial attention mechanism.
6. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 5, wherein: the channel attention module is represented as:
7. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 5, wherein: the calculation process of the spatial attention module comprises the following steps:
s301, taking the feature value output by the channel attention module as a feature map input by the space attention module, and performing global maximum pooling and average maximum pooling on the basis of the channel;
s302, concat operation is carried out on the two channels of the space attention module and the channel attention module, dimension reduction is carried out through convolution operation to obtain a feature map of only one channel, space attention weight is generated through a sigmoid function, and finally point multiplication operation is carried out on the weight and the feature map output by the channel attention module to generate final features.
8. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 7, wherein: the spatial attention module is represented as:
wherein M issRepresenting spatial attention weight, f1*7Represents 1 x 7 of the convolution layer,
representing the feature values of the spatial attention module after global average pooling;
representing the feature values of the spatial attention module after global max pooling.
9. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 4, wherein: the neurons of the output layer and the output layer of the spatial attention module are in full connection, each neuron corresponds to the working condition of bolt looseness of the steel structure bolt group, and a softmax function is adopted as an activation function of the output layer;
the activation function is:
in the formula, n and k each represent the number of each neuron, anRepresents the output of the kth neuron of the output layer, K represents the number of output nodes, Soft max (a)n) And (3) representing the output obtained after the k-th neuron is activated by the function, and finally outputting a vector of 1 x 9 by the Lenet5-CBAM depth network, wherein each value can be regarded as the confidence probability of the bolt looseness corresponding to the steel structure bolt group, and the tightness state corresponding to the maximum probability is taken as the result of fault diagnosis.
10. The bolt group loosening and positioning monitoring method based on deep learning and piezoelectric active sensing as claimed in claim 2, wherein: the step S4 specifically includes the following steps:
s401, repeatedly acquiring piezoelectric sensing signals under different tightening conditions of the bolt, obtaining a large amount of data of a training and verification model, and calculating according to the step S2 to obtain standardized multichannel wavelet packet component energy;
s402, dividing the component energy of the standardized multi-channel wavelet packet into a training sample and a test sample according to a ratio of 8:2, inputting the training sample as a one-dimensional convolution neural network, and outputting a corresponding bolt loosening position label as an expectation output of a double-attention convolution neural network model;
during first training, the connection weight of each neuron of the one-dimensional convolutional neural network is set in a random initialization mode, the connection weight of the one-dimensional convolutional neural network is updated by adopting an Adam gradient descent algorithm, and finally the trained connection weight is stored to obtain a steel structure bolt group bolt loosening positioning model;
s403, inputting the test sample into the one-dimensional convolutional neural network, testing the generalization performance of the model, and if the one-dimensional convolutional neural network meets the expected requirement, storing the one-dimensional convolutional neural network; otherwise, the one-dimensional convolutional neural network is adjusted, and the weight of the one-dimensional convolutional neural network is updated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011641905.9A CN112668526A (en) | 2020-12-31 | 2020-12-31 | Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011641905.9A CN112668526A (en) | 2020-12-31 | 2020-12-31 | Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112668526A true CN112668526A (en) | 2021-04-16 |
Family
ID=75412333
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011641905.9A Pending CN112668526A (en) | 2020-12-31 | 2020-12-31 | Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668526A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113267536A (en) * | 2021-05-14 | 2021-08-17 | 大连理工大学 | Multi-frequency band impedance fusion loose bolt positioning method |
| CN113607325A (en) * | 2021-10-09 | 2021-11-05 | 武汉地震工程研究院有限公司 | Intelligent monitoring method and system for looseness positioning of steel structure bolt group |
| CN114492146A (en) * | 2022-04-02 | 2022-05-13 | 武汉地震工程研究院有限公司 | Bolt group loosening positioning and quantitative analysis method and system based on transfer learning |
| CN116592814A (en) * | 2023-07-17 | 2023-08-15 | 塔盾信息技术(上海)有限公司 | Object displacement monitoring method based on artificial intelligence self-adaptive multidimensional calculation |
-
2020
- 2020-12-31 CN CN202011641905.9A patent/CN112668526A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113267536A (en) * | 2021-05-14 | 2021-08-17 | 大连理工大学 | Multi-frequency band impedance fusion loose bolt positioning method |
| CN113607325A (en) * | 2021-10-09 | 2021-11-05 | 武汉地震工程研究院有限公司 | Intelligent monitoring method and system for looseness positioning of steel structure bolt group |
| CN114492146A (en) * | 2022-04-02 | 2022-05-13 | 武汉地震工程研究院有限公司 | Bolt group loosening positioning and quantitative analysis method and system based on transfer learning |
| CN114492146B (en) * | 2022-04-02 | 2022-07-08 | 武汉地震工程研究院有限公司 | Bolt group loosening positioning and quantitative analysis method and system based on transfer learning |
| CN116592814A (en) * | 2023-07-17 | 2023-08-15 | 塔盾信息技术(上海)有限公司 | Object displacement monitoring method based on artificial intelligence self-adaptive multidimensional calculation |
| CN116592814B (en) * | 2023-07-17 | 2023-10-13 | 塔盾信息技术(上海)有限公司 | Object displacement monitoring method based on artificial intelligence self-adaptive multidimensional calculation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112668526A (en) | Bolt group loosening positioning monitoring method based on deep learning and piezoelectric active sensing | |
| CN104712542B (en) | A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults | |
| Li et al. | Damage identification in civil engineering structures utilizing PCA‐compressed residual frequency response functions and neural network ensembles | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| CN111737909B (en) | Structural health monitoring data anomaly identification method based on space-time graph convolutional network | |
| KR102073323B1 (en) | AI Structure Integrity Management System | |
| CN112036042B (en) | Power equipment abnormality detection method and system based on variational modal decomposition | |
| CN102265227A (en) | Method and apparatus for creating state estimation models in machine condition monitoring | |
| CN112785091A (en) | Method for performing fault prediction and health management on oil field electric submersible pump | |
| CN114757309A (en) | Multi-physical-field monitoring data collaborative fusion engineering disaster early warning method and system | |
| CN111898644B (en) | Intelligent identification method for health state of aerospace liquid engine under fault-free sample | |
| CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
| Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
| CN112816122B (en) | Bolt tightness degree monitoring method based on deep learning and piezoelectric active sensing | |
| CN112418529B (en) | Outdoor advertisement online collapse prediction method based on LSTM neural network | |
| CN111982256B (en) | Automobile balance pressure sensor fault early warning method based on multichannel CNN model | |
| CN111291481B (en) | Bayesian model-based structure early warning analysis method | |
| CN117609836A (en) | Electromagnetic sensitivity prediction and health management method for integrated module | |
| CN116821804A (en) | Fracturing pump check valve fault on-line monitoring method based on multi-feature fusion | |
| CN114492146B (en) | Bolt group loosening positioning and quantitative analysis method and system based on transfer learning | |
| Tao et al. | A structural monitoring data processing model based on signal musicalization | |
| CN115659271A (en) | Sensor abnormality detection method, model training method, system, device, and medium | |
| CN113722308B (en) | Acceleration response data complement method and device based on EEMD-MultiCNN-LSTM | |
| Yanez-Borjas et al. | Convolutional neural network-based methodology for detecting, locating and quantifying corrosion damage in a truss-type bridge through the autocorrelation of vibration signals | |
| CN113570111A (en) | On-chip monitoring method for bridge health state based on lightweight network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |