CN112966355B - Method for predicting residual service life of shield machine cutter based on deep learning - Google Patents
Method for predicting residual service life of shield machine cutter based on deep learning Download PDFInfo
- Publication number
- CN112966355B CN112966355B CN202110339632.0A CN202110339632A CN112966355B CN 112966355 B CN112966355 B CN 112966355B CN 202110339632 A CN202110339632 A CN 202110339632A CN 112966355 B CN112966355 B CN 112966355B
- Authority
- CN
- China
- Prior art keywords
- data set
- data
- machine cutter
- shield machine
- sequence
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Excavating Of Shafts Or Tunnels (AREA)
Abstract
The invention discloses a method for predicting the residual service life of a shield machine cutter based on deep learning, which aims to solve the problem of accuracy of residual service life prediction and comprises the following steps: 1. constructing a convolutional neural network; 2. constructing a long-time memory network; 3. generating a data set of the full life cycle of the shield machine cutter; 4. performing dimensionality reduction on the data set by using a principal component analysis method; 5. generating a label data set of the full life cycle of the shield machine cutter; 6. generating a training set; 7. training a convolutional neural network; 8. predicting the health index of the shield machine cutter; 9. generating a health index sequence data set; 10. training a long-time memory network; 11. and predicting the residual service life of the shield machine cutter. The method has the advantage of high accuracy in predicting the residual service life of the shield machine cutter under complex working conditions.
Description
Technical Field
The invention belongs to the technical field of machinery, and further relates to a method and a technology for predicting the residual service life of a shield machine cutter based on deep learning in the technical field of shield machine service life prediction. The method can be used for predicting the residual service life of the shield tunneling machine cutter.
Background
The shield cutter is positioned at the foremost end of the shield machine and is an important part of the shield machine. During the shield tunneling process, the problem caused by the degradation of the cutter is one of the main problems in the shield construction process. Therefore, accurately predicting the residual service life of the cutter of the shield machine is extremely important in the aspect of shield construction management. At present, the tool replacement strategy adopted in China is to manually estimate the residual service life of the shield machine tool according to the experience of a shield driver and experts by combining the geological type, the construction time and the key parameters of the shield machine, and then plan such as regular shutdown inspection, tool replacement and the like is made, so that the method has great subjectivity and randomness. At present, a few research methods are directly used for modeling the service life of a shield machine cutter, and the residual service life of the cutter is usually replaced by easily monitored data such as abrasion loss, the number of cutters, tunneling mileage or deviation of certain sensitive parameters. Therefore, there is a need for improvement in conventional tool replacement strategies and shield machine tool modeling approaches.
The patent document of the southwest university of petroleum, "a shield hob cutter wear prediction method" (patent application number: 2020100095064, publication number: CN111005733 a) proposes a shield hob cutter wear prediction method. The method comprises the following steps: firstly, acquiring unit displacement abrasion loss caused by a plasticity removal mechanism, unit displacement abrasion loss caused by a brittle fracture mechanism, unit displacement abrasion loss caused by adhesion abrasion, unit displacement abrasion loss caused by fatigue abrasion, rock breaking arc length after one point on the front surface of the hob rotates for one circle, space rock breaking arc length after one point on the side surface of the hob rotates for one circle, normal cutting force of the hob and horizontal force applied to the hob; secondly, calculating the radial abrasion and abrasion volume of the front surface of the cutter ring and the abrasion amount and abrasion volume of the side surface of the cutter ring caused by one rotation of the hob; then, multiplying the four kinds of abrasion by respective fitting coefficients and adding to obtain the abrasion volume of the front surface of the cutter ring and the abrasion volume of the side surface of the cutter ring; and finally, generating a loss prediction model and predicting the wear loss. The method can predict the service life of the shield cutter, and solve the problems of serious cutter abrasion, frequent warehouse opening and the like. However, the method still has the defects that the abrasion amount and the abrasion volume formula of the front face and the side face of the cutter ring are obtained by pushing on the basis of an ideal geometric relationship, the severe construction environment of the shield is not considered, and the abrasion form of the cutter is different when the shield is constructed in different geological environments. And the influence of the tool wear forms in different geological environments on wear prediction results in poor accuracy of the life prediction result.
Han Bingyu et al, in the published article, "analysis and prediction of wear of shield cutters in compound strata" (civil engineering report, 2020,53 (S1): 137-142+ 161.), propose a method for predicting wear of shield cutters based on a model of genetic algorithm optimized BP neural network. The method comprises the following steps: firstly, analyzing the wear rules of different types of cutters in a composite stratum; then, the wear coefficients of different types of cutters are obtained according to an empirical formula, and the maximum tunneling distance of the shield tunneling machine under similar stratum conditions is calculated according to the wear coefficients; and finally, optimizing the BP neural network model by using the established genetic algorithm for prediction. The method can accurately predict the abrasion loss of the cutter. However, the method still has the defects that the wear coefficients of different types of cutters of the method are obtained through an empirical formula, the empirical formula is usually too ideal for the assumption of external conditions such as working conditions, the research problem is only limited to cutter wear estimation, and the method is difficult to be used for predicting the residual service life of the shield machine cutter with complicated working conditions.
Disclosure of Invention
The invention aims to provide a method for predicting the residual service life of a shield machine cutter based on deep learning aiming at the defects in the prior art, which is used for solving the problem that the accuracy of the prediction result of the residual service life of the shield machine cutter is poor caused by not considering geological environment and only depending on an empirical formula to estimate the cutter abrasion in the prior art.
The method comprises the steps of firstly, constructing a geological feature set of a numerical value type, fully considering the geological environment corresponding to each shield construction operation data to improve the prediction accuracy of a model for predicting the residual service life of the shield machine cutter due to the fact that the geological environment has a large influence on the accuracy of a prediction result, secondly, reducing the dimension of a data set by using a principal component analysis method, removing data with small influence on cutter abrasion by using the principal component analysis method, improving the stability of a health index result, then constructing a health index of the shield machine cutter by using the data set after dimension reduction, and finally, using a long-time memory network as the model for predicting the residual service life of the shield machine cutter, wherein the long-time memory network has a good effect on the service life prediction problem and ensures the accuracy of the prediction of the residual service life of the shield machine cutter.
In order to achieve the purpose, the technical scheme adopted by the invention comprises the following steps:
(1) Constructing a convolutional neural network:
(1a) A9-layer convolutional neural network is built, and the structure sequentially comprises the following steps: the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the fourth convolution layer, the fourth pooling layer and the full-connection layer;
(1b) Setting the sizes of convolution kernels of the first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer to be 3 x 3, setting the number of the convolution kernels to be 4,8, 16 and 32 respectively, setting the step length to be 1, setting the first pooling layer, the second pooling layer and the fourth pooling layer to be in a maximum pooling mode, setting the sizes of the kernels of the pooling areas to be 2 x 2, setting the step lengths to be 1 and setting the number of neurons of the full connection layer to be 26;
(2) Constructing a long-time memory network:
building a long-term memory network consisting of an input layer, a hidden layer and an output layer, and setting the number of output neurons of the hidden layer to be 5;
(3) Generating a data set of the full life cycle of the shield machine cutter:
(3a) Selecting data with cutter head torque larger than 150 and cutter head rotating speed larger than 0 from shield construction operation data to form a tunneling state data set;
(3b) The geological feature set of the structure is constructed through single heat coding and is combined with the tunneling state data set to form a data set D of the full life cycle of the cutter 1 ；
(4) Using principal component analysis on data set D 1 And (4) performing dimensionality reduction treatment:
(4a) For data set D 1 Carrying out normalization processing to obtain a normalized data set D 2 ；
(4b) Using principal component analysis to normalize the data set D 2 Performing dimensionality reduction to obtain a dimensionality reduced data set D 3 ；
(5) Generating a label data set of the full life cycle of the shield tunneling machine cutter:
and reduced dimension data set D 3 Is constructed to a linear label data set L from 1 to 0 1 ；
(6) Generating a training set:
the data set D after dimensionality reduction 3 Forming a training set by 80% of the data and the label data corresponding to each row of data;
(7) Training a convolutional neural network:
inputting the training set into a convolutional neural network, and updating the weight of the convolutional neural network 1000 times by using a gradient descent method to obtain a trained convolutional neural network;
(8) Predicting the health indexes of the shield machine cutter:
the data set D after dimensionality reduction 3 Inputting the data into a trained convolutional neural network, and outputting a health index data set of the shield machine cutter;
(9) Generating a health index sequence data set:
(9a) Setting the length of a sliding window to be m, setting the moving step length of the window to be 1, performing sliding window on a health index data set, taking all data in the window which slides each time as a sequence, and forming the sequence into a two-dimensional sequence data set, wherein m represents any positive integer taken in a [50,250] interval;
(9b) Forming a label data set L from the m +1 th data to the last data in the health index data set 2 ；
(9c) Sequence dataset and tag dataset L 2 Forming a health index sequence data set;
(10) Training a long-time and short-time memory network:
inputting the first 60% sequence of the health index sequence data set into the long-time and short-time memory network, and updating the weight of the long-time and short-time memory network by a gradient descent method for 100 times to obtain a trained long-time and short-time memory network;
(11) Predicting the residual service life of the shield machine cutter:
(11a) Collecting the health index sequence data, forming a life prediction set by 40% of the sequences, sequentially inputting each sequence in the life prediction set into a trained long-time memory network, comparing each output prediction result with a threshold value, and stopping inputting the sequences in the life prediction set when the output prediction result is smaller than the threshold value;
(11b) Using RUL = | T-T 0 And calculating a residual service life result RUL of the shield machine cutter by using an equation, wherein T represents the acquisition time corresponding to the input sequence of the output prediction result smaller than the threshold value, and T 0 Indicating the acquisition time corresponding to the first sequence in the life prediction set.
Compared with the prior art, the invention has the following advantages:
firstly, the geological feature set is constructed through the unique heat coding, the geological feature set and the tunneling state data set are used for convolutional neural network training, the influence of cutter abrasion forms in different geological environments on abrasion prediction is emphasized, the problem that the service life prediction result is poor in accuracy due to the fact that the shield construction geological environment is not considered in the prior art is effectively solved, and the accuracy of the prediction of the residual service life of the shield machine cutter is improved.
Secondly, the health index of the shield machine cutter is predicted by constructing the convolutional neural network, and the residual service life of the shield machine cutter is predicted by constructing the long-term memory network, so that the problem that the prior art is only limited to cutter abrasion estimation and is difficult to predict the residual service life of the shield machine cutter under complex working conditions is effectively solved, and the accuracy of the prediction result of the residual service life under the complex working conditions is improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph of the results of convolutional neural network prediction according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the prediction results of the long term memory network according to an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments.
The steps implemented by the present invention are described in further detail with reference to fig. 1.
Step 1, constructing a convolutional neural network.
A9-layer convolutional neural network is constructed, and the structure of the convolutional neural network is as follows in sequence: the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the fourth convolution layer, the fourth pooling layer and the full connection layer.
The sizes of convolution kernels of the first convolution layer, the second convolution layer, the third convolution layer, the fourth convolution layer and the fourth convolution layer are all set to be 3 x 3, the number of the convolution kernels is set to be 4,8, 16 and 32, step length is set to be 1, the first pooling layer, the second pooling layer and the fourth pooling layer are set to be in the maximum pooling mode, the size of a pooling area kernel is set to be 2 x 2, the step length is set to be 1, and the number of neurons of the full connection layer is set to be 26.
And 2, constructing a long-term memory network.
And constructing a long-term memory network consisting of an input layer, a hidden layer and an output layer, and setting the number of output neurons of the hidden layer to be 5.
And 3, generating a data set of the full life cycle of the shield tunneling machine cutter.
And selecting data with cutter head torque larger than 150 and cutter head rotating speed larger than 0 from shield construction operation data to form a tunneling state data set.
The geological feature set of the structure is constructed through single heat coding and is combined with the tunneling state data set to form a data set D of the full life cycle of the cutter 1 。
The geological feature set constructed through the single hot coding means that the geological type corresponding to each tunneling state data is converted into a numerical value format from a text format, the numerical value format is represented by a vector with the length of n, the value of n is equal to the total number of the geological types, only one dimension in the vector is 1, the rest is 0, and the geological feature set is formed by the geological type vectors corresponding to all the tunneling state data.
Step 4, using the principal componentAnalysis of data set D 1 And (5) performing dimensionality reduction treatment.
For data set D 1 Carrying out normalization processing to obtain a normalized data set D 2 ；
Data set D using the following normalization formula 1 And (3) carrying out normalization treatment:
wherein D is 2 (i) Representing a data set D 1 Data normalized by the i-th column data in (1), D 1 (i) Representing a data set D 1 The ith column in (1), min (-) represents the min operation, and max (-) represents the max operation.
Using principal component analysis to normalize the data set D 2 Performing dimensionality reduction to obtain a dimensionality-reduced data set D 3 。
The normalized data set D is analyzed by principal component analysis 2 The steps of performing the dimension reduction treatment are as follows:
step 1, according to the following formula, normalizing the data set D 2 Performing centralization operation on each column of data, and forming a matrix X by all the data after the centralization operation.
X(i)＝D 2 (i)-mean(D 2 (i))
Wherein X (i) represents the normalized data set D 2 Line i data-centric data, D 2 (i) Representing the normalized data set D 2 Column i of (2), mean (-) represents the averaging operation.
And step 2, calculating a covariance matrix C of the matrix X according to the following formula.
Where n represents the total number of columns in matrix X.
Step 3, using λ i α i ＝Cα i Formula, calculation assistant formulaThe eigenvalue of each column in the difference matrix C and the eigenvector corresponding to each eigenvalue, wherein, lambda i Represents the ith characteristic value, alpha i Representing a characteristic value λ i The corresponding feature vector.
And 4, arranging all the characteristic values according to the power reduction.
And 5, calculating the variance contribution rate of each characteristic value in the sequence according to the following formula.
Wherein p is j Represents the variance contribution, λ, of the jth eigenvalue in the rank j Representing the jth eigenvalue in the ranking and m representing the total number of eigenvalues.
Step 6, according to
Formula, calculating the cumulative variance contribution rate of the first t characteristic values in the sequence, wherein, 0.5<η<1。
7, forming a data set D with reduced dimensions by using eigenvectors corresponding to the first t eigenvalues in the sequence 3 。
And 5, generating a label data set of the full life cycle of the shield machine cutter.
And reduced dimension data set D 3 Is in one-to-one correspondence with each line of data, a linear label data set L from 1 to 0 is constructed 1 。
And 6, generating a training set.
D, reducing the dimension of the data set D 3 Forming a training set by 80% of the data and the label data corresponding to each row of data;
and 7, training the convolutional neural network.
Inputting the training set into a convolutional neural network, and updating the weight of the convolutional neural network for 1000 times by using a gradient descent method to obtain the trained convolutional neural network.
And 8, predicting the health index of the shield tunneling machine cutter.
The data set D after dimensionality reduction 3 And inputting the data into a trained convolutional neural network, and outputting a health index data set of the shield machine cutter.
And 9, generating a health index sequence data set.
Setting the length of a sliding window as m, setting the moving step length of the window as 1, performing sliding window on the health index data set, taking all data in the window which slides each time as a sequence, and combining all the sequences into a two-dimensional sequence data set, wherein m represents any positive integer taken in a [50,250] interval.
Forming a label data set L from the m +1 th data to the last data in the health index data set 2 。
Sequence dataset and tag dataset L 2 A health index sequence dataset is composed.
And step 10, training the long-time memory network.
And inputting the first 60% sequence of the health index sequence data set into the long-time and short-time memory network, and updating the weight of the long-time and short-time memory network for 100 times by using a gradient descent method to obtain the trained long-time and short-time memory network.
And 11, predicting the residual service life of the shield tunneling machine cutter.
And (3) concentrating the health index sequence data to form a life prediction set by 40 percent of sequences, sequentially inputting each sequence in the life prediction set into a trained long-time and short-time memory network, comparing each output prediction result with a threshold, and stopping the input of the sequences in the life prediction set when the output prediction result is smaller than the threshold.
Using RUL = | T-T 0 And calculating a residual service life result RUL of the shield machine cutter by using an equation, wherein T represents the acquisition time corresponding to the input sequence of the output prediction result smaller than the threshold value, and T 0 Indicating the acquisition time corresponding to the first sequence in the life prediction set.
The effects of the present invention will be further described below with reference to examples of the present invention.
The data set used by the embodiment of the invention is construction data generated in the full life cycle of the shield machine tool in the shield interval from the No. 3 line Liu Wu shop station of the Xiamen subway to the left line of the east boundary station from 9 month and 10 days in 2017 to 10 month and 1 day in 2017 of the Zhongtiesan group Limited company.
1) And constructing a convolutional neural network.
1.1 To build a 9-layer convolutional neural network, whose structure is as follows: the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the fourth convolution layer, the fourth pooling layer and the full connection layer.
1.2 The sizes of convolution kernels of the first to fourth convolution layers are all set to be 3 x 3, the numbers of the convolution kernels are respectively set to be 4,8, 16 and 32, the step sizes are all set to be 1, the first to fourth pooling layers are set to be in the maximum pooling mode, the sizes of the pooling region kernels are all set to be 2 x 2, the step sizes are all set to be 1, and the number of neurons of the full connection layer is set to be 26.
2) And constructing a long-term memory network.
And constructing a long-time memory network consisting of an input layer, a hidden layer and an output layer, and setting the number of output neurons of the hidden layer to be 5.
3) And generating a data set of the full life cycle of the shield machine cutter.
3.1 653082 data which are respectively higher than 150kN m in cutterhead torque and higher than 0rpm in construction data of a shield interval from a No. 3 line Liu five station to an east boundary station of a building subway to form a tunneling state data set. The selected heading state data set is shown in table 1.
TABLE 1 heading status data set
3.2 A geological feature set is constructed through single heat coding and is combined with a tunneling state data set to form a data set D of the full life cycle of the cutter 1 。
The construction of the geological feature set through the single hot coding means that the geological type corresponding to each tunneling state data is converted into a numerical value format from a text format, the numerical value format is represented by a vector with the length of n =5, only one dimension of the vector is 1, and the rest of the vector are 0.
TABLE 2 geological types and corresponding one-hot coded results
| Serial number | Type of geology | One-hot coded results |
| 1 | Moderately/slightly weathered |
10000 |
| 2 | Slightly weathered granite | 01000 |
| 3 | Strongly weathered granite/moderately weathered granite/slightly weathered granite | 00100 |
| 4 | Strongly/moderately weathered granite | 00010 |
| 5 | Strongly weathered granite | 00001 |
And forming a geological feature set by the geological type vectors corresponding to all the tunneling state data.
4) Using principal component analysis on data set D 1 And (5) performing dimensionality reduction treatment.
4.1 For data set D) 1 Carrying out normalization processing to obtain a normalized data set D 2 。
Data set D using the following normalization formula 1 And carrying out normalization processing.
Wherein D is 2 (i) Representing a data set D 1 Normalized data of the ith column, D 1 (i) Representing a data set D 1 The ith column in (1), min (-) represents the min operation, and max (-) represents the max operation.
4.2 Using principal component analysis, on the normalized data set D 2 Performing dimensionality reduction to obtain a dimensionality-reduced data set D 3 。
The normalized data set D is analyzed by principal component analysis 2 The steps of performing the dimension reduction treatment are as follows:
4.2.1 D) normalized data set D according to 2 The centralization operation is carried out on each column of data, and all the data after the centralization operation form a matrix X.
X(i)＝D 2 (i)-mean(D 2 (i))
Wherein X (i) represents the normalized data set D 2 Data after centralization in column i, D 2 (i) Representing the normalized data set D 2 Column i of (2), mean (-) represents the averaging operation.
4.2.2 The covariance matrix C of the matrix X is calculated according to the following equation.
Where n represents the total number of columns in matrix X.
4.2.3 By λ) i α i ＝Cα i Calculating the eigenvalue of each column in the covariance matrix C and the eigenvector corresponding to each eigenvalue, wherein lambda i Denotes the ith characteristic value, α i Representing a characteristic value λ i The corresponding feature vector.
4.2.4 All feature values are sorted by decreasing power.
4.2.5 The variance contribution rate for each eigenvalue in the ranking is calculated according to the following equation.
Wherein p is j Represents the variance contribution ratio, lambda, of the jth eigenvalue in the rank j Representing the jth eigenvalue in the ranking and m representing the total number of eigenvalues.
4.2.6 According to
Formula, calculating the cumulative variance contribution rate of the first t characteristic values in the sequence, wherein, 0.5<η<1。
4.2.7 The eigenvectors corresponding to the first t eigenvalues in the sorting are combined into a reduced-dimension data set D 3 。
5) And generating a label data set of the full life cycle of the shield machine cutter.
And reduced dimension data set D 3 Is in one-to-one correspondence with each line of data, a linear label data set L from 1 to 0 is constructed 1 。
6) And generating a training set.
D, reducing the dimension of the data set D 3 Forming a training set by 80% of the data and the label data corresponding to each row of data;
7) And training the convolutional neural network.
Inputting the training set into the convolutional neural network, and updating the weight of the convolutional neural network 1000 times by using a gradient descent method to obtain the trained convolutional neural network.
8) And (4) predicting the health index of the shield tunneling machine cutter.
The data set D after dimensionality reduction 3 And inputting the data into a trained convolutional neural network, and outputting the health index of the shield machine cutter.
In the embodiment of the present invention, the 10885 healthy fingers are plotted as a broken line as shown in fig. 2, the abscissa in fig. 2 represents time in minutes, and the ordinate represents the value of the healthy index.
9) And generating a health index sequence data set.
The length of a sliding window is set to be m =100, the moving step length of the window is set to be 1, the sliding window is carried out on the health index data set, all data in the window which slides each time are used as a sequence, and all the sequences form a two-dimensional sequence data set.
Forming a label data set L from the m +1=101 data to the last data in the health index data set 2 。
Sequence dataset and tag dataset L 2 A health index sequence dataset is composed.
10 To train a long-short memory network.
And inputting the first 60% sequence of the health index sequence data set into the long-time and short-time memory network, and updating the weight of the long-time and short-time memory network for 100 times by using a gradient descent method to obtain the trained long-time and short-time memory network.
11 Predicting the remaining service life of the shield machine cutter.
And (3) concentrating the health index sequence data to form a life prediction set by 40 percent of sequences, sequentially inputting each sequence in the life prediction set into a trained long-time and short-time memory network, replacing the health index values with the same acquisition time in the life test set by using each output result, comparing each output prediction result with a threshold, and stopping inputting the sequence in the life prediction set when the output prediction result is less than the threshold = 0.15.
In the embodiment of the present invention, 10885 health indicators and 3653 total output prediction results are plotted as a broken line as shown in fig. 3. The abscissa in fig. 3 represents time in minutes, and the ordinate represents a health index value.
Using RUL = | T-T 0 And calculating a residual service life result RUL of the shield machine cutter by using an equation, wherein T represents the acquisition time corresponding to the input sequence of the output prediction result smaller than the threshold value, and T 0 Indicating the acquisition time corresponding to the first sequence in the life prediction set.
Claims (4)
1. A method for predicting the residual service life of a shield machine cutter based on deep learning is characterized in that a geological feature set is constructed through unique hot coding, a health index of the shield machine cutter is predicted through constructing a convolutional neural network, and then the residual service life of the shield machine cutter is predicted through constructing a long-term memory network and a short-term memory network, and the method comprises the following steps:
(1) Constructing a convolutional neural network:
(1a) A9-layer convolutional neural network is constructed, and the structure of the convolutional neural network is as follows in sequence: the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the third convolution layer, the third pooling layer, the fourth convolution layer, the fourth pooling layer and the full-connection layer;
(1b) Setting the sizes of convolution kernels of the first convolution layer, the second convolution layer, the third convolution layer and the fourth convolution layer to be 3 x 3, setting the number of the convolution kernels to be 4,8, 16 and 32 respectively, setting the step length to be 1, setting the first pooling layer, the second pooling layer and the fourth pooling layer to be in a maximum pooling mode, setting the sizes of the kernels of the pooling areas to be 2 x 2, setting the step lengths to be 1 and setting the number of neurons of the full connection layer to be 26;
(2) Constructing a long-time and short-time memory network:
building a long-term memory network consisting of an input layer, a hidden layer and an output layer, and setting the number of output neurons of the hidden layer to be 5;
(3) Generating a data set of the full life cycle of the shield machine cutter:
(3a) Selecting data with cutter head torque larger than 150 and cutter head rotating speed larger than 0 from shield construction operation data to form a tunneling state data set;
(3b) The cutter is formed by combining the geological feature set of the single-heat-coded structure and the tunneling state data setFull lifecycle dataset D 1 ；
(4) Using principal component analysis on data set D 1 And (4) performing dimensionality reduction treatment:
(4a) For data set D 1 Carrying out normalization processing to obtain a normalized data set D 2 ；
(4b) Using principal component analysis to normalize the data set D 2 Performing dimensionality reduction to obtain a dimensionality-reduced data set D 3 ；
(5) Generating a label data set of the full life cycle of the shield machine cutter:
and the reduced dimension data set D 3 Is constructed to a linear label data set L from 1 to 0 1 ；
(6) Generating a training set:
d, reducing the dimension of the data set D 3 Forming a training set by 80% of the data and the label data corresponding to each row of data;
(7) Training a convolutional neural network:
inputting the training set into a convolutional neural network, and updating the weight of the convolutional neural network 1000 times by using a gradient descent method to obtain a trained convolutional neural network;
(8) Predicting the health indexes of the shield machine cutter:
d, reducing the dimension of the data set D 3 Inputting the data into a trained convolutional neural network, and outputting a health index data set of the shield machine cutter;
(9) Generating a health index sequence data set:
(9a) Setting the length of a sliding window to be m, setting the moving step length of the window to be 1, sliding the health index data set, taking all data in the window slid every time as a sequence, and forming a two-dimensional sequence data set by all the sequences, wherein m represents any positive integer taken in a [50,250] interval;
(9b) Forming a label data set L from the m +1 th data to the last data in the health index data set 2 ；
(9c) Set of sequence data and tag data L 2 Forming a health index sequence data set;
(10) Training a long-time memory network:
inputting the first 60% sequence of the health index sequence data set into the long-time and short-time memory network, and updating the weight of the long-time and short-time memory network by a gradient descent method for 100 times to obtain a trained long-time and short-time memory network;
(11) Predicting the residual service life of the shield machine cutter:
(11a) Collecting the health index sequence data, forming a life prediction set by 40% of the sequences, sequentially inputting each sequence in the life prediction set into a trained long-time memory network, comparing each output prediction result with a threshold value, and stopping inputting the sequences in the life prediction set when the output prediction result is smaller than the threshold value;
(11b) Using RUL = | T-T 0 And calculating a residual service life result RUL of the shield machine cutter by using an equation, wherein T represents the acquisition time corresponding to the input sequence of the output prediction result smaller than the threshold value, and T 0 Indicating the acquisition time corresponding to the first sequence in the life prediction set.
2. The method for predicting the remaining service life of the shield tunneling machine cutter based on the deep learning of claim 1, wherein: the construction of the geological feature set through the one-hot coding in the step (3 b) refers to converting the geological type corresponding to each tunneling state data from a text format to a numerical format, and representing the numerical format by using a vector with the length of n, wherein the value of n is equal to the total number of the geological types, only one dimension in the vector takes a value of 1, and the rest are 0, and forming the geological feature set by the geological type vectors corresponding to all the tunneling state data.
3. The method for predicting the remaining service life of the shield tunneling machine cutter based on the deep learning of claim 1, wherein: the normalization process described in step (4 a) is performed by the following equation:
wherein D is 2 (i) Representing a data set D 1 Normalized data of the ith column, D 1 (i) Representing a data set D 1 Column i in (1), min (·) represents a min operation, and max (·) represents a max operation.
4. The method for predicting the remaining service life of the shield tunneling machine cutter based on the deep learning of claim 1, wherein: normalizing the normalized data set D using principal component analysis as described in step (4 b) 2 The steps of performing the dimension reduction treatment are as follows:
first, the normalized data set D is normalized as follows 2 Performing centering operation on each column of data, and forming a matrix X by all the data subjected to centering operation:
X(i)＝D 2 (i)-mean(D 2 (i))
wherein X (i) represents the normalized data set D 2 Line i data-centric data, D 2 (i) Representing the normalized data set D 2 Column i, mean (-) represents the averaging operation;
secondly, calculating a covariance matrix C of the matrix X according to the following formula:
wherein n represents the total number of columns in matrix X;
third step, using λ i α i ＝Cα i Calculating the eigenvalue of each column in the covariance matrix C and the eigenvector corresponding to each eigenvalue, wherein lambda i Denotes the ith characteristic value, α i Representing the characteristic value lambda i A corresponding feature vector;
fourthly, arranging all the characteristic values according to the power reduction;
fifthly, calculating the variance contribution rate of each eigenvalue in the sequence according to the following formula:
wherein p is j Represents the variance contribution ratio, lambda, of the jth eigenvalue in the rank j Representing the jth eigenvalue in the ranking, m representing the total number of eigenvalues;
a sixth step according to
Formula, calculating the cumulative variance contribution rate of the first t characteristic values in the sequence, wherein, 0.5<η<1；
Seventhly, forming a data set D with reduced dimensions by using the eigenvectors corresponding to the first t eigenvalues in the sequence 3 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110339632.0A CN112966355B (en) | 2021-03-30 | 2021-03-30 | Method for predicting residual service life of shield machine cutter based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110339632.0A CN112966355B (en) | 2021-03-30 | 2021-03-30 | Method for predicting residual service life of shield machine cutter based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112966355A CN112966355A (en) | 2021-06-15 |
| CN112966355B true CN112966355B (en) | 2023-01-06 |
Family
ID=76279676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110339632.0A Active CN112966355B (en) | 2021-03-30 | 2021-03-30 | Method for predicting residual service life of shield machine cutter based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112966355B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113553759B (en) * | 2021-06-17 | 2024-04-30 | 浙江工业大学之江学院 | Water pump residual service life prediction method, device, equipment and storage medium |
| CN113722833B (en) * | 2021-09-09 | 2023-06-06 | 湖南工业大学 | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778010A (en) * | 2016-12-29 | 2017-05-31 | 中铁十八局集团隧道工程有限公司 | TBM cutter life Forecasting Methodologies based on data-driven support vector regression |
| CN111337234A (en) * | 2020-03-09 | 2020-06-26 | 西南交通大学 | TBM scraper service life prediction system and method based on real-time monitoring |
| CN111687689A (en) * | 2020-06-23 | 2020-09-22 | 重庆大学 | Cutter wear state prediction method and device based on LSTM and CNN |
| CN112257333A (en) * | 2020-09-24 | 2021-01-22 | 浙江工业大学 | Mechanical equipment internal component service life prediction method based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3594859A1 (en) * | 2018-07-09 | 2020-01-15 | Tata Consultancy Services Limited | Failed and censored instances based remaining useful life (rul) estimation of entities |
-
2021
- 2021-03-30 CN CN202110339632.0A patent/CN112966355B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778010A (en) * | 2016-12-29 | 2017-05-31 | 中铁十八局集团隧道工程有限公司 | TBM cutter life Forecasting Methodologies based on data-driven support vector regression |
| CN111337234A (en) * | 2020-03-09 | 2020-06-26 | 西南交通大学 | TBM scraper service life prediction system and method based on real-time monitoring |
| CN111687689A (en) * | 2020-06-23 | 2020-09-22 | 重庆大学 | Cutter wear state prediction method and device based on LSTM and CNN |
| CN112257333A (en) * | 2020-09-24 | 2021-01-22 | 浙江工业大学 | Mechanical equipment internal component service life prediction method based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| Convolution and Long Short-Term Memory Hybrid Deep Neural Networks for Remaining Useful Life Prognostics;Zhengmin Kong等;《Applied Science》;20191003;第9卷(第19期);第1-18页 * |
| 基于LSTM网络的设备健康状况评估与剩余寿命预测方法的研究;陈自强;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20190930;第2019年卷(第9期);第A002-461页 * |
| 基于长短时记忆卷积神经网络的刀具磨损在线监测模型;何彦等;《中国机械工程》;20200930;第2020年卷(第16期);第1959-1967页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112966355A (en) | 2021-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112966355B (en) | Method for predicting residual service life of shield machine cutter based on deep learning | |
| US10233728B2 (en) | Method and apparatus for drilling a new well using historic drilling data | |
| CN103711523B (en) | Based on the gas density real-time predicting method of local decomposition-Evolutionary Neural Network | |
| CN104794550B (en) | Forecast of Gas Emission method based on WT-KPCA-SVR coupling model | |
| Leng et al. | A hybrid data mining method for tunnel engineering based on real-time monitoring data from tunnel boring machines | |
| CN108470095B (en) | TBM (tunnel boring machine) propulsion prediction method based on data-driven radial basis function model | |
| CN104778378A (en) | Method for analyzing influence factors of production decline of oil and gas field | |
| CN113414638B (en) | Variable working condition milling cutter wear state prediction method based on milling force time sequence diagram deep learning | |
| CN113404502B (en) | Shield hob abrasion monitoring device and method based on ballast piece morphology | |
| CN115422740A (en) | Method for predicting height of water flowing fractured zone of layered fully-mechanized caving mining of huge thick coal seam | |
| CN113431635A (en) | Semi-supervised shield tunnel face geological type estimation method and system | |
| CN112948996A (en) | Cutter head fault risk analysis method based on fault tree and improved analytic hierarchy process | |
| CN110826718B (en) | Method for filling large-section unequal-length missing data based on naive Bayes | |
| CN110472774A (en) | Variable working condition bottom tool method for predicting residual useful life based on LSTM network | |
| CN115774912A (en) | Mechanism and data combined driving TBM hob abrasion prediction method | |
| Li et al. | Product quality evaluation method based on product gene theory | |
| CN107679330B (en) | Real-time evaluation method for rock breaking performance loss degree of TBM cutter head system | |
| CN114818493A (en) | Method for quantitatively evaluating integrity degree of tunnel rock mass | |
| CN113887049A (en) | Drilling speed prediction method and system for petroleum drilling based on machine learning | |
| CN109386235B (en) | Well drilling track optimization method for horizontal well | |
| CN105975564A (en) | Relative entropy similarity-based knowledge recommendation method | |
| CN115563872B (en) | TBM hob abrasion loss prediction method based on DE-SVR algorithm | |
| CN113657515B (en) | Classification method for judging and improving surrounding rock grade of FMC model tunnel based on rock opportunity sense parameters | |
| CN116595334B (en) | Efficient water diversion fracture zone height prediction method based on relevance learning | |
| CN114912335B (en) | Coal gas generation amount prediction method based on missing data |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |