GB2616996A - Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of LSTM-CNN - Google Patents
Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of LSTM-CNN Download PDFInfo
- Publication number
- GB2616996A GB2616996A GB2308984.0A GB202308984A GB2616996A GB 2616996 A GB2616996 A GB 2616996A GB 202308984 A GB202308984 A GB 202308984A GB 2616996 A GB2616996 A GB 2616996A
- Authority
- GB
- United Kingdom
- Prior art keywords
- data
- eddy current
- lstm
- signal
- cnn
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 239000013598 vector Substances 0.000 claims abstract description 38
- 230000007547 defect Effects 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000011176 pooling Methods 0.000 claims description 29
- 239000010410 layer Substances 0.000 claims description 25
- 238000012546 transfer Methods 0.000 claims description 25
- 238000005070 sampling Methods 0.000 claims description 6
- 239000002356 single layer Substances 0.000 claims description 6
- 238000012512 characterization method Methods 0.000 claims description 5
- 101100538446 Aspergillus terreus (strain NIH 2624 / FGSC A1156) trt2 gene Proteins 0.000 claims 1
- 238000012549 training Methods 0.000 abstract description 12
- 238000013527 convolutional neural network Methods 0.000 description 34
- 230000006870 function Effects 0.000 description 13
- 238000012360 testing method Methods 0.000 description 9
- 230000004913 activation Effects 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 230000015654 memory Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000001681 protective effect Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 241001296405 Tiso Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000010200 folin Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
- G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
- G01N27/90—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws using eddy currents
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N27/00—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means
- G01N27/72—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables
- G01N27/82—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws
- G01N27/90—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws using eddy currents
- G01N27/9046—Investigating or analysing materials by the use of electric, electrochemical, or magnetic means by investigating magnetic variables for investigating the presence of flaws using eddy currents by analysing electrical signals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
- Y02E30/30—Nuclear fission reactors
Abstract
A method for recognizing the type of a vortex signal of an evaporator of a nuclear power plant on the basis of an LSTM-CNN. The method comprises: calibrating data of each channel of a vortex signal; processing the calibrated data by using a time window; processing time sequence data in a differential manner; extracting time feature information of a time sequence by means of an LSTM network; a CNN network extracting local feature information of the time sequence; fusing the feature information of the LSTM network and that of the CNN network, wherein after the training and learning of a large amount of data, the feature information thereof can be represented by means of an input signal in vector form by using a triple loss principle; and constructing a defect signal feature database, representing same in vector form, comparing same to obtain the Euclidean distance between the vector feature of the input signal and the vector feature in a defect library, determining, according to the magnitude of the Euclidean distance, the category to which the signal belongs, and ultimately achieving the aim of classifying the vortex signal.
Description
SPECIFICATION
METHOD FOR RECOGNIZING TYPE OF VORTEX SIGNAL OF
EVAPORATOR OF NUCLEAR POWER PLANT ON BASIS OF
L S TM-CNN
FIELD
The present disclosure belongs to the field of nuclear power detection equipment, and in particular, relates to a method for recognizing the type of an eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN.
BACKGROUND
In-service inspection of nuclear power plants usually uses technology of eddy current testing(ET) to track and detect heat transfer tube degradation phenomena, and the accurate and rapid analysis of a large number of steam generator heat transfer tube eddy current data has become one of the main tasks and difficulties. The traditional data analysis method comprises steps of multi-channel data comparison of signals by qualified personnel, and finally a definite conclusion is given. The analysis work is repetitive and complicated, and the analysts are easily fatigued and increase the possibility of human error.
SUMMARY
The purpose of the present disclosure is to provide a method for recognizing the type of an eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN, which intelligently identifies the type of a signal characterized by the time series data of the eddy current signal of an evaporator of a nuclear power plant, and realizes the application of LSTM-CNN technology in nondestructive testing.
To solve the above technical problems, the present disclosure employs a technical solution as follows: a method for recognizing the type of an eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN, is characterized in that, comprising the steps of following: A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant; B. calibrating collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi, chi, chh, ch2y, chNy), that is, recording the eddy current data at ti as (chit, chi, ch2x, chh, .chNx, chNy), recording the eddy current data at t2 as (chi, chi, ch2r, ch2y, ...chNr, chNy), and so on; C. using calibrated data to construct time series based on a time window, wherein a preferred parameter for length of the time window is M sampling points, and the data of this time window can be represented as [ti, t2, tm]; D. processing the data of the time series in step C in a differencing manner, that is, differencing the data of a current moment of each channel from the data of a previous moment of each channel, represented as [t2-ti, t342, . tm-Ti-i], namely 6t2, Otm-i; E. extracting time series feature information using an LSTM network whose input tune series is Sti, ot2, Stm,i; input time series of a CNN network is oti, ot2, 444, the CNN network consists of L convolutional layers, L max-pooling layers and a fully connected layer; fusing outputs of the LSTM network and the CNN network, optimizing network parameters by a principle of triplet loss function to achieve a characterization of an input signal in a vector foim; F. constructing a defect signal feature database using step E, characterizing the database in the vector form, calculating Euclidean distances between vector feature of the input signal and every defect vector feature in the database, comparing the magnitudes of all defect Euclidean distances one by one to determine the category to which the signal belongs, and finally achieving the classification of the eddy current signal; the LSTM network is configured with a single layer unidirectional architecture and consists of a plurality of LSTM units, each of the LSTM units comprises a forget gate, an input gate and an output gate, the LSTM units continuously update and remember information of a previous time series, retains valid information and forgets useless information, and adds the number of Gaussian noise extended data samples with a normal distribution to the data, the Gaussian noise extended data is 5 % -30 % of the number of historical samples.
The present disclosure further provides a method for recognizing the type of an eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN, which comprises steps of A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant; B. calibrating collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi, chi, ch2x, ch2y, ...chNx, chNy), that is, recording the eddy current data at ti as (chi, chi, chn, chn, ...chN,, chNy), recording the eddy current data at t2 as (chi, chi, chn-, ch2y, chNy), and so on; C. using calibrated data to construct time series based on a time window, wherein a preferred parameter for length of the time window is M sampling points, and the data of this time window can be represented as RI, t2, ..., hi]; D. processing the data of the time series in step C in a differencing manner, that is, differencing the data of a current moment of each channel from the data of a previous moment of each channel, represented as [b-ti, b-b, tm-Tm_i], namely Sty, 6t2, Otm-i; E. extracting time series feature information using an LSTM network whose input time series is 6t1, 6t2, 6tm4; input time series of a CNN network is 6t2, 6t2, &An, the CNN network consists of L convolutional layers, L max-pooling layers and a fully connected layer; fusing outputs of the LSTM network and the CNN network, optimizing network parameters by a principle of triplet loss function to achieve a characterization of an input signal in a vector foitti; F. constructing a defect signal feature database using step E, characterizing the database in the vector form, calculating Euclidean distances between vector feature of the input signal and every defect vector feature in the database, comparing the magnitudes of all defect Euclidean distances one by one to determine the category to which the signal belongs, and finally achieving the classification of the eddy current signal.
Optimally, the LSTM network is configured with a single layer unidirectional architecture and consists of a plurality of LSTM units, and each of the LSTM units comprises a forget gate, an input gate and an output gate Further, the LSTM units continuously update and remember information of a previous time series, and retains valid information and forgets useless information.
Optimally, the number of Gaussian noise extended data samples with a normal distribution is added to the data.
Further, the Gaussian noise extended data is 5 % -30 % of the number of historical samples.
The beneficial effects of the present disclosure are as follows: 1. A method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN (Long Short Tenn Memory Network combined with Convolutional Neural Network) provided by the present disclosure, adopts a time series processing mode, and differencing processing of the time series signal can effectively filter out the noise in the signal and extract signal feature information from the variation trend of the signal.
2. A method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN provided by the present disclosure takes the evaporator heat transfer tube eddy current signal containing multiple channels as input, and the constructed LSTM-CNN network can extract the feature information of multiple channels simultaneously to achieve the purpose of recognizing and classifying the eddy current signal.
3. A method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN provided by the present disclosure, by constructing an LSTM-CNN network, wherein the LSTM network is able to solve the signal classification problem from the time series perspective and then extract the time series feature information, and the CNN network is able to extract the local feature information in the time series feature information, fuses the feature information of LSTM and CNN networks, optimizes the network parameters by principle of triplet loss function, and after training and learning from a large amount of data, represents the feature information of an input signal in a vector form; constructs a defect signal feature database, represents the defect signal feature database in the vector form, compares the Euclidean distances between the vector features of the input signal and the vector feature in the database, and according to the magnitudes of the Euclidean distances, determines the classification of the signal.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 shows a flow chart of the network of the present disclosure; Figure 2 shows a graph of the variation of the training loss of the present disclosure.
DETAILED DESCRIPTION
In the following, the present disclosure is described in details combining with embodiments shown in the accompanying drawings: LSTM: LSTM (Long Short-term Memory) refers to a long short-term memory network with a unique design structure that has significant advantages when processing time series data, CNN: (Convolutional Neural Network) refers to a convolutional neural network, which is constructed to emulate the visual mechanisms of living things and has representational learning capabilities.
A method for recognizing the type of an eddy currentsignal of an evaporator of a nuclear power plant on basis of LSTM-CNN, comprises steps of: A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant; B. calibrating the collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi., chi, chn, ch2y, ...chm, chNy), that is, recording the eddy current data at ti as (chix, chi, chn, ch2y, ...chm, chNy), recording the eddy current data at t2 as (chn, chi, chn ch2T, chNy), and so on, the feature information of the N channels can be extracted simultaneously to achieve the purpose of recognizing and classifying the eddy current signal, C. using the calibrated data to construct time series based on a time window, where the preferred parameter for the length of the time window is M sampling points, and the data of this time window can be represented as [li, t2, hi]; D. processing the time series data in step C in a differencing manner, that is, differencing the data of the current moment of each channel from the data of the previous moment of each channel, represented as [t2-ti, t3-t2, tm-Tx14], namely 5t2, Otm_2, which can effectively filter out the noise in the signal and extract signal feature information from the variation trend of the signal; E. the LSTM network adopts a single-layer unidirectional architecture and consists of a plurality of LSTM units, each of the LSTM units comprises a forget gate, an input gate and an output gate, the input time series of the LSTM network is 56, 6t2, otm_2; using the time series feature information extracted by the LSTM network, the LSTM units continuously update and remember the infaimation of the previous time series, to retain valid information as well as to forget useless information; input time series of a CNN network is Oti., Ot2, ..., &km, the CNN network consists of L convolutional layers, L max-pooling layers and a fully connected layer; outputs of the LSTM network and the CNN network are fused, network parameters are optimized using the principle of triplet loss function to achieve the characterization of an input signal in a vector form, and after training and learning from a large amount of data, the feature information of an input signal can be represented in a vector form; F. constructing a defect signal feature database using step E, characterizing the database in the vector form, calculating Euclidean distances between vector feature of the input signal and every defect vector feature in the database, comparing the magnitudes of all defect Euclidean distances one by one to determine the category to which the signal belongs, and finally achieving the classification of the eddy current signal.
Embodiment 1 Assuming that eddy current signal of the evaporator heat transfer tube of a nuclear power plant can be integrated into 10 channels, data of each channel contains a horizontal component and a vertical component; the preferred parameter for the length of the time window is 150 sampling points; the CNN network consists of a convolutional layer-1, a max-pooling layer-1, a convolutional layer-2, a max-pooling layer-2, a convolutional layer-3, a max-pooling layer-3, a convolutional layer-4, a max-pooling layer-4, and a fully connected layer Then, in this embodiment, a flow chart of a method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LS TM-CNN, specifically comprises the following steps: A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant over a period of time; B. calibrating the collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi, chit, ch21, ch2y, ...chiox, chioy), that is, recording the eddy current data at ti as (chi, chi, ch21, ch2y, ...chNx, ChM), recording the eddy current data at t2 as (chi', any, ch2x,, ch2y, . , chNy), and so on; C. using the calibrated data to construct time series based on a time window, the data of this time window can be represented as [ti, t2, tiso]; D. processing the data of the time series in step C in a differencing manner, that is, differencing the data of the current moment of each channel from the data of the previous moment of each channel, represented as [642, t342, tiso-t149] , namely IMF At2' At1191, judging the signal type from the variation trend of time series data can not only effectively filter out the noise in the signal, but also comply with the signal trend recognition method used in manual analysis, and the data processed by time series in a differencing manner can be regarded as a 149*20 matrix; E. according to steps A-D, calibrating and differencing the historical data of the eddy current signal of the evaporator heat transfer tube of the nuclear power plant, and manually labeling according to the category to which the signal feature in the time window belongs, classifying signals in the database according to requirements, including but not limited to heat transfer tube NDD (free section signals), TSP (support signals), and manufacturing stage grinding trace (MBM), using a 3-dimensional one-hot vector to represent the category to which the signal in each time window belongs, representing NDD as [1, 0, 0], TSP as [0, 1, 0], and MBM as [0, 0, 1]; dividing the historical data into training data and test data, wherein the training data is configured to train the model, and the test data is configured to verify the reliability of the model, and for the categories with small number of samples in the historical data, extending the number of samples by adding N(0, 0.001) not mal distribution Gaussian noise, the number of which is 15% of the number of the historical samples to increase the robustness of the model; F. the input of the LSTM network is a time series [t1, M2, *** ,At149], time series feature information could be learnt by using the LSTM network; the input of the CNN network is a time series [ati, At2, *** ,L449], and the CNN network consists of a convolutional layer-1, a max-pooling layer-1, a convolutional layer-2, a max-pooling layer-2, a convolutional layer-3, a max-pooling layer-3, a convolutional layer-4, a max-pooling layer-4, and a fully connected layer; fusing the LSTM network and the CNN network, optimizing network parameters using the principle of triplet loss function, a network that represent the input signal features with vector features could be generated; the LSTM network is configured with a single layer unidirectional architecture and consists of a plurality of LSTM units, wherein each of the LSTM units comprises a forget gate, an input gate and an output gate, the number of LSTMs is 149, the output of the LSTM network is a vector of 1*1*512, and the LSTM units can continuously update and remember the information of the previous time series. The forget gate determines how many of the unit states from the previous moment are preserved to the current moment, the input gate determines how much of the input to the network at the current moment is preserved to the unit states, and the output gate controls how many of the unit states are output to the current output values of the LSTM: Forget gate: ft = cr(Wr * + 14) Input gate: it = 0-01VT ' xti + bi) Memory update: er= taniz(W, [ht_i,xt] + be) Ct = Ct_1+ it 4' et (VT * [111_1, xi] + b0)Output gate: °t = ot tanh(Ct) 147 and We are weight matrixes; hfand hc are bias vectors; tr represents a sigmoid function, which takes a range of values 0-1; tank represents a hyperbolic tangent function, which takes a range of values -1-1; Ct represents the memory update, which takes a value related to the input information, the forgetting gate information, and the memory of the previous moment CL-11; hr. is the output of an implicit layer; The input of the CNN network model is the output result of the LSTM, and the preferred output matrix size is 149*149, the CNN network consists of a convolutional layer-I, a max-pooling 1ayer-1, a convolutional layer-2, a max-pooling layer-2, a convolutional layer-3, a max-pooling layer-3, a convolutional layer-4, a max-pooling layer-4, and a fully connected layer, and a triplet loss function is used to train the network; The convolutional layer-1 of CNN network adopts a rectified linear unit (Relu) activation function, with a preferred convolutional kernel size of 7*1, a preferred number of convolutional kernels of 64, and a preferred step length of 1; The max-pooling layer-2 of CNN network adopts a max-pooling manner, with a preferred pooling layer size of 2*1 and a preferred step length of 2; The convolutional layer-2 of CNN network adopts a Relu activation function, with a preferred convolutional kernel size of 7*1, a preferred number of convolutional kernels of 64, and a preferred step length of; The max-pooling layer-2 of CNN network adopts a common max-pooling manner, with a preferred pooling layer size of 2*1 and a preferred step length of 2; The convolutional layer-3 of CNN network adopts a common Relu activation function, with a preferred convolutional kernel size of 5*1, a preferred number of convolutional kernels of 128, and a preferred step length of 1; The max-pooling layer-3 of CNN network adopts a common max-pooling manner, with a preferred pooling layer size of 2*1 and a preferred step length of 2; The convolutional layer-4 of CNN network adopts a common Relu activation function, with a preferred convolutional kernel size of 3* I, a preferred number of convolutional kernels of 256, and a preferred step length of 1; The max-pooling layer-2 of CNN network adopts a common max-pooling manner, with a preferred pooling layer size of 2*1 and a preferred step length of 2; The number of neurons of the fully connected layer of CNN network is preferred, and the output of the CNN network is a vector of 1 *1*5 1 2; The output of the LSTM network and the CNN network use the fully connected layer to fuse the output results, number of neurons is preferred, and an output vector of 1*1*128 could be constituted; The triplet loss function is: Li = [I; f(r) -f(xn 1;1 fA-a-il (x) -f (4) Ilitk where, ilJ (4) (x) L represents the Euclidean distance between the IIf (4) same signal category, -f (x11) I represents the Euclidean distance between different signal categories, represents the minimum interval between the Euclidean distance between the same signal category and the Euclidean distance between different signal categories The manually labeled training data is fed into the network model of step E, and during a process of one forward pass and one reverse pass, the update of the network parameters can be completed, i.e., one training is completed. After several training sessions, the parameters of the network model can be obtained, and the network structure, weights and other information of the model can be saved as a file in a specified format. The input signal can be characterized in a vector form of 1*1*128 using the network model: The defect signals that have been identified as defect types are selected and characterized as a vector form of 1*1*128 to folin a defect signal feature database; a new input signal is represented as a vector of 1*1*128 and compared one by one with the Euclidean distances between the defect signals in the defect feature database, and the signal category corresponding to the one with the smallest Euclidean distance is considered to be the category to which the new input signal belongs; To evaluate the effectiveness of the model trained in step F, the present disclosure introduces precision, recall rate, and balanced F score (Fl-Score) indexes in the model testing process to evaluate the model. TP (True Positive): an actual positive sample predicted by the model to be a positive sample; TN (True Negative): an actual negative sample predicted by the model to be a negative sample; FP (False Positive): an actual negative sample predicted by the model to be a positive sample; FN (False Negative): an actual positive sample predicted by the model to be a negative sample Precision -Precision: TP+F Recall = Recall rate: TP+F 2precisum. Re r oil Fl-Score: precision+Re In this example, the evaporator heat transfer tube eddy current history data of different units of a nuclear power plant are adopted and divided into training data and test data. Take NDD, TSP and MBM classification test as an example: the total training dataset is 132,320 items, of which 76,875 items are NDD data, 36,134 items are TSP data and 19,311 items are MBM data; the total test dataset is 144,676 items, of which 82,319 items arc NDD data, 36,981 items arc TSP data, and 25,376 items are MBM data. The network was trained 100 times by step 5, step 6 and step 7, and the loss variation of the training process is shown in Figure 2. The effect of the trained model on signal recognition and classification in the test dataset is shown in Table 1, and the precision of prediction of the model is 0.9889, and the classification recognition accuracy of the model is high. From Table 2, it can be seen that the model has high accuracy, recall, and Fl -Score on all classifications, and it has good classification effect on NDD, TSP, and MBM signal recognition.
Confusion matrix Predicted value
NDD TSP MBM
True value NDD 82221 5 93 TSP 36 35475 1470 MBM 0 0 25376 Table I -Test set recognition results of the eddy current signal recognition and classification method provided by the present disclosure Category Precision (%) Recall (%) Fl-Score NDD 99.96 99.88 0.9992 TSP 99.96 95.93 0.9790 MBM 94.20 100 0.9701 Table 2 -Performance comparison of the eddy current signal recognition and classification method provided by the present disclosure The embodiments described above are only for illustrating the technical concepts and features of the present disclosure, and are intended to make those skilled in the art being able to understand the present disclosure and thereby implement it, and should not be concluded to limit the protective scope of this disclosure. Any equivalent variations or modifications according to the spirit of the present disclosure should be covered by the protective scope of the present disclosure.
Claims (6)
- CLAIMS1. A method for recognizing type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN, comprising the steps of: A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant; B. calibrating collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi, chi, chh, ch2y, chNy), that is, recording the eddy current data at ti as (chit, chi, ch2x, chh, .chNx, chNy), recording the eddy current data at t2 as (chi, chi, ch2r, ch2y, ...chNr, chNy), and so on; C. using calibrated data to construct time series based on a time window, wherein a preferred parameter for length of the time window is M sampling points, and the data of this time window can be represented as [ti, t2, tm]; D. processing the data of the time series in step C in a differencing manner, that is, differencing the data of a current moment of each channel from the data of a previous moment of each channel, represented as [t2-ti, t342, . tm-Ti-i], namely 6t2, Otm-i; E. extracting time series feature information using an LSTM network whose input tune series is Sti, ot2, Stm,i; input time series of a CNN network is oti, ot2, 444, the CNN network consists of L convolutional layers, L max-pooling layers and a fully connected layer; fusing outputs of the LSTM network and the CNN network, optimizing network parameters by a principle of triplet loss function to achieve a characterization of an input signal in a vector foini, F. constructing a defect signal feature database using step E, characterizing the database in the vector form, calculating Euclidean distances between vector feature of the input signal and every defect vector feature in the database, comparing the magnitudes of all defect Euclidean distances one by one to determine the category to which the signal belongs, and finally achieving the classification of the eddy current signal; the LSTM network is configured with a single layer unidirectional architecture and consists of a plurality of LSTM units, each of the LSTM units comprises a forget gate, an input gate and an output gate, the LSTM units continuously update and remember information of a previous time series, retains valid information and forgets useless information, and adds the number of Gaussian noise extended data samples with a normal distribution to the data, the Gaussian noise extended data is 5 % -30 % of the number of historical samples.
- 2. A method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN, comprising the steps of: A. collecting eddy current data of an evaporator heat transfer tube of a nuclear power plant; B. calibrating collected eddy current data of the evaporator heat transfer tube of the nuclear power plant, wherein the eddy current data of the evaporator heat transfer tube contains N channels, the data of each channel contains a horizontal component and a vertical component, the calibrating is representing the eddy current data at time t as (chi,, chi!, ch2x, ch2y, chN-y), that is, recording the eddy current data at ti as (chi, chit, ch21, ch2y, chNy), recording the eddy current data at t2 as (chit', any., ch2,, ch2y., ...chN, chNy.), and so on; C. using calibrated data to construct time series based on a time window, wherein a preferred parameter for length of the time window is M sampling points, and the data of this time window can be represented as [ti, t2, tm]; D. processing the data of the time series in step C in a differencing manner, that is, differencing the data of a current moment of each channel from the data of a previous moment of each channel, represented as [b.-6, trt2, tm-T3,44], namely 6t2, Otm4; E. extracting time series feature information using an LSTM network whose input time series is 66, otz, Stm_t; input time series of a CNN network is aft, eatz, ohm, the CNN network consists of L convolutional layers, L max-pooling layers and a fully connected layer; fusing outputs of the LSTM network and the CNN network, optimizing network parameters by a principle of triplet loss function to achieve a characterization of an input signal in a vector form; F. constructing a defect signal feature database using step E, characterizing the database in the vector form, calculating Euclidean distances between vector feature of the input signal and every defect vector feature in the database, comparing the magnitudes of all defect Euclidean distances one by one to determine the category to which the signal belongs, and finally achieving the classification of the eddy current signal.
- 3. The method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN according to claim 2, wherein the LSTM network is configured with a single layer unidirectional architecture and consists of a plurality of LSTM units, and each of the LSTM units comprises a forget gate, an input gate and an output gate.
- 4. The method for recognizing the type of a eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN according to claim 3, wherein the LSTM units continuously update and remember information of a previous time series, and retains valid information and forgets useless information.
- 5. The method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN according to claim 2, further comprising adding the number of Gaussian noise extended data samples with a normal distribution to the data.
- 6. The method for recognizing the type of eddy current signal of an evaporator of a nuclear power plant on basis of LSTM-CNN according to claim 5, wherein the Gaussian noise extended data is 5 -30 % of the number of historical samples.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110377204.7A CN113283288B (en) | 2021-04-08 | 2021-04-08 | Nuclear power station evaporator eddy current signal type identification method based on LSTM-CNN |
PCT/CN2021/129548 WO2022213600A1 (en) | 2021-04-08 | 2021-11-09 | Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of lstm-cnn |
Publications (1)
Publication Number | Publication Date |
---|---|
GB2616996A true GB2616996A (en) | 2023-09-27 |
Family
ID=77276404
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
GB2308984.0A Pending GB2616996A (en) | 2021-04-08 | 2021-11-09 | Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of LSTM-CNN |
Country Status (3)
Country | Link |
---|---|
CN (1) | CN113283288B (en) |
GB (1) | GB2616996A (en) |
WO (1) | WO2022213600A1 (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113283288B (en) * | 2021-04-08 | 2023-08-18 | 中广核检测技术有限公司 | Nuclear power station evaporator eddy current signal type identification method based on LSTM-CNN |
CN116187203A (en) * | 2023-04-20 | 2023-05-30 | 武汉大学 | Watershed water quality prediction method, system, electronic equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020048119A1 (en) * | 2018-09-04 | 2020-03-12 | Boe Technology Group Co., Ltd. | Method and apparatus for training a convolutional neural network to detect defects |
CN111274395A (en) * | 2020-01-19 | 2020-06-12 | 河海大学 | Power grid monitoring alarm event identification method based on convolution and long-short term memory network |
US20200201313A1 (en) * | 2018-12-21 | 2020-06-25 | Canvass Analytics | Predictive analysis of industrial processes |
CN112232483A (en) * | 2020-10-15 | 2021-01-15 | 昆明理工大学 | Flight average fare prediction method combining CNN and LSTM |
CN112561165A (en) * | 2020-12-16 | 2021-03-26 | 南京航空航天大学 | Multidimensional time series data prediction method based on combined model |
CN113283288A (en) * | 2021-04-08 | 2021-08-20 | 中广核检测技术有限公司 | Nuclear power station evaporator eddy current signal type identification method based on LSTM-CNN |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109820525A (en) * | 2019-01-23 | 2019-05-31 | 五邑大学 | A kind of driving fatigue recognition methods based on CNN-LSTM deep learning model |
CN110209715B (en) * | 2019-05-31 | 2022-09-13 | 中广核检测技术有限公司 | Data pre-analysis method for eddy current detection data analysis |
CN110806692A (en) * | 2019-10-21 | 2020-02-18 | 上海海事大学 | Wave compensation prediction method based on CNN-LATM combined model |
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 |
CN111832417B (en) * | 2020-06-16 | 2023-09-15 | 杭州电子科技大学 | Signal modulation pattern recognition method based on CNN-LSTM model and transfer learning |
-
2021
- 2021-04-08 CN CN202110377204.7A patent/CN113283288B/en active Active
- 2021-11-09 GB GB2308984.0A patent/GB2616996A/en active Pending
- 2021-11-09 WO PCT/CN2021/129548 patent/WO2022213600A1/en unknown
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020048119A1 (en) * | 2018-09-04 | 2020-03-12 | Boe Technology Group Co., Ltd. | Method and apparatus for training a convolutional neural network to detect defects |
US20200201313A1 (en) * | 2018-12-21 | 2020-06-25 | Canvass Analytics | Predictive analysis of industrial processes |
CN111274395A (en) * | 2020-01-19 | 2020-06-12 | 河海大学 | Power grid monitoring alarm event identification method based on convolution and long-short term memory network |
CN112232483A (en) * | 2020-10-15 | 2021-01-15 | 昆明理工大学 | Flight average fare prediction method combining CNN and LSTM |
CN112561165A (en) * | 2020-12-16 | 2021-03-26 | 南京航空航天大学 | Multidimensional time series data prediction method based on combined model |
CN113283288A (en) * | 2021-04-08 | 2021-08-20 | 中广核检测技术有限公司 | Nuclear power station evaporator eddy current signal type identification method based on LSTM-CNN |
Also Published As
Publication number | Publication date |
---|---|
CN113283288B (en) | 2023-08-18 |
WO2022213600A1 (en) | 2022-10-13 |
CN113283288A (en) | 2021-08-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105487526B (en) | A kind of Fast RVM sewage treatment method for diagnosing faults | |
CN102122133B (en) | Self-adaption wavelet neural network abnormity detection and fault diagnosis classification system and method | |
GB2616996A (en) | Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of LSTM-CNN | |
CN115828140B (en) | Method, system and application for detecting fault by fusing neighborhood mutual information and random forest | |
CN112756759B (en) | Spot welding robot workstation fault judgment method | |
CN108647707A (en) | Probabilistic neural network creation method, method for diagnosing faults and device, storage medium | |
CN116610998A (en) | Switch cabinet fault diagnosis method and system based on multi-mode data fusion | |
Gao et al. | Blnn: multiscale feature fusion-based bilinear fine-grained convolutional neural network for image classification of wood knot defects | |
CN114015825A (en) | Method for monitoring abnormal state of blast furnace heat load based on attention mechanism | |
CN116383727A (en) | Method, system, equipment and medium for identifying coarse errors in power plant system measurement | |
CN107132515A (en) | A kind of point mark screening technique constrained based on multidimensional information | |
CN114330486A (en) | Power system bad data identification method based on improved Wasserstein GAN | |
CN114239397A (en) | Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning | |
CN116738332A (en) | Aircraft multi-scale signal classification recognition and fault detection method combining attention mechanism | |
CN117216522A (en) | Bearing fault type diagnosis method, device and equipment based on machine learning | |
CN109886288A (en) | A kind of method for evaluating state and device for power transformer | |
CN115985411A (en) | Soft measurement method for polymer melt index based on Gaussian process regression model | |
CN112836719B (en) | Indicator diagram similarity detection method integrating two classifications and triplets | |
CN114647231A (en) | Data-driven industrial process fault detection method | |
CN113657726A (en) | Personnel risk analysis method based on random forest | |
CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
CN105699043A (en) | Method for improving measuring stability and precision of wind tunnel sensor | |
Wu et al. | Network Construction for Bearing Fault Diagnosis Based on Double Attention Mechanism | |
Li et al. | Intelligent diagnosis and recognition method of GIS partial discharge data map based on deep learning | |
CN112732549A (en) | Test program classification method based on cluster analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
789A | Request for publication of translation (sect. 89(a)/1977) |
Ref document number: 2022213600 Country of ref document: WO |