CN111783362B - Method and system for determining residual service life of electric gate valve - Google Patents
Method and system for determining residual service life of electric gate valve Download PDFInfo
- Publication number
- CN111783362B CN111783362B CN202010655443.XA CN202010655443A CN111783362B CN 111783362 B CN111783362 B CN 111783362B CN 202010655443 A CN202010655443 A CN 202010655443A CN 111783362 B CN111783362 B CN 111783362B
- Authority
- CN
- China
- Prior art keywords
- determining
- network model
- convolution
- data
- convolution network
- 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]
- G06F30/20—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
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F16—ENGINEERING ELEMENTS AND UNITS; GENERAL MEASURES FOR PRODUCING AND MAINTAINING EFFECTIVE FUNCTIONING OF MACHINES OR INSTALLATIONS; THERMAL INSULATION IN GENERAL
- F16K—VALVES; TAPS; COCKS; ACTUATING-FLOATS; DEVICES FOR VENTING OR AERATING
- F16K31/00—Actuating devices; Operating means; Releasing devices
- F16K31/02—Actuating devices; Operating means; Releasing devices electric; magnetic
-
- 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
- G06N3/048—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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mechanical Engineering (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a method and a system for determining the residual service life of an electric gate valve. The remaining service life determining method includes: acquiring original data of the electric gate valve, performing data characteristic engineering, and determining two-dimensional input data processed by the characteristic engineering; converting two-dimensional input data into three-dimensional stacked data blocks; determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise reduction self-encoder, and further determining first splicing features; establishing a time convolution network model and determining a second splicing characteristic; determining an updated time convolution network model according to the second splicing characteristics; and determining an optimized convolution noise reduction self-encoder and an optimized time convolution network model according to the convolution noise reduction self-encoder and the updated time convolution network model, and predicting the residual service life of the actual electric valve operation data to determine a residual service life value. By adopting the method or the system for determining the residual service life, provided by the invention, the accuracy of predicting the residual service life can be improved.
Description
Technical Field
The invention relates to the field of prediction of the remaining service life of an electric gate valve, in particular to a method and a system for determining the remaining service life of the electric gate valve.
Background
Remaining Useful Life (RUL) defines the length from the current time to the end of Useful Life, while the main task of state of health assessment and Life prediction is to predict the Remaining time of the machine before it loses operational capability based on state monitoring information. According to the application range, the prediction precision and the related cost, the residual service life prediction method can be divided into the following three types, namely a physical model-based method, a data driving method and a reliability analysis method, as shown in figure 2, each RUL prediction method has the advantages and disadvantages, and the reliability-based RUL prediction refers to that observation data are fitted by using methods such as probability theory, mathematical statistics and the like under the condition of not depending on any physical mechanism, so that the method has the widest applicability. However, this method requires an assumption of a lifetime distribution, which is often quite different from the actual situation. Furthermore, the transition probability matrix is typically estimated from a large amount of training data.
The physical process model-based method establishes a mathematical model in combination with the failure and degradation mechanisms of an object to describe the aging and degradation processes thereof, and finally predicts the RUL value thereof according to the model. Oppenheimer et al used the Forman crack propagation law to predict RUL values for rotor shaft cracking processes. Chan et al established a physical time-varying crack propagation model and applied it to RUL prediction for turbine propulsion systems. El-Tawil et al introduce a random description into the nonlinear impairment law for RUL prediction of pipes and related accessories. In China, King and Rayama et al convert the PE model into an empirical model for RUL prediction; qiujing et al established a bearing stiffness RUL prediction model based on vibration analysis and damage mechanics. In general, modeling and establishing are tedious and complex processes for physical model-based RUL prediction. When the equipment is too complex, the degradation mechanism is difficult to understand, limiting the application of these methods. Furthermore, even if a physical model is established, some parameters in the model are related to material properties and stress levels, and need to be determined through specific experiments.
The data-driven method can avoid the physical modeling difficulty of complex equipment, and therefore has better universality. Deep learning is a new branch of machine learning, and extracts deep abstract features by stacking neurons, expressing complex nonlinear relationships. Deep Neural Networks (DNNs) have greater pattern recognition capabilities than shallow neural networks, and their accuracy is significantly improved when the amount of historical data is sufficient. The accuracy and universality of the existing data-driven-side-based RUL prediction method are low because the continuous time series characteristics of the characteristic parameters in the RUL prediction process are not considered.
Disclosure of Invention
The invention aims to provide a method and a system for determining the residual service life of an electric gate valve, so as to solve the problem that the conventional RUL prediction method is low in accuracy and universality.
In order to achieve the purpose, the invention provides the following scheme:
a method for determining the residual service life of an electric gate valve comprises the following steps:
acquiring original data of the electric gate valve; the original data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor;
performing data feature engineering on the original parameters, and determining two-dimensional input data after feature engineering processing;
converting the two-dimensional input data processed by the characteristic engineering into a three-dimensional stacked data block;
determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise reduction self-encoder, splicing the high-level features and the three-dimensional stacked data blocks in rows, and determining a first splicing feature;
establishing a time convolution network model, processing the new feature combination according to the time convolution network model, and determining a second splicing feature;
transmitting the second splicing characteristics to the time convolution network model, and determining an updated time convolution network model;
optimizing the convolutional noise reduction self-encoder and the updated time convolutional network model, and determining an optimized convolutional noise reduction self-encoder and an optimized time convolutional network model;
predicting the residual service life of actual electric valve operation data according to the optimized convolution self-encoder and the optimized time convolution network model, and determining a residual service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
Optionally, the converting the two-dimensional input data processed by the feature engineering into a three-dimensional stacked data block specifically includes:
converting two-dimensional input data into three-dimensional stacked data blocks of (N-num _ steps +1) ((num _ steps) D) by adopting a sliding time window with the length of num _ steps; wherein, N is the data length related to the sampling time and the sampling frequency; d is the feature dimension.
Optionally, the transmitting the second splicing feature to the time convolution network model, and determining an updated time convolution network model specifically include:
and processing the time convolution network model by using a dropout operation to determine an updated time convolution network model.
Optionally, the transmitting the second splicing feature to the time convolution network model, determining an updated time convolution network model, and then further including:
and adjusting the activation function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a Leaky ReLU function.
Optionally, the optimizing the convolution noise reduction self-encoder and the updated time convolution network model, and determining the optimized convolution noise reduction self-encoder and the optimized time convolution network model specifically include:
and optimizing the weights and the offsets in the convolutional noise reduction self-encoder and the updated time convolutional network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolutional noise reduction self-encoder and the optimized time convolutional network model.
A system for determining remaining service life of a power gate valve, comprising:
the original data acquisition module is used for acquiring original data of the electric gate valve; the original data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor;
the two-dimensional input data determining module is used for performing data characteristic engineering on the original parameters and determining two-dimensional input data processed by the characteristic engineering;
the conversion module is used for converting the two-dimensional input data processed by the feature engineering into a three-dimensional stacked data block;
the high-level feature determination module is used for determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise reduction self-encoder, splicing the high-level features and the three-dimensional stacked data blocks in rows and determining first splicing features;
the second splicing characteristic determining module is used for establishing a time convolution network model, processing the new characteristic combination according to the time convolution network model and determining a second splicing characteristic;
the updating module is used for transmitting the second splicing characteristics to the time convolution network model and determining an updated time convolution network model;
the optimization module is used for optimizing the convolution noise reduction self-encoder and the updated time convolution network model and determining an optimized convolution noise reduction self-encoder and an optimized time convolution network model;
the residual service life value determining module is used for predicting the residual service life of actual operation data of the electric valve according to the optimized convolution self-encoder and the optimized time convolution network model and determining a residual service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
Optionally, the conversion module specifically includes:
the conversion unit is used for converting two-dimensional input data into a three-dimensional stacked data block of (N-num _ steps +1) × (num _ steps × D) by adopting a sliding time window with the length of num _ steps; wherein, N is the data length related to the sampling time and the sampling frequency; d is the feature dimension.
Optionally, the update module specifically includes:
and the updating unit is used for processing the time convolution network model by using dropout operation and determining the updated time convolution network model.
Optionally, the method further includes:
and the adjusting module is used for adjusting the active function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a LeakyReLU function.
Optionally, the optimization module specifically includes:
and the optimization unit is used for optimizing the weight and the bias in the convolution noise reduction self-encoder and the updated time convolution network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolution noise reduction self-encoder and the optimized time convolution network model.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention provides a method and a system for determining the residual service life of an electric gate valve, which adopt an improved Time Convolution Network (TCN) to carry out the system structure design and realization of RUL prediction, solve the problem of service life prediction by starting from methods such as machine learning and deep learning, and carry out service life prediction on the electric gate valve after optimizing the time convolution network. Firstly, deeply extracting features by combining the idea of a convolution denoising self-encoder; then, based on the features after deep extraction, the improved time convolution network is assisted to carry out regression prediction on the global time series data, so that the accuracy of prediction of the residual service life is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a flowchart of a method for determining the remaining service life of an electric gate valve according to the present invention;
FIG. 2 is a complete technical flow diagram of the time-based convolutional network provided by the present invention;
FIG. 3 is a flow chart of a modeling process in a time convolution network module provided by the present invention;
fig. 4 is a structural diagram of a remaining service life determining system of the electric gate valve provided by the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The invention aims to provide a method and a system for determining the residual service life of an electric gate valve, which can improve the accuracy of predicting the residual service life.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flowchart of a method for determining the remaining service life of an electric gate valve, as shown in fig. 1, the method for determining the remaining service life of an electric gate valve includes:
step 101: acquiring original data of the electric gate valve; the original data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor.
Step 102: and performing data characteristic engineering on the original parameters, and determining two-dimensional input data after characteristic engineering processing.
Step 103: and converting the two-dimensional input data processed by the characteristic engineering into a three-dimensional stacked data block.
The step 103 specifically includes: converting two-dimensional input data into three-dimensional stacked data blocks of (N-num _ steps +1) × (num _ steps × D) by adopting a sliding time window with the length of num _ steps; wherein, N is the data length related to the sampling time and the sampling frequency; d is a characteristic dimension; the feature dimension is the number of features.
Step 104: and determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise reduction self-encoder, splicing the high-level features and the three-dimensional stacked data blocks in rows, and determining a first splicing feature.
Step 105: and establishing a time convolution network model, processing the new feature combination according to the time convolution network model, and determining a second splicing feature.
Step 106: and transmitting the second splicing characteristics to the time convolution network model, and determining an updated time convolution network model.
The step 106 specifically includes: and processing the time convolution network model by using a dropout operation to determine an updated time convolution network model.
Step 106 is followed by: and adjusting the activation function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a LeakyReLU function.
Step 107: and optimizing the convolution noise reduction self-encoder and the updated time convolution network model, and determining an optimized convolution noise reduction self-encoder and an optimized time convolution network model.
The step 107 specifically includes: and optimizing the weights and the offsets in the convolutional noise reduction self-encoder and the updated time convolutional network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolutional noise reduction self-encoder and the optimized time convolutional network model.
Step 108: predicting the residual service life of actual electric valve operation data according to the optimized convolution self-encoder and the optimized time convolution network model, and determining a residual service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
Based on the method for determining the remaining service life of the electric gate valve provided by the invention, in practical application, the specific implementation process is as follows:
step 1: the method comprises the steps of obtaining original data by utilizing an acoustic emission sensor, an acceleration sensor, a pressure difference sensor, a temperature sensor, a flow sensor and the like on the electric gate valve, and storing the original data into a computer through a data acquisition board card.
Step 2: and (2) performing data characteristic engineering on the original data acquired in the step (1), removing characteristics irrelevant to life prediction, and performing data normalization and standardization on the data to avoid the influence of dimension on subsequent life prediction.
And step 3: spatial reconstruction of the input data; in order to make the algorithm of the present invention fully consider the time sequence characteristics contained in the feature data in the calculation process, it is necessary to convert the two-dimensional input data (N × D dimension) processed by the feature engineering in step 2 into a three-dimensional stacked data block of (N-num _ steps +1) × (num _ steps × D), wherein N is the total time sequence length, D is the dimension of various sensor characteristic parameters corresponding to the total time sequence length, num _ steps refers to the time sequence number in a sliding window, the invention adopts the sliding time window with the length of num _ steps to slide and convert two-dimensional input data (N x D dimension) into a three-dimensional stacked data block of (N-num _ steps +1) x (num _ steps x D), since there is overlap between the data during each sliding, the total data input length is (N-num _ steps +1) for the algorithm of the present invention. Therefore, the input data at each moment is not an isolated characteristic parameter at a certain moment, but a combination of data in a period of time can better represent the time sequence characteristic in the degradation process.
By converting the original two-dimensional data into the three-dimensional data set with the time series attribute, the subsequent life prediction can be focused not only on a certain single instant but also on a time series, and the data characteristics of the fault process can be reflected.
And 4, step 4: and carrying out unsupervised nonlinear feature extraction according to the convolutional denoising self-coding principle. Due to the fact that the sensor on the experiment table of the electric gate valve has large noise interference in the measuring process.
Therefore, the invention firstly adopts noise reduction convolution self-coding, the actual structure of which is shown in figure 2, and multilayer convolution and pooling in the coding process and multilayer deconvolution and upsampling in the decoding process are built through a Tensorflow framework to form a deep abstract feature representation with invariable time sequence, which is more beneficial to the subsequent change of learning features of a long-term and short-term memory network layer. The resulting high-level data feature may be denoted as C ij Where i represents the time series data length and j represents the dimension of the characteristic parameter.
The convolution denoising self-encoder is adopted for unsupervised feature extraction, so that compared with the condition that feature extraction is not carried out, the feature denoising self-encoder can reflect essential features of data better, and the expressive force of the features is improved.
And 5: high-level feature C obtained from convolutional denoise self-encoder ij Column-wise concatenating the original data block (num _ steps × D) corresponding to step 2 to obtain a new feature combination, which may be denoted as C num_steps,(j+D) The data has stronger expressive ability and can assist the improved time convolution network to fully mine the data characteristics.
The convolution noise reduction self-coding result is combined with the corresponding original data, so that the feature dimensionality of the original data is enriched, the difference between features at different moments is increased, and finally the accuracy of life prediction can be improved.
Step 6: establishing a time convolution network model; for each block of the dashed time convolutional network TCN on the right side of fig. 2, the correlation model is built using the steps shown in fig. 3.
The data obtained in step 5 is first processed using dilation convolution and causal convolution, which can introduce a fixed step size between two adjacent convolution filters. When the dilation factor d is 1, the convolution is only a common convolution. The larger the expansion coefficient, the longer the input range. Therefore, a better convolution network receptive field can be obtained. The invention can freely adjust the structure of TCN by changing the size of the expansion factor and the convolution kernel, so that the TCN has a flexible receiving domain. In order to solve the problems of gradient disappearance and the like possibly caused by a depth TCN model, the TCN structure is more universal by referring to a residual convolution structure in a residual network, as shown in FIG. 2. On the basis of residual convolution, the invention initiatively optimizes the network structure of the residual convolution, forms a series-parallel connection structure of input data and output results, enriches data characteristic dimensionality and more effectively memorizes characteristic information.
By adopting the time convolution network model, the size of the receptive field of the time convolution network can be adjusted at will by adjusting the expansion factor and the convolution kernel size, and meanwhile, the causal time sequence relation among data can be deeply memorized by setting the serial-parallel structure of input and output in each layer of residual convolution on the basis of the residual convolution, so that a better service life prediction effect is achieved; compared with a long-time memory network and a short-time memory network, the calculation time is faster, the needed computer resources are less, and the problem of gradient explosion of the long-time memory network during parameter optimization can be solved.
And 7: and (4) transferring the new characteristics spliced in the step (6) to the time convolution network, and using dropout operation on the formed time convolution network model on the basis of the step (6) to make the time convolution network more robust and determine the updated time convolution network model.
By adopting dropout operation in the neural network, overfitting of the neural network result can be prevented, so that the obtained life prediction result is more stable, and excessive fluctuation can not be generated.
And 8: the activation functions involved in the convolutional self-encoder and the time convolutional network are adjusted to LeakyReLU, so that dead nodes can be avoided on the basis of the ReLU activation functions, and the nonlinear characteristics in data can be reflected.
The LeakyReLU can avoid dead nodes on the basis of the ReLU activation function, so that the sparse model can better mine relevant features, the training data is fitted, and the nonlinear features in the data can be reflected.
And step 9: initializing parameters of a convolution noise reduction self-encoder and an updated time convolution network and training the network; in the training process of the model, all data are split into a plurality of batches of training samples in order to improve the training speed and efficiency, and the processed data are randomly disturbed to reduce uncertainty and then input into a convolution self-encoder and a time convolution network for training.
Step 10: defining a loss function and optimizing parameters; the present invention uses Mean Square Error (MSE) function as the loss function. In order to optimize the weights and the offsets in the convolution self-encoder and the time convolution network, an SGD optimization algorithm is adopted to solve the network in the training process, so that the loss function value is as small as possible, and finally, the network structure parameters which best meet the service life prediction characteristics of the electric valve are obtained.
In the calculation process of each back propagation, the learning rate of the first 5 iterations is set to be 0.001, the learning rate is not attenuated, and the attenuation rate of the learning rate of each subsequent iteration is set to be 0.99. With the increase of the number of training rounds and the reduction of training errors, the convolution self-encoder and the long-time and short-time memory network prediction model can continuously approach to the actual fault and aging characteristics.
In the calculation process of each back propagation, the learning rate of the first 5 iterations is set to be 0.001, the learning rate is not attenuated, and the attenuation rate of the learning rate of each subsequent iteration is set to be 0.99. Through the change of the learning rate, the most appropriate weight and bias can be found more accurately in the back propagation calculation process, and finally the accuracy of the model is improved.
Step 11: after the off-line training process is completed, the optimized convolution self-encoder and the time convolution network model can be used for predicting the RUL of the actual operation process of the electric valve. Repeating the data normalization and data preprocessing in step 2 and step 3 on the actual degradation data as in the data feature engineering in the training process, and determining the three-dimensional stacked data block of (N-num _ steps +1) ((num _ steps) D) consistent with step 3.
Step 12: and (3) predicting the residual service life of the actual electric valve operation data obtained in the step (11) by adopting the convolution self-encoder and the time convolution network model which are trained and optimized in the steps (1) to (11), finally obtaining a residual service life value, wherein the related result can be referred by maintenance and decision-making personnel, and related measures can be taken in time, so that the safety is ensured, and the economy can be improved.
Fig. 4 is a structural diagram of a remaining service life determining system of an electric gate valve provided by the present invention, and as shown in fig. 4, the remaining service life determining system of an electric gate valve includes:
an original data acquisition module 401, configured to acquire original data of the electric gate valve; the raw data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor.
A two-dimensional input data determining module 402, configured to perform data feature engineering on the original parameters, and determine two-dimensional input data after feature engineering processing.
A conversion module 403, configured to convert the two-dimensional input data after the feature engineering process into a three-dimensional stacked data block.
The conversion module 403 specifically includes: the conversion unit is used for converting two-dimensional input data into a three-dimensional stacked data block of (N-num _ steps +1) × (num _ steps × D) by adopting a sliding time window with the length of num _ steps; wherein N is the data length related to the sampling time and the sampling frequency; d is the characteristic dimension.
And a high-level feature determining module 404, configured to determine a high-level feature according to the three-dimensional stacked data block by using a convolutional denoising autoencoder, and splice the high-level feature and the three-dimensional stacked data block in a row to determine a first splicing feature.
And a second splicing feature determining module 405, configured to establish a time convolution network model, process the new feature combination according to the time convolution network model, and determine a second splicing feature.
An updating module 406, configured to transfer the second splicing feature to the time convolution network model, and determine an updated time convolution network model.
And the adjusting module is used for adjusting the activation function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a LeakyReLU function.
The update module 406 specifically includes: and the updating unit is used for processing the time convolution network model by using dropout operation and determining the updated time convolution network model.
And an optimizing module 407, configured to optimize the convolutional denoising autoencoder and the updated time convolutional network model, and determine an optimized convolutional denoising autoencoder and an optimized time convolutional network model.
The optimization module 407 specifically includes: and the optimization unit is used for optimizing the weights and the offsets in the convolution noise reduction self-encoder and the updated time convolution network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolution noise reduction self-encoder and the optimized time convolution network model.
A remaining service life value determining module 408, configured to perform remaining service life prediction on actual operation data of the electrically operated valve according to the optimized convolutional self-encoder and the optimized time convolutional network model, and determine a remaining service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
The method deeply extracts the characteristics by combining the idea of a convolution noise reduction self-encoder; based on the features after deep extraction, the improved time convolution network is assisted to carry out regression prediction on global time series data, so that the accuracy of residual service life prediction is improved.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method for determining the residual service life of an electric gate valve is characterized by comprising the following steps:
acquiring original data of the electric gate valve; the original data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor;
performing data characteristic engineering on the original data, and determining two-dimensional input data processed by the characteristic engineering;
converting the two-dimensional input data processed by the characteristic engineering into a three-dimensional stacked data block;
converting the two-dimensional input data processed by the feature engineering into a three-dimensional stacked data block specifically comprises:
converting two-dimensional input data into three-dimensional stacked data blocks of (N-num _ steps +1) × (num _ steps × D) by adopting a sliding time window with the length of num _ steps; wherein N is the data length related to the sampling time and the sampling frequency; d is a characteristic dimension;
determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise reduction self-encoder, splicing the high-level features and the three-dimensional stacked data blocks in rows, and determining first splicing features;
establishing a time convolution network model, processing the first splicing characteristic according to the time convolution network model, and determining a second splicing characteristic;
the establishing of the time convolution network model specifically comprises the following steps:
for each time convolution network TCN, processing the first splicing characteristic by adopting expansion convolution and causal convolution, and introducing a fixed step length between two adjacent convolution filters; freely adjusting the structure of the TCN by changing the sizes of the expansion factor and the convolution kernel;
the input and output serial-parallel structure in each layer of residual convolution is set on the basis of the residual convolution, so that the causal time sequence relation among data can be deeply memorized;
transmitting the second splicing characteristics to the time convolution network model, and determining an updated time convolution network model;
optimizing the convolutional noise reduction self-encoder and the updated time convolutional network model, and determining an optimized convolutional noise reduction self-encoder and an optimized time convolutional network model;
predicting the residual service life of actual electric valve operation data according to the optimized convolution self-encoder and the optimized time convolution network model, and determining a residual service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
2. The method for determining the remaining service life of a power gate valve according to claim 1, wherein the step of transmitting the second splicing feature to the time convolution network model to determine an updated time convolution network model specifically comprises:
and processing the time convolution network model by using a dropout operation to determine an updated time convolution network model.
3. The method for determining remaining service life of a power gate valve according to claim 2, wherein the step of transferring the second splicing feature to the time convolution network model to determine an updated time convolution network model further comprises the steps of:
and adjusting the activation function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a Leaky ReLU function.
4. The method for determining the remaining service life of a power gate valve according to claim 1, wherein the optimizing the convolutional noise reduction self-encoder and the updated time convolutional network model to determine an optimized convolutional noise reduction self-encoder and an optimized time convolutional network model specifically comprises:
and optimizing weights and offsets in the convolution noise reduction self-encoder and the updated time convolution network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolution noise reduction self-encoder and the optimized time convolution network model.
5. A system for determining remaining service life of a power gate valve, comprising:
the original data acquisition module is used for acquiring original data of the electric gate valve; the original data comprises data detected by an acoustic emission sensor, an acceleration sensor, a differential pressure sensor, a temperature sensor and a flow sensor;
the two-dimensional input data determining module is used for performing data characteristic engineering on the original data and determining two-dimensional input data processed by the characteristic engineering;
the conversion module is used for converting the two-dimensional input data processed by the characteristic engineering into a three-dimensional stacked data block;
the conversion module specifically comprises:
the conversion unit is used for converting two-dimensional input data into a three-dimensional stacked data block of (N-num _ steps +1) × (num _ steps × D) by adopting a sliding time window with the length of num _ steps; wherein N is the data length related to the sampling time and the sampling frequency; d is a characteristic dimension;
the high-level feature determining module is used for determining high-level features according to the three-dimensional stacked data blocks by using a convolution noise-reduction self-encoder, splicing the high-level features and the three-dimensional stacked data blocks in rows and determining first splicing features;
the second splicing characteristic determining module is used for establishing a time convolution network model, processing the first splicing characteristic according to the time convolution network model and determining a second splicing characteristic;
the establishing of the time convolution network model specifically comprises the following steps:
for each time convolution network TCN, processing the first splicing characteristic by adopting expansion convolution and causal convolution, and introducing a fixed step length between two adjacent convolution filters; freely adjusting the structure of the TCN by changing the sizes of the expansion factor and the convolution kernel;
the input and output serial-parallel structure in each layer of residual convolution is set on the basis of the residual convolution, so that the causal time sequence relation among data can be deeply memorized;
the updating module is used for transmitting the second splicing characteristics to the time convolution network model and determining an updated time convolution network model;
the optimization module is used for optimizing the convolution noise reduction self-encoder and the updated time convolution network model and determining an optimized convolution noise reduction self-encoder and an optimized time convolution network model;
the residual service life value determining module is used for predicting the residual service life of actual operation data of the electric valve according to the optimized convolution self-encoder and the optimized time convolution network model and determining a residual service life value; the actual electric valve operation data comprises ringing count, amplitude, duration, energy, root mean square value, average signal level, pressure difference of a fluid inlet end and a fluid outlet end of the valve, flow behind the valve and opening degree of the valve, wherein the ringing count, the amplitude, the duration, the energy, the root mean square value and the average signal level are acquired by the acoustic emission sensor.
6. The system for determining the remaining service life of a power gate valve according to claim 5, wherein the updating module specifically comprises:
and the updating unit is used for processing the time convolution network model by using dropout operation and determining the updated time convolution network model.
7. The system for determining remaining service life of a power gate valve according to claim 6, further comprising:
and the adjusting module is used for adjusting the activation function involved in the convolution noise reduction self-encoder and the updated time convolution network model into a Leaky ReLU function.
8. The system for determining the remaining service life of a power gate valve according to claim 5, wherein the optimizing module specifically comprises:
and the optimization unit is used for optimizing the weight and the bias in the convolution noise reduction self-encoder and the updated time convolution network model according to the loss function by adopting a mean square error function as the loss function, and determining the optimized convolution noise reduction self-encoder and the optimized time convolution network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010655443.XA CN111783362B (en) | 2020-07-09 | 2020-07-09 | Method and system for determining residual service life of electric gate valve |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010655443.XA CN111783362B (en) | 2020-07-09 | 2020-07-09 | Method and system for determining residual service life of electric gate valve |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111783362A CN111783362A (en) | 2020-10-16 |
| CN111783362B true CN111783362B (en) | 2022-07-26 |
Family
ID=72759433
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010655443.XA Active CN111783362B (en) | 2020-07-09 | 2020-07-09 | Method and system for determining residual service life of electric gate valve |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111783362B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI780961B (en) * | 2020-12-03 | 2022-10-11 | 矻頎投資股份有限公司 | Valve with capability to self-anticipate failure |
| CN112881518B (en) * | 2021-01-08 | 2022-07-12 | 东冶及策河北能源技术有限公司 | Method for predicting residual life of dynamic filter compensator |
| CN113570138B (en) * | 2021-07-28 | 2022-04-26 | 朗坤智慧科技股份有限公司 | Method and device for predicting residual service life of equipment of time convolution network |
| CN113660236B (en) * | 2021-08-10 | 2023-05-09 | 和安科技创新有限公司 | Abnormal flow detection method based on optimized stacked noise reduction convolution self-coding network, memory and processor |
| CN113655817B (en) * | 2021-10-20 | 2022-03-08 | 广东奥迪威传感科技股份有限公司 | Flow control method, device, equipment and storage medium |
| FR3138179A1 (en) * | 2022-07-21 | 2024-01-26 | L'air Liquide, Societe Anonyme Pour L'etude Et L'exploitation Des Procedes Georges Claude | Device for detecting a valve anomaly |
| CN116400675B (en) * | 2023-06-09 | 2023-09-15 | 西南交通大学 | Fault diagnosis system and method based on improved CNN-LSTM model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543905A (en) * | 2018-11-23 | 2019-03-29 | 西安电子科技大学 | Rolling bearing method for predicting residual useful life based on improved two dimension CNN model |
| CN110377984A (en) * | 2019-07-02 | 2019-10-25 | 中国科学院深圳先进技术研究院 | A kind of industrial equipment remaining useful life prediction technique, system and electronic equipment |
| CN111274737A (en) * | 2020-02-25 | 2020-06-12 | 山东大学 | Method and system for predicting remaining service life of mechanical equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11226374B2 (en) * | 2017-10-17 | 2022-01-18 | The Board Of Trustees Of The Leland Stanford Junior University | Data-driven model for lithium-ion battery capacity fade and lifetime prediction |
| CN109992872B (en) * | 2019-03-27 | 2020-07-28 | 西安交通大学 | Mechanical equipment residual life prediction method based on stacked separation convolution module |
-
2020
- 2020-07-09 CN CN202010655443.XA patent/CN111783362B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543905A (en) * | 2018-11-23 | 2019-03-29 | 西安电子科技大学 | Rolling bearing method for predicting residual useful life based on improved two dimension CNN model |
| CN110377984A (en) * | 2019-07-02 | 2019-10-25 | 中国科学院深圳先进技术研究院 | A kind of industrial equipment remaining useful life prediction technique, system and electronic equipment |
| CN111274737A (en) * | 2020-02-25 | 2020-06-12 | 山东大学 | Method and system for predicting remaining service life of mechanical equipment |
Non-Patent Citations (2)
| Title |
|---|
| Remaining useful life prediction for mechanical equipment based on Temporal convolutional network;Ji Wenqiang等;《2019 14th IEEE International Conference on Electronic Measurement & Instruments (ICEMI)》;20200527;1192-1199 * |
| 基于深度时序特征迁移的轴承剩余寿命预测方法;陈佳鲜等;《控制与决策》;20200331;1699-1706 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111783362A (en) | 2020-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111783362B (en) | Method and system for determining residual service life of electric gate valve | |
| CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
| Tao et al. | Bearing fault diagnosis method based on stacked autoencoder and softmax regression | |
| Roy et al. | A stacked autoencoder neural network based automated feature extraction method for anomaly detection in on-line condition monitoring | |
| Chen et al. | Diagnosing planetary gear faults using the fuzzy entropy of LMD and ANFIS | |
| CN105868557A (en) | Online prediction method for remaining life of electromechanical equipment under situation of two-stage degradation | |
| Chen et al. | Vibration-based gearbox fault diagnosis using deep neural networks | |
| CN106055579B (en) | Vehicle performance data cleaning system and method based on artificial neural network | |
| CN111507046A (en) | Method and system for predicting remaining service life of electric gate valve | |
| Sun et al. | Fault diagnosis for bearing based on 1DCNN and LSTM | |
| CN111709577B (en) | RUL prediction method based on long-range correlation GAN-LSTM | |
| Martin-del-Campo et al. | Dictionary learning approach to monitoring of wind turbine drivetrain bearings | |
| CN115935834A (en) | History fitting method based on deep autoregressive network and continuous learning strategy | |
| CN114166509A (en) | Motor bearing fault prediction method | |
| CN116227586A (en) | Meta learning fault diagnosis method and device based on depth residual error shrinkage prototype network | |
| CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN114548154A (en) | Intelligent diagnosis method and device for important service water pump | |
| CN113151842B (en) | Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production | |
| CN112257694B (en) | Sparse constraint generation countermeasure network implementation method for rotary mechanical vibration signals | |
| CN112348158B (en) | Industrial equipment state evaluation method based on multi-parameter deep distribution learning | |
| CN113469013A (en) | Motor fault prediction method and system based on transfer learning and time sequence | |
| Jiang et al. | Enhancing convolutional neural network deep learning for remaining useful life estimation in smart factory applications | |
| CN116935128A (en) | Zero sample abnormal image detection method based on learning prompt | |
| CN116992757A (en) | Wellhead pressure prediction method and device based on deep learning and rolling optimization | |
| Luo et al. | A novel method for remaining useful life prediction of roller bearings involving the discrepancy and similarity of degradation trajectories |
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 |