CN114580101B - Method and system for predicting residual service life of rotary machine - Google Patents
Method and system for predicting residual service life of rotary machine Download PDFInfo
- Publication number
- CN114580101B CN114580101B CN202210166709.3A CN202210166709A CN114580101B CN 114580101 B CN114580101 B CN 114580101B CN 202210166709 A CN202210166709 A CN 202210166709A CN 114580101 B CN114580101 B CN 114580101B
- Authority
- CN
- China
- Prior art keywords
- chebgcn
- service life
- lstm
- representing
- rotary machine
- 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
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000010586 diagram Methods 0.000 claims abstract description 59
- 238000012549 training Methods 0.000 claims abstract description 33
- 238000006731 degradation reaction Methods 0.000 claims abstract description 12
- 230000015556 catabolic process Effects 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 19
- RVRCFVVLDHTFFA-UHFFFAOYSA-N heptasodium;tungsten;nonatriacontahydrate Chemical compound O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.O.[Na+].[Na+].[Na+].[Na+].[Na+].[Na+].[Na+].[W].[W].[W].[W].[W].[W].[W].[W].[W].[W].[W] RVRCFVVLDHTFFA-UHFFFAOYSA-N 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 4
- 230000007812 deficiency Effects 0.000 abstract 1
- 238000005070 sampling Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a method and a system for predicting the residual service life of rotary machinery, wherein the method comprises the following steps: acquiring a real-time vibration signal of the rotary machine to be predicted, taking the vibration signal as a prediction sample, and constructing a prediction path diagram according to the prediction sample; inputting the predicted path diagram into a ChebGCN-LSTM prediction model after training to obtain a predicted value of the residual service life of the rotary machine to be predicted; the method for obtaining the ChebGCN-LSTM predictive model after training comprises the following steps: and constructing a ChebGCN-LSTM prediction model based on ChebGCN and BiLSTM residual service life of the rotary machine, inputting a training path diagram into the ChebGCN-LSTM prediction model for training, and obtaining a trained ChebGCN-LSTM prediction model. The invention uses the path diagram to represent the time and space dependence of the signals, and the degradation state of the rotary machine is more comprehensively represented; by using ChebGCN-LSTM prediction model, the defect and deficiency of the traditional ChebGCN in the aspect of representing the global time correlation of signals are overcome, and the prediction precision is improved.
Description
Technical Field
The invention belongs to the technical field of prediction of residual service life of rotary machinery, and particularly relates to a method and a system for predicting residual service life of rotary machinery.
Background
Rotary machines have been widely used as key components of transmission systems in modern manufacturing and industrial processes, and their operating conditions directly affect overall plant performance. If the residual service life of the rotary machine can be accurately predicted before the rotary machine fails, certain measures can be timely taken to maintain the equipment, so that serious economic loss and even casualties are avoided.
The remaining service life prediction methods of the rotary machine can be generally divided into two main categories, namely a method based on an analytical model and a method based on data driving. The method based on the analytical model predicts the residual service life by constructing a mathematical or physical model of the degradation process of the rotary mechanical performance, but the method for constructing the analytical model becomes more difficult as the running environment of the rotary mechanical becomes more complex; the data-driven method benefits from the development of machine learning, and a fuzzy function of the degradation process of the rotating machinery performance can be simulated from massive monitoring data, so that the residual service life of the rotating machinery is predicted. Therefore, a method for predicting the remaining service life of a rotary machine based on data driving is popular and studied by a plurality of students at home and abroad in recent years, and is becoming a mainstream method.
The prediction of the residual service life of the rotating machinery is essentially a time series prediction problem, while the new generation artificial intelligence technology represented by the cyclic neural network has obvious advantages in time series processing, however, the methods pay more attention to the time dependence of the signals and ignore the spatial dependence of the signals. In fact, the spatial dependence of the signal provides more characteristic information. The graph is used as non-Euclidean data, and compared with the traditional Euclidean data, the graph edge can be used for representing the spatial dependence relation of signals, so that the graph data is used for analyzing and predicting the residual service life, and the effect is further improved. Based on the advantages of the comprehensive graph data representing the signal space dependence and the signal capturing time information of the cyclic neural network, the invention provides a method for predicting the residual service life of the rotary machine.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides a method and a system for predicting the residual service life of a rotary machine, and aims to solve the problem that the prediction precision of the residual service life of the rotary machine is low because the spatial dependence of vibration signals of the rotary machine is not fully focused in the prior art.
To achieve the above object, according to one aspect of the present invention, there is provided a method for predicting remaining life of a rotary machine, comprising:
Acquiring a real-time vibration signal of the rotary machine to be predicted, taking the vibration signal as a prediction sample, and constructing a prediction path diagram according to the prediction sample;
And inputting the predicted path diagram into a trained ChebGCN-LSTM prediction model to obtain a predicted value of the residual service life of the rotary machine to be predicted.
Further, the method for obtaining the ChebGCN-LSTM prediction model after training comprises the following steps:
obtaining vibration signals of the rotary machine in different life periods, taking the vibration signals as training samples, and constructing a training path diagram according to the training samples;
And constructing a ChebGCN-LSTM prediction model of the residual service life of the rotary machine based on ChebGCN (chebyshev diagram convolutional network) and BiLSTM (bidirectional long-short-time memory network), inputting a training path diagram into the ChebGCN-LSTM prediction model for training, and obtaining a trained ChebGCN-LSTM prediction model.
Further, the construction method of the path diagram comprises the following steps:
Extracting time domain statistical indexes of samples, regarding vibration signals as nodes, regarding time domain statistical indexes as node characteristics, regarding the sequence of life time periods of the vibration signals as edges, and connecting each node with adjacent nodes before and after the node by edges to form a path diagram;
wherein the path graph is denoted g= { V, E, a, F }, where V denotes the path graph node, E denotes the edge connection between two nodes, F denotes the node characteristics, and a denotes the adjacency matrix.
Further, the time domain statistical indexes are a mean value T 1, a standard deviation T 2, a square root amplitude T 3, a root mean square T 4, a peak value T 5, a slope T 6, a kurtosis factor T 7, a peak coefficient T 8, a clearance coefficient T 9, a shape factor T 10 and an impact factor T 11;
given a vibration signal of x= (X 1,x2,…,xm), where m is the data length:
The mean T 1 is:
The standard deviation T 2 is:
square root amplitude T 3 is:
root mean square T 4 is:
The peak-to-peak value T 5 is: t 5＝0.5*(max(xi)-min(xi);
The slope T 6 is:
Kurtosis factor T 7 is:
the peak coefficient T 8 is:
the clearance coefficient T 9 is:
the shape factor T 10 is:
the impact factor T 11 is:
Where i denotes a vibration signal number.
Further, the edges of the path graph only exist between adjacent nodes; for each path graph, the connection between all nodes can be represented as an adjacency matrix corresponding to the path graph.
Further, chebGCN-LSTM prediction model consists of three ChebGCN convolutional layers, one BiLSTM layer and one FC (Fully Connected network) layer;
Chebyshev convolution operation is expressed as:
wherein F ε R n×m is a graph signal representing the input of a Chebyshev convolution operation, Representing the output of the chebyshev convolution operation, K representing the highest order of the chebyshev polynomial, K representing the order of the chebyshev polynomial,/>Representing a feature matrix,/>Representing a K-order chebyshev polynomial, and theta' k∈RK representing a chebyshev coefficient vector;
The output Y εR n×s of ChebGCN convolutional layers is represented as:
Wherein W ε R m×s represents the weight matrix learned by ChebGCN convolution layer;
LSTM is expressed as:
ft＝σ(Wf·[ht-1,yt]+bf)
Wherein e t,、ft and o t represent an input gate, a forget gate and an output gate of the LSTM at the current time, σ (·) and tanh (·) represent an activation function, y t represents information of the inflow of the LSTM circulation unit at the current time, h t-1 represents an output of the circulation unit at the previous time, c t represents a current time state, Indicating a candidate state in the input gate at the current time, h t indicating the output of the loop unit at the current time, and t indicating the current time.
Further, in the ChebGCN-LSTM predictive model after training, the remaining service life ratio is set as the label of the training sample, and the remaining service life ratio is the ratio of the remaining service life to the full service life of the rotary machine.
Further, inputting a ChebGCN-LSTM prediction model after the prediction path diagram is trained, and outputting a residual service life prediction value corresponding to each diagram node; one path diagram corresponds to a plurality of residual service life values, namely the prediction method is suitable for predicting the residual service life of the node level.
Another aspect of the present invention provides a system for predicting remaining useful life of a rotary machine, comprising:
A signal acquisition unit for acquiring vibration signals of the rotary machine in different life periods;
the signal characteristic extraction unit is used for extracting a time domain statistical index of the vibration signal;
The path diagram construction unit is used for constructing a path diagram based on the time domain statistical index so as to represent the degradation state of the rotary machine;
the prediction model construction unit is used for constructing a ChebGCN-LSTM prediction model based on the residual service lives ChebGCN and BiLSTM of the rotary machine, training the ChebGCN-LSTM prediction model and obtaining a trained ChebGCN-LSTM prediction model;
And the residual service life prediction unit is used for inputting a prediction path diagram constructed by the real-time vibration signal of the rotary machine to be predicted into a ChebGCN-LSTM prediction model after training so as to obtain a residual service life prediction value of the rotary machine to be predicted.
Another aspect of the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method for predicting remaining useful life of a rotary machine as described above.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
(1) Based on the graph theory, the invention constructs the path graph (non-Euclidean data) based on the time domain statistical index, and compared with the traditional Euclidean data, the invention can effectively represent the time dependence and the space dependence between signals and comprehensively represent the degradation state of the rotary machine;
(2) The invention provides a ChebGCN-LSTM network, which utilizes ChebGCN and LSTM to respectively mine space information and time information of features in a path diagram, and overcomes the defect and the defect of the traditional ChebGCN on the global time correlation of a characterization signal;
(3) Compared with the traditional residual service life prediction method, the method has stronger performance and higher accuracy, and proves that the method is feasible and has more superiority in residual service life prediction.
Drawings
FIG. 1 is a flow chart of a method for predicting remaining useful life of a rotary machine according to the present invention;
FIG. 2 is a schematic diagram of a path diagram construction process according to the present invention;
FIG. 3 is a schematic diagram of a ChebGCN-LSTM prediction model provided by the present invention;
FIG. 4 is a root mean square error diagram comparing different residual life prediction methods according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of average absolute errors comparing different residual life prediction methods according to an embodiment of the present invention;
Fig. 6 is a schematic diagram showing a comparison of a straight line fit between a predicted value and a true value of remaining service life according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
The invention provides a method for predicting the residual service life of rotary machinery, which comprises the steps of firstly extracting a vibration signal time domain statistical index, and constructing a path diagram as an input characteristic, wherein nodes of the path diagram represent signals, and edge connection represents time sequence; compared with the traditional Euclidean data, the path diagram can effectively represent the time and space dependence of signals, and the degradation state of the rotary machine can be more comprehensively represented; and excavating a complex mapping relation between the input characteristics and the residual service life by using ChebGCN-LSTM prediction model, and effectively improving the prediction precision of the residual service life of the rotary machine.
Fig. 1 is a flowchart of a method for predicting remaining service life of a rotary machine according to the present invention, which specifically includes the following steps:
S1: obtaining vibration signals of the rotary machine in different life periods, and dividing the vibration signals into a training sample and a test sample;
specifically, vibration signals of the rotary machine under different service lives are obtained through a single accelerometer in the horizontal direction, and sample points containing different residual service life information are obtained.
S2: extracting the vibration signal time domain statistical index obtained in the step S1;
Specifically, given a vibration signal x= (X 1,x2,…,xm), where m is the data length,
The mean T 1 is calculated as:
standard deviation T 2 was calculated as:
Square root amplitude T 3 is calculated as:
Root mean square T 4 was calculated as:
peak-to-peak value T 5 is calculated as:
T5＝0.5*(max(xi)-min(xi))
the slope T 6 is calculated as:
Kurtosis factor T 7 is calculated as:
The peak coefficient T 8 is calculated as:
the clearance coefficient T 9 is calculated as:
The shape factor T 10 is calculated as:
the impact factor T 11 is calculated as:
s3: taking the vibration signal as a node, taking a time domain statistical index as a node characteristic, arranging the nodes according to the sequence of sampling time, and constructing a path diagram by connecting edges between each node and adjacent nodes before and after the node so as to represent the degradation state of the rotary machine;
Specifically, the path graph is represented as g= { V, E, a, F }, where V represents the path graph node, E represents the edge connection between two nodes, F represents the node characteristics, and a represents the adjacency matrix;
In the path diagram, each signal sample is regarded as a node, and the node characteristics are represented by the 11 time domain statistical indexes extracted in the step S2;
The nodes are arranged according to the sequence of sampling time, and the edges of the path diagram only exist between adjacent nodes and are used for representing the time sequence of signal samples;
For each path graph, the connection condition between all nodes can be expressed as an adjacency matrix corresponding to the path graph;
For dataset d= { (X 1,Z1),(X2,Z2),…,(Xn,Zn) }, a path graph dataset G' = { G 1,G2,…,GN } is constructed with these signals, where one path graph contains M signals. If the number of signals in the last path graph is less than M, the last sample spread dataset is replicated until the number of signals reaches M.
The construction process from the original vibration signal to the path diagram is shown in fig. 2.
S4: constructing and training a ChebGCN-LSTM prediction model of the residual service life of the rotary machine based on ChebGCN and BiLSTM;
specifically, the ChebGCN-LSTM prediction model consists of three ChebGCN convolution layers, one BiLSTM layer and one FC layers, and the structure of the prediction model is shown in FIG. 3.
Chebyshev convolution operation is expressed as
Wherein F ε R n×m is a graph signal representing the input of a Chebyshev convolution operation,Representing the output of the chebyshev convolution operation, K representing the highest order of the chebyshev polynomial, K representing the order of the chebyshev polynomial,/>Representing a feature matrix,/>Representing a K-th order chebyshev polynomial, θ' k∈RK representing a chebyshev coefficient vector.
The output Y εR n×s of ChebGCN convolutional layers is represented as:
w ε R m×s represents the weight matrix learned by ChebGCN convolutional layers.
LSTM is expressed as:
ft＝σ(Wf·[ht-1,yt]+bf)
Wherein e t,、ft and o t represent an input gate, a forget gate and an output gate of the LSTM at the current time, σ (·) and tanh (·) represent an activation function, y t represents information of the inflow of the LSTM circulation unit at the current time, h t-1 represents an output of the circulation unit at the previous time, c t represents a current time state, Indicating a candidate state in the input gate at the current time, h t indicating the output of the loop unit at the current time, and t indicating the current time.
Table 1 includes parameters for setting ChebGCN-LSTM predictive models:
When the ChebGCN-LSTM prediction model is trained, the remaining service life ratio is set as a training sample label, and represents the ratio of the remaining service life of the rotary machine at the current sampling moment to the actual full service life.
S5: inputting ChebGCN-LSTM predictive model into the path diagram constructed by the sample to be tested, and processing the sample to be tested to obtain the residual service life predictive value of the monitored object.
Specifically, the sample to be tested also needs to be preprocessed, the original vibration signals in different life periods are processed in the steps S1-S3 to obtain a path diagram, the path diagram is substituted into the residual service life prediction model constructed in the step S4, the prediction model processes the path diagram, and the residual service life ratio prediction value of the current moment corresponding to each diagram node is output.
The q-th path graph constructed by the sample to be tested is denoted as G q＝{Vq,Eq,Aq,Fq }:
Wherein, Weight matrix respectively representing training of first, second and third ChebGCN convolution layers of predictive model,/>, andRespectively representing the output of the first, second and third ChebGCN convolution layers of the prediction model; construction of a novel feature matrix/>Inputting the F ChebGCN∈RMN×g into the BiLSTM layer to obtain an output characteristic matrix F BiLSTM∈RMN×j; it is input to the FC layer and the resulting output F FC∈RMN×1 represents the residual life ratio prediction.
The utility of the present invention is further verified as follows.
In order to verify the effectiveness of the residual service life prediction method provided by the invention, the effectiveness of the XJTU-SY rolling bearing accelerated life test data set provided by the Western A traffic university is used for verification. Bearing operation fault data are collected in a bearing accelerated degradation test, wherein bearing failure types comprise inner ring abrasion, outer ring fracture and the like. A single accelerometer was used to acquire vibration signals in the horizontal direction at a sampling frequency of 25.6KHz with 1 minute sampling intervals of 1.28 seconds each, and the acquired sample data segment length contained 32768 data points.
To verify the effectiveness of the proposed method, four sets of experiments were performed in this example;
table 2 includes detailed experimental settings for this example:
Performance evaluation was performed using two evaluation indexes of RMSE (root mean square error, root error) and MAE (mean absolute error, both absolute errors), and the calculation formula was as follows.
Where n is the number of samples,And/>Respectively a predicted value and a true value of the remaining service life ratio.
The method comprises the following specific steps:
(1) Data acquisition and time domain index extraction
Rotating machine bearings 1-1 had a total of 123 data samples at (2100 rpm,12 kN) representing a full life cycle of 123 minutes. Similarly, rotary machines bear1_2, bear1_3, and bear1_5 have 161, 158, and 52 data samples, respectively, at (2100 rpm,12 kn), representing full life cycles of 161, 158, and 52 minutes, respectively.
Similarly, rotating machines bearing2_1 and bearing2_4 have 161 and 42 data samples, respectively, at (2250 rpm,11 kn), representing a full life cycle of 161 and 42 minutes, respectively.
In the experiment, the time domain statistical index is extracted from each data sample, and 11 time domain statistical indexes can be obtained.
(2) Constructing a path graph to characterize a state of degradation of a rotating machine
In the path diagrams, each signal sample is regarded as a node, and 30 data samples are sequentially selected according to the sampling time sequence in each path diagram to form 30 nodes; representing node characteristics by using 11 time domain statistical indexes extracted in the step (1); the edges of the path graph exist only between adjacent nodes to represent the time sequence of the signal samples, and the connection condition between all the nodes can be represented as an adjacent matrix corresponding to the path graph.
(3) Residual service life prediction model construction
Constructing a ChebGCN-LSTM prediction model of the residual service life of the rotary machine based on ChebGCN and BiLSTM; the ChebGCN-LSTM predictive model consists of three ChebGCN convolutional layers, one BiLSTM layer and one FC layer.
When the ChebGCN-LSTM prediction model is trained, the remaining service life ratio is set as a training sample label, and represents the ratio of the remaining service life of the rotary machine at the current sampling moment to the actual full service life.
(4) Residual life prediction
And (3) processing the original vibration signals in different life periods through the steps (1) and (2) to obtain a path diagram, substituting the path diagram into the residual service life prediction model constructed in the step (3), processing the path diagram by the prediction model, and outputting a residual service life ratio prediction value at the current moment corresponding to each diagram node.
The present invention is compared with three popular residual life prediction methods, including long-short-term memory network LSTM, convolutional neural network CNN, and convolutional-long-term memory network CNN-LSTM. The residual service life prediction results of the four methods are shown in fig. 4 and 5, and it can be seen that RMSE and MAE of all the test set comparison methods are not lower than 0.2. However, the RMSE of the method provided in this example is less than 0.15 and the mae is less than 0.1, indicating that the method is feasible in residual life prediction and has better prediction performance.
Fig. 6 is a comparison of the predicted values of the embodiment with the true value fitting straight line, and it can be seen that the predicted values are distributed around the true value.
Based on the graph theory, the invention constructs the path graph (non-Euclidean data) based on the time domain statistical index, and compared with the traditional Euclidean data, the path graph can effectively represent the time dependence and the space dependence between signals and comprehensively represent the degradation state of the rotating machinery. The invention has good practicability by using ChebGCN and LSTM to respectively mine the space information and the time information of the characteristics in the path diagram and using the prediction of the residual service life of the rolling bearing of the XJTU-SY dataset as an example verification.
Another aspect of the present invention provides a system for predicting remaining useful life of a rotary machine, comprising:
the signal acquisition unit is used for acquiring vibration signals of the rotary machine in different life periods and dividing the vibration signals into a training sample and a sample to be tested;
The signal characteristic extraction unit is used for extracting a vibration signal time domain statistical index;
The path diagram construction unit is used for constructing a path diagram based on the time domain statistical index so as to represent the degradation state of the rotary machine;
And the prediction model construction unit is used for constructing a ChebGCN-LSTM prediction model based on the ChebGCN and BiLSTM residual service lives of the rotary machine and training ChebGCN-LSTM prediction models by using the path diagram constructed by the training samples.
And the residual service life prediction unit is used for inputting a prediction path diagram constructed by the real-time vibration signal of the rotary machine to be predicted into the ChebGCN-LSTM prediction model after training, processing the sample to be detected and outputting a residual service life ratio predicted value of the current moment of the monitored object.
Another aspect of the present invention provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method for predicting remaining useful life of a rotary machine as described above.
It will be readily appreciated by those skilled in the art that the foregoing is merely a preferred embodiment of the invention and is not intended to limit the invention, any modifications may be made within the spirit and principles of the invention: equivalent substitutions and modifications and the like are intended to be included in the scope of the present invention.
Claims (8)
1. A method for predicting remaining service life of a rotary machine, comprising:
Acquiring a real-time vibration signal of the rotary machine to be predicted, taking the vibration signal as a prediction sample, and constructing a prediction path diagram according to the prediction sample;
inputting the predicted path diagram into a ChebGCN-LSTM prediction model after training to obtain a predicted value of the residual service life of the rotary machine to be predicted;
The construction method of the path diagram comprises the following steps:
Extracting time domain statistical indexes of samples, regarding vibration signals as nodes, regarding time domain statistical indexes as node characteristics, regarding the sequence of life time periods of the vibration signals as edges, and connecting each node with adjacent nodes before and after the node by edges to form a path diagram;
Wherein, the path graph is expressed as G= { V, E, A, F }, wherein V represents the path graph node, E represents the edge connection between two nodes, F represents the node characteristic, A represents the adjacency matrix;
the ChebGCN-LSTM prediction model consists of three ChebGCN convolution layers, one BiLSTM layer and one FC layer;
Chebyshev convolution operation is expressed as: ；
Wherein, Is a graph signal representing the input of chebyshev convolution operation,/>Representing the output of the chebyshev convolution operation, K representing the highest order of the chebyshev polynomial, K representing the order of the chebyshev polynomial,/>Representing a feature matrix,/>Representing a K-order chebyshev polynomial,/>Representing chebyshev coefficient vectors;
ChebGCN output of convolutional layer Expressed as:
；
Wherein, Representing ChebGCN the weight matrix learned by the convolutional layer;
LSTM is expressed as:
；
Wherein e t,、ft and o t represent the input gate, the forget gate and the output gate of the current time LSTM respectively, And tanh (·) represents an activation function, y t represents information of the current time flowing into the LSTM circulation unit, h t-1 represents output of the circulation unit at the previous time, c t represents the current time state,/>Indicating a candidate state in the input gate at the current time, h t indicating the output of the loop unit at the current time, and t indicating the current time.
2. The method for predicting remaining service life according to claim 1, wherein the trained ChebGCN-LSTM prediction model acquisition method is as follows:
obtaining vibration signals of the rotary machine in different life periods, taking the vibration signals as training samples, and constructing a training path diagram according to the training samples;
And constructing a ChebGCN-LSTM prediction model based on ChebGCN and BiLSTM residual service life of the rotary machine, inputting a training path diagram into the ChebGCN-LSTM prediction model for training, and obtaining a trained ChebGCN-LSTM prediction model.
3. The method of claim 1, wherein the time domain statistical indicators are a mean value T 1, a standard deviation T 2, a square root amplitude T 3, a root mean square T 4, a peak-to-peak value T 5, a slope T 6, a kurtosis factor T 7, a peak coefficient T 8, a clearance coefficient T 9, a shape factor T 10, and an impact factor T 11;
Giving vibration signals as X= (X 1, x2, …, xm), wherein m is the number of the vibration signals;
The mean T 1 is: ；
The standard deviation T 2 is: ；
square root amplitude T 3 is: ；
root mean square T 4 is: ；
The peak-to-peak value T 5 is: ；
The slope T 6 is: ；
Kurtosis factor T 7 is: ；
the peak coefficient T 8 is: ；
the clearance coefficient T 9 is: ；
the shape factor T 10 is: ；
the impact factor T 11 is: ；
Wherein, Indicating the vibration signal number.
4. The remaining useful life prediction method as claimed in claim 1, wherein the edges of the path graph exist only between adjacent nodes; for each path graph, the connection between all nodes can be represented as an adjacency matrix corresponding to the path graph.
5. The method for predicting remaining service life according to claim 1, wherein in the ChebGCN-LSTM prediction model after training, a remaining service life ratio is set as a label of a training sample, and the remaining service life ratio is a ratio of remaining service life to full service life of the rotating machine.
6. The method for predicting remaining service life according to claim 1, wherein the predicted path graph inputs a trained ChebGCN-LSTM prediction model, and outputs a predicted value of remaining service life corresponding to each graph node; one path diagram corresponds to a plurality of residual service life values, namely the prediction method is suitable for predicting the residual service life of the node level.
7. A system for predicting remaining useful life of a rotary machine, comprising the following elements:
A signal acquisition unit for acquiring vibration signals of the rotary machine in different life periods;
the signal characteristic extraction unit is used for extracting a time domain statistical index of the vibration signal;
The path diagram construction unit is used for constructing a path diagram based on the time domain statistical index so as to represent the degradation state of the rotary machine;
the prediction model construction unit is used for constructing a ChebGCN-LSTM prediction model based on the residual service lives ChebGCN and BiLSTM of the rotary machine, training the ChebGCN-LSTM prediction model and obtaining a trained ChebGCN-LSTM prediction model;
The residual service life prediction unit is used for inputting a prediction path diagram constructed by the real-time vibration signal of the rotary machine to be predicted into a ChebGCN-LSTM prediction model after training so as to obtain a residual service life prediction value of the rotary machine to be predicted;
The path diagram construction unit is specifically configured to: extracting time domain statistical indexes of samples, regarding vibration signals as nodes, regarding time domain statistical indexes as node characteristics, regarding the sequence of life time periods of the vibration signals as edges, and connecting each node with adjacent nodes before and after the node by edges to form a path diagram;
Wherein, the path graph is expressed as G= { V, E, A, F }, wherein V represents the path graph node, E represents the edge connection between two nodes, F represents the node characteristic, A represents the adjacency matrix;
the ChebGCN-LSTM prediction model consists of three ChebGCN convolution layers, one BiLSTM layer and one FC layer;
Chebyshev convolution operation is expressed as: ；
Wherein, Is a graph signal representing the input of chebyshev convolution operation,/>Representing the output of the chebyshev convolution operation, K representing the highest order of the chebyshev polynomial, K representing the order of the chebyshev polynomial,/>Representing a feature matrix,/>Representing a K-order chebyshev polynomial,/>Representing chebyshev coefficient vectors;
ChebGCN output of convolutional layer Expressed as:
；
Wherein, Representing ChebGCN the weight matrix learned by the convolutional layer;
LSTM is expressed as:
；
Wherein e t,、ft and o t represent the input gate, the forget gate and the output gate of the current time LSTM respectively, And tanh (·) represents an activation function, y t represents information of the current time flowing into the LSTM circulation unit, h t-1 represents output of the circulation unit at the previous time, c t represents the current time state,/>Indicating a candidate state in the input gate at the current time, h t indicating the output of the loop unit at the current time, and t indicating the current time.
8. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the remaining service life prediction method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210166709.3A CN114580101B (en) | 2022-02-23 | 2022-02-23 | Method and system for predicting residual service life of rotary machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210166709.3A CN114580101B (en) | 2022-02-23 | 2022-02-23 | Method and system for predicting residual service life of rotary machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114580101A CN114580101A (en) | 2022-06-03 |
| CN114580101B true CN114580101B (en) | 2024-06-14 |
Family
ID=81770108
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210166709.3A Active CN114580101B (en) | 2022-02-23 | 2022-02-23 | Method and system for predicting residual service life of rotary machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114580101B (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11000236B2 (en) * | 2017-03-24 | 2021-05-11 | Medtronic Minimed, Inc. | Patient data management systems and conversational interaction methods |
| US11340306B2 (en) * | 2017-11-16 | 2022-05-24 | Semiconductor Energy Laboratory Co., Ltd. | Lifetime estimation device, lifetime estimation method, and abnormality detection method of secondary battery |
| CN110377984B (en) * | 2019-07-02 | 2023-08-18 | 中国科学院深圳先进技术研究院 | Industrial equipment residual effective life prediction method and system and electronic equipment |
| CN112129534B (en) * | 2020-09-17 | 2021-10-29 | 上海交通大学 | Method and system for predicting service life of main shaft bearing of machine tool |
| CN113884937B (en) * | 2021-11-22 | 2022-08-30 | 江南大学 | Lithium ion battery residual life prediction method and system based on decomposition integration strategy |
-
2022
- 2022-02-23 CN CN202210166709.3A patent/CN114580101B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| Deep-convolution-based LSTM network for remaining useful life prediction;Ma M et al.;《IEEE Trans.Ind.Inform》;20211231;第17卷;全文 * |
| 基于MSCNN-LSTM的滚动轴承剩余寿命预测方法;胡城豪;胡昌华;司小胜;杜党波;高旭东;;中国测试;20200930(第09期);第103-110页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114580101A (en) | 2022-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110361176B (en) | Intelligent fault diagnosis method based on multitask feature sharing neural network | |
| CN112418277B (en) | Method, system, medium and equipment for predicting residual life of rotating machine parts | |
| CN110703057B (en) | Power equipment partial discharge diagnosis method based on data enhancement and neural network | |
| CN111237134A (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
| CN109858140B (en) | Fault diagnosis method for water chilling unit based on information entropy discrete Bayesian network | |
| CN113505655A (en) | Bearing fault intelligent diagnosis method for digital twin system | |
| CN111459144A (en) | Airplane flight control system fault prediction method based on deep cycle neural network | |
| CN112033463A (en) | Nuclear power equipment state evaluation and prediction integrated method and system | |
| CN117034143B (en) | Distributed system fault diagnosis method and device based on machine learning | |
| CN114444582A (en) | Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network | |
| CN115146718A (en) | Depth representation-based wind turbine generator anomaly detection method | |
| Oh et al. | Fault detection for lubricant bearing with CNN | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN116295948A (en) | Abnormality detection method, system and storage medium of industrial temperature sensor in large temperature difference environment | |
| CN111126477A (en) | Learning and reasoning method of hybrid Bayesian network | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN114580101B (en) | Method and system for predicting residual service life of rotary machine | |
| CN116451121A (en) | Switch fault diagnosis method, system, equipment and medium | |
| CN116793666A (en) | Wind turbine generator system gearbox fault diagnosis method based on LSTM-MLP-LSGAN model | |
| Zhang et al. | A flexible monitoring framework via dynamic-multilayer graph convolution network | |
| Jing et al. | Two-layer PSDG based fault diagnosis for wind turbines | |
| CN113191556A (en) | Nuclear power Loca event fault prediction and diagnosis method | |
| Xiong et al. | Rolling bearing fault diagnosis based on multi-modal variational autoencoders | |
| Chen et al. | A multimode anomaly detection method based on oc-elm for aircraft engine system | |
| Guo et al. | Rolling bearing fault diagnosis based on the combination of improved deep convolution network and gated recurrent unit |
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 |