CN114462459A - Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model - Google Patents
Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model Download PDFInfo
- Publication number
- CN114462459A CN114462459A CN202210377316.7A CN202210377316A CN114462459A CN 114462459 A CN114462459 A CN 114462459A CN 202210377316 A CN202210377316 A CN 202210377316A CN 114462459 A CN114462459 A CN 114462459A
- Authority
- CN
- China
- Prior art keywords
- fault
- hydraulic machine
- 1dcnn
- lstm
- network model
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000003745 diagnosis Methods 0.000 title claims abstract description 29
- 230000015654 memory Effects 0.000 claims abstract description 14
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 24
- 239000013598 vector Substances 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 15
- 238000005457 optimization Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 13
- 230000002159 abnormal effect Effects 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 241000283153 Cetacea Species 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims 1
- 230000006403 short-term memory Effects 0.000 claims 1
- 238000013136 deep learning model Methods 0.000 abstract description 3
- 230000001419 dependent effect Effects 0.000 abstract description 3
- 230000002123 temporal effect Effects 0.000 abstract description 3
- 238000013461 design Methods 0.000 abstract description 2
- 239000000284 extract Substances 0.000 abstract description 2
- 210000002569 neuron Anatomy 0.000 description 9
- 210000004027 cell Anatomy 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 229910000831 Steel Inorganic materials 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 229910001339 C alloy Inorganic materials 0.000 description 1
- 229910000975 Carbon steel Inorganic materials 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 239000000956 alloy Substances 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000019637 foraging behavior Effects 0.000 description 1
- 238000005242 forging Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model, which belongs to the field of hydraulic machine fault diagnosis methods, improves the precision, speed and the like of fault diagnosis, is different from a neural network which artificially extracts features as input, and the 1DCNN adopts an original time domain signal as input, thereby greatly simplifying the design and application of diagnosis; spatial and temporal features are extracted through a one-dimensional convolution long-short term memory (1 DCNN-LSTM) network, and features in a larger range are effectively extracted to improve the fault diagnosis rate; in the proposed solution, the LSTM layer follows a one-dimensional convolutional neural network (1 DCNN), which makes the number of time steps in the LSTM layer much smaller than the length of the input segment. Therefore, the computational complexity of the LSTM layer is greatly reduced; in order to overcome the problem that a deep learning model is very dependent on professional knowledge and manual debugging, an IWOA algorithm is used for solving the problem of automatically selecting hyper-parameters in a 1DCNN-LSTM model.
Description
Technical Field
The invention relates to the field of hydraulic machine fault diagnosis methods, in particular to a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model.
Background
With the continuous development and progress of the industrial informatization technology, the hydraulic machine is developed more and more towards the direction of precision, complication and intellectualization, which brings new challenges to the fault diagnosis and management of the hydraulic machine while creating convenience for the production and life of human beings, and the existing hydraulic technology becomes one of the key technologies in the industrial fields of all countries in the world, according to incomplete statistics, more than 95% of mechanical equipment adopts the hydraulic technology and device, the hydraulic machine is a core device which must be adopted for processing and forging various high-strength steels, carbon steels and alloy steels, is widely used in equipment in the heavy industrial fields of aerospace, steels, large-scale bearing parts, nuclear industry, military, ships, lifting machines, artificial boards and the like, is key equipment in the national economic strut industry of energy, petroleum, metallurgy and the like, and some are strategic equipment required by industrial systems and national defense, the hydraulic press is a basic device for developing large military equipment and large industrial equipment in China, marks the comprehensive production capacity and the technical development level of the country, is of vital importance in reliability and safe operation, and once the hydraulic press fails, the healthy operation of the whole production system is influenced, so that serious production events are caused, and even more, serious loss of life and property is caused.
The hydraulic machine is essentially a system integrating electromechanical control and hydraulic control, the hydraulic machine has the characteristics of concealment, staggering, randomness, difference and the like in fault, and the fault shutdown not only reduces the production efficiency of enterprises, but also causes huge economic loss. In the absence of effective fault diagnosis means, technicians usually perform fault location by using isolation methods, logic analysis methods, component swapping methods, and the like. However, even experienced technicians require a long time to troubleshoot. Therefore, the signal processing method based on mathematical analysis is widely applied to the fault diagnosis of the hydraulic machine in the early stage. Although the methods have remarkable effect on some diagnostic problems due to the diversity and complexity of faults, the feature extraction methods mainly rely on expert knowledge, have high threshold values, are very complex and time-consuming in operation, and adopt the 1DCNN-LSTM network to extract spatial and temporal features, so as to effectively extract a wider range of features and improve the fault diagnosis rate; and greatly reduces the computational complexity of the LSTM layer.
Disclosure of Invention
The invention provides a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model to solve the problems.
In order to achieve the purpose, the invention adopts the following technical scheme:
a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model comprises the following steps:
s1: collecting a fault signal of the hydraulic machine;
s2: making a model training data set (a training set and a testing set), classifying and marking;
s3: extracting the fault signal characteristics of the hydraulic machine through a one-dimensional convolution long-short term memory (1 DCNN-LSTM) network;
s4: optimizing 1DCNN-LSTM network hyper-parameters by adopting an Improved Whale Optimization Algorithm (IWOA);
s5: utilizing Softmax as a classifier, and training and verifying the deep network model;
s6: and inputting the measured signal into a model to obtain fault type information.
Preferably, the hydraulic machine fault signals collected in the step S1 include a hydraulic pressure signal, a hydraulic flow signal, a hydraulic temperature signal, a solenoid valve power failure condition and a vibration signal in a fault mode.
Preferably, the hydraulic pressure signal is acquired by a PPM-T322H pressure sensor, the hydraulic flow signal is acquired by an FT-330 type sensor, and the vibration signal is acquired by an SG2000 vibration sensor.
Preferably, the fault signals under the various fault modes collected in step S1 are not less than 100 groups, where 80% of the fault signals are used as a training set, and 20% of the fault signals are used as a verification set, and the training sets are respectively labeled.
Preferably, the one-dimensional convolution long-short term memory (1 DCNN-LSTM) network of step S3 includes a one-dimensional convolution neural network and a long-short term memory neural network; the one-dimensional convolutional neural network is not less than 1 convolutional module, and the convolutional module comprises a convolutional layer, a pooling layer and batch standardization; inputting the fault signal into a 1DCNN-LSTN model, learning the characteristic quantity of the fault signal through the convolution layer, realizing the reconstruction of the characteristic quantity of the fault through the pooling layer, and carrying out standardization processing on each hidden layer through batch standardization; inputting the reconstructed characteristic quantity into an LSTM network, and performing secondary extraction on the characteristic time sequence information through the LSTM network; and inputting the feature vector processed by the LSTM into a Global Average Poolling layer to realize the dimension reduction of the feature vector.
Preferably, the hyper-parameters to be optimized by the 1DCNN-LSTM algorithm in step S4 are mapped into the IWOA optimization algorithm, and the obtained classification precision of the test set is used as a fitness function of the IWOA optimization algorithm.
Preferably, the Softmax classifier of step S5 classifies the failure into 5 modes, i.e., abnormal downward sliding, no lifting of the press, no pressurization, no pressure maintaining, and abnormal pressure maintaining, which are respectively represented by F1, F2, F3, F4, and F5 in a one-to-one correspondence.
Preferably, the loss function for training and verifying the deep network model in step S5 is as follows:
wherein,
is an event
The probability of (a) of (b) being,
is an event
The amount of information of (a) is,
is the ith eigenvector after convolution processing.
The hydraulic machine fault diagnosis system comprises a hydraulic machine fault signal acquisition system, a fault identification system and a hydraulic machine control system. The hydraulic machine fault signal acquisition system comprises a PPM-T322H pressure sensor, an FT-330 type flow sensor and an SG2000 vibration sensor for acquisition.
Preferably, the fault identification system uses a PC or FPGA processing card as a processing module of the 1DCNN-LSTM network model, the input end of the fault identification system is connected with the hydraulic machine fault information acquisition system, and the output section is connected with the hydraulic machine control system.
Compared with the prior art, the invention provides a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model, which has the following beneficial effects:
1. the invention has the beneficial effects that: the accuracy, the speed and the like of fault diagnosis are improved, and different from a neural network with artificially extracted features as input, the 1DCNN adopts an original time domain signal as input, so that the design and the application of diagnosis are greatly simplified; spatial and temporal features are extracted through a one-dimensional convolution long-short term memory (1 DCNN-LSTM) network, and features in a larger range are effectively extracted to improve the fault diagnosis rate; in the proposed solution, the LSTM layer follows a one-dimensional convolutional neural network (1 DCNN), which makes the number of time steps in the LSTM layer much smaller than the length of the input segment. Therefore, the computational complexity of the LSTM layer is greatly reduced; in order to overcome the problem that a deep learning model is very dependent on professional knowledge and manual debugging, an IWOA algorithm is used for solving the problem of automatically selecting hyper-parameters in a 1DCNN-LSTM model.
Drawings
FIG. 1 is a flowchart of a method for diagnosing a fault of a hydraulic machine based on a 1DCNN-LSTM network model according to an embodiment of 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.
In the description of the present invention, it is to be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention.
Example 1:
a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model comprises the following steps:
s1: collecting a fault signal of the hydraulic machine;
s2: making a model training data set (a training set and a testing set), classifying and marking;
s3: extracting the fault signal characteristics of the hydraulic machine through a one-dimensional convolution long-short term memory (1 DCNN-LSTM) network;
s4: optimizing 1DCNN-LSTM network hyper-parameters by adopting an Improved Whale Optimization Algorithm (IWOA);
s5: utilizing Softmax as a classifier, and training and verifying the deep network model;
s6: and inputting the measured signal into a model to obtain fault type information.
The hydraulic machine fault signals collected in the step S1 include a hydraulic pressure signal, a hydraulic flow signal, a hydraulic temperature signal, a solenoid valve power failure condition, and a vibration signal in a fault mode.
The hydraulic pressure signal is collected through a PPM-T322H pressure sensor, the hydraulic flow signal is collected through an FT-330 type sensor, and the vibration signal is collected through an SG2000 vibration sensor.
The fault signals under various fault modes collected in step S1 are not less than 100 groups, wherein 80% of the fault signals are used as a training set, and 20% of the fault signals are used as a verification set, and the training sets are respectively labeled.
The one-dimensional convolution long-short term memory (1 DCNN-LSTM) network of the step S3 comprises a one-dimensional convolution neural network and a long-short term memory neural network; the one-dimensional convolutional neural network is not less than 1 convolutional module, and the convolutional module comprises a convolutional layer, a pooling layer and batch standardization; inputting the fault signal into a 1DCNN-LSTN model, learning the characteristic quantity of the fault signal through the convolution layer, realizing the reconstruction of the fault characteristic quantity through the pooling layer, and carrying out standardization processing on each hidden layer through batch standardization; inputting the reconstructed characteristic quantity into an LSTM network, and performing secondary extraction on the characteristic time sequence information through the LSTM network; the feature vector processed by the LSTM is input into a Global Average Pooling layer to realize the dimension reduction of the feature vector.
And step S4, mapping the hyper-parameters needing to be optimized by the 1DCNN-LSTM algorithm into the IWOA optimization algorithm, and using the obtained classification precision of the test set as a fitness function of the IWOA optimization algorithm.
The Softmax classifier of step S5 classifies the failure into 5 modes of abnormal downslide, no lifting of the press, no pressurization, no pressure holding, and abnormal pressure holding, which are respectively represented by F1, F2, F3, F4, and F5 in a one-to-one correspondence.
The loss function for training and verifying the deep network model in step S5 is as follows:
wherein,
is an event
The probability of (a) of (b) being,
is an event
The amount of information of (a) is,
is the ith eigenvector after convolution processing.
The hydraulic machine fault diagnosis system comprises a hydraulic machine fault signal acquisition system, a fault identification system and a hydraulic machine control system. The hydraulic machine fault signal acquisition system comprises a PPM-T322H pressure sensor, an FT-330 type flow sensor and an SG2000 vibration sensor for acquisition.
The fault recognition system uses a PC or an FPGA processing card as a processing module of the 1DCNN-LSTM network model, the input end of the fault recognition system is connected with the hydraulic machine fault information acquisition system, and the output section is connected with the hydraulic machine control system.
Example 2: based on example 1, but with the following differences:
a hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model comprises the following steps:
s1: collecting a fault signal of the hydraulic machine;
s2: making a model training data set (a training set and a testing set), classifying and marking;
s3: extracting the fault signal characteristics of the hydraulic machine through a one-dimensional convolution long-short term memory (1 DCNN-LSTM) network;
s4: optimizing 1DCNN-LSTM network hyperparameters by adopting an Improved Whale Optimization Algorithm (IWOA);
s5: utilizing Softmax as a classifier, and training and verifying the deep network model;
s6: and inputting the measured signals into a model to obtain fault type information.
The hydraulic machine fault signals collected in the step S1 include hydraulic pressure signals, hydraulic flow signals, hydraulic temperature signals, power failure conditions of the solenoid valve, and vibration signals of the hydraulic machine in fault modes such as abnormal downward sliding, no lifting of the hydraulic machine, no pressurization, no pressure maintaining, and abnormal pressure maintaining. The working process of the press comprises working states of descending, pressurizing, pressure maintaining, pressure releasing, lifting and the like of the press. Each working state corresponds to the power on and power off of different electromagnetic valves respectively, and simultaneously corresponds to the change of pressure and flow at different points of the system. When the hydraulic machine works, the hydraulic machine can run in a reciprocating mode in a normal state, when a certain element goes wrong, the hydraulic machine can enter a fault state, common faults of the hydraulic machine mainly include abnormal sliding, no lifting, no pressurization, no pressure maintaining, poor pressure maintaining and the like, and the fault reasons mainly include seal damage, valve blockage, electromagnetic coil burnout and the like. The hydraulic pressure signal is collected by a PPM-T322H pressure sensor, the hydraulic flow signal is collected by an FT-330 type sensor, and the vibration signal is collected by an SG2000 vibration sensor. Collecting not less than 100 groups of fault signals under various fault modes, wherein 80% of fault signals are used as a training set, 20% of fault signals are used as a verification set, and the training set is respectively marked.
Step S3, the one-dimensional convolution long-short term memory (1 DCNN-LSTM) network comprises a one-dimensional convolution neural network and a long-short term memory neural network; the one-dimensional convolutional neural network is not less than 1 convolutional module, and the convolutional module comprises a convolutional layer, a pooling layer and batch standardization; inputting the fault signal into a 1DCNN-LSTN model, learning the characteristic quantity of the fault signal through the convolution layer, realizing the reconstruction of the fault characteristic quantity through the pooling layer, and carrying out standardization processing on each hidden layer through batch standardization; inputting the reconstructed characteristic quantity into an LSTM network, and performing secondary extraction on the characteristic time sequence information through the LSTM network; and inputting the feature vector processed by the LSTM into a Global Average Poolling layer to realize the dimension reduction of the feature vector.
The convolution layer is mainly composed of a plurality of convolution kernels, the convolution kernels have the characteristics of local perception and parameter sharing, and model parameters can be greatly reduced while various feature expressions are learned. Since the input of 1DCNN is a one-dimensional vector, the convolution kernel of the network is also one-dimensional, and the one-dimensional convolution operation can be expressed as
In the formula:
input and bias for the kth neuron of the l layer respectively,
is a convolution kernel between the ith neuron of the l-1 layer and the kth neuron of the l layer,
is the output of the ith neuron at layer l-1,
the number of the first-1 layer neurons,
is a one-dimensional convolution operation.
To increase the non-linearity of the network, an activation function is typically applied after each convolution operation. In order to prevent gradient disappearance and accelerate convergence of the network, the activation function adopts a modified linear unit (ReLU) with the expression of
The final output of each neuron in the convolutional layer is therefore:
the pooling layer adopts maximum pooling, namely taking the maximum value in an adjacent area of a certain position as the final output of the position:
in the formula: h is the convolution kernel width;
the pooling layer reduces the size of the feature map and the parameter amount of the network by using the overall statistical features of the regions adjacent to a certain location as the output of the network at the location, and at the same time network overfitting can be effectively avoided,
batch normalization normalizes each hidden layer such that each layer has the same gaussian distribution with mean 0 and variance 1:
in the formula:
、
mean and standard deviation of batch (batch) data, respectively;
to avoid infinite decimals with denominators of zero;
、
respectively carrying out input before batch normalization and intermediate output of batch normalization on the ith neuron;
、
scale transformation factors and offset factors introduced for recovering network expression capacity are obtained by network learning;
and (4) outputting the final output after neuron batch standardization.
Batch standardization can avoid internal covariate change and gradient dispersion of each layer of neurons, and accelerate the network convergence speed;
the LSTM network architecture deletes or adds information to the cell state by introducing a "gated" structure. The output of the sigma layer is a value of 0-1, which represents how much information can flow through the sigma layer. Neither of 0 and 1 indicates that both cannot pass. LSTM controls the cell state by three gates, called forgetting gate, input gate and output gate, respectively. Forget to check through door (ft)
And
information to output a vector between 0-1, the 0-1 values within the vector representing the state of the cell
How much information is retained or discarded. 0 means no reservation and 1 means both reservations.
Input gate (it) to decide update
And
which information of, then utilize
And
obtaining new candidate cell information through Tanh layer
. The update rule is as follows:
by forgetting to remember the door (f)t) selecting a part of the forgotten old cell information, and adding candidate cell information through the input gate
Part of which obtains new cellular information
. The update rule is as follows:
the output gate (ot) controls how much memory information will be used in the next phase of updating.
Wherein,
-an input at the time t,
-a weight matrix, b-a bias matrix,
-the candidate vector at the time t,
-the update value at the time t,
all outputs of the t, t-1 time point model
The LSTM network architecture can overcome the disadvantages of extinction and explosion gradients in RNN models.
Step S4, mapping the hyper-parameters to be optimized by the 1DCNN-LSTM algorithm to the IWOA optimization algorithm, and using the obtained classification precision of the test set as the fitness function of the IWOA optimization algorithm
In the IWOA algorithm, the automatic super-parameter searching is controlled by setting the Number of the operations and the size of the Searchgents parameters of the IWOAAnd obtaining a global optimal solution, namely the network hyper-parameter of the CNN-LSTM. Assuming the size of the whale population is N, the search space is d-dimensional, and the position of the nth whale in the d-dimensional space can be expressed as
The location of the prey corresponds to the global optimal solution to the problem. By simulating whale foraging behavior, the bounding predation phase can be represented by the following mathematical model:
wherein t is the number of current iterations;
is an individual position vector;
as prey location vector (current optimal solution); a and D1Are coefficient vectors, respectively, and have:
wherein Max _ iter is the maximum number of iterations, and r is [0, 1 ]]The random number of (2);
for the control parameter of the current iteration t times, the control parameter is linearly decreased from 2 to 0 along with the increase of the iteration times, namely:
introducing a dynamic balance factor:
wherein,
in order to balance the factors, the method comprises the following steps of,
is a dynamic balance factor.
Creating a spiral equation between whale and prey positions
As follows:
wherein,
linearly increasing the control parameter of the current iteration t times from-1 to 0 along with the increase of the iteration times;
is [ -1, 1 [ ]]A random number in between; d2Is composed of
A coefficient vector of time; b is a constant defining the shape of a logarithmic spiral;
the whale optimization algorithm can overcome the defect that a deep learning model is very dependent on professional knowledge and manual debugging, so that the 1DCNN-LSTM model can select reasonable hyper-parameters while achieving high accuracy. And more accurate network hyper-parameters are obtained, so that more comprehensive measuring point signal data characteristics are obtained, and the accuracy of model fault diagnosis is improved.
The step S5Softmax classifier classifies the fault into 5 modes of abnormal gliding, no lifting of the press, no pressurization, no pressure maintaining and abnormal pressure maintaining, and is respectively represented by F1, F2, F3, F4 and F5.
The loss function for training and verifying the deep network model in step S5 is as follows:
wherein,
is an event
The probability of (a) of (b) being,
is an event
The amount of information of (a) is,
the ith characteristic vector is subjected to convolution processing;
in step S5, the deep convolutional network diagnostic performance is measured by cross validation using three performance indicators, namely accuracy, precision and recall. Comprehensively evaluating the classification effect of the model through the three indexes, and defining the classification effect as follows:
wherein TP represents the prediction of the positive class as the number of the positive classes; FN denotes predicting positive class as a negative class number; FP denotes the prediction of negative classes as positive class numbers; TN denotes predicting a negative class as a negative class number. Accuracy, recall, and accuracy vary from 0 to 1. The larger the value, the better the trained model fault diagnosis performance.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (10)
1. A hydraulic machine fault diagnosis method based on a 1DCNN-LSTM network model is characterized by comprising the following steps:
s1: collecting a fault signal of the hydraulic machine;
s2: making a model training data set, wherein the data set comprises a training set and a test set, and classifying and marking;
s3: extracting the fault signal characteristics of the hydraulic machine through a one-dimensional convolution long-term and short-term memory network;
s4: optimizing 1DCNN-LSTM network hyper-parameters by adopting an improved whale optimization algorithm;
s5: utilizing Softmax as a classifier, and training and verifying the deep network model;
s6: and inputting the measured signal into a model to obtain fault type information.
2. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 1, wherein the method comprises the following steps: the hydraulic machine fault signals collected in the step S1 include a hydraulic pressure signal, a hydraulic flow signal, a hydraulic temperature signal, a power failure condition of the solenoid valve, and a vibration signal in a fault mode.
3. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 2, wherein the method comprises the following steps: the hydraulic pressure signal is collected through a PPM-T322H pressure sensor, the hydraulic flow signal is collected through an FT-330 type sensor, and the vibration signal is collected through an SG2000 vibration sensor.
4. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 1, wherein the method comprises the following steps: the fault signals collected in the step S1 under various fault modes are not less than 100 groups, wherein 80% of the fault signals are used as a training set, and 20% of the fault signals are used as a verification set, and the training sets are respectively labeled.
5. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 1, wherein the method comprises the following steps: the one-dimensional convolution long-short term memory (1 DCNN-LSTM) network of the step S3 comprises a one-dimensional convolution neural network and a long-short term memory neural network; the one-dimensional convolutional neural network is not less than 1 convolutional module, and the convolutional module comprises a convolutional layer, a pooling layer and batch standardization; inputting the fault signal into a 1DCNN-LSTN model, learning the characteristic quantity of the fault signal through the convolution layer, realizing the reconstruction of the characteristic quantity of the fault through the pooling layer, and carrying out standardization processing on each hidden layer through batch standardization; inputting the reconstructed characteristic quantity into an LSTM network, and performing secondary extraction on the characteristic time sequence information through the LSTM network; the feature vector processed by the LSTM is input into a Global Average Pooling layer to realize the dimension reduction of the feature vector.
6. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 1, wherein the method comprises the following steps: and mapping the hyper-parameters needing to be optimized by the 1DCNN-LSTM algorithm to the IWOA optimization algorithm in the step S4, and using the obtained classification precision of the test set as a fitness function of the IWOA optimization algorithm.
7. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model according to claim 1, wherein the method comprises the following steps: the Softmax classifier of the step S5 classifies the failure into 5 modes, i.e., abnormal downward sliding, no lifting of the press, no pressurization, no pressure maintaining, and abnormal pressure maintaining, which are respectively represented by F1, F2, F3, F4, and F5 in a one-to-one correspondence.
8. The method for diagnosing the fault of the hydraulic machine based on the 1DCNN-LSTM network model as claimed in claim 1, wherein: the loss function for training and verifying the deep network model in step S5 is as follows:
wherein,
is an event
The probability of (a) of (b) being,
is an event
The amount of information of (a) is,
is the ith eigenvector after convolution processing.
9. The hydraulic machine fault diagnosis system applied to the hydraulic machine fault diagnosis method based on the 1DCNN-LSTM network model is characterized in that: the hydraulic machine fault signal acquisition system, the fault recognition system and the hydraulic machine control system are included; the hydraulic machine fault signal acquisition system comprises a PPM-T322H pressure sensor, an FT-330 type flow sensor and an SG2000 vibration sensor for acquisition.
10. The hydraulic machine fault diagnosis system based on the 1DCNN-LSTM network model according to claim 9, wherein: the fault recognition system uses a PC or an FPGA processing card as a processing module of the 1DCNN-LSTM network model, the input end of the fault recognition system is connected with the hydraulic machine fault information acquisition system, and the output section is connected with the hydraulic machine control system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210377316.7A CN114462459A (en) | 2022-04-12 | 2022-04-12 | Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210377316.7A CN114462459A (en) | 2022-04-12 | 2022-04-12 | Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114462459A true CN114462459A (en) | 2022-05-10 |
Family
ID=81416671
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210377316.7A Withdrawn CN114462459A (en) | 2022-04-12 | 2022-04-12 | Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114462459A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115099264A (en) * | 2022-05-26 | 2022-09-23 | 哈尔滨工程大学 | Ship part fault diagnosis method and device, computer and computer storage medium |
| CN115859090A (en) * | 2023-02-23 | 2023-03-28 | 华东交通大学 | Turnout fault diagnosis method and system based on 1DCNN-LSTM |
| CN118068820A (en) * | 2024-04-19 | 2024-05-24 | 四川航天电液控制有限公司 | Intelligent fault diagnosis method for hydraulic support controller |
| DE102023200194A1 (en) | 2023-01-11 | 2024-07-11 | Hawe Hydraulik Se | Method for automated, measurement data-based design of an electronic controller for a hydraulic system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139514A (en) * | 2021-05-14 | 2021-07-20 | 安徽三禾一信息科技有限公司 | Hydraulic system fault diagnosis method based on fuzzy neural network |
| CN113822139A (en) * | 2021-07-27 | 2021-12-21 | 河北工业大学 | Equipment fault diagnosis method based on improved 1DCNN-BilSTM |
-
2022
- 2022-04-12 CN CN202210377316.7A patent/CN114462459A/en not_active Withdrawn
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139514A (en) * | 2021-05-14 | 2021-07-20 | 安徽三禾一信息科技有限公司 | Hydraulic system fault diagnosis method based on fuzzy neural network |
| CN113822139A (en) * | 2021-07-27 | 2021-12-21 | 河北工业大学 | Equipment fault diagnosis method based on improved 1DCNN-BilSTM |
Non-Patent Citations (2)
| Title |
|---|
| YUANYUAN JIANG 等: "A Fault Feature Extraction Method for DC-DC Converters Based on Automatic Hyperparameter-Optimized One-Dimensional Convolution and Long Short-Term Memory Neural Networks", 《JOURNAL OF LATEX CLASS FILES》 * |
| 刘忠雨 等: "《深入浅出图神经网络 GNN原理解析》", 30 April 2020, 机械工业出版社 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115099264A (en) * | 2022-05-26 | 2022-09-23 | 哈尔滨工程大学 | Ship part fault diagnosis method and device, computer and computer storage medium |
| DE102023200194A1 (en) | 2023-01-11 | 2024-07-11 | Hawe Hydraulik Se | Method for automated, measurement data-based design of an electronic controller for a hydraulic system |
| CN115859090A (en) * | 2023-02-23 | 2023-03-28 | 华东交通大学 | Turnout fault diagnosis method and system based on 1DCNN-LSTM |
| CN118068820A (en) * | 2024-04-19 | 2024-05-24 | 四川航天电液控制有限公司 | Intelligent fault diagnosis method for hydraulic support controller |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114462459A (en) | Hydraulic machine fault diagnosis method based on 1DCNN-LSTM network model | |
| Li et al. | A directed acyclic graph network combined with CNN and LSTM for remaining useful life prediction | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| CN108363896B (en) | Fault diagnosis method for hydraulic cylinder | |
| Miao et al. | A novel real-time fault diagnosis method for planetary gearbox using transferable hidden layer | |
| CN108921230A (en) | Method for diagnosing faults based on class mean value core pivot element analysis and BP neural network | |
| CN112147432A (en) | BiLSTM module based on attention mechanism, transformer state diagnosis method and system | |
| CN113743016B (en) | Engine residual life prediction method based on self-encoder and echo state network | |
| CN111046961B (en) | Fault classification method based on bidirectional long-time and short-time memory unit and capsule network | |
| CN112504682A (en) | Chassis engine fault diagnosis method and system based on particle swarm optimization algorithm | |
| CN112734131A (en) | Fan blade icing state prediction method based on deep learning algorithm | |
| CN113989550A (en) | Electric vehicle charging pile operation state prediction method based on CNN and LSTM hybrid network | |
| CN115618732A (en) | Nuclear reactor digital twin key parameter autonomous optimization data inversion method | |
| CN113011248A (en) | Nuclear power plant fault diagnosis method based on LSTM | |
| CN114169091A (en) | Method for establishing prediction model of residual life of engineering mechanical part and prediction method | |
| Liu et al. | Deep & attention: A self-attention based neural network for remaining useful lifetime predictions | |
| CN113126489B (en) | Intelligent BIT design method of heavy-duty gas turbine control system based on CNN-GRU-BINN | |
| CN113673774A (en) | Aero-engine remaining life prediction method based on self-encoder and time sequence convolution network | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN115048873B (en) | Residual service life prediction system for aircraft engine | |
| CN114841063A (en) | Aero-engine residual life prediction method based on deep learning | |
| CN114818811A (en) | Aircraft engine rolling bearing fault diagnosis method based on twin network metric learning | |
| Zhou et al. | Remaining useful life prediction of aero-engine using CNN-LSTM and mRMR feature selection | |
| CN115753102A (en) | Bearing fault diagnosis method based on multi-scale residual error sub-domain adaptation | |
| Gao et al. | Fault detection of electric vehicle charging piles based on extreme learning machine algorithm |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220510 |