CN112699597A - Nuclear power starting water pump rolling bearing fault detection method and system - Google Patents
Nuclear power starting water pump rolling bearing fault detection method and system Download PDFInfo
- Publication number
- CN112699597A CN112699597A CN202011433136.3A CN202011433136A CN112699597A CN 112699597 A CN112699597 A CN 112699597A CN 202011433136 A CN202011433136 A CN 202011433136A CN 112699597 A CN112699597 A CN 112699597A
- Authority
- CN
- China
- Prior art keywords
- water pump
- nuclear power
- fault
- rolling bearing
- starting water
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005096 rolling process Methods 0.000 title claims abstract description 84
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 77
- 238000001514 detection method Methods 0.000 title claims description 19
- 238000003745 diagnosis Methods 0.000 claims abstract description 72
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 39
- 238000000034 method Methods 0.000 claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000004088 simulation Methods 0.000 claims abstract description 8
- 238000012544 monitoring process Methods 0.000 claims description 31
- 230000001133 acceleration Effects 0.000 claims description 25
- 230000004913 activation Effects 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 16
- 238000010586 diagram Methods 0.000 claims description 15
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000014509 gene expression Effects 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 4
- 230000003213 activating effect Effects 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 abstract description 47
- 238000002474 experimental method Methods 0.000 abstract description 3
- 230000007246 mechanism Effects 0.000 abstract description 3
- 238000004458 analytical method Methods 0.000 abstract description 2
- 238000011160 research Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 42
- 230000000875 corresponding effect Effects 0.000 description 25
- 230000036541 health Effects 0.000 description 16
- 238000011156 evaluation Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010892 electric spark Methods 0.000 description 2
- 238000003754 machining Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Acoustics & Sound (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Monitoring And Testing Of Nuclear Reactors (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a method and a system for detecting faults of a rolling bearing of a nuclear power starting water pump, wherein a bearing fault diagnosis model based on a convolutional neural network is established, a simulation experiment based on a control variable method is adopted to determine model parameters, the bearing fault diagnosis model is trained by utilizing original operation data of the rolling bearing of the nuclear power starting water pump and working condition labels corresponding to the original operation data based on the determined model parameters, the method is used for optimizing the parameters of the convolutional neural network based on experimental research, the model training efficiency is further improved, the links of internal mechanism analysis, Fourier transform and the like are skipped by utilizing the end-to-end characteristics of the convolutional neural network, the links directly act on the original data, the loss of original information is avoided, the fault diagnosis model is optimized by combining the experiment and the theory, the accurate maintenance of the rolling bearing of the nuclear power starting water pump is realized according to the fault diagnosis, further improving the operation efficiency of nuclear power enterprises.
Description
Technical Field
The invention relates to the technical field of intelligent operation and maintenance, in particular to a method and a system for detecting faults of a rolling bearing of a nuclear power starting water pump.
Background
The nuclear power is one of energy guarantees for the development of the economic society of China, the nuclear power starting water pump is mainly used for providing starting water supply of a steam generator under the normal operation condition and standby water supply when a booster pump/a main water supply pump fails, and the health condition of the nuclear power starting water pump is concerned about the safety and the economic benefit of the nuclear power station during the starting and shutdown. The rolling bearing is used as an important supporting component of the nuclear power unit starting water pump, and the health of the rolling bearing directly influences the performance, power, service life and the like of the starting water pump equipment.
At present, nuclear power enterprises mainly adopt a periodic maintenance strategy to maintain the reliability of important equipment such as a starting water pump and prevent major accidents. However, the periodic maintenance has the problems of difficult finding of potential defects, excessive maintenance of equipment and the like. Many devices designed to be extremely reliable fail unexpectedly well below their expected life, while others are subject to repair or even replacement while still operating healthily and reliably.
The rolling bearing monitoring system is a key problem affecting the economic safety of nuclear power enterprises, and aims to carry out real-time fault monitoring diagnosis and health assessment on a nuclear power starting water pump rolling bearing, realize intelligent decision of on-the-spot maintenance based on assessment information, effectively guarantee normal and stable operation of the nuclear power starting water pump and prevent major accidents.
As one of key technologies of intelligent operation and maintenance of a nuclear power starting water pump, the core of a bearing fault diagnosis technology is signal feature extraction and mode classification. The former is mainly used for extracting characteristics capable of reflecting fault states from bearing vibration signals, and the latter identifies bearing fault types from characteristic information. By extracting the features of the monitoring data, sometimes, in order to suppress fitting and reduce the calculation amount, dimension reduction operation is performed on the extracted features, and finally, mode classification is performed, so that bearing fault diagnosis is realized. In addition, a data-driven feature extraction technology is also widely applied gradually, as shown in fig. 8, a stacked pre-trained deep neural network performs a dimension reduction function by using pre-training of the network itself on the basis of feature extraction, so that mode classification is realized, and a diagnosis link is further simplified.
The key of the existing rolling bearing fault diagnosis technology lies in finding an optimal combination of feature extraction and classification algorithms to enable the fault recognition rate to reach the highest, but the following problems still exist at present:
in different environments, the universality of the algorithm cannot be guaranteed, and a fault diagnosis method special for a nuclear power starting water pump rolling bearing is not reported yet.
Secondly, the conventional bearing fault diagnosis technology needs to analyze the internal mechanism of the fault and extract key characteristic signals, so that the technical threshold is high and the difficulty is high.
The existing data-driven feature extraction method needs Fourier transform or wavelet transform, and has the possibility of losing important time domain features.
In addition, the current nuclear power generating unit mainly depends on engineering experience in maintenance scheme decision, the operation and maintenance management efficiency is not high, and related inventions of an intelligent operation and maintenance method for a nuclear power starting water pump rolling bearing are not reported.
Disclosure of Invention
The invention aims to provide a method and a system for detecting faults of a rolling bearing of a nuclear power starting water pump, so as to overcome the defects of the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a nuclear power starting water pump rolling bearing fault detection method comprises the following steps:
s1, establishing a bearing fault diagnosis model based on a convolutional neural network, determining model parameters by adopting a simulation experiment based on a control variable method, and training the bearing fault diagnosis model by utilizing the original operation data of the rolling bearing of the nuclear power starting water pump and working condition labels corresponding to the original operation data based on the determined model parameters to obtain a bearing fault diagnosis model;
and S2, monitoring and collecting the nuclear power starting water pump rolling bearing operation data in real time, transmitting the collected operation data to a bearing fault diagnosis model, and analyzing the collected operation data through the bearing fault diagnosis model to obtain corresponding operation state information so as to complete the nuclear power starting water pump rolling bearing fault detection.
Further, original operation data of the rolling bearing of the nuclear power starting water pump and a working condition label corresponding to the original operation data are acquired through field data acquisition.
Further, the original operation data comprises the type of the bearing, the sampling frequency, and the vibration acceleration and the rotation speed of the inner ring, the outer ring and the rolling body of the bearing.
Further, the working condition labels corresponding to the original working data comprise fault, no fault, fault diameter and fault position.
Further, the bearing fault diagnosis model comprises an input layer, a convolution layer, a pooling layer, a full connection layer and a softmax output layer.
Further, on the basis of a Tensorflow platform, original operation data are used as input of a general convolutional neural network model, a single control variable method is adopted, the influence of the type of an activation function, the size of a convolutional kernel and the number of network layers on the accuracy of the model is comprehensively analyzed, and model parameters are determined;
the accuracy rate is as follows:
wherein TP is the number of samples for which the prediction result is true and the prediction is correct, TN is the number of samples for which the prediction result is true and the prediction is false, FP is the number of samples for which the prediction result is true and the prediction is incorrect, and FN is the number of samples for which the prediction result is true and the prediction is incorrect.
Further, the model for determining the bearing fault diagnosis model comprises an activation function with a parameter value of ReLU, a characteristic diagram with a parameter value of [1,24,24,1], a convolution kernel with a parameter value of [5,5], a step length with a parameter value of [1,1,1,1], a first convolution layer with a parameter value of [5,5,1,32], a pooling layer 1 with a parameter value of [1,2,2,1], a second convolution layer with a parameter value of [5,5,32,64], a pooling layer with a parameter value of [1,2,2,1] and a scanning mode with a parameter value of SAME.
Further, after convolution is performed on the input image by using a plurality of convolution kernels and a bias term is added, a corresponding feature map of the image is obtained through an activation function, and the mathematical expression of the convolution is as follows:
wherein the content of the first and second substances,
is the jth element of the ith layer; mjThe jth convolution region of the l-1 layer feature map;
is an element therein;
is a corresponding weight matrix;
is a bias term; f (-) is an activation function; convolutional neural network model through training
The weight matrix values and
implementing classification tasks by biasing item values
Adopting a maximum pooling method to carry out maximum value taking operation on the characteristic diagram output by the convolutional layer in each non-overlapping region with the size of n multiplied by n;
unfolding the feature map into a one-dimensional feature vector, weighted summing and activating the function to obtain:
yk＝f(wkxk-1+bk)
wherein k is the serial number of the network layer; y iskIs the output of the full link layer; x is the number ofk-1Is a one-dimensional feature vector; w is akIs a weight coefficient; bkIs a bias term;
the fault diagnosis model is trained by adopting a back propagation algorithm, the gradient of each weight is calculated by utilizing a chain type derivative calculation loss function, the weight is updated according to a gradient descent algorithm, and a cost function used for solving the convolutional neural network is a cross entropy function, and the formula is as follows:
where C represents the cost, x represents the sample, n represents the total number of samples, a represents the model output value, and y represents the sample actual value.
A nuclear power starting water pump rolling bearing fault detection system comprises an online monitoring module, a fault diagnosis module and an alarm module;
the linear monitoring module is used for monitoring parameters of the inner ring, the outer ring, the rolling body vibration data, the acceleration data, the bearing rotating speed and the transmission power of the rolling bearing of the nuclear power starting water pump in real time on line and transmitting the obtained parameters to the fault diagnosis module; and the fault diagnosis module carries out fault monitoring diagnosis according to the parameters and obtains a corresponding diagnosis result, and if the diagnosis result shows that a fault exists, the alarm module gives an alarm.
Furthermore, the on-line monitoring module comprises on-site detection equipment and a storage server, wherein the on-site detection equipment comprises an acceleration sensor and a torque sensor, and the acceleration sensor is positioned on a bearing seat at the fan end and the driving end of the water pump motor and used for acquiring vibration acceleration signals of the rolling bearing; the torque sensor is connected with a motor load through an elastic pin coupler and is used for measuring power and rotating speed; the storage server is used for storing the acquired data.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention relates to a fault detection method for a rolling bearing of a nuclear power starting water pump, which comprises the steps of establishing a bearing fault diagnosis model based on a convolutional neural network, determining model parameters by adopting a simulation experiment based on a control variable method, and training the bearing fault diagnosis model by utilizing original operation data and working condition labels corresponding to the original operation data of the rolling bearing of the nuclear power starting water pump based on the determined model parameters.
Furthermore, on the basis of a Tensorflow platform, original operation data are used as input of a general convolutional neural network model, model parameters are determined by adopting a single control variable method, calculation is simple and fast, corresponding model parameters can be obtained fast, and fast detection of faults of a rolling bearing of the nuclear power starting water pump is facilitated.
According to the rolling bearing fault detection system for the nuclear power starting water pump, a real-time monitoring and diagnosing system for bearing faults is adopted, so that the reliability and stability of a nuclear power unit can be improved, excessive maintenance is avoided, and the operation efficiency and economic benefit of enterprises are improved.
Drawings
FIG. 1 is a schematic diagram of a convolutional neural network model architecture in an embodiment of the present invention.
FIG. 2 is a schematic flow chart of a back propagation algorithm in an embodiment of the present invention.
FIG. 3 is a diagram illustrating the effect of different activation functions on model accuracy in an embodiment of the present invention.
FIG. 4 is a diagram illustrating the effect of convolution kernel size on model accuracy in an embodiment of the present invention.
Fig. 5 is a schematic diagram illustrating an influence of the number of network layers on the accuracy of the model in the embodiment of the present invention.
FIG. 6 is a flow chart illustrating a maintenance scenario decision-making process according to an embodiment of the present invention.
FIG. 7 is a schematic view of an operation and maintenance system of a rolling bearing of the nuclear power starting water pump in the embodiment of the invention.
FIG. 8 is a diagram of a conventional stacked pre-trained deep neural network.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
a nuclear power starting water pump rolling bearing fault detection method comprises the following steps:
s1, establishing a bearing fault diagnosis model based on a convolutional neural network, determining model parameters by adopting a simulation experiment based on a control variable method, and training the bearing fault diagnosis model by utilizing original operation data of a rolling bearing of the nuclear power starting water pump and working condition labels corresponding to the original operation data to obtain the bearing fault diagnosis model with the diagnosis result confidence coefficient of more than 0.95;
specifically, original operation data of a rolling bearing of the nuclear power starting water pump and working condition labels corresponding to the original operation data are acquired through field data acquisition, and the working condition labels corresponding to the original operation data and the original operation data are divided into training and testing data sets according to a ratio of 3: 1;
in the method, original operation data of a rolling bearing of the nuclear power starting water pump and a working condition label corresponding to the original operation data are acquired through field data acquisition, bearing damage is single-point damage of electric spark machining, the fault diameters are respectively 0.007, 0.014, 0.021, 0.028 and 0.040 millimeters, and damage points are respectively arranged on an inner ring, an outer ring and a rolling body of the bearing;
the original operation data comprises the model number of the bearing, the sampling frequency, the vibration acceleration and the rotating speed of the inner ring, the outer ring and the rolling body of the bearing; working condition labels corresponding to the original working data comprise fault, no fault, fault diameter and fault positions (inner ring, outer ring and rolling body);
the raw running data was partitioned into training and test sets at a 3:1 ratio and converted to 24 x 24 gray scale feature maps.
The established bearing fault diagnosis model comprises an input layer, a convolution layer, a pooling layer, a full-connection layer and a softmax output layer;
on the basis of the parameters of the general convolutional neural network model, a single control variable method is adopted to comprehensively analyze the influence of the type of an activation function, the size of a convolutional kernel and the number of network layers on the accuracy rate of the model and determine the parameters of the model;
the accuracy rate is as follows:
wherein TP is the number of samples with true prediction results and correct prediction, TN is the number of samples with true prediction results and false prediction (i.e. correct prediction), FP is the number of samples with true prediction results and incorrect prediction, and FN is the number of samples with true prediction results and incorrect prediction;
training the model by utilizing a training data set based on a Tensorflow platform to obtain a nuclear power starting water pump rolling bearing fault diagnosis model with a diagnosis result confidence coefficient of more than 0.95;
the model parameters of the bearing fault diagnosis model comprise an activation function (parameter value: ReLU), a characteristic diagram (parameter value: 1,24,24, 1), a convolution kernel (parameter value: 5, 5), a step length (parameter value: 1,1, 1), a convolutional layer 1 (parameter value: 5,5,1, 32), a pooling layer 1 (parameter value: 1,2,2, 1), a convolutional layer 2 (parameter value: 5,5,32, 64), a pooling layer (parameter value: 1,2,2, 1) and a scanning mode (parameter value: SAME);
and S2, monitoring and collecting the nuclear power starting water pump rolling bearing operation data in real time, transmitting the collected operation data to a bearing fault diagnosis model, and analyzing the collected operation data through the bearing fault diagnosis model to obtain corresponding operation state information so as to complete the nuclear power starting water pump rolling bearing fault detection.
And if the detected operation data is a fault signal, recording and early warning.
Acquiring vibration acceleration signals, power and rotating speed of a rolling bearing in real time on line through an acceleration sensor and a torque sensor which are arranged at a fan end and a drive end bearing seat of a starting water pump motor; the collected data are transmitted to a server through a data line, and the server comprises a central processing unit and a data storage center, and can store and process the data; and inputting real-time operation data of the bearing into the fault diagnosis model to obtain a diagnosis conclusion. The health assessment should be based on bearing wear diameter.
And according to the constraint conditions of bearing health, maintenance cost and maintenance time, different maintenance suggestions are provided according to different faults.
Before further detailed description of the embodiments of the present invention, terms and expressions referred to in the embodiments of the present invention are described, and the terms and expressions referred to in the embodiments of the present invention are applicable to the following explanations:
example (b):
s1, acquiring original operation data and working condition labels of the rolling bearing through field experiments, and dividing the original operation data and the working condition labels into training and testing data sets according to a ratio of 3: 1;
in S1.1, the field experiment includes:
bearing type: the supporting shaft of the bearing to be detected is a rotating shaft of a nuclear power starting water pump motor, and the sampling frequency is 12kHz and 48kHz。
And (3) fault setting: the damage of the bearing is single-point damage of electric spark machining. The failure diameters were 0.007, 0.014, 0.021, 0.028, and 0.040 millimeters, respectively. The damage points are respectively arranged on the bearing inner ring, the rolling body and the outer ring.
Signal sampling: an acceleration sensor and a torque sensor are respectively arranged above a bearing seat at the fan end and the driving end of the motor and used for collecting vibration acceleration signals and power of a fault bearing, a 16-channel data recorder is used for collecting vibration acceleration signal data of a test bed, the data collection frequency is 12kHz, and the rotating speed of a shaft is 1700 r/min.
Data acquisition: 8 normal samples, 53 outer ring damage samples, 23 inner ring damage samples and 11 rolling body damage samples are obtained.
In S1.2, the data condition tag includes: fault, no fault, fault diameter, fault location (inner ring, outer ring, rolling body)
In S1.3, the original operation data is divided into a training set and a test set in a ratio of 3: 1.
The raw operating data should be converted into 24 × 24 two-dimensional feature maps. The specific method comprises the following steps:
taking 24 × 24 ═ 576 points as input lengths, adopting a sampling mode that the data overlap degree is 0, and segmenting the original vibration data by 576 point step length; one-dimensional data is sequentially converted into a 24 x 1 feature map in a column ordering by applying tf.reshape (xs, [ -1,24,24,1]) in Tensorflow as input during training.
Adding a working condition label corresponding to the characteristic diagram for the processed characteristic diagram, in this embodiment, 5 working condition labels are provided, which are respectively: the A-inner ring failure diameter is 0.007, the B-inner ring failure diameter is 0.014, the C-inner ring failure diameter is 0.021, the D-inner ring failure diameter is 0.028, and the E-is normal.
S2, establishing a convolutional neural network bearing fault diagnosis model, determining model parameters by adopting a simulation experiment based on a control variable method, and training the convolutional neural network model by utilizing a training data set to obtain a nuclear power starting water pump rolling bearing fault diagnosis model with a diagnosis result confidence coefficient of more than 0.95;
s2.1, establishing a convolutional neural network model, wherein the model framework schematic diagram is shown in figure 1, and the convolutional neural network comprises a convolutional layer, a pooling layer, a full-connection layer hidden layer and a softmax output layer.
And (3) rolling layers: after convolution is carried out on the input image by utilizing a plurality of convolution kernels and a bias term is added, a series of characteristic graphs can be obtained through an activation function, and the mathematical expression of the convolution is as follows:
wherein the content of the first and second substances,
is the jth element of the ith layer; mjThe jth convolution region of the l-1 layer feature map;
is an element therein;
is a corresponding weight matrix;
is the bias term. f (-) is the activation function.
Convolutional neural network model through training
The weight matrix values and
the bias term values implement the classification task.
A pooling layer: the pooling layer is used for reducing the dimension of the feature map and ensuring the translation invariance of the features; in this example, a maximum pooling method is selected. Specifically, the pooling layer performs a maximum value operation on the feature map output by the convolutional layer in each non-overlapping region of size n × n, thereby realizing n-fold reduction of the feature map in two dimensions.
Full connection layer: unfolding the feature map into a one-dimensional feature vector, weighted summing and activating the function to obtain:
yk＝f(wkxk-1+bk)
wherein k is a netThe number of the network layer; y iskIs the output of the full link layer; x is the number ofk-1Is a one-dimensional feature vector; w is akIs a weight coefficient; bkFor the bias term, softmax activation function is used in the present embodiment.
The convolutional neural network fault diagnosis model training method is a back propagation algorithm, as shown in fig. 2, the algorithm uses a chain derivation calculation loss function to calculate the gradient of each weight, the weight is updated according to a gradient descent algorithm, a cost function used for solving the convolutional neural network is a cross entropy function, and the formula is as follows:
where C represents the cost, x represents the sample, n represents the total number of samples, a represents the model output value, and y represents the sample actual value.
S2.2, simulation experiment: based on a Tensorflow platform, training set data is used as input of a general convolutional neural network model, and the influence of the type of an activation function, the size of a convolutional kernel and the number of network layers on the final accuracy of the model is comprehensively analyzed based on a single control variable method, so that a theoretical basis is provided for subsequent model training parameter adjustment.
The simulation experiment specifically comprises:
evaluation indexes are as follows: the evaluation index of the convolutional neural network model is accuracy, which is specifically as follows:
where TP is the number of samples for which the prediction result is true and the prediction is correct, TN is the number of samples for which the prediction result is actually false and the prediction is false (i.e., the prediction is correct), FP is the number of samples for which the prediction result is true and the prediction is incorrect, and FN is the number of samples for which the prediction result is true and the prediction is incorrect.
The general convolutional neural network model parameter settings are shown in table 2:
TABLE 2 parameter settings for CNN model
The influence of the activation function type on the accuracy of the nuclear power starting water pump rolling bearing fault diagnosis model is as follows:
common activation functions include: sigmoid function, tanh function, ReLU function, LReLU function, ELU function; the method comprises the following specific steps:
sigmoid function:
tan h function:
the RuLU function:
R(x)＝max(0,x)
LReLU function:
LR(x)＝max(αx,x)
ELU function:
and taking the training set data as the input of the general convolutional neural network model, and calculating the classification accuracy of the corresponding model by changing the type of the activation function.
As shown in FIG. 3, the ReLU function has the highest average accuracy, the function convergence rate is high, the prediction performance is good, and the method is suitable for processing fault data of the rolling bearing of the nuclear power starting water pump.
Influence of the size of the convolution kernel on the accuracy of a nuclear power starting water pump rolling bearing fault diagnosis model is as follows:
the sizes of convolution kernels are respectively set as follows: and (1, 1), (2, 2), (3, 3) … (15, 15), wherein an activation function is ReLU, other parameters are the same as those in the table 2, and the model accuracy is tested by changing the size of a convolution kernel.
From fig. 4, the model accuracy rate is positively correlated to the convolution kernel increase, and the accuracy rate remains about 0.97 after the convolution kernel is increased to [5,5 ]. Considering that a large convolution kernel can cause problems of long calculation time, large memory occupation and the like, the optimal convolution kernel size is determined to be [5,5 ].
Influence of the network layer number on the accuracy of a nuclear power starting water pump rolling bearing fault diagnosis model is as follows:
the number of layers is set to be 1,2,3,4 and 5 respectively, the activation function is ReLU, other parameters are the same as those in the table 2, and the model accuracy is tested by changing the number of layers.
From fig. 5, the model accuracy is positively correlated with the increase of the number of network layers, but a large number of layers can result in the problems of multiplied data amount, increased calculated amount, increased occupied memory, reduced training speed and the like, and the number of network layers is determined to be 2.
S2.3, the final structure and parameters of the nuclear power starting water pump rolling bearing fault diagnosis model are as follows:
TABLE 3 parameter determination for CNN model
S3, monitoring the rolling bearing carrying state of the nuclear power starting water pump in real time, inputting the monitoring data acquired by real-time monitoring into the fault diagnosis model, carrying out fault monitoring diagnosis on the rolling bearing of the nuclear power starting water pump, evaluating the health condition of the rolling bearing, recording abnormal data and giving an alarm;
and S3.1, acquiring vibration acceleration signals, power and rotating speed of the rolling bearing in real time on line through an acceleration sensor and a torque sensor which are arranged on a fan end and a driving end bearing seat of a starting water pump motor.
The field monitoring equipment comprises an acceleration sensor and a torque sensor;
the acceleration sensor is positioned on a bearing seat at the fan end and the driving end of the water pump motor and is used for acquiring vibration acceleration signals of the rolling bearing; the torque sensor is connected with a motor load through an elastic pin coupler and is used for measuring power and rotating speed; the corresponding data was collected using a 16-channel data recorder.
And S3.2, transmitting the acquired data to a server through a data line, wherein the server comprises a central processing unit and a data storage center, and the data storage center can store and process the data.
And S3.3, inputting real-time running data of the bearing into the convolutional neural network fault diagnosis model to monitor and diagnose the fault of the rolling bearing of the nuclear power starting water pump and obtain a bearing fault diagnosis conclusion.
S3.4, adopting a bearing health evaluation method based on the fault diameter, wherein the results are shown in FIGS. 6 and 7, and are specifically shown in Table 4:
TABLE 4 Nuclear power starting water pump rolling bearing fault assessment standard
S4, comprehensively considering the conditions of bearing health condition, maintenance cost, maintenance time and the like, and making a corresponding operation and maintenance scheme;
s4.1, taking bearing health, maintenance cost, maintenance time and the like as optimization targets, based on nuclear power expert knowledge, proposing different maintenance suggestions according to different faults, and establishing an operation and maintenance information base, wherein the operation and maintenance information base is specifically shown in a table 5:
TABLE 5 Nuclear power starting water pump rolling bearing operation and maintenance information base
S4.2, finding out corresponding operation and maintenance suggestion information from the operation and maintenance suggestion information base and providing the operation and maintenance suggestion information for nuclear power maintenance personnel, wherein test and verification results are as follows:
The invention also provides an intelligent operation and maintenance system for the rolling bearing of the nuclear power starting water pump, which comprises a bearing online monitoring subsystem and a bearing intelligent operation and maintenance subsystem.
The bearing on-line monitoring subsystem comprises an on-line monitoring module, a fault diagnosis module, a health evaluation module and an alarm module.
The on-line monitoring module is used for monitoring the inner ring, the outer ring, the rolling body vibration data and the acceleration data of the rolling bearing of the nuclear power starting water pump, the rotating speed and the transmission power of the bearing in real time on line.
The on-line monitoring module comprises an on-site detection device: the system comprises an acceleration sensor and a torque sensor, wherein the acceleration sensor is positioned on a bearing seat at the fan end and the driving end of a motor of a starting water pump and is used for acquiring vibration acceleration signals of a rolling bearing; the torque sensor is connected with a motor load through an elastic pin coupler and is used for measuring power and rotating speed; the corresponding data was collected using a 16-channel data recorder.
A server: the collected data are transmitted to the server through the data line, the server comprises a central processing unit and a data storage center, the collected data can be stored, and data reading is facilitated.
The fault diagnosis module is used for carrying out fault monitoring diagnosis according to the real-time monitoring data and the convolutional neural network bearing fault diagnosis model and obtaining a corresponding diagnosis result;
and the health evaluation module is used for evaluating the health condition of the bearing by taking the bearing wear diameter as a standard according to the fault diagnosis result.
The alarm module is used for giving an alarm prompt for a fault bearing and transmitting fault information to the intelligent operation and maintenance subsystem;
the bearing intelligent operation and maintenance subsystem comprises an operation and maintenance suggestion information base and an intelligent maintenance scheme decision-making module.
The operation and maintenance suggestion information base is used for providing corresponding maintenance schemes according to constraint conditions such as bearing health, maintenance cost and maintenance time and according to different faults of the rolling bearing of the nuclear power starting water pump based on nuclear power expert knowledge.
And the intelligent maintenance scheme decision module is used for finding out corresponding operation and maintenance suggestion information from the operation and maintenance suggestion information base according to the health state of the bearing and providing the operation and maintenance suggestion information for nuclear power maintenance workers.
Wherein 1) downsampling processing: for a sample sequence sampled once every several sample values, the new sequence thus obtained is a down-sampling of the original sequence, for example: for an image I with size M N, S times down sampling is carried out to obtain resolution image with size (M/S) N/S, wherein S is common divisor of M and N
2) Convolutional Neural Networks (CNN) are a type of feed-forward Neural network that contains convolution calculations and has a deep structure, and are one of the representative algorithms for deep learning. The convolutional neural network has the capability of representing learning, and can carry out translation invariant classification on input information according to the hierarchical structure of the convolutional neural network.
3) Model training: multi-class learning is performed on the image dataset. The model is constructed by adopting a TensorFlow deep learning framework, a multi-classification model is formed by combining multiple layers of neural network layers such as CNN (neural network) layers, the input of the model is a two-dimensional characteristic diagram, the output of the model is multi-classification probability, and classification categories are finally output through algorithms such as softmax. During training, the model approaches to a correct trend through an objective function such as cross entropy and the like.
4) softmax function: the normalized exponential function is a generalization of the logistic function, and is defined as follows:
wherein, ViIs the output of the pre-stage output unit of the classifier. i represents a category index, and the total number of categories is C. SiIndicates that it is currentThe ratio of the index of an element to the sum of the indices of all elements. It can "compress" a K-dimensional vector containing any real number into another K-dimensional real vector, so that each element is in the range of [0, 1]]And the sum of all elements is 1. Namely: the output values of the multiple classes can be converted into relative probabilities by the softmax index.
The invention utilizes the end-to-end characteristic of the convolutional neural network, skips the links of internal mechanism analysis, Fourier transform and the like, directly acts on the original operation data, and avoids the loss of the original information. The selection of hyper-parameters and activation functions in the traditional convolutional neural network is based on a trial and error method, and the method optimizes the parameters of the convolutional neural network based on experimental research, so that the model training efficiency is further improved. In order to solve the current situation of nuclear power extensive operation and maintenance management, a real-time monitoring and diagnosis and health dynamic evaluation method of bearing faults is developed, and a scientific and effective health management strategy is formulated based on evaluation information, so that the reliability and stability of a nuclear power unit are finally improved, excessive maintenance is avoided, and the operation efficiency and economic benefit of enterprises are improved.
Claims (10)
1. A nuclear power starting water pump rolling bearing fault detection method is characterized by comprising the following steps:
s1, establishing a bearing fault diagnosis model based on a convolutional neural network, determining model parameters by adopting a simulation experiment based on a control variable method, and training the bearing fault diagnosis model by utilizing the original operation data of the rolling bearing of the nuclear power starting water pump and working condition labels corresponding to the original operation data based on the determined model parameters to obtain a bearing fault diagnosis model;
and S2, monitoring and collecting the nuclear power starting water pump rolling bearing operation data in real time, transmitting the collected operation data to a bearing fault diagnosis model, and analyzing the collected operation data through the bearing fault diagnosis model to obtain corresponding operation state information so as to complete the nuclear power starting water pump rolling bearing fault detection.
2. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 1, characterized in that original operation data of the rolling bearing of the nuclear power starting water pump and a working condition label corresponding to the original operation data are acquired through field data acquisition.
3. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 1, wherein the original operation data comprises the model number and the sampling frequency of the bearing, and the vibration acceleration and the rotation speed of an inner ring, an outer ring and a rolling body of the bearing.
4. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 3, wherein working condition labels corresponding to original working data comprise fault, no fault, fault diameter and fault position.
5. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 1, wherein the bearing fault diagnosis model comprises an input layer, a convolution layer, a pooling layer, a full connection layer and a softmax output layer.
6. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump is characterized in that based on a Tensorflow platform, original operation data are used as input of a general convolutional neural network model, a single control variable method is adopted, the influence of an activation function type, the size of a convolutional kernel and the number of network layers on the accuracy of the model is comprehensively analyzed, and model parameters are determined;
the accuracy rate is as follows:
wherein TP is the number of samples for which the prediction result is true and the prediction is correct, TN is the number of samples for which the prediction result is true and the prediction is false, FP is the number of samples for which the prediction result is true and the prediction is incorrect, and FN is the number of samples for which the prediction result is true and the prediction is incorrect.
7. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump as claimed in claim 6, wherein the model for determining the bearing fault diagnosis model comprises an activation function with a parameter value of ReLU, a characteristic diagram with a parameter value of [1,24,24,1], a convolution kernel with a parameter value of [5,5], a step length with a parameter value of [1,1,1,1], a first convolution layer with a parameter value of [5,5,1,32], a pooling layer 1 with a parameter value of [1,2,2,1], a second convolution layer with a parameter value of [5,5,32,64], a pooling layer with a parameter value of [1,2,2,1] and a scanning mode with a parameter value of SAME.
8. The method for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 5, characterized in that a plurality of convolution kernels are convoluted with an input image, a bias term is added, a corresponding characteristic diagram of the image is obtained through an activation function, and the mathematical expression of the convolution is as follows:
wherein the content of the first and second substances,
is the jth element of the ith layer; mjThe jth convolution region of the l-1 layer feature map;
is an element therein;
is a corresponding weight matrix;
is a bias term; f (-) is an activation function; convolutional neural network model through training
The weight matrix values and
implementing classification tasks by biasing item values
Adopting a maximum pooling method to carry out maximum value taking operation on the characteristic diagram output by the convolutional layer in each non-overlapping region with the size of n multiplied by n;
unfolding the feature map into a one-dimensional feature vector, weighted summing and activating the function to obtain:
yk＝f(wkxk-1+bk)
wherein k is the serial number of the network layer; y iskIs the output of the full link layer; x is the number ofk-1Is a one-dimensional feature vector; w is akIs a weight coefficient; bkIs a bias term;
the fault diagnosis model is trained by adopting a back propagation algorithm, the gradient of each weight is calculated by utilizing a chain type derivative calculation loss function, the weight is updated according to a gradient descent algorithm, and a cost function used for solving the convolutional neural network is a cross entropy function, and the formula is as follows:
where C represents the cost, x represents the sample, n represents the total number of samples, a represents the model output value, and y represents the sample actual value.
9. A nuclear power starting water pump rolling bearing fault detection system for the nuclear power starting water pump rolling bearing fault detection method of claim 1 is characterized by comprising an online monitoring module, a fault diagnosis module and an alarm module;
the linear monitoring module is used for monitoring parameters of the inner ring, the outer ring, the rolling body vibration data, the acceleration data, the bearing rotating speed and the transmission power of the rolling bearing of the nuclear power starting water pump in real time on line and transmitting the obtained parameters to the fault diagnosis module; and the fault diagnosis module carries out fault monitoring diagnosis according to the parameters and obtains a corresponding diagnosis result, and if the diagnosis result shows that a fault exists, the alarm module gives an alarm.
10. The system for detecting the fault of the rolling bearing of the nuclear power starting water pump according to claim 9, wherein the online monitoring module comprises an on-site detection device and a storage server, the on-site detection device comprises an acceleration sensor and a torque sensor, and the acceleration sensor is positioned on a bearing seat at the fan end and the driving end of a motor of the starting water pump and is used for collecting a vibration acceleration signal of the rolling bearing; the torque sensor is connected with a motor load through an elastic pin coupler and is used for measuring power and rotating speed; the storage server is used for storing the acquired data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011433136.3A CN112699597A (en) | 2020-12-09 | 2020-12-09 | Nuclear power starting water pump rolling bearing fault detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011433136.3A CN112699597A (en) | 2020-12-09 | 2020-12-09 | Nuclear power starting water pump rolling bearing fault detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112699597A true CN112699597A (en) | 2021-04-23 |
Family
ID=75507002
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011433136.3A Pending CN112699597A (en) | 2020-12-09 | 2020-12-09 | Nuclear power starting water pump rolling bearing fault detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112699597A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113486586A (en) * | 2021-07-06 | 2021-10-08 | 新智数字科技有限公司 | Equipment health state evaluation method and device, computer equipment and storage medium |
| CN116304819A (en) * | 2023-03-15 | 2023-06-23 | 四川大学 | Nuclear reactor operation condition judging method based on LeNet-5 algorithm |
| CN116522998A (en) * | 2023-06-16 | 2023-08-01 | 嘉洋智慧安全科技(北京)股份有限公司 | Information processing method, device, equipment and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107526853A (en) * | 2016-06-22 | 2017-12-29 | 北京航空航天大学 | Rolling bearing fault mode identification method and device based on stacking convolutional network |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN111368885A (en) * | 2020-02-24 | 2020-07-03 | 大连理工大学 | Aero-engine gas circuit fault diagnosis method based on deep learning and information fusion |
-
2020
- 2020-12-09 CN CN202011433136.3A patent/CN112699597A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107526853A (en) * | 2016-06-22 | 2017-12-29 | 北京航空航天大学 | Rolling bearing fault mode identification method and device based on stacking convolutional network |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN111368885A (en) * | 2020-02-24 | 2020-07-03 | 大连理工大学 | Aero-engine gas circuit fault diagnosis method based on deep learning and information fusion |
Non-Patent Citations (5)
| Title |
|---|
| 巢娅 等: ""基于语义分割与脑-机接口的机械手抓取方法研究"", 《重庆理工大学学报（自然科学）》, pages 128 - 151 * |
| 张俊鹏 等: "卷积神经网络在轴承故障诊断中的可解释性探讨", 《轴承》 * |
| 张俊鹏 等: "卷积神经网络在轴承故障诊断中的可解释性探讨", 《轴承》, no. 7, 31 July 2020 (2020-07-31), pages 54 - 60 * |
| 张西宁 等: "一种深度卷积自编码网络及其在滚动轴承故障诊断中的应用", 《西安交通大学学报》 * |
| 张西宁 等: "一种深度卷积自编码网络及其在滚动轴承故障诊断中的应用", 《西安交通大学学报》, vol. 52, no. 7, 31 July 2018 (2018-07-31), pages 1 - 8 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113486586A (en) * | 2021-07-06 | 2021-10-08 | 新智数字科技有限公司 | Equipment health state evaluation method and device, computer equipment and storage medium |
| CN113486586B (en) * | 2021-07-06 | 2023-09-05 | 新奥新智科技有限公司 | Device health state evaluation method and device, computer device and storage medium |
| CN116304819A (en) * | 2023-03-15 | 2023-06-23 | 四川大学 | Nuclear reactor operation condition judging method based on LeNet-5 algorithm |
| CN116522998A (en) * | 2023-06-16 | 2023-08-01 | 嘉洋智慧安全科技(北京)股份有限公司 | Information processing method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Luo et al. | Convolutional neural network based on attention mechanism and Bi-LSTM for bearing remaining life prediction | |
| Wang et al. | A method for rapidly evaluating reliability and predicting remaining useful life using two-dimensional convolutional neural network with signal conversion | |
| CN112699597A (en) | Nuclear power starting water pump rolling bearing fault detection method and system | |
| US20220283576A1 (en) | Automatic diagnosis method, system and storage medium for equipment | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| CN111340238A (en) | Fault diagnosis method, device, equipment and storage medium of industrial system | |
| CN114282579A (en) | Aviation bearing fault diagnosis method based on variational modal decomposition and residual error network | |
| CN111899905B (en) | Fault diagnosis method and system based on nuclear power device | |
| CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
| CN112729834B (en) | Bearing fault diagnosis method, device and system | |
| CN111881627A (en) | Nuclear power device fault diagnosis method and system | |
| KR20230021272A (en) | Bridge health monitoring system using artificial intelligence | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| Wang et al. | Fault diagnosis of industrial robots based on multi-sensor information fusion and 1D convolutional neural network | |
| Jia et al. | Multitask convolutional neural network for rolling element bearing fault identification | |
| Tian et al. | Multi-scale deep coupling convolutional neural network with heterogeneous sensor data for intelligent fault diagnosis | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| CN114580472B (en) | Large-scale equipment fault prediction method with repeated cause and effect and attention in industrial internet | |
| Xu et al. | Deep Transfer learning remaining useful life prediction of different bearings | |
| Bharatheedasan et al. | An intelligent of fault diagnosis and predicting remaining useful life of rolling bearings based on convolutional neural network with bidirectional LSTM | |
| KR20230075150A (en) | Method for managing system health | |
| CN111368379B (en) | Rolling bearing reliability evaluation method and system for improving HWPSO-WPHM model | |
| Sepulvene et al. | Analysis of machine learning techniques in fault diagnosis of vehicle fleet tracking modules | |
| Li et al. | A Novel Unsupervised Graph Wavelet Autoencoder for Mechanical System Fault Detection | |
| Latuny | A sensitivity comparison of Neuro-fuzzy feature extraction methods from bearing failure signals |
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 |