CN117349491A - Fault diagnosis algorithm, method and system of vibrating screen - Google Patents
Fault diagnosis algorithm, method and system of vibrating screen Download PDFInfo
- Publication number
- CN117349491A CN117349491A CN202311638412.3A CN202311638412A CN117349491A CN 117349491 A CN117349491 A CN 117349491A CN 202311638412 A CN202311638412 A CN 202311638412A CN 117349491 A CN117349491 A CN 117349491A
- Authority
- CN
- China
- Prior art keywords
- fault diagnosis
- layer
- vibrating screen
- data
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000012544 monitoring process Methods 0.000 claims abstract description 29
- 230000036541 health Effects 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000004364 calculation method Methods 0.000 claims abstract description 18
- 238000012423 maintenance Methods 0.000 claims abstract description 16
- 238000004519 manufacturing process Methods 0.000 claims abstract description 9
- 238000003062 neural network model Methods 0.000 claims abstract description 5
- 238000002360 preparation method Methods 0.000 claims abstract description 4
- 238000012216 screening Methods 0.000 claims description 15
- 238000004458 analytical method Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000007405 data analysis Methods 0.000 claims description 4
- 238000012806 monitoring device Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 abstract description 2
- 239000010410 layer Substances 0.000 description 97
- 230000006870 function Effects 0.000 description 12
- 230000004913 activation Effects 0.000 description 11
- 238000012800 visualization Methods 0.000 description 10
- 238000007781 pre-processing Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000001133 acceleration Effects 0.000 description 4
- 238000007689 inspection Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 238000005336 cracking Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013016 damping Methods 0.000 description 2
- 238000007599 discharging Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- -1 smelting Substances 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000003723 Smelting Methods 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000005489 elastic deformation Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 239000011819 refractory material Substances 0.000 description 1
- 238000005549 size reduction Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention relates to a fault diagnosis algorithm, a fault diagnosis method, a fault diagnosis system and a fault diagnosis device for a vibrating screen, and belongs to the technical field of intelligent detection of working conditions of large mechanical equipment in a production line; the fault diagnosis algorithm of the vibrating screen comprises the following steps: and (3) data acquisition: collecting original data of equipment; data preparation: unifying, expanding and distributing the data; training a neural network model: the method comprises the steps of utilizing the learning and calculation judgment of a stacked self-coding neural network and a convolutional neural network to judge the health state of equipment and accurately classify faults; the fault diagnosis method of the vibrating screen comprises data monitoring, fault diagnosis and maintenance strategies; the invention relates to a fault diagnosis system of a vibrating screen, which comprises sensor enabling equipment, a fault diagnosis subsystem and a strategy decision subsystem; and the vibration sieve health state is judged by utilizing the learning and calculation judgment of a stacked self-coding algorithm and a convolutional neural network, and the accurate classification of faults is realized.
Description
Technical Field
The invention relates to a fault diagnosis algorithm, a fault diagnosis method, a fault diagnosis system and a fault diagnosis device for a vibrating screen, and belongs to the technical field of intelligent detection of working conditions of large mechanical equipment in a production line.
Background
The vibrating screen is impacted by materials all the time in operation, and the problems of feeding deviation, uneven feeding, elastic deformation of a damping spring, beam cracking, vibration beam cracking, side plate cracking and loosening or tearing of each structural member can exist. If the maintenance is not timely in the daily use process, the existing problems cannot be found in time or the pre-judgment of the existing problems is inaccurate, the operation of the vibrating screen is suddenly stopped, so that small faults are changed into large faults, and the normal production is affected.
The current solution is:
1. inspection tour by field operator experience judgment mode
Disadvantages: the method is influenced by inspection staff, and has the phenomena of inaccurate judgment, undiscovered and hysteresis, and finally causes the problems that faults are not found in time and small faults become large faults.
2. Spring fault monitoring sensor for mounting screen machine
The monitoring to the spring state is realized, the fault can be found in time and the alarm can be given out after the spring is broken, and the problems that the working efficiency of the vibrating screen is reduced and other faults are caused when personnel cannot find out the fault of the spring in time are solved.
Disadvantages: the state of the springs can only be monitored, and faults at other parts of the vibrating screen cannot be judged.
3. Vibrating screen fault monitoring system
The state of the screen is pre-warned by installing a vibration sensor, a temperature sensor, an inertia measurement unit and the like on the vibrating screen. The running state of the screening machine is sensed by each sensor, so that the collection of the movement data of the screening machine is realized, the judgment of the inspection personnel is not needed, and the capability of finding faults in advance is improved by combining the judgment of the inspection personnel.
Disadvantages: the collected data is not deeply processed, and the problems of inaccurate prediction and short early warning time of faults exist seriously only through simple waveform analysis of vibration, temperature and gesture, so that a plurality of faults can be predicted only a few minutes before the faults occur, and the purpose of early warning of the faults cannot be achieved.
Disclosure of Invention
The invention aims to solve the technical problems that: the fault diagnosis algorithm, the fault diagnosis method and the fault diagnosis system for the vibrating screen are provided, the health state of the vibrating screen is judged by utilizing the learning and the calculation judgment of a stacked self-coding algorithm and a convolutional neural network based on the data collected by the sensor in the motion state of the vibrating screen, the accurate classification of faults is realized, and the fault monitoring accuracy of the vibrating screen is improved.
The fault diagnosis algorithm of the vibrating screen comprises the following steps:
and (3) data acquisition: collecting original data of equipment;
data preparation: unifying, expanding and distributing the data;
training a neural network model: and the learning and calculation judgment of the stacked self-coding neural network and the convolutional neural network are utilized to judge the health state of the equipment and accurately classify faults.
Based on the data collected in the motion state of the vibrating screen, the judgment of the health state of the vibrating screen and the accurate classification of faults are realized by utilizing the learning and calculation judgment of a stacked self-coding algorithm and a convolutional neural network.
Preferably, the learning and calculation using stacked self-encoding neural networks and convolutional neural networks is specifically as follows:
the original data is subjected to autonomous training on the self-encoder through a stacked neural network, model parameters are continuously and autonomously optimized, and higher-layer operation of a deep network is realized; pushing the operation result of each layer to an n or n+1 layer as input, and operating each layer to realize the budget of the convolutional neural network, and adding a softMax layer at the end of the self-coding algorithm model to map the final calculation output to a target; then, training is performed in a supervised manner by using the marked training data, so as to realize model operation.
And accurate model operation is realized, and the system fault early warning accuracy is improved.
Preferably, the self-encoder adopts three layers of structures of an input layer, an hidden layer and an output layer for feature learning.
The self-encoder is an unsupervised neural network, which adopts a three-layer structure to perform feature learning: input layer, hidden layer (composed of encoder and decoder) and output layer. The encoder transforms the input x into a hidden representation h using a nonlinear mapping:
h=f(Wx+b)；
where f is a nonlinear activation function. By doing so, the auto-encoder learns the abstract representation from the input. The next step is to decode the hidden representation using a decoder. The decoder maps the hidden representation back to the original data:
z=f(Wˆx+bˆ)；
the automatic encoder neural network is trained in an unsupervised manner to minimize the reconstruction error between z and x by adjusting the model parameters θ= [ w, w ˆ, B ]. The auto-encoder learning may be stacked to form a higher level representation of the deep network. The stacking of layers is by feeding the output to the inputs of the n-th to n+1-th layers and training each layer separately. This step is called pre-training of the convolutional neural network. Pre-training of convolutional neural networks initializes model weights and then performs supervised training to fine tune the model.
Preferably, the convolutional neural network comprises a convolutional layer, a pooling layer and a fully-connected layer.
Preferably, the input and output process of the convolutional neural network follows the following steps:
z1 = ReLU(x ∗ ω1 + b1)
conv1 = pool(z1)
zi = ReLU(convi−1 ∗ωi + bi)
convi = pool(zi)
h1 = ReLU(convi∗ ωconv1 + bconv1)
y = softmax(ωconv2 ∗ h1 + bconv2 )
wherein: x, y are the input and output vectors, respectively, reLU is the rectified linear unit activation function, zi is the I convolution matrix, ∗ is the convolution operation, ω and bi are the weight matrix and the bias vector, convi is the convolution layer after the pooling operation is applied, and h1 is the hidden layer.
The invention relates to a fault diagnosis method of a vibrating screen, which comprises the following steps:
step S1, data monitoring: monitoring real-time data of the operation and health of the equipment;
step S2, fault diagnosis: the fault diagnosis algorithm of the vibrating screen is adopted to judge the health state of equipment and accurately classify faults to form screening machine fault information;
step S3, maintaining a strategy: and (3) carrying out fault mode and influence analysis and screen life operation data analysis according to the screen fault information to determine an optimal maintenance and production scheduling strategy.
The early warning of faults of the vibrating screen body and the vibration exciter is realized, the early warning can be realized more than 24 hours before the faults occur, and the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
Preferably, the operation and health real-time data of the monitoring device in step S1 is stored in a history database of the CMMS/EMA subsystem.
CMMS (computerized maintenance management system) is suitable for facility maintenance management, for which EAM adds additional functional modules such as project management, which becomes a true enterprise-level solution. CMMS/EMA = database + data analysis software + integrated services.
The invention relates to a fault diagnosis system of a vibrating screen, which comprises sensor enabling equipment, a fault diagnosis subsystem and a strategy decision subsystem,
sensor-enabling device: real-time data for monitoring the operation and health of the device;
fault diagnosis subsystem: the fault diagnosis algorithm is used for judging the health state of equipment and accurately classifying faults to form screening machine fault information;
policy resolution subsystem: the method is used for carrying out fault mode and influence analysis according to the fault information of the screening machine and analyzing the whole service life operation data of the screening machine to determine the optimal maintenance and production scheduling strategy.
The early warning of faults of the vibrating screen body and the vibration exciter is realized, the early warning can be realized more than 24 hours before the faults occur, and the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
Preferably, the system also comprises a CMMS/EMA subsystem, wherein the CMMS/EMA subsystem comprises a history database.
CMMS (computerized maintenance management system) is suitable for facility maintenance management, for which EAM adds additional functional modules such as project management, which becomes a true enterprise-level solution. CMMS/EMA = database + data analysis software + integrated services.
Compared with the prior art, the invention has the following beneficial effects:
the fault diagnosis algorithm of the vibrating screen is based on the data collected in the moving state of the vibrating screen, and realizes the judgment of the health state of the vibrating screen and the accurate classification of faults by utilizing the learning and calculation judgment of the stacked self-coding algorithm and the convolutional neural network.
The fault diagnosis method of the vibrating screen realizes early warning of faults of the vibrating screen body and the vibration exciter, and the early warning can be realized more than 24 hours before the faults occur generally, so that the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
The fault diagnosis system of the vibrating screen, disclosed by the invention, realizes early warning of faults of the vibrating screen body and the vibration exciter, and can realize early warning more than 24 hours before the faults occur, and the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
Drawings
FIG. 1 is a visual flow chart of a fault diagnosis algorithm for a shaker of the present invention;
FIG. 2 is a block diagram of a self-encoder of the present invention;
FIG. 3 is a block diagram of a convolutional neural network of the present invention;
FIG. 4 is a simplified operational flow diagram of a fault diagnosis algorithm for a shaker of the present invention;
FIG. 5 is a visual representation of the automatic encoder input of the present invention;
FIG. 6 is a first layer visualization of an automatic encoder of the present invention;
FIG. 7 is a second layer visualization of an automatic encoder of the present invention;
FIG. 8 is a visual illustration of a convolutional neural network sample of the present invention;
FIG. 9 is a second layer visualization of a convolutional neural network of the present invention;
FIG. 10 is a third layer of a convolutional neural network of the present invention;
FIG. 11 is a first layer visualization of a signal convolutional neural network of the present invention 2;
FIG. 12 is a first layer visualization of a signal convolutional neural network of the present invention 12;
FIG. 13 is a simplified workflow diagram of a method of fault diagnosis of a vibrating screen of the present invention;
FIG. 14 is a diagram of a CNN class confusion matrix according to the present invention;
fig. 15 is a graph of ROC of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in further detail with reference to the accompanying drawings.
Example 1
The invention is suitable for industries such as mines, coal, smelting, building materials, refractory materials, light industry, chemical industry, medicines, foods and the like, and as shown in fig. 1 and 4, the fault diagnosis algorithm of the vibrating screen comprises the following steps:
and (3) data acquisition: collecting original data of equipment;
data preparation: unifying, expanding and distributing the data;
training a neural network model: and the learning and calculation judgment of the stacked self-coding neural network and the convolutional neural network are utilized to judge the health state of the equipment and accurately classify faults.
Based on the data collected in the motion state of the vibrating screen, the judgment of the health state of the vibrating screen and the accurate classification of faults are realized by utilizing the learning and calculation judgment of a stacked self-coding algorithm and a convolutional neural network.
Here, based on the existing data sensing hardware of the vibrating screen, vibration and inertia measurement data of the vibration exciter, vibration data of the vibrating screen body and vibration data of the vibration isolation system.
Further, the learning and calculation using stacked self-encoding neural networks and convolutional neural networks is specifically as follows:
on the basis of applying multiple algorithm models, the original data are autonomously trained on the self-encoder through a stacked neural network, model parameters are continuously and autonomously optimized, and higher-layer operation of a deep network is realized; pushing the operation result of each layer to an n or n+1 layer as input, and operating each layer to realize the budget of the convolutional neural network, and adding a softMax layer at the end of the self-coding algorithm model to map the final calculation output to a target; then, training is performed in a supervised manner by using the marked training data, so as to realize model operation.
The method comprises the following steps:
first, data preprocessing: the collected vibration signals are often affected by environmental noise, interference signals and other factors, and signal preprocessing such as filtering, amplifying, denoising and the like is required to ensure the accuracy and reliability of the signals.
Second, preprocessing data and calculating: there are many methods for calculating the preprocessing data, and the preprocessing data is calculated according to the requirements.
The Fourier transform is used for decomposing the vibration waveform signal into a series of sine waves with different frequencies, the frequency spectrum can be obtained through the Fourier transform, and the frequency components contained in the signal can be analyzed.
The wavelet transformation decomposes the vibration waveform signal into a series of wavelet functions, can analyze the local characteristics of the signal at different scales, and is suitable for the analysis of non-stationary signals.
Short-time Fourier transform is performed to divide the vibration waveform signal into a series of short-time windows, and Fourier transform is performed on each window to obtain frequency spectrum of each window, so that frequency characteristics of the signal in different time periods can be analyzed.
Empirical mode decomposition decomposes a vibration waveform signal into a series of intrinsic mode functions IMFs, each IMF representing a local feature in the signal, which can be used to analyze the local feature and time-frequency distribution of the signal.
Random process model-the vibration waveform signal is treated as a random process, and the signal is modeled and analyzed using a random process model, such as an autoregressive model (AR), an autoregressive moving average model (ARMA), and the like.
Different models have advantages and disadvantages. And selecting different models according to different fault data, and pushing the data to the self-encoder through stacking after calculation.
Third, constructing a self-encoder and training:
the self-encoder model is constructed, a stacked neural network architecture is selected as a basic model of the self-encoder, and a suitable convolutional neural network is constructed as an encoder part of the self-encoder. The encoder comprises a convolution layer, a pooling layer, an activation function layer and the like. Stacked neural networks are typically composed of multiple hidden layers, with an encoder for compressing input data into a low-dimensional representation and a decoder for reconstructing the low-dimensional representation into the original data. Network structures using convolutional neural networks, fully-connected neural networks, or other suitable problems may be matched.
The decoder section is constructed as well as the deconvolution section from the encoder. The goal of the decoder is to restore the low-dimensional representation obtained by the encoder to the original input data.
The loss function is defined, typically a reconstruction error, from the encoder, and a Mean Square Error (MSE) or other suitable loss function may be used. For classification tasks, the difference between the model output and the actual label is measured using cross entropy as a loss function.
Autonomous training, which is performed by means of unsupervised learning, may use a combination of training data sets and labels, thus eliminating the need for labeled training data. The training data set includes input data as input from the encoder and target data equivalent to the input data for guiding reconstruction from the encoder.
Iterative optimization, updating the weights and offsets from the encoder by an optimization algorithm (e.g., random gradient descent) and a loss function, so that the reconstruction error is reduced as much as possible.
Through repeated iterative training, and each time the last result is pushed to the n or n+1 layer, the parameters of the self-encoder are continuously adjusted, and the compression and reconstruction capability of input data is improved.
A Softmax layer was added. After the decoder section, a fully connected layer (also called dense layer) is added and connected to the Softmax layer. The output of this fully connected layer will serve as the input to the classifier.
The model can be used for classification tasks after the Softmax layer is added, and the Softmax layer outputs a set of probability values which represent the probability that the input data belongs to each category. The classification decision may be made using an appropriate threshold or choosing the highest probability.
It should be noted that some adjustments and optimizations may be needed in the modification of the self-encoder model, such as adjusting network architecture, adding regularization techniques, adjusting learning rate, etc. The model is tested and evaluated for optimal performance based on specific problems and data.
And fourthly, training in a supervision mode by using the marked training data, and finally realizing accurate model operation and improving the system fault early warning accuracy.
For example: the screen body motion monitors that the output end at one side of the screen body is abnormal, the existing monitoring operation result is that the amplitude value is compared with the operation result of the multi-layer perceptron network system through simple Fourier transformation to find similar waveforms to alarm, and even many operations are incomplete to the operation of the multi-layer perceptron network. The existing algorithm similar amplitude value and waveform cannot accurately judge faults, a system can judge that the screening machine has feeding deviation, damping spring faults, structural damage and the like, a specific problem also needs to be confirmed by personnel on site, and in the initial stage of the faults, because fault characteristics are not obvious and interference of other vibrations of equipment is received, even after early warning of the system, the personnel cannot effectively judge the cause of the abnormality. However, the diagnosis algorithm is based on a stacked automatic coding algorithm and a convolutional neural network, which can perform multi-level operation and comparison on original data, the convolutional neural network has an operation structure more complex than that of a multi-level perceptron, and performs multi-level filtering operation, and performs fine splitting operation on signals acquired by each sensor, so as to realize comparison operation of fault waveforms and corresponding fault waveforms. Through uninterrupted data operation, the corresponding fault guiding result of the system algorithm model is continuously updated, and finally, the accurate early warning of the faults of the vibrating screen is realized, and hidden danger is eliminated before the accident through perception and prejudgment.
Experimental data: aiming at vibrating screens of different types, the early warning accuracy of the faults of the vibrating screens on the current market is about 60% -70%, and the fault accuracy of the vibrating screens is improved by more than 50% through an algorithm.
Vibrating screen sensor arrangement: the number of sensors installed for different types of vibrating screens is slightly different, and 6-10 sensors are generally arranged. The single-layer double-drive vibrating screen is provided with 6 sensors, the double-layer double-drive vibrating screen is provided with 8 sensors, and the three-layer double-drive vibrating screen is provided with 10 sensors. Taking the most common single-layer double-drive vibrating screen as an example: the side plates at the left and right supporting seats of the feeding end and the discharging end of the vibrating screen are respectively provided with a triaxial acceleration sensor, the center position of the vibration exciter of the vibrating screen is respectively provided with a nine-axis inertial navigation sensor, and 6 sensors are arranged in total.
And (3) vibrating screen data acquisition: the sensors for feeding and discharging the vibrating screen collect acceleration information of an X axis, a Y axis and a Z axis of a monitoring position; inertial navigation sensors on vibration exciters driven by a vibrating screen collect acceleration information and angular velocity information of X axis, Y axis and Z axis of a monitoring position. The data acquired by the acceleration sensor and the vibration exciter inertial navigation sensor of the screen body of the vibrating screen are not less than 4096 respectively per second, and 1800 data are extracted according to the corresponding period for analysis and calculation.
First, data unification is performed for each observation data using minimum maximum normalization as shown in the following equation. The data was then shuffled, spread (for 2D CNN only) and split into two subsets, 66.6% and 33.3% for training and testing, respectively;
Pnormi=[Pi−min(P)]÷[max(P)−min(P)]。
as shown in fig. 2, the self-encoder adopts three layers of input layer, hidden layer and output layer to perform feature learning. The first and last layers are input and output layers, the others being multi-layer perceptron (MLP).
The self-encoder is an unsupervised neural network, which adopts a three-layer structure to perform feature learning: input layer, hidden layer (composed of encoder and decoder) and output layer. The encoder transforms the input x into a hidden representation h using a nonlinear mapping:
h=f(Wx+b)；
where f is a nonlinear activation function. By doing so, the auto-encoder learns the abstract representation from the input. The next step is to decode the hidden representation using a decoder. The decoder maps the hidden representation back to the original data:
z=f(Wˆx+bˆ)；
the self-encoding neural network is trained in an unsupervised manner to minimize the reconstruction error between z and x by adjusting the model parameters θ= [ W, W ˆ, b, b ]. Higher level representations learned from encoders to form deep networks can be stacked. The stacking of layers is by feeding the output to the inputs of the n-th to n+1-th layers and training each layer separately. This step is called pre-training of the convolutional neural network. Pre-training of convolutional neural networks initializes model weights and then performs supervised training to fine tune the model.
As shown in fig. 3, the convolutional neural network includes a convolutional layer, a pooling layer, and a fully-connected layer.
The first layer is an input layer that receives an input signal containing 800 data points. The first convolution layer is composed of a 2D convolution layer (layer 2) and an active layer (layer 3) and applies 64 filters to each input signal. The max-pooling layer (layer 4) basically compresses the data of the previous layer by calculating a maximum value of a window size of 50. The second and third convolution layers run 16 filters and consist of layers 5, 6, 7 and 8.
The input and output process of the convolutional neural network follows the following steps:
z1 = ReLU(x ∗ ω1 + b1)
conv1 = pool(z1)
zi = ReLU(convi−1 ∗ωi + bi)
convi = pool(zi)
h1 = ReLU(convi∗ ωconv1 + bconv1)
y = softmax(ωconv2 ∗ h1 + bconv2 )
wherein: x, y are input and output vectors, respectively. ReLU is a rectified linear unit activation function. zi is the I (TH language) convolutional layer matrix. ∗ is a convolution operation. Pool refers to pool operation. Omega and bi are weight matrices and bias vectors. convi is the convolution layer after the pooling operation is applied. h1 is a hidden layer. In each training iteration, the back propagation algorithm minimizes the loss function and updates ω and Bi.
As shown in fig. 5-7, the visualization of the automatic encoder results:
neural Network (CNN) models are generally considered as black boxes. One way to understand the feature extraction process is by quantitative analysis of the weight matrix (the kernel in the case of CNNs). Our innovation is to study the importance of activating and emphasizing reconstructed signals and their relation to classification accuracy. 5-7 (1 signal: vibrating screen body polarization; 2 signal: vibrating screen excitation beam crack propagation; 3 signal: vibrating screen beam crack propagation; 4 signal: vibrating screen side plate crack propagation; 5 signal: vibrating screen auxiliary shaft bearing seat looseness; 6 signal: vibrating screen driving shaft misalignment; 7 signal: vibrating exciter auxiliary shaft bearing outer ring; 8 signal: vibrating exciter main shaft bearing; 9 signal: vibrating exciter gear wear; 10 signal: motor rotor unbalance; 11 signal: vibrating exciter rail seat bolt looseness; 12 signal: vibrating exciter output shaft gear eccentricity) show the output signals of the first two layers of the proposed automatic encoder neural network. The similarity between classes is so high, as the MLP-01 and MLP-02 outputs show activation. Thus, if the signal is corrupted by noise or if the SNR is low, the network cannot learn the key features from different classes.
Visualization results of CNN:
we studied the classification accuracy from the point of view of the second and third layers visualization. Fig. 8-10 show one sample per class (fig. 8), the average signal visualization after the second layer has applied 64 filters (fig. 9), and the third layer activation (fig. 10). Clearly, signals belonging to classes 5, 6, 8, 9 and 12 (i.e. signals 5, 6, 8, 9 and 12) show significant fault pulses, and it is difficult to infer interpretations from other signals. The second layer applies 64 filters and the average signal appears in the middle, it is clear that in the 2, 3, 4 and 7 signals pulse patterns are noted that were not noted before the filtering was applied in the second layer. After application of RELU activation, it can be seen that there are only pulses in all signals with minimal noise, and that those signatures represent key features extracted from layers 2 and 3. However, the activation from class 2, 3 and 4, does not show significant pulses compared to the other classes, requiring more careful observation of deeper layers.
Fig. 11 and 12 show a visualization of the CNN8 layer in the convolutional layer considering the 2 signal and the 12 signal, respectively. Layers 2 and 3 apply filters and calculate activation as previously described. However, the maximum pooling layer with a window size of 50 reduces the input dimension from 400 data points to 8. It is apparent that the size reduction in class 12 preserves the pulse characteristics after maximum merging and all subsequent layers. However, the pattern of 2 signals appears more complex and the largest pool does not retain the main features. Thus, the reduced classification accuracy of the 2 signal is affected by the maximum combining window size. The gradual reduction of the window size can preserve the content of the original signal, thereby achieving better classification accuracy.
As shown in fig. 13, the fault diagnosis method of the vibrating screen includes the following steps:
step S1, data monitoring: monitoring real-time data of the operation and health of the equipment;
step S2, fault diagnosis: the fault diagnosis algorithm of the vibrating screen is adopted to judge the health state of equipment and accurately classify faults to form screening machine fault information;
step S3, maintaining a strategy: and (3) carrying out fault mode and influence analysis and screen life operation data analysis according to the screen fault information to determine an optimal maintenance and production scheduling strategy.
The early warning of faults of the vibrating screen body and the vibration exciter is realized, the early warning can be realized more than 24 hours before the faults occur, and the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
Further, the operation and health real-time data of the monitoring device in step S1 are stored in a history database of the CMMS/EMA subsystem.
CMMS (computerized maintenance management system) is suitable for facility maintenance management, for which EAM adds additional functional modules such as project management, which becomes a true enterprise-level solution. CMMS/EMA = database + data analysis software + integrated services.
The invention relates to a fault diagnosis system of a vibrating screen, which comprises sensor enabling equipment, a fault diagnosis subsystem and a strategy decision subsystem,
sensor-enabling device: real-time data for monitoring the operation and health of the device;
fault diagnosis subsystem: the fault diagnosis algorithm is used for judging the health state of equipment and accurately classifying faults to form screening machine fault information;
policy resolution subsystem: the method is used for carrying out fault mode and influence analysis according to the fault information of the screening machine and analyzing the whole service life operation data of the screening machine to determine the optimal maintenance and production scheduling strategy.
The early warning of faults of the vibrating screen body and the vibration exciter is realized, the early warning can be realized more than 24 hours before the faults occur, and the early warning accuracy is improved by 50% compared with the existing algorithm. The fault type and the fault position can be accurately judged, and the position accuracy is more than 90%. For the vibrating screen which is subjected to comprehensive informatization monitoring and upgrading, the fault early warning accuracy can be improved only by updating an algorithm, and for the vibrating screen which is not subjected to or is subjected to partial informatization monitoring and upgrading, only the corresponding hardware is required to be upgraded, the equipment structure is not required to be changed, and the upgrading cost is low.
Further, the system also comprises a CMMS/EMA subsystem, wherein the CMMS/EMA subsystem comprises a history database.
CMMS (computerized maintenance management system) is suitable for facility maintenance management, for which EAM adds additional functional modules such as project management, which becomes a true enterprise-level solution. CMMS/EMA = database + data analysis software + integrated services.
The present invention was evaluated as shown in fig. 14 and 15 to demonstrate the performance of CNN-based classifiers using confusion matrices and ROC curves. From the graph, the overall accuracy is 97.5%, and the accuracy of almost all categories except categories 2, 3 and 4 is 100%. From our back-end fault data analysis dataset it was confirmed that these three categories corresponded to the same faults with different severity levels and that the severity levels were very similar, which resulted in similar fault signatures between these categories, judging errors.
We studied the learning mechanism of different neural network models. The end result shows that without preprocessing, the automatic encoder network cannot regenerate a differentiated representation of the input signal. Thus, poor reconstruction performance of the auto-encoder network leads to reduced classification accuracy. On the other hand, the outputs of the convolutional layer and the active layer in CNN show a clear distinction between different categories, which results in a significant improvement in classification accuracy. Care should be taken in selecting the largest pooling layer dimension because it has been shown that significant reductions may lead to loss of key features in different classes. The method disclosed by the invention provides a direct method for improving NN classification accuracy by using MLP, convolution and activation layer output, and realizes that the final calculation result is greatly improved.
Claims (7)
1. A fault diagnosis algorithm for a vibrating screen, comprising:
and (3) data acquisition: collecting original data of equipment;
data preparation: unifying, expanding and distributing the data;
training a neural network model: the method comprises the steps of utilizing the learning and calculation judgment of a stacked self-coding neural network and a convolutional neural network to judge the health state of equipment and accurately classify faults;
the learning and calculation using stacked self-encoding neural networks and convolutional neural networks is specifically as follows:
the original data is subjected to autonomous training on the self-encoder through a stacked neural network, model parameters are continuously and autonomously optimized, and higher-layer operation of a deep network is realized; pushing the operation result of each layer to an n or n+1 layer as input, and operating each layer to realize the budget of the convolutional neural network, and adding a softMax layer at the end of the self-coding algorithm model to map the final calculation output to a target; then, training is performed in a supervised manner by using the marked training data, so as to realize model operation.
2. The fault diagnosis algorithm of the vibrating screen according to claim 1, wherein the self-encoder performs feature learning by adopting a three-layer structure of an input layer, an hidden layer and an output layer.
3. The fault diagnosis algorithm of the vibrating screen according to claim 1, wherein the convolutional neural network comprises a convolutional layer, a pooling layer and a fully-connected layer.
4. The fault diagnosis method of the vibrating screen is characterized by comprising the following steps of:
step S1, data monitoring: monitoring real-time data of the operation and health of the equipment;
step S2, fault diagnosis: the fault diagnosis algorithm of the vibrating screen according to any one of claims 1-3 is adopted to judge the health state of equipment and accurately classify faults to form screening machine fault information;
step S3, maintaining a strategy: and (3) carrying out fault mode and influence analysis and screen life operation data analysis according to the screen fault information to determine an optimal maintenance and production scheduling strategy.
5. The method of claim 4, wherein the operation and health real-time data of the monitoring device of step S1 is stored in a historical database of CMMS/EMA subsystems.
6. A fault diagnosis system of a vibrating screen is characterized by comprising sensor enabling equipment, a fault diagnosis subsystem and a strategy decision subsystem,
sensor-enabling device: real-time data for monitoring the operation and health of the device;
fault diagnosis subsystem: the method is used for judging the health state of equipment by adopting the fault diagnosis algorithm of the vibrating screen according to any one of claims 1-3, and accurately classifying faults to form screening machine fault information;
policy resolution subsystem: the method is used for carrying out fault mode and influence analysis according to the fault information of the screening machine and analyzing the whole service life operation data of the screening machine to determine the optimal maintenance and production scheduling strategy.
7. The system of claim 6, further comprising a CMMS/EMA subsystem, the CMMS/EMA subsystem including a historical database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311638412.3A CN117349491A (en) | 2023-12-04 | 2023-12-04 | Fault diagnosis algorithm, method and system of vibrating screen |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311638412.3A CN117349491A (en) | 2023-12-04 | 2023-12-04 | Fault diagnosis algorithm, method and system of vibrating screen |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117349491A true CN117349491A (en) | 2024-01-05 |
Family
ID=89367751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311638412.3A Pending CN117349491A (en) | 2023-12-04 | 2023-12-04 | Fault diagnosis algorithm, method and system of vibrating screen |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117349491A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114354194A (en) * | 2021-12-09 | 2022-04-15 | 重庆邮电大学 | Rolling bearing fault diagnosis method based on full convolution self-encoder and optimized support vector machine |
| CN114371002A (en) * | 2021-12-30 | 2022-04-19 | 天津理工大学 | Planetary gearbox fault diagnosis method based on DAE-CNN |
| CN116578940A (en) * | 2023-05-18 | 2023-08-11 | 山东省计算中心（国家超级计算济南中心） | Bearing fault new type identification and diagnosis method based on mixed depth self-coding |
-
2023
- 2023-12-04 CN CN202311638412.3A patent/CN117349491A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114354194A (en) * | 2021-12-09 | 2022-04-15 | 重庆邮电大学 | Rolling bearing fault diagnosis method based on full convolution self-encoder and optimized support vector machine |
| CN114371002A (en) * | 2021-12-30 | 2022-04-19 | 天津理工大学 | Planetary gearbox fault diagnosis method based on DAE-CNN |
| CN116578940A (en) * | 2023-05-18 | 2023-08-11 | 山东省计算中心（国家超级计算济南中心） | Bearing fault new type identification and diagnosis method based on mixed depth self-coding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109726524B (en) | CNN and LSTM-based rolling bearing residual service life prediction method | |
| Xiang et al. | Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder | |
| Cabrera et al. | Automatic feature extraction of time-series applied to fault severity assessment of helical gearbox in stationary and non-stationary speed operation | |
| Jardine et al. | A review on machinery diagnostics and prognostics implementing condition-based maintenance | |
| Bafroui et al. | Application of wavelet energy and Shannon entropy for feature extraction in gearbox fault detection under varying speed conditions | |
| Kankar et al. | Rolling element bearing fault diagnosis using wavelet transform | |
| Kwan et al. | A novel approach to fault diagnostics and prognostics | |
| Lou et al. | Bearing fault diagnosis based on wavelet transform and fuzzy inference | |
| EP3136255A1 (en) | Method and system for automatically detecting defects on a rotating shaft | |
| Liang et al. | Convolutional recurrent neural network for fault diagnosis of high-speed train bogie | |
| CN111695209A (en) | Rotary machine small sample health assessment method driven by meta-depth learning | |
| CN106326929A (en) | Rail train running part rolling bearing fault diagnosis method | |
| Bhadane et al. | Bearing fault identification and classification with convolutional neural network | |
| CN111581880A (en) | Rolling bearing residual life prediction method based on APSO-DBN | |
| Tibaduiza Burgos | Design and validation of a structural health monitoring system for aeronautical structures. | |
| CN116380445A (en) | Equipment state diagnosis method and related device based on vibration waveform | |
| CN114563150A (en) | Bridge health online detection module generation method, detection method, tool box and device | |
| Babouri et al. | Rolling bearing faults severity classification using a combined approach based on multi-scales principal component analysis and fuzzy technique | |
| Nasser et al. | A hybrid of convolutional neural network and long short-term memory network approach to predictive maintenance | |
| CN114861349A (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| Mosallam | Remaining useful life estimation of critical components based on Bayesian Approaches. | |
| Lu et al. | Dynamic genetic algorithm-based feature selection scheme for machine health prognostics | |
| Sony | Towards multiclass damage detection and localization using limited vibration measurements | |
| CN117349491A (en) | Fault diagnosis algorithm, method and system of vibrating screen | |
| Manghai et al. | Brake fault diagnosis through machine learning approaches–a review |
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 |