CN111310830B - Blocking fault diagnosis system and method for combine harvester - Google Patents
Blocking fault diagnosis system and method for combine harvester Download PDFInfo
- Publication number
- CN111310830B CN111310830B CN202010096661.4A CN202010096661A CN111310830B CN 111310830 B CN111310830 B CN 111310830B CN 202010096661 A CN202010096661 A CN 202010096661A CN 111310830 B CN111310830 B CN 111310830B
- Authority
- CN
- China
- Prior art keywords
- data
- sdae
- model
- training
- formula
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000003745 diagnosis Methods 0.000 title claims abstract description 39
- 230000000903 blocking effect Effects 0.000 title claims abstract description 23
- 230000002452 interceptive effect Effects 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 71
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 14
- 210000002569 neuron Anatomy 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000003756 stirring Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000006378 damage Effects 0.000 claims 2
- 238000002203 pretreatment Methods 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 13
- 230000000007 visual effect Effects 0.000 abstract description 5
- 201000010099 disease Diseases 0.000 abstract description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 2
- 230000002427 irreversible effect Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012271 agricultural production Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 238000003306 harvesting Methods 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 239000012535 impurity Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000010813 municipal solid waste Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a blocking fault diagnosis system and method of a combine harvester, wherein the system comprises a sensor module, a data acquisition module, an interactive display screen, an audible and visual alarm device and an embedded industrial personal computer, wherein the sensor module is used for monitoring all parameter information of the combine harvester on line, the data monitored by the sensor module are acquired through the data acquisition module and then are processed through the embedded industrial personal computer, firstly, an SDAE-BP model is trained through the acquired data to obtain weight parameters and bias parameters of the SDAE-BP model, then the SDAE-BP model is prefabricated in the embedded industrial personal computer, and then the operation state of the combine harvester can be judged through the SDAE-BP model in the embedded industrial personal computer. The invention can complete the online monitoring of the combine harvester, has high accuracy of blocking fault monitoring, and effectively prevents the equipment from running with diseases for a long time so as to cause low efficiency and irreversible equipment loss.
Description
Technical Field
The invention belongs to the field of agricultural machinery, relates to an agricultural machinery fault monitoring technology, and in particular relates to a system and a method for diagnosing blockage faults of a combine harvester. It can be applied to the fields of agricultural machinery, mining machinery, vehicle machinery, industrial machinery and the like.
Background
Combine harvester is one of the important mechanical equipment for agricultural production. Because the combine harvester has complex structure and bad operation environment, the key parts such as threshing cylinder, feeding screw auger and conveying chain harrow are easy to be blocked. Therefore, the state of the key operation part of the combine harvester is monitored in real time, the occurrence of serious blockage faults of the combine harvester can be effectively prevented, and the method has important significance for ensuring the harvesting operation efficiency.
The existing blockage fault diagnosis system of the combine harvester in China adopts the scheme that sensors are arranged on some key operation components, and signals acquired by the sensors are collected and processed to obtain the blockage result of the combine harvester. The fault diagnosis method and the fault (201310493990.2) of the tangential and longitudinal flow combine harvester based on the FNN provide that the fault analysis is carried out through the acquisition of the rotating speed signals of the header screw feeder, the conveying groove, the tangential roller and the grain conveying screw feeder by a fault diagnosis algorithm based on the FNN; the fault diagnosis system (201721245251.1) of the combine harvester based on the CAN bus provides that the collected signal data of the engine speed sensor, the gearbox speed sensor, the header input speed sensor, the conveying winch Long Zhuaisu sensor, the elevator speed sensor and other sensors are transmitted to the controller through the CAN bus to carry out fault diagnosis. Because the combine harvester has the problems of bad operation environment, complex signal monitoring interference source and the like, the existing fault diagnosis system cannot effectively early warn the blocking fault of the combine harvester, and the fault early warning accuracy rate also needs to be further improved. In order to solve the above problems, a blocking fault diagnosis system of a combine harvester with noise immunity is provided.
Disclosure of Invention
The invention aims to design a system for monitoring and diagnosing the blocking condition of a combine harvester based on rotation speed. Through a scientific and reasonable method, the problem that the existing fault diagnosis system has lower accuracy in fault diagnosis of blockage when the combine harvester operates under the condition of complex interference is effectively solved. The invention provides a method for monitoring and diagnosing the blocking condition of a combine harvester through deep learning. The blocking fault diagnosis of the combine harvester with high precision can be realized.
In order to achieve the above object, the present invention proposes a combine harvester jam fault diagnosis model of a combination of a denoising self-encoder and a BP neural network (Stacked Denoising Autoencoder-BP, SDAE-BP). The model extracts data characteristics through a stacked denoising self-encoder (SDAE), and realizes the fault diagnosis of blockage by combining a BP neural network, and the algorithm flow is shown in figure 2.
In order to meet the above requirements, the present invention adopts the following technical scheme:
a method for diagnosing the blocking fault of a combine harvester is characterized in that A speed sensors are used for respectively monitoring k transmission parts of the combine harvester, a data acquisition module is used for acquiring data of the A speed sensors, and an industrial personal computer is used for processing the data, and the method is characterized in that:
step 1, continuously acquiring analog signals of A speed sensors through a data acquisition module, and converting the analog signals into digital signals;
step 2, manually setting fault types, collecting data of A speed sensors as training data T, establishing an SDAE-BP model based on combination of a stacked denoising self-encoder and a BP neural network by using the training data T, and obtaining a weight parameter W and a bias parameter b of the SDAE-BP model;
step 3, prefabricating or writing data processing software in the industrial personal computer according to the parameters of the data acquisition, the data processing and the SDAE-BP model;
step 4, adopting a certain sampling interval, and conveying and collecting sample data to an SDAE-BP model in the industrial personal computer by utilizing a data collecting module;
step 5, normalizing the sample data acquired in the step 4, wherein the normalized data value is between 0 and 1;
and 6, judging the running state of the combine harvester by using the trained SDAE-BP model.
Further, in the step 2, the training data T is composed of n pieces of tagged history data and K pieces of untagged history data, and the tagged data is denoted as T 1 ＝{(M 1 ,y 1 ),(M 2 ,y 2 ),…,(M j ,y j ),…,(M n ,y n ) Wherein n is the total number of labeled samples, wherein M j For the j-th sample, y j The label of the j sample is the blocking fault type of the combine harvester; the data without label is denoted as T 2 ＝{M 1 ,M 2 ,…,M j ,…,M K K is the total number of unlabeled samples; m is M j ＝(m 1 ,m 2 ,…m i …,m A )，m i For the data collected by the ith speed sensor in the jth sample, the SDAE-BP model building method is as follows:
step 2.1, data processing, for sample data M j ＝(m 1 ,m 2 ,…m i …,m A ) Each acquired data of (1) is normalized by 0-1 to obtain normalized sample data X j ＝(x 1 ,x 2 ,…,x A ) The normalized dataset is
T 1X ＝{(X 1 ,y 1 ),(X 2 ,y 2 ),…,(X j ,y j ),…,(X n ,y n)}, wherein Xj Is the j-th sample after normalization, wherein y j J is more than or equal to 1 and less than or equal to n as the label of the j sample;
T 2X ＝{X 1 ,X 2 ,…,X j ,…,X K} wherein Xj J is more than or equal to 1 and less than or equal to K for the j sample after normalization;
step 2.2 Using the data set T without tag 2X Training is sequentially carried out under different Gaussian noise by utilizing an SDAE model to obtain a pre-trained SDAE model;
step 2.3, followed by a labeled dataset T 1X Importing into a pre-trained SDAE model, extracting a dataset { (h) 1 ,y 1 ),(h 2 ,y 2 ),…,(h j ,y j ),…,(h n ,y n )}，h j For the characteristic value extracted from the jth data through the SDAE model, y j For extracted tags
And 2.4, fine-tuning the pre-trained SDAE model through the BP model.
Further, in the step 2.1, the training of the SDAE-BP model comprises two processes of unsupervised pre-training and supervised fine tuning;
the unsupervised pre-training process for the SDAE is as follows:
using unlabeled samples T 2X Performing unsupervised pre-training;
a. initial training procedure
1) Given an initial Gaussian noise center of gamma 0 And gaussian noise variance σ=1, using gaussian noiseThe original input data x is destroyed to a certain extent, and the destroyed data is obtained as +.> wherein />
2) First, data containing noise is processed according to the formula (1-1)Coding to obtain deep feature codes
wherein ,sf A nonlinear activation function, wherein W is a weight matrix, and b is a bias vector;
then from the deep features using equation (1-2)Encoding h 1 Is decoded to obtain the reconstructed output of the original input data
Finally, training of the denoising self-encoder can be completed by optimizing the objective function formula (1-3);
k is the total number of unlabeled training samples until all parameters of the DAE0 model are converged; at this time, an initial optimal weight W of the DAE0 model is obtained 1 And an optimum bias value b 1 At the same time, the initial deep feature representation h of the original input data is obtained 1 The method comprises the steps of carrying out a first treatment on the surface of the Similarly, h is represented by using initial deep features 1 Training DAE1 to obtain initial optimal weight W of DAE1 model 2 Optimum bias value b 2 Characteristic representation h 2 The method comprises the steps of carrying out a first treatment on the surface of the Using the characteristic representation h 2 Training DAE2 to obtain initial optimal weight W of DAE2 model 3 Optimum bias value b 3 Characteristic representation h 3 .
b) In the multi-noise training process, the ith Gaussian noise center is set as gamma i
1) At the noise center gamma i Then the original input data x is destroyed to obtain destroyed data as
2) Continuing to train the model SDAE obtained in the last time i-1 Training, after K iterations, obtaining a new model SDAE i The method comprises the steps of carrying out a first treatment on the surface of the The weight updating mode is W 1 (i) ←W 1 (i-1) +ΔW 1 (i) ，W 2 (i) ←W 2 (i-1) +ΔW 2 (i) ，W 3 (i) ←W 3 (i-1) +ΔW 3 (i) ；
Updating the bias value to b 1 (i) ←b 1 (i-1) +Δb 1 (i) ，b 2 (i) ←b 2 (i-1) +Δb 2 (i) ，b 3 (i) ←b 3 (i-1) +Δb 3 (i) The deep level feature representation is updated to h 1 (i) ←h 1 (i-1) +Δh 1 (i) ，h 2 (i) ←h 2 (i-1) +Δh 2 (i) ，h 3 (i) ←h 3 (i-1) +Δh 3 (i) ；
3) Calculating a new Gaussian noise center:
γ i ＝γ i-1 delta gamma formula (1-4)
Wherein delta gamma is the step length of updating the Gaussian noise center, and training is continuously performed on the model for a new round according to the step b) until the S-th Gaussian noise is generated, wherein S is the total number of the Gaussian noise, and the training is finished;
supervised fine tuning training process for SDAE
Using labelled samples T 1X Performing supervised fine tuning training; the supervised fine tuning training process of the SDAE is to add a BP classification layer on the top layer of the SDAE after the layer-by-layer unsupervised training, fine tune all parameters of the whole depth network by utilizing label information of input data, and train a data set T 1X The data set { (h) is obtained after the pre-trained SDAE model 1 ,y 1 ),(h 2 ,y 2 ),…,(h j ,y j ),…,(h n ,y n )}；
Assuming that the BP network is trained using samples { (x, y) }, then the j-th neuron of the hidden layer is input as:
the output of the j-th neuron is:
O j ＝f(net j ) Formula (1-6)
F (net) in the formulas (1-6) j ) To activate the function:
wherein ,θj Representing the bias value;
the total input to the kth neuron of the output layer is:
the output of the kth neuron of the output layer is:
the error signal is calculated as:
if the output of the BP neural network is inconsistent with the expected output value y, carrying out back propagation on an error signal from an output end, and continuously correcting a weighting coefficient and a bias value by using a gradient descent method formula (1-11) and a formula (1-12) in the back propagation process;
wherein eta is learning rate, eta=0.01 until all samples are trained, obtaining weight parameters W and bias parameters b of an SDAE-BP model at the moment, and storing the SDAE-BP model.
Further, the total sample amount n of the training data T ranges from 1500 to 3000.
Further, the speed sensors are eight in number and are respectively a cutter frequency sensor, a feeding screw feeder rotating speed sensor, a conveying chain harrow rotating speed sensor, a threshing cylinder rotating speed sensor, a fan rotating speed sensor, a manuscript-by-manuscript frequency sensor, a miscellaneous stirring Long Zhuaisu sensor and a grain conveying screw feeder rotating speed sensor, wherein the cutter frequency and the manuscript-by-screw frequency are equivalently converted into rotating speeds.
Further, in the step 2.1, the data normalization method is as follows:
wherein Nmax The maximum engine speed is 2000-3000r/min.
Further, the label y of the sample j The values represent the fault type of the sample combine harvester, in particular the following table:
a combine blocking fault diagnosis system, characterized in that: the intelligent combine harvester comprises a sensor module, a data acquisition module, an interactive display screen, an audible and visual alarm device and an embedded industrial personal computer, wherein the sensor module comprises a cutter frequency sensor, a feeding auger rotating speed sensor, a conveying chain harrow rotating speed sensor, a threshing cylinder rotating speed sensor, a fan rotating speed sensor, a manuscript-by-manuscript frequency sensor, a surplus stirring Long Zhuaisu sensor and a grain conveying auger rotating speed sensor, data monitored by the sensor module are acquired through the data acquisition module, then data processing is carried out through the embedded industrial personal computer, firstly the acquired data are trained through an SDAE-BP model, a weight parameter W and a bias parameter b of the SDAE-BP model are acquired, then the SDAE-BP model is prefabricated in the embedded industrial personal computer, and then the acquired data can judge the fault type of the combine harvester through the SDAE-BP model in the embedded industrial personal computer.
The invention has the beneficial effects that:
the invention relates to a fault diagnosis system for blocking a combine harvester, which is a fault diagnosis system with high diagnosis precision and strong anti-interference capability. The invention can diagnose the state of the combine harvester on line by utilizing the diagnosis model combining the denoising self-encoder and the BP neural network by utilizing a plurality of speed sensors, plays a good early warning role, and can effectively prevent equipment from running with diseases for a long time, thereby causing low efficiency and irreversible equipment loss and causing great loss.
Drawings
FIG. 1 is a combine blocking fault diagnostic system.
Fig. 2 is a flow chart of an algorithm of the blocking fault diagnosis system of the combine harvester.
FIG. 3 speed sensor mounting position.
FIG. 4 is a schematic diagram of the training process of the SDAE-BP model.
Fig. 5 topology of a BP neural network.
FIG. 6 is a block diagram of a fault diagnosis system for a combine harvester.
Reference numerals: 1-a cutter rotating shaft, 2-a feeding auger belt wheel, 3-a fan belt wheel, 4-a conveying auger belt wheel, 5-a threshing cylinder belt wheel, 6-a grain conveying auger belt wheel, 7-a miscellaneous auger belt wheel, 8-a tensioning belt wheel, 9-a vibrating screen belt wheel and 10-a threshing cylinder bearing.
Detailed Description
As shown in FIG. 1, the blocking fault diagnosis system of the combine harvester comprises a cutter frequency sensor, a feeding screw feeder rotating speed sensor, a conveying chain harrow rotating speed sensor, a threshing cylinder rotating speed sensor, a fan rotating speed sensor, a manuscript-by-manuscript frequency sensor, a miscellaneous stirring Long Zhuaisu sensor, a grain conveying stirring Long Zhuaisu sensor, a data acquisition module, an interactive display screen, an embedded industrial personal computer and the like, wherein the cutter frequency sensor and the manuscript-by-side frequency sensor are all in reciprocating motion, so the invention adopts the rotating speed sensor to measure the frequency.
The mounting positions of the speed sensors are shown in fig. 3, and the monitoring points of the combine harvester comprise a cutter rotating shaft 1, a feeding auger belt wheel 2, a fan belt wheel 3, a conveying auger belt wheel 4, a threshing cylinder belt wheel 5, a grain conveying auger belt wheel 6, a trash auger belt wheel 7, a tensioning belt wheel 8, a vibrating screen belt wheel 9 and a threshing cylinder bearing 10.
The invention provides a combined harvester blockage fault diagnosis model, namely an SDAE-BP model, by combining a stacked denoising self-encoder and a BP neural network (Stacked Denoising Autoencoder-BP, SDAE-BP). The model extracts data characteristics through a stacked denoising self-encoder (SDAE), and realizes the fault diagnosis of blockage by combining a BP neural network, and the algorithm flow is shown in figure 2.
The working process of the blocking fault diagnosis system of the combine harvester comprises the following steps:
analog signal E () = (E) on rotation speed sensor to be mounted on 8 monitoring points of the combine harvester 1 ,e 2 ,…,e 8 ) Input to a data acquisition module, convert it to a digital signal M () = (M 1 ,m 2 ,…,m 8 ) The digital signal M ()'s are passed through SDAE-BP model to obtain the type f of the operation state of the combine harvester; and the fault state is displayed on the interactive display screen, if the fault state is currently the fault state, an alarm is sent out through the audible and visual alarm device, the position where the fault occurs is displayed on the display screen, and a driver feeds back whether the diagnosis result is correct or not through the interactive display screen. The installation position of the rotation speed sensor in the data acquisition system is shown in fig. 3, the detection position of the sensor is shown in table 1, and the installation position of the sensor is shown in fig. 1.
Table 1 rotational speed sensor detection site
| Monitoring point name | Monitoring sensor |
| Cutting knife | Speed sensor 1 |
| Feeding screw feeder | Speed sensor 2 |
| Conveying chain harrow | Speed sensor 3 |
| Disengaging roller | Speed sensor 4 |
| Fan fan | Speed sensor 5 |
| Manuscript-by-manuscript holder | Speed sensor 6 |
| Impurity auger | Speed sensor 7 |
| Grain conveying auger | Speed sensor 8 |
And (3) constructing a fault early warning model:
the training data T consists of labeled historical data and unlabeled historical data, and 2000 labeled historical data form the training data T 1 For supervised training of models, noted as
Sample T 1 ＝{(M 1 ,y 1 ),(M 2 ,y 2 ),…,(M j ,y j ),…,(M n ,y n )}，n＝2000，M j For the j-th sample, y j The label of the j sample is the blocking fault type of the combine harvester; 500 history data without labels form training data T 2 For unsupervised training of models, sample T 2 ＝{M 1 ,M 2 ,…,M j ,…,M K }，K＝500，M j ＝(m 1 ,m 2 ,…m i …,m 8 )，m i Data collected for the ith speed sensor in the jth sample.
1) Data preprocessing
Normalizing the data can improve the convergence speed, running speed and accuracy of the model, so that the sample data M is obtained by the formula (1-0) j ＝(m 1 ,m 2 ,…,m 8 ) Performing 0-1 normalization (including all training data and measured data are processed according to the method) to obtain normalized sample data X j ＝(x 1 ,x 2 ,…,x 8 )。
wherein Nmax At maximum engine speed, N max ＝2500r/min。
2) SDAE-BP algorithm
Compared to conventional single noise SDAEs, SDAEs of different gaussian noise can learn different deep feature representations of the original input data. The SDAE algorithm provided by the invention can learn the global coarse grain characteristic and the local fine grain characteristic of the original input data at the same time, so that the deep feature representation capability of the model on the rotating speed of the combine harvester is improved. FIG. 4 shows a schematic diagram of the training process of the SDAE-BP model used in this patent. It can be seen that the SDAE used in this patent is trained by the SDAE in turn under a plurality of different gaussian noise. Wherein gamma is i Is the center of the ith Gaussian noise, S is more than or equal to i is more than or equal to 1, and gamma is the same time i-1 >γ i 。
By data set T 1X and T2X Training an SDAE-BP model, and marking label-free data after normalization processing as T 2X ＝{X 1 ,X 2 ,…,X j ,…,X K}, wherein Xj Is the j-th sample after normalization, wherein y j And j is more than or equal to 1 and less than or equal to K as the label of the j sample. X is X j ＝(x 1 ,x 2 ,…x i …,x 8 )，x i And acquiring normalized data for an ith speed sensor in the jth sample.
The normalized labeled data is marked as T 1X ＝{(X 1 ,y 1 ),(X 2 ,y 2 ),…,(X j ,y j ),…,(X n ,y n)}, wherein Xj For the j-th sample, y j The label (j is more than or equal to 1 and less than or equal to n) of the j sample is the fault type of the blockage of the combine harvester. The build up of the jam fault table is shown in table 2.
Table 2 blocking fault table
The parameter settings for the SDAE-BP model are shown in Table 3.
TABLE 3 parameter settings of SDAE-BP model
3) SDAE-BP model training
The SDAE-BP model training comprises two processes of unsupervised pre-training and supervised fine tuning.
3.1 unsupervised pretraining procedure of SDAE is as follows:
using unlabeled samples T 2X Performing unsupervised pre-training;
a) Initial training procedure
1) Given an initial Gaussian noise center of gamma 0 And gaussian noise variance σ=1, use Gaussian noiseThe original input data x is destroyed to a certain extent, and the destroyed data is obtained as +.> wherein />
2) First, data containing noise is subjected to the following formula (1-1)Coding to obtain deep feature codes
wherein ,sf Is a nonlinear activation function, is a conventional sigmoid function, W is a weight matrix, and b is a bias vector.
Then h is encoded from the deep features using equation (1-2) 1 Is decoded to obtain the reconstructed output of the original input data
Finally, training of the denoising self-encoder can be completed by optimizing the objective function formula (1-3).
Where K is the total number of training samples. Until all parameters of the DAE0 model have converged; at this time, an initial optimal weight W of the DAE0 model is obtained 1 And an optimum bias value b 1 At the same time, the initial deep feature representation h of the original input data is obtained 1 The method comprises the steps of carrying out a first treatment on the surface of the Similarly, h is represented by using initial deep features 1 Training DAE1 to obtain initial optimal weight W of DAE1 model 2 Optimum bias value b 2 Characteristic representation h 2 The method comprises the steps of carrying out a first treatment on the surface of the Using the characteristic representation h 2 Training DAE2 to obtain initial optimal weight W of DAE2 model 3 Optimum bias value b 3 Characteristic representation h 3 。
b) In the multi-noise training process, the ith Gaussian noise center is set as gamma i
1) At the noise center gamma i Then the original input data x is destroyed to obtain destroyed data as
2) Continuing to train the model SDAE obtained in the last time i-1 Training, after K iterations, obtaining a new model SDAE i The method comprises the steps of carrying out a first treatment on the surface of the The weight updating mode is W 1 (i) ←W 1 (i-1) +ΔW 1 (i) ，W 2 (i) ←W 2 (i-1) +ΔW 2 (i) ，W 3 (i) ←W 3 (i-1) +ΔW 3 (i) The method comprises the steps of carrying out a first treatment on the surface of the Updating the bias value to b 1 (i) ←b 1 (i-1) +Δb 1 (i) ，b 2 (i) ←b 2 (i-1) +Δb 2 (i) ，b 3 (i) ←b 3 (i-1) +Δb 3 (i) The deep level feature representation is updated to h 1 (i) ←h 1 (i-1) +Δh 1 (i) ，h 2 (i) ←h 2 (i-1) +Δh 2 (i) ，h 3 (i) ←h 3 (i-1) +Δh 3 (i) ；
3) Calculating a new Gaussian noise center:
γ i ＝γ i-1 delta gamma formula (1-4)
Wherein delta gamma is the step length of the Gaussian noise center update, and the step (b) is returned to continue to train the model for a new round until the S-th Gaussian noise is reached, and the training is ended.
3.2 supervised Fine tuning training procedure for SDAE
Using labelled samples T 1X Performing supervised fine tuning training;
the supervised fine tuning training process of the SDAE is to add a BP classification layer on the top layer of the SDAE after layer-by-layer unsupervised training. And (3) fine-tuning all parameters of the whole depth network by using tag information of the input data. Training the dataset through an unsupervised SDAE model to obtain the dataset { (h) 1 ,y 1 ),(h 2 ,y 2 ),…,(h j ,y j ),…,(h n ,y n ) -a }; the topology of the BP neural network is shown in FIG. 5.
Assume that a BP network is trained using samples { (x, y) }. The input to the j-th neuron of the hidden layer is:
the output of the j-th neuron is:
O j ＝f(net j ) Formula (1-6)
F (net) in the formulas (1-6) j ) To activate the function:
wherein ,θj Representing the bias value.
The total input to the kth neuron of the output layer is:
the output of the kth neuron of the output layer is:
the error signal is calculated as:
if the output of the BP neural network is inconsistent with the expected output value y, the error signal is reversely transmitted from the output end, and the weighting coefficient and the bias value are continuously corrected by using a gradient descent method (1-10) in the process of reverse transmission.
Wherein eta is learning rate, eta=0.01 until all samples are trained, obtaining weight parameters W and bias parameters b of an SDAE-BP model at the moment, and storing the SDAE-BP model.
The fault early warning system flow comprises the following steps:
before the SDAE-BP model is used for diagnosing the blocking faults of the combine harvester, corresponding fault diagnosis system software is required to be deployed in the embedded industrial personal computer, wherein the system software comprises a network weight W and a bias value b of the SDAE-BP model which are acquired, preprocessed and trained offline, the number of nodes of an input layer and an output layer of the model, the learning rate eta of the model and an activation function f (net j ) Offset value theta of (2) j . The workflow of the fault diagnosis system is as follows:
step 1: current state collection of combine harvester
In order to ensure the real-time performance of the blocking fault diagnosis and improve the fault diagnosis accuracy, the sampling rate of the signal is 300Hz, and the sampling length is 200ms. I.e. data M () = (M) each time the data acquisition system delivers to the SDAE-BP 1 ,m 2 …m i …m 8 ) Is (1, 8) dimensional data.
Step 2: data preprocessing
Normalizing the sample M () 'by the formula (1-0), and obtaining normalized sample data as X' () = (X 1 ′,x 2 ′,…x i ′…,x 8′), wherein xi ' is data obtained by normalizing data acquired by an ith speed sensor, and i is more than or equal to 1 and less than or equal to 8.
Step 3: plug fault diagnosis using SDAE-BP
The SDAE-BP model running on the embedded industrial personal computer receives normalized sample data X' (-), where m i And (3) for the data 1 which is less than or equal to i and less than or equal to 8 and is acquired by the ith speed sensor, sending the sample to be diagnosed into the SDAE-BP model to obtain the operation state type f of the combine harvester. The combine blocking fault diagnosis system is shown in fig. 6.
Application example:
1. sample input to be diagnosed
And storing 8 data acquired by the data acquisition module into M (), namely, an input vector M (), of the SDAE-BP model. E.g., M () = (8.25,1342.89,987.97,1542.51,476.81,634.26,244.98,8.25), where M 1 For manuscript-by-manuscript frequency, m 2 Rotation speed, m of the miscellaneous auger 3 For the rotation speed of the grain conveying auger, m 4 Is the rotation speed of the fan, m 5 For conveying chain harrow rotation speed, m 6 To disengage the rotation speed of the drum, m 7 For feeding screw conveyer rotating speed, m 8 Is the cutter frequency.
2. Data normalization
Normalizing sample M ()'s by equation (1-0) to obtain normalized sample data of X' () = (0.0033,0.537156,0.395188,0.617004,0.190724,0.25374,0.097992,0.0033)
3. DAE-BP occlusion fault diagnosis
Taking the current X' (-) as input to the SDAE-BP model, the SDAE model passes the characteristic h=h of equation (4-1) 1 ,h 2 ,h 3 ,h 4 The method comprises the following steps:
wherein sf As Sigmoid function, W 1 ，W 2 ，W 3 Weight matrix of first layer, second layer and third layer of SDAE model respectively, b 1 ，b 2 ，b 3 The bias value matrices of the first layer, the second layer and the third layer of the SDAE model are respectively. The output characteristic h= (0.95256,0.1192,0.0174,0.88131) is calculated, and then classified by using the BP neural network, and the Sigmoid function is used as an excitation function of the BP neural network.
wherein W4 ，W 5 Weight matrix of input layer and hidden layer of BP neural network respectively, b 1 ，b 2 And the bias value matrixes of the input layer and the hidden layer of the BP neural network are respectively. f (f) g The type of occlusion fault that is ultimately calculated.
Step 4: result output
And displaying the state f of the combine harvester on a display screen, and if the state f is the current fault state, sending out an alarm through an audible and visual alarm device and displaying the position of the fault on the display screen, wherein a driver feeds back whether the diagnosis result of the fault diagnosis system of the combine harvester is correct through an interactive display screen.
The final occlusion fault type was calculated as f= (0, 1, 0) by the SDAE-BP model for input sample X' (). From table 2 it can be seen that the detection from the current monitoring point can determine that the combine is slightly jammed in failure with the take-off roller. The display is displayed on an interactive display screen, and an audible and visual alarm device is activated to light a warning lamp to generate beeping sound.
Claims (5)
1. A method for diagnosing the jam fault of combine harvester features that A speed sensors are used to monitor the combine harvesterkThe transmission part utilizes a data acquisition module to acquire data of A speed sensors and utilizes an industrial personal computer to process the data, and is characterized in that the diagnosis method comprises the following steps:
step 1, continuously acquiring analog signals of A speed sensors through a data acquisition module, and converting the analog signals into digital signals;
step 2, manually setting fault types, and collecting data of A speed sensors as training dataTUsing training dataTEstablishing an SDAE-BP model based on combination of a stacked denoising self-encoder and a BP neural network, and obtaining weight parameters of the SDAE-BP modelAnd bias parameter->；
Step 3, prefabricating or writing data processing software in the industrial personal computer according to the parameters of the data acquisition, the data processing and the SDAE-BP model;
step 4, adopting a certain sampling interval, and conveying and collecting sample data to an SDAE-BP model in the industrial personal computer by utilizing a data collecting module;
step 5, normalizing the sample data acquired in the step 4, wherein the normalized data value is between 0 and 1;
step 6, judging the running state of the combine harvester by using the trained SDAE-BP model;
in the step 2, training dataTFrom the following componentsnHistorical data with labelsKThe history data without label is formed, and the data with label is recorded as, wherein nIs the total number of samples with labels, wherein +.>Is the firstjSample number->Is the firstjLabels of the samples, namely blocking fault types of the combine harvester; the data without tag is recorded +.>，KIs the total number of samples without labels; />，/>Is the firstjThe first sample ofiThe method for establishing the SDAE-BP model by the data collected by the speed sensors comprises the following steps:
step 2.1, data processing, for sample dataIs normalized by 0-1 to obtain normalized sample data +.>The normalized dataset is
, wherein />Is normalized to the firstjSamples of whichIs the firstjLabel of individual samples->；
wherein />Is normalized to the firstjSample number->；
Step 2.2, first utilizing the data set without tagTraining is sequentially carried out under different Gaussian noise through an SDAE model to obtain a pre-trained SDAE model;
step 2.3, followed by a labeled datasetImporting to a pre-trained SDAE model, extracting a data set，/>Is the firstjCharacteristic values extracted from the sample data through an SDAE model,is the extracted label;
step 2.4, fine tuning the pre-trained SDAE model through the BP model;
in the step 2, the SDAE-BP model training comprises two processes of unsupervised pre-training and supervised fine tuning;
the unsupervised pre-training process for the SDAE is as follows:
using samples without labelsProceeding withUnsupervised pre-training;
a. initial training procedure
1) Giving the initial Gaussian noise center asAnd Gaussian noise variance->Gaussian noise is used->For original input data->Performing certain degree of destruction to obtain destroyed data of +.>, wherein />；
2) For data containing noise according to formula (1-1)Coding to obtain deep feature code ++>
Formula (1-1)
wherein ,a non-linear activation function is used,Was a matrix of weights, the weight matrix,bis a bias vector;
then encoded from deep features using equation (1-2)Is decoded to obtain the reconstructed output of the original input data；
Formula (1-2)
Finally, training of the denoising self-encoder can be completed by optimizing the objective function formula (1-3);
formula (1-3)
in the formula KThe total number of training samples without labels is counted until all parameters of the DAE0 model are converged; at this time, an initial optimal weight of the DAE0 model is obtainedAnd an optimal bias value->At the same time, the initial deep feature representation of the original input data is also obtained +.>The method comprises the steps of carrying out a first treatment on the surface of the The same applies to the representation of the initial deep features +.>Training DAE1 to obtain initial optimal weight of DAE1 model>Optimal bias value->Characteristic representation ++>The method comprises the steps of carrying out a first treatment on the surface of the Use of the characteristic representation +.>Training DAE2 to obtain initial optimal weight of DAE2 model>Optimal bias value->Characteristic representation ++>.
b) Training process of multiple noise, set upiThe Gaussian noise center of each is
1) At the center of noiseThe original input data of the lower pairxPerforming destruction to obtain destroyed data of +.>；
2) Continuing to train the model SDAE obtained in the last time i-1 Training is performed after passingKAfter a plurality of iterations, a new model SDAE is obtained i The method comprises the steps of carrying out a first treatment on the surface of the The weight updating mode is that，/>，/>；
The bias value is updated to，/>，/>The deep level feature representation is updated to +.>，/>，/>；
3) Calculating a new Gaussian noise center:
formula (1-4)
in the formula Is the step length of Gaussian noise center update, and the model is continuously trained for a new round until the first round according to the step b)SEnding training when Gaussian noise is generated;
supervised fine tuning training process for SDAE
Using labeled samplesPerforming supervised fine tuning training; the supervised fine tuning training process of the SDAE is to add a BP classification layer on the top layer of the SDAE after the layer-by-layer unsupervised training, fine tune all parameters of the whole depth network by utilizing label information of input data, and train a data set->Through pre-treatment ofObtaining data set after training SDAE model；
Assume that a sample is usedTraining BP network, hidden layerjThe inputs to the individual neurons are:
formula (1-5)
First, thejThe output of the individual neurons is:
formula (1-6)
In the formulas (1-6)To activate the function:
formula (1-7)
wherein ,representing the bias value;
output layer (th)kThe total input to the individual neurons is:
formula (1-8)
Output layer (th)kThe output of the individual neurons is:
formula (1-9)
The error signal is calculated as:
formula (1-10)
If the output of BP neural network and the expected output valueyIf the error signals are inconsistent, the error signals are reversely transmitted from the output end, and the weighting coefficients and the offset values are continuously corrected by using the gradient descent method formulas (1-11) and formulas (1-12) in the process of reverse transmission;
formula (1-11)
Formula (1-12)
in the formula For learning rate, let->Until all samples are trained, all weight parameters of the SDAE-BP model are obtained at the moment>And bias parameter->And saving the SDAE-BP model.
2. A method for diagnosing a jam fault in a combine harvester as in claim 1, wherein: the training dataTIs not included in the total sample amount of (2)nIn the range of 1500-3000.
3. A method for diagnosing a jam fault in a combine harvester as in claim 1, wherein: the speed sensors are eight in number and are respectively a cutter frequency sensor, a feeding auger rotating speed sensor, a conveying chain harrow rotating speed sensor, a threshing cylinder rotating speed sensor, a fan rotating speed sensor, a manuscript-by-manuscript-oriented sensor, a residue stirring Long Zhuaisu sensor and a grain conveying auger rotating speed sensor, wherein the cutter frequency and the manuscript-by-wire-oriented sensor are equivalently converted into rotating speeds.
4. A method for diagnosing a jam fault in a combine harvester as in claim 3, wherein: in the step 2.1, the data normalization method comprises the following steps:
wherein The maximum engine speed is 2000-3000r/min.
5. A combine-harvester-jam fault diagnosis system employing the combine-harvester-jam fault diagnosis method of any one of claims 1 to 4, characterized in that: the intelligent automatic control system comprises a sensor module, a data acquisition module, an interactive display screen and an embedded industrial personal computer, wherein the sensor module comprises a cutter frequency sensor, a feeding auger rotating speed sensor, a conveying chain harrow rotating speed sensor, a threshing cylinder rotating speed sensor, a fan rotating speed sensor, a manuscript-by-manuscript frequency sensor, a residue stirring Long Zhuaisu sensor and a grain conveying auger rotating speed sensor, data monitored by the sensor module are acquired through the data acquisition module, then data processing is carried out through the embedded industrial personal computer, firstly acquired data are trained through an SDAE-BP model, and weight parameters of the SDAE-BP model are acquiredAnd bias parameter->The SDAE-BP model is then prefabricated in the embedded stateAnd in the industrial personal computer, the fault type of the combine harvester can be judged by collecting data through an SDAE-BP model in the embedded industrial personal computer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010096661.4A CN111310830B (en) | 2020-02-17 | 2020-02-17 | Blocking fault diagnosis system and method for combine harvester |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010096661.4A CN111310830B (en) | 2020-02-17 | 2020-02-17 | Blocking fault diagnosis system and method for combine harvester |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111310830A CN111310830A (en) | 2020-06-19 |
| CN111310830B true CN111310830B (en) | 2023-10-10 |
Family
ID=71145051
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010096661.4A Active CN111310830B (en) | 2020-02-17 | 2020-02-17 | Blocking fault diagnosis system and method for combine harvester |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111310830B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112099540B (en) * | 2020-09-01 | 2023-03-31 | 四机赛瓦石油钻采设备有限公司 | Sand concentration control method based on stepping type single neuron PID control algorithm |
| CN112631250B (en) * | 2020-12-15 | 2022-06-24 | 中国计量大学 | Fault isolation and identification method in nonlinear process based on denoising autoencoder |
| CN113945380B (en) * | 2021-09-24 | 2024-04-09 | 江苏大学 | Threshing cylinder bearing vibration fault analysis method and system and combine harvester |
| CN117628417B (en) * | 2024-01-25 | 2024-03-26 | 深圳市晶湖科技有限公司 | Intelligent safety control system for gas field |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104914851A (en) * | 2015-05-21 | 2015-09-16 | 北京航空航天大学 | Adaptive fault detection method for airplane rotation actuator driving device based on deep learning |
| CN107957551A (en) * | 2017-12-12 | 2018-04-24 | 南京信息工程大学 | Stacking noise reduction own coding Method of Motor Fault Diagnosis based on vibration and current signal |
| CN207319055U (en) * | 2017-09-27 | 2018-05-04 | 黑龙江省农业机械工程科学研究院 | Combine harvester fault diagnosis system based on CAN bus |
| CN108363382A (en) * | 2018-02-09 | 2018-08-03 | 哈尔滨工业大学（威海） | A kind of complex equipment fault diagnosis method and system |
| CN108398268A (en) * | 2018-03-15 | 2018-08-14 | 哈尔滨工业大学 | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps |
| CN109000930A (en) * | 2018-06-04 | 2018-12-14 | 哈尔滨工业大学 | A kind of turbogenerator performance degradation assessment method based on stacking denoising self-encoding encoder |
| CN110779745A (en) * | 2019-10-12 | 2020-02-11 | 杭州安脉盛智能技术有限公司 | Heat exchanger early fault diagnosis method based on BP neural network |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105973595B (en) * | 2016-04-27 | 2019-03-19 | 清华大学深圳研究生院 | A kind of diagnostic method of rolling bearing fault |
| CN109690577A (en) * | 2016-09-07 | 2019-04-26 | 皇家飞利浦有限公司 | Classified using the Semi-supervised that stack autocoder carries out |
| CN106682688B (en) * | 2016-12-16 | 2020-07-28 | 华南理工大学 | Particle swarm optimization-based stacked noise reduction self-coding network bearing fault diagnosis method |
| CN108416103A (en) * | 2018-02-05 | 2018-08-17 | 武汉大学 | A kind of method for diagnosing faults of electric automobile of series hybrid powder AC/DC convertor |
| CN108918137B (en) * | 2018-06-08 | 2021-04-27 | 华北水利水电大学 | Gear box fault diagnosis device and method based on improved WPA-BP neural network |
| CN109211546B (en) * | 2018-08-28 | 2020-05-26 | 电子科技大学 | Rotary machine fault diagnosis method based on noise reduction automatic encoder and increment learning |
| CN109858345B (en) * | 2018-12-25 | 2021-06-11 | 华中科技大学 | Intelligent fault diagnosis method suitable for pipe expansion equipment |
| CN110044554A (en) * | 2019-04-16 | 2019-07-23 | 重庆大学 | A kind of online test method of the metal pressure container leakage based on acoustic emission signal |
| CN110146282A (en) * | 2019-06-06 | 2019-08-20 | 南京航空航天大学 | Epicyclic gearbox method for diagnosing faults based on multiple domain storehouse denoising autocoding network |
-
2020
- 2020-02-17 CN CN202010096661.4A patent/CN111310830B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104914851A (en) * | 2015-05-21 | 2015-09-16 | 北京航空航天大学 | Adaptive fault detection method for airplane rotation actuator driving device based on deep learning |
| CN207319055U (en) * | 2017-09-27 | 2018-05-04 | 黑龙江省农业机械工程科学研究院 | Combine harvester fault diagnosis system based on CAN bus |
| CN107957551A (en) * | 2017-12-12 | 2018-04-24 | 南京信息工程大学 | Stacking noise reduction own coding Method of Motor Fault Diagnosis based on vibration and current signal |
| CN108363382A (en) * | 2018-02-09 | 2018-08-03 | 哈尔滨工业大学（威海） | A kind of complex equipment fault diagnosis method and system |
| CN108398268A (en) * | 2018-03-15 | 2018-08-14 | 哈尔滨工业大学 | A kind of bearing performance degradation assessment method based on stacking denoising self-encoding encoder and Self-organizing Maps |
| CN109000930A (en) * | 2018-06-04 | 2018-12-14 | 哈尔滨工业大学 | A kind of turbogenerator performance degradation assessment method based on stacking denoising self-encoding encoder |
| CN110779745A (en) * | 2019-10-12 | 2020-02-11 | 杭州安脉盛智能技术有限公司 | Heat exchanger early fault diagnosis method based on BP neural network |
Non-Patent Citations (10)
| Title |
|---|
| Guoqian Jiang ; Haibo He ; Ping Xie ; Yufei Tang.Stacked Multilevel-Denoising Autoencoders: A New Representation Learning Approach for Wind Turbine Gearbox Fault Diagnosis.《IEEE Transactions on Instrumentation and Measurement》.2017,第66卷(第9期),全文. * |
| 刘景,杨光友,马志艳,陈学海.基于随机森林的联合收割机故障诊断技术.《中国农机化学报》.2019,第40卷(第10期),全文. * |
| 基于栈式降噪自编码的故障诊断;朱煜奇;黄双喜;杨天祺;孙洁香;;制造业自动化(第03期);全文 * |
| 时培明, 梁凯, 赵娜, 等.基于深度学习特征提取和粒子群支持向量机状态识别的齿轮智能故障诊断.《中国机械工程》.2017,第28卷(第9期),全文. * |
| 朱煜奇 ; 黄双喜 ; 杨天祺 ; 孙洁香 ; .基于栈式降噪自编码的故障诊断.制造业自动化.2017,第39卷(第03期),全文. * |
| 栈式稀疏加噪自编码深度神经网络的滚动轴承损伤程度诊断;陈仁祥;杨星;杨黎霞;王家序;徐向阳;陈思杨;;振动与冲击(第21期);全文 * |
| 许倩文、吉兴全、张玉振、李军、于永进.栈式降噪自编码网络在变压器故障诊断中的应用.《电测与仪表》.2018,第55卷(第17期),第62-65页. * |
| 陈仁祥 ; 杨星 ; 杨黎霞 ; 王家序 ; 徐向阳 ; 陈思杨 ; .栈式稀疏加噪自编码深度神经网络的滚动轴承损伤程度诊断.振动与冲击.2017,第36卷(第21期),全文. * |
| 飞控系统传感器故障诊断的神经网络方法研究;魏慕恒;贾秋玲;;计算机测量与控制(第01期);全文 * |
| 齐建文、陈慧贤、吴彦华、武昕伟、杨国正.《通信辐射源信号细微特征分析与处理》.北京：国防工业出版社,2015,全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111310830A (en) | 2020-06-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111310830B (en) | Blocking fault diagnosis system and method for combine harvester | |
| US11741555B2 (en) | Crop yield estimation method based on deep temporal and spatial feature combined learning | |
| CN109258509B (en) | Intelligent monitoring system and method for abnormal sound of live pigs | |
| CN108319962B (en) | Tool wear monitoring method based on convolutional neural network | |
| Xia et al. | Fault diagnosis for rotating machinery using multiple sensors and convolutional neural networks | |
| CN112015153B (en) | System and method for detecting abnormity of sterile filling production line | |
| CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
| CN110109193B (en) | Intelligent eggplant greenhouse temperature detection device based on DRNN neural network | |
| CN111583962A (en) | Sheep rumination behavior monitoring method based on acoustic analysis | |
| CN115409131B (en) | Production line abnormity detection method based on SPC process control system | |
| Li et al. | Gear pitting fault diagnosis using raw acoustic emission signal based on deep learning | |
| Tyagi et al. | A hybrid genetic algorithm and back-propagation classifier for gearbox fault diagnosis | |
| CN116702053B (en) | Agricultural machinery fault detection method based on multiple information and MCSA-SVM | |
| CN112085108B (en) | Photovoltaic power station fault diagnosis algorithm based on automatic encoder and K-means clustering | |
| CN110119165B (en) | Aquaculture pond dissolved oxygen detection device | |
| CN113484489A (en) | Remote monitoring and early warning method for water eutrophication | |
| CN114152442A (en) | Rolling bearing cross-working condition fault detection method based on migration convolutional neural network | |
| CN114707577A (en) | Anomaly detection method and system based on self-confrontation variational self-encoder | |
| CN113033833A (en) | Method for diagnosing harvesting state fault of threshing cylinder of combine harvester | |
| CN116560946A (en) | Soil pollution data pushing system based on cloud computing | |
| CN112019607A (en) | Intelligent farm comprehensive control method and system based on visual cloud platform | |
| CN111523557A (en) | Wind power intelligent fault diagnosis method based on big data | |
| CN115659778A (en) | On-line construction method and on-line updating equipment of wind turbine generator bearing fault diagnosis model | |
| CN111652867A (en) | Heavy-environment-free plant growth method based on deep morphological belief network | |
| Qiu et al. | Fault prediction of combine harvesters based on stacked denoising autoencoders. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |