CN211479145U - Combustion condition image monitoring device based on flame image and deep learning - Google Patents
Combustion condition image monitoring device based on flame image and deep learning Download PDFInfo
- Publication number
- CN211479145U CN211479145U CN202020078111.5U CN202020078111U CN211479145U CN 211479145 U CN211479145 U CN 211479145U CN 202020078111 U CN202020078111 U CN 202020078111U CN 211479145 U CN211479145 U CN 211479145U
- Authority
- CN
- China
- Prior art keywords
- cooling
- image
- interlayer
- flame
- air
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The utility model discloses a combustion condition image monitoring device based on flame images and deep learning, which comprises a cooling sleeve, an optical sight glass, an industrial camera, a water interlayer, an air interlayer, a cooling water inlet, a cooling air outlet, a power interface and a video signal interface, wherein the flame images under different combustion conditions are collected by the image monitoring device; preprocessing the flame image and then dividing the flame image into a training set, a verification set and a test set; establishing convolution sparse self-coding, and performing unsupervised training by using a training set; the trained convolution sparse self-coding is used for extracting deep features of the verification set image; establishing a Soft-max classifier, and performing supervision training by using deep features and labels of the verification set images; the trained Soft-max classifier can accurately classify deep features of the image, so that the combustion working condition recognition is realized; the method has good application prospect in the field of combustion condition monitoring.
Description
Technical Field
The utility model relates to a burning operating mode image monitoring devices based on flame image and degree of depth study belongs to burning check out test set technical field.
Background
The new energy power generation system is limited by randomness and intermittence of renewable resources, and the stable operation of a power grid can be seriously and negatively influenced by large-scale grid connection. In order to provide enough available space for new energy grid connection, a thermal generator set needs to be flexibly transformed so as to improve the deep peak regulation capacity. Based on the above, the power station boiler is possibly under frequent and rapid variable load or long-term low load working conditions for a long time, and phenomena such as difficult pulverized coal ignition, unstable combustion, even boiler fire extinguishing and the like are caused, so that combustion adjustment is difficult, combustion efficiency is reduced, and the safe and stable operation of the thermal generator set is challenged. Therefore, an accurate and reliable combustion state monitoring system is established, and the method has important significance for preventing potential danger and improving the overall performance of boiler operation.
The flame imaging monitoring method is different from the traditional flame detection device, has the advantages of interference resistance, low maintenance and the like, and is a more effective monitoring method. The method generally comprises two important steps: feature extraction and state recognition. The feature extraction is to obtain key features by utilizing an image processing technology; and the state identification is to further analyze the obtained image characteristics to realize combustion state identification. At present, common image processing techniques are mainly classified into two types: 1) and the frequency domain and time domain methods infer the combustion state change trend by analyzing the variance and the power spectral density of the gray value change of the image. The method has the defects of long response time and weak generalization ability; 2) and the machine learning method is used for extracting image characteristics in an unsupervised mode by utilizing principal component analysis or partial least square method. However, the obtained features belong to shallow information of the image, and are lack of robustness, resulting in low recognition accuracy.
The deep learning method is considered as a major breakthrough in the field of artificial intelligence, and the deep network structure of the deep learning method can mine essential characteristics of data, so that the possibility is provided for the difficult problem which cannot be solved by the traditional data driving method. The deep learning network can overcome the defect of poor performance of a shallow method, avoids a complicated process of feature selection, and can be used for processing high-dimensional and large-amount image data. Self-coding is a typical deep learning network, and can extract the non-linear characteristics of an image in an unsupervised manner. However, during the training process of the self-encoding, phenomena such as gradient disappearance or gradient explosion easily occur, and the possibility of simply copying the information of the input layer exists, so that the key information of the image cannot be really acquired. Therefore, it is still necessary to develop a more intelligent flame image monitoring and identifying method to provide guidance for combustion adjustment and optimization, so that the method has industrial utilization value.
Disclosure of Invention
In order to solve the technical problem, the utility model aims at providing a burning operating mode image monitoring devices based on flame image and degree of depth study.
The utility model discloses a burning operating mode image monitoring devices based on flame image and degree of depth study, image monitoring devices comprises cooling jacket (1), optics sight glass (2) and industry camera (3) link to each other, and cooling jacket (1) cup joints in industry camera (3) and optics sight glass (2) outside;
the top end of the cooling sleeve (1) is of a 45-degree corner structure, and the cooling sleeve (1) is divided into a water interlayer (4) and an air interlayer (5);
the top end of the optical sight glass (2) is provided with a high-temperature resistant lens with a 90-degree visual angle;
the tail end of the industrial camera (3) is provided with a power interface (9) and a video signal interface (10).
Furthermore, the water interlayer (4) is of a sealing structure, the air interlayer (5) is of an open structure, and the two cooling modes of water cooling and air cooling correspond to each other in sequence.
Furthermore, the air cooling air interlayer (5) is arranged on the inner side, the water cooling water interlayer (4) is arranged on the outer side, the tail end of the cooling sleeve (1) is provided with a cooling water inlet (6) and a cooling air inlet (7), the top end of the cooling sleeve (1) is provided with a cooling air outlet (8), the cooling water inlet (6) is communicated with the water interlayer (4), and the cooling air inlet (7) and the cooling air outlet (8) are communicated with the air interlayer (5).
Compared with the prior art for monitoring the combustion working condition, the utility model discloses a method and device for monitoring the combustion working condition based on flame image and deep learning have following advantage:
1. the image monitoring device provided by the utility model has a high-efficiency cooling effect, and provides guarantee for the stable operation of the image acquisition system; the essential characteristics of the flame image can be accurately extracted without expert prior knowledge; the new self-coding loss function effectively solves the defects of difficult training and poor generalization performance; and the Soft-max classifier adopted is suitable for image feature classification, and accurate identification of the combustion working condition can be realized.
The above description is only an overview of the technical solution of the present invention, and in order to make the technical means of the present invention clearer and can be implemented according to the content of the description, the following detailed description is made with reference to the preferred embodiments of the present invention and accompanying drawings.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate a certain embodiment of the present invention, and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
Fig. 1 is the utility model relates to a burning operating mode image monitoring devices's principle flow schematic diagram based on flame image and degree of depth study.
FIG. 2 is a schematic structural diagram of a combustion condition image monitoring device based on flame images and deep learning.
Fig. 3 is a schematic diagram of a convolutional sparse self-encoding (CSAE) network structure in a combustion condition image monitoring device testing method based on a flame image and deep learning.
Fig. 4 is a data set architecture of an embodiment of the present invention.
Fig. 5 is a schematic diagram of training loss of a convolutional sparse self-encoding (CSAE) network according to an embodiment of the present invention.
Wherein, in the figure, 1, a cooling sleeve; 2. an optical sight glass; 3. an industrial camera; 4. a water interlayer; 5. a gas interlayer; 6. a cooling water inlet; 7. a cooling air inlet; 8. a cooling air outlet; 9. a power interface; 10. and a video signal interface.
Detailed Description
The following detailed description of the embodiments of the present invention is provided with reference to the accompanying drawings and examples. The following examples are intended to illustrate the invention, but are not intended to limit the scope of the invention.
Referring to fig. 1 to 5, the following description of a preferred embodiment of the present invention will be made in detail with reference to the accompanying drawings and attached tables:
a combustion condition image monitoring device based on flame images and deep learning comprises a cooling sleeve (1), an optical sight glass (2) and an industrial camera (3), wherein the optical sight glass (2) is connected with the industrial camera (3), and the cooling sleeve (1) is sleeved on the outer sides of the industrial camera (3) and the optical sight glass (2); the top end of the cooling sleeve (1) is of a 45-degree corner structure, and the cooling sleeve (1) is divided into a water interlayer (4) and an air interlayer (5); the water interlayer (4) is of a sealing structure, the air interlayer (5) is of an open structure, and the water interlayer and the air interlayer sequentially correspond to two cooling modes of water cooling and air cooling. The air cooling air interlayer (5) is arranged on the inner side, the water cooling water interlayer (4) is arranged on the outer side, the tail end of the cooling sleeve (1) is provided with a cooling water inlet (6) and a cooling air inlet (7), the top end of the cooling sleeve (1) is provided with a cooling air outlet (8), the cooling water inlet (6) is communicated with the water interlayer (4), and the cooling air inlet (7) and the cooling air outlet (8) are communicated with the air interlayer (5). The top end of the optical sight glass (2) is provided with a high-temperature resistant lens with a 90-degree visual angle; the tail end of the industrial camera (3) is provided with a power interface (9) and a video signal interface (10).
The utility model discloses a theory of operation is:
in actual detection, the schematic flow chart is shown in fig. 1, and specifically includes the following steps:
L＝LSPT+LMSE+LSSIM(1)
in the formula, LSPTRepresenting a sparse penalty term, LMSERepresents the mean square error, LSSIMRepresenting structural similarity;
and 4, classifying new image features by the trained Soft-max classifier, identifying the combustion condition of a single flame image, and inspecting the performance of the single flame image by using a test set.
The image monitoring device in the step 1:
the device consists of a cooling sleeve, an optical sight glass and an industrial camera. The cooling sleeve is different from the traditional structure, adopts a unique 45-degree corner design, effectively overcomes the limitation of the installation position, and has two cooling modes of air cooling and water cooling so as to ensure the stable operation of the image acquisition system. The optical sight glass is provided with a high-temperature resistant lens with a 90-degree visual angle and completely covers a combustion main reaction area.
The feature extraction step of the convolution sparse self-encoding (CSAE) network in the step 2 comprises the following steps:
step 2.1, the training set X is sent to convolutional coding, firstly, feature extraction is carried out by k1 convolutional filters (C1(k1@ a1 × a1+ s1)) with the window size of a1 × a1 and the step size of s1, then neuron activation is carried out by using a ReLU function (y (X) ═ max (0, X)), and finally feature dimensionality reduction is carried out by a maximum pooling layer (P1(b1 × b1+ f1)) with the window size of b1 × b1 and the step size of f1, so that a feature vector h is obtained1；
Step 2.2, feature vector h1Sending the data to an encoder No. 2 and an encoder No. 3 for further processing, wherein the processing process is similar to that of the encoder No. 1, and finally obtaining a feature vector h3；
Step 2.3, feature vector h3Sending the data to a decoder No. 1, realizing characteristic dimension promotion by an upsampling layer (U1(g1 × g1)) with the window size of g1 × g1, and then respectively processing the upsampling layer by a convolution filter (C4(k4@ a4 × a4+ s4) and a ReLU activation function to obtain a characteristic vector h4；
Step 2.4, feature vector h4Sending the data to a decoder No. 2 and a decoder No. 3 for further processing, wherein the processing process is similar to that of the decoder No. 1, and finally obtaining a reconstructed image Z; it is noted that the activation function of decoder No. 3 uses Sigmoid function (y (x) ═ 1/1+ e)-x) To ensure that the input image and the output image are consistent in numerical range.
The sparse penalty term L in the step 2SPTExpressed as:
wherein β represents sparse rate, F represents hidden layer neuron number, p represents sparse target constant, and average neuron activation amount p based on i axisjExpressed as:
in the formula, E represents the number of images in the training set; sij(i ∈ (1, E), j ∈ (1, F)) represents the activation amount of the neuron at the j-th position, and the KL divergence is expressed as:
the loss function L in step 2MSEExpressed as:
in the formula, XijAnd ZijRepresenting the gray scale at the (i, j) th position in the input image and the reconstructed image, respectively, of size a × T.
The loss function L in step 2SSIMExpressed as:
in the formula, c1＝(k1r)2And c2＝(k2r)2，k1And k is2Is a constant less than 1, r represents the dynamic range of the image gray scale; mu.sXAnd muZRepresents an average of the input image and the reconstructed image; sigmaXAnd σZRepresenting a variance of the input image and the reconstructed image; sigmaXZRepresenting the covariance of the input image and the reconstructed image.
The Soft-max classifier in the step 3 is defined as:
in the formula, yiDenotes the ith sample xiAn output of (d); k represents the number of classes of the training sample.
Example 1
The utility model discloses burning operating mode monitoring devices and method based on flame image recognition, including following step:
step 1) collecting flame images under three combustion conditions by using an image monitoring device, wherein the image resolution is 384 x 260 x 3. Fig. 4 shows the data set structure of the present embodiment. The data set contains 1500 (500 for each condition), and after image preprocessing (image size compression to 256 × 3, and normalization to a range of 0-1), is divided into a training set, a validation set, and a test set. The training set contains 1080(360 × 3) unlabeled images for unsupervised training of the CSAE network; the verification set comprises 120(40 multiplied by 3) label images and is used for supervising and training the Soft-max classifier; the test set contained 300 (100X 3) label images for CSAE-Soft-max model performance testing.
Step 2) the structure of the convolution sparse self-coding network is shown in fig. 3, and the network parameters are summarized in table 1. The sparse penalty term, the mean square error and the structural similarity are combined to be used as a loss function of the CSAE, and the training loss under different training times is shown in FIG. 5. The result shows that the loss value of the CSAE network is converged at the 80 th training time, so that the training time is determined to be 80 times;
| |
|
|
|
|
|
| C1(8@3×3+1) | C2(4@3×3+1) | C3(1@3×3+1) | U1(32×32) | U2(64×64) | U3(128×128) |
| ReLU | ReLU | ReLU | C4(4@3×3+1) | C5(4@3×3+1) | C6(3@3×3+1) |
| P1(2×2+2) | P2(2×2+2) | P3(2×2+2) | ReLU | ReLU | Sigmiod |
TABLE 1
Step 3) extracting deep-layer characteristics h of verification set images by using the trained CSAE network3The characteristic dimensionality is 64, and the supervision training is carried out on the Soft-max classifier by combining with the corresponding image label;
and 4) the trained Soft-max classifier can classify new image features to realize the combustion condition recognition of a single flame image. The performance of the device is tested by using a test set, and the identification precision reaches 99.3%.
Experimental result shows, the utility model discloses the intelligent monitoring model of the sparse self-encoding network of convolution established can accurately discern the burning operating mode of sola flame image.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and it should be noted that, for those skilled in the art, a plurality of modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (3)
1. The utility model provides a burning operating mode image monitoring devices based on flame image and degree of depth study which characterized in that: the image monitoring device consists of a cooling sleeve (1), an optical sight glass (2) and an industrial camera (3), wherein the optical sight glass (2) is connected with the industrial camera (3), and the cooling sleeve (1) is sleeved outside the industrial camera (3) and the optical sight glass (2);
the top end of the cooling sleeve (1) is of a 45-degree corner structure, and the cooling sleeve (1) is divided into a water interlayer (4) and an air interlayer (5);
the top end of the optical sight glass (2) is provided with a high-temperature resistant lens with a 90-degree visual angle;
the tail end of the industrial camera (3) is provided with a power interface (9) and a video signal interface (10).
2. The combustion condition image monitoring device based on the flame image and the deep learning as claimed in claim 1, wherein: the water interlayer (4) is of a sealing structure, the air interlayer (5) is of an open structure, and the water interlayer and the air interlayer sequentially correspond to two cooling modes of water cooling and air cooling.
3. The combustion condition image monitoring device based on the flame image and the deep learning as claimed in claim 1 or 2, wherein: the air interlayer (5) is arranged on the inner side, the water-cooling water interlayer (4) is arranged on the outer side, the tail end of the cooling sleeve (1) is provided with a cooling water inlet (6) and a cooling air inlet (7), the top end of the cooling sleeve (1) is provided with a cooling air outlet (8), the cooling water inlet (6) is communicated with the water interlayer (4), and the cooling air inlet (7) and the cooling air outlet (8) are communicated with the air interlayer (5).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020078111.5U CN211479145U (en) | 2020-01-15 | 2020-01-15 | Combustion condition image monitoring device based on flame image and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202020078111.5U CN211479145U (en) | 2020-01-15 | 2020-01-15 | Combustion condition image monitoring device based on flame image and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN211479145U true CN211479145U (en) | 2020-09-11 |
Family
ID=72372113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202020078111.5U Active CN211479145U (en) | 2020-01-15 | 2020-01-15 | Combustion condition image monitoring device based on flame image and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN211479145U (en) |
-
2020
- 2020-01-15 CN CN202020078111.5U patent/CN211479145U/en active Active
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110163278B (en) | Flame stability monitoring method based on image recognition | |
| CN111401126A (en) | Combustion condition monitoring method and device based on flame image and deep learning | |
| CN109086793B (en) | Abnormity identification method for wind driven generator | |
| CN109359697A (en) | Graph image recognition methods and inspection system used in a kind of power equipment inspection | |
| Jin et al. | A multi-scale convolutional neural network for bearing compound fault diagnosis under various noise conditions | |
| CN116625438A (en) | Gas pipe network safety on-line monitoring system and method thereof | |
| CN110765587A (en) | Complex petrochemical process fault diagnosis method based on dynamic regularization judgment local retention projection | |
| CN115953666B (en) | Substation site progress identification method based on improved Mask-RCNN | |
| CN113032912A (en) | Ship diesel engine fault detection method based on association rule | |
| Gu et al. | An improved sensor fault diagnosis scheme based on TA-LSSVM and ECOC-SVM | |
| CN110909881A (en) | Knowledge representation method for cross-media knowledge reasoning task | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| CN116821811A (en) | Coal mill unit fault diagnosis method and system based on multi-layer graph convolution neural network | |
| Han et al. | A hybrid deep neural network based prediction of 300 MW coal-fired boiler combustion operation condition | |
| CN115314287B (en) | Counter anomaly detection system based on deep clustering | |
| CN116451022A (en) | Adaptive bearing fault diagnosis method based on depth discrimination reactance domain | |
| CN211479145U (en) | Combustion condition image monitoring device based on flame image and deep learning | |
| Daogang et al. | Anomaly identification of critical power plant facilities based on YOLOX-CBAM | |
| Xu et al. | A novel positive–negative graph convolutional network-based fault diagnosis method with application to complex systems | |
| CN111539950A (en) | Solar panel defect detection system based on machine vision | |
| CN116844016A (en) | Flame detection method based on Fire-MCANet | |
| CN115481656A (en) | Diesel engine fuel injector fault diagnosis method based on vibration and rail pressure signals | |
| Lyu et al. | Image-based process monitoring using deep belief networks | |
| CN115830701A (en) | Human violation behavior prediction method based on small sample learning | |
| CN116994104B (en) | Zero sample identification method and system based on tensor fusion and contrast learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| GR01 | Patent grant | ||
| GR01 | Patent grant |