CN114037916A - Visual fault identification and positioning method for high-pressure air cooler of petrochemical device - Google Patents
Visual fault identification and positioning method for high-pressure air cooler of petrochemical device Download PDFInfo
- Publication number
- CN114037916A CN114037916A CN202111246542.3A CN202111246542A CN114037916A CN 114037916 A CN114037916 A CN 114037916A CN 202111246542 A CN202111246542 A CN 202111246542A CN 114037916 A CN114037916 A CN 114037916A
- Authority
- CN
- China
- Prior art keywords
- air cooler
- pressure air
- tube bundle
- convolution
- block
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000000007 visual effect Effects 0.000 title claims abstract description 24
- 238000001931 thermography Methods 0.000 claims abstract description 55
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 238000001514 detection method Methods 0.000 claims abstract description 26
- 238000010586 diagram Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000012423 maintenance Methods 0.000 claims abstract description 15
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000011176 pooling Methods 0.000 claims description 14
- 238000005406 washing Methods 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000004519 manufacturing process Methods 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 6
- 238000002347 injection Methods 0.000 claims description 6
- 239000007924 injection Substances 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000000903 blocking effect Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000003449 preventive effect Effects 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 2
- 238000005520 cutting process Methods 0.000 claims description 2
- 238000011010 flushing procedure Methods 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 238000003860 storage Methods 0.000 claims description 2
- 230000001131 transforming effect Effects 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims 4
- 239000000463 material Substances 0.000 abstract description 3
- 238000005984 hydrogenation reaction Methods 0.000 description 2
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- NINIDFKCEFEMDL-UHFFFAOYSA-N Sulfur Chemical compound [S] NINIDFKCEFEMDL-UHFFFAOYSA-N 0.000 description 1
- 150000003863 ammonium salts Chemical class 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 239000010779 crude oil Substances 0.000 description 1
- 238000002425 crystallisation Methods 0.000 description 1
- 230000008025 crystallization Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 229910052739 hydrogen Inorganic materials 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 229910052717 sulfur Inorganic materials 0.000 description 1
- 239000011593 sulfur 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device comprises the following steps: preprocessing a tube bundle infrared thermal imaging image of the high-pressure air cooler; constructing a neural network for extracting the characteristics of the infrared thermal imaging image of the high-pressure air cooler tube bundle; constructing a neural network for identifying a fault target and positioning and detecting a fault area of the high-pressure air cooler tube bundle infrared thermal imaging image; training a neural network for extracting the characteristics of the infrared thermal imaging image of the high-pressure air cooler tube bundle, and training the neural network for identifying the fault target and positioning and detecting the fault area of the infrared thermal imaging image of the high-pressure air cooler tube bundle by using the obtained image characteristic diagram to form a model; detecting and identifying an infrared visual image acquired by an infrared video acquisition device; and controlling the water supply valve according to the detection and identification result, and formulating a maintenance strategy. The invention realizes the automatic identification of the air cooler tube bundle fault by the unmanned aerial vehicle, so that the detection of the air cooler tube bundle in the operation stage is more comprehensive and delicate, a large amount of manpower and material resources are saved, and the safe operation and maintenance cost is reduced.
Description
Technical Field
The invention relates to a fault identification and positioning method. In particular to a visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device.
Background
The petrochemical industry is a high-energy-consumption industry, and little improvement in any production process can bring huge economic benefits. The air cooler is one of the common equipment in petrochemical production, the air cooler makes air and air cooler tube bank produce the transverse flow by large-scale air-blower and cools off the intraductal medium, and the air cooler is the high-pressure air cooler of the hydrogenation distillate in the petrochemical plant mostly, work in high pressure, in the hydrogen and corrosive medium environment, the operating mode condition is harsher, especially in recent years, under the many processing high sulfur crude oil circumstances of petrochemical enterprise, the hydrogenation distillate in the high-pressure air cooler tube bank very easily generates the ammonium salt crystallization under the low temperature and separates out, cause the scale deposit of air cooler tube bank, can block up the pipeline after the dirt in the tube bank piles up to a certain extent, influence technology production, the corrosion problem under the dirt is easily taken place to the tube bank metal simultaneously, lead to the air cooler to take place the tube bank leakage failure, cause major incident.
Therefore, the scaling condition of the tube bundle of the air cooler of the petrochemical device is detected, the tube bundle with serious scaling is found in time, and a worker is reminded to take corresponding measures in time, which is particularly important.
At the present stage, the general air cooler tube bank detects mostly that the manual hand-held infrared thermal imager detects near the air cooler tube bank, but this kind of method can only detect the partial tube bank of whole air cooler, and this kind of mode is not careful comprehensive not only, has wasted a large amount of manpower and materials moreover, consequently, develops a petrochemical industry device high pressure air cooler infrared machine vision fault discernment and positioner and just seems especially necessary. To date, no complete system and method exists for identifying and positioning the fault target of the air cooler tube bundle of the large petrochemical device.
Disclosure of Invention
The invention aims to solve the technical problem of providing a visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device, which can realize the predictive maintenance of an air cooler pipe bundle and overcome the defects of the prior art.
The technical scheme adopted by the invention is as follows: a visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device comprises the following steps:
1) collecting a tube bundle infrared thermal imaging image of the high-pressure air cooler by an infrared thermal imager, and preprocessing the infrared thermal imaging image;
2) constructing a neural network for extracting the characteristics of the infrared thermal imaging image of the high-pressure air cooler tube bundle;
3) extracting a neural network by combining the characteristics of the constructed high-pressure air cooler tube bundle infrared thermal imaging image, and constructing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network;
4) training the built high-pressure air cooler tube bundle infrared thermal imaging image feature extraction neural network by using the training set in the standard data set file obtained in the step 1) to obtain an image feature diagram, and then training the high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network by using the image feature diagram to obtain a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
5) detecting and identifying an infrared visual image acquired by an infrared video acquisition device by utilizing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
6) and controlling the water supply valve according to the detection and identification result, and formulating a maintenance strategy to realize preventive maintenance of the high-pressure air cooler.
The visual fault identification and positioning method for the high-pressure air cooler of the petrochemical device has the following advantages:
1. the invention realizes that the unmanned aerial vehicle automatically identifies the failure of the air cooler tube bundle, so that the detection of the air cooler tube bundle in the operation stage is more comprehensive and delicate, and a large amount of manpower and material resources are saved;
2. the invention can assist the target image data to identify more effectively by using the marked sample of the target image under the condition of limited image data set.
3. According to the automatic identification method, the working personnel are reminded to take corresponding measures to the fault tube bundle in time according to the automatically identified tube bundle scaling degree and the tube bundle fault distribution positioning diagram, and the predictive maintenance of the air cooler tube bundle is realized;
4. before the large-scale maintenance of the equipment, the maintenance scheme is determined according to the scaling and blocking conditions of a certain tube bundle in the air cooler tube bundle fault distribution positioning diagram in different periods, the safe operation and maintenance cost is reduced, and the long-period operation of the air cooler is realized.
Detailed Description
The visual fault identification and positioning method for the high-pressure air cooler of the petrochemical device is described in detail with reference to the embodiment.
The invention discloses a visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device, which comprises the following steps of:
1) collecting a tube bundle infrared thermal imaging image of the high-pressure air cooler by an infrared thermal imager, and preprocessing the infrared thermal imaging image; the preprocessing of the infrared thermal imaging image comprises the following steps:
(1) numbering and naming the infrared thermal imaging images in a naming format of 1. jpg-n.jpg to form a high-pressure air cooler tube bundle infrared thermal imaging image group;
(2) utilizing an image characteristic judgment method in the air cooler tube bundle infrared detection method to perform scaling and blockage fault identification on the high-pressure air cooler tube bundle infrared thermal imaging image group: when the heat exchange tubes in the tube bundle are scaled, the corresponding parts in the image are displayed as green areas, when the heat exchange tubes in the tube bundle are blocked, the corresponding parts in the image are displayed as blue areas, and when the heat exchange tubes in the tube bundle normally run, the tube bundle in the image is displayed as yellow areas;
(3) establishing a high-pressure air cooler tube bundle infrared anomaly detection data set, marking the occurred infrared anomaly into 3 types, and dividing into the following steps according to the abnormal form: dividing the abnormal length which is less than or equal to the length 1/10 of the heat exchange tube into short strips, dividing the abnormal length between 1/10 and 1/4 of the length of the heat exchange tube into long strips, and dividing the abnormal length which is more than or equal to the length 1/4 of the heat exchange tube into long strips;
(4) utilizing a labelling software tool to frame a green area and a blue area in each image in the high-pressure air cooler tube bundle infrared thermal imaging image group by adopting a rectangular marking frame, generating an XML file corresponding to the image in the high-pressure air cooler tube bundle infrared thermal imaging image group for storage, wherein the XML file comprises the file name, the size, the rectangular marking frame position and the marked target category information of the image, and finally, marking the marked data file according to 70: 15: 15 into a training set, a validation set and a test set, and establishing a standard data set file.
2) Constructing a neural network for extracting the characteristics of the infrared thermal imaging image of the high-pressure air cooler tube bundle;
the high-pressure air cooler tube bundle infrared thermal imaging image feature extraction neural network is of a residual convolution neural network ResNet50 structure and comprises 1 image preprocessing block and 5 stacking blocks divided from top to bottom, wherein the 1 st stacking block is formed by a standard convolution layer and a maximum pooling layer, the 2 nd stacking block is a residual module comprising 3 convolution blocks, the 3 rd stacking block is a residual module comprising 4 convolution blocks, the 4 th stacking block is a residual module comprising 6 convolution blocks, the 5 th stacking block is a residual module comprising 3 convolution blocks, 1 average pooling and 1 full connection layer. Wherein,
(1) the image preprocessing block is used for cutting and edge expanding the received image and changing the original image into an image of 227 multiplied by 3;
(2) the 1 st stacked block comprises 64 convolution kernels and 64 max pooling devices;
(3) the 2 nd stacking block is divided into two paths, one path is an expanded dimension convolution kernel, the other path comprises 3 convolution blocks, each convolution block is provided with three convolution layers from top to bottom, the 1 st layer is 64 convolution kernels of 1 × 1, the 2 nd layer is 64 convolution kernels of 3 × 3, the 3 rd layer is 256 convolution kernels of 1 × 1, the output result of the 1 st convolution block is added with the output result of the expanded dimension convolution kernel and then is used as the input of the 2 nd stacking block, the output result of the 2 nd convolution block is added with the output result of the expanded dimension convolution kernel and then is used as the input of the 3 rd stacking block, and the output result of the 3 rd stacking block enters the 3 rd stacking block;
(4) the 3 rd stacking block comprises: a dimension-extended convolution kernel and 4 volume blocks from top to bottom, wherein each volume block has three volume layers from top to bottom, the 1 st volume layer is 128 convolution kernels of 1 × 1, the 2 nd volume layer is 128 convolution kernels of 3 × 3, the 3 rd volume layer is 512 convolution kernels of 1 × 1, the output result of the 1 st volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 2 nd volume block, the output result of the 2 nd volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 3 rd volume block, the output result of the 3 rd volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 4 th volume block, and the output of the 4 th volume block enters the 4 th stack block;
(5) the 4 th stacking block comprises: a dimension-expanding convolution kernel and 6 convolution blocks from top to bottom, wherein each convolution block has three convolution layers from top to bottom, the 1 st convolution layer is 256 convolution kernels of 1 × 1, the 2 nd convolution layer is 256 convolution kernels of 3 × 3, the 3 rd convolution layer is 1024 convolution kernels of 1 × 1, the output result of the first 1 convolution block in the 6 convolution blocks is added with the output result of the dimension-expanding convolution kernel to obtain a result which is used as the input of the next convolution block, and the output result of the 6 th convolution block enters the 5 th stacking block;
(6) the 5 th stack includes: a dimension-extended convolution kernel and 3 convolution blocks from top to bottom, wherein each convolution block has three convolution layers from top to bottom, the 1 st convolution layer is 512 convolution kernels of 1 × 1, the 2 nd convolution layer is 512 convolution kernels of 3 × 3, the 3 rd convolution layer is 2048 convolution kernels of 1 × 1, the output result of the 1 st convolution block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 2 nd convolution block, the output result of the 2 nd convolution block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 3 rd convolution block, and the output result of the 3 rd convolution block is the extracted feature;
each 1 convolutional layer in each stack is activated using a batch normalization operation, the output of the 1 st convolutional layer and the output of the 2 nd convolutional layer of each convolutional block in each stack are activated using a Relu function, and the output of each convolutional block is activated using a Relu function, wherein the batch normalization operation is formulated as follows:
wherein xaiFor the input of a batch of n elements, x, at layer a of the convolutional networkai＝{xa1,xa2,…,xan}，μaIs the average value of the elements of one batch,
is the variance of the elements of a batch,
zeta is the term that guarantees a non-zero denominator, y, as a result of normalization of all elements of the same batchaiFor the final output, γaiFor scale scaling, betaaiA parameter of the offset;
the activation function Relu is expressed as:
wherein, f (y)ai) The Relu activation function value, which is the output of layer a of the convolutional network.
3) Extracting a neural network by combining the characteristics of the constructed high-pressure air cooler tube bundle infrared thermal imaging image, and constructing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network;
the method comprises the steps of constructing a neural network of the high-pressure air cooler tube bundle infrared thermal imaging image fault target recognition and fault region positioning detection, wherein the neural network is based on a FasterR-CNN structure and comprises four parts, namely a high-pressure air cooler tube bundle infrared thermal imaging image feature extraction neural network, a region production module RPN, a region suggestion pooling module and a region method module RCNN based on convolutional neural network features. The following steps:
(1) the region production module RPN comprises the following operation processes:
(1.1) sliding on a feature map extracted by a neural network for extracting the features of the infrared thermal imaging image of the high-pressure air cooler tube bundle by using a 3 x 2048 sliding window, finding out the pixel position corresponding to each three-dimensional sliding window on an original image, then generating k anchor frames on the original image, wherein the centers of the k anchor frames are superposed and are the same as the centers corresponding to the three-dimensional sliding windows, calculating the offset between the center of each anchor frame and the coordinate at each coordinate point in the feature map, and calculating all the anchor frame positions in each feature map through the relative offset between the anchor frames and the generated coordinates;
(1.2) inputting all anchor frame positions in each feature map into 1 x 1 convolutional layer, outputting 2k x 20 x 16 anchor frame image features by the 1 x 1 convolutional layer, then mapping the 2k x 20 x 16 anchor frame image features into 2 x 20k x 16 by using a reshaping function reshape to obtain a foreground background score of the anchor frame, sending the result into a softmax function for probability calculation, transforming the obtained features into the 2k x 20 x 16 anchor frame image features, and finally outputting the probability that each anchor frame belongs to the foreground and the background; inputting all the anchor frame positions in each feature map into another 1 × 1 convolutional layer, wherein the 1 × 1 convolutional layer outputs 4k × 20 × 16 features, 4k includes predictions of k anchor frames, each anchor frame has 4 data respectively representing the horizontal and vertical coordinates of the center point of each anchor frame and the width and height of each anchor frame, and the 4 offset values relative to the rectangular mark frame;
(1.3) the RPN module classification and the rectangular labeling box calculation process of the regression offset are as follows: calculating the intersection and intersection ratio IOU (A) of the rectangular marking frame and the generated anchor framei，Bj) Cross-over ratio IOU (A)i，Bj) Expressed as:
wherein IOU (A)i，Bj) Representing the intersection ratio of the area of the ith rectangular marking frame marked in the image and the jth anchor frame; a. theiRepresenting the area of the ith rectangular labeling frame labeled in the image; b isjRepresenting the area of the generated jth anchor frame;
(1.4) determining whether each anchor frame is a positive sample or a negative sample according to the cross-over ratio, wherein the positive and negative sample judgment criteria are as follows: intersection ratio IOU (A) with all rectangular label boxesi，Bj) All anchor frames smaller than 0.3 are determined as negative samples; intersection ratio IOU (A) of any one rectangular mark boxi，Bj) The anchor frame which is greater than or equal to 0.7 is a positive sample; the anchor frames that do not belong to either the positive or negative samples are ignored;
(1.5) calculating the integral loss function of the RPN module and optimizing parameters of each layer of the network according to the loss function;
the loss function of the RPN module comprises classification loss and regression loss, and the formula is as follows:
wherein
Representing the classified cross entropy loss of the screened anchor frame; pkA category truth value for the nth anchor box;
for the prediction class of the nth anchor frame, when the nth anchor frame is a positive sample
Value 1, negative sample
A value of minus 1; n is a radical ofclsTo classify the loss weight value, NregIs the weight value of the regression loss,
represents the regression loss, tk＝{tx，ty，tw，thThe distance between the center coordinate and the width and the height of the anchor frame relative to the center coordinate and the width and the height of the rectangular marking frame,
as the center coordinates and width and height phases of the anchor framePredicting values of the center coordinates and the width and height distances of the rectangular marking frame;
(1.6) carrying out parameter training on the RPN module according to a rectangular marking frame calibrated in the training set of the high-pressure air cooler tube bundle infrared thermal imaging image, alternately optimizing the RPN module parameters by a random gradient descent method, fixing the regression parameters when optimizing the classification parameters, and fixing the classification parameters when optimizing the regression parameters;
(2) the operation steps of the area suggestion pooling module are as follows:
taking 256 anchor frames out of the anchor frames forming the positive sample and the negative sample, mapping the anchor frames to a characteristic diagram extracted by a neural network through the high-pressure air cooler tube bundle infrared thermal imaging image characteristic extraction, equally dividing a characteristic area mapped by each anchor frame into squares with the size of 7 multiplied by 7, selecting 4 sampling points in each small square, then respectively carrying out bilinear interpolation on 4 points in each small square through a bilinear interpolation method to obtain 4 values, then carrying out maximized pooling to obtain the value of each small square, and obtaining a maximized pooled characteristic diagram;
(3) the operation process of the region method module RCNN based on the convolutional neural network characteristics comprises the following steps:
and (3) performing full connection operation on the feature graph after the maximized pooling, predicting the classification of each anchor frame in the 256 anchor frames, predicting the offset of each anchor frame and the corresponding rectangular marking frame, wherein the offset calculation is the same as the (1.3) RPN module classification and the rectangular marking frame calculation method of regression offset.
4) Training the built high-pressure air cooler tube bundle infrared thermal imaging image feature extraction neural network by using the training set in the standard data set file obtained in the step 1) to obtain an image feature diagram, and then training the high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network by using the image feature diagram to obtain a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
5) detecting and identifying an infrared visual image acquired by an infrared video acquisition device by utilizing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
6) and controlling the water supply valve according to the detection and identification result, and formulating a maintenance strategy to realize preventive maintenance of the high-pressure air cooler. The maintenance strategy making comprises the following steps:
(1) according to the testing result, form the distribution positioning diagram of high-pressure air cooler tube bank trouble, according to the bundle scale deposit degree of discernment, remind the staff in time to take corresponding measure, specifically be:
when the fault is a short strip-shaped area and the accumulation of the areas on a single tube bundle is less than one third of the length of the tube bundle, the opening of a water injection valve of the high-pressure air cooler is increased for flushing;
when the faults appearing in the detection process are short strip-shaped areas and long strip-shaped areas appearing on a single tube bundle, the sum of the short strip-shaped areas and the long strip-shaped areas is more than two thirds of the length of the tube bundle, the opening of a water injection valve of the high-pressure air cooler is increased for washing, and when the faults are detected after washing, the fault areas are not reduced, high-pressure water washing is recommended to be carried out to dredge the tube bundle when the machine is shut down;
when the fault is a large strip-shaped area on a single tube bundle, the opening of a water injection valve of the high-pressure air cooler is completely opened for washing, and when the fault is detected after washing, the area is not reduced, and the treatment of pipe blocking is recommended;
(2) before the large overhaul of the equipment, the overhaul scheme is determined according to the scaling and blocking conditions of the tube bundle with the fault in the high-pressure air cooler tube bundle fault distribution positioning diagram in different periods.
Claims (8)
1. A visual fault identification and positioning method for a high-pressure air cooler of a petrochemical device is characterized by comprising the following steps:
1) collecting a tube bundle infrared thermal imaging image of the high-pressure air cooler by an infrared thermal imager, and preprocessing the infrared thermal imaging image;
2) constructing a neural network for extracting the characteristics of the infrared thermal imaging image of the high-pressure air cooler tube bundle;
3) extracting a neural network by combining the characteristics of the constructed high-pressure air cooler tube bundle infrared thermal imaging image, and constructing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network;
4) training the built high-pressure air cooler tube bundle infrared thermal imaging image feature extraction neural network by using the training set in the standard data set file obtained in the step 1) to obtain an image feature diagram, and then training the high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection neural network by using the image feature diagram to obtain a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
5) detecting and identifying an infrared visual image acquired by an infrared video acquisition device by utilizing a high-pressure air cooler tube bundle infrared thermal imaging image fault target identification and fault area positioning detection model;
6) and controlling the water supply valve according to the detection and identification result, and formulating a maintenance strategy to realize preventive maintenance of the high-pressure air cooler.
2. The visual fault identification and location method for the high-pressure air cooler of the petrochemical plant according to claim 1, wherein the preprocessing of the infrared thermal imaging image in the step 1) comprises:
(1) numbering and naming the infrared thermal imaging images in a naming format of 1. jpg-n.jpg to form a high-pressure air cooler tube bundle infrared thermal imaging image group;
(2) utilizing an image characteristic judgment method in the air cooler tube bundle infrared detection method to perform scaling and blockage fault identification on the high-pressure air cooler tube bundle infrared thermal imaging image group: when the heat exchange tubes in the tube bundle are scaled, the corresponding parts in the image are displayed as green areas, when the heat exchange tubes in the tube bundle are blocked, the corresponding parts in the image are displayed as blue areas, and when the heat exchange tubes in the tube bundle normally run, the tube bundle in the image is displayed as yellow areas;
(3) establishing a high-pressure air cooler tube bundle infrared anomaly detection data set, marking the occurred infrared anomaly into 3 types, and dividing into the following steps according to the abnormal form: dividing the abnormal length which is less than or equal to the length 1/10 of the heat exchange tube into short strips, dividing the abnormal length between 1/10 and 1/4 of the length of the heat exchange tube into long strips, and dividing the abnormal length which is more than or equal to the length 1/4 of the heat exchange tube into long strips;
(4) utilizing a labelling software tool to frame a green area and a blue area in each image in the high-pressure air cooler tube bundle infrared thermal imaging image group by adopting a rectangular marking frame, generating an XML file corresponding to the image in the high-pressure air cooler tube bundle infrared thermal imaging image group for storage, wherein the XML file comprises the file name, the size, the rectangular marking frame position and the marked target category information of the image, and finally, marking the marked data file according to 70: 15: 15 into a training set, a validation set and a test set, and establishing a standard data set file.
3. The visual fault identification and location method for the high-pressure air cooler of the petrochemical plant according to claim 1, wherein the neural network for extracting the image features of the high-pressure air cooler bundle in the step 2) is based on a residual convolutional neural network ResNet50 architecture, and comprises 1 image preprocessing block and 5 stacking blocks from top to bottom, wherein the 1 st stacking block is formed by a standard convolutional layer and a maximal pooling layer, the 2 nd stacking block is a residual block comprising 3 convolutional blocks, the 3 rd stacking block is a residual block comprising 4 convolutional blocks, the 4 th stacking block is a residual block comprising 6 convolutional blocks, the 5 th stacking block is a residual block comprising 3 convolutional blocks, 1 average pooling and 1 full connection layer.
4. The visual fault identification and location method for the high-pressure air cooler of the petrochemical device according to claim 3, wherein the visual fault identification and location method comprises the following steps:
(1) the image preprocessing block is used for cutting and edge expanding the received image and changing the original image into an image of 227 multiplied by 3;
(2) the 1 st stacked block comprises 64 convolution kernels and 64 max pooling devices;
(3) the 2 nd stacking block is divided into two paths, one path is an expanded dimension convolution kernel, the other path comprises 3 convolution blocks, each convolution block is provided with three convolution layers from top to bottom, the 1 st layer is 64 convolution kernels of 1 × 1, the 2 nd layer is 64 convolution kernels of 3 × 3, the 3 rd layer is 256 convolution kernels of 1 × 1, the output result of the 1 st convolution block is added with the output result of the expanded dimension convolution kernel and then is used as the input of the 2 nd stacking block, the output result of the 2 nd convolution block is added with the output result of the expanded dimension convolution kernel and then is used as the input of the 3 rd stacking block, and the output result of the 3 rd stacking block enters the 3 rd stacking block;
(4) the 3 rd stacking block comprises: a dimension-extended convolution kernel and 4 volume blocks from top to bottom, wherein each volume block has three volume layers from top to bottom, the 1 st volume layer is 128 convolution kernels of 1 × 1, the 2 nd volume layer is 128 convolution kernels of 3 × 3, the 3 rd volume layer is 512 convolution kernels of 1 × 1, the output result of the 1 st volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 2 nd volume block, the output result of the 2 nd volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 3 rd volume block, the output result of the 3 rd volume block is added with the output result of the dimension-extended convolution kernel to be used as the input of the 4 th volume block, and the output of the 4 th volume block enters the 4 th stack block;
(5) the 4 th stacking block comprises: a dimension-expanding convolution kernel and 6 convolution blocks from top to bottom, wherein each convolution block has three convolution layers from top to bottom, the 1 st convolution layer is 256 convolution kernels of 1 × 1, the 2 nd convolution layer is 256 convolution kernels of 3 × 3, the 3 rd convolution layer is 1024 convolution kernels of 1 × 1, the output result of the first 1 convolution block in the 6 convolution blocks is added with the output result of the dimension-expanding convolution kernel to obtain a result which is used as the input of the next convolution block, and the output result of the 6 th convolution block enters the 5 th stacking block;
(6) the 5 th stack includes: and each convolution block comprises three convolution layers from top to bottom, wherein the 1 st convolution layer is 512 convolution kernels of 1 × 1, the 2 nd convolution layer is 512 convolution kernels of 3 × 3, the 3 rd convolution layer is 2048 convolution kernels of 1 × 1, the output result of the 1 st convolution block is added with the output result of the convolution kernel to be used as the input of the 2 nd convolution block, the output result of the 2 nd convolution block is added with the output result of the convolution kernel to be used as the input of the 3 rd convolution block, and the output result of the 3 rd convolution block is the extracted characteristic.
5. The visual fault identification and location method for the high-pressure air cooler of the petrochemical device according to claim 4, wherein the visual fault identification and location method comprises the following steps: each 1 convolutional layer in each stack is activated using a batch normalization operation, the output of the 1 st convolutional layer and the output of the 2 nd convolutional layer of each convolutional block in each stack are activated using a Relu function, and the output of each convolutional block is activated using a Relu function, wherein the batch normalization operation is formulated as follows:
wherein xaiFor the input of a batch of n elements, x, at layer a of the convolutional networkai＝{xa1,xa2,…,xan}，μaIs the average value of the elements of one batch,
is the variance of the elements of a batch,
zeta is the term that guarantees a non-zero denominator, y, as a result of normalization of all elements of the same batchaiFor the final output, γaiFor scale scaling, betaaiA parameter of the offset;
the activation function Relu is expressed as:
wherein, f (y)ai) The Relu activation function value, which is the output of layer a of the convolutional network.
6. The visual fault identification and location method for the high-pressure air cooler of the petrochemical plant according to claim 1, wherein the step 3) of constructing the neural network for target fault identification and location detection of the fault region of the high-pressure air cooler tube bundle infrared thermal imaging image is based on a FasterR-CNN structure, and comprises four parts, namely a neural network for extracting the characteristic of the high-pressure air cooler tube bundle infrared thermal imaging image, a region production module RPN, a region suggestion pooling module, and a region method module RCNN based on the characteristic of a convolutional neural network.
7. The visual fault identification and location method for the high-pressure air cooler of the petrochemical plant according to claim 6, wherein the visual fault identification and location method comprises the following steps:
(1) the region production module RPN comprises the following operation processes:
(1.1) sliding on a feature map extracted by a neural network for extracting the features of the infrared thermal imaging image of the high-pressure air cooler tube bundle by using a 3 x 2048 sliding window, finding out the pixel position corresponding to each three-dimensional sliding window on an original image, then generating k anchor frames on the original image, wherein the centers of the k anchor frames are superposed and are the same as the centers corresponding to the three-dimensional sliding windows, calculating the offset between the center of each anchor frame and the coordinate at each coordinate point in the feature map, and calculating all the anchor frame positions in each feature map through the relative offset between the anchor frames and the generated coordinates;
(1.2) inputting all anchor frame positions in each feature map into 1 x 1 convolutional layer, outputting 2k x 20 x 16 anchor frame image features by the 1 x 1 convolutional layer, then mapping the 2k x 20 x 16 anchor frame image features into 2 x 20k x 16 by using a reshaping function reshape to obtain a foreground background score of the anchor frame, sending the result into a softmax function for probability calculation, transforming the obtained features into the 2k x 20 x 16 anchor frame image features, and finally outputting the probability that each anchor frame belongs to the foreground and the background; inputting all the anchor frame positions in each feature map into another 1 × 1 convolutional layer, wherein the 1 × 1 convolutional layer outputs 4k × 20 × 16 features, 4k includes predictions of k anchor frames, each anchor frame has 4 data respectively representing the horizontal and vertical coordinates of the center point of each anchor frame and the width and height of each anchor frame, and the 4 offset values relative to the rectangular mark frame;
(1.3) the RPN module classification and the rectangular labeling box calculation process of the regression offset are as follows: calculating the intersection and intersection ratio IOU (A) of the rectangular marking frame and the generated anchor framei，Bj) Cross-over ratio IOU (A)i，Bj) Expressed as:
wherein IOU (A)i，Bj) Representing the intersection ratio of the area of the ith rectangular marking frame marked in the image and the jth anchor frame; a. theiRepresenting the area of the ith rectangular labeling frame labeled in the image; b isjRepresenting the area of the generated jth anchor frame;
(1.4) determining whether each anchor frame is a positive sample or a negative sample according to the cross-over ratio, wherein the positive and negative sample judgment criteria are as follows: intersection ratio IOU (A) with all rectangular label boxesi，Bj) All anchor frames smaller than 0.3 are determined as negative samples; intersection ratio IOU (A) of any one rectangular mark boxi，Bj) The anchor frame which is greater than or equal to 0.7 is a positive sample; the anchor frames that do not belong to either the positive or negative samples are ignored;
(1.5) calculating the integral loss function of the RPN module and optimizing parameters of each layer of the network according to the loss function;
the loss function of the RPN module comprises classification loss and regression loss, and the formula is as follows:
wherein
Representing the classified cross entropy loss of the screened anchor frame; pkA category truth value for the nth anchor box;
for the prediction class of the nth anchor frame, when the nth anchor frame is a positive sample
Value 1, negative sample
A value of minus 1; n is a radical ofclsTo classify the loss weight value, NregIs the weight value of the regression loss,
represents the regression loss, tk＝{tx，ty，tw，thThe distance between the center coordinate and the width and the height of the anchor frame relative to the center coordinate and the width and the height of the rectangular marking frame,
the predicted values of the distance between the center coordinate and the width and the height of the anchor frame relative to the center coordinate and the width and the height of the rectangular marking frame are obtained;
(1.6) carrying out parameter training on the RPN module according to a rectangular marking frame calibrated in the training set of the high-pressure air cooler tube bundle infrared thermal imaging image, alternately optimizing the RPN module parameters by a random gradient descent method, fixing the regression parameters when optimizing the classification parameters, and fixing the classification parameters when optimizing the regression parameters;
(2) the operation steps of the area suggestion pooling module are as follows:
taking 256 anchor frames out of the anchor frames forming the positive sample and the negative sample, mapping the anchor frames to a characteristic diagram extracted by a neural network through the high-pressure air cooler tube bundle infrared thermal imaging image characteristic extraction, equally dividing a characteristic area mapped by each anchor frame into squares with the size of 7 multiplied by 7, selecting 4 sampling points in each small square, then respectively carrying out bilinear interpolation on 4 points in each small square through a bilinear interpolation method to obtain 4 values, then carrying out maximized pooling to obtain the value of each small square, and obtaining a maximized pooled characteristic diagram;
(3) the operation process of the region method module RCNN based on the convolutional neural network characteristics comprises the following steps:
and (3) performing full connection operation on the feature graph after the maximized pooling, predicting the classification of each anchor frame in the 256 anchor frames, predicting the offset of each anchor frame and the corresponding rectangular marking frame, wherein the offset calculation is the same as the (1.3) RPN module classification and the rectangular marking frame calculation method of regression offset.
8. The visual fault identification and location method for the high-pressure air cooler of the petrochemical plant according to claim 1, wherein the step 6) of establishing the maintenance strategy comprises:
(1) according to the testing result, form the distribution positioning diagram of high-pressure air cooler tube bank trouble, according to the bundle scale deposit degree of discernment, remind the staff in time to take corresponding measure, specifically be:
when the fault is a short strip-shaped area and the accumulation of the areas on a single tube bundle is less than one third of the length of the tube bundle, the opening of a water injection valve of the high-pressure air cooler is increased for flushing;
when the faults appearing in the detection process are short strip-shaped areas and long strip-shaped areas appearing on a single tube bundle, the sum of the short strip-shaped areas and the long strip-shaped areas is more than two thirds of the length of the tube bundle, the opening of a water injection valve of the high-pressure air cooler is increased for washing, and when the faults are detected after washing, the fault areas are not reduced, high-pressure water washing is recommended to be carried out to dredge the tube bundle when the machine is shut down;
when the fault is a large strip-shaped area on a single tube bundle, the opening of a water injection valve of the high-pressure air cooler is completely opened for washing, and when the fault is detected after washing, the area is not reduced, and the treatment of pipe blocking is recommended;
(2) before the large overhaul of the equipment, the overhaul scheme is determined according to the scaling and blocking conditions of the tube bundle with the fault in the high-pressure air cooler tube bundle fault distribution positioning diagram in different periods.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111246542.3A CN114037916A (en) | 2021-10-26 | 2021-10-26 | Visual fault identification and positioning method for high-pressure air cooler of petrochemical device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111246542.3A CN114037916A (en) | 2021-10-26 | 2021-10-26 | Visual fault identification and positioning method for high-pressure air cooler of petrochemical device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114037916A true CN114037916A (en) | 2022-02-11 |
Family
ID=80135414
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111246542.3A Pending CN114037916A (en) | 2021-10-26 | 2021-10-26 | Visual fault identification and positioning method for high-pressure air cooler of petrochemical device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114037916A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115311533A (en) * | 2022-08-12 | 2022-11-08 | 哈尔滨市科佳通用机电股份有限公司 | Vehicle door sliding track breaking fault detection method |
| CN117538658A (en) * | 2023-11-16 | 2024-02-09 | 深圳市美信检测技术股份有限公司 | Artificial intelligence fault positioning method and device based on infrared spectrum and thermal imaging |
-
2021
- 2021-10-26 CN CN202111246542.3A patent/CN114037916A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115311533A (en) * | 2022-08-12 | 2022-11-08 | 哈尔滨市科佳通用机电股份有限公司 | Vehicle door sliding track breaking fault detection method |
| CN115311533B (en) * | 2022-08-12 | 2023-04-18 | 哈尔滨市科佳通用机电股份有限公司 | Vehicle door sliding track breaking fault detection method |
| CN117538658A (en) * | 2023-11-16 | 2024-02-09 | 深圳市美信检测技术股份有限公司 | Artificial intelligence fault positioning method and device based on infrared spectrum and thermal imaging |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN114037916A (en) | Visual fault identification and positioning method for high-pressure air cooler of petrochemical device | |
| CN111223088B (en) | Casting surface defect identification method based on deep convolutional neural network | |
| CN115439458A (en) | Industrial image defect target detection algorithm based on depth map attention | |
| CN112464846B (en) | Automatic recognition method for abnormal faults of freight train carriage of station | |
| CN112861646B (en) | Cascade detection method for oil unloading worker safety helmet in complex environment small target recognition scene | |
| CN114037914A (en) | Unmanned aerial vehicle inspection pipeline thermal insulation fault identification and positioning method | |
| CN115965915B (en) | Railway wagon connecting pull rod breaking fault identification method and system based on deep learning | |
| CN111738206A (en) | Excavator detection method for unmanned aerial vehicle inspection based on CenterNet | |
| CN113393426A (en) | Method for detecting surface defects of rolled steel plate | |
| CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
| CN116844147A (en) | Pointer instrument identification and abnormal alarm method based on deep learning | |
| CN111080599A (en) | Fault identification method for hook lifting rod of railway wagon | |
| CN111652846A (en) | Semiconductor defect identification method based on characteristic pyramid convolution neural network | |
| CN113313678A (en) | Automatic sperm morphology analysis method based on multi-scale feature fusion | |
| CN117593244A (en) | Film product defect detection method based on improved attention mechanism | |
| CN116503398A (en) | Insulator pollution flashover detection method and device, electronic equipment and storage medium | |
| CN113284103B (en) | Substation equipment defect online detection method based on space transformation fast R-CNN model | |
| CN115830302A (en) | Multi-scale feature extraction and fusion power distribution network equipment positioning identification method | |
| CN114581440B (en) | Method for rapidly positioning leakage point based on image recognition | |
| CN113469984B (en) | Method for detecting appearance of display panel based on YOLO structure | |
| CN114037915A (en) | Method for identifying and positioning faults of pipe bundle of key air cooler for unmanned aerial vehicle inspection | |
| CN114581697A (en) | Indicator diagram-based method for monitoring liquid supply capacity of oil extraction motor-pumped well | |
| CN114120043A (en) | Method for detecting abnormal pumping well based on production dynamic data and indicator diagram | |
| CN112037197A (en) | Hot-rolled bar cold-shearing material accumulation detection method, system and medium |
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 |