CN118190761A - Inter-crystal corrosion control device and method based on deep learning image recognition - Google Patents
Inter-crystal corrosion control device and method based on deep learning image recognition Download PDFInfo
- Publication number
- CN118190761A CN118190761A CN202211601013.5A CN202211601013A CN118190761A CN 118190761 A CN118190761 A CN 118190761A CN 202211601013 A CN202211601013 A CN 202211601013A CN 118190761 A CN118190761 A CN 118190761A
- Authority
- CN
- China
- Prior art keywords
- module
- temperature
- test
- solution
- image acquisition
- 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
- 238000005260 corrosion Methods 0.000 title claims abstract description 51
- 230000007797 corrosion Effects 0.000 title claims abstract description 51
- 238000013135 deep learning Methods 0.000 title claims abstract description 22
- 238000000034 method Methods 0.000 title claims abstract description 22
- 239000013078 crystal Substances 0.000 title description 2
- 238000009835 boiling Methods 0.000 claims abstract description 101
- 238000012360 testing method Methods 0.000 claims abstract description 89
- 239000007788 liquid Substances 0.000 claims abstract description 36
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 21
- 238000010438 heat treatment Methods 0.000 claims abstract description 21
- 239000000523 sample Substances 0.000 claims abstract description 13
- 239000000243 solution Substances 0.000 claims description 59
- 238000012549 training Methods 0.000 claims description 41
- 230000033001 locomotion Effects 0.000 claims description 35
- 238000013507 mapping Methods 0.000 claims description 20
- 238000001816 cooling Methods 0.000 claims description 13
- 238000013500 data storage Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 238000011176 pooling Methods 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 9
- 238000013145 classification model Methods 0.000 claims description 7
- 238000010561 standard procedure Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 239000000498 cooling water Substances 0.000 claims description 5
- 230000000877 morphologic effect Effects 0.000 claims description 5
- 239000012085 test solution Substances 0.000 claims description 5
- 239000002360 explosive Substances 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 239000005388 borosilicate glass Substances 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 2
- 230000001105 regulatory effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 9
- 230000007246 mechanism Effects 0.000 description 6
- 238000010998 test method Methods 0.000 description 6
- 238000009833 condensation Methods 0.000 description 5
- 230000005494 condensation Effects 0.000 description 5
- 230000001276 controlling effect Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 229910001220 stainless steel Inorganic materials 0.000 description 5
- 238000007655 standard test method Methods 0.000 description 5
- 239000002184 metal Substances 0.000 description 4
- 229910052751 metal Inorganic materials 0.000 description 4
- 239000010935 stainless steel Substances 0.000 description 4
- QAOWNCQODCNURD-UHFFFAOYSA-N Sulfuric acid Chemical compound OS(O)(=O)=O QAOWNCQODCNURD-UHFFFAOYSA-N 0.000 description 3
- 239000011521 glass Substances 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 150000002739 metals Chemical class 0.000 description 2
- 230000000630 rising effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 229910000838 Al alloy Inorganic materials 0.000 description 1
- 229910045601 alloy Inorganic materials 0.000 description 1
- 239000000956 alloy Substances 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000007769 metal material Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
Landscapes
- Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)
Abstract
The invention discloses an intergranular corrosion control device and method based on deep learning image recognition, comprising a main body bracket and an image acquisition module; the main body support is provided with a test container which is used for carrying out an intergranular corrosion test; an image acquisition module is fixed on the main body support, and a probe of the image acquisition module is aligned to a test position of the test container; the bottom of the test container is provided with a heating module, the outer side of the test container is provided with a temperature sensor, and the heating module and the temperature sensor are electrically connected with a temperature control unit; the image acquisition module is connected with the control system, the control system identifies the boiling state of the solution through the built-in deep convolutional neural network through the solution surface image acquired by the image acquisition module, and the temperature control unit is controlled to carry out temperature regulation of the intergranular corrosion test according to the identification result. The device is used for solving the problem that the liquid level state cannot be identified and the temperature can be regulated independently in the prior art so as to keep micro-boiling of the solution.
Description
Technical Field
The invention belongs to the field of computer image processing, and particularly relates to an intergranular corrosion control device and method based on deep learning image recognition.
Background
Intergranular corrosion of metals is one of the localized corrosion, i.e., corrosion that propagates inward along the interfaces between metal grains. In order to evaluate corrosion resistance and quality of grain boundaries of materials such as stainless steel, an intergranular corrosion test is required. The intergranular corrosion test is a test method for testing the intergranular corrosion sensitivity of a metal material under specific medium conditions to accelerate the metal corrosion. Intergranular corrosion test methods for stainless steel, aluminum alloys, etc. are standardized in many countries, and the test procedures and test methods are specified in GB/T4334-2008 methods for intergranular corrosion test of corrosion stainless steel for metals and alloys, and ASTM A262-15"Standard Practices for Detecting Susceptibility to Intergranular Attack in Austenitic Stainless Steels".
In the intergranular corrosion test, it is possible to perform the test by optionally one of a plurality of test methods (up to 5) provided in the above-mentioned standards, most of which (more than 3) require the test solution to be kept in a boiling or micro-boiling state for a long time (15 h to 240 h), which puts high demands on stability and control accuracy of the test instrument. At present, the intergranular corrosion test device is built by oneself by laboratory or detection mechanism using electric heater and glassware more, and the test device of building often has following problem: firstly, the integration and modularization degree of the device are low, so that a large amount of test space is occupied, and the expandability of the whole device is reduced; secondly, the device in the test often has difficulty in maintaining the micro-boiling state of the solution for a long time, and since definition of micro-boiling is not specified in the standard, manual adjustment is required at fixed time in the whole test process according to experience of a tester, which leads to an increase in test cost and affects test accuracy. Therefore, in order to solve the above problems and to realize high-quality and high-efficiency evaluation of corrosion resistance and quality of grain boundaries of materials such as stainless steel, there is a need for an intergranular corrosion test solution which has a high integration level and can autonomously recognize the boiling state of a liquid surface and make a correct response.
At present, among existing products in the domestic market, an ICT-4 type intergranular corrosion apparatus developed by Shanghai detection technology (Shanghai) is typically represented, and the apparatus consists of a heating part, a control part and a cooling part. The device is provided with four heating bases, can heat four beakers simultaneously, and the temperature can be set arbitrarily through the control part. The intelligent degree of the instrument is low, the temperature needs to be set during the test, and the liquid level state cannot be identified and regulated independently so as to keep micro-boiling of the solution; in addition, the glass vessel of the instrument is small and can not carry out the intergranular corrosion test of more or larger samples.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an intergranular corrosion control device and method based on deep learning image recognition, which are used for solving the problem that the liquid level state cannot be recognized and the temperature cannot be regulated independently so as to keep micro-boiling of a solution in the prior art.
In order to achieve the above purpose, the present invention provides the following technical solutions:
an intergranular corrosion control device based on deep learning image recognition comprises a main body bracket and an image acquisition module;
The main body support is provided with a test container which is used for carrying out an intergranular corrosion test; an image acquisition module is fixed on the main body support, and a probe of the image acquisition module is aligned to a test position of the test container;
the bottom of the test container is provided with a heating module, the outer side of the test container is provided with a temperature sensor, and the heating module and the temperature sensor are electrically connected with a temperature control unit;
the image acquisition module is connected with the control system, the control system identifies the boiling state of the solution through the built-in deep convolutional neural network through the solution surface image acquired by the image acquisition module, and controls the temperature control unit to carry out temperature regulation of the intergranular corrosion test according to the identification result.
Preferably, the image acquisition module is an industrial camera, and the industrial camera is fixed on the main body bracket through the graph acquisition bracket; the graph acquisition bracket is of a double-hinge mechanical arm structure; the top of main part support is provided with the crossbeam, be provided with the guide rail on the crossbeam, the bottom of figure collection support is provided with the pulley, and pulley embedding guide rail carries out sliding connection.
Preferably, the test container is conical flaring, the test container is made of high borosilicate glass, a top cover is arranged at the top of the test container, and the top cover and the test container are sealed through a flange.
Preferably, the control system comprises a solution boiling state identification module, a cooperative control module and a lower control module;
The solution boiling state identification module is used for processing the solution surface image obtained by the image acquisition module to obtain a motion characteristic image with characteristics such as solution ripple, bubbles and liquid level turbulence; identifying the shape, distribution, quantity and other information of the characteristics such as the wave, the bubble, the liquid level turbulence and the like in the motion characteristic image through a built-in deep convolution neural network, predicting the boiling state of the solution according to the information, and generating a corresponding control instruction;
the cooperative control module is used for converting the control instruction to generate a temperature control instruction;
The lower control module is used for receiving the temperature control instruction and controlling the temperature control unit to adjust the temperature according to the control instruction.
Preferably, the solution boiling state identification module comprises a central processing unit module, a neural network processor module and a data storage array;
The central processing unit module is used for carrying out data processing on the image acquired by the image acquisition module;
The neural network processor module is used for training and predicting a deep convolutional neural network model;
the data storage array is used for data storage.
Preferably, the cooling system further comprises a cooling circulation module, wherein the cooling circulation module comprises a condensation pipe, a circulation pipeline and a cooling water tank; the condenser pipe sets up on the top of experimental container, the condenser pipe passes through circulation pipeline and connects the cooling water tank.
An intergranular corrosion control method based on deep learning image recognition comprises the following steps,
Step 1, continuously acquiring n Zhang Yemian form pictures by an image acquisition module, and subtracting the n Zhang Yemian form pictures frame by frame according to the acquisition sequence of the image acquisition module to obtain a motion characteristic image of solution ripple, bubbles and liquid level turbulence;
step 2, identifying characteristics of waves, bubbles, liquid level turbulence and the like in the motion characteristic image in the step 1 through a deep convolution neural network so as to predict a boiling state;
And step 3, generating a control instruction according to the boiling state of the solution, and controlling the temperature control unit to perform temperature adjustment of the intergranular corrosion test.
Preferably, the prediction of the boiling state of the solution by the deep convolutional neural network specifically comprises the following steps,
Step 2.1, constructing a training data set; the motion characteristic images are used as training samples, and are divided into four types according to liquid level motion characteristics in different boiling states, namely non-boiling, micro-boiling, boiling and explosive boiling;
Data enhancement is carried out on the training sample, and the training sample is transformed through overturning, rotating, cutting and zooming to generate an enhancement sample; selecting data according to the proportion to form a training set, a verification set and a test set;
2.2, constructing a liquid level state classification model based on a convolutional neural network by using a deep learning framework, wherein the model comprises an input layer, a convolutional-pooling layer array, a full-connection layer and a normalized exponential function class output layer;
Step 2.3, inputting the training set and the verification set constructed in the step 2.1 into the liquid level state classification model in the step 2.2 for training, outputting the probability that the input motion characteristic image belongs to each boiling state as the output result, and taking the boiling state label value corresponding to the maximum probability as the final output of prediction; the class prediction result output vector Pre＝[y1,y2,y3,y4,y5,y6,y7,y8,y9]; of all images is trained by adopting an adaptive gradient optimizer at the global learning rate of 0.00001;
Step 2.4, averaging and rounding the output vector Pre of the control instruction to obtain a final class predicted value y pre; and outputting a control command value out according to the test solution and the standard method, wherein the control command value is 0,1 and 2, which respectively represent the reduced temperature, the maintained temperature and the raised temperature.
Preferably, the control instruction is generated according to the boiling state of the solution, the temperature control unit is controlled to carry out the temperature adjustment of the intergranular corrosion test specifically comprises the following steps,
Step 3.1, receiving a control instruction value out vector, and converting the control instruction value out vector into a format table;
step 3.2, mapping the out value to a correct target SV, and calculating the target SV;
Step 3.3, mapping the index number of the image acquisition module to a correct index number of the temperature control unit, and combining the index number with a target SV to form an output vector;
And 3.4, sending the output vector to a lower control module through a modbus TCP protocol, and controlling a corresponding temperature control unit to regulate the temperature by the lower control module.
Compared with the prior art, the invention has the following beneficial technical effects:
The invention provides an intergranular corrosion control device based on deep learning image recognition, which is used for realizing the autonomous classification of solution boiling states by constructing and training a convolutional neural network classification model by using a deep learning technology to extract and analyze the characteristics of surface morphology features and motion information (including waves, bubbles, liquid level turbulence and motion thereof) of the solution in different boiling states. And analyzing the classification result of the convolutional neural network through a cooperative control algorithm, converting the classification result into a temperature regulation strategy, and sending temperature control information to a lower control system through a modbus TCP protocol to carry out temperature dynamic regulation and control, so that the boiling state of the solution is ensured to meet the standard method requirement and is kept stable for a long time. Therefore, the method utilizes a certain number of motion feature diagrams of the boiling state of the corrosive solution to train a deep convolutional neural network model to learn morphological features and motion rules of the surface of the solution in different boiling states, and realizes the autonomous classification of the boiling state of the solution. The method utilizes the deep convolutional neural network to realize the prediction of the boiling state according to the identification of the forms, the respective and the number of the characteristics such as the wave, the bubble, the liquid level turbulence and the like in the liquid level motion characteristic image; the lower control system is called to dynamically adjust the temperature according to the boiling state identification result through the cooperative control algorithm, so that the boiling state of the solution always meets the requirements of the standard method and is kept stable for a long time, the stability and accuracy of controlling the boiling state of the solution in the test process are greatly improved, and the requirements of the standard method are met. The problem that solution state control depends on uninterrupted manual adjustment and control accuracy depends on manual experience in an intergranular corrosion test is solved.
According to the intergranular corrosion control method, the surface morphological characteristics and the movement information (including the ripples, the bubbles, the liquid level turbulence and the movement thereof) of the solution in different boiling states are identified by training the deep convolutional neural network model, intelligent analysis of the boiling state of the solution is realized, a cooperative control algorithm is built to convert an intelligent analysis result into a temperature regulation instruction and send the temperature regulation instruction to a lower control system to guide the lower control system to dynamically regulate the test temperature, and the boiling state of the solution is ensured to meet the requirements of a standard method and is kept stable for a long time.
Drawings
FIG. 1 is a schematic diagram of an intergranular corrosion control test apparatus according to the present invention;
FIG. 2 is a side view of the structure of the intergranular corrosion control test apparatus according to the present invention;
FIG. 3 is a schematic side view of a rail beam and a pattern acquisition bracket base structure and assembly;
FIG. 4 is a schematic top view of a rail beam and a pattern acquisition bracket base structure and assembly;
FIG. 5 is a schematic illustration of the structure of a test vessel;
FIG. 6 is a top view of the test vessel structure;
FIG. 7 is a schematic diagram of a heating module configuration in combination with a test vessel;
FIG. 8 is a schematic diagram of a unit control loop configuration;
FIG. 9 is a schematic diagram of a cooling cycle module configuration;
FIG. 10 is a schematic diagram of the control system A module architecture and flow;
Fig. 11 is a liquid level moving image of different boiling states;
FIG. 12 is a schematic diagram of a liquid level boiling state intelligent classification model;
FIG. 13 is a schematic flow diagram of a control system B module;
FIG. 14 is a schematic diagram of a control system C module architecture and flow scheme;
In the accompanying drawings: the device comprises a graph acquisition bracket 1, an image acquisition module 2, a heating module 3, a main body bracket 4, a condensing tube 5, a test container 6, a cross beam 7, a temperature control unit 8, a temperature sensor 9, a cooling tank 10, a control system 11, a display device 12, a clamping mechanism 13, a thermocouple 14, a top cover 15 and a flange 16.
Detailed Description
The invention will now be described in further detail with reference to specific examples, which are intended to illustrate, but not to limit, the invention.
As shown in fig. 1 and 2, the intergranular corrosion control device based on deep learning image recognition of the present invention includes a main body support 4, a test container 6, an image acquisition module 2, a heating module 3, a temperature control unit 8 and a cooling circulation module.
The main body bracket 4 is provided with a test container 6, and the test container 6 is used for carrying out an intergranular corrosion test; an image acquisition module 2 is fixed on the main body bracket 4, and a probe of the image acquisition module 2 is aligned with a test position of the test container 6;
The bottom of the test container 6 is provided with a heating module 3, the outer side of the test container 6 is provided with a temperature sensor 9, and the heating module 3 and the temperature sensor 9 are electrically connected with a temperature control unit 8;
The image acquisition module 2 is connected with the control system 11, the control system 11 recognizes the boiling state of the solution through the built-in deep convolutional neural network through the solution surface image acquired by the image acquisition module 2, and the temperature control unit 8 is controlled to perform temperature regulation of the intergranular corrosion test according to the recognition result.
As shown in fig. 1, the main body support 4 has a double-layer structure, wherein a control host and a circulating cooling unit are placed at the bottom layer, an experimental platform is placed at the top layer, and a test container 6 and a temperature control unit 8 are placed. The cross beams 7 with guide rail structures are distributed around the bracket and used for installing the image acquisition bracket and hanging the temperature sensor 9. The image acquisition module 2 is an industrial camera, the image acquisition bracket main body is of a double-hinge mechanical arm structure, the bottom of the image acquisition bracket main body is connected with a base with pulleys, the image acquisition bracket main body can be embedded into a bracket guide rail to move freely, and the upper end surface and the lower end surface of the base are provided with clamping mechanisms 13 which can be fixed on the guide rail cross beam 7. The end face of the base is provided with a clamping mechanism 13 capable of clamping the temperature sensor 9, and the clamping mechanism is used for fixing and accommodating the temperature sensor 9; the clamping mechanism 13 is provided with a thermocouple 14; as shown in fig. 3 and 4. The top of the bracket is connected with a clamp with universal wheels for fixing the image acquisition module 2.
The test container 6 is in a conical expansion shape, is made of borosilicate glass, is matched with a top cover 15 with the same caliber and material, and is sealed and closed by a flange 16. The top cover 15 is provided with an interface for connecting the condensation pipe 5. The overall form of the container is shown in fig. 5 and 6.
The image acquisition module 2 is composed of a lens and an imaging sensor, can be fixed at an observation position at the top or at the side of a glass instrument through the image acquisition bracket 1 and is used for acquiring the surface morphology image of the corrosive solution. The image acquisition module 2 corresponding to each heating position has a single index number.
The heating module 3 is of a half-wrapping structure, and wraps the bottom of the test container 6 in the heating area, as shown in fig. 7.
The temperature control module temperature sensor 9 and the temperature control terminals are formed, wherein the temperature control terminals are arranged on one side of the test container 6 in a stacked mode, and each temperature control terminal is connected with one heating module 3 and the temperature sensor 9 to form a unit control loop, as shown in fig. 8. The unit control loops adopt a parallel connection mode, and the number of the unit control loops is consistent with that of the test bits.
The cooling circulation module is mainly composed of a condensation pipe 5, a circulation line and a cooling tank 10, as shown in fig. 9. The condensation pipe 5 is connected to the glass top cover 15, the pipeline starts from a cooling water tank, and the condensation pipe 5 is sequentially connected in series and then returns to the circulating cooling unit.
The control system 11 mainly comprises a solution boiling state identification module (A module), a cooperative control module (B module) and a lower control module (C module). The solution boiling state is identified through a built-in deep convolutional neural network by the module A, a control instruction is sent to the module B according to an identification result, the module B receives the instruction and converts the instruction into a temperature control instruction which can be identified by the module C, and the module C adjusts the temperature after receiving the temperature control instruction. Specific:
The solution boiling state identification module (A module) mainly comprises an intelligent identification integrated machine and an intelligent identification algorithm, and the basic structure and flow are shown in figure 10. Specific:
The intelligent identification all-in-one machine comprises a central processing unit module, a neural network processor module and a data storage array. The central processing unit module is a control core of the integrated machine and is a final execution unit for information processing and program running; the neural network processor module is a main parallel computing unit and bears main operation tasks in the training and prediction process of the deep convolutional neural network model; the Data storage array is the primary storage unit that assumes the storage tasks of Data and algorithms, and its primary file structure includes three primary directories, "Raw Data", "model", "INFERDATA", and "TRAINING DATASET".
The intelligent recognition algorithm runs on the intelligent recognition integrated machine in the composition 1, and the algorithm flow is shown in the figure 10 and comprises the following main steps:
Step one, image acquisition and pretreatment:
N Zhang Yemian morphological pictures are continuously acquired through observation of the image acquisition module 2 and are marked as I 0 (0-n). And (3) subtracting the pictures in the I 0 frame by frame according to the photographing sequence of the formula (1) to obtain a motion characteristic image of solution ripple, bubbles and liquid level turbulence, and marking the motion characteristic image as I 1 (0-n-1). All images in I 1 were resized to W 0*W0 and then stored in the data storage array "INFERDATA" directory.
Step two, intelligent identification of the liquid level boiling state:
and (3) constructing a deep Convolutional Neural Network (CNN) model to predict the boiling state type of the image in the liquid level moving image set I1 (0-n-1) in the data storage array 'prediction data' catalog of the module 1 in the S2. Because the motion image captures the motion characteristics of solution waves, bubbles, and liquid surface turbulence, when there is motion in the above characteristics, the pixels at the corresponding positions in the image will be illuminated. Since the motion characteristics of waves, bubbles, and surface turbulence are different, they have different morphologies in the motion characteristic image. When the solution is in a micro-boiling state, the motion characteristics of the waves and the bubbles are more obvious, and the characteristic quantity is less; when in a boiling state, the characteristics of bubbles and liquid level turbulence are more obvious, and the characteristic quantity is increased compared with that of micro boiling; in the explosive boiling state, the liquid level turbulence characteristics are dominant, and the characteristic quantity is further increased. Meanwhile, as the boiling of the solution increases, the movement of the waves, bubbles and liquid surface turbulence increases, and the number of pixels other than 0 in the moving feature image increases. In summary, classification labeling and prediction can be performed on four boiling states by identifying different morphological features and the total number of lighting pixels in a moving image. The principle is the basic basis of training data labeling and intelligent liquid level state identification.
The intelligent identification of the liquid level boiling state mainly comprises the following steps: 2a) Construction of training data sets.
Firstly, obtaining surface morphology images of the solution in different boiling states through an image acquisition module 2, preprocessing according to the operation in the first step, obtaining a motion characteristic image of W 0×W0 as a training sample, and storing the motion characteristic image in a Raw Data catalog in a Data storage array.
Second, the images are classified into four types according to the liquid level motion characteristics (as shown in fig. 11) of different boiling states, namely, non-boiling, micro-boiling, boiling and pop-boiling. The tag values of the four categories are respectively 0,1, 2 and 3, the label after one-bit valid (one-hot) encoding is sequentially 'no boiling' = [1, 0] T "micro boiling" = [0,1, 0] T, "boiling" = [0,1, 0] T, and "pop boiling" = [0, 1] T.
Then, the samples in the training data catalog are screened according to the principle, and the samples in the same category are respectively stored in the sub-catalogs of '0-non-boiling', '1-micro-boiling', '2-boiling' and '3-explosive boiling' under the 'TRAINING DATASET' catalog, and the first number of each sub-catalog name is the corresponding category label value. Then, data enhancement is performed, and training samples are transformed in a manner of overturning, rotating, cutting, scaling and the like, so that enhanced samples are generated. All training samples of the data set are subjected to the operation, and the data set is amplified. The capacity of each class of sample in the amplified data set is not less than 1000. The training samples are renamed in a random sampling mode, and the naming format is label value_image name and serial number. And renaming the data under all the category sub-directories according to the rule, and storing the renamed data under the total data sub-directory in the TRAINING DATASET directory. Finally, the data are selected according to the proportion of (7-8): (1-2) under the total data sub-directory to form a training set, a verification set and a test set, and the training set, the verification set and the test set are respectively stored under the tracking, the verification and the testing sub-directories under the TRAINING DATASET directory. And renaming the data in the subdirectory for the second time, adding the data set type to which the data set belongs to the original name, wherein the naming format is a label value-data set type-image name and a serial number, and storing the new name in a text file with the names of tracking, verifying or testing respectively to form a label document.
The sampling and labeling process of the training data and the file structure of the training data set are shown in fig. 12. 2b) And (6) constructing an intelligent recognition model. A liquid level state classification model based on a convolutional neural network is built by using a deep learning framework, and mainly comprises an input layer, a convolutional-pooling layer array, a full-connection layer and a normalized exponential function (softmax) class output layer, wherein the model structure is shown in fig. 12. Wherein, the convolution layer (Conv) performs feature extraction on the image through convolution operation by using a convolution kernel with the size of k×k×c, and obtains a feature map with a three-dimensional (channel c×height h×width W) structure. The pooling layer (pooling) concentrates and abstracts the typical features while reducing the feature map size (H and W) through a max-pooling or average pooling operation. The pooling layer is typically located after the two convolution layers to form a convolution-pooling array as the main building block of the CNN. The full connection layer (FC) is used for classifying the feature extraction result and finally outputting the probability of the category to which the current input image belongs through a softmax function;
2c) Training the model constructed in 2 b) by using the training set constructed in 2 a) and the verification set, wherein the training set provides data for model learning characteristics, and the verification set tests the prediction accuracy of the model after each training cycle. Training is performed with an adaptive gradient optimizer (AdamW) at a global learning rate of 0.00001. And (3) adjusting parameters such as an L 2 regularization coefficient, a batch size (batch size) and the like according to a training result, and finally enabling the accuracy of the sample on a testing set to be maximum (more than or equal to 90%) after training for a plurality of weeks. Storing the trained model under a model catalog in a data storage array for predicting the boiling state of the solution; 2d) And sequentially inputting the super pixels in the INFERDATA catalog into the trained CNN model, outputting the probability that the input motion characteristic image belongs to each boiling state as a result, and taking the boiling state label value corresponding to the maximum probability as the final output of prediction. The category prediction results of all the motion feature images constitute an output vector pre= (y i)i＝1～n-1,yi =0, 1,2,3.
Step three, outputting control instructions:
calculating a control command value according to the output vector in the step 2 d), specifically: 3a) Averaging and rounding the output vector Pre obtained in 2 d) to obtain a final class prediction value of
re＝f[mean(Pre)] (2)
Wherein mean represents an average value, f [ cndot ] is a rounding function defined as:
X 1 in formula (3) is an integer portion of x, and 0.1x 2 is a fractional portion of x; 3b) And outputting a control command value according to the test solution and the standard method requirement, and marking as out. The control command values can only be 0, 1 and 2, and represent the reduced temperature, the holding temperature and the raised temperature respectively. When the standard test method requires that the boiling state of the CuSO 4+H2SO4 solution is micro-boiling, outputting a control command value according to the formula (4):
for other corrosive solutions, when the standard test method requires that the boiling state is boiling, a control instruction value is output according to the formula (5):
In particular, when S1-5) the number of unit control loops is not 1 (i.e., there are a plurality of test units), out becomes a vector, and has the following structure:
out＝[o1,o2,o3,...,oi,...,on] (6)
the subscript is the index number of the corresponding image acquisition module 2 in each unit control loop, which is the same as the index number of the temperature control unit 8 in the loop.
S2-2) the cooperative control module (B module) is an algorithm for establishing the mapping between the A module and the C module, and the flow is shown in FIG. 13. The method is used for analyzing the out value of the A module in the step S1-a) into a temperature control instruction and sending the temperature control instruction to the C module through a modbus TCP protocol, and specifically:
and 1, hardware mapping. The mapping is preset in an algorithm, and one-to-one mapping is built between index numbers of the image acquisition module 2 and index numbers of the temperature control unit 8 corresponding to each S1-4) heating module 3, and the index numbers are shown in a table 1. Storing the mapping information list in the data storage array of the component 1 in the S2-1).
Table 1 hardware mapping format table
| Camera index number | Index number of temperature control unit 8 |
| C1 | ch1 |
| C2 | ch2 |
| C3 | ch3 |
| ... | ... |
2, Temperature control mapping. The mapping is preset in an algorithm and comprises a series of temperature regulation strategies, wherein the temperature regulation strategies are used for selecting corresponding temperature regulation strategies according to the output out value of the A module in S2-1). Specifically, two schemes are included.
Scheme 1, when the standard test method requires that the boiling state of the CuSO 4+H2SO4 solution is micro-boiling, the temperature adjustment is carried out according to the formula (7):
wherein SV (step+1) represents the adjusted temperature set point; SV (step) is the current temperature setpoint; delta T is the temperature adjustment step size. For other corrosive solutions, the standard test method requires temperature adjustment according to formula (8) when the boiling state is boiling:
Scheme 2, set up SV value presets list, see Table 2. Wherein, the value of the undetermined SV is set according to the temperature range in the test process, the value of the SV is increased along with the sequence number, and the increment is delta T. According to Table 2, for CuSO 4+H2SO4 solution, standard
Table 2 SV value preset list
| Serial number (i) | Preset temperature set point (DEG C) |
| 1 | SV1 |
| 2 | SV2 |
| 3 | SV3 |
| ... | ... |
| i | SVi |
| ... | ... |
| max | SVmax |
When the test method requires that the boiling state is micro-boiling, the temperature adjustment is performed according to the formula (9):
for other corrosive solutions, the standard test method requires temperature adjustment according to formula (10) when the boiling state is boiling:
And 3, forming an instruction module. The module maps the out value to the correct target SV by receiving the out value output by the A module from S2-1), maps the SV to the correct lower temperature control unit 8 index by calling component 2 (temperature control mapping), and sends the SV adjusting value and the corresponding temperature control unit 8 index to the C module by calling component 1 (hardware mapping). The method specifically comprises the following steps:
Step one, receiving the out vector, converting it into the format of table 3 according to equation (6).
Table 3out vector conversion format table
| Camera index number | Out value |
| C1 | o1 |
| C2 | o2 |
| C3 | o3 |
| ... | ... |
And step two, calling the composition 2 (temperature control mapping) of the step S2-2, and calculating the target SV according to the formula (8), the formula (9) or the formula (10).
Step three, calling S2-2) the component 1 (hardware mapping), mapping the camera index number in the table 2 to the correct index number of the temperature control unit 8, and combining the target SV to form an output vector, see table 4.
Table 4 synergistic control output vector format table
| Index number of temperature control unit | Target SV |
| ch1 | SV1 |
| ch2 | SV2 |
| ch3 | SV3 |
| ... | ... |
And step four, sending the information of the table 4 to the module C through a modbus TCP protocol.
The lower control module (C module) mainly includes an integrated temperature control unit 8, a temperature control terminal, and a temperature sensor 9, as shown in fig. 14. The integrated temperature control unit 8 is connected with all the unit control loops of S1-5) and consists of a display device 12 and a singlechip. The lower control module is communicated with the B module. Wherein the display device 12 displays all test receptacles 6 bottom temperature sensor 9 readings (TV) and set temperature values (SV); the singlechip is connected with and controls all temperature control terminals, and has the functions of receiving TV input, communicating with the B module through a modbus TCP protocol to obtain a target SV, processing received information, outputting temperature control instructions and the like. The singlechip controls the temperature through a negative feedback mechanism, so that the actual temperature TV is as close as possible to the target temperature SV, and a power adjustment instruction is output based on the magnitude and sign of the TV-SV value in each control period; the temperature control terminal adjusts the power of the heating element according to the power adjusting instruction output by the singlechip to realize temperature adjustment.
And (5) total control flow. Specifically, the method comprises the following steps:
step one, a preparation stage. Test requirements boiling status, test time, operating camera and temperature control terminal index information are entered S2-1) into the control system 11. Establishing the hardware mapping in the S2-2) composition 1 for the working equipment according to the input information;
And step two, a conventional temperature rising stage. S2-3) the C module reads the TV value of the temperature sensor 9, and when the TV is smaller than the critical temperature T, the singlechip sends a temperature rising instruction to all the temperature control terminals working according to the hardware mapping of the B module and sets SV=T. When the TV is equal to the critical temperature T, turning to the step III;
and thirdly, intelligent sensing and controlling of the boiling state. The solution boiling state identification module and the algorithm are utilized to identify the liquid level moving image obtained by the image acquisition module 2 under each working index number, and a control instruction vector out is output;
step four, calling a module B to analyze the out vector according to the method of S2-2), outputting a target SV (table 3) corresponding to the temperature control terminal under each working index, and sending the target SV to a module C through a modbus TCP protocol;
And fifthly, the module C sends power adjusting instructions to the corresponding temperature control terminals according to the hardware mapping and the target SV output by the module B in the step S2-2), and the temperature control terminals adjust the power of the heating module 3 according to the instructions to change the heating temperature. And the module C performs negative feedback adjustment according to the target SV to enable the temperature to approach the target SV. Waiting for the module B to send out a new target SV;
and step six, executing the second to fifth steps at regular time intervals delta t until the test is finished.
And step seven, after the total test duration is finished, the module C, the module B and the module A are sequentially closed, and the power supply is closed after relevant data are stored.
Claims (9)
1. The intergranular corrosion control device based on the deep learning image recognition is characterized by comprising a main body bracket (4) and an image acquisition module (2);
The main body support (4) is provided with a test container (6), and the test container (6) is used for carrying out an intergranular corrosion test; an image acquisition module (2) is fixed on the main body support (4), and a probe of the image acquisition module (2) is aligned to a test position of the test container (6);
The bottom of the test container (6) is provided with a heating module (3), the outer side of the test container (6) is provided with a temperature sensor (9), and the heating module (3) and the temperature sensor (9) are electrically connected with a temperature control unit (8);
The control system (11) is connected with the image acquisition module (2), the control system (11) identifies the boiling state of the solution through a built-in deep convolution neural network through the solution surface image acquired by the image acquisition module (2), and the temperature control unit (8) is controlled to carry out temperature regulation of an intergranular corrosion test according to the identification result.
2. The intergranular corrosion control device based on deep learning image recognition according to claim 1, wherein the image acquisition module (2) is an industrial camera, and the industrial camera is fixed on the main body support (4) through the image acquisition support (1); the graph acquisition bracket (1) is of a double-hinge mechanical arm structure; the top of main part support (4) is provided with crossbeam (7), be provided with the guide rail on crossbeam (7), the bottom of figure collection support (1) is provided with the pulley, and pulley embedding guide rail carries out sliding connection.
3. The intergranular corrosion control device based on deep learning image recognition according to claim 1, wherein the test container (6) is in a conical flaring shape, the test container (6) is made of borosilicate glass, a top cover (15) is arranged at the top of the test container (6), and the top cover (15) and the test container (6) are sealed through a flange (16).
4. The intergranular corrosion control device based on deep learning image recognition according to claim 1, wherein the control system (11) comprises a solution boiling state recognition module, a cooperative control module and a lower control module;
the solution boiling state identification module is used for processing the solution surface image obtained by the image acquisition module (2) to obtain a motion characteristic image of characteristics such as solution ripple, bubbles and liquid level turbulence; identifying the shape, distribution, quantity and other information of the characteristics such as the wave, the bubble, the liquid level turbulence and the like in the motion characteristic image through a built-in deep convolution neural network, predicting the boiling state of the solution according to the information, and generating a corresponding control instruction;
the cooperative control module is used for converting the control instruction to generate a temperature control instruction;
The lower control module is used for receiving a temperature control instruction and controlling the temperature control unit (8) to perform temperature adjustment according to the control instruction.
5. The intergranular corrosion control device based on deep learning image recognition of claim 1, wherein the solution boiling state recognition module comprises a central processor module, a neural network processor module, and a data storage array;
the central processing unit module is used for carrying out data processing on the image acquired by the image acquisition module (2);
The neural network processor module is used for training and predicting a deep convolutional neural network model;
the data storage array is used for data storage.
6. The intergranular corrosion control device based on deep learning image recognition according to claim 1, further comprising a cooling circulation module, wherein the cooling circulation module comprises a condenser tube (5), a circulation pipeline and a cooling water tank; the condensing pipe (5) is arranged on the top of the test container (6), and the condensing pipe (5) is connected with the cooling water tank through a circulating pipeline.
7. An intergranular corrosion control method based on deep learning image recognition is characterized by comprising the following steps,
Step 1, an image acquisition module (2) continuously acquires n Zhang Yemian morphological pictures under observation, and performs frame-by-frame subtraction according to the acquisition sequence of the image acquisition module (2) to obtain a motion characteristic image of solution ripple, bubbles and liquid level turbulence;
step 2, identifying characteristics of waves, bubbles, liquid level turbulence and the like in the motion characteristic image in the step 1 through a deep convolution neural network so as to predict a boiling state;
and step3, generating a control instruction according to the boiling state of the solution, and controlling a temperature control unit (8) to perform temperature adjustment of the intergranular corrosion test.
8. The method for controlling intergranular corrosion based on deep learning image recognition according to claim 7, wherein predicting the boiling state of the solution by the deep convolutional neural network comprises the steps of,
Step 2.1, constructing a training data set; the motion characteristic images are used as training samples, and are divided into four types according to liquid level motion characteristics in different boiling states, namely non-boiling, micro-boiling, boiling and explosive boiling;
Data enhancement is carried out on the training sample, and the training sample is transformed through overturning, rotating, cutting and zooming to generate an enhancement sample; selecting data according to the proportion to form a training set, a verification set and a test set;
2.2, constructing a liquid level state classification model based on a convolutional neural network by using a deep learning framework, wherein the model comprises an input layer, a convolutional-pooling layer array, a full-connection layer and a normalized exponential function class output layer;
Step 2.3, inputting the training set and the verification set constructed in the step 2.1 into the liquid level state classification model in the step 2.2 for training, outputting the probability that the input motion characteristic image belongs to each boiling state as the output result, and taking the boiling state label value corresponding to the maximum probability as the final output of prediction; the class prediction result output vector Pre＝[y1,y2,y3,y4,y5,y6,y7,y8,y9]; of all images is trained by adopting an adaptive gradient optimizer at the global learning rate of 0.00001;
Step 2.4, averaging and rounding the output vector Pre of the control instruction to obtain a final class predicted value y pre; and outputting a control command value out according to the test solution and the standard method, wherein the control command value is 0,1 and 2, which respectively represent the reduced temperature, the maintained temperature and the raised temperature.
9. The method for controlling intergranular corrosion based on deep learning image recognition according to claim 7, wherein generating a control command according to the boiling state of the solution, controlling the temperature control unit (8) to perform the temperature adjustment of the intergranular corrosion test specifically comprises the steps of,
Step 3.1, receiving a control instruction value out vector, and converting the control instruction value out vector into a format table;
step 3.2, mapping the out value to a correct target SV, and calculating the target SV;
step 3.3, mapping the index number of the image acquisition module (2) to the index number of the correct temperature control unit (8), and combining the target SV to form an output vector;
and 3.4, sending the output vector to a lower control module through a modbus TCP protocol, and controlling a corresponding temperature control unit (8) to regulate the temperature by the lower control module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211601013.5A CN118190761A (en) | 2022-12-13 | 2022-12-13 | Inter-crystal corrosion control device and method based on deep learning image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211601013.5A CN118190761A (en) | 2022-12-13 | 2022-12-13 | Inter-crystal corrosion control device and method based on deep learning image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118190761A true CN118190761A (en) | 2024-06-14 |
Family
ID=91405654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211601013.5A Pending CN118190761A (en) | 2022-12-13 | 2022-12-13 | Inter-crystal corrosion control device and method based on deep learning image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118190761A (en) |
-
2022
- 2022-12-13 CN CN202211601013.5A patent/CN118190761A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
| CN107229930B (en) | Intelligent identification method for numerical value of pointer instrument | |
| JP5783043B2 (en) | Cell mass state discrimination technique, image processing program and image processing apparatus using the technique, and cell mass production method | |
| CN108846413B (en) | Zero sample learning method based on global semantic consensus network | |
| CN107657603A (en) | A kind of industrial appearance detecting method based on intelligent vision | |
| CN103439342A (en) | Infrared nondestructive testing method based on thermal image time sequence characteristics | |
| CN111192236B (en) | SVI measurement method based on activated sludge phase-contrast microscopic image | |
| CN108395986A (en) | Human papilloma virus automatic parting direction detection device based on deep learning | |
| CN115216553B (en) | Digital plasmodium falciparum detection system and detection method based on deep learning | |
| CN116681962A (en) | Power equipment thermal image detection method and system based on improved YOLOv5 | |
| Winata et al. | Prediction of concentration for microalgae using image analysis | |
| CN112115826A (en) | Face living body detection method and system based on bilateral branch network | |
| CN103954548A (en) | Testing device used for testing metal corrosivity of liquid | |
| CN114882501A (en) | Circulating tumor cell detection system based on interpretable deep learning algorithm | |
| CN118190761A (en) | Inter-crystal corrosion control device and method based on deep learning image recognition | |
| CN104237067A (en) | Glass hydrometer detecting/calibrating device | |
| CN117310962A (en) | Microscopic imaging focusing control method and device | |
| Nielsen et al. | Novel strategies for predictive particle monitoring and control using advanced image analysis | |
| CN115187302A (en) | Water dispenser putting prediction method and system based on GA-BP neural network | |
| CN110045691A (en) | A kind of multitasking fault monitoring method of multi-source heterogeneous big data | |
| CN105987866A (en) | Heterogeneous liquid settlement automatic-monitoring method and device | |
| Deng et al. | Automatic estimation of rice grain number based on a convolutional neural network | |
| CN115330705A (en) | Skin paint surface defect detection method based on adaptive weighting template NCC | |
| CN114494850A (en) | Village unmanned courtyard intelligent identification method and system | |
| CN208857289U (en) | Human papilloma virus automatic parting direction detection device based on deep learning |
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 |