CN110222814A - The pipe recognition methods again of Ethylene Cracking Furnace Tubes based on embedded DCNN - Google Patents
The pipe recognition methods again of Ethylene Cracking Furnace Tubes based on embedded DCNN Download PDFInfo
- Publication number
- CN110222814A CN110222814A CN201910339874.2A CN201910339874A CN110222814A CN 110222814 A CN110222814 A CN 110222814A CN 201910339874 A CN201910339874 A CN 201910339874A CN 110222814 A CN110222814 A CN 110222814A
- Authority
- CN
- China
- Prior art keywords
- dcnn
- pipe
- data
- temperature
- boiler tube
- 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.)
- Granted
Links
Classifications
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
- Radiation Pyrometers (AREA)
Abstract
The present invention discloses a kind of pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN, the method includes acquisition outer surface of furnace tube temperature and distance and inboard wall of burner hearth temperature and distance, the DCNN again building of pipe identification model, the DCNN again reconstruct of pipe identification model and the metering of outer surface of furnace tube temperature, the outer surface of furnace tube temperature being calculated to upload to Cloud Server, the distance feature difference with non-heavy pipe is managed again first with the data collected, and training generates DCNN pipe identification model again;Secondly it is directed to the characteristic of embeded processor, DCNN again pipe identification model is transplanted to the embeded processor inside intelligent temperature measurement instrument;Finally by the data processing algorithm and temperature value metering method inside DCNN again pipe identification model combination intelligent temperature measurement instrument, the temperature managed again with non-heavy pipe is calculated.This method energy high accurancy and precision differentiates that cracking furnace tube is managed again and non-heavy pipe, the accuracy of raising cracking furnace tube hull-skin temperature measurement realize the edge calculations function of intelligent temperature measurement instrument.
Description
Technical field
The present invention relates to cracking furnace tube hull-skin temperature monitoring technical fields, more particularly, to one kind based on insertion
The Ethylene Cracking Furnace Tubes of formula DCNN again pipe identification and cracking furnace tube hull-skin temperature calculation method.
Background technique
Core of the ethylene industry as petrochemical industry, in the world using ethylene yield as one national oil of measurement
The one of the important signs that of development of chemical industry level.Pyrolysis furnace is the core equipment of ethylene industry, and cracking furnace tube is its key portion again
Part.Since cracking furnace tube is chronically in the flue gas of thermal-flame, unavoidably occur coking inside boiler tube in production and
Caused by boiler tube hot-spot phenomena such as, cause boiler tube carburizing, cracking, leak stopping, fistulae, be thinned etc. forms failure.It is cracking
In the various forms of furnace boiler tube failure, there is direct relation with temperature mostly, therefore ethylene production enterprise extremely payes attention to outside furnace boiler tube
The monitoring of surface temperature.
Currently, widely applied cracking furnace tube hull-skin temperature thermometric side examination is predominantly contactless by man-hour manually hand-held
Infrared temperature-measuring gun carries out positioning thermometric to boiler tube.Since boiler tube often has shaking to pyrolysis furnace in the process of running, so by such
The disadvantages of mode carries out manually aiming at thermometric, and poor, randomness that there are measurement accuracy is by force, surveyed boiler tube number is few.For above-mentioned survey
Deficiency existing for warm mode, present invention applicant have developed a kind of cracking furnace tube outer surface contactless smart temperature measuring equipment
(hereinafter referred to as " intelligent temp-measuring device "), realizes the automatic measurement of cracking furnace tube hull-skin temperature.Intelligent temperature measurement dress
It sets and is made of four parts, respectively cracking furnace tube outer surface contactless smart temperature measurer (hereinafter referred to as " intelligent temperature measurement
Instrument "), driving platform, remote controler and guide rail.By applying the intelligent temp-measuring device in actual production, pyrolysis furnace is effectively increased
The measurement accuracy of outer surface of furnace tube temperature significantly reduces the workload of operator.However, using the intelligent temp-measuring device
During to pyrolysis furnace outer surface of furnace tube temperature monitoring, can also find boiler tube can be frequently present of block mutually, overlapping
Phenomenon, this phenomenon are also referred to as pipe of attaching most importance to.The presence of pipe phenomenon again will lead to and distinguish and calculating outside every cracking furnace tube
There are certain difficulty during surface temperature.It is to pass through in the recognizer of pipe again of the original proposition of present invention applicant
Pipe collection point number range and frontier distance jump threshold value again for setting, differentiate whether boiler tube attaches most importance to pipe as Rule of judgment.
However, in the actual production process, the strong air flow and vibration that cracking furnace tube drives in high temperature, high pressure, burner jet flames
It under rotating ring border, can make the position moment of boiler tube that the variation within the scope of small distance occur, while also result in the shape of boiler tube overlapping
Condition is varied, only only in accordance with pipe collection point number again number and frontier distance jump threshold value go to judge whether to attach most importance to pipe, deposit
In very big unreliability.
Summary of the invention
The present invention is directed to overcome at least one of the above-mentioned prior art insufficient, a kind of cracking of ethylene based on embedded DCNN is provided
The pipe recognition methods again of furnace boiler tube, this method energy high accurancy and precision differentiate that pyrolysis furnace is managed and non-heavy pipe, raising cracking furnace tube appearance again
The accuracy of face temperature measurement, realizes the edge calculations function of intelligent temperature measurement instrument.Collection inside intelligent temperature measurement instrument in the present invention
At technical grade infrared measurement of temperature module and laser ranging module, for realizing the synchro measure of temperature and distance.
The technical solution adopted by the present invention is that:
A kind of pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN is provided, comprising the following steps:
S101. the acquisition of data: it is internally integrated the intelligent temperature measurement instrument acquisition furnace of infrared measurement of temperature module and laser ranging module
The original one-dimensional data of tube outer surface temperature and distance and inboard wall of burner hearth temperature and distance;
S102. based on depth convolutional neural networks (Deep Convolution Neural Network, DCNN), pipe is known again
The building of other model: the collected original one-dimensional data of intelligent temperature measurement instrument is transferred to the end PC, and the end PC is by an original dimension for acquisition
According to being transformed to two-dimensional histogram;In two-dimensional histogram to boiler tube range data carry out feature extraction, obtain boiler tube pipe again and
The distance feature figure of non-heavy pipe, to constitute the data set of DCNN network model training;Again by the DCNN net at the data set input end PC
Network model is trained, in the end PC building DCNN again pipe identification model;
S103. the reconstruct based on DCNN again pipe identification model: the trained DCNN in the end PC again pipe identification model is migrated to
Embeded processor inside intelligent temperature measurement instrument, DCNN again pipe identification model to be reconstructed in embeded processor;
S104. the metering of outer surface of furnace tube temperature: intelligent temperature measurement instrument acquire in real time outer surface of furnace tube temperature and distance and
The original one-dimensional data of inboard wall of burner hearth temperature and distance extracts the boiler tube range data in original one-dimensional data, and returns to boiler tube
Starting position coordinates of the range data in original one-dimensional data recycle the DCNN pipe identification model again in embeded processor
Judge whether the corresponding boiler tube of boiler tube range data extracted attaches most importance to pipe, return to the identification types label of every boiler tube, according to returning
The starting position coordinates of the boiler tube identification types label and boiler tube range data that return, to the non-heavy pipes of corresponding starting position coordinates with
The outer surface of furnace tube temperature value of pipe is measured again;
S105. be uploaded to Cloud Server: the outer surface of furnace tube temperature being calculated is uploaded to cloud service by intelligent temperature measurement instrument
Device.
The present invention is preferably the insertion used using Cortex-M7 as the STM32F767VET6 of kernel as intelligent temperature measurement instrument
Formula processor, the optimization software kernel that neural network is disposed on embeded processor is CMSIS-NN, wherein in CMSIS-NN
Verify the turnover of materials stored the power functions such as depth convolution, Chi Hua needed for containing DCNN network implementations, activation and full connection so that we
DCNN is reconstructed in embeded processor, and pipe identification model is possibly realized again.
The invention firstly uses the distance feature difference managed with non-heavy pipe again, training generates DCNN pipe identification model again,
DCNN again pipe identification model is transplanted to the embedded processing inside intelligent temperature measurement instrument by the secondary characteristic for embeded processor
Device, then, then by the data processing algorithm and temperature value metering method inside DCNN again pipe identification model combination intelligent temperature measurement instrument,
The temperature managed again with non-heavy pipe is calculated.The DCNN that present invention training generates again manage again by the differentiation of pipe identification model energy high accurancy and precision
With non-heavy pipe, the accuracy of cracking furnace tube hull-skin temperature measurement is improved, the edge calculations function of intelligent temperature measurement instrument is realized.
Preferably, in step S102, during carrying out feature extraction to boiler tube range data in two-dimensional histogram, first
Remove the temperature data of outer surface of furnace tube and inboard wall of burner hearth in two-dimensional histogram, then remove inboard wall of burner hearth in two-dimensional histogram away from
After data, invalid threshold data low with the boiler tube feature degree of correlation in boiler tube range data is removed, extraction obtains the weight of boiler tube
Thus pipe and non-heavy pipe distance feature figure constitute the data set of DCNN network model training.
Preferably, in step S102, the DCNN again pipe identification model DCNN network structure include 1 input layer, 3
Convolutional layer, 3 pond max layers, 1 full-mesh layer and an output layer, each layer are successively distributed in the following order: input
Layer, the first convolutional layer, the first pond layer, the second convolutional layer, the second pond layer, third convolutional layer, third pond layer, full-mesh layer
And output layer.The input of DCNN network structure is the boiler tube signature grey scale figure of a 32x32 pixel, and output is attached most importance to pipe and non-heavy
The class probability of pipe.
It, can be by the data set of extraction managed and non-heavy pipe distance feature figure is constituted again after setting DCNN network structure
Input DCNN network model is trained, and in the training process, the parameter by modifying DCNN network structure promotes DCNN network
The recognition accuracy of model finally obtains the DCNN of desired accuracy rate pipe identification model again.
It preferably, further include by 32 floating-points of the trained DCNN in the end PC again pipe identification model before step S103
After type weight fixed point turns to 8 integer type weights, then the step being transplanted in embeded processor.
DCNN network model is in the training process at the end PC, usually using the weight and activation primitive progress of 32 floating types
Training, however the usual Installed System Memory of embeded processor is limited, and the DCNN network model training weight of 32 floating point types is determined
Point turns to 8 weights and then is transplanted in embeded processor again, and the scale of DCNN network model can be made to reduce 4 times, this
Outside, in embeded processor, the speed of integer arithmetic is faster than floating-point operation very much, so, we are in transplanting DCNN network
Before model, need 32 floating point type weights fixed point of the trained DCNN in the end PC again pipe identification model turning to 8 integers
Then type, then is transplanted in embeded processor.
It is further preferred that 32 floating point type weights fixed point of the trained DCNN in the end PC again pipe identification model is turned to
The specific steps of 8 integer type weights are as follows:
32 floating numbers of the trained DCNN in the end PC again pipe identification model are first converted to 8 approximations by S201;
S202 carries out complement code processing to resulting approximation again, obtains the fixed-point number of approximation by fixed point formula, finally
The weight and activation value of all trained DCNN pipe identification model again are indicated by 8 fixed-point numbers.
11. it is further preferred that 32 floating numbers of the trained DCNN in the end PC again pipe identification model are turned in step S301
Being changed to 8 fixed-point numbers, detailed process is as follows:
The representation of fixed-point number are as follows: [QI:QF], wherein QI and QF corresponds respectively to integer and fractional part, and fixed-point number also wraps
A sign bit is included, is used to numeral positive and negative;Length FL, the symbol of the length IL of fixed-point number integer part, fractional part
Shown in the relationship such as formula (1) of the bit wide B of the length and fixed-point number digit of position:
B=FL+IL+1 (1)
For given set of number S, required integer part length is provided by formula 2:
In formulaExpression rounds up, and x indicates any value in given set of number S;
For being converted to the fixed-point number that specified bit wide is N, the method for determination of integer part length IL is as shown in Equation 3:
It is by the length that formula (1) can calculate fractional part
FL=N-IL-1 (4)
The minimum positive number that fixed-point number mode indicates is defined, the as expression precision of fixed-point number is ε, and formula is as follows:
ε=2-FL (5)
The floating point number given for one, converts thereof into the approximation of specified bit wide, conversion formula according to the following formula are as follows:
It will in formulaIt is defined as less than the value equal to x and is the maximum integer times about ε, fixed (x) indicates the specific bit of x
Wide approximation;
For giving the fixed-point number of bit wide, the approximate value range of floating point number are as follows:
After raw value approximation finishes, the fixed point of approximation is next realized, fixed point formula is shown below:
X represents the complement of two's two's complement form of approximation in formula, and i indicates the value in [0, B-2] section, and value indicates approximation
Fixed point value.
After the completion of the weight fixed point of DCNN training pattern, the weight after fixed point is added to the insertion of intelligent temperature measurement instrument
In formula processor program, the transplanting of weight is completed, further according to the DCNN network model of the end PC training, the structure in embeded processor
The DCNN network model of same structure is built, then in conjunction with the DCNN network weight after fixed point, DCNN can be realized, and pipe is known again
Reconstruct of the other model in embeded processor.
Preferably, after step S103, the original one-dimensional boiler tube range data of acquisition is converted using data processing algorithm
For the 2-D data of DCNN network structure input form, to realize DCNN fortune of the pipe identification model in embeded processor again
Row.
It is further preferred that the specific steps of the data processing algorithm are as follows:
S401 is according to the distance difference feature of inboard wall of burner hearth and outer surface of furnace tube, out of, acquisition outer surface of furnace tube and burner hearth
The distance value of each boiler tube is extracted in wall range data;
The boiler tube distance value extracted is done eigentransformation by S402, obtains one-dimensional boiler tube characteristic;
The one-dimensional boiler tube characteristic that transformation obtains is done dimension transformation by S403, obtains that the input of DCNN network structure can be used for
2-D data.
Preferably, in step S104, specific steps that the temperature of non-heavy Guan Yuchong pipe is measured are as follows:
S501 extracts the corresponding temperature data of boiler tube range data starting position coordinates, removal when boiler tube is non-heavy pipe
Lip temperature point, then the average value of residuals temperatures data is calculated, obtain the temperature value of current boiler tube;
S502 first extracts corresponding temperature number according to the range data starting position coordinates of pipe again when boiler tube attaches most importance to pipe
According to then finding by the boiler tube edge trip point in temperature data the segmentation boundary of furnace tube temperature data, and by pipe temperature data again
Cutting is several segments, according to non-heavy pipe temperature treatment method, calculates the furnace tube temperature data that cutting is completed, obtains each furnace in pipe again
The different temperatures value of pipe.
Preferably, the intelligent temperature measurement instrument is built-in with LoRa wireless module, and the outer surface of furnace tube temperature being calculated passes through
Built-in LoRa wireless module uploads to Cloud Server, or, the intelligent temperature measurement instrument is built-in with LoRa wireless module, is calculated
Outer surface of furnace tube temperature Edge Server is uploaded to by built-in LoRa wireless module, Edge Server is uploaded to cloud service
Device;Or, Cloud Server diagnose to the boiler tube operation conditions in production process to receiving outer surface of furnace tube temperature data
To diagnosis situation, or, Cloud Server to receive outer surface of furnace tube temperature data to the boiler tube operation conditions in production process into
Row diagnosis obtains diagnosis situation, and diagnosis situation is transmitted to control workshop and/or operator's terminal.
Compared with prior art, the invention has the benefit that
One, distance feature difference of the present invention using pipe and non-heavy pipe again, training generation DCNN pipe identification model, and needle again
To the characteristic of embeded processor, DCNN again pipe identification model is transplanted to the embeded processor inside intelligent temperature measurement instrument, then
By the data processing algorithm and temperature value metering method inside DCNN again pipe identification model combination intelligent temperature measurement instrument, weight is calculated
The temperature of pipe and non-heavy pipe, pipe identification model energy high accurancy and precision differentiates pipe and non-heavy pipe again to the DCNN that present invention training generates again,
The accuracy for improving the measurement of cracking furnace tube hull-skin temperature value, realizes the edge calculations function of intelligent temperature measurement instrument.
Two, the edge calculations function of intelligent temperature measurement instrument is by intensive calculating task from the centralized node of industrial Cloud Server
Move to the network edges such as intelligent temperature measurement instrument, the network edges sides such as intelligent temperature measurement instrument close to mobile device and data source header just
The nearly Edge intelligence for providing cracking furnace pipe temperature value calculates service, reduce the round-trip cloud of magnanimity initial data waiting time and
Network cost improves the real-time and high efficiency of data processing while reducing the data processing amount of industrial Cloud Server.
Three, the outer surface of furnace tube temperature that intelligent temperature measurement instrument is calculated in real time can upload to Cloud Server by network,
Monitoring and diagnosis by Cloud Server to temperature can in real time feed back diagnostic result to operation workshop and operator institute
In the terminal held, so as in the case where the temperature value of cracking furnace pipe is abnormal, operating workshop and operator can be with
Corresponding movement is made, in time to guarantee the normal operation of cracking production process.
Detailed description of the invention
Fig. 1 is the flow chart of the cracking furnace tube hull-skin temperature metering method.
Fig. 2 is the flow chart of cracking furnace tube pipe recognition methods specific steps again.
Fig. 3 is the two-dimensional histogram of raw measurement data.
Fig. 4 is the flow chart that boiler tube range data carries out feature extraction.
Fig. 5 (a) is the characteristic pattern of the non-heavy pipe of boiler tube.
Fig. 5 (b) is the characteristic pattern of boiler tube pipe again.
Fig. 6 is the DCNN network structure of DCNN pipe identification model again.
Fig. 7 is the data processing algorithm flow chart.
Fig. 8 is the furnace tube temperature change curve using pyrolysis furnace of the present invention weight tube temperature degree recognition methods acquisition.
Fig. 9 is the furnace tube temperature change curve one acquired using the method for document 1.
Figure 10 is the furnace tube temperature change curve two acquired using the method for document 1.
Figure 11 is the furnace tube temperature change curve using the acquisition of traditional artificial measurement method.
Specific embodiment
Attached drawing of the present invention only for illustration, is not considered as limiting the invention.It is following in order to more preferably illustrate
Embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent the size of actual product;For art technology
For personnel, the omitting of some known structures and their instructions in the attached drawings are understandable.
Embodiment 1
As shown in Figure 1, a kind of pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN, including following step
It is rapid:
S101. it the acquisition of data: is internally integrated outside the intelligent temperature measurement instrument acquisition boiler tube of infrared measurement of temperature module and laser ranging module
The original one-dimensional data of surface temperature and distance and inboard wall of burner hearth temperature and distance;Intelligent temperature measurement instrument has temperature and apart from same
The function of pacing amount, meeting synchro measure boiler tube is apart from intelligent temperature measurement instrument while intelligent temperature measurement instrument measures outer surface of furnace tube temperature
Distance, also can distance of the synchro measure inboard wall of burner hearth apart from intelligent temperature measurement instrument when measuring inboard wall of burner hearth temperature;
S102. the building based on depth convolutional neural networks DCNN again pipe identification model: intelligent temperature measurement instrument collected original one
Dimension data is transferred to the end PC, and the original one-dimensional data of acquisition is transformed to two-dimensional histogram by the end PC;To furnace in two-dimensional histogram
Pipe range data carries out feature extraction, the distance feature figure of boiler tube managed again with non-heavy pipe is obtained, to constitute DCNN network model
Trained data set;The DCNN network model at the data set input end PC is trained again, to know in the end PC building DCNN again pipe
Other model;
S103. the reconstruct based on DCNN again pipe identification model: the trained DCNN in the end PC again pipe identification model is migrated to
Embeded processor inside intelligent temperature measurement instrument, DCNN again pipe identification model to be reconstructed in embeded processor;
S104. the metering of outer surface of furnace tube temperature: intelligent temperature measurement instrument acquire in real time outer surface of furnace tube temperature and distance and
The original one-dimensional data of inboard wall of burner hearth temperature and distance extracts the boiler tube range data in original one-dimensional data, and returns to boiler tube
Starting position coordinates of the range data in original one-dimensional data recycle the DCNN pipe identification model again in embeded processor
Judge whether the corresponding boiler tube of boiler tube range data extracted attaches most importance to pipe, return to the identification types label of every boiler tube, according to returning
The starting position coordinates of the boiler tube identification types label and boiler tube range data that return, to the non-heavy pipes of corresponding starting position coordinates with
The outer surface of furnace tube temperature value of pipe is measured again;
Be uploaded to Cloud Server: the outer surface of furnace tube temperature being calculated is uploaded to Cloud Server by intelligent temperature measurement instrument.
In recent years, with the proposition of intelligent plant concept, industrial Internet of Things (IIoT) and edge calculations become grinds instantly
The hot spot studied carefully.Under this overall background, cracking of ethylene industry is no exception, gradually towards using edge calculations as the industrial object of core
Networking process strides forward.Cracking furnace tube temperature monitoring is as the necessary links for ensureing that cracking of ethylene normally produces, with existing intelligence
The phenomenon that cracking furnace tube is overlapped can be often found during temperature measurer thermometric, cause the temperature of different boiler tubes to be difficult to differentiate between, nothing
Method precisely detects the temperature variations of every boiler tube.
With the development of artificial intelligence, convolutional neural networks are widely used by feat of its powerful ability in feature extraction
In automatic control, pattern-recognition, computer vision, sensor signal processing etc..The LeNet-5 of the propositions such as Yann LeCun
Convolutional neural networks model is successfully applied to the identification of handwritten numeral on banker's check, this is that convolutional neural networks are big for the first time
Range is applied in industrial practice, achieves good application effect.The Alex-Net convolutional Neural of the designs such as Krizhevsky
Network model further improves convolutional neural networks in the accuracy rate of field of image recognition.
While Internet of Things rapid development, using Cortex-M as the embeded processor (Cortex-M CPU) of kernel
Significant progress is achieved, especially has very great Cheng in operation dominant frequency, memory size using Cortex-M7 as the processor of kernel
The promotion of degree, and it is directed to Cortex-M CPU, ARM company proposes one kind and is exclusively used for disposing nerve on Cortex-M CPU
The optimization software kernel CMSIS-NN of network.Cortex-M CPU is based on CMSIS-NN kernel and carries out ANN Reasoning operation,
4.6 times of promotion will be had for runing time, handling capacity, will also have 4.9 times of promotion for efficiency, so that embedded processing
Device, which incorporates Internet of Things and is embedded in neural network, becomes possibility.The present invention preferably uses Cortex-M as intelligent temperature measurement instrument
Embeded processor, wherein Cortex-M contains DCNN net preferably using CMSIS-NN as kernel in CMSIS-NN kernel library
Network realizes the power functions such as required convolution, Chi Hua, activation and full connection, so that we reconstruct DCNN in embeded processor
Pipe identification model is possibly realized again.
The invention firstly uses the distance feature difference managed with non-heavy pipe again, training generates DCNN pipe identification model again,
DCNN again pipe identification model is transplanted to the embedded processing inside intelligent temperature measurement instrument by the secondary characteristic for embeded processor
Device, then, then by the data processing algorithm and temperature value metering method inside DCNN again pipe identification model combination intelligent temperature measurement instrument,
The temperature managed again with non-heavy pipe is calculated, the DCNN that present invention training generates again manage again by the differentiation of pipe identification model energy high accurancy and precision
With non-heavy pipe, the accuracy of cracking furnace tube hull-skin temperature value measurement is improved, the edge calculations function of intelligent temperature measurement instrument is realized
Energy.The detailed process of entire method is as shown in Figure 2.
The edge calculations function of intelligent temperature measurement instrument moves intensive calculating task from the centralized node of industrial Cloud Server
The network edges such as intelligent temperature measurement instrument are moved on to, in network edges sides such as the intelligent temperature measurement instruments of close mobile device and data source header, are melted
Close network, calculating, storage, application core ability new network architecture open platform, cracking furnace pipe temperature value is provided nearby
Edge intelligence calculates service, to reduce waiting time and the network cost in the round-trip cloud of magnanimity initial data, is reducing work
The real-time and high efficiency of data processing are improved while the data processing amount of industry Cloud Server.
For intelligent temperature measurement instrument of the invention, have a function of temperature and distance synchronous measurement, this functional purpose be for
The outer surface of furnace tube temperature and inboard wall of burner hearth temperature for including in discriminating measurement temperature data.In temperature distance synchronous measurement process
In, the data of intelligent temperature measurement instrument acquisition are one-dimensional data.When all the time, by artificially proofreading data, usually measurement is obtained
The histogram that one-dimensional data is transformed to two dimensional form is identified and is managed again.The two dimension of the obtained one-dimensional initial data of original measurement
The histogram form of expression is as shown in Figure 3.
In histogram as shown in Figure 3, abscissa indicates the collected temperature of a pyrolysis furnace observation window and apart from number
The number at strong point, ordinate then indicate the scale value of acquisition distance and temperature, and light black area data field represents distance at the middle and upper levels,
Lower layer's grey area data field represents temperature.By histogram it is found that the temperature of outer surface of furnace tube and inboard wall of burner hearth in acquisition data
It is more difficult to distinguish degree for discrimination very little, and there is significant difference, upper layer light black areas for the distance of boiler tube and inboard wall of burner hearth
Data field depressed section is the boiler tube measured.So differentiation outer surface of furnace tube can be reached according to the feature of distance difference
With the purpose of inboard wall of burner hearth temperature.It is to manage again shown in the bar mark frame being known that in Fig. 3 histogram through analysis, by Fig. 3
Bar mark frame in histogram has also been found that a problem, and the degree of the boiler tube overlapping in measurement process is not quite similar, by
One problem of this bring is exactly to be overlapped boiler tube and how to pass through algorithm to be identified.It is original by adopting during actual measurement
Integrate the jump threshold value of data point number and distance as standard to determine whether the algorithm for pipe of attaching most importance to, often exists and fail to judge and misjudge
The problem of, thus propose that a kind of high-accuracy identifies that again the method for pipe is particularly important.
Gone by artificially observing histogram identification pipe again accuracy be it is very high, this is heavily dependent on two dimension
The intuitive of data, and the input of DCNN network is bivector, is inspired by this, proposes and carries out weight based on DCNN network
Pipe knows method for distinguishing.
Wherein, as shown in figure 4, in step S102, feature extraction is carried out to boiler tube range data in two-dimensional histogram
In the process, the temperature data of outer surface of furnace tube and inboard wall of burner hearth in two-dimensional histogram is first removed, then removes furnace in two-dimensional histogram
After the range data of thorax inner wall, invalid threshold data low with the boiler tube feature degree of correlation in boiler tube range data is removed, is extracted
To boiler tube manage again and the distance feature figure of non-heavy pipe (wherein, Fig. 5 (a) is the distance feature figure of the non-heavy pipe of boiler tube, and Fig. 5 (b) is
The distance feature figure of boiler tube pipe again), thus constitute the data set of DCNN network model training.Wherein, invalid threshold value be boiler tube away from
From minimum point following data in data, specifically, first extracting boiler tube after the range data of inboard wall of burner hearth in removal two-dimensional histogram
Minimum value in range data makes the difference each range data of boiler tube with the minimum value, finally obtain boiler tube and again pipe and it is non-heavy
The distance feature figure of pipe.
Wherein, as shown in fig. 6, in step S102, DCNN DCNN network structure packet 1 of pipe identification model input again
Layer, 3 convolutional layers, 3 pond max layers, 1 full-mesh layer and an output layer, each layer are successively distributed in the following order:
Input layer, the first convolutional layer, the first pond layer, the second convolutional layer, the second pond layer, third convolutional layer, third pond layer, Quan Lian
Logical layer and output layer.The input of DCNN network structure is the boiler tube signature grey scale figure of a 32x32 pixel, output attach most importance to pipe and
The class probability of non-heavy pipe.
It, can be by the data of extraction managed and the distance feature figure of non-heavy pipe is constituted again after setting DCNN network structure
Collection input DCNN network model is trained, and in the training process, the parameter by modifying DCNN network structure promotes DCNN net
The recognition accuracy of network model finally obtains the DCNN of desired accuracy rate pipe identification model again.
It wherein, further include by 32 floating-point classes of the trained DCNN in the end PC again pipe identification model before step S103
After type weight fixed point turns to 8 integer type weights, then the step being transplanted in embeded processor.
DCNN network model is in the training process at the end PC, usually using the weight and activation primitive progress of 32 floating types
Training, however the usual Installed System Memory of embeded processor is limited, and the DCNN network model training weight of 32 floating point types is determined
Point turns to 8 weights and then is transplanted in embeded processor again, and the scale of DCNN network model can be made to reduce 4 times, this
Outside, in embeded processor, the speed of integer arithmetic is faster than floating-point operation very much, so, we are in transplanting DCNN network
Before model, need 32 floating point type weights fixed point of the trained DCNN in the end PC again pipe identification model turning to 8 integers
Then type, then is transplanted in embeded processor.
Specifically, by 32 floating point type weights of the trained DCNN in the end PC again pipe identification model fixed point turn to 8 it is whole
The specific steps of number type weight are as follows:
S301., 32 floating numbers of the trained DCNN in the end PC again pipe identification model are first converted to 8 approximations;
S302. complement code processing is carried out to resulting approximation again, the fixed-point number of approximation is obtained by fixed point formula, by 8
The fixed-point number of position indicates the weight and activation value of all trained DCNN pipe identification model again.
More specifically, 32 floating numbers of the trained DCNN in the end PC again pipe identification model are converted to 8 in step S301
Detailed process is as follows for position fixed-point number:
The representation of fixed-point number are as follows: [QI:QF], wherein QI and QF corresponds respectively to integer and fractional part, and fixed-point number also wraps
A sign bit is included, is used to numeral positive and negative;Length FL, the symbol of the length IL of fixed-point number integer part, fractional part
Shown in the relationship such as formula (1) of the bit wide B of the length and fixed-point number digit of position:
B=FL+IL+1 (1)
For given set of number S, required integer part length is provided by formula 2:
In formulaExpression rounds up, and x indicates any value in given set of number S;
For being converted to the fixed-point number that specified bit wide is N, the method for determination of integer part length IL is as shown in Equation 3:
It is by the length that formula (1) can calculate fractional part
FL=N-IL-1 (4)
The minimum positive number that fixed-point number mode indicates is defined, the as expression precision of fixed-point number is ε, and formula is as follows:
ε=2-FL (5)
The floating point number given for one, converts thereof into the approximation of specified bit wide, conversion formula according to the following formula are as follows:
It will in formulaIt is defined as less than the value equal to x and is the maximum integer times about ε, fixed (x) indicates the specific bit of x
Wide approximation;
For giving the fixed-point number of bit wide, the approximate value range of floating point number are as follows:
After raw value approximation finishes, the fixed point of approximation is next realized, fixed point formula is shown below:
X represents the complement of two's two's complement form of approximation in formula, and i indicates the value in [0, B-2] section, and value indicates approximation
Fixed point value.
After the completion of the weight fixed point of DCNN training pattern, the weight after fixed point is added to the insertion of intelligent temperature measurement instrument
In formula processor program, the transplanting of weight is completed, further according to the DCNN network model of the end PC training, the structure in embeded processor
The DCNN network model of same structure is built, then in conjunction with the DCNN network weight after fixed point, DCNN can be realized, and pipe is known again
Reconstruct of the other model in embeded processor.
Wherein, as shown in fig. 7, after step S103, using data processing algorithm by the original one-dimensional boiler tube distance of acquisition
Data are converted to the 2-D data of DCNN network structure input form, with realize DCNN again pipe identification model in embeded processor
Interior operation.
Specifically, the specific steps of the data processing algorithm are as follows:
S401 is according to the distance difference feature of inboard wall of burner hearth and outer surface of furnace tube, from the boiler tube and inboard wall of burner hearth distance of acquisition
The distance value of each boiler tube is extracted in data, the Midtone data that boiler tube extracts in data represent tube skin distance,
Both sides color area data source is in inboard wall of burner hearth distance, for representing the boundary of boiler tube and inboard wall of burner hearth;
The boiler tube distance value extracted is done eigentransformation by S402, obtains one-dimensional boiler tube characteristic, processing method are as follows:
Tube skin distance in boiler tube data is individually subtracted minimum value therein, the cut off value of outer surface of furnace tube and inboard wall of burner hearth is then
It is transformed to 300;
The one-dimensional boiler tube characteristic that transformation obtains is done dimension transformation by S403, obtains that the input of DCNN network structure can be used for
2-D data, dimension transformation algorithm description such as (Algorithm 1).
Wherein, in step S104, specific steps that the temperature value of non-heavy Guan Yuchong pipe is measured are as follows:
S501 extracts the corresponding temperature data of boiler tube range data starting position coordinates, removal when boiler tube is non-heavy pipe
Lip temperature point, then the average value of residuals temperatures data is calculated, obtain the temperature value of current outer surface of furnace tube;
S502 first extracts corresponding temperature number according to the range data starting position coordinates of pipe again when boiler tube attaches most importance to pipe
According to then finding by the boiler tube edge trip point in temperature data the segmentation boundary of boiler tube data, and by pipe temperature data cutting again
The outer surface of furnace tube temperature data that cutting is completed is calculated according to non-heavy pipe temperature treatment method for several segments, is obtained each in pipe again
The temperature value of outer surface of furnace tube.
Intelligent temperature measurement instrument is built-in with LoRa wireless module, the outer surface of furnace tube temperature being calculated by built-in LoRa without
Wire module uploads to Edge Server, and Edge Server is uploaded to Cloud Server;The same of data storage is carried out by Cloud Server
When, temperature data can also be diagnosed by the diagnostic system run in Cloud Server, export diagnostic result, and will knot
Fruit returns to control workshop.
Cloud Server can analyze every cracking in production process by diagnosing to cracking furnace tube temperature data in real time
The operation conditions of boiler tube, and diagnosis situation is transferred in control workshop and operator's hand in time.In the temperature of cracking furnace pipe
In the case that degree value is abnormal, corresponding movement can be made in time, to guarantee the normal operation of cracking production process.
Embodiment 2
Cracking furnace tube of the present invention again pipe recognition methods is applied to the ethylene cracker of Large-Scale Petrochemical Companies.
The ethylene cracker has multiple pyrolysis furnaces, and each pyrolysis furnace has 8 peep holes and 96 boiler tubes, and 12 furnaces can be observed in every hole
Pipe.In order to verify effectiveness of the invention, the actual measurement experiment of two aspects is carried out, and analyzed experimental result.
1, the DCNN training of pipe identification model and emulation again
During cracking of ethylene, pyrolysis furnace is run in the environment of high temperature, high pressure and sharp pounding, leads to boiler tube position
Variation occurs for the moment, and the boiler tube position that different periods measurement obtains has difference in varying degrees.This experiment is used for
DCNN pipe identification model is trained again test set and training set are collected in different periods, the sample structure of training set and test set
At as shown in table 1:
The verification process of DCNN pipe identification model again: DCNN network model is trained at the end PC first with test set
And it verifies, the accuracy rate of verified training pattern are as follows: 99.85%;Next the DCNN network model completed to training is weighed
It converts again, generates the data type that can run in embeded processor, the DCNN network model after converting collects after tested
It verifies again, accuracy rate are as follows: 99.70%.Accuracy rate is analyzed, pipe identification model complies fully with actual production to the DCNN of generation again
The accuracy rate range being applicable in.
1 experimental data structure composition of table
Sample | Non- heavy pipe | It manages again | It amounts to |
Training sample | 980 | 560 | 1540 |
Test sample | 620 | 230 | 850 |
2, the comparative experiments of Ethylene Cracking Furnace Tubes based on embedded DCNN pipe recognition methods and other methods again
In order to verify the superiority of method proposed by the invention, we are real with No. 5 pyrolysis furnaces in ethylene cracker
Object is tested, has done comparative test using method of the present invention and document 1 and traditional artificial measurement method.By actual production
Situation it is found that distribution and outer surface of furnace tube temperature of the cracking furnace tube in different periods be it is changed, in order to true
The real reliability of experiment is protected, in an experiment, this experiment uses different measurement methods, is measured outside 7 days boiler tubes stage by stage
Surface temperature data.The outer surface of furnace tube temperature obtained according to 3 kinds of method measurements, the variation for depicting outer surface of furnace tube temperature are bent
Line.Since during actual measurement, there is again the probability highest of pipe in No. 6 peep holes of No. 5 pyrolysis furnaces, in order to embody the present invention
Advantage of the method in terms of pipe recognition accuracy again, Fig. 8-11 only delineate 12 furnaces that No. 6 peep holes are observed
The hull-skin temperature change curve of pipe.
As seen from Figure 8, when being measured using method of the invention, in 7 days of measurement, No. 6 observations of No. 5 furnaces
12 cracking furnace tube hull-skin temperatures observed by hole are all in the trend of rising, and true via cracking of ethylene factory technique person
Recognize, diagram trend meets the changing rule of outer surface of furnace tube temperature in practical cracking production process;By the survey introduced through document 1
The outer surface of furnace tube temperature variation curve (shown in Fig. 9) that amount method obtains is it can be found that there are multiple temperature anomaly values and missings
Value, missing values occur in measurement the 3rd day and the 6th day data the 12nd with boiler tube, and measured value is all zero as shown in Figure 10.It is logical
Analysis is crossed, the reason of problem occurs is that the measurement method introduced by document 1 pipe identification occurs again during actual measurement
The case where failure, thus the hull-skin temperature value of two boiler tubes in pipe again can not be distinguished, so that the temperature of back boiler tube
The number order of angle value is one progressive forward, and the hull-skin temperature value positioned at No. 12 last boiler tubes is caused to lack, and calculates
It as a result is zero.And in other measurement number of days, the method for document 1 can successfully identify the pipe again in all boiler tubes, finally
The outer surface of furnace tube temperature value of calculating is similarly in normal range (NR).In conclusion the method for the document 1 is facing boiler tube position
It sets in the case where changing constantly, the accuracy rate of identification need to be improved;As seen from Figure 11, the side of traditional artificial measurement
Outer surface of furnace tube temperature measured by formula can have abnormal data value, and variation has exceeded normal range, can make to boiler tube
The health status of operation causes to judge by accident, measures the amount of labour that equally can also aggravate manual measurement again.
By above-mentioned experimental result it is found that a kind of Ethylene Cracking Furnace Tubes weight based on embedded DCNN proposed by the invention
Pipe recognition methods has highly very big on the recognition accuracy of pipe again compared to existing measurement method, and DCNN is moved
Embeded processor is planted, ethylene chemical plant edge device is realized --- the edge calculations of intelligent temperature measurement instrument, also in certain journey
The amount of labour that cracking of ethylene worker is alleviated on degree is provided greatly convenient and is ensured to cracking of ethylene production.
Wherein, the above-mentioned document 1 referred to refers to Peng Z, He J, Tan Y, et al.Study of dual-phase
drive synchronization method and temperature measurement algorithm for
measuring external surface temperatures of ethylene cracking furnace tubes
[J].Applied Petrochemical Research,2018,8(3):163-172。
Obviously, the above embodiment of the present invention is only intended to clearly illustrate technical solution of the present invention example, and
It is not the restriction to a specific embodiment of the invention.It is all made within the spirit and principle of claims of the present invention
Any modifications, equivalent replacements, and improvements etc., should all be included in the scope of protection of the claims of the present invention.
Claims (10)
1. a kind of pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN, which comprises the following steps:
S101. it the acquisition of data: is internally integrated outside the intelligent temperature measurement instrument acquisition boiler tube of infrared measurement of temperature module and laser ranging module
The original one-dimensional data of surface temperature and distance and inboard wall of burner hearth temperature and distance;
S102. pipe identification model constructs the boiler tube based on DCNN again: the collected original one-dimensional data of intelligent temperature measurement instrument is transferred to
The original one-dimensional data of acquisition is transformed to two-dimensional histogram by the end PC, the end PC;In two-dimensional histogram to boiler tube range data into
Row feature extraction obtains the distance feature figure of boiler tube managed again with non-heavy pipe, to constitute the data set of DCNN network model training;
The DCNN network model at the data set input end PC is trained again, in the end PC building DCNN again pipe identification model;
S103. pipe identification model reconstructs the boiler tube based on DCNN again: the trained DCNN boiler tube in the end PC again pipe identification model is moved
It plants to the embeded processor inside intelligent temperature measurement instrument, to carry out weight in embeded processor to DCNN again pipe identification model
Structure;
S104. the metering of outer surface of furnace tube temperature: intelligent temperature measurement instrument acquires outer surface of furnace tube temperature and distance and burner hearth in real time
The original one-dimensional data of inner wall temperature and distance extracts the boiler tube range data in original one-dimensional data, and returns to boiler tube distance
Starting position coordinates of the data in original one-dimensional data, recycling the DCNN in embeded processor, pipe identification model judges again
Whether the corresponding boiler tube of boiler tube range data of extraction attaches most importance to pipe, the identification types label of every boiler tube is returned to, according to return
The starting position coordinates of boiler tube identification types label and boiler tube range data manage the non-heavy Guan Yuchong of corresponding starting position coordinates
Outer surface of furnace tube temperature measured;
S105. be uploaded to Cloud Server: the outer surface of furnace tube temperature being calculated is uploaded to Cloud Server by intelligent temperature measurement instrument.
2. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 1, feature exist
In, in step S102, during carrying out feature extraction to boiler tube range data in two-dimensional histogram, the two-dimentional histogram of first removal
The temperature data of outer surface of furnace tube and inboard wall of burner hearth in figure, then remove in two-dimensional histogram after the range data of inboard wall of burner hearth, it goes
Except invalid threshold data low with the boiler tube feature degree of correlation in boiler tube range data, extracts and obtain managing again and non-heavy pipe for boiler tube
Distance feature figure.
3. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 1, feature exist
In, in step S102, the DCNN again pipe identification model DCNN network structure include 1 input layer, 3 convolutional layers, 3
The pond max layer, 1 full-mesh layer and an output layer, each layer are successively distributed in the following order: input layer, the first convolution
Layer, the first pond layer, the second convolutional layer, the second pond layer, third convolutional layer, third pond layer, full-mesh layer and output layer.
4. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 1, feature exist
In further including pinpointing 32 floating point type weights of the trained DCNN in the end PC again pipe identification model before step S103
After turning to 8 integer type weights, then be transplanted in intelligent temperature measurement instrument embeded processor the step of.
5. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 4, feature exist
In 32 floating point type weights of the trained DCNN in the end PC again pipe identification model fixed point is turned to 8 integer type weights
Specific steps are as follows:
32 floating numbers of the trained DCNN in the end PC again pipe identification model are first converted to 8 approximations by S301;
S302 carries out complement code processing to resulting approximation again, the fixed-point number of approximation is obtained by fixed point formula, finally by 8
Fixed-point number indicate the weight and activation value of all trained DCNN pipe identification model again.
6. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 5, feature exist
In in step S301,32 floating numbers of the trained DCNN in the end PC again pipe identification model are converted to the specific of 8 fixed-point numbers
Process is as follows:
The representation of fixed-point number are as follows: [QI:QF], wherein QI and QF corresponds respectively to integer and fractional part, and fixed-point number also wraps
A sign bit is included, is used to numeral positive and negative;Length FL, the symbol of the length IL of fixed-point number integer part, fractional part
Shown in the relationship such as formula (1) of the bit wide B of the length and fixed-point number digit of position:
B=FL+IL+1 (1)
For given set of number S, required integer part length is provided by formula 2:
In formulaExpression rounds up, and x indicates any value in given set of number S;
For being converted to the fixed-point number that specified bit wide is N, the method for determination of integer part length IL is as shown in Equation 3:
It is by the length that formula (1) can calculate fractional part
FL=N-IL-1 (4)
The minimum positive number that fixed-point number mode indicates is defined, the as expression precision of fixed-point number is ε, and formula is as follows:
ε=2-FL (5)
The floating point number given for one, converts thereof into the approximation of specified bit wide, conversion formula according to the following formula are as follows:
It will in formulaIt is defined as less than the value equal to x and is the maximum integer times about ε, fixed (x) indicates the specific bit of x
Wide approximation;
For giving the fixed-point number of bit wide, the approximate value range of floating point number are as follows:
After raw value approximation finishes, the fixed point of approximation is next realized, fixed point formula is shown below:
X represents the complement of two's two's complement form of approximation in formula, and i indicates the value in [0, B-2] section, and value indicates approximation
Fixed point value.
7. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 1, feature exist
In the original one-dimensional boiler tube range data of acquisition is converted to DCNN network knot using data processing algorithm after step S103
The 2-D data of structure input form, to realize DCNN operation of the pipe identification model in embeded processor again.
8. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 7, feature exist
In the specific steps of the data processing algorithm are as follows:
S401 is extracted from the boiler tube of acquisition and inboard wall of burner hearth range data according to the distance difference feature of inboard wall of burner hearth and boiler tube
The distance value of each boiler tube out;
The boiler tube distance value extracted is done eigentransformation by S402, obtains the distance feature data of one-dimensional boiler tube;
The one-dimensional boiler tube distance feature data that transformation obtains are done dimension transformation by S403, obtain that the input of DCNN network structure can be used for
2-D data.
9. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN according to claim 1, feature exist
In, in step S103, specific steps that the temperature of non-heavy Guan Yuchong pipe is measured are as follows:
S401 extracts the corresponding temperature data of boiler tube range data starting position coordinates, removes edge when boiler tube is non-heavy pipe
Temperature spot, then the average value of residuals temperatures data is calculated, obtain the temperature value of current outer surface of furnace tube;
S402 first extracts corresponding temperature number according to the boiler tube range data starting position coordinates of pipe again when boiler tube attaches most importance to pipe
According to then finding by the boiler tube edge trip point in temperature data the segmentation boundary of furnace tube temperature data, and by pipe temperature data again
Cutting is several segments, according to non-heavy pipe temperature treatment method, calculates the outer surface of furnace tube temperature data that cutting is completed, is managed again
In each outer surface of furnace tube temperature.
10. the pipe recognition methods again of the Ethylene Cracking Furnace Tubes based on embedded DCNN described in -9 any one according to claim 1,
It is characterized in that, the intelligent temperature measurement instrument is built-in with LoRa wireless module, and the outer surface of furnace tube temperature being calculated passes through built-in
LoRa wireless module uploads to Cloud Server, or, the intelligent temperature measurement instrument is built-in with LoRa wireless module, the boiler tube being calculated
Hull-skin temperature uploads to Edge Server by built-in LoRa wireless module, and Edge Server is uploaded to Cloud Server;Or,
Cloud Server diagnoses the boiler tube operation conditions in production process to receiving outer surface of furnace tube temperature data
Situation, or, Cloud Server diagnoses the boiler tube operation conditions in production process to receiving outer surface of furnace tube temperature data
Diagnosis situation is obtained, and diagnosis situation is transmitted to control workshop and/or operator's terminal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910339874.2A CN110222814B (en) | 2019-04-25 | 2019-04-25 | Ethylene cracking furnace tube heavy tube identification method based on embedded DCNN |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910339874.2A CN110222814B (en) | 2019-04-25 | 2019-04-25 | Ethylene cracking furnace tube heavy tube identification method based on embedded DCNN |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110222814A true CN110222814A (en) | 2019-09-10 |
CN110222814B CN110222814B (en) | 2022-11-25 |
Family
ID=67820005
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910339874.2A Active CN110222814B (en) | 2019-04-25 | 2019-04-25 | Ethylene cracking furnace tube heavy tube identification method based on embedded DCNN |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110222814B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020253484A1 (en) * | 2019-06-18 | 2020-12-24 | 广东石油化工学院 | Cracking furnace tube intelligent coking monitoring edge-cloud collaboration platform and working method therefor |
CN113344909A (en) * | 2021-07-01 | 2021-09-03 | 中国石油大学(北京) | Method and device for identifying and displaying coking of flame-permeable high-temperature filter of thermal power boiler |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1316484A (en) * | 2001-04-11 | 2001-10-10 | 中国石化集团扬子石油化工有限责任公司 | Intelligent control method of cracking furnace in ethylene preparing equipment |
CN102798294A (en) * | 2012-06-19 | 2012-11-28 | 合肥瑞石测控工程技术有限公司 | Real-time monitoring and safety warning device for temperature of furnace tube of tubular industrial furnace |
CN104374478A (en) * | 2014-11-20 | 2015-02-25 | 广东石油化工学院 | Method for distinguishing temperature of outer wall of cracking furnace tube from temperature of inner wall of hearth and measurement device |
CN204554771U (en) * | 2015-04-01 | 2015-08-12 | 西安西热电站信息技术有限公司 | A kind of boiler pipe furnace inwall temperature measurement device being applicable to extra-supercritical unit |
CN108072448A (en) * | 2017-12-13 | 2018-05-25 | 广东石油化工学院 | A kind of data processing method on the metering of cracking furnace tube outside wall temperature |
-
2019
- 2019-04-25 CN CN201910339874.2A patent/CN110222814B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1316484A (en) * | 2001-04-11 | 2001-10-10 | 中国石化集团扬子石油化工有限责任公司 | Intelligent control method of cracking furnace in ethylene preparing equipment |
CN102798294A (en) * | 2012-06-19 | 2012-11-28 | 合肥瑞石测控工程技术有限公司 | Real-time monitoring and safety warning device for temperature of furnace tube of tubular industrial furnace |
CN104374478A (en) * | 2014-11-20 | 2015-02-25 | 广东石油化工学院 | Method for distinguishing temperature of outer wall of cracking furnace tube from temperature of inner wall of hearth and measurement device |
CN204554771U (en) * | 2015-04-01 | 2015-08-12 | 西安西热电站信息技术有限公司 | A kind of boiler pipe furnace inwall temperature measurement device being applicable to extra-supercritical unit |
CN108072448A (en) * | 2017-12-13 | 2018-05-25 | 广东石油化工学院 | A kind of data processing method on the metering of cracking furnace tube outside wall temperature |
Non-Patent Citations (4)
Title |
---|
PENG Z ET AL.: "Study of dual-phase drive synchronization method and temperature measurement algorithm for measuring external surface temperatures of ethylene cracking furnace tubes", 《APPLIED PETROCHEMICAL RESEARCH》 * |
侯晶晶: "乙烯裂解炉炉管内插入扭曲片的强化传热研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》 * |
孙建兵等: "裂解炉炉管表面温度的测量", 《乙烯工业》 * |
李保有 等: ""U 型"双排排列辐射炉管的工业应用", 《计算机与应用化学》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020253484A1 (en) * | 2019-06-18 | 2020-12-24 | 广东石油化工学院 | Cracking furnace tube intelligent coking monitoring edge-cloud collaboration platform and working method therefor |
US11415469B2 (en) | 2019-06-18 | 2022-08-16 | Guangdong University Of Petrochemical Technology | Edge-cloud collaboration platform for intelligent coking monitoring of cracking furnace tubes and working method thereof |
CN113344909A (en) * | 2021-07-01 | 2021-09-03 | 中国石油大学(北京) | Method and device for identifying and displaying coking of flame-permeable high-temperature filter of thermal power boiler |
CN113344909B (en) * | 2021-07-01 | 2023-12-08 | 中国石油大学(北京) | Method and device for identifying and displaying flame penetration height Wen Lvjing coking of thermal power boiler |
Also Published As
Publication number | Publication date |
---|---|
CN110222814B (en) | 2022-11-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106761668A (en) | Oil well failure intelligent analysis decision system and method | |
CN110343541B (en) | Intelligent coking monitoring boundary cloud cooperative platform for cracking furnace tube and working method thereof | |
CN110352389A (en) | Information processing unit and information processing method | |
CN110222814A (en) | The pipe recognition methods again of Ethylene Cracking Furnace Tubes based on embedded DCNN | |
CN109374631A (en) | A kind of tunnel state evaluating method | |
CN105388884A (en) | Alarm system for detecting leakage fault of heat supply network based on identification algorithm driven by data and method | |
CN112906288A (en) | Switch cabinet state evaluation method based on normal sample data | |
CN114519923A (en) | Intelligent diagnosis and early warning method and system for power plant | |
CN106971058A (en) | A kind of pumping station operation monitoring data abnormal point detecting method | |
CN117523177A (en) | Gas pipeline monitoring system and method based on artificial intelligent hybrid big model | |
CN116595327A (en) | Sluice deformation monitoring data preprocessing system and method | |
CN110197199A (en) | Embedded DCNN and the weight tube temperature degree recognition methods of the pyrolysis furnace of edge calculations | |
CN108050397B (en) | Layering based on optical fiber source signal is sequential than pipe leakage monitoring method and system | |
CN112529003A (en) | Instrument panel digital identification method based on fast-RCNN | |
CN105222885A (en) | Optical fiber vibration detection method and device | |
CN110095204B (en) | Ethylene cracking furnace tube temperature diagnostic system | |
CN112131796A (en) | Real-time hard rock TBM hob abrasion evaluation method based on gene expression programming | |
CN116796275A (en) | Multi-mode time sequence anomaly detection method for industrial equipment | |
CN116429269A (en) | Infrared intelligent analysis system for ethylene cracking furnace tube | |
CN113973403A (en) | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measuring point topology | |
CN109492705A (en) | Method for diagnosing faults of the one kind based on mahalanobis distance (MD) area measurement | |
Zhang et al. | Fault pattern recognition of rolling bearing based on smoothness prior approach and dual-input depth spatial–temporal fusion | |
Liu et al. | Noisy labels co-detection and co-relabeling twin prototype networks for fault diagnosis of wind turbine gearboxes | |
CN110398288A (en) | A kind of cracking furnace tube hull-skin temperature metering method based on DBSCAN | |
CN114626162B (en) | Quantitative recognition method for loss degree of contact ball bearing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |