NL2033450A - A Real Time Measurement Method for Rheological Parameters of Drilling Fluid Based on Machine Learning - Google Patents
A Real Time Measurement Method for Rheological Parameters of Drilling Fluid Based on Machine Learning Download PDFInfo
- Publication number
- NL2033450A NL2033450A NL2033450A NL2033450A NL2033450A NL 2033450 A NL2033450 A NL 2033450A NL 2033450 A NL2033450 A NL 2033450A NL 2033450 A NL2033450 A NL 2033450A NL 2033450 A NL2033450 A NL 2033450A
- Authority
- NL
- Netherlands
- Prior art keywords
- drilling fluid
- rheological parameters
- real
- model
- machine learning
- Prior art date
Links
- 238000005553 drilling Methods 0.000 title claims abstract description 158
- 239000012530 fluid Substances 0.000 title claims abstract description 151
- 238000010801 machine learning Methods 0.000 title claims abstract description 24
- 238000000691 measurement method Methods 0.000 title abstract description 9
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 7
- 238000012360 testing method Methods 0.000 claims description 25
- 238000000034 method Methods 0.000 claims description 23
- 238000005259 measurement Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 17
- 230000003068 static effect Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 12
- 238000003756 stirring Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 239000007788 liquid Substances 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims 1
- 230000008676 import Effects 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 9
- 239000011159 matrix material Substances 0.000 description 4
- 238000010200 validation analysis Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 239000003795 chemical substances by application Substances 0.000 description 3
- 238000005520 cutting process Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 239000002002 slurry Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000003638 chemical reducing agent Substances 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005755 formation reaction Methods 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000000518 rheometry Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 239000003637 basic solution Substances 0.000 description 1
- 229920001222 biopolymer Polymers 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000034994 death Effects 0.000 description 1
- 231100000517 death Toxicity 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000003077 lignite Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 239000011347 resin Substances 0.000 description 1
- 229920005989 resin Polymers 0.000 description 1
- 239000004094 surface-active agent Substances 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
- 230000009974 thixotropic effect Effects 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 229920001285 xanthan gum Polymers 0.000 description 1
- 229940082509 xanthan gum Drugs 0.000 description 1
- 235000010493 xanthan gum Nutrition 0.000 description 1
- 239000000230 xanthan gum Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH DRILLING; MINING
- E21B—EARTH DRILLING, e.g. DEEP DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B21/00—Methods or apparatus for flushing boreholes, e.g. by use of exhaust air from motor
- E21B21/08—Controlling or monitoring pressure or flow of drilling fluid, e.g. automatic filling of boreholes, automatic control of bottom pressure
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N11/00—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties
- G01N11/10—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties by moving a body within the material
- G01N11/14—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties by moving a body within the material by using rotary bodies, e.g. vane
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N11/00—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties
- G01N2011/006—Determining flow properties indirectly by measuring other parameters of the system
- G01N2011/008—Determining flow properties indirectly by measuring other parameters of the system optical properties
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Abstract
A Real Time Measurement Method for Rheological Parameters of Drilling Fluid Based on Machine Learning 5 A real-time measurement method of drilling fluid rheological parameters based on machine learning, which adopts image recognition and machine learning to output the rheological parameters of drilling fluid in a flowing state in real time. It includes establishing a database of drilling fluid pictures, and preparing drilling fluids with the same volume but different 10 rheological parameters. The pictures of the free fluid surface of the drilling fluid with different rheological parameters were collected respectively at the same set rotation speed. A convolutional neural network model is constructed, and the model is compiled and trained. Finally, import the real-time images of the drilling fluid under the flow state of the same rotational speed into the trained model, and the model outputs the rheological parameter values 15 of the drilling fluid. Fig. l
Description
A Real Time Measurement Method for Rheological Parameters of Drilling
Fluid Based on Machine Learning
The invention belongs to the technical field of drilling exploration, in particular to a method for real-time measurement of drilling fluid rheological parameters.
Background technique
In the process of energy exploration and development and deep geological drilling, drilling fluid is inseparable from maintaining wellbore stability. During the drilling process, a large amount of cuttings and debris will be produced. Through the circulation of drilling fluid, the cuttings broken by the drill bit are carried to the surface to keep the wellbore clean. Care is taken that the drill bit always contacts and breaks new formations at the bottom of the hole, so as to achieve safe and fast drilling.
The carrying and suspension of cuttings, wellbore stability and hydraulic jet are closely related to the rheology of drilling fluid. There are two main methods for measuring drilling fluid rheological parameters. One is funnel viscometer and the other is six-speed rotational viscometer. The viscosity value measured by the funnel viscosity agent is time, and the unit is s. And the six-speed rotational viscometer can obtain the shear rate and shear stress value, so as to calculate the plastic viscosity and apparent viscosity of the drilling fluid, and the unit is mPa-s.
During the drilling process, the drilling fluid density and rheological parameters will be monitored in real time in order to ensure drilling safety and wellbore stability. However, it is impossible to monitor the rheological properties of drilling fluid in real time during the actual operation on site. Therefore, in the field of geological engineering and petroleum engineering, drilling fluid rheological parameters are mainly measured and recorded by mud engineers every hour. Due to site conditions, the hourly measurement data is basically the funnel viscosity value. The plastic viscosity and apparent viscosity testing of drilling fluid basically only occurs in major drilling projects.
Drilling fluid testing is turning to automation with the continuous advancement of society and technology. Therefore, some instruments for automatic measurement of drilling fluid rheological parameters have also been introduced. For example, the online drilling fluid testing system developed by Petrobras can automatically monitor the rheology of drilling fluids.
Another example is that the intelligent detection of the Great Wall Drilling Engineering
Institute will gradually replace the manual detection of drilling fluid performance. The construction of the system platform combines the traditional drilling fluid testing technology in the oil and gas industry with the internet of things and big data platform. The intelligent analysis system provides a large amount of accurate reference data for the real-time return of field data and the formulation of plans in the laboratory, and provides technical support for the digital transformation of the drilling fluid industry.
At present, the prior art related to the automatic testing of drilling fluid viscosity includes a method and a device for obtaining drilling fluid viscosity and density during drilling (CN102140911A) and continuous automatic measurement of displacement, specific gravity and viscosity of drilling mud (CN101446198A). The problems and disadvantages of these traditional testing devices are mainly that the testing process takes a long time, cannot achieve real-time monitoring, and requires a lot of manpower and material resources. The main measure of the emerging technical solution is to replace manual testing with machines.
However, the disadvantage is that the testing method is based on mechanization and automation. The test is complex and still requires mechanized equipment, just replacing manual measurements. In addition, the cost of the product is high, although the test speed can be increased, the data monitoring per second cannot be achieved. And most of the measured data are funnel viscosity values.
The invention aims to provide a real-time measurement method of drilling fluid rheological parameters based on machine learning, which solves the problems of poor real-time performance, such as the need to wait for a long time for sampling and measurement, complex testing equipment and high cost in the existing drilling fluid measurement methods.
The core of the present invention is to apply image acquisition, image recognition and machine learning to real-time measurement of drilling fluid rheological parameters. Combined with the shape, in particular topography, characteristics of the free surface of the drilling fluid in a flowing state, such as stable rotating at a certain speed, as the distribution density characteristics of the free surface ripple, ripple width, wave crest, wave trough, wavelength,
etc.. By training a high-accuracy prediction model, the rheological parameters, such as apparent viscosity, plastic viscosity, dynamic shear force, static shear force, thixotropy, are compared with the real-time acquisition of the drilling fluid free surface map of the flowing state at the same rotational speed. Compare classification and output accurate drilling fluid rheological parameter results.
To achieve the above object, the present invention has adopted the following scheme:
A real-time measurement method of drilling fluid rheological parameters based on machine learning, which adopts the combination of image acquisition and image recognition to output real-time rheological parameters of drilling fluid in a flowing state.
Further, the combination of image acquisition and image recognition includes the following.
Step 1: Build a database of drilling fluid pictures. Prepare drilling fluids with the same volume but different rheological parameters, and collect pictures of the free fluid surface when the drilling fluids with different rheological parameters flow at the same set rotational speed.
And collect multiple pictures under the same rheological parameters, then measure the rheological parameters of drilling fluid. A data set of one-to-one correspondence between rheological parameters and drilling fluid pictures is established. For example, the apparent viscosity, plastic viscosity, dynamic shear force, static shear force and thixotropy of the drilling fluid corresponding to the picture are measured after a picture is collected. And it is a one-to- one corresponding data set.
Step 2: Data preprocessing. Load and format the images in the dataset obtained in step 1.
Resize the image, and decode the image into a tensor. Label each picture in the dataset in the form of a label, and establish a dataset in the form of for example a picture or a label. And split the dataset into a training set and a test set, where the label is the rheological parameter corresponding to the image. Here picture, label, refers to a one-to-one mapping relationship between a drilling fluid picture and a rheological parameter. The apparent viscosity, plastic viscosity, dynamic shear force, static shear force and thixotropy corresponding to a picture in step 1 are divided into five data sets: picture, apparent viscosity, picture, plastic viscosity, picture, dynamic shear, picture, static shear and picture, thixotropic.
Step 3: Build a model. The convolutional neural network model is adopted, and the output of the model is the rheological parameters of the drilling fluid.
Step 4: compiling the model and the training model, wherein compiling the model includes setting the optimizer, the loss function and the evaluation index. The training model includes evaluating the difference between the accuracy of the model on the training set and the test set, so as to adjust the model.
Step 5: Test the rheological parameters of the drilling fluid, import the real-time drilling fluid pictures obtained in the same flow state as the speed set in step 1 into the model that has been trained in step 4, and the model outputs the rheological parameter values of the drilling fluid.
Further, the rheological parameters include apparent viscosity, plastic viscosity, dynamic shear force, static shear force and thixotropy.
As an option, use a six-speed rotational viscometer to read the readings of the drilling fluid at rotational speeds of 9600, 300, 9200, 9100, 96 and 93 and calculate the drilling fluid rheological parameters when measuring the rheological parameters of the drilling fluid in step 5.
As another option, when collecting pictures of drilling fluid flow with different rheological parameters, pictures are collected from different angles above the free fluid surface of the drilling fluid for the drilling fluid with the same rheological parameter in step 1. The different angles here refer to different positions on the circumference with the normal line of the liquid surface center when the drilling fluid is stationary as the rotation axis, i.e. the circumference can be moved up and down along the rotation axis to change the height of the picture collection, or the diameter of the circumference can be enlarged or reduced to change the collection position. For example, taking one picture per increment along the circumference.
As an option, the picture is adjusted to a grayscale picture with a size of 28x28 pixels in step 2.
As an option, the convolutional neural network model includes at least two convolutional layers and one pooling layer in step 3.
Preferably, the size of the convolution kernel is 33 in the two convolution layers, and the number of the convolution kernel is 32 and 64 respectively.
It should be noted that the size of the convolution kernel in the convolutional neural network model, the settings of the convolutional layer and the pooling layer are adjusted according to the accuracy of the model prediction, and the above solution is only the most basic solution.
As an option, the evaluation index is the accuracy rate, and the accuracy rate is the percentage of correctly classified images in step 4.
Asan option, the drilling fluid to be measured is placed in the stirring vessel and continuously stirred at a stable rotational speed in step 5. And the free drilling fluid in the flowing state is acquired in real time through the image acquisition device arranged at the drilling fluid stirring vessel. The picture of the free fluid surface of the drilling fluid in the flowing state is acquired in real time through the image acquisition device set at the drilling fluid stirring vessel. . For example, the drilling fluid to be measured is stored in a stirring tank, and the drilling fluid is rotated and stirred at a stable rotation speed.
When collecting pictures of drilling fluid flow, the following rules should be followed, 5 which can minimize environmental factors and ensure the accuracy of machine learning image recognition. (1) In terms of drilling fluid, the volume of drilling fluid should be consistent. For example, the volume of drilling fluid is uniformly 350 mL when collecting pictures, and the volume of drilling fluid is the same every time a picture 1s collected. (2) In terms of the color of the experimental pictures, the pictures collected by the camera are all color images at present. However, all the collected pictures are processed in grayscale to ensure that the pictures become gray due to the limitation of the huge amount of computer memory and data. The degree map is then trained on big data. (3) In terms of drilling fluid configuration, the drilling fluid systems used in different formations and different oil and gas resources are not the same, so the treatment agents added are also different. For example, XC xanthan gum is white solid particles, and SPNH sulfonated lignite resin is brown solid particles. After adding to the drilling fluid system, the color of the drilling fluid is different. Therefore, grayscale processing of pictures for all drilling fluid systems can reduce the misjudgment of color for rheological parameter prediction. (4) In terms of drilling fluid flow rate, the key point of picture collection is that the drilling fluid flows at a fixed rotational speed. There is no limit to the specific rotation speed, as long as the drilling fluid can be stirred stably and continuously. For example, a drilling fluid stirring speed of 300 rpm is used. The reason why the drilling fluid in the stirring state is used as the flow state is that in the process of oil and gas field development and unconventional energy development, the actual drilling fluid circulation process will pass through a stirring tank, and the mixer rotates at a fixed speed. Therefore, image acquisition equipment can be placed next to the mixing tank to obtain real-time drilling fluid pictures and predict and output rheological parameters. Therefore, after replacing the existing drilling fluid sampling, it is tested by equipment such as a six-speed rotational viscometer, thereby realizing intelligence.
(5) The light in the image collection environment and the angle of the collected image, these two factors do not affect the final output result of the drilling fluid rheological parameters.
Because the collected image will be processed in grayscale, the light in the collection environment will not be affected. The impact is also in line with the actual production process.
Sunny, cloudy and rainy weather does not affect the final result. In terms of image acquisition angle, images from any angle need to be acquired to ensure that the amount of acquired image data is large enough to reduce unlearned and untrained situations in special circumstances. For example, under the same rheological parameters, pictures of the free surface of the drilling fluid are collected along different angles for each drilling fluid.
The invention is a drilling fluid rheological parameter testing method based on image acquisition, image recognition and deep machine learning. The drilling fluid plastic viscosity, apparent viscosity, dynamic shear, static shear and thixotropy based on the drilling fluid flow image are obtained by the image acquisition equipment through the deep learning algorithm.
The invention can obtain the rheological parameters of drilling fluid without using any mechanized equipment and sensors. And it can be measured in real time, and the measurement frequency can reach the second level. It can measure the plastic viscosity, apparent viscosity, dynamic shear force, static shear force and thixotropy of drilling fluid system per second, and save economic cost, time and human resources.
For the traditional drilling fluid rheological parameter testing method, a six-speed rotational viscometer needs to be used if one wants to obtain the complete rheological parameters of the drilling fluid system. Also, the static shear and thixotropy tests take about 20 minutes. However, drilling engineers need to master the properties of the drilling fluid to accurately understand the drilling conditions during the drilling process. The casualties and economic losses caused by major accidents such as blowouts in a single well are difficult to predict. Take the Gulf of
Mexico blowout incident as an example, causing 11 deaths and a compensation plan of 20 billion US dollars, and the economic losses cannot be estimated. Therefore, the present invention is more significant in terms of time cost, economic cost and safety improvement.
The drilling fluid rheological parameter data generated by the real-time measurement method of the present invention can provide fast and effective reference for drilling fluid parameter adjustment and safe drilling in the fields of geology, resources, energy and engineering.
Based on the above analysis, the present invention has the following characteristics compared with the existing drilling fluid testing method. (1) The rheological parameters of drilling fluid can be obtained without using any mechanical equipment and sensors. (2) It can be measured in real time, and the measurement frequency can reach the second level, which can measure the plastic viscosity, apparent viscosity, dynamic shear force, static shear force and thixotropy of the drilling fluid system per second. (3) The image-based deep learning algorithm of drilling fluid rheological properties is established, it can be infinitely copied to different application scenarios, such as engineering slurry, geotechnical slurry, trenchless slurry, etc., and has a wide range of applications.
Figure 1 is the schematic flow chart of the real-time measurement method of drilling fluid rheological parameters based on image recognition and machine learning in the invention.
Figure 2 is the experimental result when the model is verified to output the rheological parameter accuracy in real time during the model training process.
Figure 3 shows the working principle of the convolution kernel.
The implementation, functional characteristics and advantages of the present invention will be further described with reference to the following embodiments and accompanying drawings, but the protection scope of the present invention is not limited to the following.
It is the implementation process of the present invention as shown in Figure 1, and the concrete steps are as follows: 1. Build database. 1.1 Use different additives such as tackifiers, fluid loss reducers, viscosity reducers, weighting agents, surfactants, and biopolymers to prepare drilling fluid systems with the same volume but different rheological parameters in a beaker. 1.2 Stir the drilling fluid in the beaker at a speed of 300rpm, take pictures of the drilling fluid system when it flows, and then use a six-speed rotational viscometer to measure the drilling fluid system at ¢ 600, 9 300, 9 200, 9 100, 9 6 and ¢ 3, the apparent viscosity, plastic viscosity, dynamic shear force, static shear force and thixotropy of the drilling fluid system are calculated. 1.3 Establish 6000 sets of pictures of drilling fluid systems with different rheological parameters. Each picture corresponds to a set of rheological parameters (apparent viscosity, plastic viscosity, dynamic shear force, static shear force and thixotropy) for the establishment of the data set. 1.4 Use 80% of the dataset images for training and the remaining 20% for validation. 2. Data preprocessing. 2.1 Load and format pictures. 2.2 Decode the picture into a tensor. 2.3 Adjust the model size. 2.4 Slice the string array to get a string dataset. 2.5 Label the dataset with picture, label to obtain data such as picture, apparent viscosity, picture, plastic viscosity, picture, dynamic shear, picture, static shear and picture, touch degeneration dataset. 2.6 The data sets can be packaged to obtain the dimensions and types of the data sets since the data sets are in the same order. 3. Build the model and set up the layers and compile the model. 3.1 In the convolutional neural network, the image is first converted from a two- dimensional array (28 x 28 pixels) to a one-dimensional array (28 x 28 = 784 pixels).
Take this layer as an unstacked row of pixels in the image and line it up. The layer has no parameters to learn, it just reformats the data. 3.2 After flattening the pixels, the network consists of a sequence of two layers. The first layer has 128 nodes, or neurons, and the second layer returns an array of length 60. 3.3 Compile the model. There are a few more settings that need to be done before the model is ready to be trained. Firstly, add a loss function, which is used to measure the accuracy of the model during training while minimizing the loss function. Also, set up the optimizer, which decides how the model updates based on the data it sees and its own loss function. 3.4 The model consists of two convolutional blocks with a max-pooling layer in each convolutional block. The fully connected layer has one hundred and twenty eight units.
3.5 The convolution operation is implemented by layers Conv2D, and the pooling operation is implemented by the function layers MaxPooling2D of the code is as follows: layers. Input ((28,28,1)), 3 layers. Conv 2D(32,3,padding="same'), layers. Conv 2D(64,3,padding='same"), layers. MaxPooling 2D(), layers. Flatten (), layers.Dense (10), 3.6 Define two classes named “ds train” and “ds validation”. Data collection uses “import os” and “image dataset from directory” functions, data processing function “augment”, model definition and prediction function “optimizer”, prediction result function uses “predictions” and “loss” and defines accuracy. Then define ten convolutional layers, in which the activation function uses the rectified linear unit “layers.relu”, and the maximum pooling “layers. MaxPooling2D” is done twice. 3.7 Use the “Inception” function to convert RGB values from [0, 255] to [-1, 1]. The main code is as follows: ds train = ds train / 255.0 ds validation = ds validation / 255.0 4. Train the model. 4.1 The data is sufficiently scrambled. The data is then migrated to the model. The data is split into different batches. Repeat the training data set. The training variables include two parameters: weight and bias. Each calculation step is 100 steps, and all data is stored in memory to increase the calculation speed. 4.2 The model learns to associate images and labels, feeds the training data to the model, and verifies that the predictions match the labels in the array. 4 3 Assuming an image whose size is M*N, given a convolution kernel W whose size is m*n, the formula of convolution can be defined as:
VEN DW Kga (IS ESM -m +1, 1<j <N-n +1) u=l v=}
The shape of the convolution kernel has a high degree of coincidence with the original image, so as to extract image features. If an area of the image is very similar to the features that the convolution kernel can detect, then that area will activate the convolution kernel and get a high value. Conversely, if a region of the image is not similar to the features detected by the convolution kernel, the value of that region will be relatively low. The diversity of convolutions is increased by sliding stride and zero padding of the convolution kernel.
In the convolutional layer, the feature map is the feature output obtained after the image, or other feature maps, of the input layer is convolved. A convolution kernel is only responsible for extracting a certain type of features. In order to fully extract the information in the image, multiple convolution kernels are used. The general structure of a convolutional layer can be expressed as follows: (1) The input feature map group is a three-dimensional tensor (tensor), where each slice matrix is an input feature map.
X( Y ell MND )
The size of each feature map is M*N, and D is the number of input feature maps.
XW eM 1<d<D) (2) Output feature map group.
YY c 0 5D) where the output feature map group is also a 3D tensor, where each slice matrix is similar to the input matrix.
Ye ell VN 1<d <P) (3) Convolution kernel: Each slice matrix of the convolution kernel is a two- dimensional tensor. wrd eq 1< p< P1<d <D)
The convolution output calculation formula is:
Y= {WP QX +H)
D
Red 1 ]
VPO WX +b") d=1
Referring to Figure 3, it is a working principle diagram of the convolution kernel.
4.4 The subsampling layer, after the convolutional layer, acts as feature selection to reduce the number of features, thereby reducing the number of parameters in the network. The feature map is divided into multiple regions, and then the maximum neuron in the region is selected as the generalization of the region. 4.5 In order to improve the efficiency of model training, feature extraction is adopted.
Extract meaningful features from new samples using representations learned by previous networks. Adding a new classifier on top of a pre-trained model allows it to be trained from scratch, reusing previously learned feature maps for the dataset, eliminating the need to (re)train the entire model and improving model training efficiency. 5. Drilling fluid rheological parameter test. drilling fluid is stirred at a speed of 300 rpm in the on-site mixing tank, and the photos of the drilling fluid flow under the stirring state are collected in real time through the camera. The photos are processed and imported into the trained model, and the model outputs the values of drilling fluid rheological parameters, including plastic viscosity, apparent viscosity, dynamic shear force, static shear force and thixotropy. 5.1 Import the drilling fluid flow photos into the trained model after preprocessing.
And the model outputs the drilling fluid rheological parameter values, including plastic viscosity, apparent viscosity, dynamic shear force, static shear force and thixotropy. 5.2 Taking the plastic viscosity of the drilling fluid as an example, the plastic viscosity value of the drilling fluid system is output by identifying the pictures of the drilling fluid flowing. As shown in Figure 2, it is a drilling fluid plastic viscosity test structure based on image recognition and machine learning. Through the test of the six-speed rotational viscometer, the plastic viscosity of the drilling fluid in the six pictures is 8, 10, 13, 15, 21 and 25mPa-s in Figure 2. The histogram on the right side of the experimental graph is the prediction result obtained by the machine learning model.
The results show that the prediction accuracy of the machine learning model for the plastic viscosity of the drilling fluid in this scenario can reach 100%.
The above are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structural transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied in other related technical fields, are the same.
Included in the scope of patent protection of the present invention.
Claims (10)
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111295919.4A CN114004999A (en) | 2021-11-03 | 2021-11-03 | Drilling fluid rheological parameter real-time measurement method based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
NL2033450A true NL2033450A (en) | 2023-06-01 |
NL2033450B1 NL2033450B1 (en) | 2024-01-18 |
Family
ID=79927003
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
NL2033450A NL2033450B1 (en) | 2021-11-03 | 2022-11-03 | A Real Time Measurement Method for Rheological Parameters of Drilling Fluid Based on Machine Learning |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN114004999A (en) |
NL (1) | NL2033450B1 (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115021986A (en) * | 2022-05-24 | 2022-09-06 | 中国科学院计算技术研究所 | Construction method and device for Internet of things equipment identification deployable model |
CN115907236B (en) * | 2023-02-17 | 2023-11-03 | 西南石油大学 | Underground complex situation prediction method based on improved decision tree |
CN116212722B (en) * | 2023-05-08 | 2023-09-01 | 中国汽车技术研究中心有限公司 | Automobile collision dummy skin raw material treatment control method |
CN116297016B (en) * | 2023-05-23 | 2023-09-19 | 武汉誉城千里建工有限公司 | Full-automatic online detection device and detection method for drilling fluid performance |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446198A (en) | 2008-07-11 | 2009-06-03 | 杨晖 | Apparatus for automatic and continuous measurement of delivery capacity, specific gravity and viscosity of drilling mud |
CN102140911A (en) | 2010-10-13 | 2011-08-03 | 中国石油天然气股份有限公司 | Method and device for acquiring viscosity and density of drilling fluids in drilling process |
WO2020221644A1 (en) * | 2019-04-29 | 2020-11-05 | Bayer Aktiengesellschaft | Determination of the rheological behavior of a fluid |
-
2021
- 2021-11-03 CN CN202111295919.4A patent/CN114004999A/en active Pending
-
2022
- 2022-11-03 NL NL2033450A patent/NL2033450B1/en active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101446198A (en) | 2008-07-11 | 2009-06-03 | 杨晖 | Apparatus for automatic and continuous measurement of delivery capacity, specific gravity and viscosity of drilling mud |
CN102140911A (en) | 2010-10-13 | 2011-08-03 | 中国石油天然气股份有限公司 | Method and device for acquiring viscosity and density of drilling fluids in drilling process |
WO2020221644A1 (en) * | 2019-04-29 | 2020-11-05 | Bayer Aktiengesellschaft | Determination of the rheological behavior of a fluid |
Also Published As
Publication number | Publication date |
---|---|
CN114004999A (en) | 2022-02-01 |
NL2033450B1 (en) | 2024-01-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
NL2033450B1 (en) | A Real Time Measurement Method for Rheological Parameters of Drilling Fluid Based on Machine Learning | |
Zhang et al. | Application of deep learning algorithms in geotechnical engineering: a short critical review | |
CN107727592B (en) | Coal rock interface identification method based on high spectral reflectance characteristics of coal rock | |
CN109155064A (en) | Automate core description | |
CN103026202A (en) | Method for obtaining consistent and integrated physical properties of porous media | |
CN111596978A (en) | Web page display method, module and system for lithofacies classification by artificial intelligence | |
Bruna et al. | A new methodology to train fracture network simulation using multiple-point statistics | |
Dong et al. | Innovative method for pavement multiple damages segmentation and measurement by the Road-Seg-CapsNet of feature fusion | |
Lan et al. | Log facies identification in carbonate reservoirs using multiclass semi-supervised learning strategy | |
Rabbani et al. | Review of data science trends and issues in porous media research with a focus on image‐based techniques | |
Alzubaidi et al. | Automatic fracture detection and characterization from unwrapped drill-core images using mask R–CNN | |
Huo et al. | Novel lithology identification method for drilling cuttings under PDC bit condition | |
Yang et al. | High-fidelity permeability and porosity prediction using deep learning with the self-attention mechanism | |
CN114119361A (en) | Method, system, equipment and medium for reconstructing underground image based on super-resolution of TESRGAN network | |
Chandan et al. | An intelligent model for indian soil classification using various machine learning techniques | |
Chen et al. | A hierarchical DCNN-based approach for classifying imbalanced water inflow in rock tunnel faces | |
Li et al. | Design and optimization of the fiber-optic probe array for measuring gas holdup in oil-gas-water multiphase flow | |
Jin et al. | Image processing and machine learning based cavings characterization and classification | |
Liu et al. | Automatic fracture segmentation and detection from image logging using mask R-CNN | |
Mokhtari et al. | Feasibility study of EPB shield automation using deep learning | |
Li et al. | A real-time multi-defect automatic identification framework for concrete dams via improved YOLOv5 and knowledge distillation | |
Zheng et al. | Laboratory-on-a-smartphone for estimating angularity of granular soils | |
Zhao et al. | Pore-scale hydraulic properties of virtual sandstone microstructures: spatial variations and voxel scale effects | |
Zhou et al. | Knowledge-based U-Net and transfer learning for automatic boundary segmentation | |
Li et al. | The identification of slope crack based on convolutional neural network |