Method for predicting fatigue damage degree and residual life of coiled tubing

Abstract

The invention belongs to the technical field of coiled tubing life prediction, and particularly discloses a coiled tubing fatigue damage degree prediction method and a coiled tubing remaining life prediction method. The fatigue damage degree prediction method comprises the following steps: carrying out a bending fatigue test on a coiled tubing sample, and collecting a series of magnetic memory signals; preprocessing the series of magnetic memory signals and extracting a plurality of magnetic memory signal characteristic values from the magnetic memory signals; sequentially arranging the plurality of magnetic memory signal characteristic values according to the acquisition time to serve as an original data column, and establishing a grey prediction model; based on the grey prediction model and in combination with the Miner fatigue linear accumulation theory, a coiled tubing fatigue damage degree prediction model representing the relation between the magnetic memory signal characteristic value and the coiled tubing fatigue damage degree is established; and based on the established model, predicting the fatigue damage degree of the oil pipe to be detected according to the detected characteristic value of the magnetic memory signal. The fatigue damage degree prediction method can conveniently and efficiently predict the fatigue damage degree of the oil pipe to be detected.

Description

Technical Field

[0001] The present invention belongs to the technical field of coiled tubing life prediction, and particularly relates to a method for predicting the fatigue damage degree of coiled tubing and a method for predicting the remaining life of coiled tubing. Background Art

[0002] When coiled tubing is in use, three bending deformations are involved during a single run-in or run-out process (under the driving force of the injector head, the tubing drives the drum to rotate, and the tubing continuously leaves the drum, changing from bent to straight; the tubing enters the guide arc and advances along the circular arc trajectory in the guide arc, changing the direction of motion, from straight to bent; the tubing leaves the guide arc and enters the injector head, and under the clamping action of the injector head chain, it leaves the injector head vertically and enters the oil well, changing from bent to straight). Each bending will affect the plastic tension of the pipe and cause fatigue, resulting in changes in the wall thickness and ovality of the coiled tubing. After long-term use, defects such as cracks may even appear on the pipe body. Coiled tubing is used for operations such as drilling, fishing, and sand washing inside the wellbore, resulting in the coiled tubing being in different stress states and fatigue states. During operation, under the combined action of the guide arc, injector head, buoyancy of wellbore fluid, torque, and self-weight, the coiled tubing simultaneously bears tensile loads and bending loads, including multiple non-linear factors such as contact, friction, large displacement, bending, and large deformation, and is prone to fatigue damage. The appearance of defects will significantly shorten the service life of the coiled tubing. In severe cases, it may even cause downhole operation accidents, increasing the operation cost of the oilfield and reducing work efficiency. If detection means can be used to timely detect tubing defects and take measures such as cutting or discarding, the service life of the tubing can be effectively extended and the occurrence of accidents can be significantly reduced.

[0003] For traditional coiled tubing fatigue life analysis, it is necessary to evaluate the fatigue life of coiled tubing in a complex working environment, simulating hundreds of load conditions, which will consume a large amount of time throughout the process. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for predicting the fatigue damage degree of coiled tubing and a method for predicting the remaining life of coiled tubing, which can save labor costs and conveniently and efficiently determine whether there are defects in the coiled tubing.

[0005] According to the first aspect of the present invention, a method for predicting the fatigue damage degree of coiled tubing is provided, which includes the following steps:

[0006] Conduct a bending fatigue test on a coiled tubing specimen, and collect a series of magnetic memory signals of the coiled tubing specimen during the bending fatigue test;

[0007] Preprocess the collected series of magnetic memory signals and extract multiple magnetic memory signal characteristic values from them;

[0008] Arrange the multiple extracted magnetic memory signal eigenvalue in the order of acquisition time to form an original data column, establish a grey prediction model, and obtain the relationship between the magnetic memory signal eigenvalue and the serial number of the original data column;

[0009] Based on the relationship between the magnetic memory signal eigenvalue and the serial number of the original data column, and in combination with Miner's fatigue linear cumulative theory, establish a prediction model for the fatigue damage degree of coiled tubing that represents the relationship between the magnetic memory signal eigenvalue and the fatigue damage degree of coiled tubing;

[0010] Based on the established prediction model for the fatigue damage degree of coiled tubing, predict the fatigue damage degree of the coiled tubing to be measured according to the detected magnetic memory signal eigenvalue of the coiled tubing to be measured.

[0011] According to some embodiments of the present invention, preprocess a series of collected magnetic memory signals, and extract multiple magnetic memory signal eigenvalues therefrom, including:

[0012] Perform noise reduction processing on a series of collected magnetic memory signal values;

[0013] Perform gradient processing on a series of magnetic memory signal values that have undergone noise reduction processing;

[0014] Extract multiple magnetic memory signal eigenvalues from a series of magnetic memory signal values that have undergone gradient processing.

[0015] According to some embodiments of the present invention, perform noise reduction processing by taking the average value of the magnetic memory signal values collected at each sampling point within a noise reduction window. The noise reduction processing uses the following formula:

[0016]

[0017] Wherein, is the magnetic memory signal value at the i-th sampling point after noise reduction processing; j is the noise reduction window, that is, the number of data within the detection area; H n is the initial magnetic memory signal value at the n-th sampling point.

[0018] According to some embodiments of the present invention, perform gradient processing by calculating the unit distance change rate of the magnetic memory signal value at each sampling point. The gradient processing uses the following formula:

[0019]

[0020] Wherein, K i is the magnetic field gradient value at the i-th sampling point, ΔH is the change amount of the magnetic memory signal value, ΔL is the change amount of the distance, H i 、H i+kThey are respectively the magnetic memory signal values at the i-th sampling point and the (i + k)-th sampling point, where k is the distance between the i-th sampling point and the (i + k)-th sampling point.

[0021] According to some embodiments of the present invention, the magnetic memory signal eigenvalue is the peak value of the gradient or the peak-to-peak value of the gradient. The peak value of the gradient is the maximum value of the absolute value of the magnetic field gradient value within the detection region, and the peak-to-peak value of the gradient is the absolute value of the difference between the positive peak value of the magnetic field gradient and the negative peak value of the magnetic field gradient within the detection region.

[0022] According to some embodiments of the present invention, the grey prediction model is the grey prediction GM(1,1) model. Arrange the multiple extracted magnetic memory signal eigenvalues in the order of acquisition time as the original data column, establish a grey prediction model, and obtain the relationship between the magnetic memory signal eigenvalue and the serial number of the original data column, including:

[0023] Perform a first-order accumulation process on the original data column X (0) to obtain the accumulated data column X (1) :

[0024] X (0) ={X (0) (1), X (0) (2), X (0) (3), …, X (0) (n)}

[0025] X (1) ={X (1) (1), X (1) (2), X (1) (3), …, X (1) (n)},

[0026]

[0027] Based on the original data column X (0) and the accumulated data column X (1) , use the least squares method to calculate the grey parameters in the grey prediction GM(1,1) model. The grey parameters include the development grey number a and the endogenous control grey number u:

[0028]

[0029]

[0030] Y N =(X (0) (2), X (0) (3), X (0) (4), …, X (0) (n)) T ;

[0031] Based on the calculated grey parameters, for the cumulative data series X (1) make a prediction to obtain a predicted cumulative data series whose terms are represented by the following formula

[0032]

[0033] For the predicted cumulative data series perform an inverse process to restore it to a predicted original data series whose terms are represented by the following formula Thus, a grey prediction GM(1,1) model is established to obtain the relationship between the magnetic memory signal characteristic value and the serial number:

[0034]

[0035] Perform a posteriori difference test on the established grey prediction GM(1,1) model. If the accuracy requirement is met, the established grey prediction GM(1,1) model is available; otherwise, the established grey prediction GM(1,1) model is not available.

[0036] According to some embodiments of the present invention, use the posteriori difference ratio C and the small error frequency P to perform a posteriori difference test on the established grey prediction GM(1,1) model. When the posteriori difference ratio C is less than the first threshold and the small error frequency P is greater than the second threshold, the established grey prediction GM(1,1) model meets the accuracy requirement.

[0037] According to some embodiments of the present invention, the posteriori difference ratio C is calculated by the following formula:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] where S1 is the standard deviation of the original data series, S2 is the standard deviation of the residuals, is the average value of the original data series, q(k) is the k-th residual, is the mean sum of squares of the residuals;

[0045] According to some embodiments of the present invention, the small error frequency P is calculated by the following formula:

[0046]

[0047] According to some embodiments of the present invention, based on the relationship between the magnetic memory signal eigenvalue and the serial number of the original data column, and in combination with the Miner fatigue linear cumulative theory, a prediction model for the fatigue damage degree of coiled tubing representing the relationship between the magnetic memory signal eigenvalue and the fatigue damage degree of coiled tubing is established, including:

[0048] According to the Miner fatigue linear cumulative theory, obtain the relationship between the fatigue damage degree of coiled tubing and the serial number;

[0049] According to the relationship between the fatigue damage degree of coiled tubing and the serial number and the relationship between the magnetic memory signal eigenvalue and the serial number, establish a prediction model for the fatigue damage degree of coiled tubing representing the relationship between the magnetic memory signal eigenvalue and the fatigue damage degree of coiled tubing.

[0050] According to some embodiments of the present invention, a bending fatigue test of coiled tubing specimens is carried out under different internal pressures, different bending deflections and different bending frequencies to establish a prediction model for the fatigue damage degree of coiled tubing under different internal pressures, different bending deflections and different bending frequencies.

[0051] According to a second aspect of the present invention, a method for predicting the remaining life of coiled tubing is provided, which includes:

[0052] Use the above prediction method for the fatigue damage degree of coiled tubing to predict the fatigue damage degree of the coiled tubing to be tested;

[0053] Input the predicted fatigue damage degree of the coiled tubing to be tested and the working condition data of the coiled tubing to be tested into the trained neural network prediction model for the remaining life of coiled tubing to obtain the remaining life of the coiled tubing to be tested.

[0054] According to some embodiments of the present invention, the neural network prediction model for the remaining life of coiled tubing is a BP neural network prediction model, which includes three parts: an input layer, a hidden layer and an output layer. The input layer is the working condition data and fatigue damage degree of the tubing, and the output layer is the remaining life.

[0055] According to some embodiments of the present invention, the working condition data of the tubing includes: the length of the coiled tubing, the depth of the wellbore fluid, the outer diameter of the coiled tubing, the mass per unit length, the tensile stiffness, the bending stiffness, the torsional stiffness, the internal pressure of the coiled tubing, and the drilling pressure.

[0056] According to some embodiments of the present invention, the BP neural network prediction model has 8 hidden layers, and the number of its neurons are: 10, 20, 40, 60, 40, 20, 10, 10.

[0057] According to some embodiments of the present invention, the BP neural network prediction model is constructed in the following manner:

[0058] Obtain multiple groups of sample data, where each group of sample data includes coiled tubing working condition data, fatigue damage degree data, and actual remaining life data;

[0059] Perform normalization processing on the multiple groups of sample data;

[0060] Divide the normalized sample data into a test data set, a training data set, and a validation data set;

[0061] Construct a first neural network model, train the first neural network model using the training data set, and perform a preliminary verification of the training results of the first neural network model using the validation data set to obtain a second neural network model;

[0062] Input the test data set into the second neural network model and perform a prediction evaluation on the second neural network model;

[0063] Based on the prediction evaluation results, determine a neural network prediction model for the remaining life of coiled tubing.

[0064] According to some embodiments of the present invention, performing normalization processing on multiple groups of sample data includes: performing normalization processing on each item of data in each group of sample data using the following formula:

[0065] y k =(x k -x min ) / (x max -x min )

[0066] where: x min is the minimum value of this item of data in multiple groups of sample data; x max is the maximum value of this item of data in multiple groups of sample data; x k is the value of this item of data in any group of sample data; y k is the value of this item of data after normalization processing.

[0067] According to some embodiments of the present invention, training the first neural network model using the training data set includes:

[0068] Input the coiled tubing working condition data and fatigue damage degree data in the training data set into the first neural network model to obtain a predicted remaining life value;

[0069] Based on the predicted remaining life value and the corresponding actual remaining life value, calculate the relative error value E, and the formula is as follows:

[0070]

[0071] where L is the number of samples in the training data set; y kis the predicted remaining life value of the k-th sample; o k is the actual remaining life value of the k-th sample;

[0072] Judge whether the relative error value E meets the threshold requirement. If the relative error value E meets the threshold requirement, the training is completed; otherwise, continue the training.

[0073] According to some embodiments of the present invention, inputting the test data set into the second neural network model and performing prediction evaluation on the second neural network model includes:

[0074] Input the coiled tubing working condition data and fatigue damage degree data in the test data set into the second neural network model, repeat the test multiple times, and obtain multiple remaining life test output values;

[0075] Calculate the average correlation coefficient between the remaining life test output value and the corresponding actual remaining life value. The formula is as follows:

[0076]

[0077] where Cov(X, Y) is the covariance of X and Y; Var[X] is the variance of X; Var[Y] is the variance of Y; X is the remaining life test output value; Y is the actual remaining life value;

[0078] Based on the calculated average correlation coefficient, perform prediction evaluation on the second neural network model. If the calculated average correlation coefficient meets the requirements, the prediction evaluation result is passed; if the calculated average correlation coefficient does not meet the requirements, the prediction evaluation result is not passed.

[0079] According to some embodiments of the present invention, based on the prediction evaluation result, determine the coiled tubing remaining life neural network prediction model, including: if the prediction evaluation result is passed, determine the second neural network model as the coiled tubing remaining life neural network prediction model; if the prediction evaluation result is not passed, retrain.

[0080] Due to the above technical solutions, the present invention has the following beneficial effects:

[0081] In the present invention, based on the magnetic memory detection method, taking the magnetic memory signal characteristic parameters as the original data and the coiled tubing fatigue damage degree as the prediction object, a coiled tubing fatigue damage degree prediction model is established. Furthermore, the fatigue damage degree of the coiled tubing to be tested can be predicted by detecting the magnetic memory signal characteristic values of the coiled tubing to be tested.

[0082] In the present invention, dynamic analysis is carried out on coiled tubing under specific working conditions. By training a neural network model, a mapping relationship between the stress state and the remaining life of the coiled tubing is established. Based on the trained neural network, real-time prediction of the remaining life of the coiled tubing can be achieved, so as to achieve the purpose of quickly analyzing the remaining life of the coiled tubing.

[0083] The method provided by the present invention can use magnetic memory detection means to timely detect defects in the tubing to be tested, facilitate timely taking remedial measures, effectively extend the service life of the tubing, greatly reduce the occurrence of accidents, improve work efficiency, and reduce the operation cost of the oilfield. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0085] Figure 1 It is a flowchart of the coiled tubing fatigue damage degree prediction method provided by the present invention;

[0086] Figure 2 It is a flowchart for establishing a grey prediction GM(1,1) model;

[0087] Figure 3 It is a flowchart of the coiled tubing remaining life prediction method provided by the present invention;

[0088] Figure 4 It is a BP neural network structure diagram of the remaining life prediction model;

[0089] Figure 5 It is a flowchart of the construction method of the BP neural network prediction model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0090] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, and are not used to limit the present invention.

[0091] Magnetic memory testing is a new non-destructive testing technology that has developed rapidly internationally since the late 1990s. It mainly determines the microscopic defects or stress concentration positions and characteristics of a workpiece, and makes a clear judgment on the early damage of the workpiece, based on the irreversible characteristic magnetic state changes that occur in the stress and deformation concentration regions of ferromagnetic materials under the excitation of the geomagnetic field. This technology does not require special magnetization equipment, is less affected by the lift-off effect, has low requirements for working conditions, is quick, simple, and convenient to operate, and has high sensitivity, repeatability, and reliability. It is more suitable for on-site use and has currently been used for the damage and various defect detections of load-bearing structural components such as aircraft landing gears, aircraft main beam screw holes, turbine disks, compressor blades, turbine blades, steam turbine blades, and pressure vessels.

[0092] In the present invention, the magnetic memory testing method is applied to the prediction of the remaining life of coiled tubing. Based on the magnetic memory testing method, using the characteristic parameters of the magnetic memory signals of coiled tubing specimens as the original data and the fatigue damage degree of coiled tubing as the prediction object, a prediction model for the fatigue damage degree of coiled tubing is established to predict the fatigue damage degree of the coiled tubing to be tested. By using the magnetic memory testing method, early fatigue diagnosis of coiled tubing can be carried out and the distribution state of residual stress can be determined. By utilizing the magnetic memory effect that is sensitive to stress and deformation concentration regions, effective detection of both early damage defects (without significant size changes) and macroscopic defects (significant size changes) of coiled tubing is achieved simultaneously.

[0093] According to the first aspect of the present invention, there is provided a method for predicting the fatigue damage degree of coiled tubing, as Figure 1 shown. This method includes the following steps: S1: Conduct a bending fatigue test on a coiled tubing specimen and collect a series of magnetic memory signals of the coiled tubing specimen during the bending fatigue test; S2: Preprocess the collected series of magnetic memory signals and extract multiple magnetic memory signal characteristic values therefrom; S3: Arrange the extracted multiple magnetic memory signal characteristic values in the order of collection time as the original data column, establish a grey prediction GM(1,1) model, and obtain the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column; S4: Based on the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column, and in combination with Miner's fatigue linear cumulative theory, establish a prediction model for the fatigue damage degree of coiled tubing that represents the relationship between the magnetic memory signal characteristic values and the fatigue damage degree of coiled tubing; S5: Based on the established prediction model for the fatigue damage degree of coiled tubing, predict the fatigue damage degree of coiled tubing according to the detected magnetic memory signal characteristic values of the coiled tubing to be tested.

[0094] The following is a specific description of each step.

[0095] S1: Conduct a bending fatigue test on the coiled tubing specimen, and collect a series of magnetic memory signals of the coiled tubing specimen during the bending fatigue test. Conducting a bending fatigue test on the coiled tubing specimen means periodically applying bending stress to the coiled tubing specimen until it breaks, that is, conducting multiple bending cycles on the coiled tubing specimen until it breaks. This process can be achieved by a bending fatigue testing machine. Collecting a series of magnetic memory signals of the coiled tubing specimen during the bending fatigue test means collecting magnetic memory signals every certain number of bending cycles during multiple bending cycles, thereby obtaining multiple magnetic memory signals as the multiple bending cycles proceed. This process can be achieved by a magnetic memory sensor. In a specific example, 50 - 200 bending cycles are conducted on the coiled tubing specimen, and during this period, the coiled tubing test is detected once every 10 - 20 bending cycles to obtain a magnetic memory signal.

[0096] S2: Preprocess the collected series of magnetic memory signals, and extract multiple magnetic memory signal characteristic values therefrom. The preprocessing includes noise reduction processing (optional) and gradient processing. In some embodiments, first perform noise reduction processing on the collected series of magnetic memory signal values; then, perform gradient processing on the series of magnetic memory signal values that have undergone noise reduction processing; finally, extract multiple magnetic memory signal characteristic values from the series of magnetic memory signal values that have undergone gradient processing.

[0097] In some embodiments, the noise reduction processing can be performed by taking the average of the magnetic memory signal values collected at each sampling point within a noise reduction window. The noise reduction processing uses the following formula:

[0098]

[0099] where, is the magnetic memory signal value at the i-th sampling point after noise reduction processing; j is the noise reduction window, that is, the number of data in the selected area; H n is the initial magnetic memory signal value at the n-th sampling point.

[0100] In some embodiments, the gradient processing can be performed by calculating the change rate per unit distance of the magnetic memory signal values at each sampling point. The gradient processing uses the following formula:

[0101]

[0102] where, K i is the magnetic field gradient value at the i-th sampling point, ΔH is the change amount of the magnetic memory signal value, ΔL is the change amount of the distance, H i 、H i+k are the magnetic memory signal values at the i-th sampling point and the i + k-th sampling point respectively, and k is the distance between the i-th sampling point and the i + k-th sampling point.

[0103] The eigenvalue of the magnetic memory signal is the peak gradient value K max or the peak-to-peak gradient value K sub . The peak gradient value K max is the maximum value of the absolute value of the magnetic field gradient value within the detection area. The peak-to-peak gradient value K sub is the absolute value of the difference between the positive peak value of the magnetic field gradient and the negative peak value of the magnetic field gradient within the detection area. Specifically:

[0104] K max = max(|K i |)

[0105] K sub = |max(K i ) - min(K i )|

[0106] where K i is the magnetic field gradient value at the i-th sampling point.

[0107] S3: Arrange the multiple extracted magnetic memory signal eigenvalues in the order of acquisition time as the original data column, establish a grey prediction model, and obtain the relationship between the magnetic memory signal eigenvalues and the serial numbers of the original data column.

[0108] In some embodiments, the grey prediction model is specifically the grey prediction GM(1,1) model. As Figure 2 shown, establishing the grey prediction GM(1,1) model includes the following steps:

[0109] S31: Determine the original data column X (0) . Arrange the multiple extracted magnetic memory signal eigenvalues in the order of acquisition time to form the original data column X (0) .

[0110] X (0) = {X (0) (1), X (0) (2), X (0) (3), …, X (0) (n)}

[0111] S32: Judge whether the original data column X (0) meets the model characteristics. If it meets the model characteristics, the method proceeds to step S33; otherwise, the method proceeds to step S310. The model characteristics refer to that the original data column generally shows an exponential change law. For example, the peak-to-peak gradient value shows an approximately exponential growth during the whole life of the oil pipe.

[0112] S33: For the original data column X (0)Perform an accumulation process to obtain an accumulated data series. During the grey prediction process, it is necessary to transform the original chaotic data into a new series with stronger regularity through mathematical processing methods. The original data series X (0) can be subjected to an accumulation generation to produce a new series X (1) .

[0113] X (1) ={X (1) (1), X (1) (2), X (1) (3), …, X (1) (n)}

[0114]

[0115] S34: Based on the original data series X (0) and the accumulated data series X (1) , use the least squares method to calculate the grey parameters in the grey prediction GM(1,1) model. The grey parameters include the development grey number a and the endogenous control grey number u.

[0116]

[0117]

[0118] Y N =(X (0) (2), X (0) (3), X (0) (4), …, X (0) (n)) T

[0119] The grey prediction GM(1,1) model is:[[]]

[0120]

[0121] Its corresponding differential equation is:[[]]

[0122]

[0123] S35: Based on the calculated grey parameters, predict the accumulated data series X (1) to obtain the predicted accumulated data series The items of the predicted accumulated data series are expressed as follows:[[]]

[0124]

[0125] S36: For the predicted accumulated data series Perform reverse processing. Reverse processing is subtraction processing. Subtraction processing is the inverse operation of addition processing. Subtract two adjacent data in the data column. This method can restore the cumulative generation column to the original data column, and is mostly used in the late stage of grey prediction to restore the prediction column obtained from the cumulative data column with strong regularity to the prediction column of the original data.

[0126] S37: Obtain the predicted original data column Predicted original data column The items are expressed as follows:

[0127]

[0128] S38: Conduct posterior difference test on the established grey prediction GM(1,1) model. The posterior difference ratio C and the small error frequency P can be used to conduct posterior difference test on the established grey prediction GM(1,1) model. The posterior difference ratio C mainly examines the discrete gap between the predicted data and the original data, that is, the ratio of the standard deviation of the residuals to the standard deviation of the original data column. The small error frequency P mainly examines the discrete degree of the residuals, that is, the frequency that the absolute value of the difference between each data residual and the average residual is less than 0.6745S1. The expressions of the two parameters are:

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135]

[0136] Among them, S1 is the standard deviation of the original data column, S2 is the standard deviation of the residuals, is the average value of the original data column, q(k) is the k-th residual, is the average sum and variance of the residuals.

[0137] S39: Determine whether the established grey prediction GM(1,1) model meets the accuracy requirements. When the posterior difference ratio C is less than the first threshold and the small error frequency P is greater than the second threshold, the established grey prediction GM(1,1) model meets the accuracy requirements; otherwise, it does not. In some cases, the prediction accuracy levels can be divided, and different thresholds can be selected according to different prediction accuracy requirements. The specific evaluation of the prediction accuracy levels is shown in Table 1. When the prediction accuracy requirement is high, the first threshold can be 0.4 and the second threshold can be 0.9; when the prediction accuracy requirement is not very high, the first threshold can be 0.5 and the second threshold can be 0.8; when the prediction accuracy requirement is low, the first threshold can be 0.6 and the second threshold can be 0.7.

[0138] Table 1 Evaluation Table of Accuracy Levels

[0139] Accuracy level Small error frequency P Posterior difference ratio C Good >0.9 <0.4 Qualified >0.8 <0.5 Marginal >0.7 <0.6 Unqualified >0.6 <0.7

[0140] When the established grey prediction GM(1,1) model meets the accuracy requirements, the grey prediction GM(1,1) model is available, and subsequent steps are continued using this model. When the established grey prediction GM(1,1) model does not meet the accuracy requirements, the grey prediction GM(1,1) model is not available, and at this time, step S310 is continued.

[0141] S310: Perform data processing on the original data series. For example, gradient processing.

[0142] S4: Based on the relationship between the magnetic memory signal eigenvalue and the serial number of the original data series, and combined with Miner's fatigue linear cumulative theory, establish a continuous tubing fatigue life prediction model representing the relationship between the magnetic memory signal eigenvalue and the continuous tubing fatigue damage degree. Specifically, according to Miner's fatigue linear cumulative theory, obtain the relationship between the continuous tubing fatigue damage degree and the serial number; according to the relationship between the continuous tubing fatigue damage degree and the serial number and the relationship between the magnetic memory signal eigenvalue and the serial number, establish a continuous tubing fatigue damage degree prediction model representing the relationship between the magnetic memory signal eigenvalue and the continuous tubing fatigue damage degree.

[0143] According to Miner's fatigue linear cumulative theory, damage accumulates gradually, and the fatigue failure criterion can be expressed as:

[0144]

[0145] where N k is the k-th fatigue loss, N Fk is the remaining life at the current load level, and k is the number of cycles.

[0146] Therefore, the fatigue damage degree D can be expressed as the ratio of the number of cycles to the total number of life cycles. Thus, the relationship between the fatigue damage degree at a certain moment and the serial number k can be obtained by combining the total number of life cycles obtained through testing with the number of cycles at a certain moment.

[0147] S5: Based on the established prediction model for the fatigue damage degree of coiled tubing, predict the fatigue damage degree of coiled tubing according to the detected characteristic values of the magnetic memory signals of the coiled tubing to be measured. After establishing the prediction model for the fatigue damage degree of coiled tubing by using coiled tubing specimens, this prediction model can be used to predict the fatigue damage degree of the coiled tubing to be measured.

[0148] In some cases, bending fatigue tests can be carried out on coiled tubing specimens under different internal pressures, different bending deflections and different bending frequencies, so as to establish a prediction model for the fatigue damage degree of coiled tubing under different internal pressures, different bending deflections and different bending frequencies. This helps to more accurately predict the fatigue damage degree based on specific working conditions, and further helps to more accurately predict the remaining life.

[0149] After obtaining the fatigue damage degree by using the above method, the remaining life of the coiled tubing can be further predicted based on the obtained fatigue damage degree. However, for the traditional fatigue life analysis of coiled tubing, it is necessary to evaluate the fatigue life of coiled tubing in a complex working environment of coiled tubing and simulate hundreds of load conditions, and the whole process will consume a lot of time. In view of this, one of the purposes of the present invention is to provide a method for predicting the remaining life of coiled tubing that can save labor costs and conveniently and efficiently judge whether there are defects in the coiled tubing. For this reason, in the present invention, based on the prediction of representative working conditions of coiled tubing by neural network, the analysis working conditions are reduced; the dynamic analysis of coiled tubing under specific working conditions is carried out, and by training the neural network model, a mapping relationship between the stress state and the remaining life of coiled tubing is established; based on the trained neural network, the real-time prediction of the remaining life of coiled tubing is realized, so as to achieve the purpose of quickly analyzing the fatigue life of coiled tubing.

[0150] According to the second aspect of the present invention, there is provided a method for predicting the remaining life of coiled tubing. As Figure 3 shown, it includes the following steps: S10: Predict the fatigue damage degree of the coiled tubing to be measured by using the above-mentioned method for predicting the fatigue damage degree of coiled tubing; S20 Input the predicted fatigue damage degree of the coiled tubing to be measured and the working condition data of the coiled tubing to be measured into the trained neural network prediction model for the remaining life of coiled tubing, and obtain the remaining life of the coiled tubing to be measured.

[0151] In some embodiments, the coiled tubing remaining life neural network prediction model is a BP neural network prediction model, which includes three parts: an input layer, a hidden layer, and an output layer. The input layer is the coiled tubing working condition data and the fatigue damage degree, and the output layer is the remaining life. The coiled tubing working condition data includes: coiled tubing length (m), wellbore fluid depth (m), coiled tubing outer diameter (mm), mass per unit length (kg / m), tensile stiffness (kN), bending stiffness (kN), torsional stiffness (kN), coiled tubing internal pressure (MPa), and drilling pressure (MPa).

[0152] The following is a brief introduction to the BP neural network prediction model.

[0153] The BP neural network mainly includes three parts: an input layer, a hidden layer, and an output layer, and the neurons in each layer affect each other layer by layer. Its basic principle is that under the stimulation of the external input sample X, the neural network adjusts the connection weights Wij in the hidden layer and the output layer to keep the network output close to the expected response value Y. The output of the hidden layer can be expressed as follows:

[0154] H = f(W H ·X + b H )

[0155] Among them, b H represents the bias coefficient of the hidden layer, f is the activation function, W H represents the connection weight of the hidden layer, H represents the output of the hidden layer, and the activation function is as follows:

[0156]

[0157] In the final output layer, multiplying the output of the hidden layer by the weight and combining the bias coefficient bo, the output of the output layer can be obtained as:

[0158] O = W o ·H + b o

[0159] Among them, b O represents the bias coefficient of the output layer, W O is the connection weight of the output layer, H represents the output of the hidden layer, and O represents the output of the output layer.

[0160] Comparing the output O of the neural network with the expected response Y to obtain the error, and updating the weights and bias coefficients of each layer of the neural network according to the error until the output value and the response error are small, and the training process is completed.

[0161] In the model verification process, the methods of mean error, error upper limit, and correlation coefficient can be used to evaluate the quality of the model. The mean error formula is expressed as follows:

[0162]

[0163] Among them: n represents the number of test sets, r i pred represents the predicted value of the response, r i act represents the actual value of the response, E abs represents the average error.

[0164] The expression for the correlation coefficient is:

[0165]

[0166] Among them, n represents the number of test sets, r i pred represents the predicted value of the response, r i act represents the actual value of the response, represents the mean of the predicted values, represents the mean of the actual values, R p represents the relative coefficient.

[0167] Figure 4 is a schematic diagram of the structure of the BP neural network prediction model. In this example, the BP neural network prediction model has 8 hidden layers, and the number of its neurons is: 10, 20, 40, 60, 40, 20, 10, 10.

[0168] Figure 5 is a flowchart of the construction method of the BP neural network prediction model. As shown in the figure, the coiled tubing remaining life neural network prediction model can be constructed in the following way:

[0169] S201: Obtain multiple sets of sample data. Each set of sample data includes coiled tubing working condition data, fatigue damage degree data, and remaining life actual data. Among them, the coiled tubing working condition data includes coiled tubing length, wellbore fluid depth, coiled tubing outer diameter, mass per unit length, tensile stiffness, bending stiffness, torsional stiffness, coiled tubing internal pressure, and drilling pressure.

[0170] S202: Perform normalization processing on multiple sets of sample data. Specifically, the following formula can be used to perform normalization processing on each item of data in each set of sample data:

[0171] y k =(x k -x min ) / (x max -x min )

[0172] Among them, x min is the minimum value of this item of data in multiple sets of sample data; x maxis the maximum value of this item of data in multiple groups of sample data; x k is the value of this item of data in any group of sample data; y k is the value of this item of data after normalization processing.

[0173] S203: Divide the normalized sample data into a test data set, a training data set, and a validation data set.

[0174] S204: Construct a first neural network model, train the first neural network model using the training data set, and perform a preliminary verification of the training results of the first neural network model using the validation data set to obtain a second neural network model. In this step, the first neural network model includes an input layer, a hidden layer, and an output layer. The input layer is used to incorporate the data sets of the pretreated coiled tubing length, wellbore fluid depth, coiled tubing outer diameter, mass per unit length, tensile stiffness, bending stiffness, torsional stiffness, coiled tubing internal pressure, and drilling pressure. The hidden layer is used to extract and process the parameter features of the data sets imported by the input layer. The output layer is used to incorporate the data set of the pretreated remaining life of the coiled tubing.

[0175] Train the first neural network model using the training data set. The specific steps are as follows: Use the training data set as the variable of the input layer and the training data set of the remaining life of the coiled tubing as the variable of the output layer, and perform repeated training to obtain the relative error value E; if the relative error value E meets the threshold requirement, the training is completed, otherwise continue training. In an example, a relative error value E less than 0.01 indicates a high degree of fitting between the predicted value and the actual measured value after passing through the first BP neural network model, and the training can be ended. The calculation formula for the relative error value E is as follows:

[0176]

[0177] In the formula: L is the number of samples in the training data set; y k is the predicted value of the remaining life of the kth sample; o k is the actual value of the remaining life of the kth sample.

[0178] The training result of the first neural network model is initially verified using a validation dataset to obtain a second neural network model. The specific steps are as follows: Input the continuous tubing working condition data and fatigue damage degree data in the validation dataset into the first neural network model to obtain the predicted remaining life values; Calculate the error value based on the predicted remaining life values and the corresponding actual remaining life values; Determine whether the error value meets the threshold requirement. If it does not meet the threshold requirement, it indicates that the verification of the first neural network model fails and retraining is required. If it meets the threshold requirement, it indicates that the verification of the first neural network model passes, and it is used as the second neural network model. Meeting the threshold requirement can be, for example, less than 0.01 or other predefined thresholds.

[0179] S205: Input the test dataset into the second neural network model to conduct a prediction evaluation on the second neural network model. The specific steps are as follows: Input the continuous tubing working condition data and fatigue damage degree data in the test dataset into the second neural network model, repeat the test multiple times to obtain multiple predicted remaining life output values; Calculate the average correlation coefficient between the predicted remaining life output values and the corresponding actual remaining life values; Based on the calculated average correlation coefficient, conduct a prediction evaluation on the second neural network model. If the calculated average correlation coefficient meets the requirement, it indicates that the second neural network model is available and the prediction evaluation result is passed. If the calculated average correlation coefficient does not meet the requirement, it indicates that the second neural network model is not available and the prediction evaluation result is not passed, and retraining is required.

[0180] In one example, the test dataset is incorporated into the second neural network model test 10 times to obtain the average correlation coefficient r. The formula for the average correlation coefficient is as follows:

[0181]

[0182] Where Cov(X, Y) is the covariance of X and Y; Var[X] is the variance of X; Var[Y] is the variance of Y; X is the predicted remaining life output value; Y is the actual remaining life value.

[0183] S206: Based on the prediction evaluation result, determine the neural network prediction model for the remaining life of the continuous tubing. Specifically, if the prediction evaluation result is passed, the second neural network model is determined as the neural network prediction model for the remaining life of the continuous tubing; if the prediction evaluation result is not passed, retraining is required.

[0184] The solution of the present invention is further described below in conjunction with specific embodiments.

[0185] Embodiment:

[0186] (1) Gradient peak

[0187] Select the peak value of the gradient as the sample to predict the fatigue life of the coiled tubing under an internal pressure of 15 MPa. A total of 1,200 tests were carried out at 15 MPa, that is, 1,200 bending cycles were performed, and a total of 13 groups of magnetic memory signal gradient peak values of the coiled tubing were detected. Using these 13 groups of gradient peak values as the original data column, the data was processed using the above gray model prediction process. The gray parameters were calculated as: a = -0.1152, u = 0.5129. That is, the differential equation is:

[0188]

[0189] The predicted cumulative data column can be calculated as:

[0190]

[0191] And further obtain the predicted original data column through inverse cumulative generation:

[0192]

[0193] The obtained original data column, cumulative data column, predicted cumulative data column, and predicted original data column are summarized in Table 2 as follows.

[0194] Table 2 Related data table

[0195] Serial number Original data column Accumulated data column Predicted accumulated data column Predicted original data column 1 0.15 0.15 0.62 0.44 2 0.67 0.82 1.13 0.47 3 0.73 1.55 1.76 0.52 4 0.75 2.3 2.34 0.55 5 0.82 3.12 3.13 1.68 6 0.83 3.95 3.92 0.72 7 0.81 4.76 4.81 0.81 8 0.82 5.58 5.87 0.92 10 0.83 6.41 8.25 1.13 11 0.95 7.36 9.61 1.22 12 1.22 8.58 111.22 1.34 13 2.25 10.83 12.95 1.61

[0196] Perform a posterior difference test on the predicted gradient peak value data sequence, and calculate its posterior difference ratio and small error frequency.

[0197] The standard deviation of the original data column S1 = 0.4770 and the standard deviation of the residual S2 = 0.1795 are calculated. Therefore, the posterior difference ratio and small error frequency are respectively:

[0198]

[0199]

[0200] In this prediction, the accuracy levels of both the posterior difference ratio and the small error frequency belong to the "good" standard, indicating that the prediction result is relatively accurate.

[0201] Damage accumulates gradually, and the fatigue failure criterion can be expressed as:

[0202]

[0203] Therefore, the fatigue damage degree D can be expressed as the ratio of the number of cycles to the total number of life cycles. Then, the relationship between the damage degree and the serial number k in this example can be expressed as:

[0204]

[0205] Substituting into the above prediction formula, the relationship between the peak gradient and the fatigue damage degree of coiled tubing can be obtained:

[0206] K max = 0.4766e 0.1078(k-1) = 0.4766e 1.216D-0.1078

[0207] Similarly, the grey prediction of the peak gradient of coiled tubing under other internal pressure conditions and bending deflection conditions based on the GM(1,1) model is carried out, and the serial number and fatigue damage degree are converted. The relationships between the fatigue damage degree of coiled tubing and the peak gradient under different internal pressures and different bending deflections are shown in Tables 3 and 4 respectively.

[0208] Table 3 Prediction model of coiled tubing fatigue damage degree under different internal pressures based on peak gradient

[0209] Internal pressure / MPa Damage model Small error frequency Posterior difference ratio 15 <![CDATA[K max = 0.2048e 1.945D-0.256 > 0.9532 0.49 20 <![CDATA[K max = 0.1415e 2.369D-0.284 > 0.9152 0.66 25 <![CDATA[K max = 0.8356e 0.5682D-0.089 > 1.0 0.38 30 <![CDATA[K max = 0.1153e 0.6582D-0.059 > 0.8952 0.34

[0210] Table 4 Prediction model of coiled tubing fatigue damage degree under different bending deflections based on peak gradient

[0211] Deflection / mm Damage model Small error frequency Posterior difference ratio 200 <![CDATA[K max = 0.3256e 0.7512D-0.0534 > 0.8623 0.4623 250 <![CDATA[K max = 0.5583e 0.1483D-0.025 > 0.8245 0.6782 300 <![CDATA[K max = 0.5945e 0.2864D-0.049 > 0.912 0.6536 350 <![CDATA[K max = 0.5863e 0.715D-0.1152 > 1.0 0.5833

[0212] By establishing the relationship between the peak gradient of the magnetic memory signal of coiled tubing and its fatigue damage degree, a prediction model of coiled tubing fatigue damage degree based on the peak gradient under different internal pressure and bending deflection conditions is obtained. When conducting the magnetic memory detection of the tubing under similar conditions, according to the detected peak gradient, the fatigue damage degree of the coiled tubing can be estimated.

[0213] (2) Peak-to-peak gradient

[0214] Select the peak-to-peak gradient as the sample to predict the fatigue damage degree of coiled tubing under a bending deflection of 200m. A total of 1178 tests were carried out under a deflection of 200mm, and 13 groups of peak-to-peak gradients of the magnetic memory signal of coiled tubing were detected. Taking these 13 groups of peak-to-peak gradients as the original data series, the data is processed using the above grey model prediction process. The grey parameters are calculated as: a = 0.0705, u = 0.4197. That is, the differential equation is:

[0215]

[0216] The predicted cumulative data series can be calculated through this equation:

[0217]

[0218] And further, the predicted original data series is obtained through inverse cumulative generation:

[0219]

[0220] The obtained original data column, cumulative data column, predicted cumulative data column, and predicted original data column are summarized in Table 5 as follows.

[0221] Table 5 Related data table

[0222]

[0223]

[0224] Perform a posterior difference test on the predicted gradient peak-to-peak data sequence, and calculate its posterior difference ratio and small error frequency. Subtract the predicted column from the original data column to obtain the residual data column q(k), and calculate its average value as q = 0.1276. The calculated residual data and the difference between the residual and the average value of the residual are summarized in Table 6.

[0225] Table 6 Grey prediction data residual

[0226] Serial number Residual Residual difference 1 0.21 0.07 2 0.06 0.08 3 0.06 0.09 4 0.02 0.11 5 0.05 0.09 6 0.06 0.07 7 0.16 0.02 8 0.08 0.07 9 0.19 0.16 10 0.24 0.15 11 0.17 0.01 12 0.05 0.03 13 0.37 0.15

[0227] The standard deviation of the original data column S1 = 0.2526 and the standard deviation of the residual S2 = 0.1052 are calculated. Therefore, the posterior difference ratio and small error frequency are respectively:

[0228]

[0229]

[0230] It can be seen from the comparison with the accuracy level table that the posterior difference ratio in this prediction belongs to the "qualified" standard, which is very close to the "good" standard, and the small error frequency accuracy level belongs to the "good" standard. Therefore, the grey prediction method based on the GM(1,1) model is relatively accurate for predicting the gradient peak-to-peak value and can meet the requirements. Similarly, if you want to predict the fatigue damage degree of coiled tubing, it is necessary to study the relationship between the gradient peak-to-peak value and the fatigue damage degree. Therefore, it is necessary to convert the serial number into the fatigue damage degree. According to Miner's fatigue linear cumulative theory, the relationship between the damage degree and the serial number k in this example can be expressed as:

[0231]

[0232] Substitute the above prediction formula to obtain the relationship formula between the gradient peak-to-peak value and the fatigue damage degree of coiled tubing:

[0233] K sub = 0.4508e 0.1078(k-2) = 0.4506e 0.8301D-0.0758

[0234] Similarly, the gray prediction based on the GM(1,1) model is carried out for the peak-to-peak values of the coiled tubing gradient under other bending deflection conditions and internal pressure conditions, and the serial numbers are converted into fatigue damage degrees. The relationships between the fatigue damage degrees of the coiled tubing and the gradient peak values under different internal pressures and different bending deflections are shown in Tables 7 and 8 respectively. By establishing the relationship between the peak-to-peak value of the magnetic memory detection gradient of the coiled tubing and its fatigue damage degree, a prediction model for the fatigue damage degree of the coiled tubing based on the gradient peak-to-peak value under different internal pressures and bending deflections is obtained. When the magnetic memory detection of the tubing under similar conditions is carried out again, according to the detected gradient peak-to-peak value, the fatigue damage degree of the coiled tubing can be estimated.

[0235] Table 7 Fatigue damage degree of coiled tubing under different internal pressures based on gradient peak-to-peak value

[0236] Internal pressure / MPa Damage model Small error frequency Posterior difference ratio 15 <![CDATA[K max = 1.152e 0.6532D-0.065 > 0.8642 0.44 20 <![CDATA[K max = 0.3592e 1.8345D-0.2156 > 0.9031 0.37 25 <![CDATA[K max = 0.6219e 0.3562D-0.0452 > 1.0 0.36 30 <![CDATA[K max = 1.2356e 0.4562D-0.0653 > 0.8851 0.49

[0237] Table 8 Fatigue damage degree of coiled tubing under different bending deflections based on gradient peak-to-peak value

[0238]

[0239]

[0240] Finally, it should be noted that the above-described embodiments are only specific implementation manners of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or easily conceive of changes, or perform equivalent replacements for some of the technical features; and these modifications, changes or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting the fatigue damage degree of coiled tubing, characterized in that, It includes the following steps: Perform a bending fatigue test on the coiled tubing sample, and collect a series of magnetic memory signals of the coiled tubing sample during the bending fatigue test; Preprocess the collected series of magnetic memory signals, and extract multiple magnetic memory signal characteristic values therefrom; Arrange the extracted multiple magnetic memory signal characteristic values in the order of collection time as the original data column, establish a grey prediction model, and obtain the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column; Based on the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column, and in combination with Miner's fatigue linear cumulative theory, establish a coiled tubing fatigue damage degree prediction model representing the relationship between the magnetic memory signal characteristic values and the coiled tubing fatigue damage degree; Based on the established coiled tubing fatigue damage degree prediction model, predict the fatigue damage degree of the coiled tubing to be measured according to the detected magnetic memory signal characteristic values of the coiled tubing to be measured.

2. The coiled tubing fatigue damage degree prediction method according to claim 1, wherein Preprocess the collected series of magnetic memory signals, and extract multiple magnetic memory signal characteristic values therefrom, including: Perform noise reduction processing on the collected series of magnetic memory signal values; Perform gradient processing on the series of magnetic memory signal values after noise reduction processing; Extract multiple magnetic memory signal characteristic values from the series of magnetic memory signal values after gradient processing.

3. The coiled tubing fatigue damage degree prediction method according to claim 2, wherein Perform noise reduction processing by taking the average value of the magnetic memory signal values collected at each sampling point within a noise reduction window. The noise reduction processing uses the following formula: Among them, is the magnetic memory signal value at the i-th sampling point after noise reduction processing; j is the noise reduction window, that is, the number of data in the detection area; H n is the initial magnetic memory signal value at the n-th sampling point.

4. The coiled tubing fatigue damage degree prediction method according to claim 2, characterized in that Perform gradient processing by calculating the unit distance change rate of the magnetic memory signal values at each sampling point. The gradient processing uses the following formula: Among them, K i is the magnetic field gradient value at the i-th sampling point, ΔH is the change in the magnetic memory signal value, ΔL is the change in distance, H i , H i+k are the magnetic memory signal values at the i-th sampling point and the (i + k)-th sampling point respectively, and k is the distance between the i-th sampling point and the (i + k)-th sampling point.

5. The method for predicting the fatigue damage degree of coiled tubing according to claim 2, wherein The magnetic memory signal characteristic value is the gradient peak value or the gradient peak-to-peak value. The gradient peak value is the maximum value of the absolute value of the magnetic field gradient value within the detection area, and the gradient peak-to-peak value is the absolute value of the difference between the positive peak value of the magnetic field gradient and the negative peak value of the magnetic field gradient within the detection area.

6. The method for predicting the fatigue damage degree of coiled tubing according to claim 1, characterized in that The grey prediction model is the grey prediction GM(1,1) model. Arrange the extracted multiple magnetic memory signal characteristic values in the order of collection time as the original data column, establish a grey prediction model, and obtain the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column, including: Perform an accumulation process on the original data column X (0) to obtain the accumulated data column X (1) : X (0) = {X (0) (1), X (0) (2), X (0) (3), …, X (0) (n)} X (1) = {X (1) (1), X (1) (2), X (1) (3), …, X (1) (n)} Based on the original data column X (0) and the cumulative data column X (1) , the grey parameters in the grey prediction GM(1,1) model are calculated using the least squares method. The grey parameters include the development grey number a and the endogenous control grey number u: Y N = (X (0) (2), X (0) (3), X (0) (4), …, X (0) (n)) T ; Based on the calculated grey parameter, for the cumulative data series X (1) make a prediction to obtain a predicted cumulative data series whose terms are represented by the following formula Reverse process the predicted cumulative data column and restore it to the predicted original data column whose items are represented by the following formula Thus, establish a grey prediction GM(1,1) model to obtain the relationship between the magnetic memory signal eigenvalue and the serial number: Perform a posteriori difference test on the established grey prediction GM(1,1) model. If the accuracy requirement is met, the established grey prediction GM(1,1) model is available; otherwise, the established grey prediction GM(1,1) model is unavailable.

7. The method for predicting the fatigue damage degree of coiled tubing according to claim 6, wherein Perform a posteriori difference test on the established grey prediction GM(1,1) model using the posteriori difference ratio C and the small error frequency P. When the posteriori difference ratio C is less than the first threshold and the small error frequency P is greater than the second threshold, the established grey prediction GM(1,1) model meets the accuracy requirement; Among them, the posteriori difference ratio C is calculated by the following formula: Among them, S1 is the standard deviation of the original data column, and S2 is the standard deviation of the residuals. is the average value of the original data column, and q(k) is the k-th residual. is the average sum and variance of the residuals. The small error frequency P is calculated by the following formula:

8. The method for predicting the fatigue damage degree of coiled tubing according to claim 1, wherein Based on the relationship between the magnetic memory signal characteristic values and the serial numbers of the original data column, and in combination with Miner's fatigue linear cumulative theory, establish a coiled tubing fatigue damage degree prediction model representing the relationship between the magnetic memory signal characteristic values and the coiled tubing fatigue damage degree, including: According to Miner's fatigue linear cumulative theory, obtain the relationship between the coiled tubing fatigue damage degree and the serial number; According to the relationship between the fatigue damage degree of the coiled tubing and the serial number, and the relationship between the characteristic value of the magnetic memory signal and the serial number, a prediction model for the fatigue damage degree of the coiled tubing is established, which represents the relationship between the characteristic value of the magnetic memory signal and the fatigue damage degree of the coiled tubing.

9. The prediction method for the fatigue damage degree of coiled tubing according to claim 1, wherein, A bending fatigue test of the coiled tubing sample is carried out under different internal pressures, different bending deflections and different bending frequencies to establish a prediction model for the fatigue damage degree of the coiled tubing under different internal pressures, different bending deflections and different bending frequencies.

10. A method for predicting the remaining life of coiled tubing, characterized in that, Including: Predict the fatigue damage degree of the coiled tubing to be tested by using the coiled tubing fatigue damage degree prediction method described in any one of the above claims 1-9; Input the predicted fatigue damage degree of the coiled tubing to be tested and the working condition data of the coiled tubing to be tested into the trained neural network prediction model for the remaining life of the coiled tubing to obtain the remaining life of the coiled tubing to be tested.

11. The continuous coiled tubing remaining life prediction method according to claim 10, wherein The neural network prediction model for the remaining life of the coiled tubing is a BP neural network prediction model, which includes three parts: an input layer, a hidden layer and an output layer. The input layer is the working condition data and fatigue damage degree of the tubing, and the output layer is the remaining life.

12. The coiled tubing remaining life prediction method according to claim 11, wherein The working condition data of the tubing includes: the length of the coiled tubing, the depth of the wellbore fluid, the outer diameter of the coiled tubing, the mass per unit length, the tensile stiffness, the bending stiffness, the torsional stiffness, the internal pressure of the coiled tubing, and the drilling pressure.

13. The coiled tubing remaining life prediction method according to claim 11, wherein The BP neural network prediction model has 8 hidden layers, and the number of its neurons are: 10, 20, 40, 60, 40, 20, 10, 10 respectively.

14. The coiled tubing remaining life prediction method according to claim 11, wherein The BP neural network prediction model is constructed in the following way: Obtain multiple groups of sample data, and each group of sample data includes the working condition data of the coiled tubing, the fatigue damage degree data, and the actual remaining life data; Perform normalization processing on multiple groups of sample data; Divide the normalized sample data into a test data set, a training data set and a validation data set; Construct a first neural network model, train the first neural network model by using the training data set, and conduct a preliminary verification of the training results of the first neural network model by using the validation data set to obtain a second neural network model; Input the test data set into the second neural network model and conduct a prediction evaluation on the second neural network model; Based on the prediction evaluation results, determine the neural network prediction model for the remaining life of the coiled tubing.

15. The coiled tubing remaining life prediction method according to claim 14, wherein Perform normalization processing on multiple groups of sample data, including: performing normalization processing on each item of data in each group of sample data by using the following formula: y k =(xk - x min ) / (x max - x min ) Among them, x min is the minimum value of this item of data in multiple groups of sample data; x max is the maximum value of this item of data in multiple groups of sample data; x k is the value of this item of data in any group of sample data; y k is the value of this item of data after normalization processing.

16. The coiled tubing remaining life prediction method according to claim 14, characterized in that Train the first neural network model by using the training data set, including: Input the working condition data and fatigue damage degree data of the coiled tubing in the training data set into the first neural network model to obtain the predicted remaining life value; Based on the predicted remaining life value and the corresponding actual remaining life value, calculate the relative error value E, and the formula is as follows: where L is the number of samples in the training dataset; y k is the predicted remaining useful life value of the k-th sample; o k is the actual remaining useful life value of the k-th sample; Judge whether the relative error value E meets the threshold requirement. If the relative error value E meets the threshold requirement, the training is completed; otherwise, continue the training.

17. The coiled tubing remaining life prediction method according to claim 14, characterized in that Input the test data set into the second neural network model and conduct a prediction evaluation on the second neural network model, including: Input the coiled tubing working condition data and fatigue damage degree data in the test dataset into the second neural network model, and repeat the test multiple times to obtain multiple remaining life test output values; Calculate the average correlation coefficient between the remaining life test output value and the corresponding actual remaining life value. The formula is as follows: Among them, Cov(X, Y) is the covariance of X and Y; Var[X] is the variance of X; Var[Y] is the variance of Y; X is the remaining life test output value; Y is the actual remaining life value; Based on the calculated average correlation coefficient, conduct a prediction evaluation on the second neural network model. If the calculated average correlation coefficient meets the requirements, the prediction evaluation result is passed; if the calculated average correlation coefficient does not meet the requirements, the prediction evaluation result is not passed.

18. The coiled tubing remaining life prediction method according to claim 17, wherein Based on the prediction evaluation result, determine the coiled tubing remaining life neural network prediction model, including: If the prediction evaluation result is passed, determine the second neural network model as the coiled tubing remaining life neural network prediction model; if the prediction evaluation result is not passed, retrain.

Images