Pipe thinning prediction system

Abstract

To provide a pipe thinning prediction system which does not require additional evaluation by another prediction means.SOLUTION: A pipe thinning prediction system comprising an arithmetic device configured to classify a plurality of pieces of learning data into a plurality of thinning events, each piece of learning data including a combination of a thinning rate and a physical factor of a pipe, when a difference between the predicted wall-thickness thinning rate and the wall-thickness thinning rate included in the learning data is equal to or more than a predetermined amount, the classification is corrected to recreate the plurality of prediction models, a predetermined physical factor is given to the plurality of prediction models to predict the wall-thickness thinning rate, and whether or not additional evaluation is necessary is determined based on the plurality of predicted wall-thickness thinning rates.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a pipe wall thinning prediction system. [Background technology]

[0002] In power plants that use a steam cycle, such as nuclear power plants and thermal power plants, the main components of the steam cycle are connected by a large number of pipes. It is known that these pipes are subject to thinning due to the flow of coolant. For example, Patent Document 1 describes a method for predicting locations with high thinning using machine learning. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-46116 Summary of the Invention [Problem to be solved by the invention]

[0004] In machine learning, if the physical conditions of the prediction target are not included in the learning data, an extrapolated evaluation is performed, which may result in a predicted value that differs significantly from the actual wall thinning rate.The pipe wall thinning prediction method described in Patent Document 1 uses all actual measurement data for learning without distinguishing between wall thinning events, making it difficult to determine in advance whether the physical conditions of the prediction target are included in the learning data, posing the issue of requiring additional evaluation using other prediction methods such as computational fluid analysis.

[0005] An object of the present invention is to provide a pipe thinning prediction system that does not require additional evaluation using other prediction means. [Means for solving the problem]

[0006] A pipe thinning prediction system according to one aspect of the present invention includes a calculation device that classifies a plurality of learning data, each of which includes a combination of pipe thinning rates and physical factors, into a plurality of thinning events, creates a plurality of prediction models corresponding to each of the thinning events, predicts the thinning rate for each of the plurality of prediction models by providing the corresponding learning data, and if the difference between the predicted thinning rate and the thinning rate included in the learning data is equal to or greater than a predetermined amount, corrects the classification and recreates the plurality of prediction models, provides predetermined physical factors to the plurality of prediction models to predict the thinning rate, and determines whether additional evaluation is necessary based on the predicted plurality of thinning rates. A pipe thinning prediction system according to one aspect of the present invention includes a calculation device that classifies a plurality of learning data, each of which includes a combination of pipe thinning rates and physical factors, into a plurality of thinning events, creates a plurality of prediction models corresponding to each of the thinning events, predicts the thinning rate for each of the plurality of prediction models by providing the corresponding learning data, and if the difference between the predicted thinning rate and the thinning rate included in the learning data is equal to or greater than a predetermined amount, corrects the classification and recreates the plurality of prediction models. A pipe thinning prediction system according to one aspect of the present invention includes a computing device that predicts the thinning rate by applying predetermined physical factors to multiple prediction models corresponding to each of multiple thinning events, and determines whether additional evaluation is necessary based on the multiple predicted thinning rates. [Effects of the Invention]

[0007] According to the present invention, additional evaluation by other prediction means is not required. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a schematic diagram showing a hardware configuration of a pipe wall-thinning prediction system according to the first embodiment. [Figure 2] FIG. 2 is a diagram schematically illustrating the pipe wall thinning prediction process according to the first embodiment. [Figure 3] FIG. 3 is an example of a table showing the relationship between wall-thinning events, related physical factors, and shape factors according to the first embodiment. [Figure 4] FIG. 4 is a table showing an example of predicted values. [Figure 5] FIG. 5 is a diagram illustrating an example of the prediction result determination list. [Figure 6] FIG. 6 is a flowchart of the learning process. [Figure 7] FIG. 7 is a flowchart of the prediction process. [Figure 8] FIG. 8 is a diagram schematically showing the pipe wall-thinning prediction process according to the second embodiment. [Figure 9] FIG. 9 is a table showing an example of the standard deviation. [Figure 10] FIG. 10 is a diagram illustrating an example of the prediction result determination list. DETAILED DESCRIPTION OF THE INVENTION

[0009] First Embodiment A pipe wall thinning prediction system according to an embodiment of the present invention will be described with reference to FIGS.

[0010] FIG. 1 is a schematic diagram showing the hardware configuration of a pipe wall-thinning prediction system according to a first embodiment. The pipe wall-thinning prediction system 1 is configured by a computer including an arithmetic device 11 such as a central processing unit (CPU), a microprocessing unit (MPU), or a digital signal processor (DSP); a non-volatile memory 12 such as a read-only memory (ROM), a flash memory, or a hard disk drive; a volatile memory 13 called a random access memory (RAM); an input / output interface 14; and other peripheral circuits. These hardware components work together to run software and realize multiple functions. The pipe wall-thinning prediction system 1 may be configured by a single computer or multiple computers. The arithmetic device 11 may be an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or the like.

[0011] The nonvolatile memory 12 stores programs capable of executing various calculations. In other words, the nonvolatile memory 12 is a storage medium (storage device) from which programs for realizing the functions of this embodiment can be read. The volatile memory 13 is a storage medium (storage device) that temporarily stores the results of calculations performed by the calculation device and signals input from the input / output interface 14. The calculation device 11 is a device that loads the programs stored in the nonvolatile memory 12 into the volatile memory 13 and executes calculations, and performs predetermined calculations on data taken in from the input / output interface 14, the nonvolatile memory 12, and the volatile memory 13 in accordance with the programs.

[0012] The input section of the input / output interface 14 converts signals input from various devices (such as the storage device 2) into data that can be calculated by the arithmetic device 11. The output section of the input / output interface 14 generates an output signal according to the calculation result of the arithmetic device 11, and outputs the signal to various devices (such as the storage device 15).

[0013] A storage device 2, such as an HDD or SSD, is connected to the pipe thinning prediction system 1. A plurality of pieces of learning data 21 are stored in the storage device 2. The arithmetic device 11 can read the learning data 21 from the storage device 2 via the input / output interface 14.

[0014] The plurality of training data 21 are data sets encompassing physical factors related to typical wall thinning phenomena. Examples of physical factors include pipe geometry, reduction ratio, bending angle, bending R ratio, upstream pipe geometry, distance to upstream element, upstream bending angle, upstream bending R ratio, pipe diameter, pipe material, wall thickness, flow state, flow velocity, temperature, pressure, density, wetness, void fraction, droplet size, dissolved oxygen concentration, solid-liquid ratio, and upstream / downstream pressure ratio. Each training data 21 includes a set of one or more physical factors and a corresponding wall thinning rate. The wall thinning rate is either an actual measurement value of an actual pipe or a theoretical value evaluated using a physical model such as computational fluid analysis.

[0015] FIG. 2 is a diagram schematically showing a pipe wall-thinning prediction process according to the first embodiment. The pipe wall-thinning prediction process 3 is a process executed by the calculation device 11. The pipe wall-thinning prediction process 3 includes a learning process 4 and a prediction process 5. The learning process 4 is a process for constructing a plurality of prediction models based on a plurality of learning data 21 prepared in advance. The prediction process 5 is a process for applying the plurality of prediction models constructed in the learning process 4 to input data 70 related to a specific pipe, and predicting the wall-thinning rate of the pipe.

[0016] The learning process 4 includes a classification process 41, a model creation process 42, and a comparison process 43. The classification process 41 classifies each piece of training data 21 into one of the wall-thinning events. In other words, the classification process 41 classifies multiple pieces of training data, each of which includes a combination of a pipe wall-thinning rate and a physical factor, into multiple wall-thinning events. Examples of wall-thinning events include flow-accelerated corrosion, two-phase flow-accelerated corrosion, droplet impingement erosion, flashing erosion, cavitation erosion, and solid particle impingement erosion. In other words, the classification process 41 estimates the factors that caused the pipe wall thinning represented by the training data 21 and classifies the training data 21 by those factors. Here, a wall-thinning event refers to the factor that caused the pipe wall thinning. For example, in Figure 2, multiple pieces of training data 21 are classified into three types of training data 21a, 21b, and 21c.

[0017] In the classification process 41, the computing device 11 refers to the piping system of the learning data 21 and classifies the learning data 21 into one of the representative wall thinning events predetermined for each piping system. For example, if the piping system is a condensate system or a feedwater system, it is classified as flow-accelerated corrosion. Alternatively, if the piping system is a main steam system or a steam extraction system, it is classified as droplet impingement erosion. Heat balance analysis or computational fluid analysis of the piping system may also be used. For example, the computing device 11 evaluates the wetness from these analyses, and if the wetness is equal to or greater than a threshold and the pressure is low, it classifies the learning data 21 as droplet impingement erosion. Furthermore, if the wetness is equal to or greater than a threshold and the pressure is high, the computing device 11 classifies the learning data 21 as two-phase flow-accelerated corrosion. On the other hand, if the wetness is equal to or less than the threshold, the computing device 11 classifies the learning data 21 as flow-accelerated corrosion.

[0018] Note that pipe wall thinning may be caused by multiple different factors, and a combination of two or more of these wall thinning events may be used as one of the wall thinning events to be classified. For example, among the training data 21 that corresponds to droplet impact erosion, training data 21 related to piping immediately after a restriction such as an orifice or valve is classified as flashing erosion. Among the training data 21 that corresponds to flow-accelerated corrosion, training data 21 in which the fluid pressure falls below the saturation pressure due to pressure loss in an orifice or valve is classified as cavitation erosion. Training data 21 in which the fluid contains fine particles such as gravel or dust is classified as solid particle impact erosion. Training data 21 that share the above characteristics is classified as a wall thinning event that combines these multiple wall thinning events.

[0019] The model creation process 42 is a process for creating multiple prediction models corresponding to each metal-thinning event. For example, a prediction model is created for each metal-thinning event to which the training data 21 is classified, such as a prediction model corresponding to flow-accelerated corrosion, a prediction model corresponding to two-phase flow-accelerated corrosion, or a prediction model corresponding to droplet impact erosion. The computing device performs machine learning using the training data 21 corresponding to each metal-thinning event to build a prediction model for each metal-thinning event. Examples of machine learning techniques that can be used include neural networks and other techniques. For example, in FIG. 2, three prediction models 60a, 60b, and 60c are created in the model creation process 42.

[0020] In the model creation process 42, the computing device 11 extracts only physical quantities related to the classified wall thinning event from the learning data 21 and uses them to train the prediction model. For example, to train a prediction model for flow-accelerated corrosion, the following data are used: wall thinning rate, pipe shape, reduction ratio, bend angle, bend R ratio, upstream pipe shape, distance to upstream element, upstream bend angle, upstream bend R ratio, pipe diameter, pipe material, wall thickness, flow state, flow velocity, temperature, and dissolved oxygen concentration. Similarly, to train a prediction model for droplet impact erosion, the following data are used: wall thinning rate, pipe shape, reduction ratio, bend angle, bend R ratio, upstream pipe shape, distance to upstream element, upstream bend angle, upstream bend R ratio, pipe diameter, pipe material, wall thickness, flow state, flow velocity, pressure, density, wetness, and droplet diameter.

[0021] 3 is an example of a table showing the relationship between wall-thinning events, related physical factors, and shape factors according to the first embodiment. By selecting related physical quantities from the learning data 21 in this way, only information related to wall-thinning events can be used for learning, thereby reducing noise during learning and improving prediction accuracy.

[0022] The comparison process 43 is a process in which, for each of the multiple prediction models, the corresponding learning data 21 is provided to predict a wall-thickness reduction rate, and the predicted wall-thickness reduction rate is compared with the wall-thickness reduction rate included in the learning data 21. That is, in the comparison process 43, the calculation device 11 first predicts a wall-thickness reduction rate from the learning data 21 using the prediction model constructed in the model creation process 42. If the difference between the predicted wall-thickness reduction rate (predicted wall-thickness reduction rate) and the wall-thickness reduction rate included in the learning data 21 (correct wall-thickness reduction rate) is less than a predetermined amount (a predetermined allowable error), the calculation device 11 determines that a prediction model with sufficient accuracy has been constructed, and terminates the comparison process 43. On the other hand, if the difference between the predicted wall-thickness reduction rate and the wall-thickness reduction rate included in the learning data 21 is equal to or greater than the predetermined amount, the calculation device 11 determines that the accuracy of the prediction model is low, corrects the classification performed by the classification process 41, and recreates the multiple prediction models. That is, the calculation device 11 re-executes the classification process 41 and the model creation process 42.

[0023] For wall-thinning events corresponding to a prediction model in which the difference between the predicted wall-thinning rate and the correct wall-thinning rate is equal to or greater than a predetermined amount, the calculation device 11 reclassifies the corresponding learning data 21 into another wall-thinning event and recreates the prediction model using the reclassified learning data 21. For example, if the calculation device 11 determines that the accuracy of the prediction model corresponding to "flow-accelerated corrosion" is low, it reclassifies the learning data 21 classified as "flow-accelerated corrosion" into "liquid impact erosion," a first predetermined reclassification candidate. The calculation device 11 then recreates the prediction model and performs the comparison process 32 again. If the accuracy is again determined to be low, the calculation device 11 further reclassifies the learning data 21 into a second predetermined reclassification candidate. The calculation device 11 then recreates the prediction model and performs the comparison process 32 again. This feedback loop allows all learning data 21 to be classified into appropriate wall-thinning events, thereby improving prediction accuracy. The allowable error may be, for example, the error of the correct wall-thinning rate (±0.1 mm / year).

[0024] The prediction process 5 includes an inference process 51 and an evaluation process 52. The inference process 51 is a process in which predetermined physical factors are applied to multiple prediction models to predict wall-thinning rates and obtain multiple predicted values. In the inference process 51, the calculation device 11 generates multiple wall-thinning rates for one piece of input data 70. For example, in FIG. 2, three prediction models 60a, 60b, and 60c were created in the learning process 4, so in the inference process 51, three predicted values ​​are generated for one piece of input data 70: a predicted value 80a by the prediction model 60a, a predicted value 80b by the prediction model 60b, and a predicted value 80c by the prediction model 60c.

[0025] Fig. 4 is a table showing an example of predicted values. For example, if there are three prediction models and 100 pieces of input data 70, 3 x 100 = 300 predicted values ​​will be obtained in the inference process 51. Fig. 4 shows, for each of the many pieces of input data 70, an example of the piping system represented by that input data 70 and the wall thinning rate (predicted value) for each wall thinning event obtained for that input data 70.

[0026] The evaluation process 52 is a process for determining whether or not additional evaluation is required for each piece of input data 70, based on the multiple predicted values ​​(i.e., multiple predicted wall-thinning rates) obtained in the inference process 51. In the evaluation process 52, if all of the multiple predicted values ​​corresponding to a certain piece of input data 70 are equal to or greater than a predetermined threshold, the calculation device 11 determines that the prediction result is invalid. On the other hand, if at least one of the multiple predicted values ​​corresponding to a certain piece of input data 70 is less than the predetermined threshold, the calculation device 11 determines that the prediction result is valid. The predetermined threshold used here is, for example, the maximum wall-thinning rate for each piping system calculated from the past input data 70 and the learning data 21. The calculation device 11 determines that additional evaluation is required for a predicted value determined to be invalid, and determines that additional evaluation is not required for a predicted value determined to be valid.

[0027] FIG. 5 is a diagram showing an example of a prediction result determination list. The prediction result determination list 90 includes the piping system corresponding to each of the plurality of input data 70, the threshold value used in the evaluation process 52, a plurality of predicted values, whether additional evaluation is necessary, and a predicted wall-thinning rate. Here, the predicted wall-thinning rate included in the prediction result determination list 90 is the maximum value among the predicted values ​​less than the threshold value. The computing device 11 supports the user in managing pipe wall-thinning by displaying the prediction result determination list 90 as shown in FIG. 5 on a display device (not shown), for example. The predicted wall-thinning rate is included in the prediction result determination list 90 to ensure maintainability in pipe wall-thinning management.

[0028] FIG. 6 is a flowchart of learning process 4. In step S100, the calculation device 11 classifies the plurality of learning data 21 into a plurality of wall-thinning events. In step S110, the calculation device 11 constructs a plurality of prediction models corresponding to each of the plurality of wall-thinning events using the plurality of learning data 21 classified in step S100. In step S120, the calculation device 11 predicts a wall-thinning rate for each of the plurality of prediction models constructed in step S110 by providing the corresponding learning data 21. In step S130, the calculation device 11 determines whether the difference between the wall-thinning rate predicted in step S120 and the wall-thinning rate included in the corresponding learning data 21 is equal to or greater than a predetermined amount. If the difference is equal to or greater than the predetermined amount, the process proceeds to step S140. In step S140, the calculation device 11 reclassifies the learning data 21 (i.e., corrects the classification of the learning data 21). In step S150, the calculation device 11 reconstructs (recreates) a plurality of prediction models using the learning data 21 reclassified in step S140. Thereafter, the process proceeds to step S 120. On the other hand, if the difference in the wall-thickness reduction rate is less than the predetermined amount in step S130, learning process 4 shown in FIG.

[0029] FIG. 7 is a flowchart of prediction process 5. In step S200, the calculation device 11 predicts a wall-thinning rate for each of a plurality of input data 70 using a plurality of prediction models, and obtains a plurality of predicted values ​​for each input data 70. In step S210, the calculation device 11 determines whether additional evaluation is necessary for each input data 70, based on the plurality of predicted values ​​(i.e., the plurality of predicted wall-thinning rates) obtained for each input data 70 in step S200. In step S220, the calculation device 11 creates a prediction result determination list 90, an example of which is shown in FIG. 5. In step S230, the calculation device 11 outputs the prediction result determination list 90, an example of which is shown in FIG. 5, to an external device, such as a display device.

[0030] According to the above-described first embodiment, the following effects are achieved.

[0031] (1) In the learning process 4, the computing device 11 classifies multiple learning data sets, each containing a combination of pipe wall thinning rates and physical factors, into multiple wall thinning events, creates multiple prediction models corresponding to each of the wall thinning events, and predicts the wall thinning rate for each of the multiple prediction models using the corresponding learning data. If the difference between the predicted wall thinning rate and the wall thinning rate included in the learning data is equal to or greater than a predetermined amount, the computing device 11 corrects the classification and recreates the multiple prediction models. In the prediction process 5, the computing device 11 predicts the wall thinning rate by applying predetermined physical factors to the multiple prediction models, and determines whether additional evaluation is necessary based on the multiple predicted wall thinning rates. This eliminates the need for additional evaluation using other prediction means. Furthermore, the accuracy of wall thinning rate prediction using machine learning can be improved without increasing the amount of learning data.

[0032] (2) The calculation device 11 includes flow-accelerated corrosion, two-phase flow-accelerated corrosion, droplet impact erosion, flashing erosion, cavitation erosion, and solid particle impact erosion, or a combination of two or more of these, in the multiple metal-thinning phenomena. By doing so, a prediction model tailored to the metal-thinning phenomenon can be prepared, thereby improving the accuracy of metal-thinning rate prediction by machine learning.

[0033] (3) The calculation device 11 creates multiple prediction models, which are neural networks. This allows for the preparation of prediction models tailored to wall-thinning events, thereby improving the accuracy of wall-thinning rate predictions using machine learning.

[0034] Second Embodiment A pipe wall-thinning prediction system according to a second embodiment of the present invention will be described with reference to Figures 8 to 10. Note that the same reference symbols are used for components that are the same as or equivalent to those described in the first embodiment, and differences will be mainly described.

[0035] Fig. 8 is a diagram similar to Fig. 2, and is a diagram schematically illustrating a pipe wall thinning prediction process according to a second embodiment. In the model creation process 42 of the second embodiment, multiple prediction models are constructed for each wall thinning event. In other words, in the first embodiment, the same number of prediction models as the number of wall thinning events to be classified are constructed, but in the second embodiment, (number of wall thinning events) x N prediction models are constructed, where N is an integer equal to or greater than 3.

[0036] The computing device 11 prepares multiple learning sets with different contents by repeating a procedure in which, for example, 80% of all learning data 21 for a certain wall-thinning event is randomly extracted to create one learning set. In the model creation process 42 according to the second embodiment, a prediction model corresponding to the multiple learning sets prepared in this manner is created for each wall-thinning event. In order to accurately obtain the standard deviation described below, it is desirable to create approximately 20 to 100 prediction models for each wall-thinning event.

[0037] In the inference process 51 of the second embodiment, multiple prediction models are used to generate multiple wall-thinning rates for each wall-thinning event for one piece of input data 70. For example, if 100 prediction models are constructed for each wall-thinning event, since there are three wall-thinning events in FIG. 8, the inference process 51 generates 100 x 3 = 300 predicted values ​​for one piece of input data 70. The calculation device 11 calculates the standard deviation of the predicted values ​​generated for each wall-thinning event. For example, since there are three wall-thinning events in FIG. 8, three standard deviations are calculated by the calculation device 11. Since there are 100 predicted values ​​for each wall-thinning event, each standard deviation is calculated from the 100 predicted values.

[0038] Fig. 9 is a table showing an example of standard deviation. For example, if there are three prediction models and 100 pieces of input data 70, 3 x 100 = 300 standard deviations will be obtained in the inference process 51. Fig. 9 shows, for each of the many pieces of input data 70, an example of the piping system represented by that input data 70 and the standard deviation of the wall thinning rate (predicted value) for each wall thinning event obtained for that input data 70.

[0039] In the evaluation process 52 of the second embodiment, whether or not additional evaluation is required for each piece of input data 70 is determined based on the multiple standard deviations obtained in the inference process 51. In the evaluation process 52, if all of the multiple standard deviations corresponding to a certain piece of input data 70 are equal to or greater than a predetermined threshold, the calculation device 11 determines that the prediction result is invalid. On the other hand, if at least one of the multiple standard deviations corresponding to a certain piece of input data 70 is less than the predetermined threshold, the calculation device 11 determines that the prediction result is valid. The predetermined threshold used here is, for example, the error in the actual measurement of the wall-thinning rate. The calculation device 11 determines that a prediction value determined to be invalid requires additional evaluation, and determines that a prediction value determined to be valid does not require additional evaluation.

[0040] Fig. 10 is a diagram showing an example of a prediction result determination list. The prediction result determination list 90 includes the piping system corresponding to each of the plurality of input data 70, the threshold value used in the evaluation process 52, a plurality of predicted values ​​and standard deviations, whether additional evaluation is required, and a predicted wall-thinning rate. Here, the predicted wall-thinning rate included in the prediction result determination list 90 is the predicted wall-thinning rate due to the wall-thinning event with the smallest standard deviation. The computing device 11 supports the user in managing wall-thinning of pipes by displaying the prediction result determination list 90 as shown in Fig. 10 on, for example, a display device (not shown).

[0041] According to the above-described second embodiment, the following advantageous effects are achieved.

[0042] (1) The calculation device 11 creates at least three or more different prediction models for the same wall-thinning event, and determines whether additional evaluation is necessary based on the standard deviation of the three or more wall-thinning rates obtained for the same wall-thinning event. In the first embodiment, the maximum wall-thinning rate of past data is used to determine the validity of the predicted value, which may result in an overly conservative evaluation of the wall-thinning rate. In contrast, in this embodiment, the validity of the predicted value is determined based on statistical evaluation, allowing for a more precise determination of the validity of the predicted value.

[0043] The following modified examples are also within the scope of the present invention, and it is possible to combine the configuration shown in the modified example with the configuration described in the above embodiment, to combine the configurations described in the different embodiments above, or to combine the configurations described in the different modified examples below.

[0044] <Variation 1> The learning process 4 and the prediction process 5 may be performed by different devices. For example, a first device may be provided that performs only the learning process 4, and a second device may be provided that performs only the prediction process 5 using the prediction model constructed by the first device. Furthermore, the prediction model constructed by the first device may be distributed to multiple second devices using a network, a portable storage medium, or the like.

[0045] <Variation 2> The learning data 21 may be input to the pipe-wall-thinning prediction system 1 from a location other than the storage device 2. For example, the learning data 21 may be input to the pipe-wall-thinning prediction system 1 via a wired or wireless network.

[0046] Although the embodiments of the present invention have been described above, the above embodiments merely illustrate some of the application examples of the present invention, and it is not intended that the technical scope of the present invention be limited to the specific configurations of the above embodiments. [Explanation of symbols]

[0047] 1...Pipe thinning prediction system, 2...Storage device, 3...Pipe thinning prediction processing, 4...Learning processing, 5...Prediction processing, 11...Calculation device, 12...Non-volatile memory, 13...Volatile memory, 14...Input / output interface, 21...Learning data, 41...Classification processing, 42...Model creation processing, 43...Comparison processing, 51...Inference processing, 52...Evaluation processing, 70...Input data, 90...Prediction result judgment list

Claims

1. A pipe thinning prediction system including a calculation device, The computing device classifying a plurality of learning data sets, each of which includes a combination of a pipe wall thinning rate and a physical factor, into a plurality of wall thinning events; creating a plurality of predictive models corresponding to each of the wall-thinning events; predicting a wall-thinning rate for each of the plurality of prediction models by providing the corresponding learning data, and if a difference between the predicted wall-thinning rate and the wall-thinning rate included in the learning data is equal to or greater than a predetermined amount, correcting the classification and recreating the plurality of prediction models; Predicting a wall-thinning rate by applying predetermined physical factors to the plurality of prediction models, and determining whether or not additional evaluation is necessary based on the plurality of predicted wall-thinning rates. Pipe thinning prediction system.

2. 2. The pipe wall thinning prediction system according to claim 1, The calculation device creates at least three or more different prediction models for the same wall thinning event, and determines whether additional evaluation is necessary based on the standard deviation of the three or more wall thinning rates obtained for the same wall thinning event.

3. 2. The pipe wall thinning prediction system according to claim 1, The calculation device includes flow-accelerated corrosion, two-phase flow-accelerated corrosion, liquid droplet impingement erosion, flashing erosion, cavitation erosion, and solid particle impingement erosion, or a combination of two or more thereof, in the plurality of wall thinning events.

4. 2. The pipe wall thinning prediction system according to claim 1, The calculation device creates the plurality of prediction models, each of which is a neural network.

5. A pipe thinning prediction system including a calculation device, The computing device classifying a plurality of learning data sets, each of which includes a combination of a pipe wall thinning rate and a physical factor, into a plurality of wall thinning events; creating a plurality of predictive models corresponding to each of the wall-thinning events; predicting a wall-thinning rate for each of the plurality of prediction models by providing the corresponding learning data, and if a difference between the predicted wall-thinning rate and the wall-thinning rate included in the learning data is equal to or greater than a predetermined amount, correcting the classification and recreating the plurality of prediction models; Pipe thinning prediction system.

6. A pipe thinning prediction system including a calculation device, The computing device predicting a wall-thinning rate by applying predetermined physical factors to a plurality of prediction models corresponding to each of a plurality of wall-thinning events, and determining whether or not additional evaluation is necessary based on the predicted plurality of wall-thinning rates; Pipe thinning prediction system.

Images