Residual life prediction device and residual life prediction method

Abstract

To provide a residual life prediction with high accuracy.SOLUTION: A residual life prediction device comprises: a simulation processing part 102 for simulating signal intensity with respect to an inspection condition being a factor of variation of a defect detection signal by a non-destructive test, and a defect condition being a condition of a defect which is an object of the non-destructive test; a probability distribution estimation part 104 for estimating a probability distribution related to the defect condition estimated from actual signal intensity, on the basis of an analysis result by the simulation processing part 102 and actual signal intensity being signal intensity of a response signal obtained by an actual inspection; and a residual life prediction part 105 for calculating a time change of the probability distribution related to the defect condition estimated by the probability distribution estimation part 104, estimating as a residual life, a time period where an overlapping part between the probability distribution related to the defect condition after lapse of a prescribed period, and a preset destruction probability distribution of an inspection object, satisfies a prescribed condition, then outputting the estimated residual life to an output part.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a remaining life prediction device and a remaining life prediction method. [Background technology]

[0002] During operation of social infrastructure equipment such as power plants and railway bogies, non-destructive testing such as ultrasonic testing is sometimes used to check for defects in component parts. If a defect is found during the inspection, it is common to temporarily suspend operation and carry out repairs.

[0003] However, rather than stopping operation and carrying out repairs every time a defect is found during an inspection, it is more reasonable to repair multiple locations simultaneously during planned operation stoppages such as regular inspections. To do this, it is necessary to predict the remaining life of the monitored object based on the defects, evaluate the risk of the defects found, and guarantee that operation can be continued until the next operation stoppage.

[0004] In response to such problems, for example, Patent Document 1 discloses a plant inspection method comprising: "a step of inspecting the evaluation target part using an inspection method set for each combination of the type of the evaluation target part, which includes at least one of a circumferential weld or a longitudinal weld of a piping, or a nozzle weld, and the thickness of the evaluation target part; and a step of selecting measurement items for additional measurement to obtain parameters necessary for improving the accuracy of a remaining life evaluation of the evaluation target part, which is performed based on the inspection results of the evaluation target part using the inspection method" (see claim 1).

[0005] Furthermore, Patent Document 2 discloses a probabilistic fatigue life prediction using ultrasonic inspection data that takes into account EIFS uncertainties, stating, "A method for probabilistically predicting the fatigue life of a material includes a step (41) of sampling a random variable related to an actual equivalent initial flaw size (EIFS), a step (42) of generating a random variable related to a parameter (lnC,m) of a fatigue crack growth equation from a multivariate distribution, and a step (43) of solving the fatigue crack growth equation using these random variables. The recorded EIFS data is obtained by scanning a target object with an ultrasonic wave, recording an echo signal from the target object, and converting the echo signal intensity into an equivalent reflector size using previously recorded values ​​from the scanned calibration block. The equivalent reflector size includes the recorded EIFS data" (see abstract). [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent No. 7125266 [Patent Document 2] Special Publication No. 2015-512526 Summary of the Invention [Problem to be solved by the invention]

[0007] The technology described in Patent Document 1 requires improvement in that additional measurements are required to improve accuracy. In addition, in the technology described in Patent Document 2, actual flaw detection signals vary depending on the inspection conditions as well as the nature of the flaws, such as the flaw shape and orientation, so it is necessary to take into consideration that remaining life prediction becomes uncertain.

[0008] The present invention has been made in view of the above background, and an object of the present invention is to perform highly accurate remaining life prediction. [Means for solving the problem]

[0009] In order to solve the above-mentioned problems, the present invention comprises a simulation processing unit that simulates inspection conditions that are factors in the variation of defect detection signals by non-destructive inspection, and signal strength for defect conditions that are conditions of defects that are the target of the non-destructive inspection, a probability distribution estimation unit that estimates a probability distribution for the defect condition estimated from the actual signal strength based on the analysis results by the simulation processing unit and an actual signal strength that is the signal strength of a response signal obtained in an actual inspection, and a remaining life prediction unit that calculates a time change in the probability distribution for the defect condition estimated by the probability distribution estimation unit, estimates the time at which the overlapping part of the probability distribution for the defect condition after a predetermined time and a predetermined fracture probability distribution of the object to be inspected satisfies a predetermined condition as the remaining life, and outputs the estimated remaining life to an output unit. Other solutions will be described in the embodiments as appropriate. Effect of the Invention

[0010] According to the present invention, it is possible to perform highly accurate remaining life prediction. [Brief description of the drawings]

[0011] [Figure 1] 1 is a functional block diagram showing a configuration of a remaining life prediction device according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating a hardware configuration of the remaining life prediction device according to the present embodiment. [Diagram 3] FIG. 2 is a conceptual diagram showing an example of a non-destructive inspection system performed in this embodiment. [Figure 4A] FIG. 1 is a conceptual diagram (part 1) showing an example of propagation analysis by the simulation processing unit. [Figure 4B] 11 is a conceptual diagram showing an example of a propagation analysis performed by a simulation processing unit. FIG. [Diagram 5] FIG. 11 is a diagram illustrating an example of an analysis result by a simulation processing unit. [Figure 6A] FIG. 2 is a conceptual diagram (part 1) showing an example of a defect shape model calculated by a simulation processing unit used in the present embodiment. [Figure 6B]FIG. 2 is a conceptual diagram (part 2) showing an example of a defect shape model calculated by the simulation processing unit used in the present embodiment. [Figure 6C] FIG. 11 is a conceptual diagram (part 3) showing an example of a defect shape model calculated by the simulation processing unit used in this embodiment. [Figure 7A] FIG. 11 is a conceptual diagram (part 1) showing an example of a reflected wave for each defect shape model by a simulation processing unit. [Figure 7B] FIG. 11 is a conceptual diagram (part 2) showing an example of a reflected wave for each defect shape model by the simulation processing unit. [Figure 7C] FIG. 11 is a conceptual diagram (part 3) showing an example of a reflected wave for each defect shape model by the simulation processing unit. [Figure 8] 4 is a conceptual diagram showing an example of a response signal input to a response signal input unit used in the present embodiment. FIG. [Figure 9] 4 is a conceptual diagram showing a method for estimating a defect condition probability distribution by a probability distribution estimation unit used in the present embodiment. FIG. [Figure 10] 11A and 11B are conceptual diagrams illustrating the estimation results of the probability distribution of the probability distribution estimation unit according to the present embodiment, taking into account the defect size and defect direction. [Figure 11] 1 is a conceptual diagram showing a change in probability distribution of defect size over time and a method of predicting remaining life. FIG. [Figure 12] 3 is a flowchart showing the steps of a remaining life prediction method according to the first embodiment. [Figure 13] FIG. 13 is a diagram showing an example of a display screen. [Figure 14] 10 is a flowchart showing the steps of a remaining life prediction method according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the drawings, the same components are designated by the same reference numerals, and detailed description of overlapping parts will be omitted.

[0013] [First embodiment] First, a first embodiment of the present invention will be described with reference to FIGS.

[0014] <Remaining life prediction device 1> FIG. 1 is a functional block diagram showing the configuration of a remaining life prediction device 1 according to this embodiment. 1, remaining life prediction device 1 includes a condition input unit 101, a simulation processing unit 102, a response signal input unit 103, a probability distribution estimation unit 104, and a remaining life prediction unit 105. Details of the processes performed by condition input unit 101 to remaining life prediction unit 105 will be described later. The condition input unit 101 receives input of inspection conditions, defect conditions, and the like that are factors that cause variations in a defect detection signal 511 (see FIG. 8) obtained by non-destructive inspection. The simulation processing unit 102 simulates the signal intensity for the inspection conditions and defect conditions that are conditions for a defect D that occurs in the inspection target T and is the subject of non-destructive inspection. The response signal input unit 103 receives a response signal 501 (see FIG. 8) obtained in an actual inspection. The probability distribution estimation unit 104 estimates a probability distribution related to the defect condition by combining the result from the simulation processing unit 102 and the defect detection signal strength 513 (see FIG. 8) input to the response signal input unit 103. The defect detection signal strength 513 is based on the response signal 501. The probability distribution related to the defect condition is output as a defect condition probability distribution 611 shown in FIG. 9 or a multidimensional defect condition probability distribution 611A shown in FIG. 10. The remaining life prediction unit 105 predicts (estimates) the remaining life of the inspection object T (see FIG. 3) based on the defect condition probability distribution 611 (or the multidimensional defect condition probability distribution 611A), and outputs the predicted remaining life to the output device 115 (see FIG. 2).

[0015] <Hardware configuration diagram> FIG. 2 is a diagram showing a hardware configuration of the remaining life prediction device 1 according to this embodiment. The remaining life prediction device 1 has a memory 111 configured with a RAM (Random Access Memory) etc. The remaining life prediction device 1 also has a calculation device 112 configured with a CPU (Central Processing Unit), a GPU (Graphics Processing Unit) etc. The remaining life prediction device 1 also has a storage device 113 configured with a HDD (Hard Disk Drive), an SSD (Solid State Drive), a ROM (Read Only Memory) etc.

[0016] Furthermore, the remaining life prediction device 1 has an input device 114 including a keyboard, a mouse, and the like, an output device 115 which is an output section including a display, and the like, and a communication device 116.

[0017] Then, the program stored in the storage device 113 is loaded into the memory 111. The loaded program is executed by the calculation device 112. This embodies the simulation processing unit 102, the probability distribution estimation unit 104, and the remaining life prediction unit 105 shown in Fig. 1. Moreover, the condition input unit 101 and the response signal input unit 103 in Fig. 1 correspond to the input device 114.

[0018] In this manner, the remaining life prediction device 1 of the present embodiment is realized as a software information processing device in a computer having hardware resources such as a CPU, a ROM, a RAM, and a HDD.

[0019] <Inspection conditions> Next, the inspection conditions input to the condition input unit 101 will be described with reference to FIG. FIG. 3 is a conceptual diagram showing an example of a non-destructive inspection system carried out in this embodiment. In this embodiment, ultrasonic testing will be particularly described as an example of non-destructive testing, but other non-destructive testing methods, such as eddy current testing and radiographic testing, can also be applied to this embodiment in the same manner.

[0020] 3 shows a schematic diagram of an inspection for detecting a defect D inside an inspection object T by using an ultrasonic sensor 121. As described above, inspection conditions, defect conditions, and material conditions are input to the condition input unit 101. The inspection conditions include the following: (A1) Variation in the installation position of the ultrasonic sensor 121 (installation position variation 201). (A2) Variation in the transmission direction of ultrasonic waves (transmission direction variation 202). The inspection conditions include either installation position variation 201 or transmission direction variation 202. Note that the inspection conditions are not limited to the above (A1) and (A2). The inspection conditions for each item such as (A1) and (A2) are input to the condition input unit 101 as a probability distribution. In addition, material conditions and allowable risk (acceptable probability of breakage) are also input to the condition input unit 101. The allowable risk is used when the remaining life prediction unit 105 predicts the remaining life. The defect conditions will be described later, and relate to the variation in size of defect D, etc. The material conditions are the variation in material constants such as the sound speed of the object T to be inspected.

[0021] <Processing of simulation processing unit 102) Next, the simulation processing unit 102 will be described with reference to FIGS. 4A to 7C. (Propagation analysis) 4A and 4B are conceptual diagrams showing an example of a propagation analysis by the simulation processing unit 102. Fig. 4A is a diagram showing the intensity distribution of an incident wave 311a by the ultrasonic sensor model 121M, and Fig. 4B is a diagram showing the intensity distribution of a reflected wave 311b by the ultrasonic sensor model 121M. Both the diagrams shown in Fig. 4A and Fig. 4B show the propagation of the reflected wave 311b by a physical simulation. The ultrasonic sensor model 121M is a model of the ultrasonic sensor 121 (see Fig. 3) used in the physical simulation.

[0022] Fig. 4A shows a propagation analysis result 301 indicating how an incident wave 311a is irradiated from the ultrasonic sensor model 121M to the defect model DM. Fig. 4B shows a propagation analysis result 302 indicating how a reflected wave 311b is reflected from the defect model DM. The defect model DM is a model of a defect D set in a physical simulation.

[0023] The simulation processing unit 102 identifies positions where the ultrasonic sensor 121 can be installed (possible installation positions) from the shape of the inspection target T. Then, the simulation processing unit 102 executes a propagation analysis for the assigned defect D according to the assigned probability distribution for each of the identified possible installation positions of the ultrasonic sensor 121. The assigned probability distribution is a probability distribution based on the variation regarding the inspection conditions. Based on the reflected wave 311b, the simulation processing unit 102 calculates the strength (signal strength) of the response signal 501 (see FIG. 8) received by the ultrasonic sensor 121. The propagation analysis by the simulation processing unit 102 can be realized by using a general physical simulation method or a data analysis method.

[0024] (Analysis results) Fig. 5 is a diagram showing an example of an analysis result by the simulation processing unit 102. Fig. 3 will be referred to as appropriate. In graph 321 shown in Fig. 5, the horizontal axis indicates the ultrasonic transmission direction (Theta) and the vertical axis indicates signal strength. Also, in each curve shown in Fig. 5, the defect size increases from the bottom to the top of the paper. In other words, each curve shown in graph 321 shown in Fig. 5 indicates the angle dependency of signal strength for each defect size. Fig. 5 shows that the strongest signal strength is obtained when ultrasonic waves are transmitted and received in a 45 degree direction.

[0025] By using a method similar to that shown in Fig. 5, it is possible to investigate the dependency of signal intensity on the installation position variation 201 shown in Fig. 3, material conditions such as the sound speed of the inspection target T, and defect conditions such as the variation in defect size. In this case, a physical simulation is used, similar to the example shown in Fig. 5.

[0026] (Defect shape) 6A to 6C are conceptual diagrams showing an example of a defect model DM calculated by the simulation processing unit 102 used in this embodiment. Fig. 6A illustrates a planar defect model DMa, Fig. 6B illustrates a spherical defect model DMb, and Fig. 6C illustrates a cylindrical defect model DMc. When the simulation processing unit 102 performs an analysis, the simulation processing unit 102 sets a defect model DM of a specified size at a specified angle at a specified position inside the object model 401. The object model 401 is a physical simulation model of the inspection target T (see Fig. 3). Then, the simulation processing unit 102 analyzes how ultrasonic waves are emitted from the ultrasonic sensor model 121M and reflected by the defect model DM in the physical simulation.

[0027] 6A to 6C also show the defect size variation (defect size variation 211), defect direction variation (defect direction variation 212), and defect position variation (defect position variation 213) that constitute the defect conditions. Also, the planar defect model DMa, the spherical defect model DMb, and the cylindrical defect model DMc shown in the defect models DMa to DMc each show the defect shape variation. Such defect conditions are set in advance in the storage device 113.

[0028] However, since the spherical defect model DMb shown in FIG. 6B cannot define the defect direction, the defect direction variation 212 is not shown.

[0029] The defect conditions include a defect size variation 211, and also any one of a defect direction variation 212, a defect position variation 213, and a defect shape variation.

[0030] (Reflected wave 311b for each defect model DM) 7A to 7C are conceptual diagrams showing an example of the reflected wave 311b for each defect model DM by the simulation processing unit 102. In FIG. 7A to 7C, as in FIGS. 6A to 6C, examples of the defect model DM are shown, which are a planar defect model DM (FIG. 7A), a spherical defect model DMb (FIG. 7B), and a cylindrical defect model DMb (FIG. 7C).

[0031] Note that, like in FIG. 4A and FIG. 4B, the reference numeral 311 indicates the state of propagation of ultrasonic waves by physical simulation. For representative defect models DM such as planar, spherical, and cylindrical defect models DM, the reflected waves 311b are stored inside the remaining life prediction device 1 (for example, the storage device 113 shown in FIG. 2). Alternatively, the user selects the reflected waves 311b of representative shapes such as planar, spherical, and cylindrical defect models DM from a list of the reflected waves 311b via an interface. For the reflected waves 311b of a defect model DM that is not in the list, the user inputs a formula or a result of a calculation in advance via an interface. Alternatively, an existence probability is set for each shape of the defect D, and the simulation processing unit 102 may select the shape of the defect D (one of the defect models DMa to DMc) according to this existence probability. The existence probabilities of these defect models DM are stored in a database (not shown) or the storage device 113. The existence probability of the defect model DM may change depending on the characteristics of the inspection target T.

[0032] In this way, the simulation processing unit 102 can perform processing by utilizing physical simulation. On the other hand, if comprehensive test data for past inspection data, calibration data, and defect conditions can be used for the state of ultrasonic wave propagation, it is also possible to store them in a database (not shown), storage device 113, etc. Then, the simulation processing unit 102 may refer to comprehensive test data for past inspection data, calibration data, and defect conditions stored in a database (not shown), storage device 113, etc. as necessary. It is possible to replace the physical simulation with such past inspection data, calibration data, and test data. In this way, an analysis can be performed taking into account the shape of the defect.

[0033] (Response signal 501) Next, the response signal input unit 103 will be described with reference to FIG. 8 is a conceptual diagram showing an example of a response signal 501 input to the response signal input unit 103 used in this embodiment. FIG. 3 will be referred to as appropriate. As shown in FIG. 8, a response signal 501 that changes over time is obtained by actual inspection. The response signal 501 is a signal obtained by actual inspection using the ultrasonic sensor 121. A defect detection signal 511 and a defect detection time 512 according to the defect conditions are obtained from the response signal 501. The defect detection signal 511 is a signal based on the response characteristics from the defect D, and the defect detection signal intensity 513 is the signal intensity of the defect detection signal 511. In other words, the defect detection signal intensity 513 corresponds to the actual signal intensity, which is the signal intensity of the response signal 501 obtained by actual inspection. In the section 521, a strong signal intensity of the response signal 501 is obtained, but this is a signal derived from the vibration of the ultrasonic sensor 121 itself.

[0034] As described above, the defect detection signal 511 and the defect detection time 512 have uncertainties that depend on the inspection conditions and defect conditions. For this reason, it is not possible to uniquely associate the defect detection signal 511 and the defect detection time 512 with each defect condition.

[0035] (Defect condition probability distribution estimation) Next, the process performed by the probability distribution estimation unit 104 will be described with reference to Fig. 9 and Fig. 10. Fig. 1 and Fig. 8 will be referred to as appropriate. FIG. 9 is a conceptual diagram showing a method for estimating the defect condition probability distribution 611 related to the defect size by the probability distribution estimation unit 104 used in this embodiment. In graph 600, the horizontal axis indicates defect size, and the vertical axis indicates signal intensity. Note that the horizontal and vertical axes of graph 600 are logarithmic. Plot points 601 in graph 600 are analysis results by simulation processing unit 102. That is, plot points 601 are the results of physical simulation analysis performed by random sampling according to a probability distribution representing defect conditions (in the example shown in FIG. 9, defect size variation 211 (see FIGS. 6A to 6C)). The probability distribution representing defect size variation 211 is graph 321 shown in FIG. 5.

[0036] In this manner, the simulation processing unit 102 simulates the signal intensity for the inspection conditions that are factors in the variation of the defect detection signal 511 by non-destructive inspection, and the defect conditions that are the conditions of the defect D that is the target of the non-destructive inspection. Specifically, the simulation processing unit 102 calculates the signal intensity corresponding to the inspection conditions and the defect conditions by performing a physical simulation. The simulation processing unit 102 simulates (calculates) the signal intensity for the inspection conditions and the defect conditions by referring to the graph 321 in Fig. 5. The results of the simulation by the simulation processing unit 102, that is, the signal intensity corresponding to the inspection conditions and the defect conditions, are plotted as plot points 601.

[0037] Moreover, a regression line 602 is a result of applying the least squares method to the plot points 601. Moreover, a horizontal dashed line 603 shown in the graph 600 corresponds to the defect detection signal intensity 513 of the defect detection signal 511 shown in Fig. 8. The plot points 601 and the regression line 602 shown in Fig. 9 indicate that the signal intensity originating from the defect D increases as the defect size increases.

[0038] 9, the probability distribution estimation unit 104 calculates a defect condition probability distribution 611, which is a probability distribution related to defect sizes (probability distribution related to defect conditions). The distribution shape of the defect condition probability distribution 611 is assumed to be a Weibull distribution or the like. The Weibull distribution includes normal distribution and log-normal distribution. The probability distribution estimation unit 104 calculates the defect condition probability distribution 611 by fitting parameters of the defect condition probability distribution 611 by a maximum likelihood estimation method or the like so as to best reproduce the plot points 601. The parameters are an average, a variance, or the like. Alternatively, the probability distribution estimation unit 104 may calculate the defect condition probability distribution 611 as an empirical distribution by converting the results of the plot points 601 into a histogram.

[0039] In this embodiment, the probability distribution estimation unit 104 calculates a histogram of plot points 601 that fall on a horizontal dashed line 603 (more precisely, in the vicinity of the horizontal dashed line 603) that corresponds to the defect detection signal intensity 513 obtained by the actual inspection. Specifically, the probability distribution estimation unit 104 divides the area in the vicinity of the horizontal dashed line 603 into a grid of a predetermined interval, and expresses the number of plot points 601 that fall within each grid as a histogram. The probability distribution estimation unit 104 sets the calculated histogram as a defect condition probability distribution 611. In the example shown in FIG. 9, the number of plot points 601 is small to make the figure easier to see, but in reality, a large number of plot points 601 exist.

[0040] In the calculated defect condition probability distribution 611, the average of the defect size indicated by the plot points 601 on (the vicinity of) the horizontal dashed line 603 coincides with the average of the defect condition probability distribution 611. In this embodiment, the defect size corresponding to the point where the regression line 602 and the horizontal dashed line 603 intersect becomes the average of the defect condition probability distribution 611. In addition, the calculated defect condition probability distribution 611 has a spread corresponding to the inspection condition (defect size variation 211 in the example shown in FIG. 9). In this embodiment, the degree of variation (standard deviation) of the plot points 601 on (the vicinity of) the horizontal dashed line 603 depends on the inspection condition (defect size variation 211 in the example shown in FIG. 9). In addition, the vertical dashed line in FIG. 9 indicates 2σ of the defect condition probability distribution 611, and indicates the defect size variation 211.

[0041] The defect condition probability distribution 611 thus obtained is input to the remaining life prediction unit 105 as an initial distribution.

[0042] 9 shows an example in which a defect condition probability distribution 611 is calculated using the variation of plot points 601 as defect size variation 211. The actual response signal 501 (see FIG. 8) varies due to the defect conditions shown in (C1) to (C2) below, in addition to the defect size variation 211. (C1) Defect shape. (C2) Defect direction variation 212 (see FIG. 6).

[0043] Therefore, it is necessary to take into consideration that remaining life prediction becomes uncertain due to defect conditions other than the defect size variation 211. That is, for each of the defect shape and defect direction variation 212 in addition to the defect size variation 211, the probability distribution estimation unit 104 performs the same process as that shown in Fig. 9. As a result, a defect condition probability distribution 611 for each of the defect conditions such as the defect shape and the defect direction variation 212 is calculated.

[0044] 9 is Bayesian estimation in which the distribution of plot points 601 is used as a prior distribution and a defect condition probability distribution 611 is estimated as a posterior distribution when defect detection signal intensity 513 is given as data. That is, the probability distribution estimation unit 104 estimates a probability distribution related to defect conditions by Bayesian estimation based on the result of the physical simulation and the input signal intensity acquired by the actual inspection. As described above, the result of the physical simulation corresponds to the plot points 601, and the input signal intensity acquired by the actual inspection corresponds to the defect detection signal intensity 513. In this way, the estimation accuracy of the defect condition probability distribution 611 can be improved.

[0045] In this way, the probability distribution estimation unit 104 estimates the defect condition probability distribution 611, which is the probability density of the defect condition estimated from the actual signal intensity. At this time, the probability distribution estimation unit 104 estimates the defect condition probability distribution 611 based on the analysis result by the simulation processing unit 102 and the actual signal intensity (defect detection signal intensity 513), which is the signal intensity of the response signal obtained in the actual inspection.

[0046] FIG. 10 is a conceptual diagram showing a probability distribution estimation result by the probability distribution estimation unit 104 according to this embodiment, taking into account defect size variation 211 (see FIGS. 6A to 6C) and defect direction variation 212 (see FIGS. 6A to 6C). 10 shows a multidimensional defect condition probability distribution 611A in which the defect direction is added to the random variables in addition to the defect size as a representative example. That is, the multidimensional defect condition probability distribution 611A shown in FIG. 10 is a defect condition probability distribution 611 shown in FIG. 9 in which the defect direction is added as a random variable. In FIG. 10, one horizontal axis shows the defect size, the other horizontal axis shows the defect direction, and the vertical axis shows the probability. That is, FIG. 10 shows a result of combining the process (Bayesian estimation) shown in FIG. 9 by the probability distribution estimation unit 104 for each of the defect size and the defect direction. In the example shown in FIG. 10, the multidimensional defect condition probability distribution 611A shows an example in which the defect size and the defect direction are independent, but the defect size and the defect direction may be correlated.

[0047] The probability distribution estimation unit 104 assumes a distribution shape of a multidimensional defect condition probability distribution 611A shown in Fig. 10 for the analysis result of the simulation processing unit 102. Then, the probability distribution estimation unit 104 fits parameters so as to best reproduce the plot points 601 (see Fig. 9) by a maximum likelihood estimation method or the like. Alternatively, the probability distribution estimation unit 104 converts the results of the plot points 601 for each defect size and defect direction into a histogram as shown in Fig. 9. This makes it possible for the probability distribution estimation unit 104 to calculate the multidimensional defect condition probability distribution 611A shown in Fig. 10 as an empirical distribution.

[0048] In the calculated multidimensional defect condition probability distribution 611A, the average of the random variables for the physical simulation results performed for the defect size and defect direction coincides with the average of the random variables in the multidimensional defect condition probability distribution 611A. The physical simulation results correspond to plot points 601 shown in FIG. 9 plotted against the defect size and defect direction. The multidimensional defect condition probability distribution 611A has a spread corresponding to the defect size variation 211 and the defect direction variation 212. Based on the multidimensional defect condition probability distribution 611A thus obtained, the remaining life prediction unit 105 calculates the growth of the defect D and evaluates the risk.

[0049] 10 shows a multidimensional defect condition probability distribution 611A having the defect size and defect direction as random variables. A multidimensional defect condition probability distribution 611A having the defect position and the like as random variables in addition to the defect size and defect direction may be estimated.

[0050] The multidimensional defect condition probability distribution 611A also takes into account the defect shape. As described above, the defect shape may be specified by the user, but may be processed probabilistically as follows. Since the defect shape is a discrete value, the probability of occurrence of each of the planar defect model DMa, the spherical defect model DMb, and the cylindrical defect model DMc shown in FIG. 7A to FIG. 7C is calculated. Then, the probability distribution estimation unit 104 associates the multidimensional defect condition probability distribution 611A with each of the planar, spherical, and cylindrical defect models DMa. That is, for example, if the probability of occurrence of a planar defect is "0.4", the multidimensional defect condition probability distribution 611A associated with the planar defect model DMa occurs with a probability of "0.4".

[0051] It should be noted that the defect condition probability distribution 611 shown in FIG. 9 is a special form of the defect condition probability distribution 611A shown in FIG.

[0052] (Remaining life prediction process) FIG. 11 is a conceptual diagram showing the change in defect condition probability distribution 611 over time and a remaining life prediction method. In the graph shown in Fig. 11, the horizontal axis represents defect size and the vertical axis represents probability. The defect condition probability distribution 611 is the defect condition probability distribution 611 for defect size estimated by the probability distribution estimation unit 104 in Fig. 9 (or the multidimensional defect condition probability distribution 611A shown in Fig. 10). For ease of explanation, the following explanation will be given using the defect condition probability distribution 611 for defect size shown in Fig. 9.

[0053] Curve 612 is a probability distribution regarding defect conditions after t time has elapsed (defect size in the example shown in FIG. 11). In FIG. 11, the outlined arrows indicate the passage of time. A specific calculation procedure for the probability distribution regarding defect conditions after t time has elapsed will be described later. Curve 621 is a destruction probability distribution of the inspection object T with respect to the defect size. The destruction probability indicates the destruction probability of the inspection object T with respect to a certain defect size. For example, if the defect size is "x1", the inspection object T will be destroyed with a probability of "p1" calculated from the destruction probability distribution. The curve 621 is obtained by a calculation method such as the finite element method (FEM).

[0054] A specific calculation procedure for the curve 612 representing the probability distribution regarding the defect condition after the lapse of time t will be described below. In general, it can be considered that a defect D grows in response to repeated stress according to the following formula (1).

[0055]

number

[0056] In equation (1), "a" is the defect size, "t" is the elapsed time, and "α", "β", "γ", and ..." are defect conditions other than the defect size. The growth rate function "f" represents the growth rate of defect D. The growth rate function "f" is calculated using a calculation method such as the finite element method.

[0057] The growth rate function “f” of the defect D has uncertainties in its parameters, and probabilistically corresponds to the stochastic differential equation shown in the following equation (2).

[0058] da = μ(a,t)dt + σ(a,t)dWt (2)

[0059] In equation (2), μ(a,t) is the mean value of the right-hand side of equation (1), σ(a,t) is the standard deviation of the right-hand side of equation (1), and Wt is the standard Wiener process.

[0060] If the change in defect size follows equation (2), the time evolution of the defect condition probability distribution 611 ("p") on the defect size is expressed by the Fokker-Planck equation, which can be expressed as the following equation (3) when the Itō integral is used.

[0061]

number

[0062] When the Stratonovich integral is used, the Fokker-Planck equations are expressed as follows:

[0063]

number

[0064] In this embodiment, any of the integration methods may be used. Note that in the Fokker-Planck equation, the random variable (the defect condition) must be a continuous random variable. The remaining life prediction unit 105 calculates the probability distribution (curve 612) of the defect size after t time has elapsed by using formula (3) or formula (4). The remaining life prediction unit 105 calculates the probability distribution (curve 612) of the defect size after t time has elapsed by numerical calculation based on formula (3) or formula (4). In this way, the remaining life prediction unit 105 calculates the time change of the probability density of the defect condition estimated by the probability distribution estimation unit 104. The calculated probability distribution (curve 612) of the defect size after t time has elapsed is the probability distribution of the defect condition after a predetermined time. In particular, in the first embodiment, the remaining life prediction unit 105 analytically calculates the probability distribution (curve 612) over time by solving the Fokker-Planck equation. At this time, the defect condition probability distribution 611 (see FIG. 9) or the multidimensional defect condition probability distribution 611A (see FIG. 10) estimated by the probability distribution estimation unit 104 is set as the initial distribution.

[0065] 11 represents the fracture probability of the inspection object T after t hours have elapsed. The remaining life prediction unit 105 compares the tolerable risk with the fracture probability after t hours have elapsed, and calculates the time t at which the fracture probability after t hours has elapsed satisfies a predetermined condition with respect to the tolerable risk as the remaining life. The predetermined condition is one of the following two: (B1) The probability of failure after time t becomes greater than the tolerable risk. (B2) The probability of failure after t time has elapsed falls within the range of the tolerable risk ±δ (δ is a predetermined value). The remaining life prediction unit 105 determines the timing of maintenance and repair using the calculated remaining life of the defect D. Note that the remaining life may be calculated for each of the multiple defects D. In this case, the remaining life prediction unit 105 determines the timing of maintenance and repair using the calculated remaining life of the multiple locations. For example, the remaining life prediction unit 105 determines the timing of maintenance and repair based on the shortest remaining life of the multiple locations.

[0066] In this way, the remaining life prediction unit 105 estimates the time when the overlapping portion of the probability distribution regarding the defect condition after a predetermined time and the preset fracture probability distribution of the inspection object T satisfies the predetermined condition as the remaining life (predicts the remaining life). Here, the probability distribution regarding the defect condition after a predetermined time is curve 612 in FIG. 11. Also, the preset fracture probability distribution of the inspection object T is curve 621 in FIG. 11. And the overlapping portion of the preset fracture probability distribution of the inspection object T is region 631 in FIG. 11.

[0067] In Fig. 11, the remaining life prediction process is performed for the defect size. However, as described above, the defect condition probability distribution 611 is actually a multidimensional defect condition probability distribution 611A having a plurality of defect conditions as random variables as shown in Fig. 10. Therefore, the curves 612 and 621 shown in Fig. 11 are also multidimensional probability distributions having the same random variables as the multidimensional defect condition probability distribution 611A. Then, the remaining life prediction unit 105 calculates a multidimensional probability distribution corresponding to the curve 612 after t time has elapsed for random variables (defect size, defect direction, etc.) that affect the destruction among these random variables. Then, the remaining life prediction unit 105 calculates the time t at which the volume of a manifold region (region 631) surrounded by the multidimensional probability distribution corresponding to the curve 612 and the multidimensional probability distribution corresponding to the curve 621 satisfies a predetermined condition as the remaining life.

[0068] (flowchart) Finally, the procedure of the remaining life prediction method according to the first embodiment will be described with reference to FIG. 12 is a flowchart showing the procedure of the remaining life prediction method according to the first embodiment, with reference to FIGS. First, in step S11, defect conditions, inspection conditions, material conditions, and an allowable risk, which is an allowable probability of breakage, are input to the condition input unit 101. Step S11 corresponds to Fig. 3. As described above, the defect conditions, inspection conditions, and material conditions are input as probability distributions for each item of variation.

[0069] Next, in step S12, which is a simulation step, the simulation processing unit 102 calculates the correspondence of the signal intensity to the defect conditions, the inspection conditions, and the material conditions. The process of step S12 is the analysis by the simulation processing unit 102 described with reference to FIGS. 4A to 7C.

[0070] That is, in step S12, the simulation processing unit 102 calculates the dependency of the signal intensity for each defect size in response to the variation 201 in the installation position of the ultrasonic sensor 121 and material conditions such as the sound speed of the inspection target T. At this time, the simulation processing unit 102 uses a physical simulation or a test database. As a result, a graph 321 shown in FIG. 5 is obtained.

[0071] In step S13, a response signal 501 obtained by an actual inspection is input to the response signal input unit 103. The process of step S13 has been described with reference to Fig. 8. In step S13, feature quantities according to the defect conditions, such as a defect detection signal 511 and a defect detection time 512, are extracted by signal processing.

[0072] In step S14, the probability distribution estimation unit 104 estimates the defect condition probability distribution 611 as a probability distribution estimation step. Step S14 is the process described with reference to Figs. 9 and 10. The probability distribution estimation unit 104 estimates the defect condition probability distribution 611 based on the defect conditions, specifically, the defect direction, defect size, shape, etc., and the defect detection signal intensity 513 of the defect detection signal 511. In step S14, first, the probability distribution estimation unit 104 assumes a distribution shape of the defect condition probability distribution 611 as shown in Fig. 9 for the analysis result in step S12. Then, the probability distribution estimation unit 104 fits parameters so as to best reproduce the plot points 601 shown in Fig. 9 by a maximum likelihood estimation method or the like. Alternatively, the probability distribution estimation unit 104 converts the results of the plot points 601 into a histogram to calculate the defect condition probability distribution 611 as an empirical distribution. Note that the defect condition probability distribution 611 estimated in step S14 is actually a multidimensional defect condition probability distribution 611A shown in FIG.

[0073] The defect condition probability distribution 611 calculated in step S14 has the following properties. With reference to the example shown in Fig. 9, the average of the random variables indicated by the plot points 601 on (the vicinity of) the horizontal dashed line 603 matches the average of the random variables of the defect condition probability distribution 611. Furthermore, the defect condition probability distribution 611 (multidimensional defect condition probability distribution 611A) calculated in step S14 has a distribution spread corresponding to the defect conditions. Based on the defect condition probability distribution 611 thus obtained, in step S15, the remaining life prediction unit 105 calculates the growth of the defect D and evaluates the risk.

[0074] In step S15, the remaining life prediction unit 105 performs analytical remaining life prediction as a remaining life prediction step. Step S15 is the process described with reference to FIG. 11. As described in the explanation of the remaining life prediction unit 105 shown in FIG. 11, the defect condition probability distribution 611 is used as an initial distribution, and the Fokker-Planck equation corresponding to the inspection system is used. As a result, the remaining life prediction unit 105 calculates the probability distribution of the defect size after t time has elapsed (curve 612 in the example shown in FIG. 11) by using the defect condition probability distribution 611 estimated by the probability distribution estimation unit 104 as an initial distribution. Furthermore, the remaining life prediction unit 105 calculates in advance the fracture probability (curve 612) of the inspection object T for the defect size "a" in FIG. 11 by a calculation method such as FEM. Note that "analytical" means analytically obtaining the probability distribution (curve 612) regarding the defect size after t time has elapsed using the Fokker-Planck equation.

[0075] As described above, the remaining life prediction unit 105 calculates the fracture probability of the inspection object T after t time has elapsed by calculating the area of ​​the region 631 surrounded by the curve 612 and the curve 621 in FIG. 11. When the defect condition probability distribution 611 is a multidimensional defect condition probability distribution 611A, the above area is the volume of a manifold region surrounded by the multidimensional probability distribution corresponding to the curve 612 and the multidimensional probability distribution corresponding to the curve 621, as described above. Then, the remaining life prediction unit 105 compares the tolerable risk with the fracture probability after t time has elapsed, as described above, and calculates the time "t" that satisfies a predetermined condition as the remaining life. The predetermined condition is either (B1) or (B2) described in the description of FIG. 11.

[0076] In step S16, a remaining life prediction step, the remaining life prediction unit 105 displays information about the calculated (estimated) remaining life on the output device 115. In this way, the calculated remaining life is presented to the user. The user uses the calculated remaining life of multiple locations to determine the timing of maintenance and repair.

[0077] (Display screen example) FIG. 13 is a diagram showing an example of a display screen displayed in step S16 of FIG. On the display screen 701, a display is provided that urges repair within Y days as information regarding remaining life. The value of "Y" is a value calculated based on the time "t" that satisfies the predetermined condition calculated in step S15 of FIG. In this way, it is possible to make a reasonable remaining life prediction that takes into account the inspection conditions and defect conditions, and it is possible to rationalize the maintenance and repair timing. As a result, it is possible to make a highly accurate remaining life prediction.

[0078] In the first embodiment, the remaining life prediction unit 105 calculates the probability distribution after t time has elapsed (curve 612 in FIG. 11) by analytically solving the Fokker-Planck equation. In this way, the probability distribution after t time has elapsed (curve 612) can be calculated with high accuracy.

[0079] Furthermore, when calculating the defect condition probability distribution 611 shown in Fig. 9 or the multidimensional defect condition probability distribution 611A shown in Fig. 10, the probability distribution estimation unit 104 sets the plot points 601 as a prior distribution. Then, the probability distribution estimation unit 104 performs Bayesian estimation using the defect detection signal intensity 513 as data. In this way, it is possible to probabilistically calculate the defect condition probability distribution 611 or the multidimensional defect condition probability distribution 611A.

[0080] In the first embodiment, as shown in Fig. 3, installation variation 201 of ultrasonic sensor 121 and variation 202 in ultrasonic transmission direction are set as inspection conditions. Furthermore, as shown in Figs. 6A to 6C, defect size variation 211, defect direction variation 212, and defect position variation 213 are set as defect conditions. In addition, defect shape is included as a defect condition. In this way, it is possible to predict remaining life taking into account actual inspection and variations that may occur due to defects.

[0081] [Second embodiment] FIG. 14 is a flowchart showing the procedure of the remaining life prediction method according to the second embodiment. In the first embodiment, a curve 612 (see FIG. 11) corresponding to the probability distribution after time t is calculated by analytically solving the Fokker-Planck equation in step S15 in FIG. 12. In contrast to this, in the second embodiment, the remaining life prediction unit 105 executes remaining life prediction by Monte Carlo (MC) calculation in step S15A in FIG. 14.

[0082] That is, in step S15A of FIG. 14, the remaining life prediction unit 105 performs a Monte Carlo calculation instead of the calculation of the Fokker-Planck equation. This makes it possible to predict the remaining life probabilistically. Specifically, the following process is performed. First, the sampling number for the Monte Carlo calculation is input to the remaining life prediction unit 105. Then, the remaining life prediction unit 105 randomly samples the defect conditions (α, β, γ,... in formula (1)) according to the probability distribution obtained by the probability distribution estimation unit 104. The probability distribution obtained by the probability distribution estimation unit 104 is the defect condition probability distribution 611 (see FIG. 9) or the multidimensional defect condition probability distribution 611A (see FIG. 10).

[0083] Next, the remaining life prediction unit 105 randomly samples the growth rate of defect D according to a probability distribution (usually a normal distribution with mean "μ(a,t)" and standard deviation "σ(a,t) ("a" is the defect size and "t" is time)). Then, the remaining life prediction unit 105 calculates the defect size after Δt time based on the randomly sampled growth rate of defect D.

[0084] The remaining life prediction unit 105 repeats this calculation to obtain a curve 612 (see FIG. 11) that represents the probability distribution of defect size after t time has elapsed. Then, the allowable risk is compared with the probability of failure after t time has elapsed, and the time t that satisfies a predetermined condition is calculated as the remaining life. The predetermined condition is either (B1) or (B2) described in the explanation of FIG. 11.

[0085] In this way, in the second embodiment, the probability distribution over time is calculated by the Monte Carlo calculation by repeating random sampling from the initial distribution to the defect condition. At this time, the probability distribution estimated by the probability distribution estimation unit 104 (defect condition probability distribution 611 (see FIG. 9) or multidimensional defect condition probability distribution 611A (see FIG. 10)) is set as the initial distribution.

[0086] In step S16, the remaining life prediction unit 105 displays the calculated remaining life on the output device 115. Note that the remaining life may be calculated for each of the multiple defects D. In this case, the remaining life prediction unit 105 uses the calculated remaining life of the multiple locations to determine the maintenance and repair timings. For example, the remaining life prediction unit 105 determines the maintenance and repair timings based on the shortest remaining life of the multiple locations. Note that the processing of step S16 is the same as that of the first embodiment.

[0087] As described above, in the Fokker-Planck equation in the first embodiment, the random variables must be continuous random variables. In contrast, in the Monte Carlo calculation in the second embodiment, the random variables may be either continuous or discrete.

[0088] According to the second embodiment, the probability distribution after t time has elapsed (curve 612 in FIG. 11) is calculated by a simulation using Monte Carlo calculations, rather than analytically calculated as in the first embodiment. This makes it possible to calculate the probability distribution after t time has elapsed (curve 612) without advanced mathematical knowledge.

[0089] In addition, the technology described in this embodiment is applicable to maintenance services for inspection objects T and plant operation plan support services.

[0090] The present invention is not limited to the above-mentioned examples, and includes various modified examples. For example, the above-mentioned examples have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. In addition, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. In addition, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

[0091] In addition, the above-mentioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as programs, tables, files, etc. that realize each function can be stored in memory 111, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

[0092] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0093] 1. Remaining life prediction device 101 Condition input section 102 Simulation processing section 103 Response signal input section 104 Probability distribution estimation unit 105 Remaining Life Prediction Department 115 Output device (output section) 121 Ultrasonic Sensor 121M Ultrasonic Sensor Model 201 Installation position variation (variation in ultrasonic sensor installation position, inspection conditions) 202 Transmission direction variation (variation in ultrasonic transmission direction, inspection conditions) 211 Defect size variation (variation of defect size, defect conditions, continuous random variables) 212 Defect orientation variation (variation of defect position, defect condition, continuous random variable) 213 Defect location variation (variation of defect direction, defect conditions, continuous random variables) 321 Graph 401 Subject Model 501 Response signal 511 Defect detection signal 513 Defect detection signal strength (actual signal strength) 600 graphs 601 Plot points (analysis results by the simulation processing unit, signal strength corresponding to inspection conditions and defect conditions) 603 Horizontal dashed line (corresponding to actual signal strength) 611 Defect condition probability distribution (Probability distribution for defect conditions) 611A Defective condition probability distribution (probability distribution for defective conditions) 612 curve (probability distribution for said defect condition after a given time) 621 curve (fracture probability distribution) 631 area (overlapping part) D. Defect DM Defect Model DMa defect model (variation of defect shape, defect conditions) DMb defect model (variation of defect shape, defect conditions) DMc defect model (variation of defect shape, defect conditions) T Inspection object S12 Calculation of signal strength correspondence to defect conditions, inspection conditions, and material conditions (simulation step) S14: Estimate the defect condition probability distribution (probability distribution estimation step) S15 Analytical remaining life prediction (remaining life prediction step) S16 Display of remaining life information (remaining life prediction step)

Claims

1. a simulation processing unit that simulates a signal intensity for an inspection condition that is a factor of variation in a defect detection signal by non-destructive inspection and a defect condition that is a condition of a defect that is a target of the non-destructive inspection; a probability distribution estimation unit that estimates a probability distribution regarding the defect condition estimated from an actual signal intensity based on an analysis result by the simulation processing unit and an actual signal intensity, which is a signal intensity of a response signal obtained in an actual inspection; and a remaining life prediction unit that calculates a time change in a probability distribution related to the defect condition estimated by the probability distribution estimation unit, estimates a time when an overlapping portion of the probability distribution related to the defect condition after a predetermined time and a preset fracture probability distribution of an inspection object satisfies a predetermined condition as a remaining life, and outputs the estimated remaining life to an output unit; A remaining life prediction device having the above structure.

2. The non-destructive testing uses an ultrasonic sensor, The inspection conditions include any one of a variation in the installation position of the ultrasonic sensor and a variation in the transmission direction of the ultrasonic wave.

2. The remaining life prediction device according to claim 1 .

3. The defect condition includes a variation in defect size, and includes any one of a variation in defect position, a variation in defect direction, and a variation in defect shape.

2. The remaining life prediction device according to claim 1 .

4. the simulation processing unit calculates signal strengths corresponding to the inspection conditions and the defect conditions by performing a physical simulation; The probability distribution estimation unit estimates a probability distribution regarding the defect condition by Bayesian estimation based on the result of the physical simulation and the input signal intensity acquired by the actual inspection.

2. The remaining life prediction device according to claim 1 .

5. the fault condition is a continuous random variable; The remaining life prediction unit is The probability distribution regarding the inspection condition estimated by the probability distribution estimation unit is used as an initial distribution, and a Fokker-Planck equation is solved to analytically calculate a probability distribution over time.

2. The remaining life prediction device according to claim 1 .

6. The remaining life prediction unit is The probability distribution estimated by the probability distribution estimation unit is used as an initial distribution, and random sampling is repeated from the initial distribution for the defect condition to calculate a probability distribution over time by Monte Carlo calculation.

2. The remaining life prediction device according to claim 1 .

7. A remaining life prediction device for predicting the remaining life of an inspection object, a simulation step of simulating a signal intensity for an inspection condition that is a cause of variation in a defect detection signal by non-destructive inspection and a defect condition that is a condition of a defect that is a target of the non-destructive inspection and occurs in the inspection object; a probability distribution estimation step of estimating a probability distribution regarding the defect condition estimated from the actual signal intensity based on an analysis result of the simulation step and an actual signal intensity, which is the signal intensity of a response signal obtained in an actual inspection; a remaining life prediction step of calculating a time change in a probability distribution related to the defect condition estimated by the probability distribution estimation step, estimating a time when an overlapping portion of the probability distribution related to the defect condition after a predetermined time and a preset fracture probability distribution of the inspection object satisfies a predetermined condition as a remaining life, and outputting the estimated remaining life to an output unit; A remaining life prediction method.

Images