Information processing device, information processing method and program

Abstract

To detect a physical quantity in the circumference of an object to be measured with higher accuracy.SOLUTION: An information processing device comprises a control part which acquires an array measured value, as a measured value of a physical quantity, measured by a sensor array having a plurality of sensors arrayed in two dimensions, acquires an estimated value of a true physical quantity through filtering for removing noise from the acquired array measured value, and outputs information related to a state of an object to be measured on the basis of the acquired estimated value, and the control part performs the filtering while eliminating an influence of sensor noise as noise based upon individual differences of the plurality of sensors, on the basis of the magnitude of the difference in the measurement values between the adjacent sensors.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Patent Document 1 describes sensing the magnetic field around a measurement object that is a ferromagnetic material, and detecting an abnormality due to thinning of the measurement object. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-163161 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the conventional configuration leaves room for improvement in the accuracy of detecting physical quantities around the object to be measured.

[0005] Therefore, an object of the present disclosure is to enable detection of physical quantities around a measurement target with higher accuracy. [Means for solving the problem]

[0006] An information processing device according to some embodiments includes: (1) acquiring array measurements, which are measurements of physical quantities, measured by a sensor array in which multiple sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain estimates of the true physical quantities; outputting information about the state of the measurement object based on the acquired estimated value; A control unit is provided, The control unit performing the filtering by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other; Information processing device.

[0007] In this way, the information processing device performs filtering based on the difference in the magnitude of the difference between the sensor noise and the true signal to be estimated between the measurement values ​​of adjacent sensors. Therefore, the information processing device can perform highly accurate filtering by removing bias due to individual differences between sensors.

[0008] In one embodiment, (2) In the information processing device of (1), The control unit acquiring a plurality of array measurements measured by the sensor array at a plurality of times; The parameters (α, β, γ) that define the statistical distribution of the sensor noise (μ) are calculated by dividing the T array measurements (g1,..., g T ) obtained by maximum likelihood method based on The filtering may be performed using the obtained parameters (α, β, γ).

[0009] In this way, the information processing device estimates the parameters of the statistical distribution of sensor noise by the maximum likelihood method based on array measurement values ​​acquired at multiple times, and performs filtering using these parameters. Therefore, the information processing device can perform highly accurate filtering based on measurement values ​​acquired at multiple times.

[0010] In one embodiment, (3) In the information processing device of (2), The control unit estimating temporal noise based on the plurality of array measurements, the temporal noise being noise due to variations in measurements at each of the plurality of sensors; The estimated temporal noise may also be used to perform the filtering.

[0011] In this way, the information processing device estimates the time noise and performs filtering, so that it is possible to remove the influence of variations in the measurement values ​​and perform filtering with high accuracy.

[0012] In one embodiment, (4) In the information processing device of (2) or (3), The control unit calculating the difference between each of the plurality of array measurements and the estimate to obtain a plurality of residuals; obtaining an evaluation index indicating the appropriateness of the parameter based on the magnitude of differences between the plurality of residual values ​​between adjacent sensors among the plurality of sensors; The parameter may be updated based on the evaluation index.

[0013] In this way, the information processing device updates the parameter based on the evaluation index indicating the appropriateness of the parameter, and therefore, it is possible to perform highly accurate filtering based on a more appropriate parameter.

[0014] In one embodiment, (5) In the information processing device of (4), The control unit may obtain, as the evaluation index, a parameter that determines a statistical distribution of common components included in the plurality of residuals by a maximum likelihood method based on the plurality of residuals.

[0015] In this way, the information processing device estimates the parameter of the statistical distribution of the common component contained in the residuals by the maximum likelihood method based on the residuals, and uses this parameter as an evaluation index. Therefore, the information processing device can accurately identify the statistical distribution of the sensor noise and perform highly accurate filtering.

[0016] In one embodiment, (6) In the information processing device according to (1) to (5), The control unit As the array measurement value, a measurement value of magnetic flux density around the measurement target, which is a magnetic body, is acquired; The presence or absence of thinning in the object to be measured and the depth of the thinning may be output as information regarding the state of the object to be measured.

[0017] In this way, the information processing device can output with high accuracy the presence or absence of thinning in the measurement object and the depth of the thinning, based on the measured value of the magnetic flux density around the measurement object.

[0018] An information processing method according to some embodiments includes: (7) An information processing method for an information processing device having a control unit, The control unit Acquiring an array measurement value, which is a measurement value of a physical quantity, measured by a sensor array in which a plurality of sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain an estimate of the true physical quantity; outputting information about the state of the object to be measured based on the obtained estimated value; Including, The control unit The filtering is performed by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other.

[0019] In this way, the information processing method performs filtering based on the difference in the magnitude of the difference between the sensor noise and the true signal to be estimated between the measurement values ​​of adjacent sensors. Therefore, the information processing method makes it possible to perform highly accurate filtering by removing bias due to individual differences between sensors.

[0020] In some embodiments, the program (8) To the computer, acquiring array measurements, which are measurements of physical quantities, measured by a sensor array in which a plurality of sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain an estimate of the true physical quantity; outputting information about the state of the object to be measured based on the obtained estimated value; Execute a procedure that includes The filtering is performed by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other.

[0021] In this way, the computer operating based on the program performs filtering based on the difference in the magnitude of the difference between the sensor noise and the true signal to be estimated between the measurement values ​​of adjacent sensors. Therefore, the program makes it possible to perform highly accurate filtering by removing bias due to individual differences between sensors. [Effects of the Invention]

[0022] According to an embodiment of the present disclosure, it is possible to detect a physical quantity around a measurement target with higher accuracy. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is a diagram illustrating a configuration of a measurement system according to an embodiment. [Figure 2] FIG. 1 is a schematic diagram showing an example of a state in which a sensor array is arranged on a measurement object. [Figure 3] FIG. 2 is a diagram schematically showing the distribution of magnetic flux density around a measurement object. [Figure 4A] FIG. 10 is a diagram illustrating an example of the X component of magnetic flux density. [Figure 4B] FIG. 10 is a diagram illustrating an example of the Y component of magnetic flux density. [Figure 4C] FIG. 10 is a diagram showing an example of the Z component of magnetic flux density. [Figure 5] 2 is a flowchart showing an example of the operation of the management device of FIG. 1; [Figure 6] FIG. 1 is a diagram showing a schematic diagram of the types of signals included in array measurement values. [Figure 7] 10 is a diagram illustrating a process flow for updating parameters used in filtering. DETAILED DESCRIPTION OF THE INVENTION

[0024] <Comparative Example> The comparative example uses a sensor array consisting of multiple sensors arranged in an array to observe leakage magnetic flux from carbon steel pipes, thereby non-destructively monitoring the volumetric loss of the pipe due to corrosion. Unlike image sensors in which many photoelectric conversion elements are densely arranged, the sensors that make up the sensor array are arranged at a certain distance from each other.

[0025] The first method according to the comparative example measures the depth of wall thinning by passing an alternating current through a carbon steel pipe and observing the leakage magnetic flux generated from the pipe surface. The leakage magnetic flux generated from the surface of a carbon steel pipe through which an alternating current flows changes over time. The first method measures the time-varying leakage magnetic flux using a sensor array installed around the pipe. The first method filters the measured values ​​of each sensor based on the frequency characteristics of the fluctuations over time to remove noise from the measured values.

[0026] The second method according to the comparative example involves magnetizing a carbon steel pipe with a permanent magnet, thereby generating leakage magnetic flux from the pipe without passing a current. When corrosion occurs in the magnetized pipe and the volume decreases, leakage magnetic flux is generated from the surface of the pipe. The second method detects this leakage magnetic flux with a magnetic sensor array to detect the presence and depth of thinning.

[0027] The first method according to the comparative example filters the measured values ​​for each sensor in the sensor array, but does not consider the discrepancy in the measured values ​​between sensors due to individual differences between sensors. As a result, the measured values ​​obtained by the first method may contain spatial noise, leaving room for improvement in the accuracy of magnetic flux density measurements.

[0028] In the first and second methods, the data measured by the sensor array is provided as a two-dimensional array. To reduce noise in such data, a spatial filter, used in image processing, can be applied. However, because multiple sensors are spatially spaced apart in the sensor array, applying a spatial filter causes aliasing of noise components at spatial frequencies, resulting in the coexistence of the signal component of the magnetic flux density to be extracted and the noise component at spatial frequencies. Therefore, it is difficult to apply a spatial filter to measurements obtained by a coarse sensor array with a certain distance between sensors. Therefore, applying a spatial filter to the first and second methods leaves room for improvement in the accuracy of magnetic flux density measurements.

[0029] As described above, in the method according to the comparative example, there is room for improvement in the measurement accuracy of a physical quantity when using a sensor array consisting of multiple sensors arranged at regular intervals. The present disclosure aims to improve the measurement accuracy of a physical quantity by removing noise due to individual differences between multiple sensors through simple calculations.

[0030] <Embodiment> Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. In each drawing, parts having the same configuration or function are denoted by the same reference numerals. In the description of this embodiment, duplicated descriptions of the same parts may be omitted or simplified as appropriate.

[0031] (Measurement System 1) 1 is a diagram showing the configuration of a measurement system 1 according to one embodiment. The measurement system 1 includes a measurement device 10 and a management device 20. The measurement device 10 and the management device 20 can communicate with each other via a network 30. The network 30 is any communication medium and may include the Internet, an intranet, a dedicated network, etc., or a combination thereof.

[0032] The measuring device 10 measures physical quantities around a measurement object 80 (see FIG. 2 ) using a sensor array 12 in which a plurality of sensors 121 are arranged two-dimensionally (in an array). In this embodiment, as an example, the measurement object 80 is a magnetized carbon steel pipe, and the sensor array 12 measures the magnetic flux density around the measurement object 80, but the type of measurement object 80 and the type of physical quantity to be measured are arbitrary. The measuring device 10 transmits the acquired measurement values ​​of the physical quantities as array measurement values ​​to the management device 20 via the network 30.

[0033] The management device 20 as an information processing device according to this embodiment performs filtering to remove noise from the array measurement values ​​received from the measurement device 10 to obtain estimates of true physical quantities, and based on these estimates, outputs information about the state of the measurement object 80. The state of the measurement object 80 may include, for example, the presence or absence of thinning in the measurement object 80 and the depth of the thinning.

[0034] As will be described later, the management device 20 performs filtering by eliminating the influence of sensor noise, which is noise due to individual differences among the multiple sensors 121, based on the magnitude of the difference in measurement values ​​between adjacent sensors 121 in the sensor array 12. Specifically, the management device 20 acquires multiple array measurement values ​​measured by the sensor array 12 at multiple times. The management device 20 acquires parameters that determine the statistical distribution of the sensor noise using a maximum likelihood method based on the multiple array measurement values, and performs filtering using the acquired parameters. Therefore, according to this embodiment, it is possible to remove bias due to individual differences among the sensors 121 and perform highly accurate filtering.

[0035] The management device 20 also calculates the difference between each of the multiple array measurement values ​​and the estimated value to obtain multiple residuals. The management device 20 obtains an evaluation index indicating the appropriateness of the parameter based on the magnitude of the difference between the multiple residual values ​​between adjacent sensors 121 among the multiple sensors 121. The management device 20 updates the parameter based on this evaluation index. Therefore, according to this embodiment, the parameter is updated based on the evaluation index indicating the appropriateness of the parameter, making it possible to perform highly accurate filtering based on a more appropriate parameter.

[0036] (Measuring device 10) The measurement device 10 includes a control unit 11, a sensor array 12, a measurement unit 13, and a communication unit 14.

[0037] The control unit 11 includes one or more processors. In one embodiment, the "processor" may be, but is not limited to, a general-purpose processor or a dedicated processor specialized for a specific process. The control unit 11 is communicably connected to each component of the measurement device 10 and controls the operation of the measurement device 10 as a whole.

[0038] The sensor array 12 includes a plurality of sensors 121 arranged in a two-dimensional array. The sensor array 12 will be described with reference to FIGS.

[0039] 2 is a schematic diagram showing an example of the arrangement of the sensor array 12 on the measurement object 80. The sensor array 12 is arranged near the measurement object 80 and measures the spatial distribution of magnetic flux density around the measurement object 80. Each sensor 121 is a magnetic sensor using a magnetically sensitive element such as a Hall element, a magnetoresistance effect element, or a magnetoimpedance effect element.

[0040] The measurement object 80 is a structure such as a pipe, a pillar, or a beam installed in a plant, a factory, a bridge, or other building. The measurement object 80 is made of a magnetic material. In this embodiment, the measurement object 80 may be magnetized by a magnetizing device that applies a magnetic field. Once magnetized, the magnetism of the measurement object 80 remains even after the magnetic field is no longer applied. However, the measurement object 80 may be made of a magnetic material that is inherently magnetic, rather than a magnetic material that generates a magnetic field when magnetized. Below, an example will be described in which the measurement object 80 is made of carbon steel, which is a ferromagnetic material.

[0041] FIG. 3 is a diagram schematically illustrating the distribution of magnetic flux density around a measurement target 80. The upper part of FIG. 3 shows the arrangement of the measurement target 80 and the sensors 121, and the lower part shows a graph 91 illustrating the relationship between position in the X direction and magnetic flux density intensity. In FIG. 3, a portion of the measurement target 80 has a wall thickness reduction 81 due to corrosion. When the wall thickness reduction 81 occurs, magnetic flux disturbance occurs due to a difference in magnetic resistance. This magnetic flux disturbance causes some of the magnetic flux to leak outside the measurement target 80. As a result, as shown in graph 91, the magnetic flux density measured by each sensor 121 increases near the portion where the wall thickness reduction 81 is located. As shown in graph 91, the magnetic flux density changes not only in the X direction but also in the Y direction. The measurement system 1 detects the wall thickness reduction 81 based on this change in magnetic flux density.

[0042] The sensor array 12 outputs the measurement values ​​of each sensor 121 as array measurement values, which are two-dimensional data according to the arrangement of the sensors 121. Each sensor 121 acquires magnetic flux density components in the X, Y, and Z directions per measurement.

[0043] 4A to 4C show array measurement values ​​acquired by the sensor array 12 as a two-dimensional map. FIG. 4A is a diagram showing an example of the X-component of magnetic flux density. FIG. 4B is a diagram showing an example of the Y-component of magnetic flux density. FIG. 4C is a diagram showing an example of the Z-component of magnetic flux density. FIGS. 4A to 4C show an example of array measurement values ​​of the sensor array 12 including 9 × 9 sensors 121. In FIGS. 4A to 4C, the displayed position of each element in the two-dimensional array corresponds to the X-coordinate and Y-coordinate of the sensor 121 that measured the measurement value corresponding to each element. In FIGS. 4A to 4C, the measured values ​​of magnetic flux density are represented by grayscale shading. Elements that are closer to white represent larger measured values, and elements that are closer to black represent smaller measured values.

[0044] In this embodiment, an example is described in which the sensor array 12 has M×N sensors 121 arranged two-dimensionally on the same plane, but the arrangement of the sensors 121 in the sensor array 12 is not limited to this. For example, the sensors 121 may be arranged two-dimensionally on the surface of a cylinder whose central axis is common to that of the measurement target 80. In this case, the distance between each sensor 121 and the surface of the measurement target 80 is the same regardless of the position of the sensor 121.

[0045] Returning to the explanation of Fig. 1, the measurement unit 13 acquires and temporarily stores array measurement values ​​measured by the sensor array 12. The measurement unit 13 includes an acquisition unit 131 and a storage unit 132.

[0046] The acquisition unit 131 acquires array measurement values ​​from the sensor array 12 under the control of the control unit 11. The storage unit 132 temporarily stores the array measurement values ​​acquired by the acquisition unit 131. The storage unit 132 includes any storage module, such as a hard disk drive (HDD), a solid state drive (SSD), a read-only memory (ROM), and a random access memory (RAM).

[0047] The communication unit 14 includes any communication module that can communicate with the management device 20 using any communication technology, whether wired or wireless. The communication unit 14 may include a communication control module for controlling communication with the management device 20, and a storage module for storing communication data such as identification information required for communication with the management device 20.

[0048] At least some of the functions of the measurement device 10 can be realized by execution by a processor included in the control unit 11. That is, the functions of the measurement device 10 can be realized by software. A program causes a computer to execute processing of steps included in the operation of the measurement device 10, thereby causing the computer to realize functions corresponding to the processing of each step. That is, the program is a computer program that causes a computer to function as the measurement device 10 according to this embodiment.

[0049] The control unit 11 acquires the array measurement values ​​of the sensor array 12 at a predetermined timing, and transmits measurement data including the acquired array measurement values ​​from the communication unit 14 to the management device 20 via the network 30. For example, in response to receiving trigger information from the management device 20 at a predetermined measurement interval (e.g., once a day), the control unit 11 may acquire the array measurement values ​​of the sensor array 12 and transmit the measurement data to the management device 20. The control unit 11 may determine that the measurement data may include information indicating the correspondence between the position of each sensor 121 and its measurement value, the measurement date and time, and the like.

[0050] (Management device 20) The management device 20 includes a control unit 21, a filtering unit 22, an application unit 23, a display processing unit 24, and a communication unit 26. The management device 20 is a general-purpose computer such as a PC (Personal Computer), a tablet terminal, or a WS (Work Station), but may also be configured as a dedicated electronic device.

[0051] The control unit 21 includes one or more processors. In one embodiment, the "processor" may be, but is not limited to, a general-purpose processor or a dedicated processor specialized for a specific process. The control unit 21 is communicably connected to each component of the management device 20 and controls the operation of the entire management device 20.

[0052] The communication unit 26 includes any communication module that can communicate with the measurement device 10 using any communication technology, whether wired or wireless. The communication unit 26 may include a communication control module for controlling communication with the measurement device 10, and a storage module for storing communication data such as identification information required for communication with the measurement device 10. The communication method of the communication unit 26 does not have to be the same as the communication method of the communication unit 14.

[0053] The filtering unit 22 performs filtering to remove noise from the array measurement values ​​measured by the sensor array 12 of the measurement device 10. The filtering unit 22 includes a measurement value storage unit 221, a filter calculation unit 222, a setting value storage unit 223, and a result storage unit 224.

[0054] The measurement value storage unit 221 holds the array measurement values ​​of the sensor array 12 acquired from the measuring device 10 via the communication unit 26. The measurement value storage unit 221 may hold a certain number (for example, T) of array measurement values ​​acquired in the past. When the measurement value storage unit 221 receives new array measurement values ​​from the measuring device 10 while holding T array measurement values, the measurement value storage unit 221 may erase the oldest array measurement values ​​in order and hold the newly acquired array measurement values. In this way, the measurement value storage unit 221 holds the latest T array measurement values.

[0055] The setting value storage unit 223 holds setting values ​​required for filter calculation. The setting values ​​may be set in advance by the user at the configuration stage.

[0056] The filter calculation unit 222 acquires a plurality of array measurement values ​​stored in the measurement value storage unit 221, and performs filtering using the setting values ​​stored in the setting value storage unit 223. The filter calculation unit 222 may perform filtering for each of the X-direction, Y-direction, and Z-direction components.

[0057] The result storage unit 224 stores the results of filtering performed by the filter operation unit 222. The filtering results are estimates of true physical quantities that are estimated based on a plurality of array measurement values ​​stored in the measurement value storage unit 221. Details of the filtering operation performed by the filtering unit 222 will be described later. When the filtering results are output as a probability density function, the result storage unit 224 may store the parameters of the probability density function (statistical distribution) to which the filtered data conforms.

[0058] The application unit 23 acquires and outputs information regarding the state of the measurement object 80 based on the filtering result in the filtering unit 22. In this embodiment, the application unit 23 outputs the presence or absence of wall-thinning 81 and the depth of the wall-thinning 81 as the state of the measurement object 80. The application unit 23 includes a calculation unit 231, a setting value storage unit 232, and a result storage unit 233.

[0059] The set value storage unit 232 holds set values ​​for acquiring information about the state of the measurement target 80 based on the filtering results. For example, the set value storage unit 232 may hold, as set values, information indicating the relationship between the magnetic flux density generated around the measurement target 80 and the presence or absence of wall thinning 81 in the measurement target 80 and the depth of the wall thinning 81. The user may set the set values ​​in advance at the configuration stage. The set value storage unit 232 may hold, as set values, model information indicating the relationship between the array measurement values ​​learned in advance by machine learning and information about the state of the measurement target 80.

[0060] The calculation unit 231 acquires information about the state of the measurement target 80 by performing calculations using the setting values ​​stored in the setting value storage unit 232 on the filtering results stored in the result storage unit 224 of the filtering unit 22. Specifically, the presence or absence of wall thinning 81 in the measurement target 80 and the depth of the wall thinning 81 may be acquired by performing calculations using the setting values ​​stored in the setting value storage unit 232 on the filtered array measurement values.

[0061] The result storage unit 233 stores the results of the calculations performed by the calculation unit 231. The application unit 23 may perform calculations based on a known method, for example, as described in Patent Document 1 or the like.

[0062] The display processing unit 24 displays information about the state of the measurement target 80 stored in the result storage unit 233 on a display unit such as a liquid crystal display. Specifically, the display processing unit 24 may plot the presence or absence of wall thinning 81 in the measurement target 80 and the depth of the wall thinning 81 on the display unit using a trend graph with error bars. The display processing unit 24 may display the error bars as, for example, a box-and-whisker plot. The display processing unit 24 may display the depth of wall thinning 81 that is determined to be dangerous on the display unit together with the measurement value of the measurement target 80.

[0063] The measurement value storage unit 221, the setting value storage unit 223, the result storage unit 224, the setting value storage unit 232, and the result storage unit 233 may be realized by the storage unit 25. The storage unit 25 includes any storage module, such as an HDD, an SSD, a ROM, or a RAM. The storage unit 25 may function as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 25 stores any information used in the operation of the management device 20. For example, the storage unit 25 may store not only the information held in the setting value storage unit 223, the result storage unit 224, the setting value storage unit 232, and the result storage unit 233, but also various types of information such as system programs and application programs. The storage unit 25 is not limited to being built into the management device 20, but may also be an external database or an external storage module.

[0064] At least some of the functions of the management device 20 can be realized by execution by a processor included in the control unit 21. That is, the functions of the management device 20 can be realized by software. The program causes a computer to execute processing of steps included in the operation of the management device 20, thereby causing the computer to realize functions corresponding to the processing of each step. That is, the program is a computer program for causing a computer to function as the management device 20 according to this embodiment.

[0065] The control unit 21 transmits trigger information to the measurement device 10 at predetermined measurement intervals (e.g., once a day) to request array measurement values ​​of the sensor array 12. Based on the array measurement values ​​transmitted from the measurement device 10 in response to the trigger information, the control unit 21 performs filtering and calculations to acquire information indicating the state of the measurement target 80.

[0066] Fig. 5 is a flowchart showing an example of the operation of management device 20 in Fig. 1. The operation of management device 20 described with reference to Fig. 5 may correspond to one of the information processing methods of an information processing device. The operation of each step in Fig. 5 may be executed based on the control of control unit 21 of management device 20.

[0067] In step S1, the control unit 21 acquires T array measurement values ​​stored in the measurement value storage unit 221 via the communication unit 26. For example, if the measurement interval of the sensor array 12 is one day (24 hours), the control unit 21 acquires array measurement values ​​for T days. In this embodiment, the measurement interval of the sensor array 12 is set to a time that is sufficiently shorter than the time required for corrosion to progress in the measurement target 80 and the time required for individual differences in measurements by each sensor 121 to change. Therefore, while the sensor array 12 performs measurements T times, the magnetic flux density around the measurement target 80 and noise due to individual differences in the sensors 121 (sensor noise) hardly change.

[0068] In step S2, the control unit 21 performs filtering processing on the T array measurement values ​​acquired in step S1 using the filtering unit 22. Details of the filtering processing will be described later.

[0069] In step S3, the control unit 21 acquires information about the state of the measurement target 80 based on the array measurement values ​​filtered in step S2 by the application unit 23.

[0070] In step S4, the control unit 21 causes the display processing unit 24 to output information about the state of the measurement target 80 acquired in step S3 to the display unit or the memory unit 25. When the processing of step S4 is completed, the control unit 21 ends the processing of the flowchart.

[0071] The filtering process of step S2 will be described with reference to Fig. 6. Fig. 6 is a diagram schematically showing the types of signals included in the array measurement values. The filtering process is performed independently for each of the X-, Y-, and Z-direction components of the array measurement values. Therefore, the following description focuses on the X-component of the array measurement values, but the same applies to the Y- and Z-components.

[0072] In Figure 6, g t (t=1,...,T) are T array measurement values ​​of the sensor array 12 acquired in step S1. T Each of these has M×N measurement values ​​(scalar quantities). Therefore, the measurement values ​​(X components) of the m-th (m=1,...,M) sensor 121 in the X direction and the n-th (n=1,...,N) sensor 121 in the Y direction are expressed as g t(m,n) =g 1(m,n) ,...,g T(m,n) It is expressed as:

[0073] f is the true value of the magnetic flux density that should be measured around the measurement target 80 and does not include any error. As described above, the magnetic flux density around the measurement target 80 changes very little while the sensor array 12 performs T measurements. Therefore, the true value f has M×N measurement values ​​(scalar quantities) that are independent of time. The true value of the magnetic flux density (X component) at the m-th sensor 121 (m=1,...,M) in the X direction and the n-th sensor 121 (n=1,...,N) in the Y direction is expressed as f m,n It is expressed as:

[0074] Array measurement g t and the difference between the true value f (g t -f) is the array measurement g t The error components include the sensor noise μ and the time noise ε t This includes:

[0075] The sensor noise μ is noise based on the individual differences of the M×N sensors 121. The sensor noise μ also changes little while the sensor array 12 performs T measurements. Therefore, the sensor noise μ has M×N measurement values ​​(scalar quantities) that are independent of time. The sensor noise (X component) of the m-th sensor 121 (m=1,...,M) in the X direction and the n-th sensor 121 (n=1,...,N) in the Y direction is expressed as μ m,n It is expressed as:

[0076] Temporal noise ε t (t=1,...,T) is noise caused by the variation of the measurement values ​​in each measurement in each of the M×N sensors 121. T Each of the sensors 121 has M×N measurement values ​​(scalar quantities). Therefore, the time noise (X component) of the m-th (m=1,...,M) sensor in the X direction and the n-th (n=1,...,N) sensor in the Y direction is expressed as ε t(m,n) =ε 1(m,n) ,...,ε T(m,n) It is expressed as:

[0077] In this case, for any t (t=1,...,T), m (m=1,...,M), and n (n=1,...,N), the following equation 1 holds:

[0078]

number

[0079] T array measurements g1,...,g T When (μ,f)=(μ i ,f i ) is the probability P(μ i ,f i |g1,...,g T ) is expressed by the following equation 2 according to Bayes' theorem:

[0080]

number

[0081] In equation 2, P(g t |μ i ,f i ) is (μ,f)=(μ i ,f i ), the array measurement g t =g1,...,g T is the probability of obtaining P(f i ) is true if f=f i This is the probability that P(μ i ) is the sensor noise μ = μ i The probability density functions (statistical distributions) P(f) and P(μ) are prior distributions.

[0082] Integrating both sides of equation 2 with respect to μ gives equation 3.

[0083]

number

[0084] Equation 3 is a function of T array measurements g1,...,g T When the true value f=f is obtained, i More generally, the probability that the array measurements g1,...,g TWhen is obtained, the probability density function of the true value f is expressed by Equation 4.

[0085]

number

[0086] Equation 4 indicates the probability that the true signal is f under the condition that T array measurement values ​​are observed. Therefore, the filtering unit 22 calculates the probability P(f|g1,...,g T ) is used as a filter. Specifically, the filtering unit 22 uses the true value f that maximizes P(f), P(μ), and P(g t Based on a model that is preset as a probability density function of |μ, f), the parameters of these probability density functions are calculated by the maximum likelihood method. P(g1,...,g T ) may be determined by the maximum likelihood method.

[0087]

number

[0088] Here, P(μ) is a probability density function observed as sensor noise μ based on individual differences among the multiple sensors 121. P(g t |μ,f) is the time noise ε when the same sensor 121 measures the same true value at different timings. t The probability density function observed is the sensor noise μ and the time noise ε t is considered to occur independently of the other sensors 121, that is, independently for each sensor 121. Therefore, the filtering unit 22 calculates P(μ) and P(g t For example, a normal distribution model may be used as |μ,f. Using a normal distribution model, μ and g t If the mean value of is 0, the parameter of P(μ) is the variance σ μ 2 and P(g t The parameter of |μ,f) is the variance σ 2Variance σ μ 2 , σ 2 is considered to have a constant value regardless of the state of the measurement target 80. Therefore, the management device 20 calculates the variance calculated based on multiple trials in advance as the variance σ μ 2 , σ 2 It may also be used as.

[0089] Alternatively, when performing the filtering process, the management device 20 may use T array measurements g1, . . . , g T Based on the temporal noise ε t Variance σ 2 Specifically, the management device 20 may calculate T array measurements g1,...,g T and the array measurements g1,...,g T Calculate the difference from the mean value of σ 2 may be calculated.

[0090] P(f) is a probability density function of the true value of the magnetic flux density around the measurement object 80. Here, it is considered that the magnetic field formed by the measurement object 80 changes continuously around the measurement object 80. In this case, it is considered that the difference in magnetic flux density at the positions of the adjacent sensors 121 is small. Therefore, the filtering unit 22 calculates, for example, f m,n and f m,n+1 The difference between m,n and f m+1,n A statistical distribution model in which the probability increases as the difference between f and f decreases may be used as P(f). More specifically, a statistical distribution model with parameters α, β, and γ may be used as P(f). Here, α is a parameter that reflects the components of the true value f of the magnetic flux density that are distributed independently for each sensor 121. β is a parameter that reflects the difference in measurement values ​​between sensors 121 adjacent in the Y direction. γ is a parameter that reflects the difference in measurement values ​​between sensors 121 adjacent in the X direction.

[0091] The filtering unit 22 calculates P(g1,...,g T) are calculated by the maximum likelihood method, and these α, β, γ and the aforementioned variance σ μ 2 , σ 2 Using P(g1,...,g T ) is maximized. As mentioned above, the result of filtering is the estimated value f^ of the true physical quantity.

[0092] In this way, the filtering unit 22 performs filtering based on the difference in the magnitude of the difference between the sensor noise μ and the true signal f to be estimated between the measurement values ​​of adjacent sensors 121. Specifically, the filtering unit 22 filters T array measurement values ​​g1,...,g2 measured by the sensor array 12 at multiple times. T The filtering unit 22 filters the parameters α, β, and γ that define the statistical distribution of the sensor noise μ, based on the T array measurements g1,...,g T The estimated value f^ is acquired by the maximum likelihood method based on the above. The filtering unit 22 performs filtering using the acquired parameters α, β, and γ. Therefore, the filtering unit 22 can perform highly accurate filtering based on measured values ​​at multiple times. Therefore, by using the estimated value f^ acquired by such highly accurate filtering, the management device 20 can detect physical quantities around the measurement target 80 with higher accuracy.

[0093] The filtering unit 22 also filters a plurality of array measurements g1,...,g T Based on this, the time noise (ε t ) and estimate the estimated time noise ε t Alternatively, the filtering unit 22 may further use a variance σ 2 Using the time noise ε t For the sensor noise, filtering may be performed using the previously acquired variance σ μ 2With this configuration, the management device 20 estimates the time noise and performs filtering, so that it is possible to remove the influence of variations in the measurement values ​​and perform filtering with high accuracy.

[0094] Furthermore, the filtering unit 22 filters the T array measurements g1,...,g T The difference between each of these and the estimated value f^ is calculated to obtain T residuals, and the variance σ of the sensor noise μ is calculated based on the obtained residuals. μ 2 The filtering unit 22 may evaluate the appropriateness of the variance σ μ 2 Based on the evaluation results of the variance σ μ 2 and update the value of the updated variance σ μ 2 The parameters α, β, and γ may be estimated again by the above-mentioned maximum likelihood method using the above.

[0095] Such updating of parameters will be described with reference to Fig. 7. Fig. 7 is a diagram showing a schematic flow of processing for updating parameters used in filtering.

[0096] In FIG. 7, T array measurement values ​​61(g t ) is provided to the filtering unit 22, where filtering processing 51 is performed. In addition to the true value f, the array measurement value 61 also includes the sensor noise μ and the time noise ε t As a result of the filtering process 51, an estimate 62 (f^) of the true value of the physical quantity is obtained. The estimate f^ is not time-series data, but is an M×N array data. The filtering process 51 performs filtering on the T array measurement values ​​61 (g t This corresponds to the process of extracting components that are commonly contained in

[0097] The filtering unit 22 filters the T array measurements 61 (g t ) and the estimated value 62(f^) that is the result of filtering, and a residual 63 is calculated (52). The residual 63 is T pieces of time series data.

[0098] The filtering unit 22 analyzes the residuals 63 to evaluate whether the filtering process is performing satisfactorily (53).

[0099] The filtering unit 22 uses the results of the evaluation 53 to perform a search (54) to update the estimation conditions in the parameter estimation 55.

[0100] The filtering unit 22 estimates the parameters included in the filter from the input data (55). The filtering unit 22 changes the estimation conditions using the results of the evaluation function.

[0101] The evaluation 53, search 54, and parameter estimation 55 based on the residual 63 will now be described in detail. When filtering in the filter process 51 is performed ideally, the estimated value f^ is almost identical to the true value f. As a result, when filtering is performed appropriately, the residual 63, which is T pieces of time-series data, will contain almost no components of the true value f. Here, since the true value f reflects the magnetic field that continuously changes around the measurement target 80, it is considered that the difference between the positions of adjacent sensors 121 is small. Therefore, the filtering unit 22 evaluates the performance of the filter process 51 based on whether the residual 63 contains many components that result in small differences between adjacent sensors 121. When the residual 63 contains many components that result in small differences between adjacent sensors 121, this means that components of the true value f that could not be extracted by the filter process 51 remain in the residual 63, and therefore the performance of the filter process 51 can be evaluated as poor. If the residual 63 contains almost no components with small differences between adjacent sensors 121 and only noise components, this means that the filter processing 51 has been able to separate the components of the true value f from the noise components, and the performance of the filter processing 51 can be evaluated as good.

[0102] Similar to the filtering process 51, the filtering unit 22 defines the components included in the residuals 63 as a probability density function, and calculates parameters that reflect the magnitude of the difference between the values ​​of adjacent sensors 121 by the maximum likelihood method. t=δ1,...,δ T , δ1,...,δ T Let d be the component that is common to all of δ1,...,δ T , and d are both M×N values. Therefore, the filtering unit 22 calculates the probability density functions P(d) and P(δ t Based on a model previously set as P(δ1,...,δ |d), the parameters of these probability density functions are calculated by the maximum likelihood method. T ) may be determined by the maximum likelihood method.

[0103]

number

[0104] Here, P(δ t |d) is a probability density function observed as noise when the same sensor 121 measures the same common component at different times. This noise is considered to occur independently for each sensor 121. Therefore, the filtering unit 22 calculates P(δ t For example, a normal distribution model may be used as |d). t If the average value of is 0, P(δ t The parameter of |d) is the variance σ' 2 The variance σ' 2 is considered to take a constant value regardless of the state of the measurement object 80. Therefore, the management device 20 calculates the variance calculated based on multiple trials in advance as the variance σ' 2 Alternatively, when performing the filtering process, the management device 20 may use T array measurements g1,...,g T Based on the noise variance σ' 2 may be calculated.

[0105] Since P(d) is a common component included in the T residuals, it is considered that the difference in magnetic flux density at the positions of the sensors 121 adjacent to each other is small. m,n and dm,n+1 The difference between m,n and d m+1,n A statistical distribution model in which the probability increases as the difference between d and d decreases may be used as P(d). More specifically, a statistical distribution model with parameters α', β', and γ' may be used as P(d). Here, α' is a parameter that reflects the components in d that are distributed independently for each sensor 121. β' is a parameter that reflects the difference between sensors 121 adjacent in the Y direction. γ' is a parameter that reflects the difference between sensors 121 adjacent in the X direction.

[0106] Using such a model, the filtering unit 22 calculates P(δ1,...,δ) of Equation 6. T ) to maximize the parameters α', β', γ', and σ' 2 When P(d) is acquired, (β'+γ') becomes a parameter indicating whether the residual 63 contains many components with small differences between adjacent sensors 121, depending on the model of P(d). Therefore, the filtering unit 22 sets α' to a predetermined fixed value, and then selects σ such that the value of (β'+γ') becomes small. μ 2 (54) and perform parameter estimation (55). Here, it is assumed that the residual 63 contains many components with small differences between adjacent sensors 121, that is, when the performance of the filter process 51 is poor, the value of (β'+γ') becomes large.

[0107] Specifically, the filtering unit 22 may estimate the parameters by repeatedly executing the following processes (1) to (7) multiple times.

[0108] (1) The filtering unit 22 initializes the estimation conditions for the filter parameters. For example, σ μ 2 For example, the filtering unit 22 generates a plurality of initial values ​​of (1.0, 2.0, 3.0, 4.0, 5.0) as a search list μ 2 may generate an initial value for

[0109] (2) The filtering unit 22 sets the condition X to be used for the parameter estimation 55. Specifically, the filtering unit 22 uses the multiple σ generated in (1) μ 2 The X-th value among the values ​​of may be substituted into the likelihood function of Equation 5.

[0110] (3) The filtering unit 22 estimates parameters in the parameter estimation 55. Specifically, the input data is substituted into the likelihood function set in (2) to estimate α, β, γ, and σ. 2 may be estimated.

[0111] (4) The filtering unit 22 applies a filter to the input data. Specifically, the filtering unit 22 applies a filter to the α, β, γ, and σ estimated in (3). 2 and the preset σ μ 2 may be substituted into Equation 4, and the input data may be substituted to perform the filtering process.

[0112] (5) The filtering unit 22 calculates the difference between the filter value and the input data. Specifically, the filtering unit 22 may calculate the residual by subtracting the filter value obtained in (4) from the original data.

[0113] (6) The filtering unit 22 analyzes the residual. Specifically, the residual calculated in (5) is substituted into the likelihood function of Equation 6, and α', β', γ', and σ' are calculated by the maximum likelihood method. 2 may be estimated.

[0114] (7) The filtering unit 22 updates the search conditions in the search 54. Specifically, the filtering unit 22 updates σ μ 2 Check β'+γ' for all the values ​​set in , and select the top three σ values ​​with the best performance. μ 2 The three selected σ μ 2 Regarding the above, the filtering unit 22 may perform the following process. a.(σ μ 2 -search radius) and (σ μ 2 + search radius) is the search list. b. Halve the search radius. c. Combine the three search lists. For example, if the initial value of the search radius is 0.5, and the result of (6) is σ μ 2 =(1.0,2.0,4.0) is selected. In this case, the updated σ μ 2 The search list will be (0.5, 1.5, 2.5, 3.5, 4.5). The next search radius will be 0.25.

[0115] The filtering unit 22 can perform filtering using more appropriate parameters (parameters of the probability density function) by repeatedly executing the above processes (1) to (7) multiple times.

[0116] As described above, the management device 20 obtains T array measurements (g1,..., g T ) and the estimated value (f^) to obtain T residuals. The management device 20 calculates the difference between the values ​​of the multiple residuals between the sensors 121 adjacent to each other among the M×N sensors 121, and obtains evaluation indices (α', β', γ', σ') that indicate the appropriateness of the parameters (α, β, γ). 2 The management device 20 acquires the evaluation indexes (α', β', γ', σ' 2 ) and updates the parameters (α, β, γ). In this way, the management device 20 updates the parameters based on the evaluation index that indicates the appropriateness of the parameters, so it is possible to perform highly accurate filtering based on more appropriate parameters.

[0117] Furthermore, the management device 20 uses parameters (α', β', γ', σ') that determine the statistical distribution of the common component d contained in the multiple residuals as an evaluation index. 2) is obtained by the maximum likelihood method based on the T residuals. In this way, the management device 20 estimates the parameter of the statistical distribution of the common component d contained in the multiple residuals by the maximum likelihood method based on the multiple residuals, and uses this parameter as an evaluation index. Therefore, it is possible to accurately identify the statistical distribution of the sensor noise and perform highly accurate filtering.

[0118] In addition, the management device 20 calculates the array measurement value g t As a result, the management device 20 acquires a measurement value of the magnetic flux density around the measurement object 80, which is a magnetic body, and outputs the presence or absence of wall thinning 81 in the measurement object 80 and the depth of the wall thinning 81 as information about the state of the measurement object 80. Therefore, the management device 20 can output the presence or absence of wall thinning 81 and the depth of the wall thinning in the measurement object 80 with high accuracy based on the measurement value of the magnetic flux density around the measurement object 80.

[0119] As described above, while the spatial filter according to the comparative example only smooths values ​​measured by adjacent sensors, the configuration according to this embodiment performs smoothing processing in the time direction. Therefore, the configuration according to this embodiment can incorporate filtering processing that reduces the effects of aliasing in spatial frequencies. Furthermore, the configuration according to this embodiment performs filtering using a probability density function that formulates the difference in spatial characteristics between the average noise component (sensor noise μ) and the true signal component f. Therefore, this embodiment distinguishes between the average noise component and the true signal component, which cannot be removed by a time filter alone, and more effectively removes noise, enabling more accurate detection of physical quantities around the measurement target. Furthermore, the configuration according to this embodiment applies an evaluation process to the filtered results, updates parameters, and improves the filter. Therefore, the configuration according to this embodiment can configure an optimized filter.

[0120] In this embodiment, the measurement system 1 detects the presence or absence of wall thinning 81 in the measurement object 80 and the depth of the wall thinning 81 based on the measured value of the magnetic flux density around the measurement object 80, but the type of physical quantity to be measured and the state of the measurement object to be detected are not limited to these. This embodiment can be applied to any configuration in which a physical quantity reflecting the state of the measurement object is measured using multiple sensors, in which the physical quantity changes continuously in space and the fluctuation of the physical quantity over time changes sufficiently slowly compared to the sampling rate. For example, the measurement system 1 can be applied to any measurement object in which a magnetic field, electric field, etc. changes as corrosion progresses.

[0121] Furthermore, since general spatial filters also have filter parameters, the method of this embodiment may be applied to such spatial filters to automatically adjust the parameters using an evaluation function. For example, in image processing, a bilateral filter requires specifying parameters such as the filter size, spatial direction, and standard deviation in the pixel value direction. By using the method of this embodiment, such parameters can be automatically adjusted. Specifically, appropriate initial values ​​may be set for the spatial filter parameters, the filter may be applied, and the residual may be analyzed using the evaluation function to evaluate the quality of the set parameters. A search list may be created for each parameter, and the peripheral values ​​of the filter parameter that shows the best performance based on the evaluation results may be added to the search list for the next time. By continuing this search, the filter parameters can be automatically set.

[0122] The present disclosure is not limited to the above-described embodiments. For example, multiple blocks shown in the block diagrams may be integrated, or one block may be divided. Multiple steps shown in the flowcharts may be executed in parallel or in a different order depending on the processing capabilities of the device executing each step, or as needed, instead of being executed in chronological order as described. Other modifications are possible without departing from the spirit of the present disclosure. [Explanation of symbols]

[0123] 1. Measurement System 10. Measuring equipment 11 Control section 12 Sensor Array 121 Sensors 13 Measuring part 131 Acquisition Department 132 Storage area 14 Communications Department 20 Management device 21 Control section 22 Filtering section 221 Measurement value storage section 222 Filter calculation section 223 Setting value storage section 224 Result storage section 23 Application Section 231 Arithmetic section 232 Setting value storage section 233 Result storage section 24 Display processing section 25 Memory section 26 Communications Department 30 Network 61 Array Measurements 62 Estimated 63 Residual 80 Measurement Objects 81 Thinning 91 graphs

Claims

1. acquiring array measurements, which are measurements of physical quantities, measured by a sensor array in which a plurality of sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain estimates of the true physical quantities; outputting information about the state of the measurement object based on the acquired estimated value; A control unit is provided, The control unit performing the filtering by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other; Information processing device.

2. The control unit acquiring a plurality of array measurements measured by the sensor array at a plurality of times; obtaining parameters defining a statistical distribution for the sensor noise by a maximum likelihood method based on the plurality of array measurements; performing the filtering using the obtained parameter; The information processing device according to claim 1 .

3. estimating temporal noise based on the plurality of array measurements, the temporal noise being noise due to variations in measurements at each of the plurality of sensors; performing the filtering further using the estimated temporal noise; The information processing device according to claim 2 .

4. The control unit calculating the difference between each of the plurality of array measurements and the estimate to obtain a plurality of residuals; obtaining an evaluation index indicating the appropriateness of the parameter based on the magnitude of differences between the plurality of residual values ​​between adjacent sensors among the plurality of sensors; updating the parameter based on the evaluation index; The information processing device according to claim 2 .

5. the control unit acquires, as the evaluation index, a parameter that determines a statistical distribution of common components included in the plurality of residuals by a maximum likelihood method based on the plurality of residuals. The information processing device according to claim 4 .

6. The control unit As the array measurement value, a measurement value of magnetic flux density around the measurement target, which is a magnetic body, is acquired; and outputting, as information regarding the state of the measurement object, whether or not there is thinning in the measurement object and the depth of the thinning. The information processing device according to claim 1 .

7. An information processing method for an information processing device including a control unit, The control unit Acquiring an array measurement value, which is a measurement value of a physical quantity, measured by a sensor array in which a plurality of sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain an estimate of the true physical quantity; outputting information about the state of the object to be measured based on the obtained estimated value; Including, The control unit performing the filtering by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other; Information processing methods.

8. On the computer, acquiring array measurements, which are measurements of physical quantities, measured by a sensor array in which a plurality of sensors are arranged two-dimensionally; filtering the acquired array measurements to remove noise and obtain an estimate of the true physical quantity; outputting information about the state of the object to be measured based on the obtained estimated value; Execute a procedure that includes performing the filtering by eliminating the influence of sensor noise, which is noise due to individual differences among the plurality of sensors, based on the magnitude of the difference in the measurement values ​​between the sensors adjacent to each other; program.

Images