Computer program, learning model, estimation device, and estimation method

Abstract

To provide a computer program, a learning model, an estimation device and an estimation method.SOLUTION: A computer program allows a computer to execute processing of acquiring waveform data obtained by a flaw detection probe when scanning an underground power transmission cable by the flaw detection probe, when inputting the waveform data obtained by the flaw detection probe, inputting the obtained waveform data to a learning model learned to output information on a cable type, and estimating the cable type of the underground power transmission cable to output an estimation result.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to a computer program, a learning model, an estimation device, and an estimation method. [Background technology]

[0002] In recent years, power transmission cables have been increasingly laid underground due to urban congestion and aesthetic reasons. Underground power transmission cables include waterproof cables with a waterproof layer and non-waterproof cables without a waterproof layer, and both types are currently used together. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Utility Model Application Publication No. 5-20210 [Patent Document 2] Japanese Patent Application Laid-Open No. 2002-75072 [Patent Document 3] Japanese Patent Application Laid-Open No. 2006-164725 Summary of the Invention [Problem to be solved by the invention]

[0004] In addition to the above-mentioned waterproof cables and non-waterproof cables, there are various types of underground power transmission cables, such as cables with different waterproof layer materials, cables with a shielding layer made of metal tape, and cables with a shielding layer made of conductor wire shield, etc. However, it is difficult to distinguish the cable type of underground power transmission cables from the outside.

[0005] An object of the present invention is to provide a computer program, a learning model, an estimation device, and an estimation method that can estimate the type of an underground power transmission cable. [Means for solving the problem]

[0006] A computer program according to one aspect of the present invention is a computer program for causing a computer to execute a process of acquiring waveform data obtained from a flaw detection probe when the flaw detection probe scans an underground power transmission cable, inputting the acquired waveform data into a learning model that has been trained to output information regarding the cable type when the waveform data obtained from the flaw detection probe is input, estimating the cable type of the underground power transmission cable, and outputting the estimation result.

[0007] A learning model according to one aspect of the present invention includes a neural network that has learned the relationship between the waveform data and the type of underground power transmission cable by using a dataset containing waveform data obtained from a flaw detection probe when the flaw detection probe scans an underground power transmission cable and information indicating the type of the underground power transmission cable as training data, and when waveform data obtained from the flaw detection probe is input, the neural network executes calculations and causes a computer to function to output information regarding the type of underground power transmission cable.

[0008] An estimation device according to one aspect of the present invention includes an acquisition unit that acquires waveform data obtained from a flaw detection probe when the flaw detection probe scans an underground power transmission cable, an estimation unit that inputs the waveform data acquired by the acquisition unit into a learning model that has been trained to output information regarding the cable type when waveform data obtained from the flaw detection probe is input, and estimates the cable type of the underground power transmission cable, and an output unit that outputs the estimation result.

[0009] An estimation method according to one aspect of the present invention involves a computer executing a process to acquire waveform data obtained from a flaw detection probe when the flaw detection probe scans an underground power transmission cable, input the acquired waveform data into a learning model that has been trained to output information about the cable type when the waveform data obtained from the flaw detection probe is input, estimate the cable type of the underground power transmission cable, and output the estimation result. [Effects of the Invention]

[0010] According to the present application, it is possible to estimate the type of underground power transmission cable. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is an explanatory diagram illustrating a configuration example of an estimation system according to a first embodiment. [Figure 2] FIG. 2 is a cross-sectional view illustrating the structure of an underground power transmission cable. [Figure 3] FIG. 2 is a block diagram illustrating the internal configuration of the estimation device. [Figure 4] FIG. 1 is a schematic diagram illustrating an example of the configuration of a learning model. [Figure 5] FIG. 1 is a conceptual diagram illustrating an example of a data set. [Figure 6] 10 is a flowchart illustrating a procedure for generating a learning model. [Figure 7] 10 is a flowchart illustrating a procedure for executing a type estimation process executed by the estimation device. [Figure 8] FIG. 10 is a schematic diagram showing a schematic configuration of a learning model in the second embodiment. [Figure 9] FIG. 11 is a schematic diagram showing a general configuration of a learning model in the third embodiment. [Figure 10] FIG. 10 is a schematic diagram showing a general configuration of a learning model in the fourth embodiment. [Figure 11] FIG. 10 is a block diagram illustrating the internal configuration of an estimation device according to a fifth embodiment. [Figure 12] FIG. 2 is a schematic diagram showing an example of the configuration of a first learning model. [Figure 13] FIG. 2 is a schematic diagram showing an example of the configuration of a second learning model. [Figure 14] 13 is a flowchart illustrating the execution procedure of a type estimation process executed by an estimation device according to a fifth embodiment. [Figure 15] 13 is a flowchart illustrating the execution procedure of a type estimation process executed by an estimation device according to a sixth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] The present invention will now be described in detail with reference to the drawings showing embodiments thereof. (Embodiment 1) 1 is an explanatory diagram illustrating an example of the configuration of an estimation system according to embodiment 1. The estimation system according to embodiment 1 includes an estimation device 1, an eddy current flaw detector 2, and an ultrasonic flaw detector 3.

[0013] The eddy current flaw detector 2 is a device for detecting defects in an object to be inspected using eddy currents, and is equipped with an eddy current flaw detector probe 20 incorporating an excitation coil and a detection coil. The eddy current flaw detector 2 supplies an AC current to the excitation coil of the eddy current flaw detector probe 20 to generate an AC magnetic field, inducing eddy currents in the surface layer of the object to be inspected. If a defect is present in the surface layer of the object to be inspected, the eddy currents are disturbed, causing a change in the magnetic field. The eddy current flaw detector 2 detects the change in the magnetic field in the object to be inspected as a change in the induced voltage appearing in the detection coil, and detects defects in the object to be inspected.

[0014] The ultrasonic flaw detection device 3 is a device for detecting defects in an inspection object using ultrasonic waves, and includes an ultrasonic flaw detection probe 30 with a built-in transmitting piezoelectric element and a built-in receiving piezoelectric element. The ultrasonic flaw detection device 3 applies a voltage to the transmitting piezoelectric element of the ultrasonic flaw detection probe 30 to generate a physical force known as the piezoelectric effect, which is then propagated as ultrasonic waves into the inspection object. If a defect exists in the ultrasonic propagation path, the ultrasonic waves will be reflected by the defect. The ultrasonic flaw detection device 3 detects the reflected waves from the defect using the receiving piezoelectric element, and detects the defect in the inspection object.

[0015] Both the eddy current flaw detector 2 and the ultrasonic flaw detector 3 are devices used to detect defects in an object to be inspected, but in this embodiment, a configuration will be described in which the cable type of an underground power transmission cable 4 is estimated based on waveform data output from these devices. Here, the underground power transmission cable 4 includes waterproof cables with a waterproof layer and non-waterproof cables without a waterproof layer. Furthermore, the underground power transmission cable 4 can include various types of cables, such as those with a different waterproof layer material, those with a shielding layer made of metal tape, and those with a shielding layer made of a conductor wire shield.

[0016] The estimation device 1 is a device for estimating the cable type of an underground power transmission cable 4 based on data obtained when the underground power transmission cable 4 is scanned by an eddy current flaw detector 2 and an ultrasonic flaw detector 3. Specifically, the estimation device 1 acquires first waveform data obtained when the underground power transmission cable 4 is scanned by an eddy current flaw detector 20 and second waveform data obtained when the underground power transmission cable 4 is scanned by an ultrasonic flaw detector 30, inputs the acquired first waveform data and second waveform data into a learning model LM (see FIG. 4 ) described below, and executes calculations using the learning model LM to estimate the cable type of the underground power transmission cable 4. As described above, underground power transmission cables 4 can include various types of cables. In the first embodiment, a configuration for estimating whether a target underground power transmission cable 4 is a waterproof cable or a non-waterproof cable will be described as an example of type estimation.

[0017] Here, the first waveform data may be data showing a raw signal waveform output from the eddy current flaw detection probe 20, or may be data showing a signal waveform after appropriate processing has been performed inside the eddy current flaw detection device 2. Similarly, the second waveform data may be data showing a raw signal waveform output from the ultrasonic flaw detection probe 30, or may be data showing a signal waveform after appropriate processing has been performed inside the ultrasonic flaw detection device 3. These waveform data may be numerical data that numerically represents the time change of the signal, or may be image data of a waveform (Lissajous waveform or time sweep waveform) in which the time change of the signal is plotted as a graph.

[0018] 1 shows a state in which a specific location of the underground power transmission cable 4 is scanned with the eddy current testing probe 20 and another location is scanned with the ultrasonic testing probe 30, but it is not necessary to scan the underground power transmission cable 4 with both testing probes 20, 30 simultaneously. For example, a configuration may be adopted in which scanning with the eddy current testing probe 20 (or the ultrasonic testing probe 30) is performed first, and then scanning with the ultrasonic testing probe 30 (or the eddy current testing probe 20) is performed. Furthermore, the location scanned by the eddy current testing probe 20 and the location scanned by the ultrasonic testing probe 30 may be the same or different.

[0019] FIG. 2 is a cross-sectional view illustrating the structure of an underground power transmission cable 4. FIG. 2A shows an example of an underground power transmission cable 4, namely, a non-waterproof cable without a water-shielding layer. A non-waterproof cable is formed, for example, by arranging a conductor 41, an insulator 42, a shielding layer 43, and a sheath 45 coaxially from the center outward. The conductor 41 is formed by twisting together multiple core wires made of copper, copper alloy, aluminum, aluminum alloy, or the like. The insulator 42 is a member for insulating the conductor 41 and is formed from a plastic such as cross-linked polyethylene, rubber, insulating oil, insulating paper, or the like. The shielding layer 43 is a member for shielding electromagnetic waves generated by current flowing through the conductor 41 and is formed by spirally winding a copper tape around the outer surface of the insulator 42. Alternatively, the shielding layer 43 may be formed as a wire shield (WS) in which conductive wires such as annealed copper wires are wound around the outer surface of the insulator 42. The sheath 45 is provided for the purpose of preventing corrosion and moisture intrusion, and is made of synthetic rubber such as chloroprene, or plastic such as vinyl chloride or polyethylene.

[0020] FIG. 2B shows a water-shielding cable having a water-shielding layer 44 as another example of the underground power transmission cable 4. The water-shielding cable includes the conductor 41, insulator 42, shielding layer 43, and sheath 45 described above, as well as the water-shielding layer 44 disposed on the outside of the shielding layer 43. The water-shielding layer 44 is a component for enhancing the water resistance of the sheath 45. For example, the water-shielding layer 44 may be a metal laminate tape in which one side of a synthetic resin substrate is coated with a metal (e.g., aluminum or lead). The water-shielding layer 44 has a width slightly longer than the circumferential length of the shielding layer 43. The water-shielding layer 44 is wound around the shielding layer 43 to cover it, and both widthwise edges are overlapped and joined in a watertight manner. The joined portion is a radially overlapping portion of the water-shielding layer 44. Hereinafter, the radially overlapping portion of the water-shielding layer 44 will be referred to as a double-layer portion 44a. On the other hand, the unbonded portions are portions where there is no overlap in the radial direction of the water-shielding layer 44. Hereinafter, the portion where there is no overlap in the radial direction of the water-shielding layer 44 will be referred to as a single portion 44b.

[0021] 3 is a block diagram illustrating the internal configuration of the estimation device 1. The estimation device 1 is a general-purpose or dedicated computer, and includes a control unit 11, a storage unit 12, an input unit 13, a communication unit 14, an operation unit 15, a display unit 16, etc.

[0022] The control unit 11 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The ROM included in the control unit 11 stores control programs and the like that control the operation of each hardware unit included in the estimation device 1. The CPU in the control unit 11 reads and executes the control programs stored in the ROM and various computer programs stored in the storage unit 12, and controls the operation of each hardware unit, thereby causing the entire device to function as the estimation device of the present application. The RAM included in the control unit 11 temporarily stores data used during execution of calculations.

[0023] In the embodiment, the control unit 11 is configured to include a CPU, a ROM, and a RAM, but may instead be one or more arithmetic circuits including a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), a quantum processor, volatile or non-volatile memory, etc. The control unit 11 may also include functions such as a clock that outputs date and time information, a timer that measures the elapsed time from when an instruction to start measurement is given until when an instruction to end measurement is given, and a counter that counts numbers.

[0024] The storage unit 12 includes a storage device such as a hard disk drive (HDD), a solid state drive (SSD), etc. The storage unit 12 stores various computer programs executed by the control unit 11 and various data used by the control unit 11.

[0025] The computer programs stored in the storage unit 12 include an estimation processing program PG1 that causes a computer to execute a process of acquiring first waveform data and second waveform data, estimating the type of underground power transmission cable based on the acquired first waveform data and second waveform data, and outputting the estimation result. The estimation processing program PG1 may be a single computer program or may be composed of multiple computer programs. Furthermore, the estimation processing program PG1 may partially use an existing library.

[0026] A computer program (program product w) including the estimation processing program PG1 is provided by a non-transitory recording medium RM on which the computer program is readably recorded. The recording medium RM is, for example, a portable memory such as a CD-ROM, a USB memory, an SD (Secure Digital) card, a micro SD card, or a CompactFlash (registered trademark). The control unit 11 reads various computer programs from the recording medium RM using a reading device (not shown) and stores the read various computer programs in the memory unit 12. The computer programs stored in the memory unit 12 may also be provided by communication via the communication unit 14. In this case, the control unit 11 acquires the computer programs through the communication unit 14 and stores the acquired computer programs in the memory unit 12.

[0027] The storage unit 12 also includes a learning model LM that has been trained to output information about the type of underground power transmission cable when the first waveform data and the second waveform data are input. The storage unit 12 stores information defining the learning model LM, such as configuration information of layers included in the learning model LM, information about the nodes included in each layer, and information such as weighting coefficients and biases between nodes. The specific configuration of the learning model LM will be described in detail later.

[0028] The input unit 13 includes a connection interface for connecting external devices. In this embodiment, the external devices connected to the input unit 13 are an eddy current flaw detector 2 and an ultrasonic flaw detector 3. The estimation device 1 acquires first waveform data obtained by scanning the underground power transmission cable 4 with an eddy current flaw detector 20 from the eddy current flaw detector 2 connected to the input unit 13, and acquires second waveform data obtained by scanning the underground power transmission cable 4 with an ultrasonic flaw detector 30 from the ultrasonic flaw detector 3 connected to the input unit 13.

[0029] In this embodiment, the eddy current flaw detector 2 and the ultrasonic flaw detector 3 are connected to the input unit 13, and the first waveform data obtained from the eddy current flaw detector probe 20 and the second waveform data obtained from the ultrasonic flaw detector probe 30 are acquired. Alternatively, an image of the waveform displayed on the display screen of the eddy current flaw detector 2 and an image of the waveform displayed on the display screen of the ultrasonic flaw detector 3 may be captured by an imaging device such as a digital camera, and the acquired waveform images may be imported into the estimation device 1 via communication or a recording medium.

[0030] The communication unit 14 includes a communication interface for transmitting and receiving various types of data. The communication interface included in the communication unit 14 is, for example, a communication interface conforming to a LAN (Local Area Network) communication standard used in WiFi (registered trademark) or Ethernet (registered trademark). When data to be transmitted is input from the control unit 11, the communication unit 14 transmits the data to be transmitted to a specified destination. Furthermore, when the communication unit 14 receives data transmitted from an external device to its own device, it outputs the received data to the control unit 11.

[0031] The operation unit 15 includes operation devices such as a touch panel, a keyboard, and switches, and receives various operations and data input by the user. The control unit 11 performs appropriate control based on various pieces of operation information provided by the operation unit 15, and stores the input data in the storage unit 12 as necessary.

[0032] The display unit 16 includes a display device such as a liquid crystal monitor or an organic EL (Electro-Luminescence) display, and displays information to be notified to the user or the like in response to an instruction from the control unit 11.

[0033] In this embodiment, the estimation device 1 is configured to include the learning model LM, but the learning model LM may be stored in an external device accessible from the estimation device 1. In this case, the estimation device 1 transmits a calculation instruction and data required for the calculation to the external device to cause the external device to execute a calculation using the learning model LM, and receives a calculation result using the learning model LM from the external device.

[0034] The estimation device 1 is not limited to a single computer, but may be a computer system including multiple computers and peripheral devices. The estimation device 1 may also be a virtual machine virtually constructed by software.

[0035] 4 is a schematic diagram showing a configuration example of the learning model LM. The learning model LM in the first embodiment is a learning model constructed by, for example, a convolutional neural network (CNN), and includes a first network layer NN1, a second network layer NN2, and a third network layer NN3.

[0036] The first network layer NN1 includes a convolution layer and a pooling layer. In Fig. 4, "Conv" represents a convolution layer, and "MaxPooling" represents a max pooling layer. When first waveform data obtained from the eddy current flaw detection probe 20 is input, the first network layer NN1 is configured to extract a first feature amount that indicates the characteristics of the first waveform data and output the extracted first feature amount to the subsequent third network layer NN3.

[0037] The second network layer NN2 includes a convolution layer and a pooling layer, similar to the first network layer NN1. When second waveform data obtained from the ultrasonic flaw detection probe 30 is input, the second network layer NN2 is configured to extract second feature amounts that indicate characteristics of the second waveform data and output the extracted second feature amounts to the subsequent third network layer NN3.

[0038] The third network layer NN3 includes a concatenation layer, a smoothing layer, a linear layer, and a dropout layer. In FIG. 4, "Concatenate" represents the concatenation layer, "Flatten" represents the smoothing layer, "Linear" represents the linear layer, and "Dropout" represents the dropout layer. The third network layer NN3 is configured to concatenate the first feature output from the first network layer NN1 and the second feature output from the second network layer NN2 in the concatenation layer, and to sequentially process the output of the concatenation layer in each subsequent layer to output information about the cable type of the underground power transmission cable 4. In this embodiment, the final linear layer (Linear4) is constructed using a softmax function, and the probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable and the probability P2 that it is estimated to be a non-waterproof cable are output as information about the cable type.

[0039] In this embodiment, the learning model LM constructed by CNN has been described, but the learning model LM is not limited to CNN, and may be a model constructed by R-CNN (Region-based CNN), RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), etc.

[0040] In the configuration example of FIG. 4 , the first network layer NN1 and the second network layer NN2 each include three convolutional layers and three pooling layers, but the number of convolutional layers and the number of pooling layers are not limited to three. The first network layer NN1 and the second network layer NN2 are designed appropriately depending on the size of the input waveform data, etc. Furthermore, in the configuration example of FIG. 4 , the third network layer NN3 includes one concatenation layer, one smoothing layer, four linear layers, and two dropout layers, but the number and arrangement of each layer are not limited to those shown in FIG. 4. The number and arrangement of each layer in the third network layer NN3 are designed appropriately depending on the input feature quantities, the information to be output, etc.

[0041] A method for generating the learning model LM will be described below. In the preparation stage for generating the learning model LM, measurements are performed using an eddy current flaw detector 2 and an ultrasonic flaw detector 3 on an underground power transmission cable 4 whose cable type is known, and a data set is created that associates type information (correct label) indicating the cable type, first waveform data obtained from the eddy current flaw detector probe 20, and second waveform data obtained from the ultrasonic flaw detector probe 30.

[0042] FIG. 5 is a conceptual diagram illustrating an example of a data set. The data set used as training data for learning includes eddy current waveform data (first waveform data), ultrasonic waveform data (second waveform data), and type information indicating the cable type. The first waveform data is numerical data or image data indicating the time series changes in eddy currents measured by the eddy current flaw detector 2 and is stored as a file within the device. Therefore, a file name specifying that file is registered in the eddy current waveform field of the data set. The second waveform data is numerical data or image data indicating the time series changes in ultrasonic waves measured by the ultrasonic flaw detector 3 and is stored as a file within the device. Therefore, a file name specifying that file is registered in the ultrasonic waveform field of the data set. Note that, for simplicity, the example data set in FIG. 5 lists only one file name for each underground power transmission cable 4. However, in practice, multiple first waveform data and second waveform data are acquired for each underground power transmission cable 4, and multiple files recording this waveform data are stored in association with type information for the cable type.

[0043] The cable type information includes information related to the waterproof layer 44. The information related to the waterproof layer 44 includes information indicating whether the underground power transmission cable 4 has a waterproof layer 44. If the underground power transmission cable 4 has a waterproof layer 44, information indicating the material of the waterproof layer 44 may be included. The waterproof layer 44 is often made of aluminum, but lead may also be used.

[0044] The type information of the cable type may further include information on the voltage, cross-sectional area, year of manufacture, shielding layer 43, manufacturer name, insulation thickness, sheath thickness, and remarks. Here, the information on the shielding layer 43 includes information on the structure of the shielding layer 43 provided in the underground power transmission cable 4. Specifically, it includes information on whether the shielding layer 43 is made of copper tape or wire shield (WS). The remarks include, for example, information indicating whether disaster prevention tape is used.

[0045] The learning data set is stored, for example, in the storage unit 12 of the estimation device 1. The estimation device 1 uses the data set stored in the storage unit 12 as training data to generate a learning model LM.

[0046] 6 is a flowchart illustrating the procedure for generating the learning model LM. The control unit 11 of the estimation device 1 selects a set of training data including first waveform data, second waveform data, and type information indicating the cable type from a data set prepared in advance (step S101). The control unit 11 inputs the first waveform data and second waveform data included in the selected training data into the learning model LM, and executes calculations using the learning model LM (step S102). Note that, before learning begins, initial values ​​are set for the model parameters of the learning model LM.

[0047] The control unit 11 evaluates the calculation results of the learning model LM (step S103) and determines whether learning is complete (step S104). The control unit 11 can evaluate the calculation results using an error function (also referred to as an objective function, loss function, or cost function) set based on the calculation results of the learning model LM and the correct label (presence or absence of the water-impermeable layer 44) included in the training data. For example, the control unit 11 determines that learning is complete when the error function becomes equal to or less than a threshold (or equal to or greater than a threshold) during the process of optimizing (minimizing or maximizing) the error function using a gradient descent method such as steepest descent.

[0048] If it is determined that learning is not complete (S104: NO), the control unit 11 updates the parameters of the learning model LM (such as weights and biases between nodes) (step S105) and returns the process to step S101. The control unit 11 can update the parameters in the learning model LM using an error backpropagation method that sequentially updates the weights and biases between nodes from the third network layer NN3 to the first network layer NN1 and the second network layer NN2.

[0049] If it is determined that the learning is completed (S104: YES), the learned learning model LM is obtained, and the control unit 11 stores the learned learning model LM in the storage unit 12 (step S106).

[0050] In this embodiment, the learning dataset is stored in the storage unit 12 of the estimation device 1, but it may also be stored in a storage device external to the estimation device 1. In this case, the estimation device 1 accesses the external storage device and learns the learning model LM using training data acquired from the external storage device. Furthermore, in this embodiment, the estimation device 1 is configured to generate the learning model LM, but the learning model LM may also be generated by an information processing device such as an external server. In this case, the estimation device 1 acquires the trained learning model LM generated externally via communication or a recording medium and stores the acquired learning model LM in the storage unit 12.

[0051] Next, a description will be given of the type estimation process that the estimation device 1 executes after the learning is completed. 7 is a flowchart illustrating the execution procedure of the type estimation process executed by the estimation device 1. An operator performs measurements on an underground power transmission cable 4 of unknown cable type using an eddy current flaw detector 2 and an ultrasonic flaw detector 3. The estimation device 1 acquires first waveform data from the eddy current flaw detector 2 via, for example, the input unit 13 (step S121), and acquires second waveform data from the ultrasonic flaw detector 3 (step S122). The processing procedures of steps S121 and S122 may be reversed.

[0052] The control unit 11 of the estimation device 1 inputs the acquired first waveform data and second waveform data to the learning model LM and executes a calculation using the learning model LM (step S123). That is, the control unit 11 inputs the first waveform data to the first network layer NN1 and extracts a feature quantity (first feature quantity) of the first waveform data by performing a calculation using the first network layer NN1. The control unit 11 also inputs the second waveform data to the second network layer NN2 and extracts a feature quantity (second feature quantity) of the second waveform data by performing a calculation using the second network layer NN2. The control unit 11 passes the first feature quantity obtained from the first network layer NN1 and the second feature quantity obtained from the second network layer NN2 to the third network layer NN3, concatenates them in the concatenation layer, and then executes calculations in each subsequent layer to obtain a final output. Specifically, the control unit 11 obtains, as final outputs, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable, and a probability P2 that the target underground power transmission cable 4 is estimated to be a non-waterproof cable.

[0053] The control unit 11 estimates the cable type of the underground power transmission cable 4 based on the calculation results of the learning model LM (step S124). That is, the control unit 11 compares the probabilities P1 and P2 output from the third network layer NN3 of the learning model LM, and estimates the cable to be a waterproof cable if the probability P1 is higher than the probability P2, and estimates the cable to be a non-waterproof cable if the probability P1 is lower than the probability P2.

[0054] The control unit 11 outputs the cable type estimation result (step S125). The control unit 11 causes information indicating the cable type (text information or image information) to be displayed on the display unit 16. Alternatively, the control unit 11 may output the information indicating the cable type to the communication unit 14, and transmit the information to an external terminal device from the communication unit 14.

[0055] The inventors of the present application divided a dataset prepared in advance into two subsets, one for training and one for validation, and generated a learning model LM using the training subset. They then evaluated the detection rate and false alarm rate for cable types using the validation subset. Here, the detection rate indicates the percentage of correct answers out of the total number of data. The false alarm rate indicates the percentage of cases where a water-sealed cable was mistakenly identified as a non-water-sealed cable, and where a non-water-sealed cable was mistakenly identified as a water-sealed cable. As a result of the validation, the learning model LM according to embodiment 1 achieved a detection rate of 99.9% and a false alarm rate of 0.1%.

[0056] As described above, in embodiment 1, the eddy current waveform data (first waveform data) obtained from the eddy current testing probe 20 and the ultrasonic waveform data (second waveform data) obtained from the ultrasonic testing probe 30 are input into the learning model LM, and the cable type is estimated based on the calculation results by the learning model LM.Therefore, even though there are various types of underground power transmission cables 4 that differ in conductor size, manufacturing year, material of the water-shielding layer 44, structure of the shielding layer 43, insulation thickness, sheath thickness, etc., it is possible to accurately estimate the presence or absence of a water-shielding layer 44 without relying on the skill of the inspector.

[0057] (Embodiment 2) In the second embodiment, a configuration for estimating the material of the water shielding layer 44 using a learning model LM will be described. The overall configuration of the system, the internal configuration of the estimation device 1, etc. are the same as those in the first embodiment, and therefore a description thereof will be omitted.

[0058] FIG. 8 is a schematic diagram showing the overall configuration of the learning model LM in the second embodiment. The configuration of the learning model LM in the second embodiment is the same as that in the first embodiment, and includes a first network layer NN1, a second network layer NN2, and a third network layer NN3. The final linear layer (Linear4) of the third network layer NN3 includes three nodes, each of which outputs a probability P1 that the water shielding layer 44 provided on the underground power transmission cable 4 is aluminum, a probability P2 that the water shielding layer 44 provided on the underground power transmission cable 4 is lead, and a probability P3 that the underground power transmission cable 4 does not include a water shielding layer 44. Such a learning model LM is generated by performing learning using, from the information contained in the dataset, information on the presence or absence of the water shielding layer 44 and information on the material of the water shielding layer 44 as correct labels. The learning procedure is the same as that in the first embodiment.

[0059] The control unit 11 of the estimation device 1 inputs the first waveform data and the second waveform data into the trained learning model LM and executes calculations using the learning model LM to estimate the cable type. For example, if the control unit 11 determines that the probability P1 is the highest as a result of calculations using the learning model LM, it estimates that the target underground power transmission cable 4 is a waterproof cable with an aluminum water-shielding layer 44. If the control unit 11 determines that the probability P2 is the highest as a result of calculations using the learning model LM, it estimates that the target underground power transmission cable 4 is a waterproof cable with a lead water-shielding layer 44. If the control unit 11 determines that the probability P3 is the highest as a result of calculations using the learning model LM, it estimates that the target underground power transmission cable 4 is a non-water-shielding cable.

[0060] As described above, in the second embodiment, by using the learning model LM, it is possible to estimate not only the presence or absence of the water shielding layer 44 but also the material of the water shielding layer 44.

[0061] (Embodiment 3) In the third embodiment, a learning model LM that can estimate the cable type regardless of the measurement location will be described. The overall configuration of the system, the internal configuration of the estimation device 1, etc. are the same as those in the first embodiment, and therefore a description thereof will be omitted.

[0062] As described above, when the underground power transmission cable 4 is provided with the water-impermeable layer 44, the water-impermeable layer 44 has a portion that overlaps in the radial direction (two-layer portion 44a) and a portion that does not overlap in the radial direction (single-layer portion 44b). Therefore, the characteristics of the waveform data obtained may differ depending on whether the two-layer portion 44a or the single-layer portion 44b is measured by the eddy current flaw detection probe 20 and the ultrasonic flaw detection probe 30.

[0063] In embodiment 3, a learning model LM capable of estimating the cable type regardless of the measurement location is generated by learning using both waveform data obtained by measuring the two-piece portion 44a and waveform data obtained by measuring the one-piece portion 44b as training data.

[0064] 9 is a schematic diagram showing the general configuration of the learning model LM in embodiment 3. The configuration of the learning model LM in embodiment 3 is the same as that in embodiment 1, and includes a first network layer NN1, a second network layer NN2, and a third network layer NN3. The final linear layer (Linear4) of the third network layer NN3 includes, for example, two nodes, and each node outputs the probability P1 that the underground power transmission cable 4 is estimated to be a waterproof cable and the probability P2 that it is estimated to be a non-waterproof cable, respectively.

[0065] The learning model LM according to the third embodiment is generated by learning using, as training data, a dataset including waveform data obtained by measuring the two-piece portion 44a and type information indicating the cable type, a dataset including waveform data obtained by measuring the one-piece portion 44b and type information indicating the cable type, and a dataset including waveform data obtained by measuring a non-waterproof cable and type information indicating the cable type. Note that it is not necessary for both the first waveform data and the second waveform data to be data obtained by measuring the two-piece portion 44a (or the one-piece portion 44b); one of the first waveform data and the second waveform data may be data obtained by measuring the two-piece portion 44a, and the other may be data obtained by measuring the one-piece portion 44b.

[0066] After the trained learning model LM is obtained, the estimation device 1 inputs waveform data obtained when scanning the underground power transmission cable 4 into the trained learning model LM, executes calculations using the learning model LM, and estimates the cable type of the underground power transmission cable 4 based on the information output from the learning model LM. The waveform data (first waveform data and second waveform data) input into the learning model LM may be data obtained by measuring the two-piece portion 44a or the single-piece portion 44b. In other words, the operator does not need to measure the two-piece portion 44a and the single-piece portion 44b separately, but can simply measure any location on the underground power transmission cable 4 and input the obtained waveform data into the learning model LM.

[0067] The learning model LM shown in Figure 3 is a learning model for estimating the presence or absence of a water-shielding layer 44, but as described in embodiment 2, it may also be a learning model for estimating the material of the water-shielding layer 44.

[0068] As described above, in embodiment 3, by using a learning model LM that has been trained using both waveform data obtained by measuring the two-piece portion 44a and waveform data obtained by measuring the one-piece portion 44b as training data, the cable type can be estimated regardless of the measurement location.

[0069] (Fourth embodiment) In the fourth embodiment, a configuration for estimating the structure of the shielding layer 43 using a learning model LM will be described. The overall configuration of the system, the internal configuration of the estimation device 1, etc. are the same as those in the first embodiment, and therefore a description thereof will be omitted.

[0070] FIG. 10 is a schematic diagram showing the general configuration of the learning model LM in the fourth embodiment. The configuration of the learning model LM in the fourth embodiment is the same as that in the first embodiment, and includes a first network layer NN1, a second network layer NN2, and a third network layer NN3. The final linear layer (Linear4) of the third network layer NN3 includes two nodes, each of which outputs a probability P1 that the shielding layer 43 is made of copper tape and a probability P2 that the shielding layer 43 is made of wire shield. Such a learning model LM is generated by performing learning using information about the structure of the shielding layer 43, which is included in the data set, as a correct label. The learning procedure is the same as that in the first embodiment.

[0071] The control unit 11 of the estimation device 1 inputs the first waveform data and the second waveform data into the trained learning model LM and executes calculations using the learning model LM to estimate the cable type. For example, if the control unit 11 determines that the probability P1 is higher than the probability P2 as a result of the calculations using the learning model LM, it estimates that the target underground power transmission cable 4 is a cable equipped with a shielding layer 43 made of copper tape. Furthermore, if the control unit 11 determines that the probability P2 is higher than the probability P1 as a result of the calculations using the learning model LM, it estimates that the target underground power transmission cable 4 is a cable equipped with a shielding layer 43 made of wire seal.

[0072] As described above, in the fourth embodiment, the structure of the shielding layer 43 can be estimated by using the learning model LM.

[0073] In embodiment 1, a configuration for estimating the presence or absence of a water-shielding layer 44 was described, in embodiment 2 a configuration for estimating the material of the water-shielding layer 44 was described, and in embodiment 4 a configuration for estimating the structure of the shielding layer 43 was described. However, a configuration for estimating two or more of the presence or absence of a water-shielding layer 44, the material of the water-shielding layer 44, and the structure of the shielding layer 43 may also be used using one learning model LM.

[0074] (Embodiment 5) In the fifth embodiment, a configuration will be described in which the cable type is estimated based on an estimation result based on the first waveform data and an estimation result based on the second waveform data.

[0075] 11 is a block diagram illustrating the internal configuration of an estimation device 1 according to embodiment 5. The estimation device 1 is a general-purpose or dedicated computer, and includes a control unit 11, a storage unit 12, an input unit 13, a communication unit 14, an operation unit 15, and a display unit 16. The configuration of each of these hardware units is the same as in embodiment 1.

[0076] In the fifth embodiment, a first learning model LM1 and a second learning model LM2, which will be described later, are installed in the storage unit 12. The control unit 11 of the estimation device 1 uses the first learning model LM1 and the second learning model LM2 to estimate the cable type.

[0077] 12 is a schematic diagram showing an example of the configuration of the first learning model LM1. The first learning model LM1 is a learning model constructed by, for example, CNN, and includes a first network layer NN11 and a second network layer NN12.

[0078] The first network layer NN11 has a convolution layer and a pooling layer, and when first waveform data obtained from the eddy current flaw detection probe 20 is input, it extracts a first feature that indicates the characteristics of the first waveform data and outputs the extracted first feature to the subsequent second network layer NN12.

[0079] The second network layer NN12 includes a smoothing layer, a linear layer, and a dropout layer, and performs calculations in each layer in sequence in response to the input of the first feature amount, outputting information about the cable type of the underground power transmission cable 4. The final linear layer (Linear4) of the second network layer NN12 outputs, as information about the cable type, for example, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable, and a probability P2 that it is estimated to be a non-waterproof cable.

[0080] That is, the first learning model LM1 according to the fifth embodiment is configured to output information about the cable type when the first waveform data obtained from the eddy current flaw detection probe 20 is input alone. The procedure for generating the first learning model LM1 is the same as that of the first embodiment, and the first learning model LM1 is generated by learning using, as training data, a data set including eddy current waveform data (first waveform data) and type information indicating the cable type.

[0081] The inventors divided a prepared dataset into two subsets, one for training and one for validation, and generated a first learning model LM1 using the training subset. They then evaluated the detection rate and false alarm rate of cable types using the validation subset. As a result of the validation, the first learning model LM1 achieved a detection rate of 94.5% and a false alarm rate of 4.4%.

[0082] 13 is a schematic diagram showing an example of the configuration of the second learning model LM2. The second learning model LM2 is a learning model constructed by, for example, CNN, and includes a second network layer NN21 and a second network layer NN22.

[0083] The first network layer NN21 has a convolution layer and a pooling layer, and when second waveform data obtained from the ultrasonic flaw detection probe 30 is input, it extracts second features that indicate the characteristics of the second waveform data and outputs the extracted second features to the subsequent second network layer NN22.

[0084] The second network layer NN22 includes a smoothing layer, a linear layer, and a dropout layer, and performs calculations in each layer in sequence in response to the input of the second feature amount, outputting information about the cable type of the underground power transmission cable 4. The final linear layer (Linear4) of the second network layer NN22 outputs information about the cable type, for example, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable, and a probability P2 that it is estimated to be a non-waterproof cable.

[0085] That is, the second learning model LM2 according to the fifth embodiment is configured to output information about the cable type when the second waveform data obtained from the ultrasonic flaw detection probe 30 is input alone. The procedure for generating the second learning model LM2 is the same as that of the first embodiment, and the second learning model LM2 is generated by learning using as training data a dataset including ultrasonic waveform data (second waveform data) and type information indicating the cable type.

[0086] The inventors divided a prepared dataset into two subsets, one for training and one for validation, and used the training subset to generate a second learning model LM2. They then evaluated the detection rate and false alarm rate for cable types using the validation subset. As a result of the validation, the second learning model LM2 achieved a detection rate of 99.5% and a false alarm rate of 0.2%.

[0087] 14 is a flowchart illustrating the execution procedure of the type estimation process executed by the estimation device 1 according to embodiment 5. An operator performs measurement of an underground power transmission cable 4 of unknown cable type using an eddy current flaw detector 2. The estimation device 1 acquires first waveform data from the eddy current flaw detector 2, for example, via the input unit 13 (step S501).

[0088] The control unit 11 of the estimation device 1 inputs the acquired first waveform data into the first learning model LM1 and executes calculations using the first learning model LM1 (step S502). As a result of the calculations, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable and a probability P2 that the target underground power transmission cable 4 is estimated to be a non-waterproof cable are obtained.

[0089] The control unit 11 estimates the cable type of the underground power transmission cable 4 based on the calculation result of the first learning model LM1 (step S503). That is, the control unit 11 compares the magnitudes of the probabilities P1 and P2 output from the second network layer NN12 of the first learning model LM1, and estimates the cable to be a waterproof cable if the probability P1 is higher than the probability P2, and estimates the cable to be a non-waterproof cable if the probability P1 is lower than the probability P2.

[0090] The control unit 11 determines whether the confidence level of the estimation result is equal to or higher than a threshold value (e.g., 95% or higher) (step S504). If the cable is estimated to be a waterproof cable in step S503, the confidence level of the estimation result is represented by a probability P1, and if the cable is estimated to be a non-waterproof cable, the confidence level of the estimation result is represented by a probability P2.

[0091] If it is determined that the confidence level is equal to or greater than the threshold (S504: YES), the control unit 11 outputs the estimation result of the cable type (step S505). Here, it is sufficient to output the estimation result using the first learning model LM1. The control unit 11 causes the display unit 16 to display information (text information or image information) indicating the cable type. Alternatively, the control unit 11 may output the information indicating the cable type to the communication unit 14, and transmit it from the communication unit 14 to an external terminal device.

[0092] If it is determined that the confidence level is less than the threshold (S504: NO), the control unit 11 acquires second waveform data (step S506). If the confidence level is less than the threshold, the second waveform data obtained from the ultrasonic flaw detector 3 is required, and therefore the control unit 11 may prompt the worker to perform measurement using the ultrasonic flaw detector 3. Specifically, the control unit 11 causes the display unit 16 to display information that the underground power transmission cable 4 should be measured using the ultrasonic flaw detector 3. Alternatively, the control unit 11 may transmit the information to a terminal device carried by the worker.

[0093] The control unit 11 of the estimation device 1 inputs the acquired second waveform data into the second learning model LM2 and executes calculations using the second learning model LM2 (step S507). As a result of the calculations, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable and a probability P2 that the target underground power transmission cable 4 is estimated to be a non-waterproof cable are obtained.

[0094] The control unit 11 estimates the cable type of the underground power transmission cable 4 based on the calculation result of the second learning model LM2 (step S508). That is, the control unit 11 compares the magnitudes of the probabilities P1 and P2 output from the second network layer NN22 of the second learning model LM2, and estimates the cable to be a waterproof cable if the probability P1 is higher than the probability P2, and estimates the cable to be a non-waterproof cable if the probability P1 is lower than the probability P2.

[0095] When an estimation result is obtained using the second learning model LM2, the control unit 11 outputs the estimation result of the cable type (step S505). Here, it is sufficient to output the estimation result using the second learning model LM2. The control unit 11 causes the display unit 16 to display information (text information or image information) indicating the cable type. Alternatively, the control unit 11 may output the information indicating the cable type to the communication unit 14, and transmit it from the communication unit 14 to an external terminal device.

[0096] As described above, in the fifth embodiment, the cable type is estimated using the first learning model LM1, and if the confidence level is equal to or higher than the threshold, the estimation result by the first learning model LM1 is output. In this case, only measurement by the eddy current flaw detector 2 is required, thereby reducing the workload of the worker. On the other hand, if the confidence level is lower than the threshold, second waveform data is acquired, and the estimation result by the second learning model LM2 is output. In this case, measurement by the eddy current flaw detector 2 and measurement by the ultrasonic flaw detector 3 are required, but if it is determined that the estimation result using the eddy current waveform data is not satisfactory, type estimation is performed using the ultrasonic waveform data, thereby improving the estimation accuracy.

[0097] In the flowchart shown in Figure 14, when an estimation result is obtained using the second learning model LM2, the procedure is to output the obtained estimation result as is, but when an estimation result is obtained in step S508, the confidence level of the estimation result may be compared with a threshold value, and the estimation result may be output only if the confidence level is equal to or greater than the threshold value, and if the confidence level is less than the threshold value, remeasurement may be prompted.

[0098] In addition, in the flowchart of Figure 14, type estimation using the first learning model LM1 is performed first, and if the confidence level of the estimation result is less than a threshold, type estimation using the second learning model LM2 is performed.However, it is also possible to perform type estimation using the second learning model LM2 first, and if the confidence level of the estimation result is less than a threshold, type estimation using the first learning model LM1.

[0099] (Sixth embodiment) In embodiment 6, a modified example of embodiment 5 will be described. The estimation device 1 is the same as in embodiment 5, and estimates the cable type using the estimation result by the first learning model LM1 and the estimation result by the second learning model LM2.

[0100] 15 is a flowchart illustrating the execution procedure of the type estimation process executed by the estimation device 1 according to embodiment 6. An operator performs measurements on an underground power transmission cable 4 of unknown cable type using an eddy current flaw detector 2 and an ultrasonic flaw detector 3. The estimation device 1 acquires first waveform data from the eddy current flaw detector 2 via, for example, the input unit 13 (step S601).

[0101] The control unit 11 of the estimation device 1 inputs the acquired first waveform data into the first learning model LM1 and executes calculations using the first learning model LM1 (step S602). As a result of the calculations, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable and a probability P2 that the target underground power transmission cable 4 is estimated to be a non-waterproof cable are obtained.

[0102] The control unit 11 estimates the cable type of the underground power transmission cable 4 based on the calculation result of the first learning model LM1 (step S603). That is, the control unit 11 compares the magnitudes of the probabilities P1 and P2 output from the second network layer NN12 of the first learning model LM1, and estimates the cable to be a waterproof cable if the probability P1 is higher than the probability P2, and estimates the cable to be a non-waterproof cable if the probability P1 is lower than the probability P2.

[0103] The estimation device 1 acquires the second waveform data from the ultrasonic flaw detector 3 via, for example, the input unit 13 (step S604).

[0104] The control unit 11 of the estimation device 1 inputs the acquired second waveform data into the second learning model LM2 and executes calculations using the second learning model LM2 (step S605). As a result of the calculations, a probability P1 that the target underground power transmission cable 4 is estimated to be a waterproof cable and a probability P2 that the target underground power transmission cable 4 is estimated to be a non-waterproof cable are obtained.

[0105] The control unit 11 estimates the cable type of the underground power transmission cable 4 based on the calculation result of the second learning model LM2 (step S606). That is, the control unit 11 compares the magnitudes of the probabilities P1 and P2 output from the second network layer NN22 of the second learning model LM2, and estimates the cable to be a waterproof cable if the probability P1 is higher than the probability P2, and estimates the cable to be a non-waterproof cable if the probability P1 is lower than the probability P2.

[0106] The control unit 11 compares the confidence level of the estimation result by the first learning model LM1 with the confidence level of the estimation result by the second learning model LM2, and selects one of the estimation results depending on the comparison result of the confidence levels (step S607). If the estimation result by the first learning model LM1 and the estimation result by the second learning model LM2 differ, i.e., if one estimates the cable as waterproof and the other estimates the cable as non-waterproof, the control unit 11 may select the estimation result with the higher confidence level.

[0107] The control unit 11 outputs the selected estimation result (step S608). Specifically, the control unit 11 causes the display unit 16 to display information (text information or image information) indicating the cable type. Alternatively, the control unit 11 may output the information indicating the cable type to the communication unit 14, and transmit the information from the communication unit 14 to an external terminal device.

[0108] As described above, in embodiment 6, the estimation result with the higher degree of certainty is output from the estimation result by the first learning model LM1 and the estimation result by the second learning model LM2. Therefore, if there is a malfunction in either the eddy current flaw detection device 2 or the ultrasonic flaw detection device 3, it is possible to avoid a deterioration in estimation accuracy due to being influenced by the waveform data output from the malfunctioning device.

[0109] In the flowchart of Figure 15, the procedure is such that type estimation using the first learning model LM1 is performed first, and then type estimation using the second learning model LM2 is performed, but the procedure may also be such that type estimation using the second learning model LM2 is performed first, and then type estimation using the first learning model LM1 is performed.

[0110] In the fifth and sixth embodiments, the configuration is such that the presence or absence of the water-shielding layer 44 is estimated, but as described in the second to fourth embodiments, the configuration may also be such that at least one of the material of the water-shielding layer 44 and the structure of the shielding layer 43 is estimated.

[0111] The embodiments disclosed herein should be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0112] 1 Estimation device 2 Eddy current flaw detection equipment 3 Ultrasonic flaw detection equipment 4. Underground power transmission cables 11 Control section 12 Storage section 13 Input section 14 Communications Department 15 Control section 16 Display section 20 Eddy current testing probe 30 Ultrasonic flaw detection probe 41 Conductor 42 Insulators 43 Shielding layer 44 Water impermeable layer 45 Sheath PG1 Estimation Processing Program LM Learning Model LM1 First Learning Model LM2 Second Learning Model

Claims

1. acquiring waveform data obtained by the flaw detection probe when scanning the underground power transmission cable with the flaw detection probe; inputting the acquired waveform data into a learning model that has been trained to output information about a cable type when waveform data obtained from a flaw detection probe is input, and estimating the cable type of the underground power transmission cable; Output the estimation results A computer program that causes a computer to execute a process.

2. The flaw detection probe includes an eddy current flaw detection probe or an ultrasonic flaw detection probe.

2. The computer program of claim 1.

3. the learning model is trained to output information about a cable type when first waveform data obtained from an eddy current flaw detection probe and second waveform data obtained from an ultrasonic flaw detection probe are input; First waveform data obtained by the eddy current testing probe when scanning the underground power transmission cable and second waveform data obtained by the ultrasonic testing probe are input to the learning model to estimate the type of the underground power transmission cable.

3. A computer program according to claim 1 or 2, for causing a computer to execute a process.

4. The information about the cable type includes information for determining whether or not a water impermeable layer is present, Based on the information output from the learning model, it is estimated whether the underground power transmission cable is a cable having a water impermeable layer.

4. A computer program according to claim 1, for causing a computer to execute a process.

5. The information about the cable type includes information about the material of the water impermeable layer, The material of the water shielding layer is estimated based on the information output from the learning model.

5. A computer program product according to claim 4, for causing the computer to execute a process.

6. the learning model is trained using, as training data, both waveform data measured for a portion of the water-shielding layer that has overlapping in the radial direction and waveform data measured for a portion of the water-shielding layer that does not have overlapping in the radial direction; Whether the acquired waveform data is waveform data measured for a portion having overlap in the radial direction or waveform data measured for a portion having no overlap in the radial direction, the presence or absence of a water shielding layer or the material of the water shielding layer is estimated based on the information output from the learning model.

6. A computer program according to claim 4 or 5, for causing the computer to execute processing.

7. the information about the cable type includes information for identifying the structure of the shielding layer; The structure of the shielding layer of the underground power transmission cable is estimated based on the information output from the learning model.

7. A computer program according to claim 1, for causing a computer to execute a process.

8. acquiring first waveform data obtained by the eddy current testing probe when scanning the underground power transmission cable with the eddy current testing probe; inputting the acquired first waveform data obtained from an eddy current flaw detection probe into a first learning model that has been trained to output information about a cable type when the acquired first waveform data is input, thereby estimating the cable type of the underground power transmission cable; If the confidence level of the estimation result by the first learning model is less than a threshold, second waveform data obtained by the ultrasonic flaw detection probe when the underground power transmission cable is scanned by the ultrasonic flaw detection probe is acquired; inputting the acquired second waveform data obtained from the ultrasonic flaw detection probe into a second learning model that is trained to output information about the cable type when the second waveform data is input, thereby estimating the cable type of the underground power transmission cable; outputting an estimation result estimated using the first learning model or the second learning model; A computer program that causes a computer to execute a process.

9. acquiring first waveform data obtained by the eddy current testing probe when scanning the underground power transmission cable with the eddy current testing probe; inputting the acquired first waveform data obtained from an eddy current flaw detection probe into a first learning model that has been trained to output information about a cable type when the acquired first waveform data is input, thereby estimating the cable type of the underground power transmission cable; acquiring second waveform data obtained by the ultrasonic flaw detection probe when the underground power transmission cable is scanned with the ultrasonic flaw detection probe; inputting the acquired second waveform data obtained from the ultrasonic flaw detection probe into a second learning model that is trained to output information about the cable type when the second waveform data is input, thereby estimating the cable type of the underground power transmission cable; Either one of the estimation results is output depending on the degree of certainty between the estimation result by the first learning model and the estimation result by the second learning model. A computer program that causes a computer to execute a process.

10. a neural network that has learned a relationship between the waveform data and the type of the underground power transmission cable by using, as training data, a dataset including waveform data obtained from the flaw detection probe when the underground power transmission cable is scanned with the flaw detection probe and information indicating the type of the underground power transmission cable; When waveform data obtained from the flaw detection probe is input, the computer is caused to execute calculations using the neural network and output information relating to the type of underground power transmission cable. Learning model.

11. The neural network a first network layer that, in response to input of first waveform data obtained by an eddy current testing probe when the underground power transmission cable is scanned with the eddy current testing probe, outputs a first feature quantity that indicates a feature of the first waveform data; a second network layer configured to, in response to input of second waveform data obtained by the ultrasonic testing probe when the underground power transmission cable is scanned with the ultrasonic testing probe, output a second feature quantity indicating a feature of the second waveform data; a third network layer including a connection layer that connects the first feature amount output from the first network layer and the second feature amount output from the second network layer, and that processes the output from the connection layer to output information about the cable type of the underground power transmission cable; Including, The computer is caused to function to output information about the type of the underground power transmission cable by executing an operation by the first network layer to extract the first feature amount from the first waveform data, an operation by the second network layer to extract the second feature amount from the second waveform data, and an operation by the third network layer to connect the first feature amount and the second feature amount by the connection layer and process an output from the connection layer. The learning model according to claim 10.

12. an acquisition unit that acquires waveform data obtained by the flaw detection probe when the flaw detection probe scans the underground power transmission cable; an estimation unit that inputs the waveform data acquired by the acquisition unit into a learning model that has been trained to output information about a cable type when waveform data obtained from a flaw detection probe is input, and estimates the cable type of the underground power transmission cable; An output unit that outputs the estimation results; An estimation device comprising:

13. acquiring waveform data obtained by the flaw detection probe when scanning the underground power transmission cable with the flaw detection probe; inputting the acquired waveform data into a learning model that has been trained to output information about a cable type when waveform data obtained from a flaw detection probe is input, and estimating the cable type of the underground power transmission cable; Output the estimation results A computer-implemented estimation method.

Images