Measuring equipment

Abstract

To make it possible to measure a thickness of a friction material of a friction brake of a vehicle.SOLUTION: A measurement device measures a thickness D of a friction material of a friction brake of a vehicle. The measurement device comprises: an imaging unit that photographs an image; an image acquisition unit 78 that acquires an image IMG photographed by the imaging unit and including other members adjacent to the friction material and the friction material, of components of the friction brake, as a measurement image IMG1; and a thickness derivation unit 80 that inputs the measurement image IMG1 to a learned model performed machine learning about estimating the thickness of the friction material, and derives the thickness D of the friction material on the basis of an index output from the learned model.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a measuring device for measuring the thickness of a friction material of a friction brake of a vehicle. [Background technology]

[0002] Patent Document 1 discloses a measuring device for measuring the thickness of brake linings for elevator brakes. This measuring device measures the thickness of the brake lining based on a measurement image that includes the brake lining and a marker provided on the brake lining or braking part. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-56391 Summary of the Invention [Problem to be solved by the invention]

[0004] In elevator brakes, it is unlikely that the brake linings will be replaced with ones of a different model or from a different manufacturer. However, in vehicle friction brakes, the friction material and other components adjacent to the friction material may be replaced with ones of a different model or from a different manufacturer. Therefore, if the measurement object is the friction material of a vehicle friction brake using the above-mentioned measuring device, the images containing the measurement object will be diverse, making measuring the thickness of the friction material cumbersome. An object of the present invention is to provide a measuring device that can easily measure the thickness of the friction material of a vehicle friction brake. [Means for solving the problem]

[0005] The measuring device for solving the above problem is a measuring device for measuring the thickness of a friction material of a friction brake of a vehicle. This measuring device comprises an imaging unit that captures an image, an image acquisition unit that acquires an image captured by the imaging unit, the image including the friction material and other members adjacent to the friction material among the components of the friction brake, as a measurement image, and a trained model that has undergone machine learning to estimate the thickness of the friction material from the image including the friction material and the other members, and outputs the measurement image acquired by the image acquisition unit. Correlation value with the thickness of the friction material and a thickness deriving section for deriving the thickness of the friction material based on the thickness deriving section.

[0006] In the above configuration, by inputting measurement images captured by the imaging unit into a trained model generated by machine learning using a wide variety of images, an index correlated with the thickness of the friction material is output from the trained model, which allows the thickness of the friction material of a vehicle's friction brake to be easily measured based on the index. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a side view schematically showing a friction brake of a vehicle. [Figure 2] FIG. 2 is a plan view showing a schematic view of a part of the friction brake. [Figure 3] FIG. 3 is a diagram showing a schematic configuration of the measurement device of the first embodiment. [Figure 4] FIG. 4 is a block diagram showing the functions of the measuring device. [Figure 5] In FIG. 5, (a) is a schematic diagram showing the positional relationship between the imaging unit and the imaging target, and (b) is a schematic diagram showing markers displayed on the imaging target. [Figure 6] FIG. 6 is a flowchart illustrating the flow of processing in the measurement device of the first embodiment. [Figure 7] FIG. 7 is an operational diagram showing how the probe head of the industrial endoscope is inserted into the inspection window. [Figure 8]FIG. 8 is a schematic diagram showing an example of an image captured by the imaging section. [Figure 9] FIG. 9 is a schematic diagram showing the learning system of the first embodiment. [Figure 10] FIG. 10 is a schematic diagram showing how images for generating learning data are captured. [Figure 11] FIG. 11 is a flowchart illustrating the flow of processing in the learning device. [Figure 12] FIG. 12 is a schematic diagram showing a schematic configuration of a measurement system according to the second embodiment. [Figure 13] FIG. 13 is a sequence diagram illustrating the flow of processing in the measurement system. [Figure 14] FIG. 14 is a flowchart illustrating the flow of processing in the measurement device of the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] (First embodiment) A first embodiment of the present invention will be described below with reference to FIGS. <Vehicle friction brakes> First, the measurement target of the measurement device of this embodiment will be described with reference to Figures 1 and 2. Figure 1 is a schematic diagram of a friction brake 10 provided on a vehicle. Figure 2 is a plan view schematically showing a portion of the friction brake 10 when viewed from the direction indicated by the outline arrow A11 shown in Figure 1.

[0009] The friction brake 10 is a disc brake. The friction brake 10 includes a caliper 11, a piston 12, a disc rotor 13, friction material 14, and a backing plate 15. The caliper 11 is supported on the vehicle body, and the piston 12 is provided within the caliper 11. The disc rotor 13 rotates integrally with the vehicle wheel. Friction materials 14 are arranged on both sides of the disc rotor 13. These friction materials 14 are supported by the backing plate 15. In this embodiment, the friction materials 14 and the backing plate 15 constitute a brake pad 16.

[0010] The friction material 14 of the friction brake 10 is the object of measurement by the measuring device of this embodiment. The disc rotor 13 and the back plate 15 are other members adjacent to the friction material 14. One of the disc rotor 13 and the back plate 15 corresponds to the "first member," and the other of the disc rotor 13 and the back plate 15 corresponds to the "second member."

[0011] 2, an inspection window 11a is provided in the caliper 11. When one looks into the caliper 11 from outside the caliper 11 through the inspection window 11a, the disc rotor 13, friction material 14, back plate 15, and piston 12 can be seen.

[0012] <Use stage of trained model> The configuration of the measuring device 20 and the method for measuring the thickness D of the friction material 14 will be described with reference to FIGS.

[0013] <Measuring equipment> 3 is a schematic diagram showing the configuration of the measurement device 20 of this embodiment. The measurement device 20 measures the thickness D of the friction material 14 using the trained model LM.

[0014] The measuring device 20 includes an industrial endoscope 30 , a calculation device 40 , an input device 50 , a display device 51 , and an alarm device 53 . The industrial endoscope 30 has a main body 31 , a connection cable 33 , a control unit 35 and a probe 37 .

[0015] The probe 37 has a connection part 371 and a probe head 373. The connection part 371 is provided at one end of the probe 37, and the probe head 373 is provided at the other end of the probe 37. The probe 37 is connected to the control unit 35 via the connection part 371. The probe 37 is configured so that the orientation of the probe head 373 can be changed by the operation of an actuator 352, which will be described later.

[0016] The probe head 373 has a light-emitting unit 374 that emits light and an imaging unit 375 that captures an image of the measurement target. The light-emitting unit 374 emits parallel light. In this embodiment, the light-emitting unit 374 is configured to include a semiconductor laser that emits laser light as a light source. The beam shape of the laser light emitted by the light-emitting unit 374 is circular. The light-emitting unit 374 irradiates the irradiation target with light, thereby displaying a marker MK, which is an image of light, on the surface of the irradiation target (see FIG. 5). When measuring the thickness D of the friction material 14, the irradiation target of the light-emitting unit 374 is the friction material 14 or another member adjacent to the friction material 14 (for example, the disc rotor 13 or the back plate 15).

[0017] The imaging unit 375 captures an image within its imaging range and forms an image. The imaging range of the imaging unit 375 is set to the display position of the marker MK and its surroundings. When measuring the thickness D of the friction material 14, the imaging unit 375 captures an image including the friction material 14, the member adjacent to the friction material 14, and the marker MK.

[0018] Hereinafter, an image including the friction material 14, the member adjacent to the friction material 14, and the marker MK will be referred to as an "image IMG." The main body 31 has a display unit 311 and an operation unit 312 .

[0019] The display unit 311 displays an image captured by the imaging unit 375 of the probe 37. The operation unit 312 is provided with a plurality of buttons that are operated by an operator using the industrial endoscope 30. The operation unit 312 includes a light-emitting button 312a and an image-capturing button 312b as operation buttons. The light-emitting button 312a is a button that is operated by the operator when making the light-emitting unit 374 emit light or stopping light emission by the light-emitting unit 374. The image-capturing button 312b is a button that is operated by the operator when making the image-capturing unit 375 capture an image or stopping image-capturing by the image-capturing unit 375.

[0020] Hereinafter, the operation of the light emission button 312a by the worker to cause the light emission unit 374 to emit light will be referred to as a "light emission instruction." Furthermore, the operation of the image capture button 312b by the worker to cause the image capture unit 375 to capture an image will be referred to as a "capture instruction."

[0021] The main body 31 may be a device dedicated to the industrial endoscope 30, or may be a general device. A general device may be a portable device such as a mobile phone. The buttons of the operation unit 312 are not limited to physical buttons. For example, the light emission button 312a and the image capture button 312b may be buttons displayed on a display device having a touch panel.

[0022] The main body 31 can communicate with the computing device 40. The communication between the main body 31 and the computing device 40 may be wired or wireless. The main body 31 transmits an image captured by the imaging unit 375 to the computing device 40.

[0023] The connection cable 33 connects the main body 31 and the control unit 35 . The control unit 35 has an operation wheel 351 and an actuator 352. The actuator 352 is built into the control unit 35. The operation wheel 351 is operated by an operator when changing the orientation of the probe head 373. When the operator operates the operation wheel 351, the actuator 352 is actuated in response to the operation, and the orientation of the probe head 373 changes.

[0024] The computing device 40 comprises a communication device 41 and a processing circuit 42 . The communication device 41 receives information transmitted from the industrial endoscope 30 and outputs it to the processing circuit 42. The communication device 41 also transmits information output from the processing circuit 42 to the industrial endoscope 30.

[0025] The processing circuit 42 has an execution unit 421, a storage unit 422, and a learning device 423. For example, the execution unit 421 is a CPU. The storage unit 422 stores a control program to be executed by the execution unit 421.

[0026] The learning device 423 is constructed by a trained model LM that has undergone machine learning to estimate the thickness D of the friction material 14. When an image IMG is input, the trained model LM outputs an index corresponding to the image IMG. This index is a value that correlates with the thickness of the friction material 14. The trained model LM is, for example, a forward propagation type neural network. The training method of the trained model LM will be described later.

[0027] Hereinafter, the image IMG input to the trained model LM for the purpose of measuring the thickness D of the friction material 14 will be referred to as the "measurement image IMG1." Furthermore, the index corresponding to the measurement image IMG1 output from the trained model LM will be referred to as the "index Y."

[0028] The processing circuit 42 derives the thickness D of the friction material 14 based on the index Y output from the learned model LM of the learning device 423. The input device 50 receives information input by a worker and outputs the received information to the calculation device 40 .

[0029] The display device 51 displays the thickness D of the friction material 14 calculated by the calculation device 40 . The notification device 53 notifies the worker of the content of the instruction from the calculation device 40. The notification device 53 may be a speaker that notifies the worker by voice, a lamp that notifies the worker by light, a screen that notifies the worker by a screen display, or a vibration generating device that notifies the worker by vibration.

[0030] 4 is a block diagram showing the functions of the measurement device 20. The execution unit 421 executes a control program stored in the storage unit 422, thereby functioning as an imaging angle acquisition unit 71, an angle condition notification unit 72, an imaging distance acquisition unit 75, a distance condition notification unit 76, an image acquisition unit 78, and a thickness derivation unit 80.

[0031] The imaging angle acquisition unit 71 analyzes the image IMG captured by the imaging unit 375 to acquire, as the imaging angle θ, an estimated value of the imaging angle of the object (for example, the friction material 14) captured by the imaging unit 375. The details of the process (hereinafter referred to as "imaging angle acquisition process") executed by the imaging angle acquisition unit 71 to acquire the imaging angle θ will be described later.

[0032] The imaging distance acquisition unit 75 analyzes the image captured by the imaging unit 375 to acquire an estimated value of the linear distance between the imaging unit 375 and the imaging target (for example, the friction material 14) as the imaging distance L. The details of the process (hereinafter referred to as "imaging distance acquisition process") executed by the imaging distance acquisition unit 75 to acquire the imaging distance L will be described later.

[0033] When the imaging angle θ acquired by the imaging angle acquisition unit 71 is within a predetermined angle range, the angle condition notification unit 72 notifies the worker via the notification device 53 that it is possible to capture a measurement image, which is an image for measuring the thickness D of the friction material 14.

[0034] Here, it is considered that the closer the imaging angle θ is to 90°, the higher the accuracy of estimating the thickness D of the friction material 14. Furthermore, by limiting the range of the imaging angle θ of the measurement image IMG1, the cost-effectiveness of generating the trained model LM can be improved. Therefore, the above-mentioned predetermined angle range is set taking into consideration the accuracy of estimating the thickness D of the friction material 14 and the cost required to generate the trained model LM. In this embodiment, the predetermined angle range is a range of imaging angles θ that includes 90°.

[0035] Hereinafter, the imaging angle θ being a value within a predetermined angle range will be referred to as an "imaging angle condition." The distance condition notification unit 76 notifies the worker by the notification device 53 that it is possible to capture a measurement image when the imaging distance L acquired by the imaging distance acquisition unit 75 is a value within a predetermined distance range.

[0036] Here, it is considered that the estimation accuracy of the thickness D of the friction material 14 based on the measurement image IMG1 varies depending on the imaging distance L. Furthermore, by limiting the range of the imaging distance L of the measurement image IMG1, the cost-effectiveness of generating the trained model LM can be improved. Therefore, the predetermined distance range is set taking into consideration the estimation accuracy of the thickness D of the friction material 14 and the cost required to generate the trained model LM.

[0037] Hereinafter, the imaging distance L being within a predetermined range will be referred to as an “imaging distance condition.” The imaging angle condition and the imaging distance condition correspond to the “imaging condition.” The image acquisition unit 78 acquires an image IMG that includes the friction material 14, a member adjacent to the friction material 14, and the marker MK and that is captured by the imaging unit 375, as a measurement image IMG1.

[0038] The thickness derivation unit 80 inputs the measurement image IMG1 acquired by the image acquisition unit 78 into the learned model LM of the learning device 423, and derives the thickness D of the friction material 14 based on the index Y output from the learned model LM of the learning device 423. For example, the thickness derivation unit 80 derives the thickness D of the friction material 14 as the product of the value indicated by the index Y and a conversion coefficient.

[0039] <Obtaining the imaging angle> The imaging angle acquisition process described above will now be described in detail with reference to Fig. 5. In Fig. 5, (a) is a schematic diagram showing the positional relationship between the imaging target TAR and the probe head 373 during imaging, and (b) is a schematic diagram showing a marker MK displayed on the imaging target TAR. Note that the two-dot chain line in Fig. 5 indicates the optical axis of the light emitting unit 374.

[0040] In Fig. 5(a), the probe head 373 is shown by solid lines and dashed lines. The solid lines indicate the position of the probe head 373 when the imaging angle θ is 90°, and the dashed lines indicate the position of the probe head 373 when the imaging angle θ is not 90°. In Fig. 5(b), the markers MK are shown by solid lines and dashed lines. The solid lines are the markers MK when the imaging angle θ is 90°, and the dashed lines are the markers MK when the imaging angle θ is not 90°.

[0041] When the imaging angle θ is 90°, the marker MK formed on the imaging target TAR has a perfect circular shape. In FIG. 5(b), the left-right direction in the drawing is the "first direction X1," and the direction perpendicular to the first direction X1 is the "second direction X2." In this case, the dimension F1 of the marker MK in the first direction X1 is equal to the dimension F2 of the marker MK in the second direction X2. Here, the dimension F1 being equal to the dimension F2 means that they are substantially the same, and some degree of error is allowed.

[0042] On the other hand, when the imaging angle θ is not 90°, the marker MK formed on the imaging target TAR has an elliptical shape, that is, the dimension F1 of the marker MK in the first direction X1 is longer than the dimension F2 of the marker MK in the second direction X2.

[0043] If the ratio of the dimension F2 in the second direction X2 to the dimension F1 in the first direction X1 is defined as the aspect ratio α, then when the imaging angle θ is 90°, the aspect ratio α will be 1. Furthermore, when the imaging angle θ is not 90°, the aspect ratio α will not be 1. Thus, the more the imaging angle θ deviates from 90°, the more the aspect ratio α deviates from 1.

[0044] Therefore, in this embodiment, the imaging angle acquisition unit 71 recognizes the marker MK in the image IMG using an existing image recognition method. Then, the imaging angle acquisition unit 71 acquires an estimated value of the imaging angle based on the aspect ratio α of the marker MK as the imaging angle θ. Specifically, the closer the aspect ratio α is to 1, the closer the imaging angle acquisition unit 71 acquires a value of the imaging angle θ that is closer to 90°. In other words, the imaging angle acquisition unit 71 acquires the imaging angle θ so that the greater the deviation of the aspect ratio α from 1, the more the imaging angle θ deviates from 90°.

[0045] <Getting the imaging distance> Next, the above-mentioned imaging distance acquisition process will be described in detail with reference to Fig. 8. Fig. 8 is a diagram showing an image IMG including the friction material 14 and other components present in the vicinity thereof, captured by the imaging unit 375. In the image IMG shown in Fig. 8, a marker MK is displayed on the surface of the disc rotor 13 adjacent to the friction material 14.

[0046] Here, the laser light emitted by the light emitting unit 374 is parallel light. Therefore, the size of the marker MK displayed on the surface of the disc rotor 13 does not change depending on the imaging distance L. On the other hand, the size (number of pixels) of the marker MK in the image IMG shown in Fig. 8 changes depending on the imaging distance L. Specifically, the longer the imaging distance L, the smaller the size.

[0047] Therefore, in this embodiment, the imaging distance acquisition unit 75 recognizes the marker MK in the image IMG using an existing image recognition method. Then, the imaging distance acquisition unit 75 acquires a value based on the size of the marker MK in the image IMG as the imaging distance L. In detail, the imaging distance acquisition unit 75 acquires a larger value as the imaging distance L as the size of the marker MK in the image IMG becomes smaller.

[0048] <Measurement of friction material thickness> A method for measuring the thickness D of the friction material 14 using the measuring device 20 will be described with reference to Figs. 6 to 8. Fig. 6 is a flowchart illustrating the flow of a series of processes executed by the measuring device 20 when measuring the thickness D of the friction material 14. Fig. 7 is a schematic diagram showing the state in which the probe head 373 is inserted into the inspection window 11a of the caliper 11. Fig. 8 is a schematic diagram showing a measurement image IMG1. Hereinafter, the series of processes shown in Fig. 6 will be referred to as the "measurement process." A control program corresponding to the measurement process is executed by the execution unit 421 of the measuring device 20 at predetermined control cycles.

[0049] 7, the worker inserts the probe head 373 of the industrial endoscope 30 into the inspection window 11a of the caliper 11. As a result, the imaging unit 375 and the light emitting unit 374 are inserted into the inspection window 11a.

[0050] More specifically, the worker inserts the probe head 373 toward the friction brake 10 through a gap formed in the wheel of the vehicle. In this state, the worker operates the operating wheel 351 of the control unit 35 of the industrial endoscope 30 to change the imaging range of the imaging unit 375, and checks the position of the inspection window 11a on the image IMG displayed on the display unit 311. Then, while looking at the image IMG displayed on the display unit 311, the worker inserts the probe head 373 into the inspection window 11a.

[0051] When the worker presses the light-emitting button 312a on the main body 31, in step S11 of Fig. 6, the execution unit 421 of the computing device 40 determines that the worker has issued an instruction to cause the light-emitting unit 374 to emit light (S11: YES), and proceeds to processing in step S13. In step S13, the execution unit 421 causes the light-emitting unit 374 to start emitting light. As a result, a marker MK is displayed on the surface of the friction material 14 or the disc rotor 13, as shown in Fig. 8. At this time, the marker MK may also be displayed on the surface of the back plate 15.

[0052] If the execution unit 421 determines in step S11 that the operator has not issued a light emission instruction (S11: NO), it ends the current processing. In step S15, the execution unit 421 acquires the imaging angle θ by functioning as the imaging angle acquisition unit 71. More specifically, the image IMG captured by the imaging unit 375 is transmitted from the industrial endoscope 30 to the calculation device 40. The execution unit 421 acquires an estimated value of the imaging angle based on the received image IMG as the imaging angle θ.

[0053] In step S17, the execution unit 421 acquires the imaging distance L by functioning as the imaging distance acquisition unit 75. In detail, the execution unit 421 acquires, as the imaging distance L, an estimated value of the imaging distance based on the image IMG transmitted from the industrial endoscope 30 to the calculation device 40.

[0054] In step S19, the execution unit 421 determines whether the imaging angle θ acquired in step S15 is a value within a predetermined angle range. If the imaging angle θ is a value outside the predetermined angle range (S19: NO), the execution unit 421 ends the current process. On the other hand, if the imaging angle θ is a value within the predetermined angle range (S19: YES), the execution unit 421 proceeds to the process of step S21.

[0055] In step S21, the execution unit 421 functions as the angle condition notification unit 72, thereby notifying the worker by the notification device 53 that the imaging angle θ is within a predetermined angle range.

[0056] In step S23, the execution unit 421 determines whether the imaging distance L acquired in step S17 is a value within a predetermined distance range. If the imaging distance L is a value outside the predetermined distance range (S23: NO), the execution unit 421 ends the current processing. On the other hand, if the imaging distance L is a value within the predetermined distance range (S23: YES), the execution unit 421 proceeds to processing in step S25.

[0057] In step S25, the execution unit 421 functions as the distance condition notification unit 76, thereby notifying the worker by the notification device 53 that the imaging distance L is within a predetermined distance range.

[0058] In step S26, the execution unit 421 determines whether or not an image capture instruction has been issued. If the execution unit 421 determines that an image capture instruction has not been issued (S26: NO), it ends the current process, and if the execution unit 421 determines that an image capture instruction has been issued (S26: YES), it proceeds to the process of step S27.

[0059] In step S27, the execution unit 421 causes the imaging unit 375 of the industrial endoscope 30 to capture an image. In this embodiment, it is assumed that the worker operates the image capturing button 312b in a state where the friction material 14, disc rotor 13, back plate 15, and marker MK are included in the image capturing range of the image capturing unit 375. Therefore, the image captured by the image capturing unit 375 in step S27 is the measurement image IMG1. In the processing of step S27, the function of the execution unit 421 to capture the measurement image IMG1 corresponds to the "image acquiring unit." The processing of step S27 corresponds to the "image acquiring step."

[0060] In steps S29 to S33, the execution unit 421 functions as the thickness derivation unit 80, thereby deriving the thickness D of the friction material 14 based on the measurement image IMG1 captured in step S27.

[0061] More specifically, in step S29, the execution unit 421 inputs the measurement image IMG1 captured in step S27 into the trained model LM. This process corresponds to the "image input step."

[0062] In step S31, the execution unit 421 acquires the index Y output from the learned model LM. The index Y is a value corresponding to the measurement image IMG1 input to the learned model LM in step S29. This process corresponds to the "index acquisition step."

[0063] In step S33, the execution unit 421 derives the thickness D of the friction material 14 based on the index Y acquired in step S31. This process corresponds to the "thickness acquisition step." In step S35, the execution unit 421 notifies the worker of the thickness D of the friction material 14 calculated in step S33. The execution unit 421 displays the thickness D of the friction material 14 on, for example, the display unit 311 of the industrial endoscope 30 or the display device 51. Then, the execution unit 421 ends this processing.

[0064] <Generation stage of trained model> The configuration of the learning system and the learning method will be described with reference to FIGS. <Learning System> 9 is a schematic diagram showing the general configuration of the learning system of this embodiment. The learning system generates a learned model LM. The learning system includes an industrial endoscope 30, a learning device 100, and an input device 130.

[0065] The industrial endoscope 30 and the learning device 100 are connected to be able to communicate with each other. The communication between the industrial endoscope 30 and the learning device 100 may be wired communication or wireless communication.

[0066] The input device 130 is connected to the learning device 100. The input device 130 accepts information input by an operator and outputs the accepted information to the processing circuit 120. The learning device 100 includes a communication device 110 and a processing circuit 120. The communication device 110 receives an image IMG captured by the imaging unit 375 of the industrial endoscope 30 from the industrial endoscope 30 and outputs the received image IMG to the processing circuit 120.

[0067] The processing circuit 120 has an execution unit 121, a first storage unit 122, and a second storage unit 123. For example, the execution unit 121 is a CPU. The first storage unit 122 stores a learning program and a learning model executed by the execution unit 121. The second storage unit 123 stores learning data LD and a learning result LR. The initial value of the learning result LR may be provided by a template of the learning model or may be provided by operator input. When re-learning is performed, the initial value of the learning result LR may be provided based on the learning result LR.

[0068] Hereinafter, the learning model will be specifically described as a neural network, but the learning model is not limited to a neural network. When the learning model is a neural network, the learning result LR is the weight of the connection between each neuron and the threshold of each neuron.

[0069] The execution unit 121 of the processing circuit 120 executes the learning program stored in the first storage unit 122, thereby functioning as a learning data generation unit and a learning unit. The learning data generation unit acquires an image IMG from the industrial endoscope 30, associates the acquired image IMG with the actual thickness of the friction material 14 (hereinafter referred to as the "actual thickness value") to generate learning data LD, and stores the learning data LD in the second memory unit 123.

[0070] Hereinafter, the image IMG captured by the industrial endoscope 30 for the purpose of generating the learning data LD will be referred to as the "image IMG2 for data generation." The learning unit performs machine learning to estimate the thickness D of the friction material 14 using multiple pieces of learning data LD stored in the second memory unit 123, and stores the learning result LR obtained by machine learning in the second memory unit 123.

[0071] Specifically, the learning unit inputs the data generation image IMG2 constituting the learning data LD into the learning model and obtains the index Y1 output from the learning model. The learning unit then derives an estimated thickness value D1, which is an estimate of the thickness of the friction material 14, based on the index Y1. The learning unit updates the parameters of the learning model so that the derived estimated thickness value D1 approaches the actual thickness value (correct value) of the learning data LD, and stores the updated parameters in the second storage unit 123 as the learning result LR. Specifically, the learning unit updates the connection weights between neurons and the thresholds of each neuron so that the error between the estimated thickness value D1 and the actual thickness value of the learning data LD is reduced. In this case, the learning unit can use well-known methods such as back propagation through time or stochastic gradient descent. The learning unit then stores the updated weights and thresholds in the second storage unit 123 as the learning result LR.

[0072] <Generating training data> The worker uses a specific friction brake 10, to which a friction material 14 of a specific thickness is attached, as the target for generating learning data LD, and captures images IMG2 for data generation from various imaging angles θ and imaging distances L using the industrial endoscope 30 (see FIG. 10 ). After capturing a specific number of images IMG2 for data generation, the worker attaches friction material 14 of different thicknesses to the friction brake 10 and captures images IMG in the same manner. By repeating this process, the worker can generate images IMG2 for data generation related to the specific friction brake 10.

[0073] Furthermore, by repeating the above process using different types of friction brakes 10 as the target for generating learning data LD, the worker can generate data generation images IMG2 for various friction brakes 10 with different combinations of friction material 14 and components adjacent to the friction material 14 (e.g., disc rotor 13 and back plate 15).

[0074] The above-described process in which the worker captures the image IMG2 for data generation will be referred to as "worker imaging process" hereinafter. The process executed by the learning device 100 during the worker imaging process (hereinafter referred to as the "learning data generation process") includes substantially the same processes as steps S11 to S27 of the measurement process shown in Fig. 6. Therefore, illustration of the learning data generation process is omitted, and the learning data generation process will be described with reference to Fig. 6. A control program corresponding to the learning data generation process is executed by the execution unit 121 of the learning device 100 at each predetermined control cycle.

[0075] In detail, when the operator issues a light emission command, the execution unit 121 causes the light emitting unit 374 of the industrial endoscope 30 to emit light (see steps S11 to S13 in FIG. 6). Then, when the execution unit 121 determines that the imaging angle condition and the imaging distance condition are met, the execution unit 121 causes the notification device 53 to notify that the imaging angle condition and the imaging distance condition are met (see steps S15 to S25 in FIG. 6).

[0076] When an image capture instruction is received from the operator with both the imaging angle condition and the imaging distance condition satisfied, the execution unit 121 captures an image IMG2 for data generation using the imaging unit 375 of the industrial endoscope 30 (see steps S26 and S27 in FIG. 6 ). Then, by functioning as a learning data generation unit, the execution unit 121 associates the captured image IMG2 for data generation with the actual thickness value of the friction material 14 attached to the friction brake 10 for which learning data is to be generated, thereby generating learning data LD, and stores the generated learning data LD in the second storage unit 123.

[0077] Here, the actual thickness value of the friction material 14 attached to the friction brake 10 is a value input by the worker in the above-mentioned worker imaging process when the worker attaches friction materials 14 of different thicknesses to the friction brake 10. The actual thickness value may be input via the input device 130 or via the operation unit 312 of the industrial endoscope 30.

[0078] <Generating a trained model> A method for generating a trained model LM based on the learning data LD generated by the training data generation process will be described with reference to Fig. 11. Fig. 11 is a flowchart illustrating the flow of a series of processes executed in the learning device 100 when generating the trained model LM. Hereinafter, the series of processes shown in Fig. 11 will be referred to as the "trained model generation process." A control program corresponding to the trained model generation process is executed by the execution unit 121 at each predetermined control cycle.

[0079] In step S51, the execution unit 121 determines whether or not the machine learning has been completed. For example, the execution unit 121 determines whether or not the number of times C of learning executions is equal to or greater than a predetermined number Cth. If the execution unit 121 determines that the machine learning has been completed (S51: YES), it ends the current process, and if it determines that the machine learning has not been completed (S51: NO), it proceeds to the process of step S53.

[0080] In step S53, the execution unit 121 acquires one piece of learning data LD, i.e., the image IMG2 for data generation and the actual thickness value, from the plurality of pieces of learning data LD stored in the second storage unit 123. This process corresponds to a "learning image acquisition step" and a "thickness actual value acquisition step."

[0081] In step S55, the execution unit 121 functions as a learning unit, thereby executing learning of the learning model using the learning data LD acquired in step S53, and storing the learning result LR in the second storage unit 123. Specifically, the execution unit 121 inputs the data generation image IMG2 acquired in step S53 into the learning model and acquires the index output from the learning model. The execution unit 121 then derives the thickness based on the acquired index and updates the parameters of the learning model so that the derived thickness approaches the actual thickness value acquired in step S53. These processes correspond to an "image input step," an "index acquisition step," and a "parameter adjustment step," respectively. The execution unit 121 then stores the updated parameters as the learning result LR in the second storage unit 123.

[0082] In step S57, the execution unit 121 counts the number of times the machine learning has been executed, ie, the number of times the process of step S55 has been executed, and ends the current process. <Effects of this embodiment> (1-1) In this embodiment, a trained model LM is generated by performing machine learning of a learning model using a wide variety of data generation images IMG2 relating to friction materials 14 of various thicknesses attached to various friction brakes 10. Furthermore, in this embodiment, a measurement image IMG1 is input to the trained model LM generated as described above, and the thickness D of the friction material 14 is derived based on the index Y output from the trained model LM. Therefore, the thickness D of various friction materials 14 of various friction brakes 10 can be easily measured.

[0083] (1-2) In this embodiment, learning data LD is generated based on data generation images IMG2 captured under conditions where both the imaging angle condition and the imaging distance condition are met, and a learned model LM is generated based on the learning data LD. Also, in this embodiment, measurement images IMG1 captured under conditions where both of the above conditions are met are input to the learned model LM. As a result, measurement images IMG1 captured under the same conditions as the data generation images IMG2 used to generate the learned model LM are input to the learned model LM, so the thickness D of the friction material 14 can be measured with high accuracy.

[0084] In addition, since the conditions for capturing the image IMG2 for data generation can be narrowed down, the cost-effectiveness of generating the trained model LM can be improved. (1-3) In this embodiment, a marker MK is included in both the image IMG2 for data generation and the image IMG1 for measurement. Here, the size of the marker MK displayed on the friction material 14 or the member adjacent to the friction material 14 does not depend on the imaging distance L. On the other hand, the size (number of pixels) of the marker MK in the image IMG2 for data generation and the image IMG1 for measurement depends on the imaging angle θ and the imaging distance L, just like the member adjacent to the friction material 14. Therefore, in the image IMG2 for data generation and the image IMG1 for measurement, the size of the marker MK is compared with the thickness D of the friction material 14, which is thought to improve the accuracy of estimating the thickness D of the friction material 14 by the trained model LM.

[0085] (1-4) In this embodiment, both the data generation image IMG2 and the measurement image IMG1 include the friction material 14, the disc rotor 13 adjacent to the friction material 14 on one side of the thickness direction of the friction material 14, and the back plate 15 adjacent to the friction material 14 on the other side of the thickness direction of the friction material 14. Therefore, both the data generation image IMG2 and the measurement image IMG1 include the boundary between the friction material 14 and the back plate 15, and the boundary between the friction material 14 and the disc rotor 13. This is thought to improve the accuracy of estimating the thickness D of the friction material 14 by the trained model LM.

[0086] (Second embodiment) A second embodiment of the present invention will be described below with reference to Figures 12 and 13. In this embodiment, differences from the first embodiment will be mainly described, and the same reference numerals will be used to designate configurations and functions that are substantially the same as those in the first embodiment, and redundant description will be omitted.

[0087] <Use stage of trained model> The configuration of the measurement system 200 of this embodiment and the method for measuring the thickness D of the friction material 14 will be described with reference to FIGS.

[0088] <Measurement system> 12 is a configuration diagram showing an outline of a measurement system 200 of this embodiment. The measurement system 200 includes a measurement device 220 and a server device 90. The measurement device 220 and the server device 90 are connected to each other via a communication network NT so that they can communicate with each other.

[0089] <Measuring equipment> The measuring device 220 includes an industrial endoscope 30 , a calculation device 240 , an input device 50 , a display device 51 , and an alarm device 53 .

[0090] The computing device 240 has a processing circuit 242, a first communication device 41A, and a second communication device 243. The first communication device 41A corresponds to the communication device 41 of the first embodiment. The first communication device 41A receives information transmitted from the industrial endoscope 30 and outputs it to the processing circuit 242, and transmits the information output from the processing circuit 242 to the industrial endoscope 30. The second communication device 243 transmits the information output from the processing circuit 242 to the server device 90, and receives information transmitted from the server device 90 and outputs it to the processing circuit 242.

[0091] The processing circuit 242 has an execution unit 421 and a storage unit 422. Unlike the processing circuit 42 of the first embodiment, the processing circuit 242 does not have a learning device 423. The execution unit 421 executes a control program stored in the storage unit 422, thereby functioning as an additional information acquisition unit 70, an imaging angle acquisition unit 71, an angle condition notification unit 72, an imaging distance acquisition unit 75, and a distance condition notification unit 76.

[0092] The additional information acquisition unit 70 acquires additional information that identifies the thickness and shape of the member adjacent to the friction material 14. The additional information acquisition unit 70 acquires, for example, information input by the worker via the operation unit 312 or the input device 50 as the additional information. The additional information acquisition unit 70 also acquires, for example, information indicated by a pattern captured by the imaging unit 375 as the additional information.

[0093] The additional information may include manufacturer information indicating the manufacturers of the friction brake 10, disc rotor 13, and brake pad 16, and model name information indicating the model names of these products. The thickness and shape of the disc rotor 13 may differ depending on the manufacturer of the friction brake 10 and disc rotor 13. The thickness and shape of the backing plate 15 that constitutes the brake pad 16 may differ depending on the manufacturer of the brake pad 16. Furthermore, the thickness and shape of the disc rotor 13 and backing plate 15 may differ depending on the model name even if the manufacturer is the same.

[0094] In this embodiment, manufacturer information and model name information of the brake pad 16 are acquired as additional information. <Server device> The server device 90 includes a communication device 91 and a processing circuit 92 .

[0095] The communication device 91 receives information transmitted from the measuring device 20, such as additional information and a measurement image IMG1, and outputs the information to the processing circuit 92. The communication device 91 also transmits information output from the processing circuit 92, such as the thickness D of the friction material 14, to the measuring device 20.

[0096] The processing circuit 92 has an execution unit 921, a memory unit 922, and a learning device 923. For example, the execution unit 921 is a CPU. The memory unit 922 stores a control program executed by the execution unit 921. The learning device 923 is constructed from a plurality of trained models LM1, LM2, .... The plurality of trained models LM1, LM2, ... are trained models LM that each correspond to a specific manufacturer and model name of the brake pad 16.

[0097] The execution unit 921 of the processing circuit 92 executes a control program stored in the storage unit 922, thereby functioning as a trained model selection unit and a thickness derivation unit. The trained model selection unit selects a trained model corresponding to the manufacturer and model name of the brake pad 16 from among multiple trained models LM1, LM2, ... based on the manufacturer information and model name information of the brake pad 16 transmitted from the measuring device 220 as additional information.

[0098] Hereinafter, the trained model selected by the trained model selection unit will be referred to as the "selected trained model." The thickness derivation unit has substantially the same function as the thickness derivation unit 80 of the first embodiment, except that it derives the thickness D of the friction material 14 based on the measurement image IMG1 transmitted from the measurement device 20 and based on the selected trained model. The thickness D of the friction material 14 is transmitted to the measurement device 220.

[0099] <Measurement of friction material thickness> A method for measuring the thickness D of the friction material 14 using the measurement system 200 will be described with reference to FIG. 13. FIG. 13 is a sequence diagram illustrating the flow of processes executed in the measurement system 200 when measuring the thickness D of the friction material 14. Hereinafter, the series of processes shown in FIG. 13 will be referred to as the "measurement process." A control program corresponding to the measurement process on the measuring device 220 side is executed by the execution unit 421 of the measuring device 220 at every predetermined control cycle. A control program corresponding to the measurement process on the server device 90 side is executed by the execution unit 921 of the server device 90 at every predetermined control cycle.

[0100] 7, the worker inserts the probe head 373 of the industrial endoscope 30 into the inspection window 11a of the caliper 11. As a result, the imaging unit 375 and the light emitting unit 374 are inserted into the inspection window 11a.

[0101] Step S71 in Fig. 13 is substantially the same as steps S11 and S13 of the measurement process of the first embodiment (see Fig. 6). When execution unit 421 of measurement device 220 determines that there is an instruction to emit light, it causes light emitter 374 to emit light.

[0102] Step S73 is substantially the same as steps S15 to S25 of the measurement process of the first embodiment (see FIG. 6). When the execution unit 421 determines that the imaging angle condition is met, it notifies the worker of this fact via the notification device 53. When the execution unit 421 determines that the imaging distance condition is met, it notifies the worker of this fact via the notification device 53.

[0103] Step S75 is substantially the same as steps S26 and S27 of the measurement process in the first embodiment (see FIG. 6). When determining that there is an image capturing instruction, the execution unit 421 causes the image capturing unit 375 of the industrial endoscope 30 to capture a measurement image IMG1.

[0104] In step S77, the execution unit 421 transmits additional information about the brake pad 16 and the measurement image IMG1 captured in step S75 to the server device 90. The additional information about the brake pad 16 is input in advance by the worker. In this process, the function of the execution unit 421 to transmit the additional information corresponds to the "additional information transmission unit." Furthermore, in this process, the function of the execution unit 421 to transmit the measurement image IMG1 corresponds to the "image provision unit."

[0105] In step S171, the execution unit 921 of the server device 90 determines whether or not the additional information and measurement image IMG1 regarding the brake pad 16 transmitted in step S77 from the measuring device 220 have been received. If the execution unit 921 determines that the additional information and measurement image IMG1 regarding the brake pad 16 have not been received (S171: NO), it ends the current process, and if it determines that the additional information and measurement image IMG1 regarding the brake pad 16 have been received (S171: YES), it proceeds to the process of step S173.

[0106] The function of the execution unit 921 to receive additional information corresponds to an “additional information receiving unit.” The function of the execution unit 921 to receive the measurement image IMG1 corresponds to an “image receiving unit.” In step S173, the execution unit 921 functions as a trained model selection unit to select a selected trained model based on the received additional information about the brake pad 16.

[0107] In steps S175 to S179, the execution unit 921 functions as a thickness deriving unit, thereby deriving the thickness D of the friction material 14 based on the received measurement image IMG1. More specifically, in step S175, the execution unit 921 inputs the received measurement image IMG1 into the selected trained model selected in step S173.

[0108] In step S177, the execution unit 921 acquires, from the selected trained model, the index Y corresponding to the measurement image IMG1 input to the selected trained model in step S175. In step S179, the execution unit 921 derives the thickness D of the friction material 14 based on the index Y acquired in step S177, in the same manner as in step S33 of the measurement process of the first embodiment.

[0109] In step S181, the execution unit 921 causes the communication device 91 to transmit the thickness D of the friction material 14 calculated in step S179 to the measurement device 220. The function of the execution unit 921 in the processing of step S173 corresponds to a "model selection unit." The function of the execution unit 921 in the processing of steps S175 and S177 corresponds to an "index acquisition unit." The function of the execution unit 921 in the processing of step S181 corresponds to an "index provision unit."

[0110] In step S79, the execution unit 421 of the measuring device 220 determines whether or not the thickness D of the friction material 14 transmitted from the server device 90 has been received. If the execution unit 421 determines that the thickness D of the friction material 14 has not been received (S79: NO), it ends the current process, and if the execution unit 421 determines that the thickness D of the friction material 14 has been received (S79: YES), it proceeds to the process of step S81.

[0111] The process in which the execution unit 421 receives the thickness D of the friction material 14 corresponds to the "index receiving unit." Step S81 is substantially the same as step S35 of the measurement process in Embodiment 1. In step S81, the execution unit 421 notifies the operator of the thickness D of the friction material 14.

[0112] <Generation stage of trained model> The learning system, learning data generation process, and trained model generation process of this embodiment are substantially the same as the learning system, learning data generation process, and trained model generation process of the first embodiment, respectively (see FIGS. 9 and 10). On the other hand, the method for generating learning data LD and the method for generating trained model LM of this embodiment are different from the method for generating learning data LD and the method for generating trained model LM of the first embodiment, respectively.

[0113] In this embodiment, a learned model LM corresponding to the manufacturer and model name of the brake pad 16 is generated for each manufacturer and model name of the brake pad 16. Here, the learned model LM corresponding to the manufacturer and model name of the brake pad 16 means that the learned model LM is a learned model dedicated to the manufacturer and model name of the brake pad 16.

[0114] Specifically, the worker uses the industrial endoscope 30 to capture images IMG2 for data generation from various imaging angles θ and imaging distances L, using a friction brake 10 equipped with brake pads 16 of a predetermined manufacturer and model name as the target for generating training data LD (see FIG. 10 ). After capturing a predetermined number of images IMG2 for data generation, the worker then attaches brake pads 16 of the same manufacturer and model name but with different thicknesses of friction material 14 to the friction brake 10, and captures images IMG2 for data generation in the same manner. By repeating this process, the worker can generate training data LD that will be used to generate a trained model LM corresponding to the brake pads 16 of the predetermined manufacturer and model name.

[0115] In this embodiment, learning data LD is generated using a data generation image IMG2 generated for each manufacturer and model name of the brake pad 16 as described above, and a learned model LM corresponding to the manufacturer and model name of the brake pad 16 is generated using the learning data LD.

[0116] <Effects of this embodiment> In this embodiment, in addition to the effects (1-1) to (1-4) of the above embodiment, the following effects can be further obtained.

[0117] (2-1) In this embodiment, the thickness D of the friction material 14 is estimated using a trained model corresponding to the manufacturer and model name of the brake pad 16 attached to the friction brake 10. Even if the thickness D of the friction material 14 differs between brake pads 16 of the same manufacturer and model name, the thickness of the backing plate 15 is the same.

[0118] Therefore, when generating a trained model LM based on a plurality of data generation images IMG2, the accuracy of estimating the thickness D of the friction material 14 using the trained model LM generated from a plurality of training data LD in which the manufacturers and model names of the brake pads 16 included in each data generation image IMG2 are the same is considered to be higher than the accuracy of estimating the thickness D of the friction material 14 using the trained model LM generated from a plurality of training data LD in which the manufacturers and model names of some of the brake pads 16 included in each data generation image IMG2 are different. Therefore, the trained model LM of this embodiment can accurately derive the thickness D of the friction material 14.

[0119] (2-2) Furthermore, when generating a trained model LM with a predetermined estimation accuracy, the number of data generation images IMG2 required when the manufacturers and model names of the brake pads 16 included in each data generation image IMG2 are the same is considered to be smaller than the number of data generation images IMG2 required when the manufacturers and model names of some of the brake pads 16 included in each data generation image IMG2 are different. Therefore, according to this embodiment, it is possible to improve the cost-effectiveness of generating a trained model LM.

[0120] (2-3) In this embodiment, the learning device 923 is provided in the server device 90. Therefore, it is not necessary to provide a configuration equivalent to the learning device 923 in the measurement device 220. This allows the processing circuit 242 of the measurement device 220 to be simplified.

[0121] (2-4) Furthermore, the server device 90 can re-learn the learned models LM1, LM2, ... and add a learned model LM corresponding to a brake pad 16 of a new manufacturer and model name. Therefore, even if it becomes necessary to re-learn the learned models LM1, LM2, ... or to add a learned model LM, there is no need to change the processing circuit 242 of the measuring device 220 or rewrite the contents of the memory unit 422 of the processing circuit 242.

[0122] (Third embodiment) A third embodiment of the present invention will be described below with reference to Fig. 14. In the third embodiment, differences from the first embodiment will be mainly described, and the same reference numerals will be used to designate configurations and functions that are substantially the same as those in the first embodiment, and redundant description will be omitted.

[0123] <Measuring equipment> The measuring device of this embodiment is substantially the same as the measuring device 20 of the first embodiment, except that the industrial endoscope captures video. Therefore, illustration of the measuring device of this embodiment will be omitted, and the industrial endoscope of this embodiment will be described below with the same reference numerals as the industrial endoscope 30 of the first embodiment.

[0124] When the image capturing button 312b on the main body 31 is operated, the industrial endoscope 30 starts capturing a moving image using the image capturing unit 375. The frames that make up the moving image are transmitted to the computing device 40 one by one.

[0125] <Measurement method> The measurement method of this embodiment will be described with reference to Fig. 14. Fig. 14 is a flowchart illustrating the flow of a series of processes executed by the measuring device 20 when measuring the thickness D of the friction material 14. Hereinafter, the flow of the series of processes shown in Fig. 14 will be referred to as the "measurement process." A control program corresponding to the measurement process is executed by the execution unit 421 of the measuring device 20 at every predetermined control cycle.

[0126] The worker inserts the probe head 373 of the industrial endoscope 30 into the inspection window 11a of the caliper 11 in the same manner as in the first embodiment (see FIG. 7). In this embodiment, the worker presses the imaging button 312b to start capturing video using the imaging unit 375 once the imaging range of the imaging unit 375 of the industrial endoscope 30 includes the friction material 14 and the adjacent members.

[0127] 14, the execution unit 421 of the computing device 40 determines that a light emission instruction has been issued to cause the light emitting unit 374 of the probe head 373 to emit light (S101: YES), and proceeds to the processing of step S103. If the execution unit 421 determines that a light emission instruction has not been issued in step S101 (S101: NO), it ends the current processing.

[0128] In step S103, the execution unit 421 causes the light emitting unit 374 to emit light. Furthermore, when the worker presses the image capturing button 312b of the industrial endoscope 30, in step S104, the execution unit 421 determines that the worker has issued an image capturing instruction to cause the image capturing unit 375 to capture a moving image (S104: YES), and proceeds to the processing of step S105. When the execution unit 421 determines that an image capturing instruction has not been issued (S104: NO), it ends the current processing.

[0129] In step S105, the execution unit 421 causes the imaging unit 375 to capture a moving image. Steps S107 and S109 are substantially the same as steps S15 and S17, respectively, of the measurement process in the first embodiment (see FIG. 6). The execution unit 421 acquires the imaging angle θ in step S107, and acquires the imaging distance L in step S109.

[0130] In step S111, the execution unit 421 determines whether the imaging angle condition is met based on the imaging angle θ acquired in step S107. If the execution unit 421 determines that the imaging angle condition is not met (S111: NO), it ends the current process, and if it determines that the imaging angle condition is met (S111: YES), it proceeds to the process of step S113.

[0131] In step S113, the execution unit 421 determines whether the imaging distance condition is met based on the imaging distance L acquired in step S109. If the execution unit 421 determines that the imaging distance condition is not met (S113: NO), it ends the current process, and if the execution unit 421 determines that the imaging distance condition is met (S113: YES), it proceeds to the process of step S115.

[0132] In step S115, the execution unit 421 acquires a frame of the moving image captured by the imaging unit 375 as a measurement image IMG1. By the processing of steps S105 to S115, a frame that satisfies both the imaging angle condition and the imaging distance condition is acquired as a measurement image IMG1 from among frames of the moving image captured by the imaging unit 375. The processing of steps S105 to S115 corresponds to an "imaging step."

[0133] Steps S117 to S123 are substantially the same as steps S29 to S35 of the measurement process of the first embodiment (see FIG. 6), so a description of the processes of steps S117 to S123 will be omitted.

[0134] <Generation stage of trained model> The learning system of this embodiment is substantially the same as the learning system of the first embodiment, except that the industrial endoscope 30 captures video. Therefore, illustrations of the learning system of this embodiment will be omitted, and the industrial endoscope of this embodiment will be described below with the same reference numerals as the industrial endoscope 30 of the first embodiment.

[0135] The image IMG2 for data generation in this embodiment is a frame that satisfies both the imaging angle condition and the imaging distance condition among frames of a moving image captured by the industrial endoscope 30. The worker captures an image of the target for generating learning data LD in the same manner as the worker imaging process in the first embodiment.

[0136] The learning data generation process of this embodiment includes substantially the same processes as steps S101 to S115 of the measurement process shown in Fig. 14. Therefore, illustration of the learning data generation process is omitted, and the learning data generation process will be described with reference to Fig. 14. A control program corresponding to the learning data generation process is executed by the execution unit 121 of the learning device 100 at every predetermined control cycle.

[0137] In detail, when the operator issues a light emission command, the execution unit 121 causes the light emitting unit 374 to emit light (see steps S101 and S103). When the operator issues an image capture command, the execution unit 121 causes the industrial endoscope 30 to capture a moving image (see steps S104 and S105). When the execution unit 121 determines that both the imaging angle condition and the imaging distance condition are met, it acquires a frame of the moving image as an image IMG2 for data generation (see steps S107 to S115 in FIG. 14).

[0138] The execution unit 121 generates learning data LD by associating the acquired data generation image IMG2 with the actual thickness value of the friction material 14 attached to the friction brake 10 for which the learning data LD is to be generated, and stores the generated learning data LD in the second memory unit 123.

[0139] The trained model generation process of this embodiment is substantially the same as the trained model generation process of the first embodiment, so a description thereof will be omitted. <Effects of this embodiment> According to this embodiment, in addition to the effects (1-1) to (1-4) of the first embodiment, the following effects can be obtained.

[0140] (3-1) In this embodiment, frames of a video that satisfy both the imaging angle condition and the imaging distance condition are acquired as the measurement image IMG 1 and the data generation image IMG 2. This eliminates the need to press the imaging button 312b at a timing when both the imaging angle condition and the imaging distance condition are satisfied, thereby reducing the effort required of the operator to capture the measurement image IMG 1 and the data generation image IMG 2.

[0141] <Example of change> The above-described embodiments can be modified as follows: The above-described embodiments and the following modifications can be combined with each other to the extent that no technical contradiction occurs.

[0142] In the above embodiments, the measuring device 20 is illustrated as including an industrial endoscope 30, a computing device 40, an input device 50, a display device 51, and an alarm device 53, but the functions of the measuring device 20 may be realized in any hardware form.

[0143] For example, the functions of the main body 31 of the industrial endoscope 30, the computing device 40, the input device 50, the display device 51, and the alarm device 53 may be realized by a single piece of hardware, or the functions of the computing device 40, the input device 50, the display device 51, and the alarm device 53 may be realized by a single piece of hardware.

[0144] Furthermore, the functions of the display device 51 and the notification device 53 may be realized by the display unit 311 of the industrial endoscope 30. In this case, the display device 51 and the notification device 53 may be omitted. The function of the input device 50 may be realized by the operation unit 312 of the industrial endoscope 30. In this case, the input device 50 may be omitted.

[0145] In the first and second embodiments, when the imaging angle condition and the imaging distance condition are both met, the worker is notified that the respective conditions are met. However, when both the imaging angle condition and the imaging distance condition are met, the worker may be notified that the conditions for capturing the measurement image IMG1 or the data generation image IMG2 are met.

[0146] In the third embodiment, when the imaging angle condition and the imaging distance condition are both met, the worker is notified that each condition has been met. However, when both the imaging angle condition and the imaging distance condition are met, the operator may be notified that the measurement image IMG1 or the data generation image IMG2 has been acquired.

[0147] In the above embodiments, the imaging conditions are the imaging angle condition and the imaging distance condition. However, either the imaging angle condition or the imaging distance condition may be used as the imaging condition, or no imaging condition may be set.

[0148] In the above-described embodiments, the light source of the light-emitting unit 374 is a semiconductor laser, but the light source of the light-emitting unit 374 does not have to be a semiconductor laser as long as it can emit parallel light and can display the marker MK on the surface of the friction material 14 or on the surface of another member located in the vicinity of the friction material 14.

[0149] In the above-described embodiments, the image capturing unit 375 and the light emitting unit 374 are unitized, but the light emitting unit 374 and the image capturing unit 375 do not have to be unitized. For example, the light emitting unit 374 and the image capturing unit 375 may be provided separately in the probe head 373, or one of the image capturing unit 375 and the light emitting unit 374 may be provided in a separate unit from the probe head 373.

[0150] In the above multiple embodiments, the light emitting unit 374 is provided, but the light emitting unit 374 may be omitted. In this case, the measurement image IMG1 and the data generation image IMG2 are images IMG that do not include markers MK.

[0151] In the second embodiment described above, multiple trained models LM were generated for each manufacturer and model name of the brake pad 16, and the manufacturer information and model name information of the brake pad 16 were used as additional information. However, multiple trained models LM may be generated for each manufacturer and model name of a component included in the measurement image IMG1 whose actual size and shape do not change (e.g., the disc rotor 13), and the manufacturer information and model name information of the component may be used as additional information.

[0152] In the second embodiment, the additional information is information that identifies the size and shape of the member adjacent to the friction material 14. However, the additional information is information for selecting a trained model from multiple trained models LM. Therefore, for example, multiple trained models LM may be generated for each imaging angle condition and imaging distance condition, and the imaging angle θ and imaging distance L may be used as additional information. In this case, the imaging angle θ as additional information can be acquired by the imaging angle acquisition process, and the imaging distance L as additional information can be acquired by the imaging distance acquisition process.

[0153] In the second embodiment, the server device 90 transmits the thickness D of the friction material 14 to the measuring device 220. However, the server device 90 may transmit the index Y or a value derived from the index Y to the measuring device 220. In this case, the measuring device 220 derives the thickness D of the friction material 14 based on the index Y or the value derived from the index Y.

[0154] In the first and third embodiments, one trained model LM is generated and no selected trained model is selected. However, even if the measuring device 20 is provided with a learning device 423, multiple trained models LM may be generated and a selected trained model may be selected according to additional information, as in the second embodiment. In this case, the additional information is input to the measuring device 20 by the operator via, for example, the input device 50 or the operation unit 312 of the industrial endoscope 30.

[0155] In the second embodiment, multiple trained models LM are generated and a trained model is selected according to the additional information. However, even if the learner 923 is provided in the server device 90, a single trained model LM may be generated, as in the first and third embodiments.

[0156] In the above embodiments, the imaging distance L is estimated based on the dimensions in the image IMG of the marker MK, the actual dimensions of which are known. However, the imaging distance L may be estimated based on the size in the measurement image IMG1 of an object whose size is unlikely to change over time and whose actual dimensions are known.

[0157] For example, the imaging distance L may be estimated based on the size of the disc rotor 13 and the size of the back plate 15 . In addition, the imaging distance L may be estimated using the principle of triangulation, or may be estimated based on the time it takes for the reflected laser light from the object being measured to be detected, or may be estimated based on the time it takes for the reflected ultrasonic wave from the object being measured to be detected.

[0158] In the first and second embodiments, still images captured by the industrial endoscope 30 are used as the measurement image IMG1 and the data generation image IMG2. However, as in the third embodiment, frames of a moving image captured by the industrial endoscope 30 may be used as the measurement image IMG1 and the data generation image IMG2.

[0159] In the third embodiment, the image for data generation IMG2 was a frame of a moving image captured by the industrial endoscope 30. However, as in the first and second embodiments, a still image captured by the industrial endoscope 30 may be used as the measurement image IMG1 or the data generation image IMG2.

[0160] The processing circuitry 42 of the computing device 40 may be configured as one or more processors that operate according to a computer program, one or more dedicated hardware circuits such as dedicated hardware that executes at least some of the various processes, or a circuit including a combination of these. Dedicated hardware may include, for example, an application-specific integrated circuit (ASIC).

[0161] Next, the technical ideas that can be understood from the above-described embodiments and modifications will be described. (i) A measuring device in which the imaging unit captures the measurement image including the friction material, a first member adjacent to the friction material on one side of the thickness direction of the friction material, and a second member adjacent to the friction material on the other side of the thickness direction.

[0162] (b) In a measurement method for measuring the thickness of the friction material of a vehicle friction brake, an image acquisition step of acquiring, as a measurement image, an image including the friction material and other members adjacent to the friction material among components of the friction brake; an image input step of inputting the measurement image acquired in the image acquisition step into a trained model that has undergone machine learning to estimate the thickness of the friction material from an image including the friction material and the other member; an index acquisition step of acquiring an index corresponding to the measurement image input in the image input step, which is output from the trained model; A measuring method comprising:

[0163] (c) A measurement method in which, in the acquisition step, an image that satisfies the same imaging conditions as or more restrictive than the imaging conditions of the image used when training the training model of the trained model is acquired as the measurement image.

[0164] (D) A learning method for learning a learning model of a trained model that estimates the thickness of a friction material of a friction brake of a vehicle from an image including the friction material, a learning image acquisition step of acquiring an image including the friction material and other members adjacent to the friction material among the components of the friction brake; a thickness actual value acquisition step of acquiring an actual thickness value of the friction material; an image input step of inputting the image acquired in the image acquisition step into the learning model; an index acquisition step of acquiring an index corresponding to the image input in the image input step, the index being output from the learning model; A learning method characterized by including a parameter adjustment step of adjusting parameters of the learning model so that the index obtained in the index step approaches the actual value of the thickness of the friction material obtained in the thickness acquisition step.

[0165] (e) A measurement system comprising a measurement device and a server device for measuring the thickness of a friction material of a vehicle friction brake, The measuring device is an imaging unit that captures an image; an image acquisition unit that acquires, as a measurement image, an image captured by the imaging unit that includes the friction material and another member adjacent to the friction material among components of the friction brake; an image providing unit that provides the measurement image acquired by the image acquiring unit to the server device, The server device an index acquisition unit that inputs the measurement image provided by the image provision unit into a trained model that has undergone machine learning to estimate the thickness of the friction material from an image including the friction material and the other member, and acquires an index output from the trained model; an index providing unit that provides the index acquired by the index acquiring unit or a value derived based on the index to the measurement device, A measurement system characterized in that:

[0166] (f) the measuring device includes an additional information acquisition unit that acquires additional information that specifies at least one of an imaging angle of the measurement image, an imaging distance of the measurement image, and a size of the other member, and an additional information provision unit that provides the additional information acquired by the additional information acquisition unit to the server device; A measurement system in which the index acquisition unit inputs the measurement image provided by the image provision unit into the trained model corresponding to the additional information provided by the additional information provision unit, and acquires the index output from the trained model.

[0167] (G) A measurement system including a measurement device and a server device, the measurement device for measuring the thickness of a friction material of a friction brake of a vehicle, an imaging unit that captures an image; an image acquisition unit that acquires, as a measurement image, an image captured by the imaging unit that includes the friction material and another member adjacent to the friction material among components of the friction brake; an image providing unit that provides the measurement image acquired by the image acquiring unit to the server device, which includes a trained model that has undergone machine learning to estimate the thickness of the friction material from an image including the friction material and the other member; an index receiving unit that receives from the server device an index corresponding to the measurement image output from the trained model or a value derived based on the index; A measuring device comprising:

[0168] (h) an additional information acquisition unit that acquires additional information that identifies at least one of an imaging angle of the measurement image, an imaging distance of the measurement image, and a size of the other component; an additional information providing unit that provides the additional information acquired by the additional information acquiring unit to the server device; The index receiving unit receives from the server device an index corresponding to the measurement image output from the trained model corresponding to the additional information provided from the additional information providing unit, or a value derived based on the index.

[0169] (i) A measurement system for measuring the thickness of a friction material of a friction brake of a vehicle, comprising a server device and a measurement device, wherein the server device: an image receiving unit that receives, from the measuring device, a measurement image including the friction material and another member adjacent to the friction material among components of the friction brake; an index acquiring unit that inputs the measurement image received by the image receiving unit into a trained model that has undergone machine learning to estimate the thickness of the friction material from an image including the friction material and the other member, and acquires an index output from the trained model; an index providing unit that provides the measurement device with the index acquired by the index acquiring unit or a value derived based on the index; A server device comprising:

[0170] (j) an additional information receiving unit that receives additional information specifying at least one of an imaging angle of the measurement image, an imaging distance of the measurement image, and a size of the other member from the measurement device; The index acquisition unit inputs the measurement image received by the image receiving unit into the trained model corresponding to the additional information received by the additional information receiving unit, and acquires the index output from the trained model. [Explanation of symbols]

[0171] 10...Friction brake 13...Disc rotor 14...Friction material 15...Back plate 20,220...Measuring equipment 373...Probe head 374...Light-emitting part 375...imaging unit 70...Additional information acquisition unit 78...Image acquisition unit 80...Thickness lead-out section

Claims

1. A measuring device for measuring the thickness of a friction material of a friction brake of a vehicle, an imaging unit that captures an image; an image acquisition unit that acquires, as a measurement image, an image captured by the imaging unit that includes the friction material and another member adjacent to the friction material among components of the friction brake; a thickness derivation unit that inputs the measurement image acquired by the image acquisition unit into a trained model that has undergone machine learning to estimate the thickness of the friction material from an image including the friction material and the other component, and derives the thickness of the friction material based on a correlation value with the thickness of the friction material output from the trained model.

2. an additional information acquisition unit that acquires additional information that specifies at least one of an imaging angle of the measurement image, an imaging distance of the measurement image, and a size of the other member; 2. The measuring device of claim 1, wherein the thickness derivation unit inputs the measurement image into the trained model corresponding to the additional information acquired by the additional information acquisition unit, and derives the thickness of the friction material based on a correlation value with the thickness of the friction material output from the trained model.

3. the image acquisition unit acquires, as the measurement image, an image captured by the imaging unit under predetermined imaging conditions; The predetermined imaging condition includes at least one of the following: an imaging angle of the image captured by the imaging unit is a value within a predetermined angle range; and an imaging distance of the image captured by the imaging unit is a value within a predetermined distance range. The measuring device according to claim 1 or 2.

4. the imaging unit captures a video, the image acquisition unit acquires, as the measurement image, a frame that satisfies a predetermined imaging condition from among frames of the moving image captured by the imaging unit; The predetermined imaging condition includes at least one of the following: an imaging angle of the image captured by the imaging unit is a value within a predetermined angle range; and an imaging distance of the image captured by the imaging unit is a value within a predetermined distance range. The measuring device according to claim 1 or 2.

5. A light-emitting unit that emits light is provided, the image acquisition unit acquires, as the measurement image, an image including a marker, which is an image of light displayed on a surface of the friction material or the other component by the light-emitting unit irradiating the surface of the friction material or the other component, the friction material, and the other component; The measuring device according to any one of claims 1 to 4.

Images