Mobile object state estimation system

Abstract

To provide a mobile body state estimating system capable of, with respect to a mobile body for which selection of a compression point and an impact point, pressure addition to the compression point, impact addition to the impact point, and measurement of displacement are difficult, estimating a state of a building frame without pressure addition to the compression point and impact addition to the impact point.SOLUTION: A mobile body state estimating system comprises: position measuring means 1 for measuring positions of displacement measurement points, such as displacement measurement points A3, B3, C3 and the like, provided on blades A-C of a wind turbine for wind power generation; initial position storage means 2a and after-passage position storage means 2b for storing, respectively, an initial position and an after-passage position of the displacement measurement point; displacement amount data calculation means 2c for calculating displacement amount data D1-D6 indicating variation of the displacement measurement point; structure model data output means 3 for generating structure models M1-Mn; three-dimensional model analyzing means 4 for performing displacement analysis by receiving gravity data G and structure data K1-Kn, for generating a large amount of pieces of learning data L1 to Ln; a neural network 5 for performing reverse analysis by receiving input of the displacement amount data D1-D6, for estimating a state of the blade A; and state display means 6.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a mobile body state estimation system that enables estimation of changes in the state of a mobile body due to long-term use and deterioration over time (partial damage, changes in the structure and strength of the mobile body, and alteration and deterioration of the materials that make up the mobile body, etc.). [Background technology]

[0002] The present inventor has invented and obtained a patent for an internal state estimation system that enables accurate estimation of the internal state (thickness, material strength, etc.) of a structure without relying on image recognition using radiation, ultrasound, electromagnetic waves, etc., in order to grasp changes in the internal structure of a structure or deterioration of the materials that make up the structure (Patent Document 1: Japanese Patent No. 6,440,892 and Patent Document 2: Japanese Patent No. 6,527,999). The former patented invention applies pressure to a pressure point on the outer surface of a stationary structure, and measures displacement before and after a predetermined pressure is applied to the pressure point at multiple displacement measurement points on the outer surface of the structure, thereby accurately estimating the internal state of a structure that is difficult to observe or inspect internally. The latter patented invention applies a single impact to an impact point on the outer surface of the structure, and measures amplitude data and time data or natural frequency after the impact is applied to the impact point at multiple vibration measurement points on the outer surface of the structure, thereby accurately estimating the internal state of a structure that is difficult to observe or inspect internally. Therefore, in order to apply the internal state estimation systems of the above-mentioned two patented inventions to a motor or prime mover rotor housed in a housing, it was necessary to remove part of the housing housing the motor or rotor and then apply a predetermined pressure or a single impact to the pressure point or impact point. Furthermore, in the case of moving bodies such as large wind turbines for wind power generation, trains, and airplanes, it is difficult to select pressure points and impact points, apply pressure to pressure points or impacts to impact points, and measure displacement, making it difficult to apply the internal state estimation systems of the two patent inventions mentioned above.

[0003] Furthermore, Patent Document 3 (JP 2016-217133 A) describes a calculation system that estimates the remaining life, etc., using environmental data, operating data, and strain information of a wind turbine (see, in particular, claim 1 and paragraph 0042), and Patent Document 4 (JP 2018-031610 A) describes a system that estimates the cumulative fatigue damage level of a specified part of a structure using sensor data (2) that measures physical quantities (displacement information, etc.) that change in response to the stress state of each part of the wind turbine, and detailed operating status data (3) such as the rotor rotation speed and the wind conditions to which the wind turbine is actually exposed (see, in particular, claim 1 and paragraphs 0013 to 0017). However, all of these systems detect the position of the rotating blades of a rotating wind turbine or rotor and predict the equipment's lifespan, and are not able to estimate where and what type of damage has occurred on the rotating body. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6440892 (Patent Publication No. 2020-38128) [Patent Document 2] Patent No. 6527999 (Patent Publication No. 2020-85631) [Patent Document 3] JP 2016-217133 A (Patent No. 6553399 A) [Patent Document 4] JP 2018-031610 A Summary of the Invention [Problem to be solved by the invention]

[0005] The problem that this invention aims to solve is to make it possible to estimate the condition of a body (damage status, changes in structure and strength, deterioration or degradation of constituent materials, etc.) of a motor or prime mover rotor housed in a housing, large wind turbines for wind power generation, trains, aircraft, and other moving bodies, where it is difficult to select pressure points or impact points, apply pressure to pressure points or impact points, and measure displacement, without applying pressure to pressure points or impact points. [Means for solving the problem]

[0009] Claims to solve the above problem 1 The invention relates to A moving body state estimation system capable of estimating a state of a moving body, comprising: A plurality of displacement measurement points provided on the moving body; a position measuring means for measuring positions of the plurality of displacement measuring points while the moving body is stationary; an initial position storage means for storing initial positions of the plurality of displacement measurement points measured by the position measurement means while the moving body is stationary in a predetermined posture at a time point when the moving body is in an initial state; a post-elapsed position storage means for storing post-elapsed positions of the plurality of displacement measurement points measured by the position measurement means while the moving body is stationary in the predetermined posture after a predetermined period has elapsed since the moving body was in the initial state; a displacement amount data calculation means for calculating displacement amount data between initial positions and subsequent positions of the plurality of displacement measurement points; a structure model data output means for generating a number of structure models each having a different damaged or deteriorated state for the moving body, and outputting structure data relating to the number of structure models; a structural model analysis means for receiving gravity data acting on the moving body while the moving body is stationary in the predetermined attitude and the structural data output from the structural model data output means, performing a displacement analysis at displacement measurement corresponding points corresponding to the plurality of displacement measurement points for each structural model using a numerical analysis means, and outputting, as learning data, specific information for identifying the structural model that was the subject of the displacement analysis and displacement analysis data at the plurality of displacement measurement corresponding points obtained by the displacement analysis performed on the structural model; a neural network that receives the displacement amount data, performs an inverse analysis, and outputs state estimation information of the moving body; a state display means for displaying a state of the moving object in accordance with the state estimation information output from the neural network; the neural network is previously provided with learning data for the multiple structure models output from the structure model analysis means and is optimized; The predetermined position includes a plurality of positions, the initial position storage means stores initial positions of the plurality of displacement measurement points in a state in which the moving body is stationary in the plurality of postures; the post-elapsed position storage means stores post-elapsed positions of the plurality of displacement measurement points in a state in which the moving body is stationary in the plurality of postures; the displacement amount data calculation means calculates, for each of the plurality of postures, displacement amount data between an initial position and a post-elapsed position of the plurality of displacement measurement points; the structure model analysis means receives, for each of the plurality of postures, gravity data acting on the moving body while the moving body is stationary and the structure data output from the structure model data output means, performs a displacement analysis at the plurality of displacement measurement corresponding points for each structure model using a numerical analysis means, and outputs, as learning data, specific information for identifying the structure model that was the subject of the displacement analysis and displacement analysis data at the plurality of displacement measurement corresponding points obtained by the displacement analysis performed on the structure model; The neural network receives the displacement amount data obtained in each of the plurality of postures, performs an inverse analysis, and outputs state estimation information of the moving body.

[0010] Claims to solve the above problem 2 The invention relates to Claim 1 In the moving object state estimation system according to the present invention, the moving body is a rotating body having a rotating shaft and a rotor, The states in which the moving body is stationary in the multiple postures include a state in which the moving body is stationary so that a rotor center line passing through the center of rotation and the tip of the rotor forms a first predetermined angle with respect to a reference horizontal line that passes through the center of rotation and is perpendicular to the rotation axis, and a state in which the moving body is stationary so that the rotor center line forms a second predetermined angle with respect to the reference horizontal line.

[0011] Claims to solve the above problem 3 The invention relates to Claim 2 In the moving object state estimation system according to the present invention, The rotating body is a wind turbine for wind power generation, The condition of the moving body is characterized by the location and size of a damaged or deteriorated part of the wind turbine for wind power generation.

[0012] Claims to solve the above problem 4 The invention relates to Claim 1 In the moving object state estimation system according to the present invention, The moving body is a movable body that is movable in at least one direction, The states in which the movable body is held stationary in the multiple postures include a state in which the movable body is held stationary so that a movable body straight line passing through the front and rear ends of the movable body forms a first predetermined angle with respect to a horizontal plane, and a state in which the movable body straight line is held stationary so that a second predetermined angle with respect to a horizontal plane. Effect of the Invention

[0016] Claim 1 The moving object state estimation system of the present invention comprises: at a time when the moving body is in an initial state, initial positions of a plurality of displacement measurement points are stored with the moving body stationary in a predetermined posture, and after a predetermined period of time has elapsed since the time when the moving body is in the initial state, post-elapsed positions of the plurality of displacement measurement points are stored with the moving body stationary in the predetermined posture, displacement amount data between the initial positions and the post-elapsed positions of the plurality of displacement measurement points is calculated, the displacement amount data is input to a neural network to perform an inverse analysis, state estimation information for the moving body is output, and the state of the moving body is displayed according to the output state estimation information, and since the neural network is given in advance learning data for a large number of structural models output from the structural model analysis means and is optimized, it is possible to estimate the state of a structure constituting the moving body by simply temporarily holding the moving body stationary in a predetermined posture, without the need to select pressure points or impact points, apply pressure to the pressure points or impacts to the impact points, and measure displacement of a moving body for which it is difficult to apply pressure or impact to the pressure points, and The initial positions and the positions after the displacement measurement points are stored when the moving body is stationary in a plurality of postures, and the displacement amount data between the initial positions and the positions after the displacement measurement points for each of the plurality of postures is calculated. The displacement amount data is input to a neural network for inverse analysis, and the state estimation information of the moving body is output, so that the state of the structure constituting the moving body can be estimated. High It can be estimated with high accuracy.

[0017] Claim 2 and 4 The moving object state estimation system according to the present invention is as follows: 1 In addition to the above effects of the invention, 2 In the present invention, the state of a rotor having a rotor shaft and rotor blades can be easily estimated with high accuracy by simply keeping the rotor blade center line at a first predetermined angle and a second predetermined angle with respect to a reference horizontal line. 4 In this method, the state of a movable body that can move in at least one direction can be easily estimated with high accuracy by simply placing the movable body in a stationary state so that the movable body straight line forms a first specified angle and a second specified angle with respect to a horizontal plane.

[0018] Claim 3 The moving object state estimation system according to the present invention is as follows: 2 In addition to the above-mentioned effects of the present invention, the position and size of a damaged or deteriorated part in a wind turbine for wind power generation can be estimated and displayed as the state of the moving body. [Brief description of the drawings]

[0019] [Figure 1] FIG. 1 is a block diagram of a state estimation system for a wind turbine for wind power generation according to a first embodiment. [Diagram 2] 1 is a diagram showing a basic configuration of a wind turbine for wind power generation according to a first embodiment and deformation due to gravity; [Diagram 3] FIG. 13 shows an example of a model in which the size of the damaged area is 4. [Figure 4] FIG. 4 is a flowchart showing an algorithm of the state estimation system according to the first embodiment. [Diagram 5] 4 is a table showing the accuracy rate of the estimation result by the state estimation system according to the first embodiment. [Figure 6] FIG. 11 is a diagram showing a stationary state in which position measurement of a displacement measurement point is performed in the second embodiment. [Figure 7] FIG. 11 is a flowchart showing an algorithm of the state estimation system according to the second embodiment. [Figure 8] 13 is a table showing the accuracy rate of the estimation result by the state estimation system according to the second embodiment. [Figure 9] 13 is a diagram showing a stationary state in which position measurement of a displacement measurement point is performed on a movable body according to a modified example. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] Hereinafter, the embodiment of the present invention will be described with reference to examples. EXAMPLES

[0021] FIG. 1 is a block diagram of a state estimation system for a wind turbine for wind power generation according to a first embodiment. As shown in FIG. 1, the state estimation system for a wind turbine for wind power generation according to the first embodiment is a wind turbine with a length of 60 m, a width of 2 m, a thickness of 0.4 m, and a specific gravity of 1500 kg / m 3The method is for estimating the condition of a wind power turbine having three blades A to C made of a material with an elastic modulus of 98 GPa (e.g., damage to the blades, changes in structure and strength, and deterioration or degradation of the constituent materials), and is composed of the following means (1) to (9). In Example 1, Blades A to C are of simplified shape and material, but when creating a state estimation system for use in an actual wind turbine for wind power generation, it is better that the shape, structure and materials of each part of Blades A to C etc. are as close as possible to those of the actual wind turbine to the extent that can be understood from design drawings etc.

[0022] Means (1): Six displacement measurement points A1 to A6, B1 to B6, and C1 to C6 are provided on blades A to C, respectively. These displacement measurement points are located on a straight line extending from the center of rotation of each blade toward the tip, and are provided on the underside of each blade, one at every 10 m from the base. Means (2): Position measurement means 1 that is provided near a wind turbine for wind power generation and measures the positions of displacement measurement points A1 to A6, B1 to B6, and C1 to C6 when the blades are stationary. The position measurement means 1 measures the position of each stationary displacement measurement point by any one of the following methods: a triangulation method, in which each displacement measurement point is optically observed from two appropriate points to determine its relative position with respect to a reference point; a method, in which each displacement measurement point is received to determine its relative position with respect to a reference point; or a method, in which each displacement measurement point is received to determine its absolute position or its relative position with respect to a reference point by receiving a signal from a positioning device (GPS, etc.) installed at each displacement measurement point. In addition, in Example 1, when measuring the positions of displacement measurement points A1 to A6, B1 to B6, and C1 to C6 of blades A to C, the pitch angle is changed so that the wind-receiving surfaces of blades A to C are parallel to the rotation axis R, and then blades A to C are successively made horizontal and stationary with their back sides facing straight down. By doing so, the position fluctuation of the displacement measurement points due to the effect of wind pressure on blades A to C can be ignored, and since gravity acts evenly in the vertical direction on the wide surfaces of blades A to C, the position fluctuation of the displacement measurement points can be increased. Furthermore, since the positional fluctuation of the displacement measurement points in the width direction on blades A to C is structurally small, by arranging each displacement measurement point on a plane perpendicular to the rotation axis R, each position can be identified using XY coordinates with the position of the rotation axis R as the origin. This can reduce the processing load on the displacement amount data calculation means 2c, the structure model analysis means 4, and the neural network 5, which will be described later. FIG. 2 shows the basic configuration of the wind turbine for wind power generation according to the first embodiment and its deformation due to gravity. FIG. 2(1) shows the state in which blade A is held stationary with its back surface facing straight down and horizontal after the pitch angle is changed so that the wind-receiving surfaces of blades A to C are parallel to the rotation axis R.

[0023] Means (3): Initial position storage means 2a which stores the positions of displacement measurement points A1 to A6, B1 to B6 and C1 to C6 (hereinafter referred to as "initial positions") measured by position measurement means 1 when blades A to C are successively horizontal and stationary with their back sides facing straight down as described above at the time when the wind turbine is in its initial state (immediately after the wind turbine is newly installed or repaired). Means (4): A post-elapsed position storage means 2b which stores the positions (hereinafter referred to as "post-elapsed positions") of displacement measurement points A1 to A6, B1 to B6, and C1 to C6 measured by the position measurement means 1 in a state in which blades A to C are successively made horizontal and held stationary with their back sides facing directly downward, similar to the means (3), after a predetermined period of time has elapsed since the wind turbine was in its initial state. Means (5): A displacement amount data calculation means 2c which calculates displacement amount data indicating changes in the positions of the displacement measuring points A1 to A6, B1 to B6, and C1 to C6 over a predetermined period of time, based on the initial positions of the displacement measuring points A1 to A6, B1 to B6, and C1 to C6 stored in the initial position storage means 2a described in the above section of means (3) and the post-elapsed positions of the displacement measuring points A1 to A6, B1 to B6, and C1 to C6 stored in the post-elapsed position storage means 2b described in the above section of means (4). Note that FIG. 1 shows a state in which the initial positions of all displacement measurement points are stored in the initial position storage means 2a, the blade A is stopped, the post-travel positions of the displacement measurement points A1 to A6 measured by the position measurement means 1 are stored, and the displacement amount data is calculated; therefore, FIG. 1 only shows the displacement amount data D1 to D6 for the displacement measurement points A1 to A6. Figure 2(2) shows how blades A to C bend due to gravity when blade A is horizontal and stationary with its back side facing straight down when the wind turbine is in its initial state, and Figure 2(3) is a superimposed diagram of Figure 2(2) showing how blade A bends due to gravity when blade A is horizontal and stationary with its back side facing straight down after a predetermined period of time has elapsed from the time when the wind turbine is in its initial state. Figure 2(2) shows the positions of blades A to C when the wind turbine is in its initial state, enlarged 100,000 times in the vertical direction only, and Figure 2(3) shows the positions of blades A to C shown in Figure 2(2) together with an example of the position of blade A after a predetermined period of time has passed since the wind turbine was in its initial state.

[0024] Means (6): A structural model data output means 3 for generating structural models M1 to Mn each having a different position and size of a damaged or deteriorated portion for a wind turbine, and outputting structural data K1 to Kn relating to each structural model. In Example 1, 2,268 models were created for seven types of damaged area location, from 0 to 6 from closest to the rotation axis R, and five types of damaged area size, from 0 to 4 (however, in all cases, 0 indicates no damaged area), and 2,041 pieces of data, or 90% of the data, were used as training data, and the remaining 227 pieces of data were used as validation data. FIG. 3 is a diagram showing an example of a model in which the size of the damaged area is 4. The position and size of the damaged or deteriorated areas are randomly generated to create more than 2000 different patterns of structural data, but it is better to set the size of the damaged or deteriorated areas so that many small patterns are included. Furthermore, blades A to C are normally made of exactly the same shape and material, and displacement measurement points A1 to A6, B1 to B6, and C1 to C6 are also positioned in the same way, so structure models M1 to Mn are created for one blade, and state estimation of blades A to C can be performed using the same structure models M1 to Mn for each blade. Therefore, the following description will be given only regarding the case where the state of blade A is estimated.

[0025] Means (7): A structural model analysis means 4 which performs displacement analysis for each structural model M1 to Mn at displacement measurement corresponding points corresponding to a plurality of displacement measurement points A1 to A6 using numerical analysis means based on the structural data K1 to Kn output from the structural model data output means 3 and the gravity data G acting on each point of the blade A of the wind power generating wind turbine when the blade A is held horizontal and stationary with the back side facing straight down as shown in FIG. 2 (1), and outputs, for the structural model M1, specific information (e.g., k1) identifying the structural data K1 and displacement analysis data D11 to D61 at the displacement measurement corresponding points as learning data L1, and outputs, for the structural model Mn, specific information (e.g., kn) identifying the structural data Kn and displacement analysis data D1n to D6n at the displacement measurement corresponding points as learning data Ln. In the three-dimensional model analysis means 4 of the first embodiment, a finite element method analysis means (hereinafter referred to as "FEM analysis means") is used as a numerical analysis means. When the FEM analysis means is used, the structural models M1 to Mn are divided into a mesh of numerous tetrahedrons, and each tetrahedron is analyzed under the condition that a force equal to the mass of the tetrahedron multiplied by the gravitational acceleration is applied in the vertical direction to the center of gravity of the tetrahedron, thereby making it possible to calculate how the position of any point (any vertex of a tetrahedron) in the target structural model changes. Each of the structure models M1 to Mn has the same shape as blade A of a wind turbine for wind power generation, and the positions of the displacement measurement corresponding points on the structure models M1 to Mn are set to correspond to the displacement measurement points A1 to A6 on the blade A, respectively. Therefore, when the structure model data output means 3 creates structure models M1 to Mn, each of which has a different position and size of damaged or deteriorated areas, and outputs structure data K1 to Kn for each structure model, displacement analysis data is obtained at displacement measurement corresponding points by holding each structure model horizontally and at rest with the back side facing straight down, as if gravity data G (vertical force due to gravity at the center of gravity of each tetrahedron) was added to each point of the structure model, and therefore a large amount of learning data L1 to Ln can be prepared in a short period of time.

[0026] Means (8): A neural network 5 that receives input of displacement amount data D1 to D6 indicating the fluctuations in the positions of the displacement measurement points A1 to A6 over a predetermined period of time calculated by the displacement amount data calculation means 2c, performs inverse analysis, and outputs state estimation information for the blade A. As shown in FIG. 1, this neural network 5 is constructed from an input layer 5I, an output layer 5O, a learning data storage means 5L, a learning data input means 5T that inputs a large amount of learning data L1 to Ln to the learning data storage means 5L, as well as multiple intermediate layers (not shown), and others. In addition, the intermediate layer is constructed by combining various layers, such as convolutional layers, pooling layers, and fully connected layers, as necessary. Furthermore, the neural network 5 is optimized by storing a large amount of learning data L1 to Ln in the learning data storage means 5L and allowing it to self-learn. In this way, by using parameters optimized by a sufficient number of learning data (several thousand to several hundred thousand depending on the complexity of the target shape), an estimation accuracy of 95% or more can be obtained. Then, when the displacement amount data D1 to D6 calculated by the displacement amount data calculation means 2c is input to the input layer 5I, the neural network 5 performs inverse analysis and outputs condition estimation information that is close to the position and size of the damaged or deteriorated part in the actual blade A with high accuracy. Means (9): Status display means 6 for displaying the status of blade A in accordance with the status estimation information of blade A output from the neural network 5. The status display means 6 of Example 1 displays the position and size of the damaged or deteriorated areas of blades A to C according to the status estimation information output from the neural network 5. In the normal display mode, the position and size of the damaged or deteriorated areas are displayed on a plan view of blades A to C, but it is also possible to select a mode in which the damaged or deteriorated areas are displayed on a perspective view of blades A to C.

[0027] FIG. 4 is a flow diagram showing an algorithm of the condition estimation system according to the first embodiment. By following the steps T1 to T4 below, a large number of learning data L1 to Ln are generated for one blade, stored in a neural network for self-learning, and then the position and size of the damaged or deteriorated part on blade A are estimated by steps 1 to 6, and the estimated position and size of the damaged or deteriorated part are displayed. In addition, using the same learning data L1 to Ln and neural network 5, the positions and sizes of the damaged or deteriorated areas on blades B and C are estimated by steps 1 to 6 similar to those for blade A, and the positions and sizes of the estimated damaged or deteriorated areas are displayed.

[0028] Step T1: Based on the current state and design drawings of the wind turbine for wind power generation, multiple structural models (M1 to Mn) are created for one blade. Step T2: Create structural data (K1 to Kn) for each structural model (M1 to Mn). Step T3: Using the FEM analysis means, a displacement analysis is performed based on the gravity data G and each structural data (K1 to Kn) for a case in which each structural model (M1 to Mn) is held horizontally and cantilevered with the back side facing straight down, and displacement analysis data at the displacement measurement corresponding points is output. Step T4: Based on the displacement analysis data output in step T3, a large number of learning data (L1 to Ln) are generated and input to the neural network 5. Step 1: When the wind turbine is in its initial state, blade A is held horizontal and stationary with its back surface pointing straight down, and the initial positions of displacement measurement points A1 to A6 are measured and stored. Step 2: After a predetermined period of time has elapsed since the initial positions were stored, the blade A is held horizontally and stationary with the back side facing straight down, and the subsequent positions of the displacement measurement points A1 to A6 are measured and stored. Step 3: Based on the initial positions and subsequent positions of the displacement measuring points A1 to A6 stored in steps 1 and 2, displacement amount data (D1 to D6) are calculated. Step 4: Based on the displacement amount data (D1 to D6) calculated in step 3, a reverse analysis is performed by the neural network 5 that has been trained using a large amount of training data (L1 to Ln). Step 5: Based on the result of the inverse analysis in step 4, state estimation information is output. Step 6: Based on the condition estimation information output in step 5, the location and size of the damaged or deteriorated area of ​​blade A are displayed.

[0029] FIG. 5 is a table illustrating the accuracy rate of the estimation result by the state estimation system according to the first embodiment. The accuracy rate of the estimation results was verified by randomly selecting 50 sample data from the 227 data remaining for verification after having the neural network 5 perform learning 30,000 times using 2,041 pieces of training data, performing inverse analysis using the neural network 5 based on the displacement data in the sample data, and comparing the state estimation information regarding the position and size of the damaged area based on the results of the inverse analysis with the position and size of the damaged area in the structural model that was the basis of the sample data. In addition, because samples with no damage and a position of 0 and a size of 0 are the most likely to occur, 14 identical samples were intentionally selected for the inverse analysis. As a result, the accuracy rate for the location of the damaged area was 96% (48 / 50), but in the two incorrect cases, locations 1 and 4 were estimated to be adjacent locations 2 and 5, respectively. The correct answer rate for the size of the damaged area was also 96% (48 / 50), but in the two incorrect cases, sizes 1 and 3 were estimated as sizes 2 and 4, which are one level larger. There were no cases in which both the location and size of the injury site were incorrect. Based on these verification results, it can be evaluated that the condition estimation system of the present invention is capable of accurately estimating the location and size of the damaged or deteriorated areas of blades A to C, and it is expected that the accuracy rate can be further increased by increasing the number of learning sessions. EXAMPLES

[0030] The state estimation system according to the second embodiment is intended to accurately estimate the positions and sizes of damaged or deteriorated parts of blades A to C of the same wind power generating turbine as the state estimation system according to the first embodiment, and the configuration of the wind power generating turbine is the same as that of the first embodiment (for details, see the explanation of the displacement measurement points A1 to A6, B1 to B6, and C1 to C6 in the means (1)). Furthermore, the position measuring means 1, the structure model data output means 3 and the state display means 6 are the same as those in the first embodiment (for details, see the explanations regarding the means (2), (6) and (9)). Example 2 differs from Example 1 in that the stationary state of the blade when measuring the position of the displacement measurement point is the state shown in FIG. 6(2) (a state in which blade A is horizontal and stationary with its back side facing directly upward) in addition to the states shown in FIG. 2(1) and FIG. 6(1) as in Example 1 (a state in which blade A is horizontal and stationary with its back side facing directly downward). Therefore, the block diagram of the state estimation system according to the second embodiment is almost the same as the block diagram of the state estimation system according to the first embodiment shown in FIG. 1, and differs only in that there are two systems of displacement amount data, so a detailed description thereof will be omitted.

[0031] FIG. 7 is a flow diagram showing an algorithm of the condition estimation system according to the second embodiment. In the following steps T21 to T24, a large number of learning data L1R to LnR and L1L to LnL are generated for one blade, L1R to LnR are stored in neural network 5R, and L1L to LnL are stored in neural network 5L for self-learning, and then in steps 21 to 26, the position and size of the damaged or deteriorated part on blade A are estimated, and the estimated position and size of the damaged or deteriorated part are displayed. In addition, using the same learning data L1R to LnR, L1L to LnL and neural networks 5R and 5L, the positions and sizes of the damaged or deteriorated areas on blades B and C are estimated by steps 21 to 26 similar to those for blade A, and the positions and sizes of the estimated damaged or deteriorated areas are displayed.

[0032] Step T21: Based on the current state and design drawings of the wind turbine for wind power generation, multiple structural models (M1 to Mn) are created for one blade. Step T22: Create structural data (K1 to Kn) for each structural model (M1 to Mn). Step T23: Using the FEM analysis means, a displacement analysis is performed based on the gravity data G and each structural data (K1 to Kn) for the cases where each structural model (M1 to Mn) is held horizontally and cantilevered with its back side facing straight down, and where it is held cantilevered with its back side facing straight up, and displacement analysis data at the displacement measurement corresponding points is output. Step T24: Based on the displacement analysis data output in step T23, a large number of learning data (L1R to LnR, L1L to LnL) are generated, and L1R to LnR are input to the neural network 5R, and L1L to LnL are input to the neural network 5L. Step 21: When the wind power generating turbine is in its initial state, the initial positions of displacement measurement points A1 to A6 are measured and stored with blade A held horizontally and stationary with its back side facing straight down, and with its back side facing straight up. Step 22: After a predetermined period of time has elapsed since the initial position was stored, the subsequent positions of displacement measurement points A1 to A6 are measured and stored while blade A is held horizontally and stationary with the back side facing straight down, and while it is held stationary with the back side facing straight up. Step 23: Based on the initial positions and subsequent positions of the displacement measuring points A1 to A6 stored in steps 21 and 22, displacement amount data (D1R to D6R, D1L to D6L) are calculated. Step 24: Based on the displacement data (D1R to D6R) calculated in step 23, an inverse analysis is performed using the neural network 5R that has been trained using a large amount of learning data (L1R to LnR), and based on the displacement data (D1L to D6L), an inverse analysis is performed using the neural network 5L that has been trained using a large amount of learning data (L1L to LnL). Step 25: Based on the results of the two inverse analyses in step 24, state estimation information is output. Step 26: Based on the condition estimation information output in step 25, the position and size of the damaged or deteriorated part of blade A are displayed.

[0033] FIG. 8 is a table illustrating the accuracy rate of the estimation result by the state estimation system according to the second embodiment. The accuracy rate of the estimation results was verified by having the neural networks 5R and 5L perform training 30,000 times using 1,404 pieces of training data, and then randomly selecting 50 sample data from the 157 pieces of data that were left as verification data. Inverse analysis was performed using the neural networks 5R and 5L based on the displacement data in the sample data, and the state estimation information regarding the location and size of the damaged area based on the results of both inverse analyses was compared with the location and size of the damaged area in the structural model that was the basis of the sample data. In addition, since a sample with no damage and a position of 0 and a size of 0 is the most likely case to occur, 10 identical samples were intentionally selected and subjected to the inverse analysis in the same manner as in Example 1. As a result, the correct answer rate for the location of the damaged area was 98% (49 / 50), with the incorrect answers estimating the location to be adjacent, and the correct answer rate for the size of the damaged area was 96% (48 / 50), with the two incorrect answers both estimating the size to be one size smaller, and there were no cases where both the location and size of the damaged area were incorrect. Furthermore, since the position accuracy has improved despite the small amount of training data and the magnitude accuracy has not changed, if displacement data is measured at different stopping positions in actual operation and an inverse analysis is performed, and the position and size of the damaged area are estimated based on the results of both inverse analyses, it is expected that the prediction accuracy will be improved compared to Example 1, in which displacement data is measured at one stopping position and the position and size of the damaged area are estimated.

[0034] Modifications of the embodiment are listed below. (1) In the first and second embodiments, the condition estimation system estimates the damage condition, changes in structure or strength, and alteration or deterioration of the constituent materials (specifically, the location and size of the damaged or deteriorated area) of Blades A to C of a wind turbine used for wind power generation. However, the present invention is not limited to wind turbines used for wind power generation, and can be applied to any rotating body having a rotating shaft and rotors. When applied to such a rotating body, a plurality of displacement measurement points are provided on the rotors, and when measuring the positions of the displacement measurement points, the rotors are brought to a standstill so that the rotor center line passing through the center of rotation and the tip of the rotor forms a predetermined angle with respect to a reference horizontal line that passes through the center of rotation and is perpendicular to the rotation axis. Also, as shown in FIG. 9, the present invention can be applied to a movable body that is movable in at least one direction (in FIG. 9, a vehicle such as a bus, a diesel railcar, or a train). When applied to such a movable body, a plurality of displacement measurement points are provided at appropriate locations on the movable body body, and when measuring the positions of the displacement measurement points, the movable body can be tilted and stopped so that the movable body straight line passing through the front and rear ends of the movable body forms a predetermined angle with respect to the horizontal plane (30 degrees in Figure 9(2)), as shown in Figure 9(2).In order to obtain an effect similar to that of Example 2, the movable body can also be tilted and stopped so that the movable body straight line forms a first predetermined angle and a second predetermined angle with respect to the horizontal plane (±30 degrees in Figures 9(2) and (3)). Furthermore, the movable body may be tilted and stopped so that a line in the width direction passing through the right and left ends of the movable body forms a predetermined angle with the horizontal plane. In the case of a vehicle, the predetermined angle should preferably be within the range of -30 degrees to 30 degrees due to constraints such as its climbing ability and braking ability. As such, the present invention can be applied not only to rotating bodies but also to movable bodies, and therefore in the claims, rotating bodies and movable bodies are collectively referred to as "moving bodies", and the state in which the blades or rotors are stationary at a predetermined angle and the state in which the movable body is tilted at a predetermined angle and stationary are collectively referred to as "the state in which the moving body is stationary in a predetermined posture".

[0035] (2) In Examples 1 and 2, six displacement measurement points were provided, one on the back side of each blade at 10 m intervals from the base, on a straight line extending from the center of rotation of each blade toward the tip. However, multiple displacement measurement points may be provided, and the location of the displacement measurement point may be anywhere on the outer surface of each blade if the position of the displacement measurement point is measured based on optical observation, or anywhere on the outer surface or inside of each blade if a transmitter and receiver are used. (3) In Examples 1 and 2, when measuring the positions of the displacement measurement points of blades A to C, the pitch angle is changed so that the wind-receiving surfaces of blades A to C are parallel to the rotation axis R, and then blades A to C are held stationary. However, blades A to C may be held stationary without changing the pitch angle.

[0036] (4) In Example 1, when measuring the positions of the displacement measurement points of blades A to C, blades A to C are horizontally stationary with their back sides facing straight down. However, instead of horizontally stationary with their back sides facing straight down, blades A to C may be stationary so that a central line passing through the center of rotation and the tip of blades A to C forms a predetermined angle within the range of -45 degrees or more and 45 degrees or less, or 135 degrees or more and 225 degrees or less, with respect to a reference horizontal line that passes through the center of rotation and is perpendicular to the rotation axis R. In addition, in Example 2, when measuring the positions of the displacement measurement points of blades A to C, blades A to C are held horizontally and stationary with their back sides facing straight down and stationary with their back sides facing straight up. However, instead of holding blades A to C horizontally and stationary with their back sides facing straight down, they may be held stationary so that their center lines form a first predetermined angle in the range of -45 degrees or more and 45 degrees or less with respect to the reference horizontal line, and instead of holding blades A to C horizontally and stationary with their back sides facing straight up, they may be held stationary so that their center lines form a second predetermined angle in the range of 135 degrees or more and 225 degrees or less with respect to the reference horizontal line. Furthermore, the blades A to C are not limited to being stopped at the first and second predetermined angles, but may be stopped at three or more predetermined angles and the positions of the displacement measurement points may be measured. For this reason, in the claims, expressions such as "a state in which the central straight lines of blades A to C are stopped so as to form a first predetermined angle and a second predetermined angle with respect to the reference horizontal line" and "a state in which the movable body straight line is stopped so as to form a first predetermined angle and a second predetermined angle with respect to the horizontal plane" are used to refer to "a state in which the movable body is stopped in a plurality of postures." (5) In the state estimation systems of the first and second embodiments, FEM analysis means is used to generate the training data and the validation data. However, the present invention is not limited to FEM analysis means, and numerical analysis means based on the finite difference method, the boundary element method, or the like may be used. [Explanation of symbols]

[0037] 1 Position measuring means 2a Initial position storage means 2b Later position storage means 2c Displacement amount data calculation means 3 Structure model data output means 4 Structural model analysis means 5 Neural network 5I Input layer 5L Learning data storage means 5O Output layer 5T Learning data input means 6 Status display means A~C Blade A1~A6, B1~B6, C1~C6 Displacement measurement points D1~D6 Displacement data D1R~D6R, D1L~D6L Displacement data D11~D61, D1n~D6n Displacement analysis data K1~Kn Structure data L1~Ln Learning data L1R~LnR, L1L~LnL Learning data M1~Mn Structure model R rotation axis

Claims

1. A moving object state estimation system capable of estimating a state of a moving object, comprising: A plurality of displacement measurement points provided on the moving body; a position measuring means for measuring positions of the plurality of displacement measuring points while the moving body is stationary; an initial position storage means for storing initial positions of the plurality of displacement measurement points measured by the position measurement means while the moving body is stationary in a predetermined posture at a time point when the moving body is in an initial state; a post-elapsed position storage means for storing post-elapsed positions of the plurality of displacement measurement points measured by the position measurement means while the moving body is stationary in the predetermined posture after a predetermined period has elapsed since the moving body was in the initial state; a displacement amount data calculation means for calculating displacement amount data between initial positions and subsequent positions of the plurality of displacement measurement points; a structure model data output means for generating a number of structure models each having a different damaged or deteriorated state for the moving body, and outputting structure data relating to the number of structure models; a structural model analysis means for receiving gravity data acting on the moving body while the moving body is stationary in the predetermined attitude and the structural data output from the structural model data output means, performing a displacement analysis at displacement measurement corresponding points corresponding to the plurality of displacement measurement points for each structural model using a numerical analysis means, and outputting, as learning data, specific information for identifying the structural model that was the subject of the displacement analysis and displacement analysis data at the plurality of displacement measurement corresponding points obtained by the displacement analysis performed on the structural model; a neural network that receives the displacement amount data, performs an inverse analysis, and outputs state estimation information of the moving body; a state display means for displaying a state of the moving object in accordance with the state estimation information output from the neural network; the neural network is previously provided with learning data for the multiple structure models output from the structure model analysis means and is optimized; The predetermined position includes a plurality of positions, the initial position storage means stores initial positions of the plurality of displacement measurement points in a state in which the moving body is stationary in the plurality of postures; the post-elapsed position storage means stores post-elapsed positions of the plurality of displacement measurement points in a state in which the moving body is stationary in the plurality of postures; the displacement amount data calculation means calculates, for each of the plurality of postures, displacement amount data between an initial position and a post-elapsed position of the plurality of displacement measurement points; the structure model analysis means receives, for each of the plurality of postures, gravity data acting on the moving body while the moving body is stationary and the structure data output from the structure model data output means, performs a displacement analysis at the plurality of displacement measurement corresponding points for each structure model using a numerical analysis means, and outputs, as learning data, specific information for identifying the structure model that was the subject of the displacement analysis and displacement analysis data at the plurality of displacement measurement corresponding points obtained by the displacement analysis performed on the structure model; The neural network receives the displacement amount data obtained in each of the plurality of postures, performs an inverse analysis, and outputs state estimation information of the moving body. A moving object state estimation system comprising:

2. the moving body is a rotating body having a rotating shaft and a rotor, The states in which the moving body is stationary in the plurality of postures include a state in which the moving body is stationary so that a rotor center line passing through the center of rotation and the tip of the rotor forms a first predetermined angle with respect to a reference horizontal line that passes through the center of rotation and is perpendicular to the rotation axis, and a state in which the moving body is stationary so that the rotor center line forms a second predetermined angle with respect to the reference horizontal line.

2. The moving object state estimation system according to claim 1 .

3. The rotating body is a wind turbine for wind power generation, The condition of the moving body is the location and size of a damaged or deteriorated part of the wind turbine.

3. The moving object state estimation system according to claim 2.

4. The moving body is a movable body that is movable in at least one direction, The states in which the movable body is stationary in the plurality of postures include a state in which the movable body is stationary so that a movable body straight line passing through a front end and a rear end of the movable body forms a first predetermined angle with respect to a horizontal plane, and a state in which the movable body is stationary so that the movable body straight line forms a second predetermined angle with respect to a horizontal plane.

2. The moving object state estimation system according to claim 1 .

Images