Crack prediction device, fault diagnosis device, crack prediction method, and fault diagnosis method for rotating electrical machine

Abstract

The invention comprises: a shape model setting unit (21) for setting a shape model of an object structure (7) to which an external force is applied, a crack candidate surface (7fc) of a crack expected to be hidden in the shape model, and an observation surface (7fo) of the shape model; an inference model generating unit (22) for sequentially changing the boundary conditions of the crack candidate surface (7fc) in a structural analysis model produced based on the shape model, and generating an inference model obtained by numerically analyzing the structural analysis model; and a crack state analyzing unit (3) for applying an observation surface deformation vector representing the deformation of the observation surface (7fo) obtained by the actual measurement value of the observation surface (7fo), an inference model, and a potential variable representing whether there is a crack (9) in the crack candidate surface (7fc), and simultaneously obtaining the distribution of the load and displacement in the crack candidate surface (7fc) through probabilistic inference, thereby inferring the position and size of the crack (9).

Description

Technical Field

[0001] The present application relates to a crack estimation device, a fault diagnosis device, a crack estimation method, and a fault diagnosis method for a rotating electrical machine. Background Art

[0002] Generally speaking, hidden cracks, invisible from the surface of rotor structures in rotating electrical machines such as those used in turbogenerators, cannot be detected visually. Consequently, cracks can go unnoticed during conventional inspections and grow, impacting the lifespan of the structure. Therefore, detecting hidden cracks is a crucial issue in structural inspections. Non-destructive methods exist for detecting hidden cracks, such as strain measurement of the structural surface, ultrasonic testing (see, for example, Patent Document 1), and X-ray inspection.

[0003] Compared to other non-destructive inspection methods, measuring strain on structural surfaces can be easily miniaturized and implemented at a low cost. However, since the cracks themselves are not directly measured, inverse analysis of the relationship between surface strain and cracks is required to infer hidden cracks that are not visible from the surface.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2012-159477 (Paragraphs 0013 to 0031, Figures 1 to 8 ) Summary of the Invention

[0007] One method for estimating cracks hidden from the surface involves preparing learning data on the relationship between cracks and surface shape changes and estimating the crack's location and size based on this data. However, this estimation involves solving an inverse problem, making it an ill-posed problem. In the case of an ill-posed problem, attempting to estimate hidden cracks from measured data using learning data and the least squares method results in reduced accuracy, making it impossible to determine when the device will stop or remain in use.

[0008] The present application discloses a technology for solving the above-mentioned problems, and its purpose is to estimate the size and position of cracks with high accuracy based on changes in surface shape.

[0009] The crack inference device disclosed in the present application is characterized in that it comprises: a shape model setting unit for setting a shape model of an object structure to be inspected and to which an external force is applied, a crack candidate surface where cracks hidden from the surface of the shape model are expected to occur, and an observation surface on the surface of the shape model that serves as a measurement object; an inference model generating unit for generating a matrix for inferring the state of the crack candidate surface based on the state of the observation surface, based on a matrix that associates the state of the crack candidate surface with the state of the observation surface obtained by numerically analyzing a structural analysis model produced based on the shape model; and a crack state analyzing unit for applying an observation surface deformation vector representing the deformation of the observation surface obtained by actual measurement values of the observation surface, the inference model, and a potential variable representing the presence or absence of a crack on the crack candidate surface, to simultaneously determine the distribution of load and displacement on the crack candidate surface through probabilistic inference, thereby inferring the position and size of the crack.

[0010] The crack inference method disclosed in the present application is characterized in that it includes: a shape model setting step of setting a shape model of an object structure to be inspected and to which an external force is applied, a crack candidate surface where cracks are expected to occur in a portion hidden from the surface of the shape model, and an observation surface on the surface of the shape model that serves as a measurement object; an inference model generating step of generating a matrix for inferring the state of the crack candidate surface based on the state of the observation surface, based on a matrix that associates the state of the crack candidate surface with the state of the observation surface obtained by numerically analyzing a structural analysis model produced based on the shape model; a step of accepting actual measurement values of the observation surface; and a crack state analysis step of applying an observation surface deformation vector representing the deformation of the observation surface obtained from the measurement value, the inference model, and a potential variable representing the presence or absence of a crack on the crack candidate surface, to simultaneously determine the distribution of load and displacement on the crack candidate surface through probabilistic inference, thereby inferring the position and size of the crack.

[0011] According to the crack estimation device or crack estimation method disclosed in the present application, latent variables representing the displacement of the surface where cracks occur and the properties of the load are used in the inverse analysis, so the state of hidden cracks can be estimated with high accuracy based on changes in surface shape. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a block diagram for explaining the configuration of the crack estimation device according to the first embodiment.

[0013] Figure 2 This is a flowchart showing the operation of the crack estimation device and the crack estimation method according to the first embodiment.

[0014] Figure 3A and Figure 3BThese are perspective views showing the operation of the crack estimation device according to the first embodiment and the relationship between the crack candidate surface and the observation surface in the crack estimation method when a tensile load and a bending moment are applied to a flat plate, respectively.

[0015] Figure 4A and Figure 4B These are schematic diagrams showing the operation of the crack estimation device according to the first embodiment and the state of dividing the crack candidate surface for each element and the state of dividing the observation surface for each element in the crack estimation method, respectively.

[0016] Figure 5 This is a flowchart showing the operation of the estimation model generating unit or the estimation model generating procedure of the crack estimation device according to the first embodiment.

[0017] Figure 6 This is a diagram showing a memory structure for storing displacement information of a crack candidate surface in the crack estimation device or the crack estimation method according to the first embodiment.

[0018] Figure 7 This is a diagram showing a memory structure for storing deformation information of an observation surface in the crack estimation device or the crack estimation method according to the first embodiment.

[0019] Figure 8 This is a diagram showing a memory structure for storing load information on a crack candidate surface in the crack estimation device or the crack estimation method according to the first embodiment.

[0020] Figure 9 This is a flowchart showing the operation of the crack state estimation unit or the crack state estimation procedure of the crack estimation device according to the first embodiment.

[0021] Figure 10 This is a diagram showing an example of a portion of the crack estimation device of each embodiment including the first embodiment executing arithmetic processing or realizing at least a portion of the function of the crack estimation method and the fault diagnosis method of the rotating electrical machine by software.

[0022] Figure 11A and Figure 11B They respectively show the operation of the crack estimation device of each embodiment including embodiment 1, a stereoscopic view showing a coordinate system when internal pressure is applied to a cylinder in a crack estimation method or a fault diagnosis method of a rotating electrical machine, and a top view showing a state where internal pressure is applied.

[0023] Figure 12 This is a diagram showing a memory structure for storing information on displacement changes of an observation surface in the crack estimation device or crack estimation method according to the second embodiment.

[0024] Figure 13This is a diagram showing a memory structure for storing information on angle changes of an observation surface in the crack estimation device or the crack estimation method according to the second embodiment.

[0025] Figure 14 This is a block diagram for explaining the configuration of a crack estimation device according to a third embodiment.

[0026] Figure 15 This is a flowchart showing the operation of the crack estimation device or the crack estimation method according to the third embodiment.

[0027] Figure 16 This is a diagram showing the overall configuration of a fault diagnosis device according to a fourth embodiment.

[0028] Figure 17 This is a flowchart showing additional operations of the fault diagnosis device or additional steps of the fault diagnosis method for a rotating electrical machine according to the fourth embodiment.

[0029] Figure 18 This is a flowchart showing additional operations of the fault diagnosis device or additional steps of the fault diagnosis method for a rotating electrical machine according to the fifth embodiment.

[0030] (Explanation of Symbols)

[0031] 1: Crack estimation device; 2: Model generation unit; 21: Shape model setting unit; 211: Inspection load setting unit; 22: Estimation model generation unit; 3: Crack state analysis unit; 31: Measurement data acquisition unit; 311: Inspection load indication unit; 32: Crack state estimation unit; 5: Fault diagnosis device; 52: Terminal; 53: Alarm device; 6: Measuring device (measuring instrument); 7: Target structure; 70: Rotating electrical machine; 7fc: Crack candidate surface; 7fo: Observation surface; 9: Crack; D: Forward coefficient matrix; Efc: Element; Efo: Element; E measure : Observation surface matrix; G: rigidity matrix; Lt: tensile load (external force); Mb: bending moment (external force); Pi: internal pressure (external force); Δ ans : variation distribution (posterior distribution); Δ crack_diff : crack surface matrix; λ: reference value (judgment criterion). DETAILED DESCRIPTION

[0032] Hereinafter, a crack estimation device, a crack estimation method, and an inspection method for a rotating electrical machine according to various embodiments of the present application will be described with reference to the accompanying drawings. In each drawing, identical or corresponding components and locations are denoted by identical reference numerals.

[0033] Implementation method 1.

[0034] Figure 1 to Figure 11 is a diagram for explaining the structure and operation of the crack estimation device and the crack estimation method according to the first embodiment. Figure 1 This is a block diagram for explaining the structure of a crack estimation device. Figure 2 FIG3 is a flow chart showing the operation of the crack estimation device or the crack estimation method. FIG3 is a perspective view showing the relationship between the surface where cracks are expected to occur (crack candidate surface) when a tensile load is applied to the flat plate and the observation surface that can be measured from the outside ( Figure 3A ) and a perspective view showing the relationship between the crack candidate surface and the observation surface when a bending moment is applied to the flat plate ( Figure 3B 4 is a schematic diagram showing a state in which the crack candidate surface is divided into each element in a grid shape for structural analysis ( Figure 4A ) and a schematic diagram showing the state of dividing the observation surface for each element ( Figure 4B ).

[0035] and, Figure 5 This is a flowchart showing the operation of the estimation model generating unit constituting the crack estimation device or the estimation model generating procedure. Figures 6 to 8 They are shown in the generated Figure 5 This is a diagram of the memory structure used to store the inference model described in the Figure 6 FIG. 1 is a diagram showing a memory structure for storing displacement information of a crack candidate surface. Figure 7 is a diagram showing a memory structure for storing deformation information of an observation surface, Figure 8 It is a diagram showing the structure of a memory for storing load information on crack candidate surfaces.

[0036] Figure 9 This is a flowchart showing the operation of the crack state estimation unit or the crack state estimation step constituting the crack estimation device. Figure 10 FIG11 is a diagram showing an example of hardware in which the computational processing portion of the crack estimation device or at least a portion of the functions of the crack estimation method according to the first or subsequent embodiments is implemented by software. FIG12 is a perspective view showing a coordinate system when internal pressure is applied to a cylinder in the crack estimation device or the crack estimation method. Figure 11A ) and a top view showing a state where internal pressure is applied ( Figure 11B ).

[0037] In the crack estimation devices, crack estimation methods, and rotating electrical machine inspection methods of various embodiments, including this one, the displacement of the surface where the cracks are occurring and the properties of the load are used as advance information to resolve ill-posed problems. This allows the location and size of hidden cracks to be estimated with high accuracy based on surface shape changes. This is described in detail below.

[0038] Crack estimation device 1 Figure 1As shown, it comprises: a model generating unit 2 that calculates learning data to generate an inference model; a crack state analyzing unit 3 that uses the generated inference model to analyze the state of cracks based on the measured values of the inspection object surface; and an analysis result output unit 4 that outputs the analysis result.

[0039] The model generation unit 2 includes a shape model setting unit 21 that sets learning conditions, and an inference model generation unit 22 that generates an inference model for use in inferring cracks based on the set learning conditions, and outputs the generated model to the crack state analysis unit 3. Furthermore, the crack state analysis unit 3 includes a measurement data acquisition unit 31 that acquires measurement data representing the surface condition of the inspection object, such as strain and displacement, and a crack state inference unit 32 that applies the measurement data to the inference model, calculates the position and size of cracks hidden from the surface, and outputs the calculated model to the analysis result output unit 4.

[0040] The shape model setting unit 21 has the function of determining the location of cracks to be inspected as learning conditions, creating a shape model of the inspection location, determining the hypothetical crack shape, and outputting it to the inference model generation unit 22. The inference model generation unit 22 gradually changes the hypothetical crack shape and location, performs numerical analysis on the structural analysis model created from the shape model, stores the displacement and load of the candidate crack surface, and the deformation of the observation surface as vectors in a primary storage unit, and represents the stored analysis results of all crack shapes in a matrix. Then, using the linear relationship between the displacement of the candidate crack surface and the deformation of the observation surface, the forward coefficient matrix between the displacement matrix of the crack and the deformation matrix of the observation surface is calculated. Furthermore, using the linear relationship between the displacement of the candidate crack surface and the load, the inference model generation unit 21 has the function of calculating the rigidity matrix of the displacement matrix of the crack and the load matrix of the crack, and outputting it to the crack state estimation unit 32.

[0041] The measurement data acquisition unit 31 acquires measurement data when measuring the deformation of the observation surface of the target structure being inspected. The crack state estimation unit 32 uses the displacement matrix calculated based on the measurement data output from the measurement data acquisition unit 31 and the load matrix and stiffness matrix of the crack output as the estimation model to estimate the size and position of the crack based on the displacement of the candidate crack surface. At this time, by using latent variables representing the sparsity of the displacement and load of the candidate crack surface, which have an inverse relationship, the displacement and load are estimated simultaneously using JE-MAP (Joint Estimation-Maximum A Posteriori). Furthermore, the usability is determined based on the estimated crack size, and the remaining usable period is calculated and output to the analysis result output unit 4.

[0042] The analysis result output unit 4 outputs the estimated size and position of the crack based on the analysis result output from the crack state estimation unit 32. Furthermore, for example, as an inspection result for a device having a structure such as a rotating electrical machine, the device can be further used and the remaining useful life of the device can be output.

[0043] according to Figure 2 The following flowchart illustrates the operation based on the above-described structure. First, the model generation unit 2 executes a learning phase (steps S2110 to S2200) in which the conditions for learning data are determined, learning data is generated according to the determined conditions, and an inference model used for inferring the crack state is generated based on the generated learning data. Then, based on the generated inference model, the crack state analysis unit 3 executes a crack state analysis phase (steps S3100 to S3200) in which the crack state is estimated based on the measured data of the surface state of the inspection object. Finally, the analysis result output unit 4 executes a result output phase (step S4000) in which the analysis result is displayed (output).

[0044] <Learning Stage>

[0045] In the learning phase, as learning data conditions, a step of determining an inspection target is performed to determine a structure to be inspected for cracks, a location hidden from the surface of the structure to be estimated for cracks, and a location on the surface of the structure to be measured for estimating cracks (step S2110). For example, Figure 3A As shown, a flat target structure 7 is determined as a structural object, and a candidate crack surface 7fc is determined as a target for estimating the presence or absence of cracks 9. Then, as a surface to be measured, an area on the surface of the target structure 7 close to the candidate crack surface 7fc is determined as an observation surface 7fo for measuring strain.

[0046] Then, assuming that a tensile load Lt is applied to the target structure 7 during inspection, a shape model of the inspection area is set for the entire target structure 7 or a portion thereof (step S2120). When modeling the entire target structure 7, in addition to the load, deformation constraints and temperature distribution imposed on the target structure 7 are considered as boundary conditions for structural analysis. On the other hand, when modeling a portion of the target structure 7, the displacement of the cut surface or the distribution of the load are reflected as boundary conditions in the structural analysis.

[0047] In addition, in the target structure 7, the tensile load Lt is not limited to the above-mentioned tensile load Lt, for example, Figure 3B The bending moment Mb is shown in the figure, but the inspection object is determined in the same way and the shape model is set. Figure 4A As shown in FIG, the crack candidate surface 7fc in the shape model is divided into a grid and decomposed into a plurality of elements Efc. Figure 4AIn the example, the crack candidate surface 7fc is divided into n pieces in the x direction and m pieces in the y direction, and the positions of the intersections of the grids that form the dividing lines are represented by coordinates (i, j). Therefore, the coordinates (i, j) are represented by numbers from (0, 0) to (n, m).

[0048] A point where the grids intersect is defined as a crack 9, and the crack 9 is sequentially moved on the points where all grids intersect (grid points: coordinates (i, j)). In step S2120, the order of movement is determined. For each structural analysis in which the boundary conditions of the crack 9 of the crack candidate surface 7fc are changed, the displacement and load obtained at the grid points of the crack candidate surface 7fc are stored in the determined order. In addition, the components of the stored displacement are for Figure 3A or Figure 3B The tensile load Lt or bending moment Mb shown is the component that is most displaced in the crack candidate surface 7fc.

[0049] Then, if Figure 4B As shown, an observation surface 7fo is determined as a measurement target for estimating the crack 9 and as a range for obtaining deformation information of the surface of the target structure 7. In this embodiment, strain is measured as the surface deformation. The observation surface 7fo is also divided into a plurality of elements Efo in a grid-like manner, similar to the crack candidate surface 7fc. Figure 4B In

[15] , the observation plane 7fo is divided into n pieces in the x-direction and p pieces in the z-direction, and the coordinates (k, l) represent the positions of the intersection points of the grids that form the dividing lines. Therefore, the coordinates (k, l) are represented by numbers from (0, 0) to (n, p). Furthermore, the observation plane 7fo can be defined as a portion of a continuous plane, a collection of multiple ranges, or a range composed of points.

[0050] Furthermore, for each structural analysis in which the boundary conditions of the crack 9 of the crack candidate surface 7fc are changed, the strains obtained at the grid points of the observation surface 7fo are stored in the determined order. The order of storage is also determined in step S2120. Figure 3A or Figure 3B The tensile load Lt or bending moment Mb shown is the component with the largest strain. Furthermore, regarding the strain when a load or moment is applied to two axes (e.g., the z-axis and the y-axis), principal strain, Tresca's (based on the yield condition) equivalent strain, and Mises' (based on the yield condition) equivalent strain, which are parameters for evaluating the strain of a multiaxial stress field generated by the influence of multiaxial loads and structures, can also be used.

[0051] When the shape model is set in steps S2110 and S2120, the set items are used to automatically generate learning data and calculate the inverse matrix generated from the learning data in the estimated model generation step S2200. The above-mentioned inspection target determination step S2110 and shape model setting step S2120 are performed by the shape model setting unit 21, while the estimated model generation step S2200 described below is performed by the estimated model generation unit 22.

[0052] refer to Figure 5 , which illustrates the detailed process in the inference model generation step S2200. First, the shape model including the crack candidate surface 7fc and the observation surface 7fo determined in steps S2110 and S2120, and the shape, position, and learning order of the crack 9 assumed for learning are read in (step S2210). Then, a structural analysis model is created based on the shape model (step S2220). Furthermore, the crack candidate surface 7fc and the observation surface 7fo in the created structural analysis model are divided into a plurality of elements Efc and Efo, respectively, and boundary conditions in which no crack 9 occurs are provided. The displacement of the crack candidate surface 7fc and the deformation of the observation surface 7fo are calculated by structural analysis (step S2230).

[0053] Next, the crack candidate surface 7fc of the structural analysis model is divided into a plurality of elements Efc. Each node included in the element Efc is used as a boundary condition for a crack 9, and the deformation of the observation surface 7fo is calculated using structural analysis (step S2240). Then, for each condition that defines a node as a crack 9, the displacements of all nodes on the crack candidate surface 7fc and the differences before and after the occurrence of the crack 9 under load are arranged in the learned order to create a vector. Furthermore, for all nodes on the observation surface 7fo, the differences before and after the occurrence of the deformed crack 9 are arranged in the learned order to create a vector (step S2250), and the created vector is stored in the primary storage unit (step S2260).

[0054] Next, for each node of the candidate crack surface 7fc, a determination is made as to whether the structural analysis is complete (step S2270). If not complete ("No" in step S2270), all nodes of the candidate crack surface 7fc are considered to be in a cracked state, and the nodes considered to be in a cracked state are changed (step S2280). The process then returns to step S2240, where the structural analysis is performed and the process of storing the vectors in the primary storage unit is repeated.

[0055] On the other hand, if the structural analysis of all nodes of the candidate crack surface 7fc as being in a cracked state is completed ("Yes" in step S2270), the process proceeds to step S2290. In step S2290, the displacement vectors of the candidate crack surface 7fc stored in the primary storage unit are arranged in the order of the read learning to create a crack surface matrix. The observation surface matrix is then created based on the deformation vectors of all nodes on the observation surface 7fo stored in the primary storage unit (step S2290).

[0056] Specifically, if Figure 6 As shown, the column vector Δ(0,0) of the displacement change of the crack candidate surface 7fc arranges the displacement data of the nodes of the crack candidate surface 7fc in the order determined in the shape model setting step S2120. δ(i,j) in the column vector Δ(0,0) represents the displacement of the node at the coordinate (i,j) on the crack candidate surface 7fc. Furthermore, the column vector Δ(i,j) is created as the coordinate (i,j) of the crack candidate surface 7fc with respect to the position (coordinate) of the crack 9 learned. i_j (i, j) represents the element in the column vector. i_j (i, j) represents the displacement of the node at coordinate (i, j) in the candidate crack surface 7fc as a result of structural analysis using the node at coordinate (i, j) as the crack state. This column vector is arranged for each row in the order of the locations of the cracks 9 determined in the shape model setting step S2120 to create a crack surface matrix Δ crack_diff .

[0057] Then, if Figure 7 As shown, the column vector E(0,0) of the strain change of the observation surface 7fo is also arranged in the order determined in the shape model setting step S2120. ε(k,l) in the column vector E(0,0) represents the strain of the node at the coordinate (k,l) on the observation surface 7fo. Furthermore, the information about the location (coordinates) of the crack occurrence is used as the coordinates (i,j) of the crack candidate surface 7fc, and the column vector E(i,j) is created, and ε is used to represent the strain data of the nodes at the coordinates (k,l) on the observation surface 7fo. i_j (k, l) represents the elements in the column vector. i_j (k, l) represents the strain at the node at coordinates (k, l) on the crack candidate surface 7fc as a result of structural analysis using the node at coordinates (i, j) on the crack candidate surface 7fc as a crack state.

[0058] The column vectors are arranged in the order of the locations where the cracks 9 are generated in the shape model setting step S2120, and the observation plane matrix E is created. measure .

[0059] In this way, when making the crack surface matrix Δ crack_diff and the observation surface matrix E measure When the linear relationship between the displacement of the candidate crack surface 7fc and the deformation of all nodes in the observation surface 7fo is used, the forward coefficient matrix D for mapping from the crack surface matrix to the observation surface matrix is defined as shown in formula (1). Then, by multiplying both sides of formula (1) from the right side by the crack surface matrix Δ as shown in formula (2), crack_diff The inverse matrix Δ crack_diff -1 , as shown in formula (3) according to the crack surface matrix Δ crack_diff and the observation surface matrix E measure Create a forward coefficient matrix D (step S2300).

[0060] [Mathematical formula 1]

[0061] Mathematical formula 1

[0062] DΔ crack_diff =E measure (1)

[0063] DΔ crack_diff [Δ crack_ddiff ] -1 =E measure [Δ crack_diff ] -1 (2)

[0064] D=E measure [Δ crack_diff ] -1 (3)

[0065] Next, the load vectors of the crack candidate surface 7fc stored in the primary storage unit are arranged in the order of learning to create a load matrix of the crack candidate surface 7fc (step S2310). Figure 8 As shown, the column vector Z(0,0) of the load change of the candidate crack surface 7fc also arranges the load data of the nodes of the candidate crack surface 7fc in the order determined in the shape model setting step S2120. ζ(i,j) in the column vector Z(0,0) represents the load of the node at the coordinate (i,j) on the candidate crack surface 7fc. Furthermore, regarding the information on the position (coordinate) where the crack 9 has been learned, the column vector of the coordinate (i,j) is created as the coordinate (i,j) of the candidate crack surface 7fc, and ζ(i,j) is used. i_j (i, j) represents the element in the column vector. i_j (i, j) represents the load (reaction force) of the node at the coordinate (i, j) position on the candidate crack surface 7fc, as a result of structural analysis using the node at the coordinate (i, j) position on the candidate crack surface 7fc as the crack state. This column vector is arranged in the order of the positions (coordinates) of the determined cracks 9 for each row to create a load matrix Zcrack_diff .

[0066] Next, based on the linear relationship between the displacement of the crack candidate surface 7fc and the load, a stiffness matrix G is created, and a load vector of the crack candidate surface 7fc in the absence of cracks is created. Load matrix Z crack_diff and the crack surface matrix Δ crack_diff The relationship is expressed as the following equation (4). no_crack is the load vector of the crack candidate surface 7fc when there is no crack. crack_diff Move to the right, so that GΔ crack_diff Move to the left and multiply the crack surface matrix Δ from the right on both sides in the same way as in formula (2). crack_diff The inverse matrix Δ crack_diff -1 , as shown in formula (5), the rigidity matrix G is obtained.

[0067] [Mathematical formula 2]

[0068] Mathematical formula 2

[0069] Z crack_diff =GΔ crack_diff +Z no_crack (4)

[0070] G=(z crack_diff -Z no_crack )[Δ crac k_diff ] -1 (5)

[0071] Then, the prepared forward coefficient matrix D, stiffness matrix G, and load vector in the absence of cracks are output to the crack state analyzer 3 as a model for estimating cracks (step S2320 ), and the learning phase ends.

[0072] <Crack Status Analysis Phase>

[0073] Once the above-mentioned learning phase is completed, a crack state analysis phase can be executed to estimate the state of cracks in the target structure 7, which can be used for fault detection. Figure 9 The flowchart of FIG. 1 illustrates the operation of the crack state analysis stage executed by the crack state analysis unit 3.

[0074] In the crack state analysis phase, first, as preparation for analyzing the crack state, the crack state estimation unit 32 reads the forward coefficient matrix D, the stiffness matrix G, and the load vector data for the absence of cracks as the estimation model generated in the learning phase (step S3210). Then, when estimating the state of cracks, for example, for fault detection, the measurement data acquisition unit 31 acquires measurement data obtained from the observation surface 7fo of the actual measurement target structure 7.

[0075] As an example of measurement data, strain will be used for explanation. Measurement methods include strain gauges and digital image correlation. Regarding strain, for example, the strain value measured before or immediately after use of the target structure 7 is retained as an initial value, assuming no cracks 9 have occurred in the portion hidden from the surface. Furthermore, the difference between the strain value measured at the time when cracks 9 are suspected to have occurred, as in fault diagnosis, and the previously retained initial value is obtained as strain data.

[0076] The measured strain data are arranged in the order determined in the inference model to form the observation surface matrix E measure The same strain column vector (step S3220). The measured strain data is set to the same position as the strain data during learning. Next, as shown in formula (6), a vector L of potential variables indicating whether there is a crack 9 on the crack candidate surface 7fc is introduced (step S3230). The i and j in lv(i, j) in the vector L represent the load matrix Z. crack_diff , crack surface matrix Δ crack_diff The same position (coordinates) within the crack candidate surface 7fc.

[0077] [Mathematical formula 3]

[0078] Mathematical formula 3

[0079]

[0080] Here, when lv(i, j) is 1, it is defined as “there are cracks”, and when it is 0, it is defined as “there are no cracks”, and it takes the value of 0 or 1.

[0081] Next, the probability distribution of the displacement and load on the candidate crack surface 7fc is calculated using the latent variable representing the sparsity of the displacement and load in the inverse relationship on the candidate crack surface 7fc. This probability distribution is used as the prior distribution (step S3240). Here, the inverse relationship means that if one exists, the other does not exist, and if the other exists, one does not exist, and the two cannot exist simultaneously.

[0082] Here, the probability distribution is assumed to be normal distribution, and the expected value and covariance of the displacement and load of the crack candidate surface 7fc are used as functions of the latent variables. The vector of the expected value of the displacement and the matrix of the covariance are respectively Δ ex , Δ cov , let the vector of expected values of the loads and the matrix of covariance be Z ex , Z covEach vector and matrix element becomes a function of the latent variable, and is set to have an appropriate expected value and variance in the crack 9 and the non-crack part. The posterior distribution is expressed by equation (7).

[0083] p(Δ ans │L)=Norm(Δ ans │Δ ex , Δ cov )Norm(Z ans │Z ex , Z cov )(7)

[0084] Next, the strain distribution of the observation result is set as the expected value E of the noise noiseex , the noise during measurement is set to variance E noisecov , using the forward coefficient matrix D of the inferred model, as shown in formula (8), the displacement distribution Δ of the inferred crack candidate surface 7fc is obtained ans and the expected value of the observation result E m The probability distribution of the difference is set as the likelihood distribution (step S3250).

[0085] L(Δ ans │E m )=Norm(DΔ ans -E m │E noiseex , E noisecov ) (8)

[0086] Finally, the displacement distribution Δ is estimated by Bayesian estimation in a manner that satisfies the following formula (9): ans Here, as Bayesian inference, the displacement distribution Δ is estimated by JE-MAP inference that maximizes the posterior probability. ans (Step S3260).

[0087] [Formula 4]

[0088] Mathematical formula 4

[0089]

[0090] According to the obtained displacement distribution Δ ans The strain distribution of the surface is obtained by using the forward coefficient matrix D, and the expected value E of the observed result is calculated using the following formula (10): m difference (step S3270).

[0091] │Δ ans DE m │≦λ (10)

[0092] Then, it is determined whether the calculated difference value is less than the predetermined convergence judgment reference value λ (step S3280). If the difference value exceeds the reference value λ ("No" in step S3280), the process proceeds to step S3240, and the displacement distribution Δ ans Then, if the difference value converges to or below the reference value λ ("Yes" in step S3280), it is determined that the displacement distribution Δ obtained at that time point is ans Convergence, according to the displacement distribution Δ of the converged crack candidate surface 7fc ans , find the size and position of the crack 9 (step S3290).

[0093] The information on the size and position of the crack 9 obtained is output to the analysis result output unit 4, and the process transitions to the analysis result output stage (step S4000). This concludes the crack state analysis stage, and the process transitions to the analysis result output stage (step S4000) using the analysis result output unit 4. In the analysis result output stage, the position and size of the crack 9 output from the crack state analysis unit 3 are displayed as inspection results.

[0094] Furthermore, it is also considered that the parts that perform calculation processing in the crack estimation device 1 of the first embodiment, particularly the model generation unit 2 and the crack state analysis unit 3, are composed of the following components: Figure 10 The hardware 10 shown includes a processor 101 and a storage device 102. Although not shown, the storage device 102 includes a volatile storage device such as a random access memory and a non-volatile auxiliary storage device such as a flash memory. Alternatively, a hard disk auxiliary storage device may be used instead of the flash memory. The processor 101 executes programs input from the storage device 102.

[0095] In this case, the program is input from the auxiliary storage device to the processor 101 via the volatile storage device. Furthermore, the processor 101 can output data such as computation results to the volatile storage device of the storage device 102, or store the data in the auxiliary storage device via the volatile storage device. This hardware 10 can be connected to a measuring instrument or retrieve the strain measurement results for use in data processing. In other words, the crack estimation method of this embodiment 1 can also be executed using this hardware 10. Of course, this can also be applied to the crack estimation device 1, crack estimation method, or rotating electrical machine inspection method of each subsequent embodiment.

[0096] In the above description, a flat plate is assumed as the target structure 7 and is represented by an xyz orthogonal coordinate system, but the present invention is not limited thereto. Figure 11A The cylindrical coordinate system shown in FIG. 7 is a cylindrical structure 7 and has coordinates RθZ. In this case, Figure 4A as well as Figure 4B The x shown corresponds to R, y corresponds to θ, and z corresponds to Z. The cylindrical structure to be targeted is as follows Figure 11B The structure in which the internal pressure Pi is applied to the inner peripheral surface 7fi as in the shrink fitting portion shown, targets a structure in which the shape of the surface changes due to cracks occurring in a portion hidden from the surface.

[0097] An example of a target structure to which the cylindrical coordinate system is applied is a shrink fit portion between a retaining ring that is shrink fit to an end portion of a rotor of a rotating electrical machine and a rotor core.

[0098] Implementation method 2.

[0099] While the first embodiment described above illustrates an example of estimating crack states based on strain changes as deformation of the observation surface, this is not limiting. Furthermore, it is impossible to learn all desired crack shapes. This second embodiment describes an example of estimating crack states based on displacement and angle changes. In particular, it is possible to efficiently learn a small amount of crack data, and based on changes in the observation surface, estimate the location and size of cracks in any portion hidden from view.

[0100] Figure 12 and Figure 13 This is a diagram showing the crack estimation device or crack estimation method according to the second embodiment. Figure 5 This is a diagram of the memory structure used to store the inference model described in the Figure 12 FIG. 1 is a diagram showing a memory structure for storing information on displacement changes of an observation surface. Figure 13 This diagram illustrates the memory structure for storing information on angle changes in the observation surface. The crack estimation device or crack estimation method according to Embodiment 2 is identical to that described in Embodiment 1, except for the operations related to the memory structure (steps S2240 to S2290). Therefore, the description will focus on the differences from Embodiment 1, and the figures used in Embodiment 1 will be used.

[0101] Even when using inflectional variations, such as Figure 12 As shown, the column vector Dis(0,0) of the displacement change of the observation surface 7fo is arranged in the order determined in the shape model setting step S2120. i_j (k, l) represents the displacement change of the node at the coordinate (k, l) on the observation surface 7fo. Furthermore, information on the location (coordinates) of the crack 9 learned is used as the coordinate (i, j) of the crack candidate surface 7fc to create a column vector Dis(i, j). i_j (k, l) represents the elements in the column vector. i_j(k, l) represents the displacement change of the node at coordinate (k, l) in the candidate crack surface 7fc as a result of structural analysis using the node at coordinate (i, j) in the candidate crack surface 7fc as the crack state. This column vector is arranged for each row in the order of the positions of the cracks 9 determined in the shape model setting step S2120 to create a matrix Dis measure .

[0102] In addition, when using angle changes, such as Figure 13 As shown, the column vector A(0,0) of the displacement change of the observation plane 7fo is arranged in the order determined in the shape model setting step S2120, and the angle change data of the nodes of the observation plane 7fo are arranged. i_j (k, l) represents the angle change of the node with the coordinate (k, l) on the observation surface 7fo. Furthermore, information on the position (coordinate) of the crack 9 is learned, and the column vector A(i, j) is created as the coordinate (i, j) of the crack candidate surface 7fc. i_j (k, l) represents the elements in the column vector. i_j (k, l) represents the angle change of the node at coordinates (k, l) on the candidate crack surface 7fc as a result of structural analysis using the node at coordinates (i, j) on the candidate crack surface 7fc as the crack state. This column vector is arranged for each row in the order of the locations of the cracks 9 determined in the shape model setting step S2120 to create a matrix A. measure .

[0103] By using displacement or angle changes in this way, the process of creating learning data corresponding to all the shapes of the cracks 9 occurring on the candidate crack surface 7fc can be automated. As a result, the position and size of cracks 9 in any portion hidden from the surface can be estimated based on changes in the observation surface 7fo, using a small amount of learning data. Furthermore, by using not only strain changes but also displacement and angle changes as the deformation of the observation surface 7fo, the variety of measurement methods can be expanded, enabling faster and more accurate measurements compared to strain measurement.

[0104] Implementation method 3.

[0105] In Embodiment 1 or 2, deformation due to cracks must be generated on the observation surface during inspection, so the target structure is limited to structures where a force is pre-applied, such as shrink-fit structures. However, even when no force is pre-applied to the target structure, measurement can be performed by applying a constant load to the target structure under crack-free conditions and during inspection. In the crack estimation device or crack estimation method of Embodiment 3, by applying a constant load to the target structure, the crack state can be estimated even when no force is pre-applied to the target structure.

[0106] Figure 14 and Figure 15 This is a diagram for explaining the structure and operation of a crack estimation device, a crack estimation method, or a rotating electrical machine inspection method according to a third embodiment. Figure 14 This is a block diagram for explaining the structure of a crack estimation device. Figure 15 This is a flowchart illustrating the operation of the crack estimation device or the crack estimation method. The crack estimation device or the crack estimation method of this third embodiment is identical to that described in the first embodiment, except for the configuration of the shape model setting unit and the measurement data acquisition unit described in the first and second embodiments, and the operation related to the inspection load (steps S2115 and S3090). Therefore, the description will focus on the parts that differ from the first embodiment, and the figures used in the first embodiment will be used.

[0107] like Figure 14 As shown, the shape model setting unit 21 is provided with a test load setting unit 211 for setting the magnitude and position of the load applied as an external force to the target structure 7. Furthermore, the measurement data acquisition unit 31 is provided with a test load indicating unit 311 for indicating the magnitude and position of the load set by the test load setting unit 211 to be applied to the target structure 7.

[0108] Moreover, if Figure 15 As shown, during the learning phase, a test load setting step (S2115) is provided to set the magnitude and position of the test load. In the shape model setting step (S2120), the set test load is added to the boundary conditions during structural analysis. Furthermore, during the crack state analysis phase, a test load indication step (S3090) is provided. In the measurement data acquisition step (S3100), an external force is applied to the target structure 7 according to the indicated magnitude and position, and the surface condition is measured. This allows inspection of target structures 7 that have not been previously subjected to a force, expanding the range of target structures 7 that can be inspected.

[0109] The configuration of the inspection load indicating unit 311 is not limited to indicating a load, and may be modified to automatically apply a load to the target structure 7. In this case, the inspection load indicating step S3090 may be rewritten as an inspection load applying step.

[0110] In this way, by applying a load according to a set size and position to the target structure 7 during inspection, it is also possible to inspect a target structure to which no force has been applied in advance, thereby expanding the scope of objects that can be inspected.

[0111] Implementation method 4.

[0112] In the above embodiment, the configuration and operation required for estimating the crack state of the target structure are described. In this fourth embodiment, a fault diagnosis device and a fault diagnosis method for a rotating electrical machine are further described for performing fault diagnosis on the target structure. Figure 16 and Figure 17 1 is a diagram for explaining the structure and operation of a fault diagnosis device and a fault diagnosis method for a rotating electrical machine according to a fourth embodiment. Figure 16 is a diagram showing the overall structure of a fault diagnosis device, Figure 17 This is a flowchart showing an additional step for the operation described in the crack state estimation device or the crack estimation method as the operation of the fault diagnosis device or the fault diagnosis method of the rotating electrical machine.

[0113] Furthermore, the fault diagnosis device or the fault diagnosis method for a rotating electrical machine of the fourth embodiment is an example of adding the necessary configuration and operation for fault diagnosis to the crack estimation device or the crack estimation method described in any of the first to third embodiments. Therefore, the description of the first to third embodiments will be used, with the description focusing on the additional portions.

[0114] The fault diagnosis device 5 of the fourth embodiment is as follows Figure 16 As shown, the system comprises: the crack estimation device 1 described in Embodiments 1 to 3; and a terminal 52 having functions for inputting operating conditions, physical property values, and other parameters required for fault diagnosis, and displaying the diagnostic results. Furthermore, the system is configured to acquire measurement data from a measuring device 6 that measures the surface condition of a rotor of a rotating electrical machine 70, serving as a target structure 7, and to estimate the size and location of cracks 9 based on the rotor's surface condition. Furthermore, the crack estimation device 1 uses, for example, the external force applied to the rotor, the rotor's physical property values, and the limit value input from terminal 52 to calculate the operating time until the limit value is reached. The alarm device 53 will be described later in Embodiment 5.

[0115] Therefore, if Figure 17As shown, for example, a step (step S3300) is added in which data such as the external force applied to the target structure 7, such as a rotor, the physical properties of the target structure, and the size and location of cracks that serve as thresholds for rendering the target structure 7 unusable are input from the terminal 52. Furthermore, this data can be obtained during the product design phase.

[0116] Furthermore, a step is added (step S3310) to calculate the crack growth rate under the target product's usage conditions based on the information on the location and size of the crack 9 calculated in step S3290, using the input data and knowledge of fracture mechanics by the crack state estimation unit 32. Furthermore, the growth rate can be estimated not only based on knowledge of fracture mechanics but also based on the results of time-series estimation of the size and location of the crack 9. Furthermore, a step is added to calculate the period of use until the crack 9 reaches a size and location that renders the target structure unusable.

[0117] Thus, in addition to outputting the estimation result, analysis result output unit 4 also outputs information about the remaining usage period to terminal 52 (step S4000). Terminal 52 presents information including the remaining usage period based on the output from analysis result output unit 4 (step S5100). Furthermore, the function of terminal 52 can be substituted by analysis result output unit 4 described in embodiments 1 to 3.

[0118] This allows not only the location and size of the crack 9 but also the remaining usable time of the device to be understood, enabling planned repair and update of the device, thereby functioning as a fault diagnosis device 5 or a fault diagnosis method for a rotating electrical machine.

[0119] Implementation method 5.

[0120] In the fourth embodiment, the remaining service life of the fault diagnosis notification device is described as an example, but the present invention is not limited thereto. In the fifth embodiment, an alarm is provided to more actively notify a fault signal. Figure 18 It is a diagram for illustrating the operation of the fault diagnosis device of embodiment 5 and the fault diagnosis method of a rotating electrical machine, and is a flowchart showing the operation of the fault diagnosis device or the fault diagnosis method of a rotating electrical machine as an additional step for the operation described in the crack inference device or the crack inference method.

[0121] Furthermore, the fault diagnosis device or the fault diagnosis method of the rotating electrical machine of the present embodiment 5 is an example of adding a structure and an operation necessary for fault diagnosis to the crack estimation device or the crack estimation method described in any of the embodiments 1 to 3. Therefore, the description of the embodiments 1 to 3 is cited, and the description will be centered on the additional parts. In addition, as the structure of the fault diagnosis device, the structure of the crack estimation device or the crack estimation method described in the embodiment 4 is cited. Figure 16 .

[0122] The fault diagnosis device 5 of the fifth embodiment includes a device for Figure 16 The alarm device 53 sounds an alarm according to the contents described above. Furthermore, the size and position of the cracks 9 that can be used as the limit value are input from the terminal 52. Furthermore, the crack estimation device 1 uses the limit value input from the terminal 52, for example, to determine whether there is a crack exceeding the limit value among the estimated cracks 9.

[0123] Therefore, if Figure 18 As shown, after step S3290, a step of inputting a limit value is added (step S3340). Furthermore, a step of determining whether a crack 9 exceeding a size (limit value) that renders it unusable exists at a certain location based on the information on the position and size of the crack 9 calculated in step S3290 and the input limit value data is added (step S3350).

[0124] If there are nine cracks exceeding the threshold value ("Yes" in step S3350), display data for displaying an alarm urging the user to stop using the device is generated (step S3360). On the other hand, if there are no nine cracks exceeding the threshold value ("No" in step S3350), display data for displaying the presence or absence of cracks or the number of cracks is generated (step S3360).

[0125] As a result, the analysis result output unit 4 outputs a warning or display data such as the presence or absence of cracks to the alarm device 53 (step S4000). Based on the output from the analysis result output unit 4, the alarm device 53 displays a warning indicating the suspension of use or the presence or absence of cracks (step S5200). Furthermore, the function of the alarm device 53 can be replaced by the analysis result output unit 4 described in Embodiments 1 to 3.

[0126] In addition, the usage period that reaches the limit value or the estimated size and position of the crack 9 described in the fourth embodiment may be displayed on the terminal 52 .

[0127] Implementation method 6.

[0128] Alternatively, the crack candidate surface 7fc described in Embodiments 1 to 5 can be set as follows. The distribution of the stress causing the crack 9 in the target structure 7 is measured in advance or determined through structural analysis. An evaluation stress suitable for determining the location of the crack 9 based on the material and stress distribution is selected, and the point where this stress reaches its maximum is defined as the crack location. Furthermore, the crack candidate surface 7fc is set perpendicular to the direction of the maximum principal stress in the location and extends through the surface of the target structure 7 facing the location of the crack 9. This setting allows for the preparation of learning data before inspection, shortening the inspection time.

[0129] In addition, this application describes various illustrative embodiments and examples, but the various features, methods, and functions described in one or more embodiments are not limited to application to specific embodiments and can be applied to the embodiments alone or in various combinations. Therefore, within the scope of the technology disclosed in this specification, countless unillustrated variations are envisioned. For example, this includes the case where at least one component is deformed, added, or omitted, and further the case where at least one component is extracted and combined with components shown in other embodiments.

[0130] As described above, the crack estimation device 1 according to each embodiment is configured to include: a shape model setting unit 21 for setting a shape model of an object structure 7 to be inspected and to which an external force (e.g., tensile load Lt, bending moment Mb, internal pressure Pi, etc.) is applied, a crack candidate surface 7fc where cracks are expected to occur in a portion hidden from the surface of the shape model, and an observation surface 7fo on the surface of the shape model to be measured; and an estimation model generating unit 22 for correlating the state of the crack candidate surface 7fc with the state of the observation surface 7fo, based on a matrix obtained by numerically analyzing a structural analysis model created based on the shape model. A matrix is generated for estimating the state of the crack candidate surface 7fc based on the state of the observation surface 7fo; and a crack state analysis unit 3 applies the observation surface deformation vector representing the deformation of the observation surface 7fo obtained through the actual measurement value of the observation surface 7fo, the estimation model, and the potential variable representing the presence or absence of the crack 9 on the crack candidate surface 7fc, and through probabilistic inference, simultaneously calculates the distribution of the load and displacement in the crack candidate surface 7fc, thereby estimating the position and size of the crack 9. Therefore, based on the shape change of the observation surface 7fo that can be directly measured, the size and position of the crack 9 occurring in the part hidden from the surface that is difficult to observe can be estimated with high precision.

[0131] In particular, the crack state analysis unit 3 is configured to use the sparseness of the load and the displacement, which are in an inverse relationship, as latent variables, and the probability distribution of the displacement of the crack candidate surface 7fc obtained simultaneously with the load as a prior distribution, and to use the probability distribution of the displacement of the observation surface 7fo calculated by the prior distribution and the estimation model (displacement distribution Δans ) as the posterior distribution, and then update the posterior distribution (the displacement distribution Δ ans ) and the posterior distribution at the convergence time point (variation distribution Δ ans ), and the size and position of the crack 9 are obtained, the size and position of the crack 9 occurring in the portion hidden from the surface that is difficult to observe can be estimated more reliably and accurately.

[0132] If Bayesian inference is used in probabilistic inference, especially MAP inference, the variation distribution Δ can be estimated with simple calculations. ans .

[0133] If any of the displacement change, strain change, and angle change of the observation surface 7fo is used as the observation surface deformation vector, it can be applied to structures of various shapes as the target structure 7 for estimating the crack 9 .

[0134] If the external force is an inspection load applied during the inspection of the target structure 7, and there is an inspection load setting unit 211 for setting the position and size of the inspection load and an inspection load indicating unit 311 for displaying the position and size of the set inspection load during the inspection, then the target structure 7 for inferring the crack 9 can also be applied to a structure to which no external force is applied in its structure.

[0135] In addition, according to the fault diagnosis device 5 of the present application, if it is constructed to include: the above-mentioned crack inference device 1, which is connected to the measuring instrument (measuring device 6) for measuring the observation surface 7fo of the target structure 7, and is provided with a measurement data acquisition unit 31 for obtaining the measurement value from the measuring instrument (measuring device 6); and a terminal 52, which accepts information on the boundary conditions of the cracks in the target structure 7, and outputs the accepted information on the boundary conditions to the crack inference device, and displays the analysis results in the crack inference device 1, the crack status analysis unit 3 determines whether the size and position of the obtained crack 9 exceed the boundary conditions. If exceeded, a warning of the occurrence of a fault is displayed on the terminal 52 (also including an alarm 53), so that the fault of the target structure 7 can be correctly diagnosed and notified.

[0136] Alternatively, according to the fault diagnosis device 5 of the present application, if it is constructed to include: the above-mentioned crack inference device 1, which is connected to a measuring instrument (measuring device 6) for measuring the observation surface 7fo of the target structure 7, and has a measurement data acquisition unit 31 for acquiring measurement values from the measuring instrument (measuring device 6); and a terminal 52, which accepts component information including the force applied to the target structure 7 during the operation of the device having the target structure 7 and the physical property values of the material constituting the target structure 7, outputs the accepted component information to the crack inference device 1, and displays the analysis results in the crack inference device 1, the crack state analysis unit 3 calculates the development life of the crack 9 based on the calculated size and position of the crack 9 and the component information, and displays the information of the remaining usage period on the terminal 52, then it is possible to correctly diagnose and inform the remaining usage period until the device having the target structure 7 fails.

[0137] In addition, the crack inference method according to each embodiment is constructed to include: a shape model setting step (step S2120) of setting a shape model of the target structure 7 to which an external force (such as a tensile load Lt, a bending moment Mb, an internal pressure Pi, etc.) is applied as an inspection object, a crack candidate surface 7fc where cracks are expected to occur in a portion hidden from the surface of the shape model, and an observation surface 7fo on the surface of the shape model to be measured; and an inference model generation step (steps S2200 (S2210 to S2320)) of sequentially changing the boundary conditions of the crack candidate surface 7fc in the structural analysis model produced based on the shape model, and associating the state of the crack candidate surface 7fc with the state of the observation surface 7fo based on the numerical analysis of the structural analysis model. A matrix is generated from the state of the observation surface 7fo to estimate the state of the crack candidate surface 7fc based on the state of the observation surface 7fo; a step of accepting the actual measurement value of the observation surface 7fo (step S3100); and a crack state analysis step (step S3200 (steps S3210~S3290)), applying the observation surface deformation vector representing the deformation of the observation surface 7fo obtained from the measurement value, the estimation model and the potential variable representing whether there is a crack 9 on the crack candidate surface 7fc, and through probabilistic inference, simultaneously calculating the distribution of load and displacement in the crack candidate surface 7fc, thereby estimating the position and size of the crack 9, so that based on the shape change of the observation surface 7fo that can be directly measured, the size and position of the crack 9 occurring in the part hidden from the surface that is difficult to observe can be estimated with high precision.

[0138] In particular, in the crack state analysis step (step S3200), if the crack surface matrix Δ is calculated from the matrix constituting the estimation model and the analysis result of the crack candidate surface 7fc is expressed in a matrix, crack_diff The observation plane matrix E that represents the analysis result of the observation plane 7fo in a matrix form measureThe forward coefficient matrix D of the mapping and the observation surface deformation vector representing the deformation of the observation surface 7fo of the object structure 7 are used to calculate the probability distribution of the displacement of the crack candidate surface 7fc inferred through the latent variables based on the observation surface deformation vector and the forward coefficient matrix D as the likelihood distribution (step S3250). Through the latent variables representing the sparsity of the load and displacement with opposite properties, the probability distribution of the load and displacement in the crack candidate surface 7fc are simultaneously inferred. The calculated probability distribution of the displacement of the crack candidate surface 7fc is used as the prior distribution (step S3240). Based on the likelihood distribution and the prior distribution, the displacement distribution of the crack candidate surface 7fc is obtained through probability inference (step S3260). This allows the size and position of the crack 9 to be reliably and more accurately inferred.

[0139] In addition, the estimation model generation step (step S2200) may include the steps of creating a structural analysis model based on the shape model, sequentially changing the boundary conditions for the occurrence of cracks 9 for all parts of the crack candidate surface 7fc, performing numerical analysis, and associating and storing information on the cracked parts generated based on the analysis results, the analysis results of the crack candidate surface 7fc, and the analysis results of the observation surface 7fo (step S2260); and obtaining a crack surface matrix Δ obtained by matrix-representing the stored analysis results of the crack candidate surface. crack_diff The observation plane matrix E that represents the analysis result of the observation plane 7fo in a matrix form measure The forward coefficient matrix D of the mapping and the rigidity matrix G that represents the relationship between the load and displacement of the analysis results of the crack candidate surface 7fc in a matrix form, and the steps of outputting the forward coefficient matrix D and the rigidity matrix G as the relationship information used in the inference (steps S2290 to S2320) make it possible to efficiently infer the crack 9 using a small amount of learning of crack data.

[0140] In addition, in the inference model generation step (step S2200), by performing numerical analysis on the disconnection between the multiple elements Efc obtained by dividing the crack candidate surface 7fc or changing the shape or conditions of the crack candidate surface 7fc to the same shape as that of the case where the crack 9 is generated, the crack 9 can also be efficiently inferred using a small amount of crack data.

[0141] If the shape model setting step (step S2100) is constructed to include: a step of measuring or determining the distribution of the generated stress in the target structure 7 based on structural analysis; a step of selecting an evaluation stress suitable for determining the occurrence location of the crack 9 based on the material constituting the target structure 7 and the distribution of the generated stress; and a step of taking the point where the generated stress becomes maximum as the occurrence location of the crack 9, and determining the crack candidate surface 7fc in a manner perpendicular to the direction of the maximum principal stress in the occurrence location and penetrating the surface opposite to the occurrence location in the target structure 7, then learning data can be prepared before the inspection, and the time spent on the inspection can be shortened.

[0142] In addition, according to the fault diagnosis method of the rotating motor of the present application, if the target structure 7 is a rotating motor component including a rotor constituting the rotating motor 70 and any component in a retaining ring thermally mounted on the end of the rotor, and is constructed to include: the steps performed in the above-mentioned crack inference method; a step of accepting component information including the force applied to the rotating motor component during the operation of the rotating motor 70 and the physical property values of the material constituting the rotating motor component (step S3300); using the information on the size and position of the crack 9 obtained in the crack state analysis step (step S3200) and the target structure information, a step of obtaining the obtained crack development life and calculating the remaining service period (steps S3310 to S3320); and a step of displaying the calculated service period (step S5100), then the remaining service period until the rotating motor 70 fails can be correctly diagnosed and notified.

[0143] Alternatively, according to the fault diagnosis method of a rotating motor of the present application, if the object structure 7 is a rotating motor component including a rotor constituting the rotating motor 70 and any component in a retaining ring hot-fitted to the end of the rotor, and is constructed to include: the steps performed in the above-mentioned crack inference method; a step of accepting information on the boundary conditions of the cracks generated in the rotating motor component (step S3340); and a step of determining whether the size and position of the crack 9 obtained in the crack state analysis step (step S3200) exceed the boundary conditions, and if so, notifying the occurrence of the fault (steps S3350 to S3370, step S5200), then the fault of the rotating motor 70 can be correctly diagnosed and notified.

Claims

1. A crack estimation device, characterized in that: have: a shape model setting unit that sets a shape model of a target structure to be inspected and to which an external force is applied, a crack candidate surface where cracks are expected to occur in a portion hidden from a surface of the shape model, and an observation surface on the surface of the shape model to be measured; an estimation model generating unit that generates an estimation matrix for estimating the state of the candidate fracture surface from the state of the observation surface, based on a matrix that relates the state of the candidate fracture surface and the state of the observation surface, obtained by numerically analyzing a structural analysis model created based on the shape model; and The crack state analysis unit uses the observation surface deformation vector representing the deformation of the observation surface obtained through the actual measurement value of the observation surface, the inference matrix, and the potential variables representing whether there is a crack on the crack candidate surface, and through probabilistic inference, simultaneously calculates the distribution of load and displacement in the crack candidate surface, thereby inferring the position and size of the crack.

2. The crack estimation device according to claim 1, wherein The crack state analysis part: By using the load and the sparsity of the displacement, which are in an inverse relationship, as the latent variables and the probability distribution of the displacement of the crack candidate surface obtained simultaneously with the load as the prior distribution, The probability distribution of the displacement of the observation surface calculated by the prior distribution and the inference matrix is used as the posterior distribution, and the size and position of the crack are calculated based on the posterior distribution at the time point when the posterior distribution converges by updating the latent variable.

3. The crack estimation device according to claim 1, wherein Bayesian inference is used in the probabilistic inference.

4. The crack estimation device according to any one of claims 1 to 3, characterized in that: As the observation surface deformation vector, any of the displacement change, strain change, and angle change of the observation surface is used.

5. The crack estimation device according to any one of claims 1 to 3, characterized in that: The external force is an inspection load applied during the inspection of the target structure. The crack estimation device comprises: checking a load setting unit to set the position and size of the load; and The inspection load indicator displays the position and magnitude of the inspection load that is set during the inspection.

6. A fault diagnosis device, characterized in that: have: The crack estimation device according to any one of claims 1 to 5, connected to a measuring instrument for measuring the observation surface of the target structure, and comprising a measurement data acquisition unit for acquiring the measurement value from the measuring instrument; as well as The terminal receives information on boundary conditions of cracks in the target structure, outputs the received boundary condition information to the crack estimation device, and displays the analysis results of the crack estimation device. The crack state analysis unit determines whether the obtained size and position of the crack exceed the limit condition, and if so, displays a warning indicating a failure on the terminal.

7. A fault diagnosis device, characterized in that: have: The crack estimation device according to any one of claims 1 to 5, connected to a measuring instrument for measuring the observation surface of the target structure, and comprising a measurement data acquisition unit for acquiring the measurement value from the measuring instrument; as well as The terminal receives component information including a force applied to the target structure during operation of a device having the target structure and a physical property value of a material constituting the target structure, outputs the received component information to the crack estimation device, and displays an analysis result of the crack estimation device. The crack state analyzing unit calculates the crack development life based on the calculated size and position of the crack and the component information, and displays information on the remaining service life on the terminal.

8. A crack estimation method, characterized in that: include: a shape model setting step of setting a shape model of a target structure to be inspected and to which an external force is applied, a crack candidate surface where cracks are expected to occur in a portion hidden from a surface of the shape model, and an observation surface on the surface of the shape model to be measured; an inference model generating step of generating, based on a matrix correlating the state of the candidate fracture surface with the state of the observation surface, a inference matrix for inferring the state of the candidate fracture surface from the state of the observation surface, based on the matrix obtained by numerically analyzing a structural analysis model created based on the shape model; a step of accepting actual measurement values of the observation surface; and The crack state analysis step applies the observation surface deformation vector representing the deformation of the observation surface obtained from the measurement value, the inference matrix, and the potential variables representing whether there is a crack on the crack candidate surface. Through probabilistic inference, the distribution of load and displacement in the crack candidate surface is simultaneously calculated, thereby inferring the position and size of the crack.

9. The crack estimation method according to claim 8, wherein: In the crack state analysis step, Calculating a forward coefficient matrix for mapping from a crack surface matrix that is included in the inference matrix and represents the analysis result of the crack candidate surface in a matrix to an observation surface matrix that is included in the inference matrix and represents the analysis result of the observation surface in a matrix, and an observation surface deformation vector that represents the deformation of the observation surface of the target structure, and calculating a probability distribution of the displacement of the crack candidate surface inferred via the latent variables based on the observation surface deformation vector and the forward coefficient matrix as a likelihood distribution, The probability distribution of the load and displacement in the crack candidate surface is estimated simultaneously through the latent variable representing the sparsity of the opposing load and displacement, and the calculated probability distribution of the displacement of the crack candidate surface is used as a prior distribution. The displacement distribution of the crack candidate surface is obtained by the probability inference based on the likelihood distribution and the prior distribution.

10. The crack estimation method according to claim 8, wherein: The inference model generation step comprises: The steps of creating the structural analysis model based on the shape model, performing the numerical analysis for all parts of the crack candidate surface while sequentially changing the boundary conditions for crack occurrence, and storing information on cracked parts generated based on the analysis results, the analysis results of the crack candidate surface, and the analysis results of the observation surface in association with each other; and A step of obtaining a forward coefficient matrix that maps a stored crack surface matrix representing the analysis results of the crack candidate surface to an observation surface matrix representing the analysis results of the observation surface, and a rigidity matrix that represents the relationship between the load and displacement of the analysis results of the crack candidate surface, and outputting the forward coefficient matrix and the rigidity matrix as relationship information used in inference.

11. The crack estimation method according to any one of claims 8 to 10, characterized in that: In the inference model generation step, In the numerical analysis, the connection between the plurality of elements obtained by dividing the crack candidate surface is disconnected or the displacement of the crack candidate surface is changed to the same shape or condition as when the crack occurs.

12. The crack estimation method according to any one of claims 8 to 10, characterized in that: The shape model setting step includes: a step of measuring or determining the distribution of stress generated in the target structure by structural analysis; a step of selecting an evaluation stress suitable for determining the occurrence location of the crack based on the material constituting the target structure and the distribution of the generated stress; and The step of determining the crack candidate surface so as to be perpendicular to the maximum principal stress direction in the crack occurrence site and to penetrate the surface of the target structure facing the crack occurrence site, taking the point where the generated stress is maximum as the crack occurrence site.

13. A method for diagnosing a fault of a rotating electrical machine, characterized in that: The target structure is a rotating electrical machine component including a rotor constituting the rotating electrical machine and a retaining ring shrink-fitted to an end portion of the rotor. The fault diagnosis method of the rotating electrical machine comprises: Each step performed by the crack estimation method according to any one of claims 8 to 12; a step of accepting component information including a force applied to the rotating electrical machine component during operation of the rotating electrical machine and a physical property value of a material constituting the rotating electrical machine component; a step of obtaining a crack growth life obtained in the crack state analysis step and calculating a remaining service life by using the information on the size and position of the crack obtained in the crack state analysis step and the component information; and Displays the calculated steps for the usage period.

14. A method for diagnosing a fault of a rotating electrical machine, characterized in that: The target structure is a rotating electrical machine component including a rotor constituting the rotating electrical machine and a retaining ring shrink-fitted to an end portion of the rotor. The fault diagnosis method of the rotating electrical machine comprises: Each step performed by the crack estimation method according to any one of claims 8 to 12; a step of receiving information on a limit condition of a crack occurring in the rotating electrical machine component; and a step of determining whether the size and position of the crack obtained in the crack state analysis step exceed the limit condition, and notifying the occurrence of a failure if the limit condition is exceeded.

Images