JP2020064024A - Rotary machine abnormality factor specifying device - Google Patents

Abstract

To provide a rotary machine abnormality factor specifying device capable of clarifying a vibration factor even in a case where an amount of accumulated data in the past is small or where there is no accumulated data in the past.SOLUTION: A rotary machinery abnormality factor specifying device (10) comprises: a past data storage unit (11) that stores past data at the time of abnormality in a rotary machine; an axial motion simulation unit (12) that simulates the axial motion of a rotor of the rotary machine with an abnormality by the transfer matrix method, to generate simulation data; a simulation data storage unit (13) that stores the simulation data generated by the axial motion simulation unit; a learning unit (14) that learns using the past data and the simulation data; and an abnormality factor specifying unit (15) that specifies an abnormality factor of the data at the time of abnormality, using the learning data of the learning unit.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a rotary machine abnormality factor identification device.

It is under study to identify the cause of abnormality from the frequency of vibration generated in a rotating machine and to promptly clarify the countermeasure.
There is a particular need for such techniques in large rotating machinery such as turbines in power plants.

  In Patent Document 1 which is an example of a conventional technique, a past similar abnormality can be easily searched and reproduced for a complicated phenomenon such as a shaft vibration abnormality, and an operation of an operator when a similar abnormality occurs A vibration diagnosis assisting apparatus for a rotary machine is disclosed which can sufficiently assist the decision in consideration of similar abnormal axis vibration, which is one of the operation judging processes.

  In Patent Document 2 which is an example of a conventional technique, when data obtained from a rotating machine is taken in and the cause of vibration is estimated from the characteristics of a vibration phenomenon occurring in the rotating machine, it is first constructed in a knowledge base of a computer. The cause of vibration is estimated based on the vibration causal matrix that correlates the characteristics of multiple phenomena and multiple causes, and then the phenomenon of the estimated vibration cause is analyzed using the vibration analysis model built into the memory of the computer. However, thereafter, a technique for evaluating a diagnostic result for a vibration phenomenon based on the analysis result and the vibration cause estimation result is disclosed.

JP-A-2-31117 Japanese Patent No. 378041

However, the above-mentioned conventional technique reveals the vibration factor based on the past data stored in the database.
Therefore, there is a problem that it is difficult to clarify the vibration factor when the amount of accumulated data in the past is small or when there is no accumulated data in the past.

  The present invention has been made in view of the above, and even if the past data storage amount is small or there is no past data storage, the rotary machine abnormality factor that can clarify the vibration factor The purpose is to obtain a specific device.

  The present invention, which solves the above problems and achieves the object, provides a past data storage unit that stores past data at the time of an abnormality in a rotating machine, and a transfer matrix method for the rotating machine having an abnormality corresponding to a vibration causal matrix. An axial motion simulation unit that simulates axial motion of a rotating body to generate simulation data, a simulation data storage unit that stores simulation data generated by the axial motion simulation unit, and the past data storage unit that stores the simulation data. A learning unit that learns using past data and the simulation data stored in the simulation data storage unit, and an abnormal factor specifying unit that specifies an abnormal factor of abnormal time data using the learning data of the learning unit. It is a rotating machine abnormality factor identification device provided.

  In the rotary machine abnormality factor identification device, the learning unit separates the vibration spectrum, which is the simulation data, into a forward component and a backward component with respect to the rotation direction, and converts the components into a full-spectrum frequency domain characteristic. Learning data is generated by the abnormal factor specifying unit, and the vibration spectrum, which is the abnormal time data acquired from the displacement sensor and the acceleration sensor installed in the two axis directions orthogonal to the rotation axis, with respect to the rotation direction. It is preferable that the forward component and the backward component are separated and converted into the characteristics of the frequency region of the full spectrum to identify the cause of abnormality of the rotation axis.

  According to the present invention, there is an effect that the vibration factor can be clarified even when the amount of accumulated data in the past is small or there is no accumulated data in the past.

It is a block diagram which shows the structure of the rotary machine abnormality factor identification device concerning embodiment. It is a figure which shows an example of the real machine of the rotary machine which is the specific target of the abnormality factor by the rotary machine abnormality factor identification device shown in FIG. FIG. 3 is a model diagram in which the actual machine shown in FIG. 2 is implemented in Modelica language. It is a figure which shows the time series simulation result when a 3rd axis model is a symmetrical axis. It is a figure which shows the time series simulation result when a 3rd axis model is an asymmetric axis. It is a figure which shows the time series simulation result when a 3rd axis model is a crack axis. It is a figure which shows the amplitude with respect to a frequency when a 3rd axial model is a symmetrical axis, an asymmetrical axis, and a case where it is a crack axis.

Embodiments for carrying out the present invention will be described below with reference to the accompanying drawings.
However, the present invention should not be limitedly interpreted by the description of the following embodiments.

<Embodiment 1>
FIG. 1 is a block diagram showing the configuration of a rotary machine abnormality factor identification device 10 according to the present embodiment.
The rotary machine abnormality factor identification device 10 shown in FIG. 1 includes a past data storage unit 11, an axial motion simulation unit 12, a simulation data storage unit 13, a learning unit 14, and an abnormality factor identification unit 15.

The past data storage unit 11 stores past data at the time of abnormality in the rotating machine.
This past data is actual data, that is, data obtained when a past abnormality occurred.
This data is, for example, a time-series frequency spectrum acquired by a vibration sensor installed in a rotating machine.
The past data storage unit 11 can be realized by a recording medium such as a semiconductor memory and a magnetic disk.

The axial motion simulation unit 12 simulates the axial motion of the abnormal rotating body by the transfer matrix method to generate simulation data.
The abnormality of the axial motion of the rotating body is an abnormality corresponding to the vibration causal matrix.
The axial motion simulation unit 12 can be realized by a processor such as an MPU (Micro Processing Unit) and a CPU (Central Processing Unit).

The vibration causal matrix is a table in which the predominant frequency generated by vibration is classified according to the cause of vibration generated in the rotating machine.
An example of the vibration causal matrix is shown in Table 1 below.

Here, “***” in Table 1 above is the main frequency, “**” is the frequency that can occur next to the main frequency, and “*” is the frequency that can be generated. .
The parenthesized “*” is a frequency that is unlikely to occur.
In Table 1, for example, when a rotor is abnormal, a vibration having a frequency equal to the rotation speed of the rotor is detected, and when a vibration having a frequency twice the rotation speed of the rotor is not detected, unbalance and shaft bending are detected. It is shown that is estimated as a vibration factor.
Then, at this time, if vibration of a low frequency lower than the rotation speed of the rotor is further detected, it is possible to specify that the vibration factor is shaft bending.

The simulation data storage unit 13 stores the simulation data generated by the axial motion simulation unit 12.
The simulation data storage unit 13 can be realized by a recording medium such as a semiconductor memory and a magnetic disk.

The learning unit 14 learns using one or both of the past data accumulated in the past data accumulation unit 11 and the simulation data stored in the simulation data storage unit 13.
Here, the learning unit 14 uses, as learning data, data converted into characteristics in the frequency domain by performing Fourier transform or wavelet transform using time-series data as a feature amount.
Examples of the learning method of the learning unit 14 include a statistical method and a machine learning method for learning correlation such as Gaussian mixture distribution, multi-class SVM (Support Vector Machine), and neural network.
The learning unit 14 learns the correlation between the time-series frequency spectrum acquired by the vibration sensor installed in the rotating machine and the vibration factor, and creates learning data.
The learning unit 14 can be realized by a processor such as an MPU and a CPU.
In addition, when the learning unit 14 performs learning using only the simulation data, the rotating machine abnormality factor identification device 10 may not include the past data storage unit 11, or the learning unit 14 may use the past data. When learning is performed using both the simulation data and the simulation data, it is possible to improve the accuracy of identifying the abnormality factor.

The abnormality factor identifying unit 15 identifies the abnormality factor in the abnormal data by using the learning data of the learning unit 14 and outputs the abnormality factor identifying result.
The abnormal time data is, for example, a time-series frequency spectrum acquired by the vibration sensor installed in the rotating machine during the abnormal time.
The abnormality factor identification unit 15 can be realized by a processor such as an MPU and a CPU.

The abnormality factor identification result is notified to a user or an administrator by being output to, for example, an output unit mounted on or connected to the rotating machine abnormality factor identification device 10.
Here, a display can be illustrated as the output unit.

<Explanation of the axis motion simulation unit 12>
Next, an example of the simulation performed by the axial motion simulation unit 12 will be described.
The axial motion simulation unit 12 performs simulation by the transfer matrix method to generate time-series simulation data of vibration due to abnormality.
The simulation data generated by the axial motion simulation unit 12 is stored in the simulation data storage unit 13.
This simulation data can be used to construct the vibration causal matrix described above.

In the transfer matrix method, the rotating shaft of the rotating machine is decomposed into a rotating body that is a rigid body, a shaft and a bearing rigidity that are elastic bodies, and the like.
Here, examples of the rotating body may include a turbine, a blade, a gear, and a bearing portion.
Further, by adding spring rigidity and a damper to the bearing portion, it is possible to simulate the dynamic characteristics of the shaft vibration of the rotating machine.
The transfer matrix method can be used to introduce imbalance or non-linearity for each classified component.
Here, examples of the non-linearity include asymmetry and cracks.
In order to model a rotating machine at the time of vibration, a method almost equivalent to the transfer matrix method and a non-causal Modelica language may be used.

FIG. 2 is a diagram showing an example of an actual machine of the rotary machine 20 that is a target for identifying an abnormality factor by the rotary machine abnormality factor identification device 10 shown in FIG.
The rotary machine 20 shown in FIG. 2 includes a power unit 21, a first shaft 22, a shaft coupling 23, a second shaft 24, a first bearing unit 25, a third shaft 26, and a rotor. 27, a fourth shaft 28, and a second bearing portion 29.

The power unit 21 outputs rotation.
The first shaft 22 is an output shaft connected to the power unit 21.
The second shaft 24 is connected to the first shaft 22 via a shaft coupling 23.
The first bearing portion 25 is fixed and supports the second shaft 24 and the third shaft 26.
The third shaft 26 is connected to the second shaft 24 via the first bearing portion 25.
The second bearing portion 29 is fixed and supports the fourth shaft 28.
The rotor 27 is arranged between the third shaft 26 and the fourth shaft 28.

FIG. 3 is a model diagram in which the actual machine shown in FIG. 2 is implemented in the Modelica language.
The rotary machine model 20M shown in FIG. 3 includes a power unit model 21M, a first shaft model 22M, a shaft coupling model 23M, a second shaft model 24M, a bearing unit model 25M, and a third shaft model 26M. A rotor model 27M, a fourth shaft model 28M, and a bearing portion model 29M.
In these configurations shown in FIG. 3, “M” is added to the reference numerals of the configurations shown in FIG. 2 to which they correspond.

Here, the axes of the rigid rotating body and the elastic body, which are the basic elements of the rotating body, will be described.
The force F _{x in} the horizontal direction of the asymmetric rotating body which is a rigid body with static imbalance is expressed by the following formula (1), and the force F _{y in the} vertical direction is expressed by the following formula (2).

The moment M _{x in} the horizontal direction of the asymmetric rotating body which is a rigid body with dynamic imbalance is represented by the following equation (3), and the moment M _{y in the} vertical direction is represented by the following equation (4).

However, u _x is the deflection in the horizontal direction, u _y is the deflection of the vertical, i _x is the deflection angle in the horizontal direction, i _y is the deflection angle in the vertical direction.
Further, θ is the rotation angle, m is the mass of the rotating body,
e is the eccentricity represented by the length, e ₀ is the phase of the eccentricity,
τ is the magnitude of the dynamic imbalance expressed by an angle, τ ₀ is the phase of the dynamic imbalance,
I _d is the diametrical moment of inertia and I _p is the polar moment of inertia.

  Further, the asymmetric elastic axis model having different rigidity is a linear beam model, and is expressed by the following equations (5) and (6).

  The relationship between the coordinate system O-ξη rotating at the rotation speed ω and the second moment of area in the stationary coordinate system is expressed by the following equation (7).

  Here, the axis with cracks is represented by the following formula (8).

Where m is the mass of the shaft, L is the length of the shaft, E is the Young's modulus, I _x and I _y are the second moments of area of the stationary coordinate system, and I _ξ and I _η are the rotations. It is the moment of inertia of area of the coordinate system, and δ is the strength of the crack expressed by the moment of inertia of area.

In a model of a rotating machine using rolling bearings, bearing rigidity can be simulated by using catalog values.
The bearing stiffness can also be calculated by Hertz's contact theory.
A model of a bearing scratch that causes a problem as a bearing abnormality can be simulated by applying an impact force of a frequency due to the scratch.
This frequency is determined by the pitch circle diameter of the bearing, the diameter of the rolling elements and the number of rolling elements.

Abnormalities of gears can be calculated by obtaining the dynamic characteristics of the shaft from the number of gears and meshing rigidity.
A model of an unbalanced rotating body is used for the gear itself.

Then, a model in a normal state is created from a model created by combining the models described above.
If impulse test data is available, the parameters of bearing rigidity or damping coefficient may be adjusted to match the data.
Abnormal data is generated by changing the parameters of axis bending by causing imbalance in the model in the normal state.

  For other abnormalities, data is generated by adding a physical phenomenon of abnormalities that may occur depending on the type of rotating machine and the bearing type as a component and performing simulation.

Next, the simulation results when the axis in the rotating machine is a symmetric axis, an asymmetric axis, and a crack axis are compared.
FIG. 4A is a diagram showing a time-series simulation result when the third axis model 26M is a symmetrical axis.
FIG. 4B is a diagram showing a time series simulation result when the third axis model 26M is an asymmetric axis.
FIG. 4C is a diagram showing a time-series simulation result when the third axis model 26M is a crack axis.
4A, 4B, and 4C, the horizontal axis represents time and the vertical axis represents amplitude.

FIG. 5 is a diagram showing amplitudes with respect to frequencies when the third axis model 26M is a symmetrical axis, an asymmetrical axis, and a crack axis.
As shown in FIG. 5, since the crack axis extends the asymmetric axis, the crack axis has the same component as the asymmetric axis.
Here, the high frequency component shows a nonlinear term.

  Here, although it is difficult to separate the spectrum of the asymmetric axis and the spectrum of the crack axis, there is a difference in the amplitude of the high frequency component as shown in FIG. It is possible to separate from the axis.

  As described above in the present embodiment, even if the amount of accumulated data in the past is small or there is no accumulated data in the past, the vibration factor can be clarified by using the simulation data as the learning data. .

<Embodiment 2>
In the present embodiment, a mode will be described in which the displacement sensor and the acceleration sensor are installed in two axial directions orthogonal to the target axis for identifying the vibration factor.

In the present embodiment, the learning unit 14 separates the vibration spectrum, which is the simulation data, into a forward component and a backward component with respect to the rotation direction, and converts the components into a full-spectrum frequency domain characteristic for learning.
In addition, in the present embodiment, the abnormality factor identifying unit 15 separates the vibration spectrum, which is the abnormal time data, into a forward component and a backward component with respect to the rotation direction, and converts the components into a full-spectrum frequency domain characteristic. To identify the cause of the abnormality.

  Thus, even if it is difficult to identify the vibration factor according to the first embodiment, the present embodiment can improve the accuracy of identifying the abnormal factor.

  It should be noted that the present invention is not limited to the above-described embodiment, and includes various modified examples in which constituent elements are added, deleted, or converted to the above-described configuration.

10 Rotating Machine Abnormality Factor Identifying Device 11 Past Data Accumulating Section 12 Axis Motion Simulation Section 13 Simulation Data Storage Section 14 Learning Section 15 Abnormality Factor Identifying Section 20 Rotating Machine 20M Rotating Machine Model 21 Power Section 21M Power Section Model 22 First Axis 22M First shaft model 23 Shaft joint 23M Shaft joint model 24 Second shaft 24M Second shaft model 25 First bearing portion 25M First bearing portion model 26 Third shaft 26M Third shaft model 27 Rotor 27M Rotor model 28 Fourth shaft 28M Fourth shaft model 29 Second bearing portion 29M Second bearing portion model

Claims (2)

A past data storage unit that stores past data when an abnormality occurs in a rotating machine,
By the transfer matrix method, an axial motion simulation unit for simulating the axial motion of the rotating body of the rotating machine having an abnormality corresponding to the vibration causal matrix to generate simulation data,
A simulation data storage unit for storing the simulation data generated by the axial motion simulation unit;
A learning unit that learns using the past data accumulated in the past data accumulation unit and the simulation data stored in the simulation data storage unit,
An abnormality factor specifying device for a rotating machine, comprising: an abnormality factor specifying part for specifying an abnormality factor of abnormal data by using the learning data of the learning part. The learning unit separates the vibration spectrum, which is the simulation data, into a forward component and a backward component with respect to the rotation direction, and converts the full spectrum into a frequency domain characteristic to generate learning data.
The abnormality factor identification unit uses a vibration spectrum, which is the abnormal time data acquired from a displacement sensor and an acceleration sensor installed in two axial directions orthogonal to the rotation axis, as a forward component and a backward component with respect to the rotation direction. 2. The rotating machine abnormality factor identifying device according to claim 1, wherein the rotating machine abnormality factor identifying device identifies the abnormality factor of the rotating shaft by converting into a full spectrum frequency domain characteristic.