JP6879873B2 - Failure probability evaluation system - Google Patents

Description

The present invention relates to a failure probability evaluation system for a plurality of machines of the same type included in a machine system.

In the past, some methods for evaluating the remaining life of mechanical system fatigue failures have been proposed. As a typical example, there is one that uses the Miner's Rule of Cumulative Damage (Non-Patent Document 1). Specifically, for example, there is a method described in Patent Document 1. On the other hand, as another method for evaluating the soundness of mechanical components, a method called abnormality diagnosis or sign detection is known (Patent Document 2).

JP 2015-229939 WO2016 / 117021

M.A. Miner: Cumulative Damage in Fatigue, J. Appl. Mech., 12 (3), ppA159-A164

In a mechanical system, mechanical elements and mechanical structures (hereinafter collectively referred to as mechanical components) exposed to a dynamic load are designed for fatigue life so as to satisfy the required life. However, in actual use, the dynamic loads acting on them vary depending on the usage environment and usage conditions. At the same time, the fatigue life of each individual of these mechanical components also potentially varies. Therefore, at the fatigue life design stage, the safety side is designed so as not to lead to fatigue fracture or fatigue failure, assuming these variations. However, in recent years, there has been a demand for the operation and maintenance of mechanical systems that can safely use up the life of mechanical components from the viewpoints of cases where mechanical systems are exposed to unexpected environments and resource saving and energy saving. .. In view of this situation, it is extremely important to more accurately grasp the remaining life of each component of the mechanical system that is actually in operation. Further, as described above, the life of a mechanical component has a variation based on a certain probability distribution. Therefore, the remaining life is defined stochastically, and evaluating the remaining life is equivalent to evaluating the failure probability when an arbitrary time (remaining life) elapses from a certain point in time. .. If the soundness of the target can be evaluated as the failure probability, the expected value of the loss amount can be theoretically calculated by multiplying the loss amount generated when the target fails and the failure probability. As a result, operation and maintenance decisions can be easily made from an economic point of view.

Against this background, several methods for evaluating the remaining life of mechanical system fatigue failures have been proposed. A typical example is to use the Miner's Rule of Cumulative Damage (Non-Patent Document 1). For example, in the case of a mechanical structure, a strain gauge or other sensor that measures strain or stress is attached to the target site, and the time history data is acquired. A waveform counting method such as the rainflow method is applied to the obtained time history data to obtain the frequency distribution of strain and stress waveforms. For this occurrence frequency distribution, refer to the fatigue diagram of the materials that make up the target site, and determine the degree of fatigue damage by the linear cumulative damage rule. Here, the degree of fatigue damage is a physical quantity that represents the consumption rate of fatigue life with respect to the average fatigue life of the target (Patent Document 1). Furthermore, the degree of fatigue damage can be obtained, and the probability of failure at any degree of fatigue damage can be obtained by referring to the life variation defined in the fatigue diagram. In addition, if it is possible to predict the degree of fatigue damage after an arbitrary time has passed using some method, it is also possible to obtain the failure probability at that time, and the remaining life is evaluated stochastically. It becomes possible. However, the fatigue life of ordinary mechanical components may vary from 1/10 to 10 times depending on the target. Therefore, it may be difficult to provide the remaining life and failure probability evaluation with the accuracy required for operation and maintenance only by this method for mechanical components with such a relatively large variation in fatigue life.

On the other hand, as another method for evaluating the soundness of mechanical components, there is a method called abnormality diagnosis or sign detection (Patent Document 2). Quantitatively define or learn the operating state of a machine component in a healthy state in advance. Here, the operating state is a physical quantity that is expected to change according to soundness, such as the vibration acceleration, frequency, and temperature of the object, or a state quantity expressed by a combination thereof. There is an application example in which the state quantity is constantly evaluated during operation, an alert is issued according to the degree of deviation from the healthy state, or the mechanical system is automatically stopped. Since this method directly monitors the soundness, it can be expected to detect soundness changes with relatively high sensitivity. However, in order to quantitatively calculate the remaining life and failure probability by applying this method, the relationship between the change in state quantity and the remaining life or failure probability when a certain state quantity change occurs is acquired in advance. Although it is necessary to keep it, it is not realistic considering the time and cost because it requires pre-testing at the mechanical component level or mechanical system level instead of the material level.

Further, in a mechanical system whose operating state changes from moment to moment, such as a wind power generation system, a method of calculating the remaining life in units of time from the beginning is not practical. For example, in a wind power generation system, the load acting on each mechanical component per unit time differs depending on the wind conditions and control conditions. Therefore, it is desirable that the remaining life and the failure probability change according to these assumed conditions, but the remaining life evaluation on an hourly basis cannot meet such requirements.

As mentioned above, as a method for evaluating the failure probability or remaining life due to fatigue of a mechanical system, a method based on the linear cumulative damage rule is known, but its accuracy is often not sufficient for practical use. In addition, the soundness evaluation method based on abnormality diagnosis can detect soundness changes with relatively high accuracy, but in order to quantitatively evaluate the failure probability, a preliminary test that requires time and cost is required. .. In addition, the method of evaluating the remaining life by time often causes practical inconvenience, especially in mechanical systems whose operating state is not constant. Therefore, the emergence of a system that provides practical and highly accurate failure probability evaluation or remaining life evaluation for mechanical systems operated in an environment with strong uncertainties such as wind power generation systems has been awaited.

In order to solve the above problems, for example, the configuration described in the claims is adopted. The present application includes a plurality of means for solving the above problems. For example, it is a system for evaluating the failure probability of a plurality of machine elements included in a mechanical system, and the mechanical elements are fatigued or damaged or aged. Based on the physical quantity 1 that changes due to change, the means for evaluating the state quantity that represents the soundness of the machine element, and the physical quantity 2 that changes depending on the load or load received by the machine element or the operation data of the machine element. A means for evaluating the cumulative fatigue damage degree of the machine element, a storage unit for storing the state amount and the fatigue damage degree, and the state amount of the machine element in which a failure has occurred among the plurality of machine elements. A failure probability evaluation system characterized by having a failure probability evaluation unit that calculates a failure probability of a machine element in which a failure has not occurred among the plurality of machine elements based on the degree of fatigue damage.

Similar to the known technique, the present invention has a function of evaluating the degree of fatigue damage of a target based on measurement and simulation by a sensor, and at the same time, has a function of evaluating a soundness state based on sensor measurement. Furthermore, by evaluating the remaining life of a mechanical system not on the time axis but on the fatigue damage degree axis, it is possible to provide an effective remaining life evaluation even for a mechanical system whose operating state is not constant. In addition, statistical modeling that assumes a probability distribution for the variation in remaining life under the condition that a certain health state is observed makes it possible to provide a failure probability in the future in which an arbitrary degree of damage is further accumulated.

The schematic diagram explaining the operation when this invention is applied to a bearing group. The schematic diagram explaining the operation after a failure occurs in one bearing by applying this invention to a bearing group. A display example by a display unit in one of the examples of the present invention. The schematic diagram explaining the failure probability calculation method in this invention. A display example by a display unit in one of the examples of the present invention.

Hereinafter, examples will be described with reference to the drawings.

FIG. 1 is a diagram schematically explaining the operation of the failure probability evaluation system 100 in the present invention, taking the rotary bearing 1 as a mechanical component (machine element) as an example. The present invention is directed to a plurality of, preferably the same type, set of mechanical components. Each mechanical component to be observed has a sensor that measures the physical quantity that changes to reflect the soundness, and the physical quantity that changes depending on the load and load received by the mechanical component for the purpose of evaluating the degree of fatigue damage. Install each sensor. In this embodiment, the former corresponds to the acceleration sensor 2, the latter corresponds to the load cell 3 and the tachometer 4, respectively. That is, the acceleration sensor 2 measures the rotational vibration acceleration generated from the damaged bearing, and the load cell 3 and the tachometer 4 acquire the amplitude and the number of repetitions of the load acting on the bearing. Since it is necessary to directly measure the physical quantity for evaluating the soundness, at least one sensor for that purpose is indispensable in this embodiment, but the load applied to the target does not necessarily have to be directly measured. For example, in the case of rotary bearings used in wind power generators and automobiles, the load on the bearings is based on the operation data of mechanical components such as the history of wind conditions and power generation in the former case and the history of speed and engine speed in the latter case. It is also possible to estimate the history of the load to be applied. Further, in this embodiment, the soundness evaluation by the acceleration sensor 2 is premised, but the type of the sensor is not limited to the acceleration sensor. For example, an AE sensor or a temperature sensor is used, or a plurality of types of sensors are combined. You may use it. Further, the sensor that measures the load and the physical quantity that changes depending on the load may be a strain sensor. Various sensors may be newly installed as a part of the failure probability evaluation system 100, or the failure probability evaluation system 100 has a signal receiver (not shown) so as to receive the detection value of the sensor installed in the mechanical component. It may be.

The vibration acceleration data obtained by the acceleration sensor 2 is transmitted to the state evaluation unit 6 via the A / D conversion unit 5. Here, the vibration acceleration data is converted into a state quantity for evaluating the soundness of the bearing 1. There are several possible methods for determining the state quantity that indicates soundness. For example, in the case of bearings, fine cracks and flaking may occur in the inner ring or outer ring due to repeated loading. Vibration is generated each time the rolling element inside passes through these damaged positions. Therefore, it is effective to use the acceleration effective value of the frequency band obtained by multiplying the number of rolling elements by the frequency corresponding to the value obtained by multiplying the number of rotations by the number of rolling elements. That is, in this embodiment, the effective acceleration value of the specific frequency band is used for the subsequent evaluation as a state quantity (hereinafter, state quantity) indicating soundness. As described above, when using a plurality of types of sensors, it is desirable to convert the obtained multiple physical quantity data into one state quantity and use it by applying, for example, clustering analysis. The obtained state quantity is transmitted to the state quantity change temporary storage unit 8 and temporarily stored as time-series data. Here, the retention period of time-series data may be set arbitrarily, but setting a period equivalent to the lead time required for maintenance or replacement work of the target suppresses the required storage area without impairing the function. It is the most effective from the viewpoint of doing.

On the other hand, the load amplitude and the number of repetitions obtained by using the load cell 3 and the tachometer 4 are transmitted to the fatigue damage evaluation unit 7. The fatigue damage evaluation unit 7 calculates the fatigue damage degree based on the linear cumulative damage rule (Non-Patent Document 1). In this embodiment, the load acquired from the load cell 3 is used as the load amplitude, and the number of rotations acquired from the tachometer 4 is used as the number of repetitions to calculate the degree of fatigue damage. Alternatively, a strain sensor such as a strain gauge may be used to directly measure the stress time-series change of the evaluation site. The calculated fatigue damage degree is transmitted to the damage degree change temporary storage unit 9 in the same manner as the state quantity, and is temporarily stored here. It is most effective to set the storage period here to be the same as the lead time for maintenance and replacement work, similar to the storage period of the state quantity change temporary storage unit 8.

The state evaluation unit 6 and the fatigue damage evaluation unit 7 shown above are shown as independent components in FIG. 1, but the present invention does not particularly limit these mounting modes. For example, even if each of them is composed of software in a single computer system, there is no problem in realizing the function. In addition, although the state quantity change temporary storage unit and the damage degree change temporary storage unit are also shown as independent components, they may be configured on the same storage device. Further, as shown in FIG. 1, the above components are prepared for all the target mechanical components (that is, all the first to nth bearings), but the state evaluation unit 6, the fatigue damage degree evaluation unit 7, and the state. The amount change temporary storage unit 8 and the damage degree change temporary storage unit 9 may be configured on common hardware.

FIG. 2 shows the failure probability of other bearings (individual number 1 in this case) that have not failed since the time when the bearing of individual number n failed during the operation of the system with the configuration of FIG. It is a figure explaining the operation of the system schematically. First, for an individual in which a failure has occurred, data transmission from each sensor is stopped, but when the failure occurs, the state quantity change stored in the state quantity change temporary storage unit 9 and the damage degree change temporary storage unit 8 is triggered. The time-series data of the damage degree change is transmitted to the failure history evaluation unit 12. The details will be described later, but in the failure history evaluation unit 12, when an arbitrary state quantity is observed, a statistical model that defines the probability that a failure will occur (failure probability) from that point to the point when an arbitrary degree of damage is further accumulated. 20 (relational expression) is constructed. This statistical model is transmitted to the failure probability evaluation unit 11, and the failure probability after the accumulation of arbitrary damage degrees is calculated together with the current state quantity of the unfailed individual. At this time, for the same individual, since the damage degree change so far is stored in the damage degree change temporary storage unit 9, the damage degree change prediction unit 10 predicts the subsequent damage degree change from the tendency of the damage degree change ( That is, the relationship between the change in damage degree and the passage of time is defined), and the failure probability evaluation unit 11 may adopt a method of evaluating the relationship between the change in failure probability and the passage of time.

The failure probability calculated by the failure probability evaluation unit 11 is a function defined as a relationship with the subsequent cumulative damage degree or time, and is not fixed to one value. Therefore, as shown in FIG. 3, the display unit 13 displays in a graph format as the cumulative status of the damage degree after the present time or the change in the failure probability with the passage of time, so that the user can operate and maintain it in the subsequent operation and maintenance. You will be able to make decisions easily. For example, in a mechanical system that can switch the operation mode, if the load on the target changes depending on the operation mode, as shown in Fig. 3, a failure occurs when the subsequent operation is performed in each operation mode. By displaying the probability change, it is possible to efficiently support the decision of the operation policy as well as the decision of the maintenance time.

If a high failure probability is calculated in the relatively near future as a result of the failure probability evaluation, not only the result should be displayed, but also a function to send a stop command or a degenerate operation command to the target control device should be provided. Is desirable. Waiting for the user's judgment about the target for which the state quantity change should be detected unless it is the latest of the failure may lead to the failure at an early stage. In that case, it is easy to prevent the occurrence of failure by providing a function to automatically stop or degenerate the system according to the evaluation result.

The method of constructing a statistical model that defines the failure probability after the above-mentioned arbitrary damage degree is accumulated will be described with reference to the schematic diagram of FIG. First, as step S1, for the bearing with the individual number n in which the failure occurred, the state quantity time series data 19 (S = f (t)) and the damage degree time series data 18 (D = f (D = f (D = f)) up to just before the failure occurred. t)) is saved.

Next, as step S2, these two time-series data are related as the relationship X (ΔD = f (S _t )) between the damage degree increment (ΔD) leading to the failure and the state quantity. In other words, it is equivalent to expressing the increment of damage degree until failure as a function with the observed state quantity as a variable. Specifically, in the state quantity time series data 19 (S = f (t)) and the damage degree time series data 18 (D = f (t)), the status quantity recorded at the same time t1 and the failure from t1. Using the damage degree difference until the occurrence occurs as a data set, data showing the relationship between the state amount and the damage degree increment in the process until the failure occurs is generated for the individual number n in which the failure occurred. Next, statistical modeling is performed on this data assuming variations according to a certain probability distribution, and statistical model 20 is acquired. As the method of statistical modeling, the least squares method that assumes a normal distribution may be used in the simplest way, but since the damage increment to which the probability distribution should be applied is defined as a non-negative value, a normal distribution is assumed. That is not strictly appropriate. In the study of the inventors on bearings, the generalized linear model (GLM), which adopted the gamma distribution defined as the non-negative distribution as the probability density function (PDF), expresses the data distribution relatively well. I know that. At this time, it is desirable to use the reciprocal function for the GLM link function.

Since the relationship between the damage increment ΔD until the failure and the observed state quantity S in the individual number n where the failure occurred was defined as the statistical model 20, the failure probability F and the failure probability F are then set as step S3. Define the relationship of damage increment ΔD until failure. It is assumed that _a certain state quantity S1 is observed in any individual that has not yet failed, and that the arbitrary degree of damage ΔD a increases thereafter. Here, the PDF of the damage degree increment until the failure in the state quantity S1 is expressed as P = f (S1). At this time, the failure probability F at the time when the damage degree is increased by the _{damage degree increment ΔD a is expressed by Equation 1.}

This is equivalent to calculating the cumulative probability using the Cumulative Distribution Function (CDF) for PDF obtained by statistical modeling. That is, the PDF to be referred to is determined by the arbitrary state quantity S observed in the individual in which the failure has not yet occurred, and then the damage degree increment ΔD _a, which is assumed to increase thereafter, is substituted into the corresponding CDF. By doing so, it is possible to calculate the failure probability F in an individual in which a failure has not yet occurred. Therefore, when a certain state quantity S is observed, the change 21 of the failure probability F with the subsequent increase in the degree of damage increment is nothing but the CDF itself corresponding to the PDF obtained from the statistical model.

By the above procedure, using the statistical model 20 which is the relationship between the damage increment ΔD until the failure and the observed state quantity S at the individual number n where the failure occurred, the machine was placed in a similar environment, especially in the same type machine. Therefore, the failure probability F can be calculated for each other individual based on the state quantity S observed in the other individuals for which the failure has not yet occurred.

Furthermore, by expressing the failure probability F based on the time based on the relationship of the damage degree ΔD _a per hour when operating under different load conditions, the fluctuation of the failure probability F when operating under different load conditions is predicted in the future. be able to. The display of the failure probability prediction 16 in FIG. 3 is based on the current state quantity S of the individual number 1 selected in the menu 151, and the failure probability F based on the operation time when operating in the high output mode, the normal mode, and the degenerate mode, respectively. The fluctuation is displayed. The overall situation summary 17 of FIG. 3 displays the failure probability F after 10 days in the operation mode selected for each of the individual numbers 1 to 10. This makes it easier for users to make subsequent operational and maintenance decisions.

In the first embodiment, as shown in FIG. 3, a method of displaying the change in the failure probability from the present to the future was adopted. This method can be said to be an effective method from the viewpoint of supporting operation and maintenance decision-making. However, since the state quantity calculated by the state evaluation unit 6 in FIG. 1 is calculated based on, for example, the frequency measured by the bearing, short-period fluctuations are observed depending on the operating conditions, apart from large trend changes. There is a possibility that If the time variation of the state quantity that is the premise for calculating the failure probability F of each individual is relatively large, the entire graph in the failure probability future prediction display 16 in FIG. 3 may be updated excessively frequently. is there. In that case, as shown in the failure probability prediction history display unit 22 in FIG. 5, the future time point for evaluating the failure probability is fixed in advance by the time or the damage degree increment value, and the failure probability evaluation result so far. It may be in a format that displays the transition of. In FIG. 5, for example, the prediction time to be displayed on the failure probability prediction history display unit 22 can be selected from the menu 152. By adopting such a display format, the history of the failure probability evaluation results so far can be easily confirmed, so that the user can easily estimate the subsequent tendency.

In the first and second embodiments, the failure probability of the unfailed target was evaluated by statistically processing the data up to the failure triggered by the failure of the mechanical component of the same type. This method enables highly accurate evaluation because the analysis is based on the state quantity, but on the other hand, the failure probability cannot be defined until the actual failure data is obtained. Therefore, as described in the background technology section, the cumulative fatigue damage degree obtained by the fatigue damage degree evaluation unit 7 in FIG. 1 until any failure occurs in the target mechanical component group. The failure probability may be defined using. During this period, the prediction accuracy cannot be expected to be as high as the prediction based on other individual data of the same group that actually failed, but even if the failure data cannot be acquired, the failure probability is evaluated without significantly changing the system configuration. It becomes possible to do.

The failure history evaluation unit 12 can also be provided separately from the failure evaluation system. In that case, the failure history evaluation unit 12 creates a statistical model for evaluation based on the individual failure data when a failure occurs, saves it in the failure evaluation system, and has a plurality of failure probability evaluation units 11 based on the statistical model. Perform failure assessment of mechanical systems including mechanical components.

1. Bearing
2. Accelerometer
3. Load cell
4. Tachometer
5. A / D converter
6. Condition evaluation department
7. Fatigue damage evaluation department
8. State quantity change temporary storage unit
9. Damage degree change temporary storage unit
10. Damage degree change prediction unit
11. Failure Probability Evaluation Department
12. Failure history evaluation department
13. Display
14. Control unit
15. Display contents on the display
16. Failure probability future forecast display
17. Summary of overall failure probability
18. Damage degree time series data
19. State quantity time series data
20. Statistical model
21. Failure probability prediction
22. Failure probability prediction history display

Claims (11)

It is a system that evaluates the failure probability of multiple machine elements included in the same type of machine component group .
Physical quantity that varies over the year changes, before Symbol physical quantity representing the change that by the load or loads the machine element is subjected, young properly evaluates the cumulative fatigue damage degree of the machine element based on operating data of the machine element Means and
A means for evaluating the state quantity representing the soundness of the machine element by associating the cumulative fatigue damage degree with the failure probability and the remaining life.
A storage unit that stores the state quantity and the cumulative fatigue damage degree,
Among the plurality of machine elements, the state quantity representing the soundness of the plurality of machine elements has not decreased based on the state quantity and the cumulative fatigue damage degree of the machine element having a reduced state quantity indicating the soundness. A failure probability evaluation unit that calculates the failure probability of machine elements, and
Among the plurality of machine elements, a failure history evaluation unit that statistically correlates the relationship between the state quantity of the machine element in which the failure has occurred and the increment of the cumulative fatigue damage degree until the failure occurs.
A failure probability evaluation system characterized by having. The failure probability evaluation system according to claim 1, wherein the storage unit is a temporary storage unit that temporarily stores the state quantity and the cumulative fatigue damage degree for an arbitrary period. The failure probability evaluation system according to any one of claims 1 to 2 , wherein the physical quantity is a load or a strain. The failure probability evaluation system according to any one of claims 1 to 3 , wherein the mechanical element is a bearing of the same type. The failure probability evaluation system according to any one of claims 1 to 4 , wherein the means for evaluating the cumulative fatigue damage degree is to calculate the fatigue damage degree based on the linear cumulative damage rule. In means for evaluating the quantity of state, and characterized by calculating a state quantity on the basis of the physical quantity of a plurality of types, failure probability evaluation system according to any one of claims 1 to 5. The failure history evaluation unit uses the time-series data of the state amount and the fatigue damage degree of the mechanical element whose state amount indicating the soundness has decreased, and uses the increment of the fatigue damage degree until the failure occurs as a function. generating a statistical model in which the state quantity as a variable, a fault of claim 1 that has been characterized for calculating the cumulative distribution function corresponding to the probability density function representing the variation of the fatigue damage of the increment in the statistical model Probability evaluation system. The failure probability evaluation system according to claim 7 , wherein the statistical model is a generalized linear model. Failure probability evaluation system according to any of the probability density function according to claim 7 or claim 8 characterized by using a gamma distribution. It has a function to display as a graph the relationship between the failure probability for any one or more of the plurality of machine elements calculated by the failure probability evaluation unit and the increment or time of the fatigue damage degree expected in the future. The failure probability evaluation system according to any one of claims 1 to 9 , characterized in that it has a display unit. Of the failure probabilities for any one or more of the plurality of machine elements calculated by the failure probability evaluation unit, at least one state in the future determined by the increment or time of the fatigue damage degree assumed in the future. The failure probability according to any one of claims 1 to 9 , wherein the evaluated failure probability has a display unit that displays the relationship between the predicted value up to the present time and the fatigue damage degree or time as a graph. Evaluation system.

Images