JP2016045852A - Abnormality diagnostic device and abnormality diagnostic method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable abnormality diagnosis of a machine plant with excellent reliability.SOLUTION: An abnormality diagnostic device 1 comprises: time series data acquisition unit 10 which acquires sensor data from a plurality of sensors installed in a machine plant as time series data; an anomaly measure calculation unit 12 which calculates an anomaly measure that is an index indicating the magnitude of difference from the normal condition of the machine plant on the basis of a statistical method using the time series data as learning data; an approximate equation calculation unit 13 which calculates an approximate equation approximating transition of the anomaly measure calculated on the basis of the time series data acquired in the past by use of a polynomial equation; and an anomaly measure estimation unit 14 which estimates the anomaly measure until a predetermined time in the future by use of the approximate equation. Whenever acquiring the time series data, the anomaly measure is calculated with respect to the latest time series data, the anomaly measure of the latest time series data is added to the transition to calculate the approximate equation, and the anomaly measure in the future is estimated by use of that approximate equation.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an abnormality diagnosis apparatus and an abnormality diagnosis method for diagnosing an abnormality in a mechanical facility.

  Construction machinery, medical equipment, power generation facilities such as wind power, solar power, and thermal power generation, water treatment, and various other mechanical equipment such as plants have failed to meet the final specifications due to lower operating rates due to abnormalities in mechanical equipment and performance and quality deterioration. Regular maintenance is performed to prevent adverse effects on customers such as lack of reliability. However, even if regular maintenance is carried out, it is inevitable that the machine equipment will be down or the performance will deteriorate due to a failure. Early detection of abnormality (detection of abnormal signs) based on the sensor data attached to the machine equipment will become important. ing.

  However, in the presence of a large amount of sensor data, a large amount of machine equipment information, and maintenance history information, it is necessary to grasp the state of the machine equipment and predict how much it can operate without any further failure (the time that the equipment can be operated continuously). Requires both design and field knowledge and a lot of data analysis, and is difficult and difficult.

  For example, in Patent Document 1, a Gaussian process, a k-NN (k-Nearest Neighbor) method, a particle filter method, and the like are applied to multidimensional sensor data acquired from a mechanical facility to estimate an abnormal measure of the mechanical facility. Furthermore, an abnormality diagnosis method for estimating RUL (Remaining Useful Life) is disclosed.

JP 2013-152655 A

However, since the state of the mechanical equipment changes from moment to moment, for example, as disclosed in Patent Document 1, if diagnosis is performed every time maintenance work is performed, abnormality diagnosis such as estimation of an abnormality measure or estimation of RUL is performed. It may be difficult to do with sufficient reliability.
Then, this invention makes it a subject to provide the abnormality diagnosis apparatus and abnormality diagnosis method which can perform abnormality diagnosis of a mechanical installation with the outstanding reliability.

  In order to solve the above-mentioned problem, the abnormality diagnosis apparatus of the present invention is an abnormality diagnosis apparatus for diagnosing abnormality of a mechanical facility, and acquires sensor data from a plurality of sensors installed in the mechanical facility as time series data. A time series data acquisition unit, a statistical method using the time series data as learning data, an abnormality measure calculation unit that calculates an abnormality measure that is an index indicating a magnitude of deviation from a normal state of the mechanical equipment, and a past An approximate expression calculation unit that calculates an approximate expression that approximates a transition of the abnormal measure calculated based on the time-series data acquired by using a polynomial, and an abnormal measure up to a predetermined time in the future using the approximate expression Each time the time series data is acquired by the time series data acquisition unit, the latest time series data is obtained by the abnormality measure calculation unit. And calculating the approximate expression obtained by adding the abnormal measure for the latest time series data to the transition by the approximate expression calculating unit, and using the approximate expression by the abnormal measure estimating unit in the future. An anomaly measure is configured to be estimated.

  According to the present invention, it is possible to perform abnormality diagnosis with excellent reliability.

It is a block diagram which shows the structure of the abnormality diagnosis apparatus which concerns on embodiment of this invention. It is a block diagram which shows the structure of the abnormality measure calculation part in the abnormality diagnosis apparatus which concerns on embodiment of this invention. It is a figure explaining the k-NN method as an example of the calculation method of the abnormality measure used in embodiment of this invention. It is a figure explaining the local subspace method as an example of the calculation method of the abnormal measure used in embodiment of this invention. It is a graph which shows typically the example of the abnormality measure used in embodiment of this invention. It is a graph which shows typically the example of the abnormal measure after the filter process used in embodiment of this invention. In embodiment of this invention, it is explanatory drawing for demonstrating the method of abnormality diagnosis. 5 is a flowchart showing a flow of abnormality diagnosis processing in the abnormality diagnosis apparatus according to the embodiment of the present invention. It is a flowchart which shows the flow of the abnormality measure calculation process in the flowchart shown in FIG.

  Embodiments of the present invention will be described in detail with reference to the drawings as appropriate. In each figure, common portions are denoted by the same reference numerals, and redundant description is appropriately omitted.

  The abnormality diagnosis apparatus according to the embodiment of the present invention has an abnormality measure that is an index indicating the state of mechanical equipment in order to maintain and improve the operation rate of mechanical equipment used in factories, commercial facilities, construction sites, hospitals, and the like. Understand and provide information for abnormality diagnosis. Therefore, in addition to sensor data or sensor data from the machine equipment to be diagnosed, information related to operation information, event information, equipment load, work reports, etc. is acquired, and the acquired information is used from now to the future. The abnormal measure of the machine equipment is estimated, and further the operation continuation time (RUL; remaining life) of the machine equipment is estimated.

[Configuration of abnormality diagnosis device]
First, the configuration of the abnormality diagnosis apparatus 1 according to the present embodiment will be described with reference to FIG.
As shown in FIG. 1, the abnormality diagnosis apparatus 1 includes a time series data acquisition unit 10, a time series data storage unit 11, an abnormality measure calculation unit 12, an approximate expression calculation unit 13, an abnormality measure estimation unit 14, A RUL calculation unit (operation continuation possible time calculation unit) 15, an abnormality sign detection unit 16, an output unit 17, a sensor data extraction unit 18, and a reference period setting unit 19 are configured.

The time-series data acquisition unit 10 is means for acquiring multi-dimensional sensor data output from the machine equipment to be diagnosed via the Internet network or the like. The sensor data is handled as time series data associated with the acquisition time (or time measured by the sensor). The time-series data acquisition unit 10 stores the latest acquired, that is, the current time-series data by sequentially storing them in the time-series data storage unit 11 every time the acquired time-series data is acquired.
Note that the current time series data may be output directly from the time series data acquisition unit 10 to the abnormality measure calculation unit 12.

  The time series data storage unit 11 stores the time series data input from the time series data acquisition unit 10. The time series data stored in the time series data storage unit 11 is appropriately referred to as past and present time series data by the abnormality measure calculation unit 12 and the sensor data extraction unit 18.

When newly acquired time-series data is added to the time-series data storage unit 11, the validity as data (not abnormal, already stored in the time-series data storage unit 11 by an evaluation unit (not shown). It is accumulated after evaluating (similarity to the data that has been made) and can be used as past time-series data in a normal state.
Further, when there are a plurality of machine facilities to be diagnosed, the time series data is stored in association with each machine facility to be a unit to be diagnosed.

  The time series data stored in the time series data storage unit 11 may include event data, operation data, load data, maintenance history data, and the like as environmental data in addition to the sensor data. Each of these data is associated with the time when each was acquired.

Here, the event data indicates the operation state of the machine facility, and indicates, for example, the control state of the operation pattern such as starting and stopping of the machine facility.
The operation data indicates an operation time such as an operation time and an operation time of the mechanical equipment and its accumulated time. For example, in the case of an excavator or the like, the detailed time of the operation such as the traveling time or the time of the turning operation corresponds.
The load data indicates the load state applied to the machine equipment, and corresponds to, for example, the load state and fuel consumption applied to the engine, the number of patients in the medical equipment, the hardness of the workpiece in the machine tool, and the like.
The maintenance history data indicates work history such as past failure contents and parts replacement with respect to the mechanical equipment, and includes a list of work items performed as maintenance work.

  The anomaly measure calculation unit 12 refers to the current and past time series data stored in the time series data storage unit 11 and uses a statistical method using the time series data as learning data, from the normal state of the machine equipment. An anomaly measure that is an index indicating the magnitude of the deviation is calculated. Further, the abnormality measure calculation unit 12 inputs the type of sensor data used when calculating the abnormality measure from the sensor data extraction unit 18, and uses the multidimensional sensor data including the specific type of sensor data to measure the abnormality measure. Is calculated. The abnormality measure calculation unit 12 outputs the calculated abnormality measure to the approximate expression calculation unit 13, the abnormality sign detection unit 16, and the reference period setting unit 19. Further, the abnormality measure calculation unit 12 accumulates the calculated abnormality measure as time series data in a storage unit (not shown), and time series data of the abnormality measure for a period according to a request from the approximate expression calculation unit 13 or the like. May be output.

In addition, as shown in FIG. 2, the abnormality measure calculation unit 12 includes a first abnormality measure calculation unit 121, a second abnormality measure calculation unit 122, an abnormality measure integration unit 123, and a filter processing unit 124. . The first abnormality measure calculation unit 121 and the second abnormality measure calculation unit 122 calculate the abnormality measure by different methods. The first abnormality measure calculation unit 121 and the second abnormality measure calculation unit 122 output the abnormality measure calculated by each to the abnormality measure integration unit 123.
As a method for calculating the anomaly measure, a vector quantization method, a local subspace method, or the like can be used. A description of the method for calculating these anomaly measures will be given later.

  Each of the first abnormality measure calculation unit 121 and the second abnormality measure calculation unit 122 may use multidimensional sensor data as a multidimensional vector as it is, but use a feature amount subjected to feature conversion. Also good. As a feature conversion method, for example, principal component analysis, independent component analysis, wavelet transform, or the like can be used. By performing the feature conversion, it is possible to reduce the number of dimensions of the time series data and improve the sensitivity of the abnormality measure.

The anomaly measure integration unit 123 receives the anomaly measures calculated from the first anomaly measure calculation unit 121 and the second anomaly measure calculation unit 122, integrates the two anomaly measures into one value, and performs a filter processing unit. It outputs to 124.
As a method of integrating the two abnormal measures, a harmonic average, a weighted average, or the like can be used. In particular, it is preferable to use a harmonic average in order to integrate the anomaly measures calculated by different methods so as not to impair the characteristics of the respective anomaly measures and to not emphasize the characteristics of the respective anomaly measures. .

  Note that the method used to calculate the abnormality measure is not limited to two, and may be one or three or more. Moreover, when calculating an abnormality measure with three or more methods, it is preferable to integrate them into one value by harmonic averaging.

The filter processing unit 124 inputs the abnormality measure integrated into one from the abnormality measure integration unit 123, and improves the sensitivity or reliability of abnormality diagnosis or removes a predetermined filter in the time axis direction in order to remove noise. Processing is performed. The abnormality measure after the filter processing by the filter processing unit 124 is output to the approximate expression calculation unit 13, the abnormality sign detection unit 16, and the reference period setting unit 19 shown in FIG.
In the present embodiment, the filtering process is performed on the integrated abnormality measure. However, the filtering process may be performed on the abnormality measure calculated by each method and then integrated.

  Examples of filter processing methods include a moving average filter that calculates a moving average for a predetermined time width, a minimum value filter that calculates a minimum value for a predetermined time width, and a maximum value for a predetermined time width. A maximum value filter can be used.

When the minimum value filter is used as the filter processing, for example, as shown in FIG. 5, the waveform 200 of the abnormal measure that changes while greatly vibrating up and down is connected to the lower limit value of the fine peak as shown in FIG. It is converted into the waveform 201 as described above. By using the minimum value filter, it is possible to prevent the abnormal measure from being excessively evaluated due to noise superposition. For this reason, abnormality diagnosis can be performed with high accuracy.
Further, when the maximum value filter is used as the filter processing, the maximum value of the abnormality measure can be grasped without missing. For this reason, abnormality diagnosis can be performed with high sensitivity.
In addition, when a moving average filter is used as the filter processing, an abnormality diagnosis of characteristics intermediate between the maximum value filter and the minimum value filter can be performed.
In these filter processes, the time width as the filter size can be appropriately determined according to the frequency characteristic of the abnormal measure, the frequency characteristic of the noise component included in the abnormal measure, and the like.

Here, two methods for calculating the abnormality measure will be described as examples.
In this embodiment, in any method, time-series data (hereinafter, also referred to as “normal data”) obtained by measuring with a sensor while the mechanical equipment to be diagnosed is operating normally is used. Is.
Note that, as described above, time-series data of feature amounts obtained by performing feature conversion on sensor data may be used.

(Vector quantization method)
First, the vector quantization method will be described with reference to FIG.
In the vector quantization method, normal data is clustered in advance as learning data, and current time-series data to be diagnosed (hereinafter also referred to as “diagnostic data”) and a cluster to which the diagnostic data belongs An anomaly measure is calculated based on the distance.
According to the vector quantization method, an abnormal measure can be calculated at high speed and with stable accuracy by using a cluster generated in advance.

As a clustering method, for example, a k-average method can be used. In addition, for example, a k-NN method can be used to determine a cluster to which diagnostic data belongs.
As shown in FIG. 3, the learning data includes a cluster A (a cluster in which the learning data indicated by a black circle “●” is a member x _A ) and a cluster B (a learning data indicated by a black square “■” is a member x _B ). Cluster). According to the k-NN method, as indicated by a broken-line circle in FIG. 3, first, k members (5 in the example of FIG. 3) xA, xB in the nearest vicinity of the diagnostic data q are extracted. Then, the number of members extracted for each cluster is counted, and the cluster with the largest number of extracted members is determined to be the cluster to which the diagnostic data q belongs. In the example shown in FIG. 3, since the members x _B belonging cluster B is most extracted three and, cluster membership diagnostic data q is determined as the cluster B.

The abnormality measure can be calculated by using a distance d between the diagnosis data q and the representative value c of the cluster B that is the belonging cluster (for example, the center of gravity of the belonging member can be used).
The anomaly measure may use the distance d between the diagnostic data q and the representative value c of the belonging cluster as it is, but is preferably normalized. Here, the normalization can be performed by dividing the distance d by a radius indicating the spread of the belonging cluster. The radius of the cluster is not particularly limited. For example, the average value of the distance from the representative value c to each member, the distance from the representative value c to the farthest member, the standard deviation or standard deviation of the members Can be used that is a constant multiple of.

  The determination of the cluster to which the diagnostic data q belongs is not limited to the k-NN method. For example, the cluster having the shortest distance between the diagnostic data q and the representative value c of each cluster can be determined as the belonging cluster.

(Local subspace method)
Next, the local subspace method will be described with reference to FIG.
In the local subspace method, k normal data nearest to the diagnostic data q are extracted, and are lowered from the diagnostic data q to a (k−1) -dimensional subspace defined by the extracted k normal data. An anomaly measure is calculated based on the length of the perpendicular.
According to the local subspace method, normal data similar to diagnostic data is extracted from the accumulated normal data and used. Therefore, even when the state of machine equipment changes drastically, it is possible to calculate an abnormal measure with high accuracy. is there.

In the example shown in FIG. 4, three normal data x ₁ , x ₂ , x ₃ are extracted with k = 3. Here, the two-dimensional subspace SS defined by the three normal data is a plane. Then, the distance d from the diagnostic data q to the foot Xb of the perpendicular drawn down to the partial space SS is calculated. The anomaly measure may use the distance d as it is, but is preferably normalized. Here, normalization can be performed, for example, by dividing the distance d by the standard deviation of k normal data extracted.

  In the present embodiment, for example, the first anomaly measure calculation unit 121 uses the vector quantization method, and the second anomaly measure calculation unit 122 uses the local subspace method to calculate the anomaly measure according to the above-described procedure. .

Returning to FIG. 1, the description of the configuration of the abnormality diagnosis apparatus 1 will be continued.
The approximate expression calculation unit 13 uses a past and present abnormality measure (an abnormality measure subjected to filter processing by the filter processing unit 124) input from the abnormality measure calculation unit 12, and approximates with a polynomial that indicates the transition of the abnormality measure. The equation is calculated. Further, the approximate expression calculation unit 13 inputs information about the reference period of the time series data of the abnormal measure to be referred to when calculating the approximate expression from the reference period setting unit 19. The approximate expression calculation unit 13 calculates an approximate expression that fits the time series data of the anomaly measure in the reference period indicated by the information. Here, calculating an approximate expression means calculating each coefficient of a polynomial that is an approximate expression. In addition, the approximate expression calculation unit 13 outputs the calculated approximate expression to the abnormality measure estimation unit 14.
Furthermore, it is preferable that the approximate expression calculation unit 13 also calculates the error of the coefficient of the approximate expression. As a result, the reliability of the approximate expression can be indicated by the magnitude of the error. When the approximate expression calculation unit 13 calculates a coefficient error, the coefficient error is also output to the abnormality measure estimation unit 14.

  Note that the order of the polynomial that is an approximate expression is not particularly limited, and a polynomial of the first order or higher can be used. Further, the degree of the polynomial may be configured to be appropriately selected according to an instruction from the operator. For example, as shown in FIG. 6, a transition of an abnormal measure (a waveform 201 indicated by a solid line) and a curve (a waveform 202 indicated by a broken line) drawn using an approximate expression of a selected order are displayed on a display device, a printing device, or the like. The graph may be displayed in a graph so that the operator can repeat the selection of the order and the recalculation of the approximate expression until an appropriate order that fits the curve indicating the transition of the abnormal measure displayed in the graph is determined. .

Here, the case of approximating a polynomial approximation of an abnormal measure by a cubic function will be described as an example.
When the anomaly measure is y and the time is x, the anomaly measure y can be expressed as Equation (1) using coefficients a, b, c, and d as a cubic function of the time x.
y = ax ³ + bx ² + cx + d (1)
The coefficients a, b, c, and d can be calculated by applying the least square method using time series data of (x, y) that is actually measured data. When the degree of the polynomial is first order, second order or fourth order or more, the number of coefficients increases or decreases, but can be calculated by the least square method.

  Moreover, the error about each most probable value of the coefficients a, b, c, and d can be obtained by a statistical method from the distribution of errors from the approximate expression of (x, y) that is a measured value. That is, when it is assumed that the error from the approximate expression of x, y follows a normal distribution, the standard error can be calculated for the coefficients a, b, c, d from the standard deviation of the error of x, y. .

As a simple example, a case where the anomaly measure y is approximated by a linear function at time x will be described. In this case, the approximate expression can be expressed as Expression (2) using the coefficients a and b.
y = ax + b (2)
The coefficients a and b are calculated by the following procedure. That is, assuming that the number of measured values is N and individual measured values for x and y are expressed as y _i and x _i using the subscript i, and the measured value x _i has almost no error. The coefficients a and b are respectively expressed as Expression (3.1) and Expression (3.2), and the standard errors σ _a and σ _b of the coefficients a and b are expressed as Expression (4.3). Using the standard deviation σ y of the error of _y , they are expressed as Equation (4.1) and Equation (4.2), respectively.

  The anomaly measure estimator 14 inputs polynomial coefficients and coefficient errors from the approximate expression calculator 13 as approximate expressions, and uses the approximate expressions to calculate a predetermined period from the present time to the future as an estimation period (in FIG. A period from time t0 to time t2), and a future abnormality measure is estimated. Here, estimating the future anomaly measure is, for example, when the above-described equation (1) is used as an approximate expression, coefficients a, b, c, d and appropriate intervals (for example, time series) The anomaly measure y for each time x is calculated at the same interval as the data sampling). The abnormality measure estimation unit 14 outputs the abnormality measure calculated for the estimation period to the RUL calculation unit 15 and the output unit 17.

  Note that the estimation period of the abnormality measure may be a predetermined period, or may be designated by an operator using appropriate input means such as a keyboard or a pointing device.

  The RUL calculation unit (operation continuation possible time calculation unit) 15 inputs the abnormality measure calculated for the estimation period from the abnormality measure estimation unit 14, and based on the transition of the abnormality measure in the estimation period, the RUL (operation continuation possible) Time). The RUL calculation unit 15 outputs the calculated RUL to the output unit 17.

Here, a method of calculating the RUL will be described with reference to FIG.
In FIG. 7, a waveform 201 indicates an actual measurement value of an abnormal measure, a waveform 202 indicated by a solid line indicates an abnormal measure up to the current time calculated using an approximate expression, and a waveform 203 indicated by a broken line is an approximation. The most probable value of the anomaly measure for the estimation period calculated using the equation is shown. The period from the past time t1 to the current time t0 is a reference period that is referred to when calculating the approximate expression.
Further, the threshold value TH indicates an upper limit value of the abnormality measure that is a limit at which the machine facility can be normally operated. That is, it indicates that the mechanical equipment can be operated until the time when the estimated waveform 203 of the abnormal measure reaches the threshold value TH. Therefore, the time from the current time t0 to the time when the estimated abnormality measure reaches the threshold value TH can be calculated as an estimated value of RUL (see FIG. 7).

Furthermore, when calculating the approximate expression, by calculating the error of the approximate expression, that is, the error of the coefficient of the polynomial that is the approximate expression, the upper limit value and the lower limit value of the approximate expression are estimated using the error. be able to. As shown in FIG. 7, during the estimation period, the waveform 203 shows the most probable value of the estimated abnormality measure, and the waveforms shown by dotted lines above and below the waveform 203 show the upper and lower limit values of the estimated value of the abnormality measure. is there.
Thus, in addition to the most probable value of the estimated anomaly measure, for example, in Equation (2), an approximate expression in which the coefficients a and b are changed from the most probable value by 1.96 times the standard errors σ _a and σ _b is obtained. The reliability of the estimated value can be grasped by calculating the upper and lower limit values of the anomaly measure in the 95% confidence interval. Further, in addition to or instead of the estimation error of the abnormality measure, the estimation error of the RUL may be calculated using a time when the waveform of the upper and lower limit values of the abnormality measurement exceeds the threshold value TH.

  Further, as shown in FIG. 7, by displaying each waveform 201, 202, 203, threshold value TH, RUL estimation value, RUL estimation error, etc. and presenting it to the operator, the future transition of the anomaly measure and its reliability ( (Relevance) can be easily grasped by the operator.

Furthermore, in this embodiment, every time new diagnostic data (time-series data) is acquired, an abnormal measure is calculated, the reference period to which the latest abnormal measure is added is reset, and an approximate expression is sequentially calculated. Furthermore, RUL is recalculated.
In this way, each time series data is acquired, the reference period is reset and the approximate expression of the abnormality measure is recalculated, so that the RUL can be estimated based on the latest approximate expression. In addition, the display content such as the waveform shown in FIG. 7 is updated with recalculation of the approximate expression and RUL. For this reason, abnormality diagnosis of mechanical equipment, such as estimation of RUL, can always be performed with high accuracy. In addition, since a polynomial is used as an approximate expression, even when time series data is acquired at a very short interval, the approximate expression can be easily calculated each time time series data is acquired.

Returning to FIG. 1, the description of the configuration of the abnormality diagnosis apparatus 1 will be continued.
The abnormality sign detection unit 16 inputs the abnormality measure calculated for the latest time series data from the abnormality measure calculation unit 12, and determines whether or not the abnormality measure exceeds a predetermined threshold value. Detect the presence or absence. When the abnormality measure exceeds a predetermined threshold, the abnormality sign detection unit 16 diagnoses that there is an “abnormal sign” in the mechanical equipment and outputs the diagnosis result to the output unit 17.
Note that the threshold TH used to calculate the RUL can be used as the threshold for detecting the abnormal sign, but the threshold estimated by RUL is lower than the threshold TH. Alternatively, an abnormality sign may be detected at an early stage.

  The output unit 17 inputs time series data of the abnormality measure calculated for the estimation period from the abnormality measure estimation unit 14, inputs RUL from the RUL calculation unit 15, and diagnoses whether there is an abnormality sign from the abnormality sign detection unit 16. And the input data is displayed. The output unit 17 outputs these data to AHM (asset health management) and EAM (enterprise asset management), which are not shown, instead of or in addition to displaying these input data. Further, the output unit 17 further inputs a past abnormal measure from the abnormal measure calculation unit 12 and an approximate equation from the approximate expression calculation unit 13, and displays data relating to the abnormal measure in a graph as shown in FIG. You may make it do.

  The sensor data extraction unit 18 extracts one or more sensor data greatly affecting the abnormality measure from the multidimensional sensor data accumulated in the time series data storage unit 11, and sets the type of the extracted sensor data as the abnormality measure. Output to the calculation unit 12. By reducing the number of dimensions of the sensor data used for calculating the abnormal measure, the processing load for calculating the abnormal measure can be reduced. Note that the abnormality measure calculation unit 12 may calculate the abnormality measure using all types of sensor data without extracting the sensor data.

As a method for extracting sensor data, for example, an impulse response between the sensor data can be used. In other words, the impulse response to the abnormal measure of each sensor data is examined using the time-series data acquired in the past accumulated in the time-series data storage unit 11, and the sensor data having a large influence on the change of the abnormal measure is obtained. It can be extracted in advance.
As another extraction method, when the abnormality measure greatly changes, sensor data having a large contribution to the abnormality measure may be extracted in advance.

  The reference period setting unit 19 receives time series data of past abnormal measures from the abnormal measure calculation unit 12 and determines a period of time series data of the abnormal measure to be referred to by the approximate expression calculation unit 13 to calculate. It is. The reference period setting unit 19 outputs the determined reference period of the past abnormal measure to the approximate expression calculation unit 13.

The reference period setting unit 19 analyzes the transition of the past abnormal measure. For example, the reference period setting unit 19 starts from the time when the slope of the change in the abnormal measure becomes a predetermined threshold or more (for example, time t1 in FIG. 7), 7 is determined as a reference period for calculating the approximate expression (see FIG. 7). In addition, the reference period setting unit 19 may set a time corresponding to an inflection point with a positive slope as a starting point of the reference period in the graph indicating the transition of the abnormality measure.
Further, the reference period setting unit 19 may input a reference period according to an instruction from an operator via an input unit such as a keyboard or a mouse, and output the input reference period to the approximate expression calculation unit 13.

[Operation of abnormality diagnosis device]
Next, referring to FIGS. 8 and 9 (refer to FIGS. 1 and 2 as appropriate), the operation of the abnormality diagnosis apparatus 1 according to the embodiment for performing abnormality diagnosis processing will be described.
As illustrated in FIG. 9, the abnormality diagnosis apparatus 1 acquires, as time series data, sensor data that is a measurement value of a sensor installed in a mechanical facility by the time series data acquisition unit 10 (step S10). The time series data acquired by the time series data acquisition unit 10 is stored in the time series data storage unit 11 as a normal data database. Note that time-series data in which the mechanical equipment is not normal, for example, time-series data whose validity has been denied by an evaluation unit (not shown), is not accumulated in the time-series data storage unit 11.
Further, it is assumed that the time series data storage unit 11 stores in advance an amount of normal data necessary for calculation of the abnormality measure.

  Next, the abnormality diagnosing device 1 uses the abnormality measure calculation unit 12 as the diagnosis data with the latest time-series data acquired in step S10 as the diagnosis data, and appropriately refers to the past normal data, and the abnormality measure for the diagnosis data. Is calculated (step S11).

Here, with reference to FIG. 9, the abnormality measure calculation processing step (S11) performed by the abnormality measure calculation unit 12 will be described.
The anomaly measure calculation unit 12 uses the first anomaly measure calculation unit 121 to calculate an anomaly measure using a first method (for example, a vector quantization method) (step S21).
Next, the anomaly measure calculation unit 12 uses the second anomaly measure calculation unit 122 to calculate an anomaly measure using a second method (for example, the local subspace method) (step S22).
Note that either step S21 or step S22 may be performed first or in parallel.
In addition, it is assumed that the type of sensor data used in step 21 and step S22 is extracted by the sensor data extraction unit 18 in advance.

Next, the abnormality measure calculation unit 12 integrates the abnormality measure calculated in step S21 and step S22 into one value by the abnormality measure integration unit 123 (step S23).
Next, the abnormality measure calculation unit 12 adds a predetermined value such as a minimum value filter to the time series data of the abnormality measure including the latest abnormality measure for the abnormality measure integrated into one value by the filter processing unit 124 in step S23. Filter processing is performed (step S24). It is assumed that the time series data of the abnormality measure calculated by the abnormality measure calculation unit 12 is accumulated in the storage unit in the abnormality measure calculation unit 12.

Returning to FIG. 8, the description of the operation of the abnormality diagnosis apparatus 1 will be continued.
In the abnormality diagnosis apparatus 1, the reference period setting unit 19 sets the reference period of the time series data of the abnormality measure used for calculating the approximate expression (step S12).

Next, the abnormality diagnosis apparatus 1 uses the approximate expression calculation unit 13 to calculate an approximate expression indicating the transition of the abnormality measure using the time series data of the reference period set in step S12 (step S13).
Next, the abnormality diagnosing device 1 uses the approximate expression calculated in step S <b> 13 to measure an abnormality measure at a predetermined time interval (for example, sampling interval of time series data) for a predetermined future period that is an estimation period. Is calculated (estimated) (step S14).

Next, the abnormality diagnosis apparatus 1 calculates the RUL by using the estimated value of the abnormality measure calculated in step S14 by the RUL calculation unit 15 (step S15).
Next, the abnormality diagnosis device 1 diagnoses the presence / absence of an abnormality sign by using the abnormality measure for the diagnosis data calculated in step S11 by the abnormality sign detection unit 16 (step S16). Note that step S16 may be performed at any timing after step S11.

  Next, the abnormality diagnosis apparatus 1 uses the output unit 17 to output diagnosis results such as the RUL calculated in step S15, the presence / absence of an abnormality sign diagnosed in step S16, and the estimated value of the abnormality measure calculated in step S14. The information is displayed on a display device (not shown) and / or output to an external host system.

  Moreover, the abnormality diagnosis apparatus 1 executes the processing from step S10 to step S17 each time new time series data is acquired as diagnostic data.

  In addition, this invention is not limited to above-described embodiment, Various modifications are included. For example, the above-described embodiment has been described in detail for easy understanding of the present invention, and is not necessarily limited to one having all the configurations described. Further, a part of the configuration of an embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of an embodiment. In addition, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. In addition, each of the above-described configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tapes, and files that realize each function can be stored in a recording device such as a memory, a hard disk, or an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
Further, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

DESCRIPTION OF SYMBOLS 1 Abnormality diagnosis apparatus 10 Time series data acquisition part 11 Time series data storage part 12 Abnormality measure calculation part 13 Approximation formula calculation part 14 Abnormality measure estimation part 15 Abnormal sign detection part 16 RUL calculation part (operation continuation time calculation part)
DESCRIPTION OF SYMBOLS 17 Output part 18 Sensor data extraction part 19 Reference period setting part 121 1st abnormality measure calculation part 122 2nd abnormality measure calculation part 123 Abnormality measure integration part 124 Filter processing part 200 Abnormality measurement waveform 201 Abnormality measurement waveform after filter processing 202 Waveform of approximate expression 203 Waveform of approximate expression of estimation period TH threshold

In order to solve the above-mentioned problem, the abnormality diagnosis apparatus of the present invention is an abnormality diagnosis apparatus for diagnosing abnormality of a mechanical facility, and acquires sensor data from a plurality of sensors installed in the mechanical facility as time series data. and time-series data acquisition unit, and the time by a statistical technique series data and training data, the anomaly measure calculator for calculating an index indicating the magnitude of abnormal measure of deviation from the normal state of the machinery, the For an abnormal measure, a minimum value filtering process for calculating a minimum value in a predetermined time width, a maximum value filtering process for calculating a maximum value for a predetermined time width, or a movement for calculating a moving average for a predetermined time width and filtering unit for performing any one of filtering of the average filter, the transition of the anomaly measure calculated based on the time series data acquired in the past, Using the anomaly measure whose serial filtering process has been performed, and the approximate expression calculation section which calculates an approximate expression that approximates a polynomial, by using the approximate expression, an abnormal measure estimation for estimating the anomaly measure up to a predetermined future point in time Each time the time-series data acquisition unit acquires the time-series data, the abnormality measure calculation unit calculates an abnormality measure for the latest time-series data, and the approximate expression calculation unit calculates the abnormality The approximate formula obtained by adding the abnormal measure for the latest time series data to the transition is calculated, and the abnormal measure estimating unit is configured to estimate a future abnormal measure using the approximate formula.

In order to solve the above-mentioned problem, the abnormality diagnosis apparatus of the present invention is an abnormality diagnosis apparatus for diagnosing abnormality of a mechanical facility, and acquires sensor data from a plurality of sensors installed in the mechanical facility as time series data. A time series data acquisition unit, and a statistical method using the time series data as learning data, an abnormality measure calculation unit that calculates an abnormality measure that is an index indicating the magnitude of deviation from the normal state of the mechanical equipment, with respect to an error measure, filter processing unit that performs either Kano filtering of the maximum value filter processing for calculating the maximum value of the minimum value filtering or predetermined time width calculating the minimum value in a predetermined time width And an approximation obtained by approximating the transition of the abnormal measure calculated based on the time series data acquired in the past by a polynomial using the abnormal measure subjected to the filtering process. And an abnormal measure estimation unit that estimates an abnormal measure up to a predetermined time point in the future using the approximate equation, and acquires the time series data by the time series data acquisition unit. Each time, the abnormal measure calculating unit calculates an abnormal measure for the latest time series data, and the approximate equation calculating unit adds the abnormal measure for the latest time series data to the transition. It is configured to calculate and estimate a future abnormal measure using the approximate expression by the abnormal measure estimation unit.

Claims (6)

An abnormality diagnosis device for diagnosing an abnormality in a mechanical facility,
A time-series data acquisition unit that acquires sensor data from a plurality of sensors installed in the mechanical equipment as time-series data;
A statistical measure using the time-series data as learning data, an abnormal measure calculation unit that calculates an abnormal measure that is an index indicating the magnitude of deviation from the normal state of the mechanical equipment,
An approximate expression calculation unit for calculating an approximate expression obtained by approximating a transition of the abnormal measure calculated based on the time series data acquired in the past by a polynomial;
An anomaly measure estimation unit that estimates an anomaly measure up to a predetermined time in the future using the approximate expression, and
Each time the time-series data is acquired by the time-series data acquisition unit, the abnormal measure calculation unit calculates an abnormal measure for the latest time-series data, and the approximate expression calculation unit calculates the latest time-series data. An abnormality diagnosis apparatus characterized by calculating the approximate expression obtained by adding the abnormal measure to the transition and estimating a future abnormal measure using the approximate expression by the abnormal measure estimation unit. The anomaly measure calculation unit calculates an anomaly measure by two or more methods,
An anomaly measure integration unit for integrating the anomaly measures calculated by the two or more methods;
The abnormality diagnosis apparatus according to claim 1, wherein the abnormality measure integration unit integrates the abnormality measures by calculating a harmonic average of abnormality measures by two or more methods. A filter processing unit that performs a minimum value filtering process for calculating a minimum value in a predetermined time width for the abnormal measure,
The abnormality diagnosis apparatus according to claim 1, wherein the approximate expression calculation unit calculates the approximate expression using an abnormality measure subjected to the minimum value filter processing.   The abnormal measure estimation unit performs the estimation by calculating an upper limit value and / or a lower limit value of a predetermined confidence interval of the abnormal measure calculated by a statistical method in addition to the most probable value of the abnormal measure. The abnormality diagnosis device according to any one of claims 1 to 3, wherein the abnormality diagnosis device is characterized in that:   The operation continuation that calculates the period from when the latest time-series data is acquired until the estimated value of the abnormality measure exceeds a predetermined threshold as the operation continuation time indicating the time during which the mechanical equipment can be operated continuously The abnormality diagnosis device according to any one of claims 1 to 4, further comprising a possible time calculation unit. An abnormality diagnosis method for diagnosing an abnormality in mechanical equipment,
Time-series data acquisition processing step for acquiring sensor data from a plurality of sensors installed in the mechanical equipment as time-series data;
Abnormal measure calculation processing step of calculating an abnormal measure that is an index indicating the magnitude of deviation from the normal state of the mechanical equipment by a statistical method using the time series data as learning data;
An approximate expression calculation processing step for calculating an approximate expression obtained by approximating a transition of the abnormal measure calculated based on the time series data acquired in the past by a polynomial;
An anomaly measure estimation processing step for estimating an anomaly measure up to a predetermined time in the future using the approximate expression, and
Every time the time series data is acquired, an abnormal measure for the latest time series data is calculated, and the approximate expression obtained by adding the abnormal measure for the latest time series data to the transition is calculated, and the approximate expression A method for diagnosing an abnormality characterized by estimating a future anomaly measure using a computer.

Images