TWI749416B - Method for diagnosing abnormality of equipment having variable rotation speeds - Google Patents

Abstract

A method for diagnosing abnormality of equipment having variable rotation speeds is provided. The method includes a model establishing stage and an online operating stage. In the model establishing stage, historic operation data of the equipment having variable rotation speeds are obtained. Each of the historic operation data includes a historic rotation speed value and a corresponding historic operation state signal. Then, feature values of historic operation state are calculated in accordance with the historic operation data. Thereafter, a gauss mixture model (GMM) is calculated in accordance with the feature values of historic operation state. In the online operating stage, at first, an online rotation value and a corresponding online operation state signal are obtained when the equipment having variable rotation speeds online operates, and a feature value of online operation state is calculated accordingly. Thereafter, it is determined whether the equipment having variable rotation speeds is in an abnormal state when online operating in accordance with the gauss mixture model and the feature value of online operation state.

Description

Monitoring and Diagnosis Method for Abnormality of Variable Speed Equipment

The invention relates to a method for monitoring and diagnosing abnormalities in variable speed equipment.

In order to maintain and manage the equipment effectively, a related monitoring and diagnosis system is usually built to detect whether the equipment is abnormal. Generally speaking, the monitoring system will monitor the equipment as a whole or in part when the equipment is operating online to obtain signals representing the operating status of the equipment, and then analyze these signals to determine whether the equipment is operating abnormally. The conventional abnormal monitoring method can detect abnormalities more efficiently for fixed-speed equipment, and diagnose the abnormalities. However, for variable-speed equipment, the conventional abnormal monitoring method cannot efficiently detect equipment abnormalities.

Therefore, there is a need for an abnormal monitoring method for variable-speed equipment to efficiently detect abnormal operation of the variable-speed equipment.

According to one aspect of the present invention, the embodiments of the present invention provide a An abnormal monitoring method for variable speed equipment. This variable speed equipment abnormal monitoring method includes a model establishment phase and an online operation phase. In this abnormal monitoring method for variable speed equipment, the model establishment stage is first carried out. In the model building stage, first obtain a plurality of historical operating data of the variable speed equipment, and each historical operating data includes the historical speed value and the corresponding historical operating state signal. Then, calculate the characteristic value of the historical operating state based on the historical operating data. Next, the Gaussian mixture model is calculated based on the historical operating state characteristic values. Then, proceed to the online work phase. In the online operation phase, first obtain the online operation data of the variable speed equipment during online operation, where the online operation data includes the online rotation speed value and the corresponding online operation status signal. Then, at least one characteristic value of the online operation state is calculated according to the online operation data. Then, according to the Gaussian mixture model and the characteristic value of the online operating state, it is judged whether the variable speed equipment is abnormal during the online operation.

In some embodiments, the historical operating state signal and the online operating state characteristic value are the vibration signals of the variable speed equipment.

In some embodiments, the model building stage further includes a first normalization step to normalize the historical operating state characteristic values.

In some embodiments, the online operation phase further includes a second normalization step to normalize the characteristic value of the online operation state.

In some embodiments, the historical operation data corresponds to the normal operation of the speed equipment.

In some embodiments, the model building stage further includes a feature screening step to remove at least one redundant feature value from the historical operating state feature value.

In some embodiments, the aforementioned feature screening step includes: selecting a target feature value, where the target feature value is one of the historical operating state feature values; calculating the entropy value of the target feature value, and determining whether the entropy value is Is less than the preset entropy threshold; and when the entropy is less than the preset entropy threshold, it is determined that the target feature value is at least one redundant feature value, so as to remove the target feature value.

In some embodiments, the aforementioned feature screening step further includes: selecting a first target feature value and a second target feature value, wherein the first target feature value and the second target feature value are two of the historical operating state feature values; Calculate the average correlation coefficient between the first target feature value and the second target feature value; determine whether the average correlation coefficient is greater than the preset coefficient threshold; when the average correlation coefficient is greater than the preset coefficient threshold, determine the first target One of the feature value and the second target feature value is a redundant feature value to be removed.

In some embodiments, the aforementioned step of judging whether the variable speed equipment is abnormal during online operation according to the Gaussian mixture model and the characteristic value of the online operating state includes: determining the standard boundary line according to the plurality of boundary points of the Gaussian mixture model; judging the line Whether the characteristic value of the operating state exceeds the standard boundary; when the characteristic value of the operating state on the line exceeds the standard boundary, it is determined that the variable speed equipment is abnormal.

In some embodiments, the step of judging whether the variable speed equipment is abnormal during online operation according to the Gaussian mixture model and the characteristic value of the online operating state includes: calculating the weighted probability according to the Gaussian mixture model and the characteristic value of the online operating state; determining the weighting Whether the probability is less than the preset probability threshold; when the weighted probability is less than the preset probability threshold, it is determined that the variable speed equipment is abnormal.

100: Abnormal monitoring method for variable speed equipment

110: Model building stage

111~113: Step

112a~112b: steps

120: Online operation stage

121~123: Steps

123a~123d: steps

123e~123h: steps

In order to make the above and other objectives, features, advantages and embodiments of the present invention more obvious and understandable, the detailed description of the attached drawings is as follows: [FIG. 1] is a diagram showing the abnormal monitoring of the variable speed equipment according to the embodiment of the present invention Schematic flow diagram of the method.

[Fig. 2] is a schematic flow chart showing the steps of calculating historical operating state characteristic values according to an embodiment of the present invention; [Fig. 3] is a drawing showing historical operating state characteristic values according to an embodiment of the present invention; [Fig. 4] is a drawing [Fig. 5] is a schematic diagram showing the flow diagram of determining whether the operation of the equipment is abnormal according to the Gaussian mixture model according to the embodiment of the present invention; [Fig. 6] A schematic diagram showing the calculation of a standard boundary line according to an embodiment of the present invention; and [FIG. 7] is a schematic flowchart showing the steps of judging whether the operation of the equipment is abnormal according to the Gaussian mixture model according to an embodiment of the present invention.

The following is a detailed description of the implementation with the accompanying drawings, but the provided implementation is not used to limit the scope of the present invention, and the description of the structure operation is not used to limit the order of its execution, any structure recombined by components , The devices produced with equal efficacy are all covered by the present invention Scope. In addition, the drawings are for illustrative purposes only, and are not drawn in accordance with the original dimensions.

Regarding the "first", "second", ... etc. used in this article, they do not particularly refer to the order or sequence, but only to distinguish elements or operations described in the same technical terms.

Please refer to FIG. 1, which is a schematic flowchart of a method 100 for monitoring abnormality of a variable speed equipment according to an embodiment of the present invention. The method 100 for monitoring and diagnosing abnormalities in variable-speed equipment is applicable to equipment whose rotational speed changes, such as cooling fans for cooling water towers or combustion fans for blowing air. The method 100 for monitoring and diagnosing abnormalities of variable speed equipment includes a model establishment phase 110 and an online operation phase 120. The model building stage 110 may collect historical operating data of the variable speed equipment, and use these historical operating data to build a model of the equipment operation. In the online operation phase 120, when the equipment is online for operation, the aforementioned model can be used to determine whether the equipment is abnormal.

In the model building stage 110, step 111 is first performed to obtain a plurality of historical operating data of the variable speed equipment. Each historical operating data includes historical rotational speed values and corresponding historical operating status signals. For example, step 111 will record the operating state signals of the variable speed equipment at various speeds. Operating status signals include but are not limited to vibration signals, sound signals, or heat signals, etc. Taking the vibration signal as an example, step 111 will record the relationship (vibration signal) of the vibration acceleration value of the variable-speed equipment at a rotation speed with respect to time. The aforementioned historical operating data can be obtained by measuring the vibration acceleration value under a number of different equipment speeds.

Next, step 112 is performed to calculate historical operating state characteristic values based on historical operating data. In some embodiments of the present invention, the historical operating state characteristic value can be calculated according to a plurality of preset frequency bandwidth ranges. Please refer to FIG. 2, which is a schematic flowchart of step 112 according to an embodiment of the present invention. In step 112, step 112a is first performed to use a filter (such as a band pass/band reject filter) to divide the frequency of the historical operating state signal according to the aforementioned preset bandwidth range. Then, step 112b is performed to calculate the historical operating state characteristic value of the historical operating state signal corresponding to each bandwidth range. In the embodiment of the present invention, the historical operating state characteristic value includes but is not limited to mean value, mean square value, root mean square value, skewness, kurtosis, kurtosis index, skewness index, and Wiener entropy. The details are shown in Table 1 below:

In Table 1, x _i represents the historical operating state signal (time domain), f _i represents the energy corresponding to the i-th frequency, β represents kurtosis, and f _x represents skewness. Furthermore, assuming that the number of signals is N and the preset bandwidth range is: 0~1K Hz, 1K~2K Hz,..., 5K~6K Hz, the characteristic values of historical operating conditions as shown in Figure 3 can be obtained. , Where the historical operating state characteristic values shown in Figure 3 have been normalized. The equation for normalization is as follows:

表示第i筆資料的第J個特徵，μ^J和σ^J表示對應此特徵所算出的平均值與標準差，

表示正規化後的值。值得一提的是，在本發明之其他實施例中，亦可以不分頻段。換句話說，圖3可以只有一個頻段，即全頻段0~6K赫茲。 In the above normalized equation,

Represents the Jth feature of the i-th ^{data, μ J} and σ ^J represent the average value and standard deviation calculated corresponding to this feature,

Represents the normalized value. It is worth mentioning that in other embodiments of the present invention, frequency bands may not be divided. In other words, Figure 3 can have only one frequency band, that is, the full frequency range is 0~6K Hz.

Please return to FIG. 1, after step 112, step 113 is performed to calculate the corresponding Gaussian mixture model according to the characteristic value of the historical operating state for the bandwidth range of step 112. In order to make the Gaussian mixture model approach the distribution of data points, the number value M of the mixture model needs to be determined first. The method of determining the mixed model quantity value M is to continuously adjust the value of the mixed model quantity value M, and then calculate the Gaussian mixture model and its corresponding Akaike Information Criterion (Akaike Information Criterion) or Bayesian Information Criterion (BIC) value. The equations of AIC and BIC are as follows:

，其係代表資料點，S _i 代表轉速，i=1~N，μ _m 代表高斯分佈中心，Σ_m代表共變異數矩陣，π _m 代表與π相關的係數。資料點與高斯分佈中心之間的馬氏距離係表示如下：

In the above equation of AIC value/BIC value, D _f and N are the number of free parameters and the number of data points of a single Gaussian distribution, respectively,

, Its system represents the data point, S _i represents the rotation speed, i =1~N, μ _m represents the Gaussian distribution center, Σ _m represents the covariance matrix, and π _m represents the coefficient related to π. The Mahalanobis distance between the data point and the Gaussian distribution center is expressed as follows:

As shown in Figure 4, after calculating the AIC value/BIC value curve, then take a straight line between the beginning and the end of the AIC value/BIC value curve, and then find the point on the curve that is the farthest from the straight line (that is, the curve reflex point ), in this way, the value of the mixed model quantity value M can be determined. In the embodiment of FIG. 4, the number value M of the mixed model is determined to be 6.

In addition, in some embodiments of the present invention, a feature screening step may be performed between steps 112 and 113 to remove useless redundant feature values. In some embodiments of the present invention, when the aforementioned historical operation data are all data of the normal operation of the equipment, the feature selection step can be performed as follows: firstly, a target feature value to be processed is selected from the aforementioned historical operating state feature values. Of course Then, the entropy value of the target feature value is calculated, and it is judged whether the entropy value is less than the preset entropy threshold. In this embodiment, the preset entropy threshold is 0.5, but the embodiment of the present invention is not limited to this. When the entropy value is less than the preset entropy threshold, it is determined that the target feature value is a redundant feature value, and the target feature value is removed.

In some embodiments of the present invention, the relationship between any two feature values can be confirmed through the average correlation coefficient. Specifically, firstly, the first target characteristic value and the second target characteristic value are randomly selected from the historical operating state characteristic values, and the average correlation coefficient between the first target characteristic value and the second target characteristic value is calculated. Then it is judged whether the average correlation coefficient is greater than the preset coefficient threshold. When the average correlation coefficient is greater than the preset coefficient threshold, it is determined that one of the first target characteristic value and the second target characteristic value is a redundant characteristic value, and it is removed.

In some embodiments of the present invention, when the aforementioned historical operation data includes data including normal operation and abnormal operation of the equipment, the feature selection step may use Analysis of Variance (ANOVA) or Mutual Information (Mutual Information); MI) to determine the discriminative eigenvalues, and remove other redundant eigenvalues that are not discriminatory.

Please return to FIG. 1, after the model building stage 110, the online operation stage 120 is followed. In the online operation stage 120, step 121 is first performed to obtain online operation data of the variable speed equipment during online operation. This online operation data includes the online speed value and the corresponding online operation status signal. Then, step 122 is performed to calculate the online operating state characteristic value based on the online operating data. Since step 122 is similar to the aforementioned step 112, I won't repeat it here. Next, proceed to step 123 to determine whether the variable speed equipment is abnormal during online operation based on the Gaussian mixture model and the characteristic value of the online operating state.

Please refer to FIG. 5, which is a schematic flowchart of step 123 according to an embodiment of the present invention. In step 123, step 123a is first performed to determine the standard boundary line according to the plurality of boundary points of the Gaussian mixture model. In this embodiment, step 123 is to define the boundary between the normal situation and the abnormal situation (ie, the aforementioned standard boundary). For example, the calculation method of the boundary with Mahalanobis distance of 2 is shown in Figure 6. The boundary points of the Gaussian mixture model are connected and the outermost points are retained to obtain the normal upper boundary. This normal upper boundary can be used as a standard boundary line. Next, proceed to step 123b to determine whether the characteristic value of the on-line operating state exceeds the standard boundary line. If the characteristic value of the operating state on the line exceeds the standard boundary line, proceed to step 123c to determine that the operation of the variable speed equipment is abnormal. For example, when the characteristic value of the operating state of a single line exceeds the standard boundary line, it can be judged that the equipment is abnormal. Otherwise, proceed to step 123d to determine that the operation of the equipment is normal. In addition, considering that the number of boundary points may be too much, the smooth curve method (Spline) or polygonalization method (Polygonalization) can be used to obtain the standard boundary line.

Please refer to FIG. 7, which is a schematic flowchart of step 123 according to another embodiment of the present invention. In this embodiment, step 123 uses the weighted probability to determine whether the device is abnormal. In step 123 of this embodiment, step 123e is first performed to calculate the weighted probability according to the Gaussian mixture model and the characteristic value of the online operating state. The formula for calculating the weighted probability P is as follows:

Then, step 123f is performed to determine whether the weighted probability is less than the preset probability threshold. In this embodiment, the preset probability threshold is 0.5, but the embodiment of the present invention is not limited to this. When the weighted probability is less than the preset probability threshold, step 123g is performed to determine that the operation of the variable speed equipment is abnormal. Otherwise, proceed to step 123h to determine that the variable speed equipment is operating normally.

，其中w為轉速的權重，其值與建立高斯混合模型時一致。 In some embodiments, the weighted probability can be used to simultaneously use multiple indicators (characteristic values) to determine whether the operation of the variable speed equipment is abnormal. Take step 123 shown in Figure 7 as an example. When step 123 shown in Figure 7 simultaneously uses multiple indicators to determine whether the operation of the variable speed equipment is abnormal, the preset probability threshold can be 0.01, and Eigenvalue z _i =[ w ＊

, Where w is the weight of the speed, and its value is the same as when the Gaussian mixture model was established.

It can be seen from the above description that the method for monitoring abnormality of the variable speed equipment in the embodiment of the present invention uses a Gaussian mixture model to determine whether the variable speed equipment is abnormal. For different historical operating data, the embodiment of the present invention provides different feature screening steps. For example, if the historical operating data only contains data about the normal operation of variable speed equipment, the entropy value or average correlation coefficient can be used to filter the characteristics. For another example, if the historical operating data includes data about normal and abnormal operations of a variable speed equipment, the variance analysis method or the mutual information method can be used to filter the characteristics. In addition, in some embodiments of the present invention, the feature screening step is not limited to feature screening, but can also target other descriptors (for example, For example, frequency band) to filter.

Although the present invention has been disclosed in several embodiments as above, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field to which the present invention pertains can do various things without departing from the spirit and scope of the present invention. Modifications and modifications, therefore, the scope of protection of the present invention shall be subject to the scope of the attached patent application.

100‧‧‧Abnormal monitoring method for variable speed equipment

110‧‧‧Model establishment stage

111~113‧‧‧Step

120‧‧‧Online operation stage

121~123‧‧‧Step

Claims (7)

A method for monitoring abnormality of a variable-speed equipment for monitoring a variable-speed equipment, wherein the method for monitoring abnormality of the variable-speed equipment includes: performing a model establishment phase, including: obtaining a plurality of historical operating data of the variable-speed equipment, each The historical operating data includes a historical rotational speed value and a corresponding historical operating state signal; a plurality of historical operating state characteristic values are calculated according to the historical operating data, wherein the historical operating state characteristic values correspond to the same operating state And calculating a Gaussian mixture model based on the characteristic values of the historical operating conditions; and performing an online operation phase, including: obtaining an online operation data of the variable speed equipment during online operation, wherein the online operation data includes an online The speed value and a corresponding online operating state signal; calculating at least one online operating state characteristic value based on the online operating data; and judging that the variable speed equipment is operating online based on the Gaussian mixture model and the at least one online operating state characteristic value Whether there is an abnormality at the time; wherein the model establishment stage further includes a feature screening step to remove at least one redundant feature value from the historical operating state feature value, and the feature screening step includes: Select a target characteristic value, where the target characteristic value is one of the historical operating state characteristic values: calculate an entropy value of the target characteristic value, and determine whether the entropy value is less than a preset entropy threshold; And when the entropy value is less than the preset entropy threshold, determining that the target feature value is the at least one redundant feature value to remove the target feature value; wherein the feature screening step further includes: selecting a first target Characteristic value and a second target characteristic value, wherein the first target characteristic value and the second target characteristic value are two of the historical operating state characteristic values; calculating the first target characteristic value and the second target characteristic An average correlation coefficient (correlation coefficient) between values; determine whether the average correlation coefficient is greater than a preset coefficient threshold; and when the average correlation coefficient is greater than the preset coefficient threshold, determine the first target feature value and the One of the second target characteristic values is the at least one redundant characteristic value to remove the one of the first target characteristic value and the second target characteristic value. In the abnormal monitoring method for variable speed equipment described in the first item of the scope of patent application, the historical operating state signal and the online operating state characteristic value are the vibration signals of the variable speed equipment. The abnormal monitoring method for variable speed equipment as described in item 1 of the scope of patent application, wherein the model establishment stage further includes a first normalization Step to normalize the characteristic values of the historical operating conditions. According to the method for monitoring abnormality of variable speed equipment described in item 3 of the scope of patent application, the online operation phase further includes a second normalization step to normalize the at least one characteristic value of the online operating state. For example, the abnormal monitoring method for variable speed equipment described in item 1 of the scope of patent application, wherein the historical operation data all correspond to the normal operation of the speed equipment. According to the method for monitoring abnormality of variable speed equipment described in item 1 of the scope of patent application, the step of judging whether the variable speed equipment is abnormal during online operation according to the Gaussian mixture model and the at least one characteristic value of the online operating state includes: Determine a standard boundary line according to the plurality of boundary points of the Gaussian mixture model; determine whether the characteristic value of the operating state on the at least one line exceeds the standard boundary line; and when the characteristic value of the operating state on the at least one line exceeds the standard boundary line, It is determined that the variable speed equipment is abnormal. According to the method for monitoring abnormality of variable speed equipment described in item 1 of the scope of patent application, the step of judging whether the variable speed equipment is abnormal during online operation according to the Gaussian mixture model and the at least one characteristic value of the online operating state includes: Calculate a weighted probability according to the Gaussian mixture model and the at least one online operating state characteristic value; determine whether the weighted probability is less than a preset probability threshold; and when the weighted probability is less than the preset probability threshold, determine the change The speed device is abnormal.

Images