Power system asset predictive health analysis

Abstract

The present disclosure relates to a method of performing a predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the method comprising: acquiring a signal of at least one item of operational data; segmenting a signal of the at least one item of operation data into a plurality of feature segments; determining, for each of the plurality of feature segments, a plurality of sets of metrics based on the signal of the at least one item of operational data in the respective plurality of feature segments, where each of the plurality of sets of metrics includes at least one metric indicating a degree of correspondence between the signal of the at least one item of operational data in the respective plurality of feature segments and a mathematical function; selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets; determining a predictive asset health based on the signal selected in at least one of the plurality of feature segments; and generating an output based on the predictive asset health status. The present disclosure also relates to a corresponding method of operating and / or maintaining an asset, a decision system operable to perform a predictive health analysis on an asset, and an industrial power system.

Description

Technical Field

[0001] The present disclosure relates to a method of performing predictive health analysis, a decision system, and an industrial or power system. Background Art

[0002] In power systems, physical devices (also known as assets) degrade over time. Degraded assets can lead to improper operation within the power system, necessitating monitoring and predictive health analysis to plan repairs, maintenance, and other tasks, thereby improving power system reliability. Specifically, estimating remaining useful life (RUL) can serve as an indicator of asset health. Traditionally, RUL is estimated from monitored data by extracting, analyzing, and selecting features that best reflect the asset's RUL or have the greatest significance relative to the asset's RUL. However, this estimation becomes challenging when assets operate in different operating modes. In such cases, monitored data often exhibit transitions, resulting in discontinuities that bias feature significance and, in turn, lead to unreliable RUL estimates. For example, a considered feature may demonstrate high significance relative to the RUL estimate when evaluated in a single operating mode, but may appear insignificant when evaluated across multiple operating modes. This can result in less significant features being used to estimate RUL, while more significant features are ignored. This in turn leads to lower accuracy and robustness of RUL estimates and may limit the success of maintenance decision planning, failures and unnecessary maintenance operations.

[0003] Therefore, there is a need to improve methods for performing predictive health analysis, decision systems, and industrial or power systems. Summary of the Invention

[0004] The present disclosure relates to a method for performing predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the method comprising: acquiring a signal of at least one item of operating data; segmenting the signal of at least one item of operating data into a plurality of feature segments; determining a plurality of metric sets for each of the plurality of feature segments based on the signal of at least one item of operating data in the corresponding plurality of feature segments, wherein each of the plurality of metric sets includes at least one metric indicating a degree of correspondence between the signal of at least one item of operating data in the corresponding plurality of feature segments and a mathematical function; selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets; determining a predictive asset health state based on the selected signal in at least one of the plurality of feature segments; and generating an output based on the predictive asset health state.

[0005] In one embodiment, segmenting the signal of at least one item of operational data into a plurality of characteristic segments is based on an operational mode of the asset. Persons skilled in the art will appreciate that the term "segmenting" may refer to separating, splitting, dividing, or the like. In one embodiment, segmenting the signal may include or include detecting a transient in the signal, and segmenting the data into different segments based on the detected transient in the signal.

[0006] In one embodiment, the method further comprises at least one of: planning maintenance based on the output; replacing at least one sensor used to monitor data based on the output, in particular, a sensor identified as being critical to asset health monitoring; removing at least one sensor used to monitor data that is identified as having no information value based on the output; sending the generated output to an end user via a graphical user interface; and optimizing the design of a monitoring system for the asset, in particular, by retaining at least one sensor identified as being useful based on the output.

[0007] In one embodiment, determining the predictive asset health includes or is determining a RUL.

[0008] In one embodiment, the method further comprises: decomposing a signal of at least one item of operation data in the plurality of feature segments into at least a trend portion and a residual portion.

[0009] In one embodiment, determining a plurality of sets of metrics for each of the plurality of feature segments is based on a trend portion of a decomposed signal of at least one item of operational data in the plurality of feature segments.

[0010] In one embodiment, the method further comprises combining the plurality of metric sets into a single metric set indicating a degree of correspondence between a signal of at least one item of operational data in the plurality of feature segments and the mathematical function.

[0011] In one embodiment, the method further comprises normalizing the plurality of metric sets in amplitude and / or time.

[0012] In one embodiment, the method further includes receiving sensor measurement data captured during operation of the asset; and updating the predictive asset health state based on the received sensor measurement data.

[0013] In one embodiment, the asset is a power transformer, a distributed energy resource (DER) unit, or a generator.

[0014] In one embodiment, each of the plurality of metric sets includes at least one metric indicating a goodness of a signal of at least one item of operational data in the corresponding plurality of feature segments for determining a predictive asset health state.

[0015] In one embodiment, the signal is selected in at least one of the plurality of signature segments for determining a predictive asset health state based on a plurality of metric sets.

[0016] In one embodiment, the method further includes combining the plurality of metric sets into a single metric set indicating how good a signal of at least one operational data item in the plurality of characteristic segments is used to determine the predictive asset health status.

[0017] The present disclosure also relates to a method for operating and / or maintaining an asset (in particular, a power system asset or an industrial asset), the method comprising: performing a predictive health analysis on the asset using any of the above methods; and automatically performing at least one of the following: scheduling downtime of the asset based on the determined predictive asset health status; scheduling maintenance work based on the determined predictive asset health status; scheduling replacement work based on the determined predictive asset health status; changing maintenance intervals based on the determined predictive asset health status.

[0018] The present disclosure further relates to a determination system operable to perform predictive health analysis on an asset (in particular, for determining the remaining useful life (RUL) of a power system asset or an industrial asset), the determination system comprising at least one integrated circuit operable to: acquire a signal of at least one item of operating data; segment the signal of at least one item of operating data into a plurality of feature segments; determine a plurality of metric sets for each of the plurality of feature segments based on the signal of at least one item of operating data in the corresponding plurality of feature segments, wherein each of the plurality of metric sets includes at least one metric indicating a degree of correspondence between the signal of at least one item of operating data in the corresponding plurality of feature segments and a mathematical function; select a signal in at least one of the plurality of feature segments based on the plurality of metric sets; determine a predictive asset health state based on the selected signal in at least one of the plurality of feature segments; and generate an output based on the predictive asset health state.

[0019] In one embodiment, the at least one integrated circuit is further operable to segment a signal of the at least one item of operational data into a plurality of characteristic segments based on the asset's operational mode.

[0020] The present disclosure also relates to an industrial or power system comprising: an asset and the above-mentioned determination system for performing predictive health analysis on the asset, optionally, wherein the determination system is a decentralized controller of the industrial or power system for controlling the asset.

[0021] The following items relate to specific embodiments of the present disclosure:

[0022] 1. A method for performing predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the method comprising:

[0023] a signal for obtaining at least one item of operational data;

[0024] Segmenting the signal of the at least one item of operational data into a plurality of characteristic segments;

[0025] determining a plurality of metric sets for each of the plurality of feature segments based on a signal of the at least one item of operational data in a corresponding plurality of feature segments,

[0026] wherein each of the plurality of metric sets comprises at least one metric indicating a degree of correspondence between a signal of the at least one item of operational data in the corresponding plurality of feature segments and a mathematical function;

[0027] selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets;

[0028] determining a predictive asset health state based on a selected signal in at least one of the plurality of feature segments; and

[0029] An output is generated based on the predictive asset health.

[0030] 2. The method according to claim 1, further comprising at least one of the following:

[0031] planning maintenance based on the output;

[0032] replacing at least one sensor used for monitoring data based on the output, in particular, a sensor identified as being critical to the health monitoring of the asset;

[0033] removing at least one sensor for monitoring data whose output is identified as having no information value based on the output;

[0034] sending the generated output to an end user via a graphical user interface; and

[0035] The design of a monitoring system for the asset is optimized, in particular, by retaining at least one sensor identified as useful based on the output.

[0036] 3. The method of clause 1 or 2, wherein determining the predictive asset health state comprises determining a RUL, or determining a RUL.

[0037] 4. The method according to any of the preceding claims, further comprising:

[0038] Decomposing a signal of at least one item of operation data in the plurality of feature segments into at least a trend portion and a residual portion; and

[0039] Wherein, determining a plurality of metric sets for each of the plurality of feature segments is based on the trend portion of the decomposed signal of at least one item of operational data in the plurality of feature segments.

[0040] 5. The method according to any of the preceding claims, further comprising combining the plurality of metric sets into a single metric set, the single metric set indicating a degree of correspondence between the signal of at least one item of operational data in the plurality of feature segments and the mathematical function.

[0041] 6. The method of clause 5, further comprising normalizing the plurality of metric sets in amplitude and / or time.

[0042] 7. The method according to any of the preceding claims, further comprising:

[0043] receiving sensor measurement data captured during operation of the asset; and

[0044] The predictive asset health status is updated based on the received sensor measurement data.

[0045] 8. The method according to any of the preceding claims, wherein the asset is a power transformer, a distributed energy resource (DER) unit, or a generator.

[0046] 9. The method according to any of the preceding claims, wherein segmenting the signal of the at least one item of operational data into a plurality of characteristic segments is based on an operational mode of the asset.

[0047] 10. A method of operating and / or maintaining an asset, in particular a power system asset or an industrial asset, comprising:

[0048] performing a predictive health analysis on the asset using any of the methods described above; and

[0049] Automatically perform at least one of: scheduling downtime for the asset based on the determined predictive asset health; scheduling maintenance work based on the determined predictive asset health; scheduling replacement work based on the determined predictive asset health; changing maintenance intervals based on the determined predictive asset health.

[0050] 11. A determination system operable to perform predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the determination system comprising at least one integrated circuit operable to:

[0051] a signal for obtaining at least one item of operational data;

[0052] Segmenting the signal of the at least one item of operational data into a plurality of characteristic segments;

[0053] determining a plurality of metric sets for each of the plurality of feature segments based on a signal of the at least one item of operational data in the corresponding plurality of feature segments,

[0054] wherein each of the plurality of metric sets comprises at least one metric indicating a degree of correspondence between a signal of the at least one item of operational data in the corresponding plurality of feature segments and a mathematical function;

[0055] selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets;

[0056] determining a predictive asset health state based on a selected signal in at least one of the plurality of characteristic segments; and

[0057] An output is generated based on the predictive asset health.

[0058] 12. An industrial or power system comprising:

[0059] assets; and

[0060] A determination system according to clause 11 to perform predictive health analysis on the asset, optionally wherein the determination system is a decentralized controller of the industrial or power system used to control the asset.

[0061] The following describes exemplary embodiments of the present disclosure. It should be noted that, unless otherwise indicated or apparent, certain aspects of any of the described embodiments may also be present in other embodiments. However, for improved clarity, each aspect will be described in detail only when first mentioned, and repeated descriptions of the same aspect will be omitted.

[0062] These and other aspects and embodiments thereof are described in more detail in the drawings, the description and the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of a method according to one embodiment of the present disclosure is shown.

[0064] Figure 2 Exemplary monitoring data according to one embodiment of the present disclosure is shown.

[0065] Figure 3 A flow chart of a method according to one embodiment of the present disclosure is shown.

[0066] Figure 4 A flow chart of a method according to an embodiment of the present disclosure is shown.

[0067] Figure 5A flow chart of a method according to an embodiment of the present disclosure is shown.

[0068] Figure 6a ) shows exemplary data with multiple signal modes. Figure 6b ) shows the combined with interference signal Figure 6a ) to simulate monitoring data of industrial or power systems under various operating conditions. Figure 6c ) shows Figure 6b ) data, which has multiple feature segments segmented according to an embodiment of the present disclosure.

[0069] Figure 7 The outputs of various stages of the method according to one embodiment of the present disclosure are shown.

[0070] Figure 8 The outputs of the stages of the method according to one embodiment of the present disclosure are shown.

[0071] Figure 9 1 shows the stages of the method according to one embodiment of the present disclosure.

[0072] Figure 10a ) shows a determination system according to an embodiment of the present disclosure. Figure 10b ) shows an industrial or power system according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0073] Figure 1 A flow chart of a method according to an embodiment of the present disclosure is shown. At S101, a signal of at least one item of operation data is acquired. At S102, the signal of at least one item of operation data is segmented into a plurality of feature segments. At S103, a plurality of metric sets are determined for each of the plurality of feature segments based on the signal of at least one item of operation data in the corresponding plurality of feature segments. In one embodiment, each of the plurality of metric sets includes at least one metric indicating a degree of correspondence between the signal of at least one item of operation data in the corresponding plurality of feature segments and a mathematical function. At S104, a signal is selected in at least one of the plurality of feature segments based on the plurality of metric sets, the plurality of feature segments being particularly used to determine a predictive asset health state. At S105, a predictive asset health state is determined based on the signal selected in at least one of the plurality of feature segments. At S106, an output is generated based on the predictive asset health state.

[0074] In one embodiment, segmenting the signal of the at least one item of operational data into a plurality of characteristic segments is based on an operational mode of the asset.

[0075] Those skilled in the art will appreciate that the term "segmentation" may refer to separation, splitting, division, etc. In one embodiment, the segmentation signal may be or may include detecting a transient in the signal and separating the data into different segments based on the detected transient in the signal.

[0076] In one embodiment, the method further comprises at least one of: planning maintenance based on the output; replacing at least one sensor used to monitor data based on the output, in particular, a sensor identified as critical for asset health monitoring; removing at least one sensor used to monitor data that is identified as having no information value based on the output, the at least one sensor; sending the generated output to an end user via a graphical user interface; and optimizing the design of the monitoring system for the asset, in particular, by retaining at least one sensor identified as useful based on the output. In one embodiment, determining the predictive asset health state includes or is determining the RUL. In one embodiment, the method further comprises receiving sensor measurement data captured during operation of the asset; and updating the predictive asset health state based on the received sensor measurement data. In one embodiment, the asset is a power transformer, a distributed energy resource (DER) unit, or a generator.

[0077] Figure 2 Exemplary monitoring data according to an embodiment of the present disclosure is shown. Specifically, first subgraph 210 is a graph of monitoring data characterized by the continuity of the data curve therein, without any transitions or jumps in the data values. In contrast, the data curve shown in second subgraph 220 includes multiple transitions (i.e., step changes), resulting in discontinuities in the data curve. This discontinuity can arise from a variety of reasons, including the various operating modes during the generation of the monitoring data. For example, during an arbitrary time window of 1 ≤ t < 6, the asset operates in a first operating mode; during an arbitrary time window of 6 ≤ t < 11, the asset operates in a second operating mode, and so on. Those skilled in the art will appreciate that the first operation may be temporally separated from the second operation, but for data processing purposes, the data monitored during the first and second operations can be concatenated, e.g., to form a data stream. An asset may operate in a certain operating mode based on operating conditions, which may depend on external factors (e.g., load and / or input) or internal factors (e.g., the maintenance status of the asset). That is, for example, when the load, input, and / or maintenance status of the asset changes, the asset may operate in a different operating mode. According to one embodiment, the monitored data pertains to an asset (particularly a power system asset or an industrial asset) and / or is associated with the health of the asset.

[0078] Figure 3A flow chart of a method according to one embodiment of the present disclosure is shown. Specifically, a high-level view of the method according to one embodiment of the present disclosure is shown. Acquired input data (S310) is processed to determine an operating mode (S320) and segment the acquired input data into multiple segments. In one embodiment, multiple features are extracted from the acquired input data, wherein a feature is a set of values ​​(particularly, a set of values ​​of smaller dimension and / or scale) that describes another set of data (particularly, a set of data of relatively larger dimension and / or scale, i.e., the acquired input data or a segment thereof). Accordingly, in one embodiment, multiple features are processed to determine an operating mode and / or segment the data into multiple segments. Then, based on the detected operating mode, at least one metric is calculated for each of the multiple features (S330), and accordingly, features are selected from the multiple features based on the metrics of the multiple features, particularly the features with the highest metric values ​​(S340). In other words, feature selection is the process of selecting the features with the most informative value for a specific task. Feature selection can also support the selection of an optimal sensing configuration for an asset, so that only sensors that provide valuable data for a specific task can be retained. Those skilled in the art will appreciate that the feature selection method is not limited to the method described herein, and may be any other method. In one embodiment, the feature that best distinguishes different operation modes is identified (S350). Figure 4 An embodiment of the metric calculation performed at S330 is shown in further detail.

[0079] refer to Figure 4 , for each data set acquired during the corresponding detected operating mode (i.e., the data set within the feature segment and / or the features extracted therefrom), the following method is performed: optionally, each data set acquired during the corresponding detected operating mode is normalized (S411) in terms of data value (y-axis) and time (x-axis); a goodness metric is calculated (S413) for each detected operating mode; optionally, a weight for each of the detected operating modes is calculated (S414); a final metric value for each feature on multiple feature segments is calculated (S420); and a goodness metric and / or final metric value is output (S430).

[0080] It should be noted that the metrics included in the metric set can also be referred to as goodness metrics. Accordingly, determining the metric can be equivalent to determining the goodness metric, and thus block S413 corresponds to the following feature: based on the signal of at least one item of operating data in the corresponding plurality of feature segments, determining a plurality of metric sets for each of the plurality of feature segments, wherein each of the plurality of metric sets includes at least one metric indicating a degree of correspondence between the signal of at least one item of operating data in the corresponding plurality of feature segments and a mathematical function.

[0081] In one embodiment, determining a metric means or includes applying input data to a mathematical function to obtain an output of the mathematical function, wherein the output indicates the goodness of the input data for determining a predictive asset health state. The term "goodness" may be synonymous with terms such as "applicability," "usefulness," "applicability," and "significance," and thus may be used interchangeably. The input data may be data operated on or features extracted therefrom. The mathematical function may define at least one feature and / or relationship, and thus its output may indicate the presence or degree of presence (or equivalently, significance) of the at least one defined feature and / or relationship present or potentially present in the input data fed into the mathematical function. Accordingly, in one embodiment, the goodness metric is the output of the mathematical function obtained by processing the input fed into the mathematical function, wherein the goodness metric indicates the presence or degree of presence of the at least one defined feature and / or relationship present or potentially present in the input data fed into the metric, wherein the at least one feature and / or relationship is defined by the mathematical function. The goodness metric may indicate the goodness of the input data for determining a predictive asset health state. A non-exhaustive list of mathematical functions includes correlation, monotonicity, and robustness as described below.

[0082] In one embodiment, each of the plurality of metric sets includes at least one metric indicating a degree of quality of a signal of at least one operational data item in the corresponding plurality of feature segments used to determine a predictive asset health state. In one embodiment, selecting a signal in at least one of the plurality of feature segments is for determining a predictive asset health state based on the plurality of metric sets. In one embodiment, the method further includes combining the plurality of metric sets into a single metric set indicating a degree of quality of a signal of at least one operational data item in the plurality of feature segments used to determine a predictive asset health state.

[0083] In one embodiment, the acquired input data and / or features extracted therefrom are decomposed into a trend part and a residual part as follows:

[0084] X(t k )=X T (t k )+X R (t k ) (1)

[0085] Among them, X(t k ), X T (t k ) and X R (t k ) represent the time t k The value of the input data / feature analyzed at time t kThe trend of the input / feature at time t k The residual part of the input data / feature at the input data / feature. According to one embodiment, the decomposition of the input data / feature includes performing a smoothing process on the signal / feature, for example, by linear local weighted regression, moving average, etc. In one embodiment, a quality metric is calculated based on the decomposed data. For example, the quality metric may include the following correlation:

[0086]

[0087] The quality metric is a calculated correlation value, i.e., the left-hand side of formula (2), which is the output obtained by applying the input data to the mathematical function described by the right-hand side of formula (2), which defines a linear characteristic or relationship between the time points and the values ​​at the time described in the input data. Input data with a highly linear relationship between the time points and the values ​​will produce a high correlation. That is, a quality metric score is obtained for such input data. Similarly, the quality metric can further include the following monotonicity:

[0088]

[0089] The right-hand side of formula (3) evaluates the consistency of the input data in increasing or decreasing, that is, the degree to which the derivative sign of the input data is maintained, and outputs a goodness metric (i.e., the left-hand side of formula (3)). Highly monotonic input data (i.e., a data set whose values ​​are mainly increasing or decreasing over time) will receive a higher score. Therefore, the resulting goodness metric will be high, indicating that the input data highly corresponds to the monotonic function in formula (3). According to one embodiment, the input data can be considered to have a high degree of monotonicity. Similarly, the goodness metric can further include the following robustness:

[0090]

[0091] It is used to evaluate the tolerance of input data to outliers. In formulas (2) to (4), K represents the total number of observations and δ(·) represents a simple unit step function. The weighted measure for the goodness or badness of each operating mode can be determined as follows:

[0092] weightedMetric(X,T)=ω1Corr(X,T)+ω2Mon(X)+ω3Rob(X)+…+ω N Metric N (5)

[0093] Among them, ω iis the i-th weighting factor for scaling each quality metric, and N represents the number of quality metrics considered. Accordingly, the final metric value can be determined as follows:

[0094]

[0095] Among them, θ i It should be understood by those skilled in the art that the calculation of the weighted metric and the final metric is not limited to the linear combination described herein, and other methods may be used to perform the calculation.

[0096] In one embodiment, multiple features are extracted from the data within each of the feature segments. In one embodiment, the features can be logarithmic features, linear features, cubic features, quadratic features, constant features, periodic features, triangular features, square wave features, etc. In one embodiment, a measure of quality is calculated for each of the multiple features of the data within each of the multiple feature segments. Accordingly, the measure of quality of the features can be combined under multiple operating modes. For example, input data having two segments, each having a first feature and a second feature, can be used to calculate correlation and monotonicity according to formula (2) and formula (3), respectively. The correlation of the first feature of the first segment can be combined with the correlation of the first feature of the second segment, and similarly, the monotonicity of the first feature of the first segment can be combined with the monotonicity of the first feature of the second segment. The same operation can be performed on the measure of quality of the second feature. Thus, the measure of quality of each feature under multiple operating modes can be compared. Accordingly, features or subsets of features can be selected based on the measure of quality of each feature under multiple operating modes. Figure 8 An embodiment showing combined goodness-of-badness metrics associated with features is disclosed in .

[0097] Figure 5 A method flow chart according to an embodiment of the present disclosure is shown. Specifically, a feature embodiment selected at S340 is shown with the aid of a flow chart. The final metric value of the feature (e.g., calculated according to formula (6)) is normalized at S510. According to one embodiment, the feature can be classified based on the normalized final metric value. Features with corresponding normalized final metric values ​​exceeding a threshold can be selected at S520 and classified into a subset of optimal features. According to one embodiment, features or a subset of features can be selected based on a threshold without normalization. The optimal subset of features is returned at S530.

[0098] Figure 6a ) shows exemplary data with multiple signal modes. Figure 6b ) shows the combined with interference signal Figure 6a) to simulate monitoring data of industrial or power systems under various operating conditions. Figure 6c ) shows a plurality of feature segments segmented according to an embodiment of the present disclosure Figure 6b ) data. Specifically, Figure 6a) to Figure 6c) Subgraphs 610 to 680 in FIG. 5 show the following characteristics, respectively: constant, linear, quadratic, cubic, logarithmic, periodic, square wave, and triangular. Figure 6a ) shows the original feature data, and Figure 6b ) shows the modified feature data, where 5 operation modes are added to the Figure 6a ), that is, 5 transitions in the data are added to the original feature data of the sub-graphs 610 to 680. Figure 6c ) shows an embodiment based on the present disclosure from Figure 6b ) in sub-graphs 610 to 680, for example, the operation mode detection at S320.

[0099] Figure 7 The outputs of different stages of the method according to one embodiment of the present disclosure are shown. The first sub-graph 710 shows Figure 6b ), and transition points (marked with dots on the data curve) based on the detection of the operating mode according to an embodiment of the present disclosure. The operating mode can be detected based on the abnormality indicator calculated by the multivariate transient detection algorithm, as shown in the second sub-graph 720. Based on the abnormality indicator of the second sub-graph 720, the cubic features of the first sub-graph 710 are segmented into 5 different operating modes. Each of the segmented operating modes is normalized and smoothed to be decomposed into a trend part and a residual part. The third sub-graph 730 shows the normalized raw data of the 5 segments and the corresponding trends. In this example, linear local weighted regression is applied for data smoothing. The fourth sub-graph 740 shows the goodness-of-goodness metric values ​​of formulas (2) to (4) calculated for each operating mode. When the feature increases, the monotonicity metric has a positive value; when the feature decreases, the monotonicity metric has a negative value, thereby indicating the direction of the trend of the feature. The fourth sub-graph 740 also shows the weighted metrics of each operating mode calculated based on the goodness-of-goodness metric of each operating mode. The weighted metrics for the various operating modes can be combined to derive a final metric for the considered feature, for example, according to formula (6).The above feature evaluation is repeated for each of the features.

[0100] Figure 8 1 shows the output of one stage of the method according to an embodiment of the present disclosure. Specifically, Figure 8The combination of goodness and weighted metrics for each of the features is shown. The final weighted metric values ​​for the features indicate that the best feature is the logarithmic feature, followed by the linear feature, the cubic feature, and the quadratic feature. The final weighted metric values ​​can be sorted and selected (e.g., by comparing with a threshold) to form a subset of the best features.

[0101] Figure 9 1 shows the output of one stage of the method according to an embodiment of the present disclosure. Specifically, Figure 9 Importance scores for features presented in FIG6 are shown. The feature importance score can be an output of block S350 and indicates the performance of the feature in distinguishing different operating modes (or equivalently, applicability, usefulness, applicability, significance, etc.). In one embodiment, the feature importance score is used for feature selection, for example, in classification. Accordingly, feature selection algorithms used for classification problems can also be applied here, for example, the ReliefF algorithm. In one embodiment, the feature importance score can be used to identify features that indicate the time for maintenance. For example, operating data obtained before maintenance of a device or system may be very different from operating data obtained after maintenance, that is, there is a transition in the values ​​of the operating data. In this case, the maintenance time can be identified based on the feature importance score or based on the features selected therefrom. Accordingly, the data can be separated into data before maintenance and data after maintenance based on the feature importance score or based on the features selected therefrom. In one embodiment, the feature importance score or based on the features selected therefrom can be used to determine a predictive asset health state.

[0102] Figure 10a )and Figure 10b ) shows a determination system and an industrial or power system according to an embodiment of the present disclosure. Specifically, Figure 10a ) shows a determination system according to an embodiment of the present disclosure. The determination system 1010 includes at least one integrated circuit 1011, which is operable to execute any of the above embodiments. Figure 10b ) shows an industrial or power system according to an embodiment of the present disclosure. The industrial or power system 1020 includes assets 1030 and a determination system 1010.

[0103] Although various embodiments of the present disclosure have been described above, it should be understood that they are presented by way of example only and not by way of limitation. Similarly, various figures may illustrate example architectures or configurations for the purpose of enabling those of ordinary skill in the art to understand the exemplary features and functions of the present disclosure. However, it should be understood by those skilled in the art that the present disclosure is not limited to the example architectures or configurations shown, but may be implemented using a variety of alternative architectures and configurations. Additionally, as will be understood by those of ordinary skill in the art, one or more features of an embodiment may be combined with one or more features of another embodiment described herein. Therefore, the breadth and scope of the present disclosure should not be limited by any of the exemplary embodiments described above.

[0104] It should also be understood that any reference to an element herein using references such as "first," "second," etc., does not limit the number or order of those elements in general. Rather, these references may be used herein as a convenient way to distinguish between two or more elements or instances of an element. Thus, a reference to a first element and a second element does not mean that only two elements may be used, or that the first element must precede the second element in some manner.

[0105] Additionally, those skilled in the art will appreciate that information and signals may be represented using a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, and symbols that may be referenced in the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[0106] Those skilled in the art will also understand that the various exemplary logical blocks, units, processors, devices, circuits, methods, and functions described in conjunction with the various aspects disclosed herein may be implemented by electronic hardware (e.g., digital implementation, analog implementation, or a combination of the two), firmware, various forms of programs or design codes including instructions (for convenience, referred to herein as "software" or "software units"), or any combination of these technologies.

[0107] To clearly illustrate this interchangeability of hardware, firmware, and software, various exemplary components, blocks, units, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware, firmware, or software, or a combination of these technologies, depends on the specific application and the design constraints imposed on the entire system. Technicians can implement the described functionality in various ways for each specific application, but such implementation decisions do not depart from the scope of this disclosure. According to various embodiments, processors, devices, components, circuits, structures, machines, units, etc. can be configured to perform one or more functions described herein. As used herein, "configured to" or "configured for" with respect to a particular operation or function refers to a processor, device, component, circuit, structure, machine, unit, etc. that is physically constructed, programmed, and / or arranged to perform a particular operation or function.

[0108] In addition, those skilled in the art will understand that the various exemplary methods, logic blocks, units, devices, components and circuits described herein may be implemented or executed within an integrated circuit (IC), which may include a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, or any combination thereof. The logic blocks, units and circuits may also include an antenna and / or a transceiver to communicate with various components within a network or within a device. The general-purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller or state machine. The processor may also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other suitable configuration for performing the functions described herein. If implemented in software, the functions may be stored as one or more instructions or codes on a computer-readable medium. Therefore, the steps of the methods or algorithms disclosed herein may be implemented as software stored on a computer-readable medium.

[0109] Computer-readable media include computer storage media and communication media, including any medium that can transfer a computer program or code from one place to another. Storage media can be any available medium that can be accessed by a computer. By way of example (but not limitation), such computer-readable media can include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired program code in the form of instructions or data structures and that can be accessed by a computer.

[0110] Additionally, memory or other storage and communication components may be used in embodiments of the present disclosure. It should be understood that, for clarity, the above description describes embodiments of the present disclosure with reference to different functional units and processors. However, it will be apparent that any suitable distribution of functionality may be employed between different functional units, processing logic elements, or domains without departing from the scope of the present disclosure. For example, functions shown as being performed by separate processing logic elements or controllers may be performed by the same processing logic element or controller. Therefore, references to specific functional units are merely references to suitable devices that provide the functionality, rather than indicating a strict logical or physical structure or organization.

[0111] Those skilled in the art will readily appreciate various modifications to the implementations described in this disclosure, and the general principles defined herein may be applied to other implementations without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the implementations shown herein, but rather should be accorded the widest scope consistent with the novel features and principles disclosed in this disclosure as set forth in the following claims.

Claims

1. A method for performing predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the method comprising: a signal for obtaining at least one item of operational data; Segmenting the signal of the at least one item of operational data into a plurality of characteristic segments; determining a plurality of metric sets for each of the plurality of feature segments based on a signal of the at least one item of operational data in a corresponding plurality of feature segments, wherein each of the plurality of metric sets comprises at least one metric indicating a degree of correspondence between a signal of the at least one item of operational data in the corresponding plurality of feature segments and a mathematical function; selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets; determining a predictive asset health state based on a selected signal in at least one of the plurality of feature segments; and An output is generated based on the predictive asset health.

2. The method according to claim 1, further comprising at least one of the following: planning maintenance based on the output; replacing at least one sensor used for monitoring data based on the output, in particular, a sensor identified as being critical to the health monitoring of the asset; removing at least one sensor for monitoring data whose output is identified as having no information value based on the output; sending the generated output to an end user via a graphical user interface; as well as The design of a monitoring system for the asset is optimized, in particular, by retaining at least one sensor identified as useful based on the output.

3. The method according to claim 1 or 2, wherein: Determining the predictive asset health state includes determining a RUL, or determining a RUL.

4. The method according to any one of the preceding claims, further comprising: Decomposing a signal of at least one item of operation data in the plurality of feature segments into at least a trend portion and a residual portion; and Wherein, determining a plurality of metric sets for each of the plurality of feature segments is based on the trend portion of the decomposed signal of at least one item of operational data in the plurality of feature segments.

5. The method of any preceding claim, further comprising combining the plurality of metric sets into a single metric set, the single metric set indicating a degree of correspondence between the signal of at least one item of operational data in the plurality of feature segments and the mathematical function. The method of claim 5 , further comprising normalizing the plurality of metric sets in amplitude and / or time.

7. The method according to any one of the preceding claims, further comprising: receiving sensor measurement data captured during operation of the asset; as well as The predictive asset health status is updated based on the received sensor measurement data.

8. A method according to any preceding claim, wherein: The asset is a power transformer, a distributed energy resource DER unit, or a generator.

9. A method according to any preceding claim, wherein: Segmenting the signal of the at least one item of operational data into a plurality of characteristic segments is based on an operational mode of the asset.

10. A method of operating and / or maintaining an asset, in particular a power system asset or an industrial asset, comprising: performing a predictive health analysis on the asset using any of the methods described above; as well as Automatically perform at least one of: scheduling downtime for the asset based on the determined predictive asset health; scheduling maintenance work based on the determined predictive asset health; scheduling replacement work based on the determined predictive asset health; changing maintenance intervals based on the determined predictive asset health.

11. A determination system operable to perform predictive health analysis on an asset, in particular for determining the remaining useful life (RUL) of a power system asset or an industrial asset, the determination system comprising at least one integrated circuit operable to: a signal for obtaining at least one item of operational data; Segmenting the signal of the at least one item of operational data into a plurality of characteristic segments; determining a plurality of metric sets for each of the plurality of feature segments based on a signal of the at least one item of operational data in the corresponding plurality of feature segments, in, each of the plurality of metric sets comprising at least one metric indicating a degree of correspondence between a signal of the at least one item of operational data in the corresponding plurality of feature segments and a mathematical function; selecting a signal in at least one of the plurality of feature segments based on the plurality of metric sets; determining a predictive asset health state based on a selected signal in at least one of the plurality of characteristic segments; and An output is generated based on the predictive asset health.

12. An industrial or power system comprising: assets; as well as The determination system of claim 11 to perform predictive health analysis on the asset, optionally wherein the determination system is a decentralized controller of the industrial or power system used to control the asset.

Images