Analysis of main causes of anomalies in turbomachines
A simplified computer-implemented method for analyzing turbomachine anomalies using comparative analysis of characteristic importance values addresses the challenge of identifying root causes in turbomachines, enhancing efficiency and reliability in anomaly detection.
Patent Information
- Authority / Receiving Office
- RU · RU
- Patent Type
- Patents
- Current Assignee / Owner
- NUOVO PIGNONE TECH SRL
- Filing Date
- 2023-12-22
- Publication Date
- 2026-06-30
AI Technical Summary
Identifying and determining the root cause of anomalies in turbomachines, particularly in turbomachines for oil and gas applications, is challenging due to their complexity and the difficulty in achieving efficiency and reliability without extensive testing and training.
A computer-implemented method for analyzing the root cause of turbomachine anomalies through comparative analysis of characteristic importance values in anomalous and non-anomalous time intervals, using a simplified approach that does not require prior knowledge or training, involving the use of performance sensors and turbomachine characteristics.
Enables efficient and reliable identification of the root cause of turbomachine anomalies without extensive testing or training, facilitating timely actions such as alerts or system adjustments.
Smart Images
Figure 00000001_ABST
Abstract
Description
AREA OF TECHNOLOGY
[0001] The subject matter of the invention described herein relates to the analysis of the root causes of anomalies in turbomachines.BACKGROUND OF THE INVENTION
[0002] Even if a turbomachinery is designed to avoid abnormalities and maintenance is carried out to prevent them, abnormalities will still occur in turbomachinery during operation.An anomaly is a situation in a machine that is far from standard functioning; the first example of an anomaly is a vibration at a certain location in a machine, the amplitude of which exceeds the normal vibration amplitude at that location; the second example of an anomaly is the rotational speed of a certain component of the machine, which exceeds the normal amplitude of the rotational speed of that component; the third example of an anomaly is a temperature at a certain location in a machine, the value of which exceeds the normal temperature at that location; the fourth example of an anomaly is a pressure in a certain channel or cavity of the machine, the value of which exceeds the normal pressure in that channel or cavity; the fifth example of an anomaly is a flow in a certain channel of the machine, the value of which exceeds the normal flow in that channel or cavity.The expression "far from" should be understood to mean that the difference between a standard value, such as a nominal value, and the actual value is greater than a specified difference, such as a specified percentage difference; such a specified difference typically differs depending on the parameter and may also depend, for example, on the operating state of the machine.
[0003] Root cause analysis of turbomachinery anomalies, i.e., identifying the cause of a turbomachinery anomaly that occurred in the past or is currently being observed, is crucial for both the manufacturer and the user, but it is extremely difficult to perform reliably. The complexity of turbomachinery operation, such as a compressor or turbine, and the conditions in which it is installed and operated, such as in the oil and gas industry, further complicate this task. In some cases, even accurately identifying the anomaly is difficult.
[0004] Technical literature describes computer-implemented methods and computer systems aimed at fully automatic anomaly detection in machines. Some anomalies are relatively easy to identify, while others are difficult. Effective and reliable execution typically requires extensive testing and training for each machine of interest. In-depth knowledge of the machine of interest and the conditions in which it is installed and operating is generally a significant advantage for a reliable solution.
[0005] Similarly, technical literature describes computer-implemented methods and computer systems that are aimed at fully automatic detection of the root causes of anomalies in machines. This task is more challenging, and achieving efficiency and reliability is much more difficult. Even then, solving the problem may require extensive testing and training.
[0006] Therefore, it would be desirable to have a simpler method for analyzing the root causes of anomalies in turbomachines, particularly turbomachines for oil and gas applications, without sacrificing efficiency and reliability. SUMMARY OF THE INVENTION
[0007] According to a first aspect, the subject matter of the invention described herein relates to a computer-implemented method for analyzing the root cause of an anomaly in a turbomachine, the method comprising the steps of: a) receiving measurement data from a plurality of performance sensors mounted on a turbomachine related to a time interval, b) receiving an identifier of a target performance sensor in whose measurement data an anomaly is detected, c) receiving a start time and an end time of a non-anomalous time sub-interval when the anomaly does not occur, d) receiving a start time and an end time of an abnormal time sub-interval when the anomaly occurs, e) receiving identifiers from a plurality of performance sensors associated with turbomachine characteristics that may represent the root causes of the anomaly, and f) obtaining at least one turbomachine characteristic that should be considered the root cause of the anomaly,Based on a "comparative analysis" of the "characteristic importance" values of the characteristic sensors in the non-anomalous and anomalous time intervals. The terms "comparative analysis" and "characteristic importance" will be explained further in the detailed description.
[0008] According to other aspects, the subject matter of the invention described herein relates to a computer system and a turbomachine arrangement in which such a method is implemented. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The embodiments of the invention described and many of the attendant advantages presented herein can be more fully appreciated and understood by studying the following detailed description taken in conjunction with the accompanying drawings, wherein: Fig. 1 shows a basic block diagram of an embodiment of the arrangement of a turbomachine according to the invention, comprising the system of the invention, Fig. 2 shows a block diagram of an embodiment of the method of the invention for analyzing the root cause of an anomaly in a turbomachine, and Fig. 3 shows a block diagram of a possible embodiment of one particular step of the method shown in Fig. 2. DETAILED DESCRIPTION OF EMBODIMENTS
[0010] As explained above, identifying anomalies in a turbomachine is difficult, and determining their root cause is even more challenging if a reliable result is required. Therefore, the goal was to limit the problem to a simplified, yet still complex, problem, while avoiding preliminary testing and training on a machine or machines. It is assumed that A) the anomalies are identified, i.e., the anomalous periods and non-anomalous periods are known, and B) the possible causes of the anomaly are known. The task is to select the best cause, i.e., the cause that is most likely to be the true root cause of the anomaly.Typically, the number of possible causes of an anomaly is large, ranging from 10 to 100, and depends on the specific anomaly; for example, higher-than-normal vibration may be caused by an abnormal flow rate in any of a set of channels, an abnormal pressure value in any of a set of channels, or an abnormal rotational speed of any of a set of components; the task is to select which of the channel flows, or which channel pressure, or which component rotational speed has caused or is causing the anomaly identified at a specific time. According to the subject matter of the invention described herein, this task is solved by "comparative analysis," i.e., by comparing anomalous and non-anomalous periods, and does not require any prior knowledge of any anomaly, in particular, does not require prior training.
[0011] Reference will now be made in detail to embodiments of the present disclosure, examples of which are shown in the drawings. The examples and drawings are provided to clarify the disclosure and are not to be construed as limiting the disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made within the scope or spirit of the disclosure. In the following description, like reference numerals are used in the figures illustrating the embodiments to indicate elements that perform the same or similar functions. Furthermore, for clarity of illustration, some reference numerals may not be repeated in all figures.
[0012] In Fig. 1, an embodiment of an arrangement 100 of a turbomachine according to the invention is shown very schematically together with its user 10, who interacts with the arrangement 100. The arrangement 100 comprises a turbomachine 180 and a computer system 140 according to the invention. The user 10 may be an employee of the company that manufactured the turbomachine 180, or an employee of the company responsible for testing the turbomachine 180, or an employee of the company that manages the installation in which the turbomachine 180 is installed. More generally, the user 10 is a person or group of persons interested in determining the root cause of an anomaly that has occurred or is occurring in the turbomachine 180; such a need may be repeated from time to time in the event of any new anomaly.
[0013] The arrangement 100 and its variants will be described in more detail below. It is now important to assume that the turbomachine 180 comprises a set of characteristic sensors 182 mounted thereon; the number of sensors 182 is significant, for example 100-1000; the sensors 182 repeatedly measure the "characteristics" (which may also be called "variables") of the turbomachine, such as, for example, temperatures, pressures, volumetric and mass flows, displacements, speeds (e.g., rotational speed), accelerations, vibrations, valve opening levels, commanded angular positions of the IGV, detected angular positions of the IGV, gas compositions, burner states. Typically, the sensors 182 are "actual" sensors, i.e., devices that perform a measurement inside the turbomachine and determine / output an (analog or digital) signal, the amplitude of which corresponds to the measured value.Alternatively, in accordance with the subject matter of the invention described herein, one or more sensors may be so-called "virtual" sensors; as is known, a "virtual" sensor is a piece of software running on a computer (this may be the same computer executing the method of the invention) that repeatedly calculates, for example, a formula, using data from one or more "actual" sensors as input and producing data from the "virtual" sensor as output, as if the machine had an "actual" sensor built in instead of a "virtual" sensor.
[0014] System 140 receives measurement data from sensors 182 in some manner. Fig. 1 shows arrows between turbomachine 180 and system 140, which can be interpreted as (wired or wireless) connection(s), so that measurement data is received directly from turbomachine 180. However, according to some embodiments, measurement data can be collected, for example, by a computer system (not shown in Fig. 1) at a certain time, and then transmitted to computer system 140 at a later time via a (wired or wireless) computer connection or via a data storage device only for analysis.
[0015] Fig. 2 shows a flow chart 200 of an embodiment of the inventive method for analyzing the root cause of an anomaly in a turbomachine, such as the turbomachine 180 shown in Fig. 1; this is a computer-implemented method that can be implemented, for example, by the computer system 140 shown in Fig. 1. The flow chart path goes from the START block 210 to the END block 280 for each identified anomaly - in general, it should be expected that during the operation of the turbomachine, as a rule, several anomalies occur one after another; therefore, it can be repeated, for example, during a study or inspection of the turbomachine (i.e., in an offline mode) or during a test of the turbomachine or during the operation of the turbomachine.
[0016] The method according to the invention comprises the following steps:a) receiving (block 220) data from a set of characteristic sensors installed on a turbomachine, wherein the data corresponds to measurements performed by the characteristic sensors of the set of characteristic sensors in a time interval,b) receiving (block 230) an identifier of a characteristic sensor of the set of characteristic sensors, wherein an anomaly appears in the measurement data from at least said characteristic sensor, and said characteristic sensor is a target characteristic sensor for the anomaly,c) receiving (step 240) a first start time and a first end time of a first sub-interval, wherein the first sub-interval is contained in the above-mentioned time interval, wherein the anomaly does not appear in the first sub-interval,d) receiving (block 250) a second start time and a second end time of the second sub-interval,wherein the second time sub-interval is contained within the above-mentioned time interval, wherein the anomaly occurs in the second time sub-interval, e) receiving (block 260) identifiers of a plurality of characteristic sensors from a set of characteristic sensors associated with characteristics of the turbomachine that may represent the main causes of the anomaly, and f) obtaining (block 270) at least one characteristic of the turbomachine that should be considered the main cause of the anomaly, based on the "comparative analysis" of the "characteristic importance" values of the characteristic sensors from the plurality of characteristic sensors in the first time sub-interval and in the second time sub-interval.
[0017] “Benchmarking” in step “f” means comparing abnormal periods and non-anomalous periods, in particular the “characteristic importance” in abnormal periods and non-anomalous periods; step “f” will be better explained later with the help of Fig. 3. The term “characteristic importance” means the degree or level of influence of an input characteristic, i.e., a variable, on an output characteristic, i.e., a variable. If a system is viewed as a black box with several input variables and one output variable, any specific output value can be considered in the context of the influence of all input values; however, each input value could have different influence on a specific output value. A system can be associated with a “predictive model”, which is usually very complex, and with an “explanatory model”, which should be quite simple to understand.A very effective type of "explanatory model" is a linear function of binary variables, allowing for the implementation of "additive feature attribution methods." More detailed information on this topic can be found, for example, in the article by M. Lundberg and Su-In Lee, "A unified approach to interpreting model predictions," in the Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS 2017).
[0018] In particular, at step "f" at least one characteristic is selected from a plurality of characteristics, i.e. characteristics associated with a plurality of characteristic sensors mentioned in step "e". In particular, at step "f" the difference value of the characteristic importance of each characteristic sensor from the plurality of characteristic sensors is calculated and the highest difference value of the characteristic importance is determined from it. The reason why more than one characteristic can be obtained at step "f" will be explained later and is related to the fact that the method according to the invention can be used as an aid to a person (or a group of people, for example technical experts), so that the final decision, which is the true root cause, can be made, for example, by such a person (or a group of people, for example technical experts) also based on his experience or may require additional testing and / or research.
[0019] The "comparative analysis" in step "f" according to the subject matter of the invention described in this document comprises: - calculating the importance value of the characteristic of each characteristic sensor from said plurality of characteristic sensors, in particular as a contribution of the characteristic to the regression of the target variable, in said first time sub-interval and in said second time sub-interval, - calculating the difference value of the importance of the characteristic of each characteristic sensor from said plurality of characteristic sensors, wherein the difference value of the importance of the characteristic is the difference between the importance value of the characteristic in said first time sub-interval and the importance value of the characteristic in said second time sub-interval, and - determining the high or highest difference value of the importance of the characteristic between the calculated difference values of the importance of the characteristic.
[0020] Preferably, the importance values of the characteristics are calculated based on a model (usually an explanatory model) of the turbomachine, which is a linear function of binary variables, where the binary variables correspond to the characteristics of the turbomachine corresponding to a specified set of characteristic sensors, as mentioned, for example, in the already cited article by Scott M. Lundberg and Su-In Lee.
[0021] According to some exemplary embodiments, one or more or all of the "receiving" stages, i.e., stages "b" and "c", as well as "d" and "e", include receiving input data from a user, such as user 10 shown in Fig. 1. This is particularly true for stages "b" and "e", i.e., identifying the target characteristic and identifying possible characteristics of the root cause.
[0022] According to some typical embodiments, the occurrence of an anomaly is assessed by a human based on human observation of the turbomachine and, for example, its measurement data.
[0023] Typically, the second time sub-interval, i.e., the "anomalous" time sub-interval, follows the first time sub-interval, i.e., the "non-anomalous" time sub-interval. Indeed, an anomaly may begin even at the end of the first time sub-interval, but its effects are not obvious in the first time sub-interval, even at the end of the first time sub-interval. Even if a person misjudges the situation, the inventive method still yields good results if only a few fragments of measurement data have been collected in the first time sub-interval at the time the anomaly occurs.
[0024] In the above steps “b” and “e”, reference is made to “identifiers” as a means for identifying the sensors and characteristics of the turbomachine.
[0025] In steps "c" and "d" above, reference is made to "start time" and "end time" as means for identifying "time subintervals." Similarly, a "time subinterval" can be identified, for example, by "start time" and "duration" or "end time" and "duration."
[0026] According to advantageous embodiments, such as the embodiment shown in Fig. 3, step "f" includes the following sub-steps: f1) creating (block 272) a model (typically an explanatory model) of a turbomachine based on the received measurement data (in particular, the measurement data in a first time sub-interval, i.e., the "non-anomalous" time sub-interval), wherein the model has as input data at least the characteristics of the turbomachine corresponding to a plurality of characteristic sensors, and as output data at least a characteristic of the turbomachine corresponding to a target characteristic sensor, f2) calculating (block 274) an importance value of the characteristic of each characteristic sensor of the plurality of characteristic sensors separately in the first time sub-interval and the second time sub-interval with respect to the measurement data from at least the target characteristic sensor based on the model created in sub-step "f1",f3) calculating (block 275) a difference value of the importance of the characteristic of each characteristic sensor of the plurality of characteristic sensors, wherein the difference value of the importance of the characteristic is the difference between the importance value of the characteristic in the first time sub-interval and the importance value of the characteristic in the second time sub-interval, based on the importance values of the characteristic calculated in sub-step "f2", f4) determining (block 276) the highest difference value of the importance of the characteristics between the difference values of the importance of the characteristics calculated in sub-step "f3", and f5) obtaining (block 278) from the highest difference value of the importance of the characteristics determined in sub-step "f4", the associated characteristic of the turbomachine, which should be considered the main cause of the anomaly.
[0027] The model in the "f1" substage is primarily a regression model, which can be implemented most preferably through a recurrent neural network, in particular an LSTM (Long Short-Term Memory) type neural network. Such a model, in particular a neural network, is trained to predict the target characteristic based on the measured data of the input characteristic sensors using the data received in stage "a."
[0028] According to advantageous embodiments, sub-step "f2" is implemented using an explainability method applied on top of the model created in sub-step "f1" and based on a method related to Shapley values, in particular SHAP values (see, for example, the already cited article by Scott M. Lundberg and Su-In Lee, which provides general explanations of methods related to Shapley values and a specific description of SHAP values). This explainability method originates from game theory and basically assigns to each feature a value corresponding to the change in the expected prediction of the model when this feature is conditioned. The basic procedure underlying this explainability method is to retrain the model on all possible subsets of features S of F, where F is the set of all features, and assign an importance value to each feature that reflects the effect of including this feature on the prediction of the model.To calculate this effect, a model is trained on the present features, and another model is trained on the retained features. Since the retention effect of a feature depends on other features in the model (the collinearity effect), prior differences are calculated for all possible subsets of features. Shapley values are then calculated and used as feature attributes. They represent a weighted average of all possible differences in predictions obtained with and without a particular feature. This calculation is typically approximated to speed up the process.
[0029] Preferably, the operating conditions of the turbomachine in the first sub-interval and the operating conditions of the turbomachine in the second sub-interval are similar. The similarity may be based on the values of the input characteristics; for example, similar conditions may mean that the compressor is within the same rotation speed range, and / or within the same suction pressure range, and / or within the same discharge pressure range. The similarity may be based on the values of the output characteristics, i.e., on the effects; the same conditions may mean that, if there should be no anomaly, the target characteristic should have the same value or be within the same range of values.From a practical point of view, similarity can be based on temporal proximity; similarity is probable if the first time sub-interval and the second time sub-interval are consecutive or close in time to each other (e.g., the temporal distance is less than 10%, or 20%, or 50%, or 100% of the duration of the first time sub-interval or the second time sub-interval).
[0030] According to some advantageous embodiments: in sub-step "f4", a set of the highest characteristic importance difference values is determined, and in sub-step "f5", the characteristic importance difference values from the set of the highest characteristic importance difference values are ranked and the associated turbomachine characteristics are accordingly ranked as the root cause of the specified anomaly. Thus, some characteristics are provided as possible root causes of the anomaly, and they are ordered according to the likelihood that they are the true root cause.
[0031] According to some advantageous embodiments, in the sub-step “f5,” a confidence value is determined for determining the root cause of the anomaly. Specifically, the sub-steps “f1” and “f2,” as well as “f3,” “f4,” and “f5” are repeated based on different initialization of the weight values of the neural network; for each repetition, a distinct ranking of the features is obtained; for each feature, the average value of its ranked position and the variance of its ranked position are determined; wherein the confidence value of the feature, which represents the root cause of the analysis, is the inverse of its variance.
[0032] The method for analyzing the root cause of an anomaly in a turbomachine described and claimed in this document can be implemented using a computer system, such as system 140 in Fig. 1, configured to implement the method. The system may comprise essentially a processor, such as processor 142 shown in Fig. 1, a memory device, such as memory device 146 shown in Fig. 1, connected to the processor 142 and configured to store a program and data, and a human input / output interface, such as human input / output interface 144 connected to the processor 142. These components 142, 144 and 146 are key components of the computer; thus, system 140 can be, for example, a so-called "workstation" or a so-called "server" or even a so-called "cluster" of computers. To implement the method according to the invention, a corresponding computer program is stored in the memory device.To implement the method, input data from user 10 is received from the human input / output interface and sent to the processor. As already described, the system according to the invention is adapted to receive measurement data from a turbomachine in any manner (see, for example, the arrows in Fig. 1). A typical embodiment is that system 140 comprises a database 148 for storing data, in particular measurement data, from one or more turbomachines. Transmission of measurement data from turbomachines to a computer located remotely from the turbomachines and their storage in a database located within the computer or connected to the computer are known in the art and are beyond the scope of the protection of the present patent application.
[0033] According to some embodiments, the system of the invention is configured to implement the method of the invention in an offline mode. In other words, the measurement data can be transferred from the turbomachines to the database well in advance (e.g., one hour, one day, or one month) of processing in accordance with the method described herein.
[0034] According to other embodiments, the system of the invention, such as the system 140 shown in Fig. 1, is configured to perform the method of the invention during operation of a turbomachine (or turbomachines), such as the turbomachine 180 in Fig. 1.
[0035] According to possible embodiments, the system of the invention, for example the system 140 shown in Fig. 1, may be configured to automatically identify one or more anomalies (not necessarily all) or one or more types of anomalies (not necessarily all) in a turbomachine, for example the turbomachine 180 shown in Fig. 1. For example, a corresponding portion of software may be stored in a memory device of the system, for example the memory device 146 shown in Fig. 1, to perform such a task and provide information to the portion of software implementing the method of the invention. The corresponding portion of software may also be capable of identifying an “anomalous” time sub-interval and a “non-anomalous” time sub-interval and providing them to the portion of software implementing the method of the invention.
[0036] As shown in Fig. 1, the system of the invention can be integrated into a turbomachine arrangement, such as the arrangement 100 shown in Fig. 1. Such a system essentially comprises a turbomachine, such as the turbomachine 180 shown in Fig. 1, and a system of the invention, such as the system 140 shown in Fig. 1.
[0037] Understanding the root cause of an anomaly in a turbomachine is useful not only for gaining more knowledge / more accurate knowledge about this turbomachine and / or for better designing this turbomachine or a similar turbomachine (in particular, one or more components of the turbomachine). In fact, based on the identified root cause (or causes), according to some embodiments, it is possible to initiate an alert / alarm (audible or visual) for the user during turbomachine operation (in the field or during testing) and / or take certain actions, for example, via a computer controlling the turbomachine or one or more subsystems that are part of the turbomachine or connected to the turbomachine. The action may be motivated, for example, by safety concerns or the goal of ensuring efficiency.The action may be, for example, the shutdown of a turbomachine, or the activation of a safety system or subsystem, or a change in the control of a component (e.g. opening or closing a valve).
Claims
1. A computer-implemented method for analyzing the root cause of an anomaly in a turbomachine, including the following stages: a) receiving (220) data from a set of characteristic sensors installed on a turbomachine, wherein the data correspond to measurements performed by the characteristic sensors from the specified set of characteristic sensors in a time interval, b) receiving (230) an identifier of a characteristic sensor from said set of characteristic sensors, wherein said anomaly appears in measurement data from at least said characteristic sensor, and said characteristic sensor is a target characteristic sensor for said anomaly, c) receiving (240) a first start time and a first end time of a first time sub-interval, wherein the first time sub-interval is contained within the specified time interval, wherein the specified anomaly does not occur in the specified first time sub-interval, d) receiving (250) a second start time and a second end time of a second time sub-interval, wherein the second time sub-interval is contained within the specified time interval, wherein the specified anomaly occurs in the second time sub-interval, e) receiving (260) identifiers of a plurality of characteristic sensors from said set of characteristic sensors associated with characteristics of the turbomachine that may represent the root causes of said anomaly, and f) obtaining (270) at least one characteristic of the turbomachine, which should be considered the main cause of said anomaly, based on a comparative analysis of the importance values of the characteristics from the characteristic sensors from said plurality of characteristic sensors in said first time sub-interval and in said second time sub-interval; wherein the said comparative analysis at stage “f” (270) includes: - calculating the importance value of the characteristic of each characteristic sensor from the specified set of characteristic sensors in the specified first time sub-interval and in the specified second time sub-interval, - calculating the difference value of the importance of the characteristic of each characteristic sensor from the specified plurality of characteristic sensors, wherein the difference value of the importance of the characteristic is the difference between the importance value of the characteristic in the specified first time sub-interval and the importance value of the characteristic in the specified second time sub-interval, and - determination of the highest or highest value of the difference in the importance of the characteristic between the calculated values of the difference in the importance of the characteristic.
2. The method according to claim 1, in which the importance values of the characteristic are calculated on the basis of a turbomachine model, which is a linear function of binary variables, wherein the binary variables correspond to the characteristics of the turbomachine corresponding to the specified set of characteristic sensors.
3. The method according to claim 1, wherein step “f” includes the following sub-steps: f1) creating (272) a model of a turbomachine based on said measurement data, wherein said model has as input data at least characteristics of a turbomachine corresponding to said plurality of characteristic sensors, and as output data at least a characteristic of a turbomachine corresponding to said target characteristic sensor, f2) calculating (274) the importance value of the characteristic of each characteristic sensor from said plurality of characteristic sensors in said first time sub-interval and in said second time sub-interval relative to the measurement data from at least said target characteristic sensor based on said model created in sub-stage “f1”, f3) calculating (275) the difference value of the importance of the characteristic of each characteristic sensor from said plurality of characteristic sensors, wherein the difference value of the importance of the characteristic is the difference between the importance value of the characteristic in said first time sub-interval and the importance value of the characteristic in said second time sub-interval, based on the importance values of the characteristic calculated in sub-step “f2”, f4) determining (276) the highest value of the difference in the importance of the characteristic between the values of the difference in the importance of the characteristic calculated in sub-stage “f3”, and f5) obtaining (278) from the highest value of the difference in the importance of the characteristic determined in sub-stage “f4”, the associated characteristic of the turbomachine, which should be considered the main cause of the said anomaly.
4. The method according to claim 3, wherein said model at sub-stage “f1” (272) is a regression model.
5. The method according to claim 4, in which the said model at substage “f1” (272) is implemented through a recurrent neural network.
6. The method according to claim 5, in which the said model at substage “f1” (272) is implemented through an LSTM type neural network.
7. The method according to claim 1, wherein the sub-step “f2” (274) is performed by means of methods related to Shapley values.
8. The method according to claim 1, wherein the operating conditions of the turbomachine in said first time sub-interval and the operating conditions of the turbomachine in said second time sub-interval are similar.
9. The method according to claim 3, wherein at sub-step “f4” (276) a set of the highest values of the difference in the importance of the characteristics is determined, and wherein at sub-stage “f5” (278) the values of the difference in the importance of the characteristics of the specified set of the highest values of the difference in the importance of the characteristics are ranked and, accordingly, the associated characteristics of the turbomachine are ranked as the main causes of the specified anomaly.
10. The method according to claim 3, wherein at sub-step “f5” (278) a confidence value is determined for determining the root cause of the said anomaly.
11. A computer system (140) configured to implement a computer-implemented method for analyzing the root cause of an anomaly in a turbomachine according to claim 1.
12. The computer system (140) according to claim 11, configured to implement the method according to claim 1 during operation of the turbomachine (180).
13. The computer system (140) according to claim 12, configured to initiate an alert / alarm based on the root cause identified by said method.
14. The computer system (140) according to claim 12, configured to perform actions on the turbomachine (180) or on one or more subsystems that are part of the turbomachine (180) or connected to the turbomachine (180), based on the root cause identified by the said method.
15. A turbomachine assembly (100) comprising a turbomachine (180) and a computer system (140) according to claim 11.
16. A turbomachine assembly (100) comprising a turbomachine (180) and a computer system (140) according to claim 12.