WO2020253950A1

WO2020253950A1 - Monitoring method, predicting method, monitoring system and computer program

Info

Publication number: WO2020253950A1
Application number: PCT/EP2019/066089
Authority: WO
Inventors: Francis FOMI WAMBA; Thomas Seitz; Gerrit GLOTH
Original assignee: Framatome GmbH
Current assignee: Framatome GmbH
Priority date: 2019-06-18
Filing date: 2019-06-18
Publication date: 2020-12-24
Anticipated expiration: 2021-12-18

Abstract

A monitoring method for monitoring at least one plant component of a power plant comprises the steps: - Receiving asset data associated with the plant component of the power plant; and - Processing the asset data received with at least one analyzing module (36) that applies machine learning and/or statistical methods, thereby generating analysis results (50), wherein the analysis results (50) comprise information on a condition of the plant component of the power plant. Further, a predicting method, a monitoring system (10) and a computer program are described.

Description

Monitoring method, predicting method, monitoring system and computer program

The invention relates to a monitoring method as well as a monitoring system for monitoring at least one plant component of a power plant. Further, the invention relates to a prediction method of predicting at least one condition of at least one plant component of a power plant. In addition, the invention relates to a computer program.

In power plants known in the state of the art, for instance nuclear power plants, Instrumentation and Control (l&C) systems, Condition Monitoring Systems (CMS) and Computerized Maintenance Management Systems (CMMS) are the key tools for the safe and reliable operation of the power plants. These systems each monitor respective plant components in order to obtain data or rather information about conditions of the plant components.

Typically, monitoring in power plants is only done for critical equipment of the power plant such as turbines and main coolant pumps. The amount of collected data is reduced to the most known fault indicators like effective and peak values of the parameters monitored according to the corresponding standards, for instance ISO 10816 and ISO 7919, as well as according to expert knowledge. Furthermore, the data gathered is stored separately and analyzed by different experts or departments separately.

Accordingly, it is known in the state of the art to monitor alarm thresholds of time-domain vibration values and to perform frequency-domain analyses from time to time to follow slow changes in the vibration behavior of the respective plant components, particularly rotating plant components. However, not all available information is used for monitoring purposes, as the current systems are specialized, for instance on vibration data. Hence, the specialized systems cannot efficiently combine their data to be monitored with information or rather data from other sources, particularly other systems, in order to obtain deeper insights of the power plant. This drawback of the current systems results in many different and separate monitoring systems that only deal with specific plant components or conditions to be monitored.

In addition, the quality of an associated diagnosis also depends on the knowledge and capacity of the respective expert performing the analysis or rather diagnosis.

Further, standard Big Data tools for handling several different data cannot be used, as they cannot handle dynamic data and, thus, work exclusively with operational static data.

Today, besides safety considerations driven by authorities, the optimization of the overall effectiveness and the reduction of operational costs are becoming more important for the plant operators. Hence, different data or rather information is required to get insights regarding the overall effectiveness and the reduction of operational costs of the power plant.

The objective of the present invention is to provide a possibility to gather more information and to make the monitoring process of a power plant less dependent on the knowledge and capacity of the expert performing the respective analyses.

The invention provides a monitoring method for monitoring at least one plant component of a power plant, particularly a nuclear power plant, comprising the following steps: - Receiving asset data associated with the plant component of the power plant, particularly the nuclear power plant; and

Processing the asset data received with at least one analyzing module that applies machine learning and/or statistical methods, thereby generating analysis results, wherein the analysis results comprise information on a condition of the plant component of the power plant, particularly the nuclear power plant.

Further, the invention provides a monitoring system for monitoring at least one plant component of a power plant, particularly a nuclear power plant, comprising at least one data input interface configured to receive asset data associated with the plant component of the power plant, particularly the nuclear power plant, and an analyzing module configured to apply machine learning and/or statistical methods on the asset data received via the data input interface.

The invention is based on the finding that more information concerning the power plant can be retrieved by processing the available asset data in an appropriate manner, namely by means of machine learning and/or statistical techniques. These techniques enable to analyze the respective asset data in a computer-aided manner and automatically. Accordingly, the analysis does not depend on the knowledge and capacity of any experts that have to perform the respective analysis.

In other words, the invention provides a computerized method for automatic analyses of data assigned to the assets of the power plant, namely its plant components.

In fact, the asset data is processed by machine learning and/or statistical methods such that more information, namely certain analysis results, can be retrieved from the overall asset data gathered, for instance gathered by different (separately formed) systems and/or sensors. The asset data obtained by different kinds of systems may be analyzed commonly while applying the machine learning and/or statistical methods, which is not possible when using the techniques known in the state of the art.

The analysis results obtained may comprise information on a condition of the at least one plant component of the power plant. However, the respective information on the condition of the plant component is not restricted to safety considerations, as it may also comprise information relating to optimization of operational equipment effectiveness and reduction of operational costs.

These benefits may be achieved by optimization of maintenance intervals, maintenance duration, spare parts management and/or component replacements that are outputted as analysis results.

Accordingly, the monitoring method as well as the monitoring system may provide cost reduction benefits.

The machine learning and/or statistical methods may relate to a deep learning technique, namely a combined feature extraction and classification. Prior to the analyzing by means of the machine learning and/or statistical methods, the asset data may be pre-processed.

For instance, the asset data is structured, merged, analyzed, filtered and/or transformed. Thus, the respective asset data may be aligned in time previously due to the pre-processing, which simplifies the following analyzing.

In general, mathematical operations and/or transformations may be used for pre-processing the asset data. The mathematical operation may relate to a grouping and/or portioning of the asset data.

The following analyzing may relate to a statistical and cluster analysis of the asset data, particularly with respect to the merged asset data, in order to perform a data quality check, a data cleaning, a data reduction and/or detection of outlier, anomalies, tendencies and/or groups.

The analyzing (machine learning and/or statistical methods applied) may also concern classification analysis and/or regression analysis, particularly with respect to the merged asset data. The classification analysis may use machine learning and/or deep learning classification methods or models. The regression analysis may use machine learning and/or deep learning regression methods or models.

Generally, the classification techniques may relate to support vector machines, discriminant analyses, Naive Bayes, nearest neighbor, decision trees, logistic regression, ensemble classifiers or LSTM RNN networks. The regression techniques may relate to linear regression, generalized linear model, support vector regression, Gaussian process regression, ensemble methods, decision trees or LSTM RNN networks.

The machine learning and/or statistical methods applied may also concern Remaining Useful Lifetime (RUL) estimations. The estimations obtained correspond to the respective analysis results when processing the asset data. Particularly, at least one specialized model, also called predictive model, is used, which is designed for computing the remaining useful lifetime of the plant component from the asset data gathered.

Moreover, the machine learning and/or statistical methods applied may concern diagnoses. Hence, the respective analysis results relate to symptom observations. The symptoms of certain conditions are determined and outputted. In other words, diagnoses are obtained as analysis results which provide the information why certain conditions occur.

Generally, the monitoring system may comprise several analyzing modules that interact with each other. Hence, analysis results may be exchanged among the analyzing modules. In other words, each of the analyzing modules may provide the respective analysis results such that at least another analyzing module may access the respective analysis results for further processing, particularly analyzing.

The monitoring is done automatically such that a cost reduction and time-saving as well as a better and easier data interpretation and decision-making can be ensured. In fact, the monitoring method as well as the monitoring system avoid unplanned outages of the power plant, as the respective plant components are monitored continuously in an appropriate manner. Thus, the user and/or operator of the power plant become(s) aware of upcoming failures in time.

The monitoring system and the monitoring method may be data-driven. Hence, the monitoring system may relate to a data-driven diagnostic and decision management system, which focus on the condition of the power component(s) rather than pre-defined schedules that match more or less with the real situation.

The monitoring method as well as the monitoring system can be used for predictive maintenance, as the respective plant component is monitored continuously such that its status is observed. Hence, plant component failures can be predicted before they occur. Accordingly, the power plant safety is ensured while superfluous maintenance is avoided. In addition, engineers and/or staff of the power plant are provided with more information and deeper insights concerning the power plant, particularly its components.

Put another way, a condition-based maintenance instead of a scheduled one is provided due to the monitoring method applied or rather the monitoring system used. Thus, the overall maintenance costs can be reduced.

In fact, condition-based replacements of plant components may take place rather than scheduled ones or rather fault-based ones. The plant components can be monitored automatically, which reduces the costs required for the staff.

As the outages can be reduced and/or predicted, the lost/reduced power can be anticipated previously.

In general, the monitoring system is a Central Asset Data Intelligence System (CADIS).

The at least one plant component may be an active component, such as a pump and/or a motor, or rather a passive component, for instance a pipe and/or a heat exchanger.

Generally, the entire monitoring system may be located within the power plant directly.

Alternatively, the monitoring system is located abroad with respect to the power plant. Hence, the asset data is forwarded to the monitoring system, for instance via the internet or any other data communication system.

Furthermore, the analysis results obtained may concern the plant component of which the asset data was gathered and/or another plant component of which no asset data was gathered.

According to an aspect, the asset data comprises data from different types of data sources, particularly from an Instrumentation and Control (l&C) system, a Condition Monitoring System (CMS) and/or a Computerized Maintenance Management System (CMMS). Thus, the respective asset data may be retrieved from different systems corresponding to respective data sources. The hardware related to the monitoring system may correspond to already existing data sources used in the power plant. Thus, it is not necessary to install additional hardware components for retrieving the respective asset data used by the monitoring method or rather the monitoring system.

Besides the above-mentioned systems, namely CMS, l&C system and/or CMMS, the monitoring system or rather the monitoring method may deal with a Supervisory Control and Data Acquisition (SCADA) system, at least one Programmable Logic Controller (PLC), a Distributed Control System (DCS), a Manufacturing Execution System (MES), Historians, also called operational historians, or rather any other data base providing asset data.

Generally, the asset data may relate to different data types, which can be processed by the monitoring method and the monitoring system, respectively. The different data types may be provided by the different data sources.

In other words, the monitoring system may be used to connect different assets of the power plant, namely components, sensor devices and/or systems, which each may gather certain asset data. The sensors may relate to mechanical, electrical, process, performance and/or digital interface sensors configured to sense respective parameters.

The data gathered from the different assets is collected and evaluated in a common and combined manner by means of the monitoring system. This is contrary to existing specialized (monitoring) systems that can only process a certain type of data.

According to another aspect, the asset data comprises numeric data, image data, video data and/or text data. The text data may relate to reports and/or logs providing information with regard to at least one plant component in a text format. The text data may also comprise equipment manuals and/or information of an Original Equipment Manufacturer (OEM), for instance a report concerning a Failure Modes and Effect Analysis (FMEA), the plant component design, the installation of the plant component, the commissioning, the operation and/or the maintenance. For instance, the logs relate to operational and/or intervention logs. The reports may relate to diagnostic reports or rather root causes reports.

When processing text data, a data mining technique may be applied in order to extract data from the text that can be processed further. For instance, text mining functionalities or rather text-recognizing techniques may be used in order to retrieve the respective information from the text data.

Further, the asset data may be provided as numerical data, for instance measurement values obtained by measurement sensors associated with the respective plant component. In addition, the asset data may correspond to image data and/or video data. The respective image/video data is processed in an appropriate manner while extracting information that can be processed further. For instance, pattern and/or image recognition techniques may be applied in order to retrieve information from the respective content.

Further, the asset data may comprise static data and/or dynamic data changing over time. This typically concerns the numeric data, which may be static or rather dynamic. Hence, a single value (static numeric data) may be provided or rather a time dependent parameter that is measured continuously so that the respective parameter relates to dynamic data, as its value measured changes over time.

Another aspect provides that the asset data comprises condition data, process data, maintenance data, operational logs and/or maintenance logs. As mentioned above, the respective asset data may be obtained from different types of data sources. In other words, different types of data may be provided and processed by the same monitoring system.

The condition data may be gathered by a Condition Monitoring System (CMS). Hence, the condition data may relate to data assigned to a condition of a plant component. Thus, the condition data may inter alia concern electrical data, for instance voltages, currents or rather powers, and/or mechanical data, such as vibration data, acoustic data, forces, strains and/or loads. Typically, the condition data corresponds to numeric dynamic data, as the respective values of the parameters change over time.

The process data, also called operational data, may be gathered by a SCADA system and/or an l&C system. The process data may relate to temperatures, pressures, flows, powers and so on. Hence, the process data typically corresponds to numeric static data, as the respective parameters do not fluctuate, particularly compared to the parameters corresponding to dynamic data.

The maintenance data may be gathered by a Computerized Maintenance Management System (CMMS). The maintenance data may relate to text data, for instance reports such as maintenance reports. However, the maintenance data may also relate to numeric data, for instance a certain date. The operational logs and/or maintenance logs may also be gathered by the CMMS. The respective logs mainly relate to text data.

Another aspect provides that the asset data is obtained by means of measuring at least one parameter with at least one measurement sensor. Particularly, the parameter corresponds to a mechanical parameter, an electrical parameter, a process parameter and/or a digital parameter. Thus, the monitoring system or rather the monitoring method may be assigned to at least one measurement sensor that measures the respective parameter in order to provide the respective asset data.

The asset data may be collected via a central data storage, wherein the central data storage is accessed to obtain the asset data for further processing, particularly subsequent analyzing. The monitoring system comprises the central data storage for the asset data that is configured to collect the asset data to be processed. The asset data may be stored in a structured manner in the central data storage. Accordingly, the central data storage may relate to a file archive, as the respective asset data is inter alia stored in the central data storage. The respective asset data may be stored automatically, which means that the asset data gathered is automatically forwarded to the central data storage without any manual input.

The central data storage relates to a single data storage platform.

Moreover, the central data storage may be configured to map between the plant components and the asset data automatically. Hence, the plant components and the asset data associated therewith are automatically matched with each other without any manual input.

The central data storage is configured to allow users to store structured and/or unstructured asset data in different formats and of any size. Accordingly, the respective asset data may be stored as numeric data, text data, image data and/or video data.

In addition, the central data storage is configured to simplify data import as well as data export. Thus, the respective different data sources, for instance the condition monitoring system, the instrumentation and control system and/or the computerized maintenance management system, can easily forward the respective asset data gathered to the central data storage. Further, the at least one analyzing module may access the central data storage easily in order to process the respective asset data that is centrally collected by the central data storage.

In order to ensure the simple data import and/or data export, the central data storage may be configured to communicate via different data communication protocols, for instance FTP, WebDAV, OPC, Modbus, MQTT, HTTP, TCP/IP and similar. The respective data exchange between the data sources and the central data storage or rather the data storage and the at least one analyzing module may take place online, namely in streaming, and/or offline, namely in batch.

The at least one analyzing module may communicate with the central data storage in a bidirectional manner such that unstructured asset data provided by the data sources may be accessed by the at least one analyzing module. Then, the unstructured asset data is processed appropriately by the at least one analyzing module in order to obtain structured asset data that can be stored in the central data storage for further processing, particularly further analyzing.

The structuring of the asset data may relate to a pre-processing of the asset data. Afterwards, the machine learning and/or statistical methods may be applied on structured asset data.

Generally, the at least one analyzing module may store its analysis results generated in the central data storage. Thus, at least another analyzing module may access the central data storage and the analysis results stored for further processing. In other words, analysis results may be shared or rather exchanged among different analyzing modules via the central data storage.

Put another way, the central data storage is the main component that can be accessed by each other component of the monitoring system.

The central data storage may be a scalable big data platform, which may comprise at least one High Performance Computing (HPC) module and/or at least one Graphic Processing Unit (GPU) computing server and/or workstation. The big data platform used may comprise several CPUs, several memory cards, several GPUs and/or several disk drives, for instance solid-state disk (SSD) drives. The respective components may be extendable such that the big data platform can be extended in a desired manner, in particular later if necessary. In addition, the monitoring system may comprise a Network Attached Storage (NAS) for data backup and long-term data storage. The analysis results and/or asset data may be backed-up in the backup data storage.

Furthermore, at least a portion of the asset data is processed by means of feature extraction techniques, thereby extracting diagnostic features of the asset data portion. The feature extraction techniques may be applied by a feature extraction module that is configured to extract diagnostic features of the asset data received. The feature extraction techniques may inter alia convert unstructured asset data into structured asset data.

Generally, the feature extraction techniques may relate to so-called data mining tools, as diagnostic features are extracted from huge amounts of data, namely the asset data obtained by the different types of data sources. In other words, the feature extracting may relate to a data mining, as the diagnostic features are generated/mined based on the raw asset data processed by means of the feature extraction module.

The diagnostic features extracted may be accessed by the at least one analyzing module so that the diagnostic features derived from the asset data is processed appropriately by applying machine learning and/or statistical methods.

The diagnostic features extracted and at least a portion of the asset data gathered may be processed together by the analyzing module(s). For instance, diagnoses may be obtained by taking raw asset data as well as diagnostic features extracted into account.

Generally, the extracted diagnostic features correspond to a reduced form of the raw asset data without losing any information required for the intended analysis. Put another way, data not required for the following analysis may be suppressed or rather disregarded. Hence, the analysis only takes the diagnostic features extracted into account, accelerating the analysis since only a portion of the overall asset data is used for the analysis.

For instance, a time series of the features extracted may be obtained. Thus, the respective features may also be aligned in time. The feature extraction techniques may also comprise a signal processing such as filtering and/or integration, which is done simultaneously.

Alternatively or additionally, a statistical analysis of the asset data may be performed in order to extract features from the asset data. For example, a histogram is generated based on the asset data, wherein the respective diagnostic features are obtained from the histogram.

Additionally or alternatively, a fast Fourier transform or any other transform may be performed, wherein the diagnostic features are extracted from the respective transform.

In general, the features extracted may relate to statistical parameters obtained from the asset data and/or values of the respective transform.

Hence, the features extracted can be a root mean square (RMS), a fast Fourier transform peak value, fast Fourier transform energy value, an orbit value and/or a harmonic vector.

The central data storage may be used to store the analysis results obtained and/or diagnostic features extracted.

In general, time-domain relevant techniques may be applied in order to extract features from the asset data, for instance data smoothing, outlier removal, resembling, signal statistics such as mean, moving average and so on. Principle component analysis (PCA), rain flow counting, time series models, recursive and batch-based models and/or Kalman filters, for instance linear, unscented and/or extended Kalman filters.

The time-domain features extracted may relate to effective values, peak values, mean values, crest values and/or kurtosis values.

In addition, frequency-domain related feature extraction techniques may be used such as filtering, time synchronous averaging, spectral analyses and statistics, for instance FFT, peak-to-peak values, bandwidth and so on, model analyses using model/frequency data, envelope analyses and/or any other suitable analyses. The frequency-domain features extracted may relate to narrow band frequency values of amplitude, phase and envelope spectra as well as broadband energy values of amplitude spectra.

Furthermore, time and frequency domain related techniques may be used such as wavelet transforms, empirical mode decomposition, Hilbert-Huang transforms, short-time Fourier transforms and/or Wigner-Ville distributions.

All these different techniques may be applied on the asset data gathered in order to extract features from the huge amount of asset data obtained.

Thus, the diagnostic features extracted may comprise time-domain features and/or frequency domain-features.

Another aspect provides that the diagnostic features extracted are stored, particularly in the central data storage, and/or processed by at least one insights module applying statistics and/or clustering techniques to the diagnostic features extracted. Thus, the insights module is also called statistics and/or clustering module. The monitoring system may comprise at least one insights module that is configured to apply statistics and/or clustering techniques to the diagnostic features extracted. Additionally or alternatively, the insights module is configured to apply statistics and/or clustering techniques to the (raw) asset data.

As mentioned earlier, the feature extraction techniques may be used to extract relevant features from the overall asset data without losing information required to the intended analysis.

The features extracted may also be processed by the at least one insights module that is configured to apply statistic and/or clustering techniques to the diagnostic features in order to reduce the amount of feature data obtained to few important features.

Generally, the statistic and/or clustering techniques may relate to screening, ranking, selection and/or transformation. The insights module ensures that groups or clusters may be discovered due to similarities between the processed asset data and/or diagnostic features extracted. Further, anomalies, outliers or fault patterns may also be detected while applying statistic and/or clustering techniques. In other words, the statistical and cluster analysis may provide a data quality check, a data cleaning, a data reduction as well as a detection of outliers, anomalies and/or groups.

This relates to a grouping and/or a partioning of the respective asset data or rather diagnostic features extracted.

The asset data may be received in real time, for instance continuously. Further, the asset data is processed in real time, particularly continuously, by the monitoring system.

Thus, a scalable computer may be used to perform the respective monitoring method in order to process the asset data in real time.

The respective plant component may be checked permanently and/or event- triggered in order to gather respective asset data continuously and/or when necessary.

Once the asset data is gathered, the asset data is forwarded to the central data storage for collecting purposes.

Another aspect provides that the analysis results are transferred automatically to a customer data platform. The customer data platform may relate to a data sink, as the respective analysis results are exported.

Additionally, the features extracted may be transferred to the customer data platform. For instance, the analysis results (and optionally the features extracted) may be exported to a cloud, an loT platform, an enterprise system (EPR) or any other database, for instance a SqL database, a big data database and/or a NoSqL database. Moreover, the information may be exported to any other data warehouse.

The monitoring system may be modularly structured enabling additional modules to be integrated into the monitoring system (later). Accordingly, additional components or equipment may be integrated, particularly additional data sources and/or analyzing modules.

Moreover, the monitoring system may have least one diagnostics module that is configured to diagnose at least the (raw) asset data. The diagnostics module relates to an analyzing module, which is configured to support comprehending why a certain condition occurs. For this purpose, the diagnostics module is inter alia configured to efficiently visualize and deeper analyze the asset data collected, the extracted diagnostics features and/or the analysis results obtained by another analyzing module, particularly the insights revealed by the insights module.

For instance, trend analysis, signal analysis and/or operational deflection shape (ODS) analysis may be provided by the diagnostics module. Generally, the diagnostics module supports to identify the root cause of a certain event detected in an easy and fast manner.

In addition, the diagnostics module may verify or rather confirm the data-driven analysis results obtained while providing recommendations of remedial actions or rather counter measures simultaneously.

Accordingly, it can be determined why a certain event/issue did happen. Moreover, a root cause analysis can be provided for the respective events/issues occurred.

The monitoring system may be an online monitoring system that provides a web-based user interface via which information obtained is visualized. For instance, the analysis results and/or features extracted by the feature extraction module may be visualized by means of the customer data platform such as the web-based user interface. The web-based user interface may be a remotely accessible web-based user interface, also called dashboard, enabling users to access the relevant information.

The analysis results may be visualized by means of the remotely accessible web-based user interface for asset condition and priority management, which inter alia enables fleet and plant managers to learn from data and insights and to make reliable and faster decisions.

The web-based user interface may communicate with the central data storage and allows users to have access from anywhere to the stored data/information. For instance, access is provided for current and historical health condition(s) as well as asset data of the plant component(s). Further, an event notification/alarming may be provided, for instance by e-mail or short message services. The event may relate to the occurrence of an outlier or generally any anomaly detected during the processing of the asset data.

Via the web-based user interface, a list of all occurred asset events may be accessed, which is stored in the central data storage. The list may comprise information concerning the time, the type of issue, the place and the root cause.

Moreover, the central data storage may comprise information with respect to health condition changes and/or operating mode changes, which can be accessed via the web-based user interface.

Data source connection errors may be logged and stored in the central data storage. Thus, such error notifications may also be accessed via the web-based user interface.

In a similar manner, ticket status for critical/aggravating events may be stored in the central data storage.

In addition, a statistical assessment of the events is stored in the central data storage. Thus, the statistical assessment can also be accessed in a remote manner via the web-based user interface.

Generally, analysis results, for instance insights and key performance indicators (KPI), from the at least one analyzing module are stored in the central data storage. These analysis results can also be accessed via the web-based user interface.

In addition, at least one prediction module may be provided that is configured to predict asset-relevant information, in particular wherein the prediction module is configured to apply classification techniques and/or regression techniques. As mentioned earlier, the classification techniques and/or regression techniques may relate to the machine learning and/or statistical methods. Hence, predictions concerning the at least one plant component may be provided by the prediction module.

The prediction module may correspond to an analyzing module. Thus, the respective predictions outputted may relate to the analysis results of the specific analyzing module. The classification analysis may also use already collected data for model training, testing and evaluation.

Further, the classification analysis may use new collected data for prediction. The regression analysis may use already collected data for model training, testing and evaluation. The evaluation analysis may use new collected data for prediction.

Furthermore, the invention provides a predicting method of predicting at least one condition of at least one plant component of a power plant, wherein the monitoring method described above is used to predict an event, a fault, a remaining useful lifetime, a maintenance date and/or a fatigue. These predictions relate to a certain condition of the plant component.

Accordingly, a predictive analysis is possible. Thus, it can be determined that a certain issue will happen as well as the respective probability with which the issue will happen. Predictions or rather estimations are obtained as the analysis results. For instance, this ensures that required spare parts are ordered in time.

The Remaining Useful Lifetime (RUL) estimation may use the asset data, particularly previously merged and structured asset data, and at least one specialized model designed for computing the remaining useful lifetime of a plant component from different types of asset data gathered. For instance, the respective model is useful when historical data and information are available concerning run-to-failure histories of plant components similar to the one to be analyzed. Moreover, known threshold value(s) of some condition indicators that indicate the respective failure and/or data about how much time or how much usage it took for similar plant components to reach the failure are taken into account.

Fatigue usage factor prediction may use data on mechanical plant components like pipes with validated finite models of the respective plant components (design data). Hence, mechanical parameters are taken into account such as vibration, temperature and/or load data. The fatigue usage factor predictions relate to the analysis results obtained by processing the asset data.

The predictions may be used to analyze historical and current data and, thus, creating a predictive model (algorithm), which is enabled to describe what has happened in the past and to use this predictive model to predict what may happen in the future.

The respective model is trained by means of labelled data, comprising input data, namely features extracted and/or asset data, as well as output data. The input data relates to so-called predictors or known data, whereas the output data relates to known responses assigned to the input data. The labelled data is used to create the predictive model. In fact, the predictive model obtained is based on machine learning and/or statistical methods.

Once the predictive model has been generated, newly measured asset data can be inputted as new input data, which is processed by the predictive model. The predictive model outputs predictive responses associated with the new input data as the analysis results. The predictive responses relate to information concerning conditions of a certain plant component.

Accordingly, the predicting method is enabled to predict at least one condition of the plant component of the power plant. As already described, the predicting method or rather the predictive model may be based on classification and/or regression techniques.

Generally, all data/information obtained by any processing, namely the analysis results of any of the analyzing modules, may be stored in the central data storage. Thus, other modules may access this data and use the respective data for their own processing/analysis.

Further, all data, namely the (raw) asset data, the analysis results and/or the features extracted, may be merged in order to obtain time aligned data that can be used for further processing.

Particularly, statistical and cluster analyses, classification analyses, regression analyses, remaining useful lifetime estimations and/or predictions may be performed based on the asset data, particularly pre-processed asset data like merged structured asset data.

The monitoring system may consist of hardware modules, for instance sensors, smart data loggers and/or centralized data servers, as well as software modules, for instance user interface(s), dashboard(s) and/or analyzing modules, particularly the algorithms/techniques associated thereto.

In general, each component in a power plant has its typical characteristics like design, function, failure modes, and operating modes. Based on these characteristics, an individual Failure Mode Symptom Analysis may be performed, which identifies all relevant failure modes with their diagnostic indicators.

The result of this analysis helps to design and configure the monitoring system concerning the selection of sensors, signal conditioning modules, monitoring parameters, data storage strategies, and data analyses for feature extraction.

Accordingly, the monitoring system is a universal one that has to configured in the intended manner.

In addition, a computer program for performing the monitoring method and/or the predicting method is provided, wherein the computer program comprises computer program code means performing the steps of the method, when the computer program is executed on a computer or a corresponding computing unit. Thus, the asset data relates to input data of the computer program that processes the asset data by means of applying machine learning and/or statistical methods, thereby generating analysis results, wherein the analysis results comprise information on a condition of the plant component of the power plant. Alternatively or additionally, the computer program predicts an event, a fault, a remaining useful lifetime, a maintenance date and/or a fatigue.

The foregoing aspects and many of the attendant advantages of the claimed subject matter will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

Figure 1 shows an overview of a monitoring system according to the present invention,

Figure 2 shows an overview representing two different embodiments with regard to the location of the monitoring system according to the present invention, Figure 3 shows another overview of a monitoring system according to the present invention,

Figure 4 shows another overview of a monitoring system according to the present invention,

Figure 5 shows an overview of a feature extraction module used by the monitoring system according to the present invention,

Figure 6 shows an overview of an insights module used by the monitoring system according to the present invention,

Figure 7 shows another overview of a monitoring system according to the present invention,

Figure 8 shows another overview of a monitoring system according to the present invention,

Figure 9 shows a schematic overview illustrating the teaching of a predictive model used by the predicting method according to the present invention,

Figure 10 shows an overview of an exemplary embodiment of a remaining useful life prediction using known historical data,

Figure 1 1 shows an overview of the exemplary embodiment of a remaining useful life prediction of Figure 10 with new data, and

Figure 12 shows an overview of an exemplary fatigue usage factor prediction.

In Figure 1 , a monitoring system 10 for monitoring at least one plant component of a power plant is shown, particularly a nuclear power plant. The monitoring system is also called Central Asset Data Intelligence System (CADIS).

In the shown embodiment, the monitoring system 10 comprises four main structures, namely a data source structure 12, a data communication protocol structure 14, a data storage platform and analytics module structure 16 as well as a data sink structure 18. As shown in Figure 1 , the data source structure 12 interacts with the data communication protocol structure 14 that in turn interacts with the data storage platform and analytics module structure 16. In addition, the data storage platform and analytics module structure 16 interacts with the data sink structure 18.

The data source structure 12 comprises several different types of data sources 20, which each may provide different kinds of data.

In the shown embodiment, the several data sources 20 relate to a Condition Monitoring System (CMS) 22, an Instrumentation and Control (l&C) system as well as a Supervisory Control and Data Acquisition (SCADA) system 24, a Computerized Maintenance Management System (CMMS) 26 as well as an Original Equipment Manufacturer (OEM) data source 28.

The respective different data sources 20 all gather asset data associated with the plant component(s) of the power plant to which the monitoring system 10 is associated.

Further, the monitoring system 10 comprises a central data storage 30 that communicates with the data source structure 12, namely the different data sources 20.

The central data storage 30 is assigned to the data storage platform and analytic modules structure 16 that has a data input interface 32 via which the respective asset data is received from the data source structure 12, namely the different data sources 20. Accordingly, the (raw) asset data provided by the several data sources 20 is collected by and stored in the central data storage 30.

For communicating with the different data sources 20, the data communication protocol structure 14 is provided, which ensures that the data storage platform and analytic modules structure 16 is enabled to communicate with the different data sources 20 via different data communication protocols, for instance FTP, WebDAV, TCP/IP, UDP, Modbus, OPC and/or other data communication protocols required.

In addition, the monitoring system 10 comprises a feature extraction module 34 that is connected with the central data storage 30. The feature extraction module 34 is enabled to access the asset data provided by the central data sources 20 and to process the asset data read from the central data source 20. The feature extraction module 34 is configured to extract diagnostic features of the asset data. The feature extraction module 34 uses feature extraction techniques that may convert unstructured asset data into structured asset data.

Generally, the feature extraction techniques may relate to a so-called data mining, as the diagnostic features are extracted from huge amount of asset data, namely the asset data obtained by the different types of data sources 20. In other words, the diagnostic features are generated/mined based on the raw asset data processed by means of the feature extraction module 34. Accordingly, the amount of asset data is reduced.

Moreover, data merging may take place by the feature extraction module 34.

The data merging may be assigned to the asset data or rather the already obtained diagnostic features extracted.

In other words, the asset data may be merged by means of the feature extraction module 34 in order to obtain merged asset data. The merged asset data may be further processed by the feature extraction module 34 such that the diagnostic features are extracted from the merged asset data. Alternatively, the merged asset data may be provided without any feature extraction afterwards.

Alternatively or additionally, the data merging may take the diagnostic features extracted into account. Thus, the diagnostic features extracted are merged in order to obtain merged diagnostic features, which are used for further processing.

Generally, the data merging may relate to structure the respective data. Thus, structured asset data or rather structured diagnostic features may be provided.

The feature extraction module 34 may communicate with the central data storage 30 in a bidirectional manner such that the diagnostic features extracted, the merged asset data, namely the structured one, or rather the structured diagnostic features can be forwarded to the central data storage 30. Hence, this data may be accessed by other modules of the monitoring system 10, as will be described later in more detail.

The monitoring system 10 further comprises several analyzing modules 36 that are located in parallel to each other as shown in Figure 1 . In the shown embodiment, a first analyzing module 36 relates to a statistics and clustering module 38, wherein a second analyzing module 36 relates to a classification module 40, wherein a third analyzing module 36 relates to a regression module 42, wherein a fourth analyzing module 36 relates to a remaining useful life (RUL) estimation module 44, wherein a fifth analyzing module 36 relates to a fatigue prediction module 46, and wherein a sixth analyzing module 36 relates to a diagnostics module 48.

The statistics and clustering module 38 is also called insights module.

All of these analyzing modules 36 are configured to apply machine learning and/or statistical methods on the asset data received, particularly the diagnosis features extracted from the asset data, so as to provide analysis results 50 that comprise information on a condition of the plant component of the power plant.

The monitoring system 10 also comprises a data backup 52 that inter alia receives the analysis results 50 so that the analysis results 50 are stored for a long time.

In addition the analysis results 50 are inter alia forwarded to the data sink structure 18, which may comprise a user interface dashboard 54, a cloud IOT platform 56, a database 58 and/or a data warehouse 60.

Moreover, the (structured and/or unstructured) asset data stored in the central data storage 30 may be accessed via the data sink structure 18, particularly the respective components shown.

In fact, the monitoring system 10 may provide a web-based user interface via the data sink structure 18 so that the information obtained by the analysis results 50 may be visualized to the user.

In other words, the data sink structure 18 generally comprises a customer data platform to which the analysis results 50 determined are transferred automatically.

In Figure 2, two different embodiments with regard to the location of the monitoring system 10 are shown, as the monitoring system 10 may be located within the power plant area 62 completely (upper overview) or partly in the power plant area 62 and an abroad area 64 (lower overview). In the upper overview, all structures discussed with respect to Figure 1 are located in the power plant area 62, namely the data source structure 12, the data communication protocol structure 14, the data storage platform and analytics module structure 16 as well as the data sink structure 18.

In contrast, the lower overview illustrates that the data source structure 12 and the data sink structure 18 are located in the power plant area 62, whereas the data storage platform and analytics module structure 16 is located elsewhere, namely the abroad area 64. The data communication protocol structure 14 provides the interface between the power plant area 62 and the abroad area 64.

Figure 3 shows the monitoring system 10 in a different overview, particularly the data storage platform and analytics module structure 16.

The respective data provided by the data sources 20 are illustrated as process data, condition data as well as maintenance logs, which are forwarded to the central data storage 30 as asset data, namely the data associated with the plant component.

The process data, the condition data as well as the maintenance logs correspond to the different kinds of data sources 20, which were already shown in Figure 1 . Accordingly, the asset data are collected by the central data storage 30 from the data sources 20.

Then, diagnostic features are inter alia extracted from the asset data at least partly by means of the feature extraction module 34. The diagnostic features extracted can be used for further analyzing as will be discussed hereinafter.

Figure 3 also shows that the asset data collected by the central data storage 30 are also processed further directly without any feature extraction.

In the shown embodiment, the (raw) asset data is directly forwarded to the diagnostics module 48. The diagnostics module 48 may further receive diagnostic features extracted by the feature extraction module 34.

Further, the (raw) asset data is forwarded to a prediction module that may be established by the classification module 40, the regression module 42, the remaining useful life (RUL) estimation module 44 and/or the fatigue prediction module 46. Deep learning techniques, for instance provided by at least one LSTM RNN network, may be applied on the (raw) asset data for the respective prediction.

Moreover, the prediction module may also take the diagnostic features extracted into account provided by the feature extraction module 34. For instance, supervised learning techniques may apply on the respective data.

Figure 3 further reveals that the analyzing modules 36 may exchange their respective analysis results 50 among each other.

For instance, the analysis results 50 are forwarded to the central data storage 30 so that the respective analysis results 50 can be accessed by several modules, particularly different analyzing modules 36.

The analysis results 50 obtained may relate to certain actions that shall be done.

These actions may be proposed or rather recommended by the respective analyzing module(s) 36. An operator may follow the proposal or rather recommendation, for instance by acknowledging.

Further, the respective action may also be performed automatically. Hence, the operator will only be informed about the action performed.

Moreover, the analysis results 50 may relate to information, for instance a root cause of certain events/issues. An expert may review the analysis results 50 in order to decide how to proceed.

In Figure 4, the monitoring system 10 is shown in a different manner, particularly the data storage platform and analytics module structure 16.

The respective flow of data is illustrated for a certain embodiment.

In a first step, the respective asset data is imported or rather received via the data input interface 32. The asset data comprises sensor data, for instance vibration parameters and/or process parameters, as well as additional information, namely video data, image data and text data, such as reports.

The asset data stored in the central data storage 30 is accessed and processed, for instance by the feature extraction module 34. For processing the asset data, configuration files and/or loading measurement files are read.

Additionally or alternatively, a signal processing, for instance filtering and/or integration, a signal analyzing, for instance generating histograms and/or transforming (FFT/STFT) and/or a feature extraction is performed, for instance a statistical parameter such as root mean square (RMS), FFT peak values, FFT energy values, orbit values and/or harmonic vectors.

Furthermore, the asset data and/or the analysis results, namely the results of the respective processing, are stored in a time series.

Accordingly, the (raw) asset data is structured while processing the asset data.

As previously mentioned, a data merging may take place.

Furthermore, machine learning and/or statistical methods are applied on the data provided, for instance the diagnosis features extracted, the (structured) asset data and/or the data obtained by the data processing.

The machine learning and/or statistical methods may relate to a statistical and cluster analysis of the asset data by means of the statistics and clustering module 38, particularly with respect to the merged asset data.

As shown in Figure 4, a time series of diagnosis features extracted and/or asset data, for instance process data, may be inputted. The statistical and cluster analysis identifies relevant changes, for instance groups and/or outliers, wherein respective information is outputted, namely clusters, outliers, important features and/or results of a Principle Component Analysis (PCA).

Alternatively or additionally, diagnostics are done by means of the diagnostics module 48. The diagnostics module 48 may also take the asset data and/or the diagnosis features extracted by the feature extraction module 34 into account.

The diagnostics module 48 performs a symptom observation while doing a trend analysis, a signal analysis and/or an operational deflection shape (ODS) analysis. Hence, it can be verified why certain events/issues did occur.

Moreover, the analyzing modules 36 and/or the diagnostics module 48 may also provide a reporting with recommendation wherein the problem and its cause are described. Furthermore, the impact of the problem as well as the respective recommendation for avoiding or rather fixing the problem may be outputted provided that the respective action has not been performed automatically.

Figure 5 shows in detail how the raw asset data is processed in order to extract the diagnostic features by means of the feature extraction module 34.

Generally, the asset data may be processed by time-domain and/or frequency- domain techniques, which may depend on the respective data to be processed. The feature extraction ensures that the amount of data used for further analysis is reduced.

In Figure 6, the usage of the statistics and clustering module 38, also called insights module, is shown. The statistics and clustering module 38 performs a Principle Component Analysis (PCA) and an additional clustering on the features table provided by the feature extraction module 34.

Afterwards, the analysis results 50 are provided in a descriptive manner simplifying the conclusion provided.

In Figure 7, the monitoring system 10 is shown in a different manner.

It reveals that the respective data is gathered from different assets of the power plant, wherein the different data sources 20 are illustrated, namely the CMS 22, the l&C system 24, the CMMS 26 and/or other systems.

The data source structure 12 relates to the CADIS hardware.

The data storage platform and analytics module structure 16 relates to the CADIS software.

Moreover, Figure 7 reveals that the single central data storage 30 relates to a file archive being the central module of the CADIS software.

All analyzing modules 36 interact with the central data storage 30 in a bidirectional manner.

Accordingly, the (raw) asset data can be accessed by each of the analyzing modules 36, which may process the asset data in a certain manner. As shown in Figure 7, the feature extraction module 34 is a specific analyzing module 36.

Since the analyzing modules 36 interact with the central data storage 30 in a bidirectional manner, the respective analysis results, for instance the diagnosis features extracted, are stored in the central data storage 30 such that other analyzing modules 36 may use the respective analysis results for further processing.

For instance, the diagnosis features extracted may be clustered or used for certain predictions.

Previously, the asset data could have been reduced or rather cleaned.

Afterwards, a trend analysis, signal analysis or rather ODS analysis of respective assets may be outputted, particularly by means of the diagnostics module 48.

Furthermore, the processed asset data may be outputted, namely statistics associated therewith and/or the asset data clustered, for instance by means of the statistics and clustering module 38.

In addition, predictions may be outputted, for instance predictions performed by means of the remaining useful life (RUL) estimation module 44 and/or the fatigue prediction module 46.

Generally, classification and/or regression techniques may be applied by any of the above-mentioned modules.

In Figure 8, the monitoring system 10 is shown in another overview.

It becomes obvious how the asset data is gathered by means of the CMS 22 monitoring plant components like control valves, heat exchangers, pumps and pipes.

In addition, the l&C/SCADA 24 is used for gathering operating and process parameters, for instance, power, mass flows, pressures and/or temperatures. The CMS 22 gathers mechanical, electrical as well as operation and process parameters, for instance voltages/currents, vibrations, temperatures, vibro- acoustics and/or temperatures.

Accordingly, the asset data is obtained by means of measuring at least one parameter with at least one measurement sensor.

The asset data gathered is processed with at least one analyzing module 36 that applies machine learning and/or statistical methods on the asset data so as to obtain insights and providing recommendations with regard to optimizations.

The respective prediction module, for instance the remaining useful life (RUL) estimation module 44 and/or the fatigue prediction module 46, may be configured to do conduct a predicting method of predicting at least one condition of the at least one plant component.

For this purpose, the asset data is received and processed by the respective module in order to obtain analysis results, which relate to a predictive condition of the component.

For this purpose, a predictive model is used that was trained previously, as shown in Figure 9.

At the beginning labelled data is used, which comprises known data and known responses, namely historical data. Then, the predictive model is trained by the known data in order to be enabled to output predicted responses when the predictive model receives new asset data.

In Figures 10 and 1 1 , the prediction done by the remaining useful life (RUL) estimation module 44 is shown as an example.

The RUL prediction module 44 using failure models that are designed for computing RUL from different types of measured data. The models are useful when following historical data and information are available as shown in Figure 10.

Run-to-failure histories of plant components similar to the one to be diagnosed, a known threshold value of some condition indicator that indicates the failure and data about how much time or how much usage it took for similar plant components to reach failure (lifetime) are used. The RUL prediction module 44 using labelled data analyses historical and current labelled data with predictors and responses, to create the predictive model which describes the relationships between predictors (condition and process features) and responses (e.g. fault types) as labelled clusters or data distributions, as described above.

Then, the RUL prediction module 44 computes an Health Index Value by means of the following equation:

HIV = 1 - D2R/D2F

Herein, D2R is the distance between the current distribution centre and the reference distribution centre, and D2F is the distance between the fault distribution centre and reference distribution centre. The time series of the HIV is finally used as input in regression methods to predict the RUL, namely when the HIV will be equal 0.

This is shown in Figure 1 1 .

In Figure 12, the usage of the fatigue prediction module 46 is shown in more detail.

Using the fatigue usage factor prediction module, continuously acquired vibration, temperature or load data on a plant component like a pipe are processed together with the finite element method (FEM) model results of the plant component. Hence, the stresses at defined positions can be estimated. Finally, the fatigue usage factor can be determined.

Using linear regression methods and the time series of the fatigue usage factor, it is possible to predict the remaining useful life of the plant component, as shown in Figure 12.

All the analysis results 50 obtained by the different analyzing modules 36 can be outputted, namely via the web-based dashboard 54, also called web-based user interface, for asset condition and priority management; please refer to Figure 1 .

This enables fleet and plant managers to learn from the asset data and the analyzing results obtained and to make reliable and faster decisions. The web-based dashboard 54communicates with the central data storage 30 and allows users to have access from anywhere to:

- current and historical health condition as well as asset data of the plant components, list of all occurred asset events (When, What, Where, How): o health condition changes, o operating mode changes, o system faults, o ticket status for critical, aggravating events,

- statistical assessment of the events,

- analysis results 50, namely insights and key performance indicators (KPI) from the analyzing modules 36.

Hereinafter, another example is described in detail.

The monitoring system 10 using several decentralized stations of the CMS 22 is used to monitor more than 35 safety and availability relevant plant components, for instance pumps. The monitoring takes place online.

The process data of these plant components are collected via l&C 24 together with the condition data inside the centralized data storage 30.

Each monitored plant component, namely each monitored pump, has about 4 - 5 vibration sensors, depending on the number of radial and axial rolling bearings and a minimum of 5 process parameters.

The plant components are permanently monitored, but the data is smartly stored, i.e. periodically or after an event. For each vibration signal, a specific data processing and feature extraction is done automatically after each data storage. In fact, two types of features are extracted:

• time-domain features like effective, peak, mean, crest, or kurtosis values, and • frequency-domain features like narrow band frequency values of amplitude, phase and envelope spectra as well as broadband energy values of amplitude spectra.

The configured data collection, processing and feature extraction results in minimum 80 features time series per plant component and more than 2800 features time series for the 35 plant components.

The unique way to deal efficiently with this huge amount of data is using data analytics to extract valuable insights and to make reliable data-based decisions.

During commissioning, collected baseline measurements were analyzed using descriptive statistics and unsupervised machine learning algorithms (clustering) in order to characterize the behavior of the plant components and to derive reference and threshold values for the different operating modes.

In a first step, gather, receive and collect the raw asset data.

In a second step, extract features time series from the raw asset data.

In a third step, clean, normalize and then reduce the number of features (using variance and correlation criteria, feature selection algorithms and feature transformation algorithms like PCA).

In a fourth step, apply cluster algorithms to identify natural groups, pattern and outliers.

In a fifth step, analyze and interpret the identified clusters by understanding the behavior of the plant components (what happened, when, where, how), identifying healthy and faulty states and define criteria for their prediction, and identifying the root causes of the abnormal or faulty states.

The respective steps mentioned above can be summarized as receiving asset data associated with the plant component and processing the asset data received with at least one analyzing module that applies machine learning and/or statistical methods, thereby generating analysis results. In fact, several processing steps are performed in the example given above.

In another example, a large number of asset data have been collected on a reactor coolant system (RCS) during the commissioning phase of a nuclear power plant at different commissioning phases, namely cold functional test (CFT) and hot functional test (HFT).

In total, more than 100 measurement files were collected. Each measurement file had more than 50 measurement channels (record length > 1800 sec, sampling frequency > 5000 Hz).

In a first processing step, features are extracted from the collected asset data.

In a second step, the features are combined with labels extracted from operational logs (e.g. pressure level, temperature level, pump status, test conditions ...).

In a third step, a statistics and clustering of the features is performed in order to find anomalies which happened during the commissioning and then to describe when, where and how they happened.

In a fourth step, the analysis of the detected events and the related data is performed in order to find out the root cause of the event.

In a last step, a predictive model is created with vibration features as predictors and the following responses:

• reactor coolant pump states,

• reactor coolant system temperature level, and

• reactor coolant system pressure level, in order to check the reproducibility of the behavior at different load states.

In general, the monitoring system 10 enables plant operators to collect huge and complex asset data at one central place, to transform the asset data into actionable insights for faster and reliable decisions.

The monitoring system 10 stores and presents all occurred events with their description (what/event type, when/event date, where/event location, how/event occurrence) and informs continuously about the (historical and current) health condition and fatigue state of the monitored assets and can predict these values if necessary (predictive maintenance). The monitoring system 10 provides easier and faster data exploration, diagnostics and root cause analyses for deeper analysis of critical events.

Claims

1 . A monitoring method for monitoring at least one plant component of a power plant, comprising the following steps:

Receiving asset data associated with the plant component of the power plant; and

Processing the asset data received with at least one analyzing module (36) that applies machine learning and/or statistical methods, thereby generating analysis results (50), wherein the analysis results (50) comprise information on a condition of the plant component of the power plant.

2. The monitoring method according to claim 1 , wherein the asset data comprises data from different types of data sources (20), particularly from an instrumentation and control system (24), a condition monitoring system (22) and/or a computerized maintenance management system (26).

3. The monitoring method according to claim 1 or 2, wherein the asset data comprises numeric data, image data, video data and/or text data.

4. The monitoring method according to any of the preceding claims, wherein the asset data comprises static data and/or dynamic data changing over time.

5. The monitoring method according to any of the preceding claims, wherein the asset data comprises condition data, process data, maintenance data, operational logs and/or maintenance logs.

6. The monitoring method according to any of the preceding claims, wherein the asset data is obtained by means of measuring at least one parameter with at least one measurement sensor, in particular wherein the parameter corresponds to a mechanical parameter, an electrical parameter, a process parameter and/or a digital parameter.

7. The monitoring method according to any of the preceding claims, wherein the asset data is collected via a central data storage (30), wherein the central data storage (30) is accessed to obtain the asset data for further processing, in particular wherein the asset data is stored in a structured manner in the central data storage (30).

8. The monitoring method according to any of the preceding claims, wherein at least a portion of the asset data is processed by means of feature extraction techniques, thereby extracting diagnostic features of the asset data portion, in particular wherein the feature extraction techniques transform unstructured asset data into structured asset data.

9. The monitoring method according to claim 8, wherein the diagnostic features extracted comprise time-domain features and/or frequency-domain features.

10. The monitoring method according to claim 8 or 9, wherein the diagnostic features extracted are stored, particularly in the central data storage (30), and/or processed by at least one insights module (38) applying statistics and/or clustering techniques to the diagnostic features extracted.

1 1 . The monitoring method according to any of the preceding claims, wherein the asset data is received and processed in real time and/or permanently.

12. The monitoring method according to any of the preceding claims, wherein the analysis results are transferred automatically to a customer data platform.

13. A predicting method of predicting at least one condition of at least one plant component of a power plant, wherein the monitoring method according to any of the preceding claims is used to predict an event, a fault, a remaining useful lifetime, a maintenance date and/or a fatigue.

14. A monitoring system (10) for monitoring at least one plant component of a power plant, comprising at least one data input interface (32) configured to receive asset data associated with the plant component of the power plant and an analyzing module (36) configured to apply machine learning and/or statistical methods on the asset data received via the data input interface (32).

15. The monitoring system (10) according to claim 14, wherein the monitoring system (10) is modularly structured enabling additional modules to be integrated into the monitoring system (10).

16. The monitoring system (10) according to claim 14 or 15, wherein a feature extraction module (34) is provided that is configured to extract diagnostic features of the asset data received.

17. The monitoring system (10) according to claim 16, wherein at least one insights module (38) is provided that is configured to apply statistics and/or clustering techniques to the diagnostic features extracted.

18. The monitoring system (10) according to any of claims 14 to 17, wherein at least one diagnostics module (48) is provided that is configured to diagnose at least the asset data.

19. The monitoring system (10) according to any of claims 14 to 18, wherein at least one prediction module (40, 42, 44, 46) is provided that is configured to predict asset-relevant information, in particular wherein the prediction module (40, 42, 44, 46) is configured to apply classification techniques and/or regression techniques.

20. The monitoring system (10) according to any of claims 14 to 19, wherein a central data storage (30) for the asset data is provided that is configured to collect the asset data to be processed.

21. The monitoring system (10) according to any of claims 14 to 20, wherein the monitoring system (10) is an online monitoring system (10) that provides a web- based user interface (54) via which information obtained is visualized.

22. A computer program for performing the monitoring method according to any of claims 1 to 12 and/or the predicting method according to claim 13, wherein the computer program comprises computer program code means performing the steps of the method, when the computer program is executed on a computer or a corresponding computing unit.