Power System Component Evaluation

Abstract

A computer system for evaluating power system components 10 is provided. [Solution] A system configured to receive time-series data 30 from a sensor that monitors a power system component, preprocess the time-series data 30, extract a plurality of features 32 from the preprocessed time-series data 30, use a statistical test method 34 on the plurality of features 32 to determine the statistical significance 36 of each of the features 32, send one or more of the features 32 having a statistical significance 36 exceeding a predetermined limit value 38 to a machine learning model 40, quantify the contribution of the prediction output 42 from the machine learning model 40, and repeat the extraction, use, sending, and quantification until the judgment criteria are met in order to obtain an evaluation 50 of the power system component.

Description

Technical Field

[0001] The present disclosure generally relates to methods for controlling power system components. In certain aspects, the present disclosure relates to power system component evaluation. The present disclosure can be applied, in particular, to heavy-duty vehicles such as trucks, buses, and ships among several vehicle types. The present disclosure can also be applied, in particular, to other application fields such as industrial applications, automotive applications, power generation facilities, stationary applications, etc. among several application fields. The present disclosure may be described with respect to a particular vehicle or application field, but the present disclosure is not limited to any particular vehicle or application field.

Background Art

[0002] Evaluating power system components often aims, among several purposes, in particular, to enhance performance and ensure compliance with regulations. Current systems can provide some insights, but are struggling to capture the complex differences between various data contents mainly due to the complexity and cost of ensuring accurate identification measurement criteria. For example, fuel types such as diesel and biofuels share many overlapping characteristics. Current solutions either require computationally intensive hardware and software systems or have low specificity required for a reliable evaluation of power system components that can be used to enhance decision-making.

Summary of the Invention

[0003] In a first aspect of this disclosure, a computer system for evaluating a power system component is provided, the computer system includes a processing circuit configured to: receive time-series data from a sensor monitoring the power system component; preprocess the time-series data; extract a plurality of features from the preprocessed time-series data; use a statistical test method on the plurality of features to determine the statistical significance of each of the features; feed one or more of the features having statistical significance exceeding a predetermined limit into a machine learning model; quantify the contribution of the prediction output from the machine learning model; and perform at least one iterative process until one or more predetermined criteria are met, wherein the iterative process includes: (i) extracting a plurality of features from the prediction output; (ii) using a statistical test method on the plurality of features from the prediction output; (iii) feeding one or more of the features from the prediction output having statistical significance exceeding a predetermined limit into a machine learning model; and (iv) quantifying the contribution of the prediction output from the machine learning model; and, in response to the predetermined criteria being met, obtain an evaluation of the power system component based on the prediction output.

[0004] The evaluation may be one of the following: (i) detecting a deviation of the functionality of the power system component from baseline functionality, or (ii) determining the calibration status of the power system component. The processing circuit may be further configured to use the evaluation to initiate one of the following: predictive maintenance of the power system component or recalibration of the power system component.

[0005] A first aspect of this disclosure may address the problem of efficiently and accurately evaluating power system components using readily available sensor data. Technical advantages may include providing a computationally efficient and interpretable evaluation method that leverages existing data and provides insights into the physical and chemical processes of power system components, unlike the prior art “black box” approach.

[0006] In several optional examples, including at least one preferred example, the processing circuit is configured to define one of the following problems—regression, clustering, or classification—based on the predicted output above, and to quantify the contribution of the predicted output by solving the regression, clustering, or classification problem above in order to obtain an evaluation. Technical advantages may include increased flexibility in model selection, enabling a more tailored evaluation based on the nature of the data and the desired outcome.

[0007] In several optional examples, including at least one preferred example, time-series data relate to a power system component that is an internal combustion engine, and the evaluation is the determination of the fuel characteristics of the fuel consumed by the aforementioned internal combustion engine. Technical benefits may include improved fuel efficiency and emissions management through accurate identification of fuel characteristics.

[0008] In some examples, optionally including at least one preferred example, fuel characterization involves the fuel composition of the fuel described above. Technical advantages may include accurately identifying the fuel composition to improve fuel efficiency and enable emissions control.

[0009] In some optional examples, including at least one preferred example, the time-series data may include engine speed data. Technical benefits may include enhanced predictive maintenance capabilities through the use of engine performance metrics.

[0010] In several optional examples, including at least one preferred example, the evaluation involves detecting deviations of the functionality of the power system components described above from baseline functionality. Technical benefits may include early detection of potential problems, reduced downtime, and reduced maintenance costs.

[0011] In some optional examples, including at least one preferred example, the evaluation is the determination of the calibration status of the power system components described above. Technical benefits may include ensuring the desired performance and accuracy of the power system components through timely recalibration.

[0012] In some optional examples, including at least one preferred example, the processing circuit is configured to preprocess the time-series data by segmenting the time-series data into multiple data chunks and downsampling the data chunks to a selected frequency in order to increase the availability of samples. Technical advantages may include streamlining data processing and reducing computational load while maintaining data integrity.

[0013] In some cases, including at least one preferred example, the statistical testing methods used may include a combination of the Mann-Whitney U test and the Benjaminy-Hochberg method. Technical advantages may include robust feature selection and control over the false positive rate in hypothesis testing.

[0014] In some cases, including at least one preferred example, the predefined criteria include a combination of cost and gain. Technical benefits may include improving resource allocation by balancing computational and financial costs with achieving desired model performance, thereby increasing the efficiency and effectiveness of the evaluation process.

[0015] In some optional examples, including at least one preferred example, the processing circuit is further configured to receive the operating state of the power system in which the power system components are located. Technical benefits may include situational insights into component performance, improving the accuracy of the evaluation.

[0016] According to a second aspect of this disclosure, a power system including the computer system of the first aspect is provided.

[0017] A second aspect of this disclosure may address the problem of efficiently and accurately evaluating power system components using readily available sensor data. Technical advantages may include providing a computationally efficient and interpretable evaluation method that leverages existing data and provides insights into the physical and chemical processes of power system components, unlike the prior art “black box” approach.

[0018] A third aspect of this disclosure provides a computer-implemented method for evaluating a power system component, the computer-implemented method comprising: receiving time-series data from a sensor monitoring the power system component by a processing circuit of the computer system; pre-processing the time-series data by the processing circuit; extracting a plurality of features from the pre-processed time-series data by the processing circuit; using a statistical test method on the plurality of features to determine the statistical significance of each of the features by the processing circuit; feeding one or more of the features having statistical significance exceeding a predetermined limit to a machine learning model by the processing circuit; and using a machine learning model by the processing circuit The process involves quantifying the contribution of the predicted output from the engine, and executing at least one iterative process by a processing circuit until one or more predetermined criteria are met, the execution of which includes (i) extracting multiple features of the predicted output, (ii) using a statistical test method on the above multiple features of the predicted output, (iii) feeding one or more of the above features of the predicted output that have statistical significance exceeding a predetermined limit to a machine learning model, and (iv) quantifying the contribution of the predicted output from the machine learning model, and the process by which, in response to the above predetermined criteria being met, obtains an evaluation of the power system component based on the predicted output.

[0019] The evaluation may be one of the following: (i) detecting a deviation of the functionality of the power system component from baseline functionality, or (ii) determining the calibration status of the power system component. The processing circuit may be further configured to use the evaluation to initiate one of the following: predictive maintenance of the power system component or recalibration of the power system component.

[0020] A third aspect of this disclosure may address the problem of efficiently and accurately evaluating power system components using readily available sensor data. Technical advantages may include providing a computationally efficient and interpretable evaluation method that leverages existing data and provides insights into the physical and chemical processes of power system components, unlike the prior art “black box” approach.

[0021] In several optional examples, including at least one preferred example, quantifying the contribution involves defining one of the regression, clustering, or classification problems based on the predicted output above, and solving the regression, clustering, or classification problem above to obtain an evaluation. Technical benefits may include increased flexibility in model selection, enabling a more fitted evaluation based on the nature of the data and the desired outcome.

[0022] In several optional examples, including at least one preferred example, time-series data relate to a power system component that is an internal combustion engine, and the evaluation is the determination of the fuel characteristics of the fuel consumed by the aforementioned internal combustion engine. Technical benefits may include improved fuel efficiency and emissions management through accurate identification of fuel characteristics.

[0023] In some optional examples, including at least one preferred example, fuel type determination involves the fuel composition of the fuel described above. Technical benefits may include an optimized combustion process and reduced engine wear due to the accurate identification of the fuel composition.

[0024] In some optional examples, including at least one preferred example, the time-series data may include engine speed data. Technical benefits may include enhanced predictive maintenance capabilities through the use of detailed engine performance metrics.

[0025] In a fourth aspect of the present disclosure, a computer program product including program code for performing the method of the third aspect is provided when executed by a processing circuit.

[0026] The fourth aspect of the present disclosure may attempt to enable a new power system and / or an existing power system to be conveniently configured by software installation / updating in order to efficiently and accurately evaluate power system components using immediately available sensor data. Technical advantages may include providing a computationally efficient and interpretable evaluation method that utilizes existing data and provides insights into the physical and chemical processes of power system components, unlike the "black box" approach of the prior art.

[0027] In a fifth aspect of the present disclosure, a non-transitory computer-readable storage medium including instructions for causing a processing circuit to perform the method of the third aspect is provided when executed by the processing circuit.

[0028] The fifth aspect of the present disclosure may attempt to enable a new power system and / or an existing power system to be conveniently configured by software installation / updating in order to efficiently and accurately evaluate power system components using immediately available sensor data. Technical advantages may include providing a computationally efficient and interpretable evaluation method that utilizes existing data and provides insights into the physical and chemical processes of power system components, unlike the "black box" approach of the prior art.

[0029] As will be apparent to those skilled in the art, aspects, examples (including any preferred examples), and / or the appended claims of the present disclosure may be appropriately combined with each other. Additional features and advantages are disclosed in the following description, the claims, and the drawings, some of which will be readily apparent to those skilled in the art from them, or will be recognized by practicing the present disclosure described herein.

[0030] This specification also discloses a computer system, a control unit, a code module, a computer-implemented method, a computer-readable medium, and a computer program product associated with the aforementioned technical advantages.

[0031] Examples will be further described in detail below with reference to the accompanying drawings.

Brief Description of the Drawings

[0032] [Figure 1] A schematic diagram of an exemplary power system with a computer system according to an embodiment is shown. [Figure 2] A flowchart detailing an exemplary process of data processing from receiving time-series data to obtaining an evaluation of a power system component according to an embodiment is shown. [Figure 3] A method implemented on a computer for evaluating a power system component is shown. [Figure 4] A schematic diagram of a computer system for implementing the embodiments disclosed in this specification.

Modes for Carrying Out the Invention

[0033] Throughout the description, like reference characters refer to like elements.

[0034] The detailed description provided below offers information and examples of the disclosed technology with sufficient detail to enable those skilled in the art to practice the present disclosure.

[0035] This disclosure seeks to address the limitations of current systems by leveraging an advanced approach to evaluating power system components. Time-series data is acquired from existing sensors monitoring power system components, enabling continuous and granular data collection. This data is preprocessed, and multiple features are extracted to identify relevant patterns and attributes that may be overlooked by conventional methods. Statistical testing methods are used to determine the statistical significance of these features, ensuring that only the most influential features are selected. These statistically significant features are fed into a machine learning model, enabling efficient differentiation of complex and overlapping characteristics, such as those found in different fuel types. The contribution of the predictive output is quantified, and an iterative process is introduced. In this iterative process, the steps of feature extraction, statistical testing, and model feeding are repeated until predetermined criteria are met, gradually refining the features and outputs. This iterative approach allows for continuous improvement and refinement of the evaluation, providing a technically advanced solution to address the challenges of current methods. These approaches reduce the need for computationally intensive resources and sensor placement while maintaining specificity, ultimately enabling reliable evaluations that enhance the decision-making process.

[0036] Figure 1 is an exemplary schematic diagram of a heavy vehicle, including a truck 3 and a trailer 5, which functions as an exemplary power system 1 in the context of this disclosure. Heavy vehicles are just one of many potential applications of the approach described herein. Beyond heavy vehicles, this disclosure is applicable to a wide variety of power systems 1 across various types of vehicles and applications, including but not limited to industrial applications such as buses, ships, construction machinery, agricultural machinery, and manufacturing plants; mobile applications such as portable power units and remote monitoring systems; fixed applications such as generator sets, power plants, and HVAC systems; and wind turbines, solar farms, and the like.

[0037] As illustrated, the power system 1 includes one or more power system components 10, and these one or more power system components 10 can encompass a wide range of elements within the power system 1 that can be evaluated. The power system components 10 are essential to the function and efficiency of the power system 1, and their evaluation is performed for purposes such as performance optimization, predictive maintenance, compliance with regulatory standards, and quality control.

[0038] In the context of power system 1, “evaluation” refers to the evaluation and analysis of various components and operating parameters to determine their condition, performance, and efficiency. This generally includes acquiring sensor data, processing sensor data, and identifying patterns, anomalies, and / or potential problems that may affect the functionality of power system components 10. The evaluation aims to provide insights into the health of power system components 10, predict potential failures, and propose maintenance or optimization actions to enhance performance and / or ensure compliance with regulatory standards. The evaluation may encompass qualitative and / or quantitative analysis, including the use of techniques such as statistical testing, feature extraction, and machine learning to inform the decision-making process and derive actionable conclusions that support the efficient and reliable operation of power system 1.

[0039] The power system component 10 may be an internal combustion engine (ICE) whose primary function is to convert fuel into mechanical energy. The performance, fuel efficiency, and exhaust output of the ICE may be areas of evaluation. Other exemplary power system components 10 include a transmission system, fuel system, electrical components, cooling system, exhaust system, hydraulic system, control system and sensors, auxiliary systems, and the like.

[0040] The power system 1 integrates one or more sensors 12 configured to acquire real-time data 30 or time-series data 30 with respect to various operating modes of the power system components 10. This disclosure advantageously utilizes existing sensors 12 to enhance data analysis without requiring the installation of additional sensors, thus maintaining cost-effectiveness and ease of implementation. Examples of such sensors 12 vary depending on which type of power system component 10 they are designed to monitor. In the case of an ICE, the sensor 12 may be an engine speed sensor configured to acquire data on the rotational speed of the ICE. Other sensors 12 may include a temperature sensor to monitor thermal conditions, a pressure sensor to measure fluid or pneumatic pressure in the system, a vibration sensor to detect mechanical imbalances, an acoustic sensor to monitor noise levels, an oxygen sensor to measure exhaust gas composition, a mass airflow sensor to evaluate intake air volume, a knock sensor to detect engine knocking, a fuel level sensor to measure fuel volume, a throttle position sensor to monitor throttle valve position, a lambda sensor to evaluate the air-fuel mixture ratio, and the like.

[0041] The computer system 100, including the processing circuit 102, is configured to process time-series data 30 acquired by the sensor 12 to extract valuable insights and detect patterns or anomalies used to derive the performance and health of the power system component 10 under evaluation. The time-series data 30 refers to a series of data points. These can be collected or recorded at consecutive, equally spaced points in time, providing a continuous flow of information that reflects changes in specific parameters over time. Alternatively, the data stream does not need to be continuous. Downsampling can result in a continuous stream based on the time-series data 30, even if the initial data stream is discontinuous (for example, due to an uncalibrated sensor). For this purpose, the time-series data 30 allows for a more complete understanding of the dynamics of the power system component 10, power system 1, and / or related systems that operationally communicate with power system 1 and / or power system component 10, all while leveraging the existing sensor infrastructure. Finally, the evaluation 50 of the power system component 10 is obtained based on the time-series data 30 and various calculation steps.

[0042] In some examples, the evaluation 50 identifies deviations in the functionality of a power system component 10 from its established baseline functionality. This baseline represents the normal operating conditions and performance metrics that the component should exhibit under standard circumstances. By continuously monitoring and analyzing time-series data 30 from the power system component 10, the computer system 100 can detect anomalies or deviations that indicate potential problems such as wear, damage, or malfunction. For example, if the rotational speed data of the ICE shows irregular fluctuations beyond the expected range or pattern, it may suggest a problem with the ICE's performance and prompt further investigation or maintenance. This type of evaluation 50 is used in predictive maintenance, enabling early detection of problems and reducing downtime by addressing issues before they lead to failure.

[0043] In some examples, evaluation 50 identifies the calibration status of the power system component 10. Calibration involves adjusting the component to ensure its output matches a standard or expected value, thereby maintaining the accuracy and reliability of its operation. By analyzing the time-series data 30 of the power system component 10, the computer system 100 can determine whether it remains within acceptable calibration parameters or whether recalibration is required. For example, if a sensor 12 in the power system 1 consistently provides data that deviates from a known standard, the evaluation may reveal that the sensor 12 is out of calibration. This evaluation 50 may enable timely recalibration, which can be important in the power system 1 where accuracy is related to overall performance and safety.

[0044] As stated above, while evaluation 50 can actually be established for any power system component 10, it should be noted that this disclosure finds a specific area of ​​use in the context of determining the fuel characteristics of the fuel consumed by the power system component 10, which is an ICE. Determining the fuel characteristics may include determining the type and / or composition of the fuel consumed. The shift from conventional diesel to biofuels such as hydrous vegetable oil (HVO) is driven by environmental considerations and the need for more sustainable fuel choices. ICE manufacturers want to offer flexibility in fuel selection rather than imposing strict obligations on the type of fuel. However, accurately determining the fuel characteristics of the fuel used in an ICE is of interest for several reasons, including emissions assessment, regulatory compliance, maintenance needs and costs, and improved fuel supply chain management through demand forecasting and improved logistics. Distinguishing between closely related substances such as biofuels and diesel is difficult due to their similarities, which, while a practical advantage, complicates the distinction. Current sensors and electronic control units (ECUs) may not obtain the data necessary to effectively distinguish between fuels. Furthermore, computationally intensive machine learning models are difficult to deploy in real-time industrial environments. Interpreting these models in relation to their physical and chemical significance presents another challenge.

[0045] To address these issues, the approach described herein utilizes a dual methodology combining time-series feature extraction and automated machine learning (AutoML). This approach leverages time-series data 30 collected by sensor 12 to develop a simpler, less resource-intensive model, providing robust and interpretable predictions in line with the advancement of Industry 4.0. In the examples herein, fuel characteristics can be established based on processed rotational speed data, but it should be understood that other power system components 10 can be evaluated using other sensor readout time-series data 30. Next, the approach used by the computer system 100 will be described in more detail with further reference to Figure 2.

[0046] Figure 2 shows a flowchart detailing an exemplary data processing process from receiving time-series data 30 to obtaining an evaluation 50 of the power system component 10. The actions involved in this exemplary process may be performed under the control of a processing circuit 102 of a computer system 100.

[0047] The first action is to receive time-series data 30 from the sensor 12 that monitors the power system component 10. As described above, the time-series data 30 provides a time series of measurements that reflect the dynamic behavior of the power system component 10 over time.

[0048] An optional action may also be to receive information regarding the operating state of the power system 1 in which the power system component 10 is located. The operating state can provide contextual information that helps to interpret the time-series data 30 more accurately. For example, in the case of an ICE, the operating state may include parameters such as rail pressure, timing of fuel injection initiation, load conditions, throttle position, or engine temperature. In the case of a transmission system, it may include the current gear selection or torque output. In the context of a generator set, the operating state may refer to the output level or frequency stability. By receiving this operating state information, it becomes possible to gain a deeper understanding of the conditions under which the time-series data 30 was collected, thereby potentially increasing the accuracy and relevance of subsequent evaluations of the power system component 10.

[0049] The next action is to preprocess the time series data 30. In the preprocessing stage, the time series data 30 is segmented into smaller, manageable data chunks, and these data chunks are downsampled to selected frequencies. This is to increase the number of samples available for machine learning tasks such as training, validation, and testing. This has been shown to be particularly important when dealing with data from internal combustion engines (ICE), where engine speed data must be carefully segmented to maintain combustion swirl consistency while ensuring a sufficient sample size.

[0050] In an experimental example using ICE engine speed data, raw data collected at a high frequency of 200 kHz was downsampled to a lower frequency. Frequencies of 10 kHz, 1 kHz, and 100 Hz were tested to investigate the effect of sampling frequency on model performance. In the tests performed, the 10 kHz data retained most of the features of the raw data, including small fluctuations and some characteristic peaks. On the other hand, when the frequency was reduced to 1 kHz, these peaks became less pronounced, but some regular small fluctuations remained. Further downsampling to 100 Hz significantly smoothed the fluctuations in the data. Based on these findings, a controlled downsampling coefficient was appropriately selected (in this case, a coefficient of 20). The controlled downsampling coefficient provided a sufficient representation of the characteristics of the raw data without increasing computational requirements or limiting the applicability of results derived from the raw data. Therefore, the raw time-series data 30 is downsampled to be manageable within the constraints of existing sensor technology and computational resources.

[0051] Taking the above into consideration, the processing circuit 102 may be configured to downsample the raw time-series data 30 using one or more pre-selected downsampling coefficients. These coefficients may be adjusted based on the results of a test from a first pre-selected downsampling coefficient. This may be done automatically in response to feedback results.

[0052] The next action is to extract multiple features 32 from the preprocessed time-series data 30. This process specifically involves feature engineering, which is used to fit the data to a selected machine learning model in the classification of the preprocessed time-series data 30. Various other approaches exist in the art of feature extraction, such as distance-based methods (e.g., dynamic time stretching (DTW)), shapelet-based approaches, or deep learning techniques (e.g., CNN / RNN). However, these all require computational resources and large datasets, and often produce "black-box" models that complicate interpretation. Therefore, feature-based methodologies are adopted due to their interpretability, efficiency with smaller datasets, and flexibility in feature selection.

[0053] Generally, feature-based methodologies involve extracting multiple features.32 This approach is advantageous when computational resources and data availability are limited, while high interoperability is required. Using scalable hypothesis testing (tsfresh) based time-series feature extraction, the feature extraction process can be automated to compute and evaluate hundreds of time-series properties or features. These properties or features may include, among other things, several peaks, maximum values, and autocorrelations.

[0054] The tsfresh method generates a comprehensive set of candidate features for further processing. Some selected features computed by the tsfresh method include the unconditional maximum likelihood of an autoregressive process at maximum lag (lag being the delay between data points), linear least squares regression, the presence of a unit root in a time series sample, and autocorrelation at a given lag. Additional features may include selecting a specific range on the graph and the mean change between consecutive points within that range, the complexity-invariant distance (CID) of the time series, the continuous wavelet transform based on Ricker wavelets, the ratio of the sum of squares of a specified segment to the sum of squares of the entire time series, the division of the series into a given number of segments, the Fourier coefficients from the discrete Fourier transform, and the binned entropy of the power spectral density using the Welsh method. Furthermore, features may include the mean of the central approximation of the second derivative, the number of peaks supporting at least a specified value, partial autocorrelation at a particular lag, transposed entropy, and cross-power spectral density. The above examples are not exhaustive; rather, other variations may apply.

[0055] In general, the tsfresh method provides three predefined methods for feature extraction: minimal, efficient, and comprehensive. The minimal method provides features for rapid testing, the comprehensive method provides all features for detailed analysis, and the efficient method balances feature richness with computational efficiency, excluding computationally intensive features. Any of these may be selected depending on various factors such as the type of time series data, the data preprocessing methodology used, and the type of power system component being evaluated. In a specific example of ICE, the efficient method proved to be the most effective method because its ability to balance feature richness with computational requirements made it suitable for a particular practical application.

[0056] The next action is to use statistical testing methods 34 on the extracted features 32 so that their respective statistical significance 36 can be determined. This may be done in combination with the Mann-Whitney U test and the Benjaminy-Hochberg method. Statistical significance 36 represents the likelihood that the observed differences or relationships in the data are not due to chance. In the context of machine learning, the Mann-Whitney U test assesses whether feature distributions differ between groups, such as binary target variables. A low p-value in this test suggests strong differentiation, making this feature a candidate for inclusion in the model. Its non-parametric nature makes it particularly useful for real-world datasets because it is robust to non-normalized data distributions. Higher downsampling frequencies tend to result in lower p-values ​​and indicate greater statistical significance 36 because more data detail is retained. This leads to steeper gradients in identifying important features, improving the accuracy and reliability of the model and thus enhancing the effectiveness of predictive maintenance.

[0057] The term "steep slope" as used above in this text is generated by ranking features based on p-values ​​obtained from statistical testing methods such as the Mann-Whitney U test.34 If features are ranked in ascending order of their p-values, a plot can be created where the x-axis represents the ranked features and the y-axis represents the corresponding p-values. The higher the downsampling frequency, the more detailed the data becomes, and as a result, generally, the p-values ​​of important features become lower. When these p-values ​​are plotted, the line connecting them forms a slope. A steep slope indicates that the p-values ​​decrease rapidly among the top features, highlighting the distinction between the most important and least important features. This slope means that important features can be identified more quickly and clearly, improving the accuracy, reliability, and effectiveness of the model in predictive maintenance. Without ranking, the p-values ​​are randomly distributed, and the plot cannot show a meaningful slope.

[0058] Following the calculation of p-values, the Benjaminy-Hochberg method can be applied to control the false positive rate (FDR) when testing multiple hypotheses. This procedure adjusts p-values ​​to reduce false positives so that the selected features are truly beneficial. The procedure involves sorting p-values, setting a predetermined limit of 38, and rejecting the null hypothesis for p-values ​​below this limit of 38. As the downsampling frequency increases, both the number and proportion of relevant features increase, but the increasing benefit decreases, suggesting a desirable frequency balance.

[0059] The predetermined limit value 38 serves as a benchmark for determining which features 32 are significant enough to be included in a machine learning model for further analysis. The predetermined limit value 38 can be varied based on several factors, such as the specific application, the desired level of confidence, and the characteristics of the dataset being analyzed. In practice, the predetermined limit value 38 is often set to 0.05 as a p-value threshold, indicating that there is only a 5% probability that the observed differences in feature distributions are due to random variation. On the other hand, the predetermined limit value 38 can be adjusted depending on the context and requirements of the analysis. For example, in scenarios where a higher level of certainty is required, a stricter predetermined limit value 38, such as 0.01, may be applied. The selection of the predetermined limit value 38 may also depend on the trade-off between accuracy and recall in the model's performance. In applications where false positives are particularly costly, a lower predetermined limit value 38 may be chosen to ensure that only the most statistically significant features are selected. Conversely, in situations where losing important features is more detrimental, a higher predetermined limit value 38 may be adopted to capture a wider range of features.

[0060] In heuristic decisions, the focus is placed on the most relevant features at each downsampling frequency, as steeper gradients indicate their higher relevance. Different frequencies highlight unique features. At 100 Hz, features relate to time-domain patterns such as trends and autocorrelation. At 10 kHz, on the other hand, features focus on frequency-based information, going beyond the scope of simple time-domain analysis. Lower p-values ​​at higher frequencies support the aforementioned discussion regarding the preservation of data detail. Potential feature redundancy, such as overlap between the real and virtual parts of the Fourier transform, is recognized, ensuring efficient and non-redundant feature selection for model development.

[0061] The next action is to feed features 32 having statistical significance 36 exceeding a predetermined limit 38 into the machine learning model 40. This may be done by using AutoML as a pipeline for the continuous integration / delivery of one or more machine learning models 40. AutoML streamlines the process by reducing the need to manually select and test multiple models. AutoML automates model selection and hyperparameter optimization, initially setting hyperparameters randomly and iterating through them to refine them and improve model performance until a predefined threshold, such as a time limit, is reached. This results in multiple models trained with various hyperparameters for each category, which are then ranked by AutoML based on their performance, eliminating the need for bootstrapping and allowing the process to be driven by the training data. As an example, AutoML may include one or more of the following: decision trees, random forests, logistic regression, XGBoost, and LightGBM, and may automatically select the best model based on the type of time series data 30, extracted features 32, power system components 10, etc.

[0062] The next action is to obtain the predictive output 42 from the best model provided by AutoML, such as the machine learning model 40, and quantify its contribution. This may be done using SHAP (SHapley Additive exPlanations) values, which are then returned to the feature extraction process for iterations of multiple features 32. Thus, the above steps (feature extraction, use of statistical methods, feeding into machine learning mode, and quantification of contribution) are repeated one or more times, with the difference being that these calculations are based on updated features each time. Thus, in each iteration, calculations are performed from the previous iteration. Thus, the calculation is performed once based on the preprocessed time series data 30, and each additional time is performed based on the quantified contribution of the predictive output 42.

[0063] The iterative process is performed one or more times until one or more predetermined criteria are met. The criteria may include a combination of cost and gain. Costs consider computational costs and / or human resource costs in terms of time and money. This includes, for example, conducting an assessment of the resources required to continue the iterative process, including the computational power and time required to train and evaluate Model 40, as well as the economic impact of continuing the process. The gains relate to model performance and include various metrics such as accuracy, F1 score, and other relevant performance indicators. The evaluation of model performance is relevant to the purpose of power component evaluation.

[0064] Costs and gains can be weighted differently based on various factors. For example, in scenarios where safety is a concern, gains may take precedence over costs. Conversely, when capturing an overall trend rather than precise accuracy is desired, cost reduction may be acceptable at the expense of lower gains.

[0065] While cost and gain are the primary considerations, other predefined criteria such as feature significance, minimum variance of predictive output, maximum computation time limit, and maximum number of iterations may potentially be considered. These considerations help fine-tune the balance between cost and gain, ensuring that the iterative process yields valuable, actionable insights.

[0066] The predetermined criteria may be fitted during the calculation. This may be done manually (for example, after n iterations, the criteria should be changed to a different value) or automatically in response to a decision by the machine learning model 40.

[0067] The SHAP values ​​discussed above provide a unified framework by quantifying the contribution of each feature 32 to the model's output 42, based on the principles of cooperative game theory, particularly the Shapley value. This method decomposes the predictions to reveal how individual features influence the model's decision-making process, thereby providing insights into the most influential features.

[0068] Quantifying contributions may involve defining and solving one of several types of machine learning problems, namely regression, clustering, or classification problems. This flexibility allows for the adaptation of the nature of the predictive output 42 and the evaluation 50 to specific purposes.

[0069] In the case of a regression problem, the goal is to predict a continuous value. The processing circuit 102 is configured to use the prediction output 42 to model the relationship between the extracted features 32 and the continuous target variable. This involves determining how the features 32 contribute to the variation of the prediction output 42 and evaluating the model performance using metrics such as mean squared error (MSE) or root mean squared error (RMSE). By iterating against these metrics, feature selection and model parameters can be refined to improve prediction accuracy, ultimately leading to a more accurate evaluation 50 of the power system component 10.

[0070] The classification problem focuses on classifying the predicted output 42 into discrete classes. The processing circuit 102 is configured to analyze which features 32 have the most significant impact on classification, using metrics such as accuracy, precision, recall, or F1 score to measure the effectiveness of the model. By re-evaluating the importance of features and tuning the model based on these metrics, iterative improvements in classification accuracy can be obtained, resulting in a reliable assessment of whether the power system component 10 is functioning as expected or requires intervention.

[0071] The clustering problem involves grouping the predicted outputs 42 into clusters based on similarity, without predefined labels. The processing circuit 102 is configured to use a clustering algorithm to identify patterns or anomalies in the predicted outputs 42 that may indicate an underlying problem or operational state. Metrics such as a silhouette score or Davies-Bouldin index may be used to evaluate the quality of the clustering and guide iterative refinement.

[0072] The goal is to enable the specification of performance metrics regardless of the type of problem. These metrics provide a quantitative measure of model performance, allowing the computer system 100 to identify areas for improvement and iterate through feature selection and model fine-tuning. This iterative process ensures that evaluations 50 are accurate and useful, supporting effective decision-making and predictive maintenance strategies for power system components 10.

[0073] In some cases, kernel SHAP is used for its computational efficiency while maintaining a high level of accuracy. Kernel SHAP minimizes loss and provides a clear explanation of model predictions by assigning weights to each feature subset based on its size and solving a weighted linear regression problem to determine SHAP values. The use of SHAP values ​​in this context not only improves the interpretability of the model but can also inform the feature extraction and selection process, enabling iterative refinement and better alignment of features with predictive targets.

[0074] The illustrated feedback loop (also referred to herein as a repeating process), which combines statistical testing methods 34 with the quantification of predictive outputs 42, provides a favorable method for evaluating power system components 10. The above approach is computationally less expensive compared to more resource-intensive methods such as neural networks, making it feasible for real-time industrial applications. Furthermore, the results are interpretable, providing insights related to underlying physical or chemical processes rather than presenting decision-makers with "black box" artificial intelligence. This interpretability can be used to understand and trust the decisions of the machine learning model 40, ultimately resulting in not only an efficient approach but also a practical and useful approach for decision-making processes in industrial environments.

[0075] In some examples, a method for training a machine learning model 40 is proposed. The training method may utilize the collection of time-series data 30 from an existing sensor 12, as described above. The time-series data 30 can undergo the preprocessing and feature extraction steps described above to extract meaningful features 32 that serve as input for the machine learning model 40. The features 32 are selected based on their statistical significance 36 (or additionally, based on results based on iterations), so that the most impactful data is used to train the machine learning model 40.

[0076] Once identified, feature 32 is fed into machine learning model 40, which is then trained using historical data representing various operating states and conditions of the power system component 10, for example, along with historical time-series data 10. During training, different machine learning algorithms, such as logistic regression, random forest, or decision tree, may be used depending on the specific requirements of the evaluation task. The training process may include cross-validation for parameter refinement. AutoML can be used to automate this process, select the optimal algorithm, and fine-tune hyperparameters to achieve the desired performance.

[0077] The goal of this training method is to develop a machine learning model 40 that predicts the behavior or state of a power system component 10 based on new input data. By incorporating features rigorously tested for statistical significance 36 and employing a structured training process, the trained machine learning model 40 can become a reliable tool for evaluating the functionality of the power system component 10 as described herein. For this purpose, the trained machine learning model 40 can be used for future evaluations, including solving one or more of the following problems: regression problems, clustering problems, or classification problems.

[0078] Figure 3 shows an exemplary computer-implemented method 200 for evaluating a power system component 10. Method 200 is performed by a processing circuit 102 of a computer system 100. The method includes receiving time-series data 30 from a sensor 12 that monitors the power system component 10 (210). Method 200 includes preprocessing the above time-series data 30 (220). Method 200 includes extracting a plurality of features 32 from the preprocessed time-series data 30 (230). Method 200 includes using a statistical test method 34 on the above features 32 to determine the statistical significance 36 of each of the above features 32 (240). Method 200 includes feeding one or more of the above features 32 having a statistical significance 36 that exceeds a predetermined limit value 38 into a machine learning model 40 (250). Method 200 includes quantifying the contribution of the predictive output 42 from the machine learning model (260). Method 200 includes performing at least one iteration until one or more predetermined criteria are met (270). Performing (270) includes extracting multiple features of the predicted output 42 (272). Performing (270) includes using a statistical test method on the above multiple features 32 of the predicted output 42 (274). Performing (270) includes feeding one or more of the above features 32 of the predicted output 42 that have a statistical significance 36 exceeding a predetermined limit value 38 into the machine learning model 40 (276). Performing (270) includes quantifying the contribution of the predicted output 42 from the machine learning model 40 (278). Method 200 includes obtaining an evaluation 50 of the power system component 10 based on the predicted output 42 in response to the above predetermined criteria being met (280).

[0079] Figure 4 is a schematic diagram of an exemplary computer system 400 for implementing the examples disclosed herein. The computer system 400 is adapted to execute instructions from a computer-readable medium to perform these and / or any of the functions or processes described herein. The computer system 400 may be connected to other machines in a LAN (Local Area Network), LIN (Local Interconnection Network), Automotive Network Communication Protocol (e.g., FlexRay), intranet, extranet, or the Internet (e.g., network connectivity). Although a single device is shown, the computer system 400 may include any set of devices that execute a set (or more sets) of instructions individually or collectively to perform any one or more of the methods described herein. Accordingly, any reference in this disclosure and / or claims to a computer system, computing system, computer device, computing device, control system, control unit, electronic control unit (ECU), processor device, processing circuit, etc., includes a reference to one or more such devices that execute a set (or more sets) of instructions individually or collectively to perform one or more of the methods described herein. For example, a control system may include a single control unit, or it may include multiple control units that are interconnected or otherwise interconnected in a communicative manner so that any function performed can be distributed among the control units as desired. Furthermore, such devices may communicate with each other or with other devices of various system architectures, for example, directly or via a controller area network (CAN) bus.

[0080] The computer system 400 may include at least one computing device or electronic device which may include firmware, hardware, and / or execute software instructions for implementing the functions described herein. The computer system 400 may include a processing circuit 402 (e.g., a processing circuit including one or more processor devices or control units), a memory 404, and a system bus 406. The computer system 400 may include at least one computing device having the processing circuit 402. The system bus 406 provides an interface to system components including, but not limited to, the memory 404 and the processing circuit 402. The processing circuit 402 may include any number of hardware components for performing data or signal processing or for executing computer code stored in the memory 404. The processing circuit 402 may include, for example, a general-purpose processor, an application-specific processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a circuit including processing components, a group of distributed processing components, a group of distributed computers configured for processing, or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, or any combination thereof, all designed to perform the functions described herein. The processing circuit 402 may further include computer executable code that controls the operation of the programmable devices.

[0081] The system bus 406 may be any of several types of bus structures that can be further interconnected with the memory bus (with or without a memory controller), peripheral bus, and / or local bus using any of the various bus architectures. Memory 404 may be one or more devices for storing data and / or computer code to complete or facilitate the methods described herein. Memory 404 may include database components, object code components, script components, or other types of information structures to support the various activities described herein. Any distributed or local memory device may be utilized by the systems and methods described herein. Memory 404 may be communicably connected to the processing circuit 402 (e.g., via the circuit or any other wired, wireless, or network connection) and may include computer code for executing one or more processes described herein. Memory 404 may include non-volatile memory 408 (e.g., read-only memory (ROM), erase-programmable read-only memory (EPROM), electrically erasable read-only memory (EEPROM), etc.), volatile memory 410 (e.g., random access memory (RAM)), or any other medium that can be used to carry or store desired program code in the form of machine-executable instructions or data structures and can be accessed by a computer or other machine having processing circuitry 402. Basic input / output system (BIOS) 412 may be stored in non-volatile memory 408 and may include basic routines that help transfer information between elements within the computer system 400.

[0082] The computer system 400 may further include, or be coupled with, non-temporary computer-readable storage media such as a storage device 414, which may include, for example, an internal or external hard disk drive (HDD) (e.g., Extended Integrated Drive Electronics (EIDE) or Serial Advanced Technology Attachment (SATA)), a storage HDD (e.g., EIDE or SATA), flash memory, etc. The storage device 414, as well as other devices associated with computer-readable media and computer-usable media, may provide non-volatile storage such as data, data structures, and computer-executable instructions.

[0083] Hardcoded or softcoded computer code may be provided in the form of one or more modules. A module(s) may be implemented as software and / or hardcoded into circuitry to perform the functions described herein, either entirely or partially. A module may be stored in a storage device 414 and / or volatile memory 410, which may include an operating system 416 and / or one or more program modules 418. All or part of the examples disclosed herein may be implemented as a computer program 420 stored in a temporary or non-temporary computer-available or computer-readable storage medium (e.g., one or more media), such as the storage device 414, including complex programming instructions (e.g., complex computer-readable program code) for causing a processing circuit 402 to perform the actions described herein. Thus, the computer-readable program code of the computer program 420 may include software instructions for implementing the functions of the examples described herein when executed by the processing circuit 402. In some examples, the storage device 414 may be a computer program product (e.g., a readable storage medium) on which the computer program 420 is stored, and at least a portion of the computer program 420 may be loadable (e.g., into a processor) to implement the functions of the examples described herein when executed by the processing circuit 402. The processing circuit 402 may function as a controller or control system for a computer system 400 that implements the functions described herein.

[0084] The computer system 400 may include an input device interface 422 that can be configured to receive inputs and selections communicated to the computer system 400 when executing commands from a keyboard, mouse, touch-sensitive surface, etc. Such input devices may be connected to the processing circuit 402 via the input device interface 422 coupled to the system bus 406, but can also be connected via other interfaces such as a parallel port, an Institute of Electrical and Electronic Engineers (IEEE) 1394 serial port, a Universal Serial Bus (USB) port, or an IR interface. The computer system 400 may include an output device interface 424 configured to transfer outputs to a display, a video display unit (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)), etc. The computer system 400 may include a communication interface 426 suitable for communicating with a network, as appropriate or desired.

[0085] The operational actions described in any of the exemplary embodiments of this specification are described for the purpose of providing examples and discussion. The actions may be performed by hardware components, embodied by machine-executable instructions that cause a processor to perform the actions, or by a combination of hardware and software. A specific order of method actions may be shown or described, but the order of actions may differ. Furthermore, two or more actions may be performed simultaneously or partially simultaneously.

[0086] Example 1: A computer system for evaluating power system components, comprising: receiving time-series data from a sensor monitoring the power system components; preprocessing the time-series data; extracting multiple features from the preprocessed time-series data; using a statistical test method on the multiple features to determine the statistical significance of each of the features; feeding one or more of the features having statistical significance exceeding a predetermined limit to a machine learning model; quantifying the contribution of the predictive output from the machine learning model; and performing at least one iterative process until one or more predetermined criteria are met. A computer system comprising a processing circuit configured to perform a process, the iterative process comprising (i) extracting a plurality of features from the prediction output; (ii) using the statistical test method on the plurality of features from the prediction output; feeding one or more of the features of the prediction output having statistical significance above a predetermined limit to the machine learning model; and quantifying the contribution of the prediction output from the machine learning model; and obtaining an evaluation of the power system components based on the prediction output in response to the predetermined criteria being met.

[0087] Example 2: The computer system according to Example 1, wherein the processing circuit is configured to define one of a regression problem, a clustering problem, or a classification problem based on the prediction output, and to quantify the contribution of the prediction output by solving the regression problem, the clustering problem, or the classification problem in order to obtain the evaluation.

[0088] Example 3: The computer system according to any one of Examples 1 to 2, wherein the time-series data relates to the power system component which is an internal combustion engine, and the evaluation is the determination of the fuel characteristics of the fuel consumed by the internal combustion engine.

[0089] Example 4: The computer system according to Example 3, wherein the fuel characteristic determination includes the fuel composition of the fuel.

[0090] Example 5: The computer system according to any one of Examples 3 to 4, wherein the time-series data includes engine speed data.

[0091] Example 6: The computer system according to any one of Examples 1 to 5, wherein the evaluation is the detection of a deviation of the functionality of the power system component from baseline functionality.

[0092] Example 7: The computer system according to any one of Examples 1 to 5, wherein the evaluation is the determination of the calibration status of the power system components.

[0093] Example 8: A computer system according to any of the prior embodiments, wherein the processing circuit is configured to preprocess the time-series data by segmenting the time-series data into a plurality of data chunks and downsampling the data chunks to a selected frequency in order to increase the availability of the samples.

[0094] Example 9: A computer system according to any of the prior embodiments, wherein the predefined criteria include a combination of cost and gain.

[0095] Example 10: The computer system described in any of the prior examples, wherein the statistical test method used includes a combination of the Mann-Whitney U test and the Benjaminy-Hochberg test.

[0096] Example 11: A computer system according to any of the prior embodiments, wherein the processing circuit is further configured to use an automated machine learning platform as the machine learning model.

[0097] Example 12: A computer system according to any of the prior embodiments, wherein the processing circuit is further configured to quantify the contribution of the predicted output by using SHAP (SHapley Additive exPlanation) values.

[0098] Example 13: A computer system according to any of the prior embodiments, wherein the processing circuit is further configured to receive the operating state of the power system in which the power system components are located.

[0099] Example 14: A power system comprising the computer system described in any of the prior embodiments.

[0100] Example 15: A method implemented in a computer for evaluating a power system component, comprising: receiving time-series data from a sensor monitoring the power system component by a processing circuit of the computer system; preprocessing the time-series data by the processing circuit; extracting a plurality of features from the preprocessed time-series data by the processing circuit; using a statistical test method on the plurality of features to determine the statistical significance of each of the features by the processing circuit; feeding one or more of the features having statistical significance exceeding a predetermined limit to a machine learning model by the processing circuit; quantifying the contribution of the prediction output from the machine learning model by the processing circuit; executing at least one iterative process until one or more predetermined criteria are met, wherein the iterative process includes: extracting features of the prediction output; using the statistical test method on these features; returning significant features to the machine learning model; and quantifying the contribution of the updated prediction output; and obtaining an evaluation of the power system component based on the prediction output when the predetermined criteria are met.

[0101] Example 16: The method of Example 15, wherein quantifying the contribution comprises defining one of a regression problem, a clustering problem, or a classification problem based on the prediction output, and solving the regression problem, the clustering problem, or the classification problem in order to obtain the evaluation.

[0102] Example 17: The method according to any one of Examples 15 to 16, wherein the time-series data relates to the power system component which is an internal combustion engine, and the evaluation is the determination of the fuel characteristics of the fuel consumed by the internal combustion engine.

[0103] Example 18: The computer system according to Example 17, wherein the fuel characteristic determination includes the fuel composition of the fuel.

[0104] Example 19: The computer system described in any of Examples 17 to 18, wherein the time-series data includes engine speed data.

[0105] Example 20: A computer program product comprising program code for performing the method described in any of Examples 15 to 19 when executed by the processing circuit.

[0106] Example 21: A non-temporary computer-readable storage medium that, when executed by the processing circuit, includes an instruction causing the processing circuit to perform the method described in any of Examples 15 to 19.

[0107] The terms used herein are for illustrative purposes only and are not intended to limit the disclosure. Where used herein, the singular forms “a,” “an,” and “the” are intended to include the plural form unless the context otherwise clearly indicates. Where used herein, the term “and / or” includes any combination of one or more and all combinations of the relevant enumerated items. Where used herein, the terms “comprises,” “comprising,” “includes,” and / or “including” identify the presence of a described feature, component, action, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other features, components, actions, steps, operations, elements, components, and / or groups thereof.

[0108] The terms, first, second, etc., may be used herein to describe various elements, but it should be understood that these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, without departing from the scope of this disclosure, the first element may be referred to as the second element, and similarly, the second element may be referred to as the first element.

[0109] Relative terms such as “downward,” “upward,” “upper,” “lower,” “horizontal,” or “vertical” may be used herein to describe the relationship between one element and another as shown in the figure. It will be understood that these terms and the terms above are intended to encompass different orientations of the apparatus in addition to the orientation shown in the figure. When an element is said to be “connected” or “joined” to another element, it will be understood that the element may be directly connected or joined to the other element, or there may be an intervening element. In contrast, when an element is said to be “directly connected” or “directly joined” to another element, there is no intervening element.

[0110] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meanings as those generally understood by those skilled in the art to which this disclosure pertains. It will be further understood that terms used herein should be construed as having meanings consistent with their meanings in the context of this specification and related art, and that no idealized or overly formal meanings shall be assumed unless explicitly defined herein.

[0111] It will be understood that this disclosure is not limited to the embodiments described above and shown in the drawings, and rather, that those skilled in the art will recognize that many changes and modifications may be made within the scope of this disclosure and the appended claims. The drawings and this specification disclose embodiments for illustrative purposes only, and not for limiting purposes, and the scope of this disclosure is set forth in the following claims.

Claims

1. A computer system for evaluating power system components, Receiving time-series data from sensors that monitor the aforementioned power system components, Preprocessing the aforementioned time-series data, Extracting multiple features from the aforementioned preprocessed time-series data, In order to determine the statistical significance of each of the aforementioned features, a statistical test method is used for the multiple features, This involves feeding one or more of the aforementioned features that have statistical significance exceeding a predetermined limit into a machine learning model, To quantify the contribution of the prediction output from the aforementioned machine learning model, Executing at least one iterative process until one or more predetermined criteria are met, wherein the iterative process is: (i) Extracting multiple features from the prediction output, (ii) Using the statistical testing method on the multiple features from the prediction output, (iii) Sending one or more of the features of the prediction output that have statistical significance exceeding a predetermined limit to the machine learning model, (iv) To quantify the contribution of the prediction output from the machine learning model, The above-mentioned actions include, In response to the fulfillment of the predetermined criteria, an evaluation of the power system component is obtained based on the predicted output, wherein the evaluation is one of the following: (i) detection of a deviation of the functionality of the power system component from baseline functionality, or (ii) determination of the calibration status of the power system component. (i) using the evaluation to initiate one of the following: predictive maintenance work on the power system components, or (ii) recalibration of the power system components. The computer system includes a processing circuit configured to perform the following.

2. The computer system according to claim 1, wherein the processing circuit is configured to define one of a regression problem, a clustering problem, or a classification problem based on the prediction output, and to quantify the contribution of the prediction output by solving the regression problem, the clustering problem, or the classification problem in order to obtain the evaluation.

3. The computer system according to claim 1, wherein the aforementioned time-series data relates to the power system component, which is an internal combustion engine.

4. The computer system according to claim 3, wherein the evaluation is the determination of the fuel characteristics of the fuel consumed by the internal combustion engine.

5. The computer system according to claim 4, wherein the determination of the fuel characteristics includes the fuel composition of the fuel.

6. The computer system according to claim 4, wherein the time-series data includes the engine speed related to the internal combustion engine.

7. The aforementioned processing circuit is To increase the availability of the samples, the aforementioned time-series data is segmented into multiple data chunks, The data chunk is downsampled to a selected frequency, The computer system according to claim 1, configured to preprocess the time-series data by...

8. The computer system according to claim 1, wherein the predefined criteria include a combination of cost and gain.

9. The computer system according to claim 1, wherein the processing circuit is further configured to receive the operating state of the power system in which the power system components are arranged.

10. The computer system according to claim 1, wherein the predefined criteria include a combination of cost and gain.

11. The computer system according to claim 1, wherein the statistical testing method used includes a combination of the Mann-Whitney U test and the Benjaminy-Hochberg test.

12. The computer system according to claim 1, wherein the processing circuit is further configured to use an automated machine learning platform as the machine learning model.

13. The computer system according to claim 1, wherein the processing circuit is further configured to quantify the contribution of the predicted output by using SHAP (Shapley Additional explanation) values.

14. The computer system according to claim 1, wherein the processing circuit is further configured to receive the operating state of the power system in which the power system components are arranged.

15. A power system comprising the computer system described in claim 1.

16. A computer-implemented method for evaluating power system components, The computer system's processing circuit receives time-series data from sensors that monitor the power system components, The processing circuit preprocesses the time-series data, The processing circuit extracts multiple features from the preprocessed time-series data, The processing circuit uses a statistical testing method on the multiple features in order to determine the statistical significance of each of the features. The processing circuit sends one or more of the features having statistical significance exceeding a predetermined limit to the machine learning model. The processing circuit quantifies the contribution of the prediction output from the machine learning model, The processing circuit executes at least one iterative process until one or more predetermined criteria are met, and the execution of this process is: (i) Extracting multiple features of the prediction output, (ii) Using the statistical testing method on the multiple features of the prediction output, (iii) Sending one or more of the features of the prediction output that have statistical significance exceeding a predetermined limit to the machine learning model, (iv) To quantify the contribution of the prediction output from the machine learning model, The above-mentioned actions include, The processing circuit, in response to the fulfillment of the predetermined criteria, obtains an evaluation of the power system component based on the predicted output, wherein the evaluation is one of (i) detecting a deviation of the functionality of the power system component from baseline functionality, or (ii) determining the calibration status of the power system component. (i) using the evaluation to initiate one of the following: predictive maintenance work on the power system components, or (ii) recalibration of the power system components. A method implemented in the computer, including the above.

17. Quantifying the aforementioned contributions is Based on the aforementioned prediction output, define one of the following: a regression problem, a clustering problem, or a classification problem. In order to obtain the aforementioned evaluation, the regression problem, clustering problem, or classification problem must be solved, The method according to claim 16, including the method described in claim 16.

18. The method according to claim 16, wherein the time-series data relates to the power system component which is an internal combustion engine, and the evaluation is the determination of the fuel characteristics of the fuel consumed by the internal combustion engine.

19. A computer program product comprising program code for performing the method described in claim 16 when executed by the processing circuit.

20. A non-temporary computer-readable storage medium, which, when executed by the processing circuit, includes an instruction causing the processing circuit to perform the method described in claim 16.

Images