Method and system for condition monitoring of electrical devices

Abstract

According to an aspect of the present inventive concept there is provided a method for condition monitoring of an electrical device. The method comprises: causing the generation of a first stimulus signal at a stimulus initiation position at the electrical device; obtaining a first response to the first stimulus signal using one or more sensors positioned at the electrical device such that the first stimulus signal propagates through at least a portion of interest of the electrical device from said stimulus initiation position to the one or more sensors; causing the generation of a second stimulus signal at the stimulus initiation position at the electrical device in an operational configuration of the electrical device; obtaining a second response to the second stimulus signal using the one or more sensors; comparing the first response and the second response to determine a change between them; and determining a condition of said at least portion of interest of the electrical device based on the comparison.

Description

[0001] METHOD AND SYSTEM FOR CONDITION MONITORING OF ELECTRICAL DEVICES

[0002] Technical field

[0003] The present disclosure relates to methods and systems for condition monitoring of electrical devices.

[0004] Background

[0005] Condition monitoring of electrical devices plays a key role in ensuring their reliability, safety, and efficiency. Various methods and systems have been developed to monitor the condition of such devices. However, prior art is limited to testing devices at a sub-system and disassembled level, which make it challenging to pinpoint the source of defects or anomalies in the electrical devices.

[0006] Moreover, existing condition testing methods are often restricted to assessing specific aspects of a device, rather than providing a comprehensive understanding of its condition. These limitations highlight the need for more effective and efficient approaches to ensure the reliability and safety of electrical devices. Therefore, there is a need for developing methods and systems that can provide accurate, reliable, and relevant information about the condition of electrical devices.

[0007] Summary

[0008] An objective of the present disclosure is to provide a method for monitoring the condition of an electrical device, such as a Medium- to high-voltage (MV / HV) electrical device or an electrical system, in a reliable and efficient manner.

[0009] A further objective is to provide a condition monitoring method that can detect changes in the condition of an electrical device without disrupting its normal operation.

[0010] Yet a further objective is to enable the identification of specific conditions or faults within an electrical device, such as cracks, defects, or wear.

[0011] Additionally, another objective is to provide a method that can generate high-quality training data for a machine learning model to enable accurate and reliable condition monitoring.These and other objectives are at least partly met by the invention as defined in the independent claims. Preferred examples are set out in the dependent claims.

[0012] According to an aspect, there is provided a condition monitoring method for an electrical device, the method comprising: causing the generation of a first stimulus signal at a stimulus initiation position at the electrical device; obtaining a first response to the first stimulus signal using one or more sensors positioned at the electrical device such that the first stimulus signal propagates through at least a portion of interest of the electrical device from said stimulus initiation position to the one or more sensors; causing the generation of a second stimulus signal at the stimulus initiation position at the electrical device in an operational configuration of the electrical device; obtaining a second response to the second stimulus signal using the one or more sensors; comparing the first response with the second response to determine a change between the first response and the second response; and determining a condition of said at least portion of interest of the electrical device based on the comparison.

[0013] Hereby, detection of variations in the condition of at least the portion of interest of the electrical device is facilitated. Depending on the variations in condition of the electrical device, the change between the first response and the second response may correspond to changes in amplitude, waveform, phase shift, frequency, energy, and / or other characteristics of the response. Generally, the use of a first stimulus signal and a, subsequent, second stimulus signal to generate a first response and a second response, allows for identification and analysis of changes between the first response and the second response.

[0014] Further, a change between the first response and the second response may be indicative of a change in the condition of the electrical device. In particular, by comparing the first response with the second response, it is possible to determine the condition of the at least portion of interest of the electrical device, enabling effective condition monitoring of said at least portion. In an example, the method may comprise determining the condition of the at least portion of interest of the electrical device based on the first response and the second response, and more specifically based on the comparison between the first response and the second response.

[0015] Furthermore, the electrical device being in an operational configuration, i.e. , any state in which the electrical device is operating, enables real-timemonitoring of the condition of electrical device during operation, e.g., allowing for faster detection of potential issues and the taking of preventive measures against failures.

[0016] The method also reduces the need for invasive or destructive testing methods, thus decreasing maintenance costs and simplifying testing of electrical devices’ condition. Particularly, by using stimulus signals and one or more sensors to measure responses of the stimulus signals having traveled through at least the portion of interest of the electrical device, a non-invasive and non-destructive method of evaluating the condition of the electrical device is provided. The method thus also enables monitoring of internal or hard-to-reach components of the electrical device, e.g., components whose condition would otherwise only possible to assess by deconstructing the electrical device or removing the electrical device from its operational state.

[0017] Moreover, the method also facilitates proactive maintenance activities and optimized identification of issues at an early stage. This can lead to improved efficiency and reduced downtime of the electrical device.

[0018] The first stimulus signal may be at least one first stimulus signal, generated at, at least one stimulus initiation position. In other words, generation of multiple stimulus signals may be caused at once from multiple positions. This may, e.g., involve the use of orchestrated stimulus sequences, where multiple stimulus signals are triggered in a coordinated manner to provide a more comprehensive understanding of the electrical device’s condition. The use of multiple simultaneous stimulus signals may enable detection of more subtle changes or defects of the electrical device.

[0019] Herein, the term "portion of interest" may refer to any area, part, or component of the electrical device.

[0020] The term “positioned at” is to be understood to entail reference to a location that is on, near, or adjacent to the electrical device. In other words, when it is stated that a sensor or stimulus signal generator is “arranged / positioned at” a particular location, it means that it is located in close proximity to that location, which may include being directly on, immediately adjacent to, or in the immediate vicinity of the electrical device.

[0021] The term “change” as used in the comparison step may refer to any difference or variation between the first response and the second response. The change may manifest in various ways, such as a change in amplitude, phase shift, waveform, frequency, energy, and / or other characteristics of the first and second responses. For example, the change may be an increase ordecrease in the magnitude, a shift in the timing or duration, or a modification in a shape or pattern of a response waveform of the second response.

[0022] The term “stimulus signal” may refer to a disturbance that propagates through at least part of the electrical device. The stimulus signal may, e.g., be a shock wave or a shock signal.

[0023] The first response and the second response may, respectively, be referred to as a first response stimulus signal and a second response stimulus signal.

[0024] The energy of the first stimulus signal and the second stimulus signal may be controlled to ensure that they do not interfere with the normal operation of the electrical device. For example, the first stimulus signal and the second stimulus signal may have lower energy than a mechanical energy resulting from the operation of the electrical device. This may ensure that the stimulus signals do not introduce any significant stress or strain on the electrical device. In other words, the stimulus signals may be configured to not alter a behaviour of the electrical device or cause any damage to the electrical device.

[0025] By introducing a signal with a relatively low energy, in context of the electrical device, it is possible to monitor the condition of the electrical device without disrupting its normal function or causing any wear and tear. Hence, the method may be applied to a variety of devices, without causing any adverse effects.

[0026] The electrical device may be an electrical switching device or a switchgear. The electrical device may, e.g., be configured to control a flow of electrical energy in a power system, reliable operation is therefore essential for maintaining grid stability and preventing power outages.

[0027] In some examples, the method can be used for a variety of families of switchgears, such as any one or more of a Gas Insulated Switchgear (GIS), an Air Insulated Switchgear (AIS), a Dead Tank Breaker (DTB), a Live Tank Breaker (LTB), and a Generator Circuit Breaker (GBS). Further, in some examples, the electrical device may be a drive, e.g., relating to any one or more of Hybrid Mechanisms (HM) and Spring-Spring Drives (SSD).

[0028] In an example, the first stimulus signal and the second stimulus signal, respectively, may have lower energy than a mechanical energy resulting from switching of the electrical switching device or switchgear.

[0029] Generally, the term “electrical device” entail any device or system that uses electrical energy, e.g., such as switchgear or circuit breaker. Aswitchgear may be electromechanical, e.g., incorporating mechanical elements controlled by electrical parts. A circuit breaker may be an electrical device designed to protect an electrical circuit from damage caused by current in excess of that which the equipment can safely carry. A circuit breaker’s function may be to interrupt current flow to protect equipment and to prevent fire. The term “electrical device” may cover actuation technology, e.g., where electrical signals control mechanical movements.

[0030] The first stimulus signal and the second stimulus signal may be steep step signals.

[0031] The steep step signals may be well-suited for causing the generation of responses that are sensitive to changes in the condition of the electrical device. The steepness of the step signals may allow for detection of different conditions. High steepness of the stimulus signals may enable surface assessments (e.g., of cracks or loose screws), while lower steepness may enable assessment of sub-surface properties or components (e.g., ball bearing integrity). Additionally, the use of steep step signals may provide a high signal-to-noise ratio, allowing for accurate and reliable detection of changes in the condition of the electrical device.

[0032] The second stimulus signal may be transmitted through the at least portion of interest at regular time intervals, wherein a response to one of said second stimulus signals may be compared to a response to any other second stimulus signal or to the first response.

[0033] Hereby, continuous monitoring of the electrical device’s condition is provided, enabling the detection of changes or trends over time. By transmitting the second stimulus signal at regular intervals, it is possible to capture responses that can be continuously compared to previous responses, providing a more comprehensive understanding of the condition of the electrical device.

[0034] The method may be designed to be repeatable, e.g., allowing for consistent and reliable measurements overtime. By transmitting stimulus signals at regular intervals, the method can provide a continuous stream of data that may be used to track changes in the condition of the electrical device. The repeatability enables the detection of subtle changes or trends in the condition of the electrical device.

[0035] The comparison of responses may, for example, be done by comparing a response to one of the second stimulus signals to a response to any other second stimulus signal, or to the initial first response. The regulartransmission of the second stimulus signal may also facilitate the development of a baseline understanding of the electrical device’s normal behaviour, thus, e.g., making it easier to detect deviations from such a norm and / or diagnose potential issues.

[0036] The term Tegular time intervals” may refer to a series of time periods between transmissions of the second stimulus signals. For example, the second stimulus signal may be transmitted every minute, every hour, or every day, etc.

[0037] It is to be understood that the wording of transmitting the second stimulus signal at regular time intervals is not to be interpreted to limit the method to only use two stimulus signals, rather the method may encompass transmission of multiple stimulus signals, including third, fourth, fifth, and / or subsequent stimulus signals, and so on. In other words, the method may involve the transmission of a series of stimulus signals, each of which can be compared to previous or subsequent responses.

[0038] Further, it is appreciated that the transmission of the second stimulus signal may be done irregularly, such as in response to a specific event or trigger, or at random intervals.

[0039] The one or more sensors may be positioned to detect localized responses representative of propagation of the first stimulus signal and the second stimulus signal within the at least portion of interest of the electrical device.

[0040] In other words, the one or more sensors may be positioned to capture responses that are unique to specific areas or components of the electrical device. Hence, the method may provide a more detailed and spatially resolved understanding of the electrical device’s condition. In particular, the method may enable identification of potential issues or anomalies that may be localized to specific regions or components of the electrical device.

[0041] Hereby, it is possible to obtain localized responses that provide insight into the stimulus signals propagating through specific portions of, or within, the electrical device. The method hence enables improved detection of potential issues or anomalies that may be limited to specific areas or components of the electrical device.

[0042] In this context, “localized responses” may refer to the specific ways in which the stimulus signals interact with, and is affected by, local, i.e., specific, structures or components of the electrical device.The first response and the second response may be detected by a plurality of sensors placed at different positions at the electrical device.

[0043] The plurality of sensors detecting the first response may, preferably, be the same plurality of sensors detecting the second response. The plurality of sensors may, e.g., be fixedly placed at the electrical device.

[0044] The use of a plurality of sensors positioned at different locations at the electrical device enables a comprehensive understanding of the condition of the electrical device. In particular, the plurality of sensors may capture the first response and the second response from a plurality of perspectives, hence enabling a detailed analysis of the condition of the electrical device.

[0045] Further, by using a plurality of sensors placed at different positions, the method can detect responses that may be unique to specific areas or components of the electrical device. The method may also identify patterns or trends that may not be apparent from a single sensor location. The plurality of sensors may enable creation of an accurate map of the electrical device’s condition, in turn allowing for effective diagnosis and maintenance of the electrical device.

[0046] The first response may be a baseline response measured when the electrical device is in a known healthy condition and / or has passed a factory acceptance test. In other words, the first response of the first stimulus signal may be used as a reference or benchmark response. The baseline response may represent a normal or expected behavior of the electrical device when it is functioning properly.

[0047] The term “healthy condition” may hence be considered as a “good” or “functioning” condition, such as when the electrical device is first installed or after it has undergone a thorough inspection and testing process.

[0048] Hereby, the baseline response provides a reference point against which subsequent second responses can be compared, allowing for the detection of changes or deviations that may indicate a problem or degradation in the condition of the electrical device.

[0049] In particular, the use of a baseline response measured during a known healthy state ensures that the comparison between the baseline response and the second response is meaningful in identifying issues or anomalies in the integrity of the electrical device. The method thus also enables detection of subtle or gradual changes in the device’s condition over time.

[0050] The method may further comprise evaluating the first response and / or the second response using a machine learning model, wherein the machinelearning model may be trained using a dataset of labelled responses from healthy and faulty electrical devices, and wherein the machine learning model may be configured to detect the change between the first response and the second response.

[0051] In other words, the machine learning model may be used to analyze the responses to the stimulus signals and automatically identify any changes or anomalies in the responses that may indicate a problem with the electrical device.

[0052] By training the machine learning model on a dataset of labelled responses from healthy and faulty devices, the machine learning model can learn to recognize patterns and characteristics that are indicative of normal or abnormal behavior of the electrical device. The method hence enables accurate detection of subtle changes in the device’s condition.

[0053] The machine learning model may be trained using data collected from the one or more sensors during different operating conditions of the electrical device, and / or from one or more sensors of a plurality of electrical devices.

[0054] In other words, the machine learning model may be trained to recognize patterns and characteristics of responses to stimulus signals under various operating conditions of the electrical device. Hereby, a comprehensive understanding of the detected responses under various operating conditions of the electrical device is provided.

[0055] Further, by training the machine learning model on data collected from one or more sensors of a plurality of electrical devices, the machine learning model can learn from a larger dataset and / or develop a more generalizable understanding of the behavior of the responses of the stimulus signals traveling through such electrical devices. By leveraging data from multiple electrical devices, the machine learning model can identify common patterns and characteristics that are indicative of healthy or faulty conditions more accurately. Improved accuracy and reliability in detecting changes or anomalies in the response of the stimulus signals are hereby provided.

[0056] The machine learning model may be configured to identify a cause of the change between the first response and the second response, and / or to provide a diagnosis or recommendation for maintenance or repair based on the determined change.

[0057] In other words, the machine learning model may be configured to identify an underlying root cause of the change. The method may hence pinpoint a specific issue or problem with the electrical device. By training themachine learning model on data that comprises labels on root causes of faulty conditions of electrical devices, the machine learning model can learn to recognize patterns and characteristics that are associated with specific types of problems or issues. For example, the machine learning model may identify whether a change in the second response is due to a loose connection, a faulty component, or some other type of issue. Thus, the method enables targeted and effective maintenance and repair actions to be taken, e.g., reducing downtime of the electrical device.

[0058] Alternatively, or additionally, the machine learning model may be configured to provide a diagnosis or recommendation for maintenance or repair based on the determined change. In an example, the diagnosis or recommendation for maintenance or repair may be based on an identified cause of the change. Hereby, the method may provide actionable guidance to maintenance personnel or operators of the electrical device.

[0059] The condition of said at least portion of interest of the electrical device may be related to one or more of: cracks or defects in a material of the electrical device, tension in the material of the electrical device, presence of particles and / or air bubbles in a fluid of the electrical device, or properties of subcomponents of the electrical device, such as ball bearings.

[0060] The tension in the material of the electrical device may be affected by the tightness of screws or other fasteners used to assemble or secure components of the electrical device. For example, if a screw is not tightened properly or has loosened overtime, e.g., due to vibration, thermal expansion, or other factors, the surface tension of adjacent materials may be affected, resulting in changes to a waveform of the stimulus signal as it travels through the portion of interest.

[0061] In other words, the method can detect changes in surface tension, allowing for the identification of screws that are not properly tightened or have loosened.

[0062] The fluid may be any type of liquid or gas that is present within the electrical device, such as hydraulic oil, coolant, or insulating oil. The presence of the fluid may affect a waveform of the stimulus signal as it travels through the portion of interest. Changes in properties or composition of the fluid may indicate potential issues with the electrical device. For example, the presence of air bubbles or particles in the fluid may alter the waveform of the stimulus signal. Further, changes in the fluid’s viscosity or temperature may affect thestimulus signal. Hence, the method can detect potential problems with the electrical device, such as leaks, contamination, or degradation of the fluid.

[0063] Cracks or defects in the material of the electrical device may cause changes in a waveform of the response corresponding to the stimulus signal traveling through said material. Similarly, changes in properties of subcomponents or internal components can cause changes in the waveform, e.g., indicating wear or degradation of such subcomponents or internal components of the electrical device.

[0064] Generally, by analyzing the waveforms of response stimulus signals, the method can identify changes that may indicate a potential problem or issue with the electrical device.

[0065] In other words, there is provided a comprehensive and non-invasive method for monitoring the condition of electrical devices, enabling the detection of potential issues and faults, without interrupting the electrical device’s operational state.

[0066] According to a second aspect, there is provided a method for controlling an operation of an electrical device, wherein the operation of the electrical device is controlled based on a condition of at least a portion of interest of the electrical device determined by the method of the first aspect.

[0067] The second aspect may generally present the same or corresponding advantages as the first aspect.

[0068] The controlling of the electrical device may be achieved by adjusting one or more parameters of the electrical device in response to the determined condition. For example, if the condition monitoring method of the first aspect detects a fault or anomaly in the electrical device, the controlling method may adjust a voltage or current supplied to the electrical device to compensate for the fault. This can help maintain performance and prevent further damage.

[0069] In an example, the controlling method may be used to optimize the operation of the electrical device, e.g., by adjusting parameters such as power consumption, efficiency, or output, based on the determined condition.

[0070] The method for controlling the operation of the electrical device provides a proactive approach to maintaining the health and reliability of the electrical device.

[0071] In other words, a closed-loop control system may be provided, where the condition monitoring method provides feedback to the controlling method, which then adjusts the operation of the electrical device accordingly. Hence,real-time optimization and control of the operation of the electrical device is enabled.

[0072] According to a third aspect, there is provided a system for condition monitoring of an electrical device, the system comprising: one or more stimulus signal generators for generating stimulus signals in at least a portion of interest of the electrical device; one or more sensors; and a processing unit configured to execute the steps of the method according to the first aspect.

[0073] The third aspect may generally present the same or corresponding advantages as the former aspects.

[0074] The stimulus signal generator may be a device or system capable of inducing a stimulus signal in or throughout at least a portion of the electrical device. In other words, the stimulus signal generator may introduce a disturbance or perturbation that generates the stimulus signal. The stimulus signal generator may, e.g., utilize a vibration-inducing mechanism to generate the stimulus signal. The stimulus signal generator may, e.g., apply a mechanical or thermal stimulus to the electrical device.

[0075] In a non-limiting example, the stimulus signal generator may be an actuator, a laser, and / or an existing component of the electrical device, such as a pilot valve.

[0076] The system facilitates the detection of changes in the condition of at least the portion of interest of the electrical device. The changes in condition may be detected by detecting changes in amplitude, phase shift, waveform, frequency, energy, and / or other characteristics of the responses of the stimulus signals generated by the one or more stimulus signal generators.

[0077] The one or more stimulus signal generators may generate stimulus signals that propagate through one or more regions of interests of the electrical device. Correspondingly, the one or more sensors may detect the responses of the one or more stimulus signal generators that are representative of the condition of one or more portions of interest of the electrical device.

[0078] In an example, the system may comprise a plurality of stimulus signal generators, e.g., placed at different locations at the electrical device. Hence, a more comprehensive and detailed analysis of the electrical device’s condition is enabled. In particular, the plurality of stimulus signal generators may allow for generation of multiple stimulus signals that can propagate through the electrical device from various angles and / or directions, e.g., providing a more complete picture of the condition of the electrical device. The use of a pluralityof stimulus signal generators may further enable detection of localized changes or anomalies in the condition of at least the portion of interest of the electrical device.

[0079] The processing unit of the system may be a separate component or part of the one or more sensors. In other words, the system may be designed to have a centralized architecture, e.g., where a single processing unit gathers data from the one or more sensors. Alternatively, a decentralized approach may be considered, e.g., where the one or more sensors have their own processing capabilities to process data locally.

[0080] For example, in a decentralized approach, a stimulus signal generator may send stimulus signals through the portion of interest, and a sensor may collect the data, process it locally, and determine whether to raise an alarm or error signal, e.g., without needing to transmit data to a centralized processing unit.

[0081] In an example, both centralized and decentralized approaches may be utilized. In such a configuration, e.g., feature point machine learning may be implemented on one or more sensors. However, if conditions arise that are not known, e.g., to a classifier or clustering algorithm, the one or more sensors may transmit data to a centralized device for further evaluation and / or cloud-based analysis.

[0082] The one or more stimulus signal generators may be built-in components of the electrical device that are configured to support operation of the electrical device, or wherein the one or more stimulus signal generators may be external devices arranged at the electrical device.

[0083] In the case of being built-in components, the stimulus signal generators may be integrated into the electrical device and / or configured to support normal operation of the electrical device while also generating stimulus signals for condition monitoring purposes. For example, a pilot valve and / or other existing mechanism within the electrical device may be used as a stimulus signal generator, e.g., allowing for efficient and compact system design. Further, when the one or more stimulus signal generators are built-in components of the electrical device, the one or more stimulus signal generators may be configured to work seamlessly, and / or in conjunction, with an existing architecture of the electrical device, e.g., minimizing the need for modifications to the electrical device.

[0084] In contrast, external devices, such as actuators, can be attached at the electrical device and be configured to generate stimulus signals in theelectrical device. When one or more stimulus signal generators are external devices arranged at the electrical device, facilitated maintenance of the stimulus signal generators may be provided. Further, external stimulus signal generators may be more versatile, enabling flexible positioning at one or more portions of interests of the electrical device.

[0085] According to a fourth aspect, there is provided a method for training a machine learning model to classify responses to stimulus signals associated with conditions of an electrical device, the method comprising: obtaining a dataset of labelled responses to stimulus signals having propagated through at least a portion of interest of the electrical device in accordance with the condition monitoring method according to the first aspect, wherein the labels indicate corresponding conditions of the electrical device; and training the machine learning model using the dataset, wherein the machine learning model learns to map the responses to the corresponding conditions of the electrical device.

[0086] The fourth aspect may generally present the same or corresponding advantages as the former aspects.

[0087] The method for training the machine learning model enables the development of an accurate and reliable machine learning model that can be used to classify responses to stimulus signals and determine the condition of at least the portion of interest of the electrical device.

[0088] The labels in the dataset of labelled responses to stimulus signals may indicate corresponding conditions of the electrical device, such as healthy or faulty states, allowing the machine learning model to learn the relationships between the responses and the conditions.

[0089] Effects and features of the first, second, third, and fourth aspects are largely analogous. Examples mentioned in relation to the first, second, third, and fourth aspects are largely compatible. It is further noted that the disclosure relates to all possible combinations of features unless explicitly stated otherwise.

[0090] Brief description of the drawings

[0091] The above, as well as additional objects, features and advantages of the present inventive concept, will be better understood through the following illustrative and non-limiting detailed description, with reference to the appended drawings. In the drawings, like reference numerals will be used for like elements unless stated otherwise.Fig. 1 is a flowchart of a method for condition monitoring an electrical device.

[0092] Fig. 2 is a flowchart of method for training a machine learning model to classify responses to stimulus signals associated with conditions of an electrical device.

[0093] Figs. 3A-B are schematic illustrations of systems for condition monitoring of at least a portion of interest of an electrical device.

[0094] Fig. 4 is a schematic illustration of a system for condition monitoring of at least a portion of interest of an electrical device where a stimulus signal travels through said at least portion of interest of the electrical device.

[0095] Figs. 5A-D are schematic illustrations of systems for condition monitoring of at least a portion of interest of electrical devices where a stimulus signal travels through said at least portion of interest and a feature of the electrical devices.

[0096] Figs. 6A-B are exemplary data plots of readings from one or more sensors connected to a system for condition monitoring of an electrical device where stimulus signals travel through the electrical device.

[0097] Fig. 7 illustrates plots of machine learning-based air bubble detection, showing classification and confidence level result.

[0098] Detailed description

[0099] Fig. 1 illustrates a flowchart of a condition monitoring method 1000 for an electrical device. The method 1000 comprises:

[0100] causing 1100 the generation of a first stimulus signal at a stimulus initiation position at the electrical device;

[0101] obtaining 1200 a first response to the first stimulus signal using one or more sensors positioned at the electrical device such that the first stimulus signal propagates through at least a portion of interest of the electrical device from said stimulus initiation position to the one or more sensors;

[0102] causing 1300 the generation of a second stimulus signal at the stimulus initiation position at the electrical device in an operational configuration of the electrical device;

[0103] obtaining 1400 a second response to the second stimulus signal using the one or more sensors;

[0104] comparing 1500 the first response with the second response to determine a change between the first response and the second response; anddetermining 1600 a condition of said at least portion of interest of the electrical device based on the comparison.

[0105] The step of comparing 1500 the first response with the second response may be performed using any suitable means, such as a processor, a computer, an algorithm, and / or machine learning model. The change between the first response and the second response may indicate a change in the condition of the portion of interest, such as a fault or degradation.

[0106] Although not shown in Fig. 1 , the method 1000 may further comprise evaluating the first response and / or the second response by using a machine learning model. The machine learning model may be trained using a dataset of labelled responses from healthy and faulty electrical devices. Further, the machine learning model may be configured to detect the change between the first response and the second response. Alternatively, or additionally, the machine learning model may be configured to detect patterns in the responses that are indicative of specific conditions, such as faults or degradation.

[0107] The machine learning model may be trained using data collected from the one or more sensors during different operating conditions of the electrical device, and / or from one or more sensors of a plurality of electrical devices.

[0108] In an example, the machine learning model may be configured to identify a cause of the change between the first response and the second response, and / or to provide a diagnosis or recommendation for maintenance or repair based on the determined change. The machine learning model may use various techniques, such as classification, regression, or clustering, to analyse the responses and determine the condition of the portion of interest of the electrical device.

[0109] Further, the condition monitoring method 1000 may comprise using a combination of sensors and / or machine learning models to determine the condition of multiple portions of interest in the electrical device. Each portion of interest may, e.g., be monitored using a different sensor and / or machine learning model.

[0110] In an example, the machine learning model may be configured to detect anomalies in the responses collected from the one or more sensors. The anomalies may indicate changes in the condition of at least the portion of interest of the electrical device, such as a sudden fault or degradation. The machine learning model may use various techniques, such as statisticalprocess control or machine learning algorithms, to detect the anomalies and, e.g., alert operators or maintenance personnel of the electrical device.

[0111] In another example, the machine learning model may be configured to provide a confidence level for the determined condition of the electrical device. The confidence level may indicate the probability that the determined condition is accurate.

[0112] In yet another example, the condition monitoring method 1000 may comprise using a plurality of machine learning models to determine the condition of the electrical device. Each machine learning model may be trained on a different dataset or using a different algorithm. The outputs from each machine learning model may be combined using various techniques, such as averaging or voting, to provide an accurate determination of the condition.

[0113] The condition monitoring method 1000 may further comprise visualizing the responses collected from the one or more sensors. The responses may, e.g., be visualized using graphs, charts, and / or other visualization tools. The visualization may assist operators or maintenance personnel to understand the condition of the electrical device and in making informed decisions about maintenance or repair.

[0114] In particular, the condition monitoring method 1000 may comprise providing alerts or notifications to operators or maintenance personnel when the determined change or condition exceeds a certain threshold or indicates a potential fault or degradation.

[0115] In this context, a threshold may refer to a specific value or level beyond which a change or anomaly is deemed significant enough to warrant attention and / or affect the condition of the electrical device. The threshold may, e.g., be defined based on any one or more of: an overall energy level of a recorded stimulus signal, a maximum shift of an energy peak to lower or higher frequencies, a maximum decrease or increase of the stimulus signal in a time domain, or a point at which a Discrete Fourier Transform (DFT) energy level surpasses a specified value, e.g., in terms of an overall energy level of a recorded frequency bandwidth or an energy level within at least one defined frequency band.

[0116] The threshold may, e.g., be adjusted based on factors such as type of electrical device being monitored, specific application or environment in which the electrical device operates, and / or a level of confidence desired in the condition monitoring results, i.e. readings form the one or more sensors.The threshold may further, or alternatively, be associated with the machine learning model. The threshold may refer to a point at which the machine learning model determines that a response to a stimulus signal is anomalous.

[0117] In the context of machine learning, the threshold may refer to a shift in a probability of a classifier network, e.g., the machine learning model, from a trained “healthy” state to a “health undefined” or “unhealthy” state. In other words, the threshold may be a point at which a confidence level of the machine learning model, indicating that the data is “healthy”, falls below a certain percentage. Further, the condition monitoring method 1000 may comprise storing the responses collected from the one or more sensors in a database or other data storage system. The stored responses may be used to train the machine learning model, predict future conditions, and / or analyse trends over time.

[0118] Fig. 2 illustrates a flowchart of a method 2000 for training a machine learning model to classify responses to stimulus signals associated with conditions of an electrical device. The method 2000 comprises:

[0119] obtaining 2100 a dataset of labelled responses to stimulus signals having propagated through at least a portion of interest of the electrical device in accordance with the condition monitoring method 1000 as discussed in relation to Fig. 1 , wherein the labels indicate corresponding conditions of the electrical device; and

[0120] training 2200 the machine learning model using the dataset, wherein the machine learning model learns to map the responses to the corresponding conditions of the electrical device.

[0121] The dataset may include responses collected under various operating conditions. The responses may be collected from multiple sensors positioned at different locations at the electrical device. The labels associated with each response may be assigned based on expert knowledge or domain expertise. The dataset may also include metadata, such as information about the electrical device, its operating history, environmental conditions, temperature, and / or any maintenance or repair activities performed.

[0122] The trained machine learning model may be updated periodically using new data collected from the electrical device, ensuring that it remains accurate and effective over time.

[0123] Furthermore, although not depicted in the drawings, it is appreciated that a method for controlling an operation of the electrical device can berealised, wherein the operation of the electrical device is controlled based on a condition, of at least a portion of interest of the electrical device, determined by the method of Fig. 1.

[0124] The controlling may be performed using any suitable means, e.g., such as a controller, a processor, an algorithm, or a machine learning model. The controlling may, e.g., involve adjusting parameters, such as voltage, current, frequency, and / or temperature, to influence or optimize the operation of the electrical device based on its condition.

[0125] Fig. 3A schematically illustrates a system 100 for condition monitoring of an electrical device 110, the system 100 comprises: a stimulus signal generator 120 for generating stimulus signals in at least a portion of interest of the electrical device 110; a sensor 130; and a processing unit (not explicitly depicted in Fig. 3A) configured to execute the steps of the method 1000 as discussed in relation to Fig. 1.

[0126] The electrical device 110 depicted in the drawings may include the portion of interest of the electrical device 110. Alternatively, the electrical device 110 depicted in the drawings may be the portion of interest of the electrical device 110.

[0127] In any case, the stimulus signals may propagate through at least the portion of interest of the electrical device 110 from a stimulus initiation position to the sensor 130 positioned at the electrical device 110. The portion of interest may be any part of the electrical device 110 that is desired to be monitored, such as a component, a subsystem, or a specific area. Further, the sensor 130 may be positioned at any suitable location at the electrical device 110 and configured to detect the responses of stimulus signals having propagated through at least the portion of interest of the electrical device 110.

[0128] In Fig. 3A, the stimulus signal generator 120 is an external device arranged on the electrical device 110. However, the stimulus signal generator 120 may be a built-in component of the electrical device 110 that is configured to support operation of the electrical device 110. In other words, the stimulus signal generator 120 may correspond to an internal mechanism and / or an integral part of the electrical device 110, such as a pilot valve.

[0129] In general, the stimulus signal generator 120 may, e.g., be an actuator, a piezo stack, a spring charged mechanics, and / or a linear motor. Further, the sensor 130 may, e.g., be a micro-electro-mechanical systems (MEMS) sensor, a magnetic sensor, and / or a piezo sensor.Although not shown in Fig. 3A, it is appreciated that there may be one or more stimulus signal generators 120, and / or one or more sensors 130. Hence, in an example, one or more stimulus signal generators 120 may be external devices to the electrical devices 110 or built-in components of the electrical device 110. In a particular example, one or more stimulus signal generators 120 may be external to the electrical device 110 while one or more stimulus signal generators 120 may be built-in components of the electrical device 110.

[0130] Further, Fig. 3A depicts the stimulus signal generator 120 being arranged on the electrical device 110. Specifically, in Fig. 3A, the stimulus signal generator 120 is attached to the electrical device 110 via a fastener 121. The fastener 121 may be any suitable fastener, e.g., a screw(s), magnet(s), and / or adhesive(s).

[0131] Similarly, in the example shown in Fig. 3A, the sensor 130 is arranged on the electrical device 110. The sensor 130 is attached to the electrical device 110 via a fastener 131. The fastener 131 of the sensor 130 may be of the same, or a different, kind of fastener as the fastener 121 for the stimulus signal generator 120.

[0132] In Fig. 3A, the stimulus signal generator 120 and sensor 130 are further depicted to have a respective arbitrary stimulus and sensing surface or volume 122, 132. The arbitrary stimulus and sensing surface or volume 122, 132 may, respectively, be a stimulus region and a sensing region. These regions may, generally, correspond to the direct, or substantially direct, portions of the electrical device 110 at which the stimulus signal generator 120 and sensor induces, and respectively senses, stimulus signals propagating through the electrical device 110.

[0133] The positioning or arrangement of the stimulus signal generator 120 and sensor 130 on the electrical device 11 , as shown in Fig. 3A, may be referred to as an invasive, or surface-touching arrangement of the stimulus signal generator 120 and sensor 130.

[0134] In Fig. 3B, an alternative schematic illustration of the system 100 discussed in relation to Fig. 3A is shown. Here, the stimulus signal generator 120 and sensor 130 are arranged at a distance from the electrical device 110. Generally, the positioning or arrangement of the stimulus signal generator 120 and sensor 130 in Fig. 3B may hence be referred to as non-invasive arrangement of the stimulus signal generator 120 and sensor 130.Non-limiting examples of a non-invasive stimulus signal generator 120 may be: a laser, an electromagnetic field, a plasma, and / or a laminar heat flow. Further, non-limiting examples of a non-invasive sensor 130 may be: a laser vibrometer, an electromagnetic sensor, a camera, and / or a microphone.

[0135] Although not explicitly shown in the drawings, it is to be understood that any combination of arrangements, i.e. , positioning, of stimulus signal generators 120 and sensors 130 may be utilized. For example, any suitable combination of invasive, non-invasive, and / or built-in arrangements of the stimulus signal generator(s) 120 and the sensor(s) 130 may be employed.

[0136] However, for brevity, only the invasive signal generator 120 and sensor 130 will be described and depicted in the following examples. Yet, as described above, it is to be readily understood that any combination of types of signal generators 120 and sensor 130 may be used. Hence, hereinafter, the stimulus signal generators 120 and sensors 130 discussed should be regarded as being arbitrary types of stimulus signal generators 120 and sensors 130.

[0137] Fig. 4 provides a schematic illustration of a similar system 100 as discussed in relation to Figs. 3A-B.

[0138] In Fig. 4, the stimulus signal generator 120 is depicted to generate a stimulus signal 140. The stimulus signal 140 propagates though the electrical device 110 and reaches the sensor 130 as a response signal. In other words, the response signal corresponds to the stimulus signal having propagated through at least part of the electrical device 110 and being sensed by the sensor 130.

[0139] The stimulus signal 140 depicted here may correspond to either one of the first stimulus signal and the second stimulus signal.

[0140] The stimulus signal 140 is generated, or induced, at a stimulus initiation position. The stimulus initiation position may be located at any point on the electrical device 110. In an example, the stimulus initiation position may be a position which optimizes the propagation of the stimulus signal 140 through the electrical device 110.

[0141] The stimulus signal 140 may take a variety of forms, in an example, the stimulus signal 140 may be a steep step signal. However, the stimulus signal 140 may, e.g., be a ramp signal or a sinusoidal signal. Further, the stimulus signal 140 may be configured to have any desired characteristics, such as amplitude, frequency, and duration.The second stimulus signal may be transmitted through the at least portion of interest at regular time intervals, wherein a response to one of said second stimulus signals is compared to a response to any other second stimulus signal or to the first response.

[0142] The regular time intervals may be predetermined. For example, the regular time intervals may be stored in a memory device (not explicitly shown in these drawings) of the system 100.

[0143] Further, as the response to one of said second stimulus signals may be compared to a response to any other second stimulus signal, the system 100 may be configured to store the responses in a memory device.

[0144] In the event the response to one of said second stimulus signals is compared to the first response, the first response may have been measured previously and stored in a memory device. The first response may hence serve as a reference point for evaluating changes in the at least portion of interest of the electrical device 110 over time.

[0145] In particular, the first response may be a baseline response measured when the electrical device 110 is in a known healthy condition and / or has passed a factory acceptance test.

[0146] The healthy condition may be verified through a factory acceptance test or other certification process. In other words, it may be ensured that the electrical device 110 operates within predetermined specifications at the time the first stimulus signal is induced and the corresponding first response is registered.

[0147] The sensor 130 may be positioned to detect localized responses representative of propagation of the first stimulus signal and the second stimulus signal within the at least portion of interest of the electrical device. In other words, the sensor 130 may be positioned in close proximity to a particular component or feature of, or within, the electrical device 110, to detect localized responses that are representative of the stimulus signal propagation in that specific area.

[0148] Alternatively, the sensor 130 may be placed at a distance from the components, allowing the sensor 130 to detect more global responses that are indicative of the condition of the electrical device 110.

[0149] However, the localized responses may be influenced by various factors, e.g., including the geometry and material properties of the surrounding structure, the presence of any interfaces or boundaries, andwaveform of the stimulus signal 140. As such, the sensor 130 may be positioned to optimize its sensitivity to specific types of localized responses.

[0150] Furthermore, in the event a plurality of sensors 130 are used, each sensor 130 may be positioned to detect localized responses at different locations at the electrical device 110.

[0151] In an example, the sensor 130 may be configured to detect localized responses in multiple dimensions, such as axial, radial, and / or tangential directions, e.g., to provide a more complete picture of the stimulus signal propagation through at least part of the electrical device 110.

[0152] Although not show in Fig. 4, the first response and the second response may be detected by a plurality of sensors 130 placed at different positions at the electrical device 110. Some, or all, of the plurality of sensors 130 may be connected to a common processing unit or may operate independently. The plurality of sensor 130 may, e.g., transmit their measurements to a central location for analysis.

[0153] Figs. 5A-D schematically illustrate systems 100 for condition monitoring of electrical devices 110. Particularly, Figs. 5A-D provide schematic visual representations of various scenarios in which the system 100 can be utilized to detect changes or anomalies in the electrical device 110. It is appreciated that these systems 100 benefit from the discussions of the systems 100 related to Figs. 3A-B and 4.

[0154] In Fig. 5A, the system 100 is depicted as being capable of detecting surface damage on the electrical device 110. Stimulus signals 140 propagating at a surface of a material of the electrical device 110 are susceptible to changes in said material. Hence, cracks or defects 112 in materials of the electrical device 110, e.g., such as aluminum or steel, can be detected.

[0155] In particular, when a crack or defect 112 is present, it alters at least the waveform of the stimulus signal, allowing the system 100 to detect and diagnose the issue. Generally, the stimulus signal propagation can be influenced by factors such as the material’s elasticity, density, and / or geometry.

[0156] Further, by analyzing the changes in, e.g., the waveforms of the stimulus signals 140, the system 100 may provide information about the location, size, and / or severity of the surface damage. Such information may, e.g., be provided based on the positions of the stimulus signal generator(s) 120 and / or sensor(s) 130.In Fig. 5B, the system 100 is shown being capable of detecting deviating torque in screws or bolts 114 of the electrical device 110. In particular, surface propagating stimulus signals 140 are affected by differences in surface tension. For example, when the torque of screws or other bolts 114 is varied, it alters the surface tension, which in turn affects at least the waveform of the stimulus signal 140.

[0157] Hence, the system 100 can detect changes in the torque of screws or bolts and diagnose potential issues related to lose or over-tightened connections.

[0158] In Fig. 5C, the system 100 is illustrated as being capable of detecting air bubbles 116 in a fluid of the electrical device 110. The fluid may, e.g., be a hydraulic oil or coolant within the electrical device 110.

[0159] Fluids can affect, at least, the waveform of a stimulus signal 140 propagating therethrough. Therefore, also objects altering the fluid, e.g., particles or air bubbles, can further affect at last the waveform of the stimulus signal 140.

[0160] Detection of the presence of air bubbles or other contaminants in a fluid of the electrical device may be indicative of issues such as leaks, blockages, or degradation of the fluid.

[0161] In Fig. 5D, the system 100 is depicted as being capable of detecting changes in subcomponents of the electrical device 110, e.g., such as broken balls in a ball bearing within the electrical device 110.

[0162] The properties of subcomponents can alter, at least, the waveform of the stimulus signals 140. In particular, changes in physical characteristics of subcomponents, such as their shape, size, material, or surface finish, can affect the way the stimulus signals 140 propagate through the electrical device 110. For example, a worn or damaged bearing may introduce additional damping or stiffness that alters the frequency content or amplitude of the stimulus signals, while a cracked or corroded component may change the acoustic impedance or reflectivity of the material, causing the stimulus signals 140 to be scattered or attenuated in a unique way. Similarly, changes in the mechanical properties of subcomponents, such as their resonance frequencies or modal shapes, may also affect the waveform of the stimulus signals 140.

[0163] Further, frequency content and / or amplitude of the stimulus signals 140 may be altered by changes in the mechanical properties of subcomponents and / or materials of the electrical device 110. For example, changes in theYoung's modulus and / or compression module of a material may affect its stiffness and density, e.g., leading to changes in the responses to the stimulus signals 140 in at least a portion of interests of the electrical device 110. Similarly, variations in the acoustic impedance or reflectivity of a material may cause the stimulus signals 140 to be scattered or attenuated in unique ways.

[0164] Hence, the system 100 can detect issues related to worn or damaged subcomponents, e.g., internal components of the electrical device 110.

[0165] Generally, the condition of the at least portion of interest of the electrical device 110 may be related to one or more of: cracks or defects in a material of the electrical device 110, tension in the material of the electrical device 110, presence of particles or air bubbles in a fluid of the electrical device 110, or properties of subcomponents of the electrical device 110.

[0166] Figs. 6A-B illustrate exemplary data plots of readings from one or more sensors 130 in accordance with the system 100 or method 1000 discussed in relation to any of Figs. 1 , 3A-B, 4, and 5A-D.

[0167] The readings in Figs. 6A-B are depicted with an arbitrary ehnergy unit. Each data point represents an energy, i.e. , response to a generated stimulus signal, detected by the sensor.

[0168] Fig. 6A illustrates a graph 200 of readings from a sensor at an electrical device where the torque of two screws, denoted ‘screw T and ‘screw 2’, of the electrical device have been varied.

[0169] In particular, the graph 200 in Fig. 6A illustrates the overall energy extracted from the stimulus signals over respective data sets in consecutive order. That is, the data sets are not randomly plotted, but rather in a timely manner, allowing for a clear understanding of how the energy changes over time. In Fig. 6A, the frequency band selected for calculating the overall energy was 12500 Hz to 22500 Hz, which here provides a sufficient resolution to capture the relevant features of the stimulus signals.

[0170] The graph 200 highlights the different clusters of different conditions that are present in the data sets, indicating that the system and method is capable of distinguishing between various types of faults and / or anomalies. Notably, in certain frequency bands, the condition monitoring results may exhibit non-uniqueness due to the high degree of similarity between cluster levels. However, this may be mitigated by selecting an alternative frequency band for analysis.A first part 210 of the graph 200 shows readings corresponding to ‘screw T having a 100% torque, i.e., 100% of the applied torque being converted into a clamping force.

[0171] In a second part 220 of the graph 200, readings are shown for when ‘screw T has a 10% torque, whereas a third part 230 shows readings associated with a quarter turn of ‘screw T.

[0172] Further, a fourth part 240 of the graph 200 shows readings corresponding to a second screw, ‘screw 2’, having 10% torque, and, in a fifth part 250 of the graph 200, readings associated with a quarter turn of ‘screw 2’ are shown.

[0173] As seen in Fig. 6A, the readings, or responses, for each different torque of ‘screw T and ‘screw 2’, correspond to distinct energies levels.

[0174] Particularly, the first part 210 shows a steady state response associated with 100% torque for ‘screw T, where the readings fluctuate within a certain threshold around a certain mean value. Similarly, the second part 220 reveals a decrease in energy level when ‘screw T is set to 10% torque, with readings fluctuating within a second threshold around a second mean value, and the third part 230 demonstrates characteristic energy levels for a quarter turn of ‘screw T, where the readings fluctuate within a third threshold around a third mean value. The fourth and fifth parts 240, 250 of the graph 200 show similar trends for ‘screw 2’, with distinct energy levels associated with each different torque setting.

[0175] The data points in these regions exhibit a consistent pattern, indicating that the sensor is accurately detecting changes in the torque of the screws.

[0176] Hence, these results demonstrate the ability of the system and method to detect even subtle changes in the electrical device’s condition related to torque of screws of the electrical device.

[0177] Fig. 6B illustrates a graph 300 of readings from a sensor at an electrical device where air bubbles have been introduced in a fluid flowing at least partly through the electrical device. The graph of Fig. 6B was plotted similarly as described in relation to Fig. 6A.

[0178] The readings in the first part 310, associated with presence of air bubbles, of the graph 300 in Fig. 6B show a consistent energy level of with some fluctuations within a certain threshold and around a mean value. In other words, the sensor is detecting a stable and predictable response from stimulus signals traveling through at least part of the electrical device.In contrast, a second part 320 of the graph 300 in Fig. 6B shows a general increase in energy level when air bubbles are absent from the fluid flowing through at least part of the electrical device. The difference between the cluster of the first part 310 and second part 320 provides information about a deviation in impedance caused by air bubbles. Thus, it may also be accurately detected when air bubbles are present in a fluid within the electrical device.

[0179] Hence, by analyzing the energy levels of the readings, the systems and methods discussed herein can provide insights into the electrical device’s condition and identify potential issues before they become major problems or result in failure of the electrical device.

[0180] Fig. 7 illustrates the application of a machine learning model for identifying air bubbles within an electrical device. The upper portion 400 of Fig. 7, demonstrates the effective classification of data sets into good and bad conditions, i.e. , absence and presence of air bubbles, respectively. The lower portion 500 of Fig. 7 shows the confidence level of a time series forest, highlighting the robustness and reliability of the machine learning model.

[0181] Hence, the machine learning model may detect even small deviations between response of stimulus signals traveling through at least part of the electrical device.

[0182] In the above the inventive concept has mainly been described with reference to a limited number of examples. However, as is readily appreciated by a person skilled in the art, other examples than the ones disclosed above are equally possible within the scope of the inventive concept, as defined by the appended claims.

Claims

27CLAIMS1. A condition monitoring method (1000) for an electrical device, the method comprising:causing (1100) the generation of a first stimulus signal at a stimulus initiation position at the electrical device;obtaining (1100) a first response to the first stimulus signal using one or more sensors positioned at the electrical device such that the first stimulus signal propagates through at least a portion of interest of the electrical device from said stimulus initiation position to the one or more sensors;causing (1300) the generation of a second stimulus signal at the stimulus initiation position at the electrical device in an operational configuration of the electrical device;obtaining (1400) a second response to the second stimulus signal using the one or more sensors;comparing (1500) the first response with the second response to determine a change between the first response and the second response; anddetermining (1600) a condition of said at least portion of interest of the electrical device based on the comparison.

2. The method according to claim 1 , wherein the electrical device is an electrical switching device or a switchgear.

3. The method according to claim 1 or 2, wherein the first stimulus signal and the second stimulus signal are steep step signals.

4. The method according to any of claims 1-3, wherein the second stimulus signal is transmitted through the at least portion of interest at regular time intervals, wherein a response to one of said second stimulus signals is compared to a response to any other second stimulus signal or to the first response.

5. The method according to any of claims 1-4, wherein the one or more sensors are positioned to detect localized responses representative of propagation of the first stimulus signal and the second stimulus signal within said at least portion of interest of the electrical device.

6. The method according to any of claims 1-5, wherein the first response and the second response are detected by a plurality of sensors placed at different positions at the electrical device.

7. The method according to any of claims 1-6, wherein the first response is a baseline response measured when the electrical device is in a known healthy condition and / or has passed a factory acceptance test.

8. The method according to any of claims 1-7, further comprising:evaluating the first response and / or the second response using a machine learning model, wherein the machine learning model is trained using a dataset of labelled responses from healthy and faulty electrical devices, and wherein the machine learning model is configured to detect the change between the first response and the second response.

9. The method according to claim 8, wherein the machine learning model is trained using data collected from the one or more sensors during different operating conditions of the electrical device, and / or from one or more sensors of a plurality of electrical devices.

10. The method according to claim 8 or 9, wherein the machine learning model is configured to identify a cause of the change between the first response and the second response, and / or to provide a diagnosis or recommendation for maintenance or repair based on the determined change.

11. The method according to any of claims 1 -10, wherein the condition of said at least portion of interest of the electrical device is related to oneor more of: cracks or defects in a material of the electrical device, tension in the material of the electrical device, presence of particles and / or air bubbles in a fluid of the electrical device, or properties of subcomponents of the electrical device such as ball bearings.

12. A method for controlling an operation of an electrical device, wherein the operation of the electrical device is controlled based on a condition of at least a portion of interest of the electrical device determined by the method (1000) of any of claims 1-11.

13. A system (100) for condition monitoring of an electrical device (110), the system (100) comprising:one or more stimulus signal generators (120) for generating stimulus signals (140) in at least a portion of interest of the electrical device (110);one or more sensors (130); anda processing unit configured to execute the steps of the method (1000) according to any one of claims 1-11.

14. The system according to claim 13, wherein the one or more stimulus signal generators are built-in components of the electrical device that are configured to support operation of the electrical device, or wherein the one or more stimulus signal generators are external devices arranged at the electrical device.

15. A method (2000) for training a machine learning model to classify responses to stimulus signals associated with conditions of an electrical device, the method comprising:obtaining (2100) a dataset of labelled responses to stimulus signals having propagated through at least a portion of interest of the electrical device in accordance with the condition monitoring method according to any of claims 1-11, wherein the labels indicate corresponding conditions of the electrical device; andtraining (2200) the machine learning model using the dataset, wherein the machine learning model learns to map the responses to the corresponding conditions of the electrical device.

Images