Predicting electrical component failures

Abstract

The goal is to predict electrical component failure. A method, system, and apparatus, including a media-encoded computer program product, for predicting electrical component failures may be provided. A first sensor measurement of a component of an electrical distribution grid taken at a first time may be obtained. A second sensor measurement of the component taken at a second time may be identified, which may be after the first time. Input, which may include the first sensor measurement and the second sensor measurement, may be processed using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a certain type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement. Data indicative of the prediction may be provided for presentation by a display.

Description

[Technical Field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims priority to U.S. Provisional Application No. 63 / 350,174, filed June 8, 2022. The disclosure of the prior application is considered part of the disclosure of this application and is incorporated by reference into the disclosure of this application.

[0002] TECHNICAL FIELD This disclosure relates to power distribution grids, and in particular to processes for predicting failures of components of a power distribution grid. [Background technology]

[0003] Electric power utilities have hundreds of thousands of assets deployed in the field. When an asset fails (e.g., a transformer explodes), the failure can cause widespread power outages and pose life-threatening risks. To prevent failures, simple heuristics can be used to determine when upgrades and replacements are recommended. For example, a utility may have a policy to replace transformers after a fixed operating period (e.g., 20 years). However, while simple heuristics can be used to make approximate predictions, they can overpredict and underpredict failures. In the case of overpredicted failures, equipment is replaced prematurely, resulting in wasted costs and materials. In the case of underpredicted failures, equipment fails unexpectedly, potentially with catastrophic results. For example, a time-based heuristic can be used to determine when to replace a transformer, but the heuristic may overpredict the failure of a lightly loaded transformer in a milder environment or underpredict the failure of a heavily loaded transformer in a hotter environment. Summary of the Invention

[0004] Generally, this specification relates to a process for predicting component failures in an electrical distribution grid, and more specifically, the present disclosure relates to using two or more time-series sensor measurements as inputs to a machine learning model configured to predict component failures.

[0005] One aspect features obtaining a first sensor measurement of a component of an electrical grid taken at a first time. A second sensor measurement of the component taken at a second time can be identified, which may be after the first time. The input, which may include the first sensor measurement and the second sensor measurement, can be processed using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of fault during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement. The time interval can be a period of time after the second time. Data indicative of the prediction can be provided for presentation by a display.

[0006] In some implementations, the sensor measurements may be images, such as optical or thermal images, hi some implementations, the sensor measurements may be acoustic recordings.

[0007] The prediction may include one or more of the following characteristics: The machine learning model may include a defect detection machine learning model and a failure prediction machine learning model. The machine learning model may include a failure prediction machine learning model. The failure prediction machine learning model may include a defect detection hidden layer. The prediction may include one or more of the likelihood that the component will fail over a single time period, the likelihood that the component will fail over each of multiple time periods, the mean time to failure, a distribution of failure probabilities, or the period in which the component is most likely to fail. The component characteristics may include one or more of bulges, tilts, loose fasteners, missing fasteners, cracks, burn marks, rust, oil leaks, missing or damaged insulation, operating noise, or thermal quality. The machine learning model may be a recurrent neural network. The recurrent neural network may be a long short-term memory machine learning model or a cross-attention-based transducer model. The input may further include characteristics of the component and characteristics of the operating environment. The characteristics of the operating environment may include a series of temperature values ​​measured at or around the component. The input, which may include a first sensor reading and characteristics of the operating environment, may be processed using a machine learning model configured to generate a prediction representing a recommended time for capturing one or more subsequent sensor readings of the component.

[0008] In some implementations, the first and second sensor measurements are images of the component. A first acoustic recording of the component of the power distribution grid taken at a first time can be obtained. A second acoustic recording of the component taken at a second time can be identified. A second input, which can include the first acoustic recording and the second recording, can be processed using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of fault during the time interval based on one or more changes in one or more characteristics of the component as represented in the second acoustic recording compared to the first acoustic recording. Data provided for presentation by the display can be determined based on a weighted combination of the prediction and the second prediction.

[0009] In some implementations, the first and second sensor measurements are optical images of the component. A first thermal image of the component of the power distribution grid taken at a first time can be acquired. A second thermal image of the component taken at a second time can be identified. The second input, including the first thermal image and the second thermal image, can be processed using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of failure during the time interval based on one or more changes in one or more characteristics of the component as depicted in the second thermal image compared to the first thermal image. Data provided for presentation by the display can be determined based on a weighted combination of the prediction and the second prediction.

[0010] Certain embodiments of the subject matter described herein can be implemented to achieve one or more of the following advantages: The techniques described below can be used to predict component failure using a series of sensor measurements, such as images of a component taken over a period of time. By using multiple images of the component, the system can determine the change to the component's defects, including the rate of change, to generate a more accurate reliability prediction. The system can also generate a more accurate reliability prediction by using predictions based on different types of sensor measurements, such as images and audio recordings, or different types of images. The system can also generate a more accurate reliability prediction by using characteristics of the component's operating environment.

[0011] The details of one or more embodiments of the subject matter described herein are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the invention will become apparent from the description, drawings, and claims. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is an illustration of component failure over a period of time. [Figure 2] 1 is an illustration of component failure over a period of time. [Figure 3A] FIG. 1 is a diagram of an exemplary system for predicting electrical component failure. [Figure 3B] FIG. 1 is a diagram of an exemplary system for predicting electrical component failure. [Figure 4] FIG. 1 is a flow diagram of an exemplary process for predicting electrical component failure. [Figure 5] 1 is an illustration of a component defect that may be detectable in a thermal image over a period of time. [Figure 6] FIG. 1 is a block diagram of an exemplary computer system. DETAILED DESCRIPTION OF THE INVENTION

[0013] This specification describes techniques for predicting the likelihood of failure of components in an electrical grid over one or more specified time periods. The techniques can include evaluating sensor measurements of the components taken at multiple times. For example, the sensor measurements can include image data. For example, FIGS. 1 and 2 are illustrations of component defects visible in images over a period of time. FIG. 1 depicts a transformer 100 over five time periods: 1990, 1995, 2000, 2005, and 2010, with the amount of rust 110, 120, 130a, 130b, 140a, 140b, and 140c increasing over time. For example, in 1990, the transformer 100 exhibits no rust. In 1995, the transformer 100 has one small rust spot 110. In 2000, the transformer 100 exhibits a larger rust spot 120. By 2005, the transformer 100 contains a large rust spot 130a and a second, smaller rust spot 130b. By 2010, the transformer 100 contains a very large rust spot 140a and two smaller rust spots 140b, 140c.

[0014] Both the presence of defects (in this example, rust) and the rate of change of the defects can be used to predict component failure. In Figure 1, the amount of rust increases over time, which can predict failure, for example, if the rust coverage exceeds a threshold value and the component is no longer able or unlikely to function properly.

[0015] In contrast, FIG. 2 shows a transformer 200 with rust 210a, 210b, 220a, and 220b for two time periods: 1990 and 2020. The amount of rust, while significant, changed little over the 30-year period. If the unit did not fail due to rust over this 30-year period, the slow rate of spread may indicate a low probability that rust will cause failure over the next several years. While FIGS. 1 and 2 illustrate rust as an example, a wide variety of defects are possible. Examples of defects visible in the images may include bulges, tilt, loose or missing fasteners, cracks, burn marks, rust, oil leaks (e.g., oil stains), missing or damaged insulation, or thermal quality, among others.

[0016] For these reasons, systems that consider only a single image and therefore cannot assess not only the presence of a defect but also the rate of change of the defect may miss predictive signals of a fault or non-fault. Accordingly, this specification describes techniques for determining predictions by using machine learning models that evaluate signals from multiple time periods.

[0017] Additionally, systems that consider other types of sensor measurements can evaluate more predictive signals. For example, the system can consider image data such as thermal images or audio recordings.

[0018] FIG. 3A is a diagram of an example system 300 for predicting electrical component failure. Briefly, the system 300 can use a defect detection machine learning model to process inputs including sensor measurement data to determine which defects, if any, exist on the component. For example, the defect detection machine learning model can be a classification machine learning model, such as a convolutional neural network or any other suitable type of classification model. The system 300 can provide the sensor measurements to the defect detection machine learning model, which can determine an output including an encoding of the sensor measurements. The encoding can include an indication of the presence and type of defect. The system 300 can use the defect detection machine learning model to process sensor measurements of the component taken at different times, as described below, and use the multiple outputs as inputs to a failure prediction machine learning model.

[0019] A sensor measurement of a component may be obtained by a sensor for a particular point in time. For example, the sensor measurement may be a photographed image of the component or an audio recording taken near the component. The audio recording may, for example, capture sounds made by the component.

[0020] 3A, the sensor measurements are images. The images may be optical images. In some implementations, the images may be other types of images, such as thermal images.

[0021] Thus, system 300 can use a defect detection machine learning model to process inputs including images to determine what, if any, defects are present on the component. System 300 can provide the images to the defect detection machine learning model, which can determine an output including an encoding of the image. The encoding can include an indication of the presence and type of defect. System 300 can use the defect detection machine learning model to process images of the component taken at different times, and the multiple outputs can be used as inputs to a failure prediction machine learning model, as described below.

[0022] Examples of defects may include bulges, tilt, loose or missing fasteners, cracks, burn marks, rust, oil leaks (e.g., oil stains), missing or damaged insulation, operating noise, or thermal quality, among others.

[0023] Image data can be obtained from a variety of sources. For example, the component owner can capture images at periodic intervals. Images can also be obtained from other parties, for example, from vehicles that include cameras, such as autonomous vehicles, from photo-sharing websites (if the photo owner approves such use), etc.

[0024] To determine the likelihood of failure, the system may process inputs including the output of the defect detection machine learning model for two images of the component with a failure prediction machine learning model configured to generate predictions related to failure of the component over a period of time.

[0025] The inputs may further include grid maps, component characteristics, and operating environment characteristics. Operating environment characteristics may include, but are not limited to, the number and timing of blackouts, brownouts, lightning strikes, and blown fuses, as well as weather conditions (e.g., temperature and humidity). Additionally, operating environment characteristics may include one or more series of values. For example, such a series may include temperature values ​​measured at or around the component location at multiple points in time.

[0026] In implementations where the sensor measurements include thermal images, the system can use characteristics of the operating environment to distinguish between changes in the component and changes in the environment. For example, thermal images may be taken at different times of the year or under different environmental conditions. Different environmental conditions may affect the temperatures present in the thermal images. Thus, the system can use characteristics such as the temperature of the environment to compare the thermal quality of the component at different times, isolated from changes in the environment.

[0027] In implementations in which the sensor measurements include thermal images, for example, the system can use characteristics of the operating environment to determine the thermal quality of the component. For example, the system can determine the ambient temperature of the component's environment using temperature values ​​measured at or around the component taken at a time within the same time window in which the component's thermal image was taken. Thus, the system can obtain temperature information by comparing the temperature present in the thermal image to the ambient temperature. As another example, the system can use weather conditions such as humidity to perform moisture analysis. For example, moist air, i.e., air with a higher humidity, has a higher heat capacity and is a better heat conductor than dry air. The moisture conditions of the air surrounding the component can affect the temperature of the component. Thus, the system can use the thermal image and humidity information to determine the thermal quality of the component in the context of the environment.

[0028] Component characteristics may include, but are not limited to, make, model, age, rating, thermal constants, winding type, and load metrics (maximum load, average load, time under maximum load, etc.) The grid map may include, for example, the components present in the grid, their interconnection patterns, and the distances between the elements.

[0029] The defect detection machine learning model may be a neural network. In some implementations, the defect detection machine learning model is a long short-term memory (LSTM) model. LSTM models differ from feedforward models in that they can process sequences of data, such as sensor measurements (or output from processing sensor measurements) of a component over multiple time periods. In some implementations, the defect detection machine learning model is a cross-attention-based transformer model.

[0030] Examples of predictions of a failure prediction machine learning model may include, but are not limited to, the likelihood that a component will fail over a single time period, the likelihood that a component will fail over each of multiple time periods, the mean time to failure, and the most likely period in which a component will fail. Additionally, the failure prediction machine learning model may be configured to generate one or more of these outputs.

[0031] Failure prediction machine learning models can be evaluated in response to various triggers, for example, the models can be evaluated whenever new data arrives (e.g., images of components), at periodic intervals, and when a user requests an evaluation (e.g., during a maintenance planning exercise).

[0032] In some implementations, the defect detection machine learning model is a component of the failure prediction machine learning model (described above). For example, defect detection may be performed by one or more hidden layers in the failure prediction machine learning model, and outputs from those layers may be used by other layers of the failure prediction machine learning model.

[0033] The system 300 can train a failure prediction machine learning model using training examples that include feature values ​​and results. The results can indicate whether a component has failed during a given period of time. For example, a value of "1" can indicate a failure, and a value of "0" can indicate no failure. The feature values ​​can include two or more images of the component, a grid map, features of the component, and features of the operating environment, as described above.

[0034] System 300 may include a feature acquisition engine 310, an image identification engine 320, an evaluation engine 330, and a prediction-providing engine 340. Engines 310, 320, 330, and 340 may be provided as one or more computer-executable software modules, hardware modules, or a combination thereof. For example, one or more of engines 310, 320, 330, and 340 may be implemented as blocks of software code having instructions that cause one or more processors of system 300 to perform the operations described herein. Additionally or alternatively, one or more of engines 310, 320, 330, and 340 may be implemented in an electronic circuit, such as, for example, a programmable logic circuit, a field programmable logic array (FPGA), or an application-specific integrated circuit (ASIC).

[0035] The feature acquisition engine 310 can acquire feature data related to a component failure. The feature data can include, but is not limited to, images 305a, 305b of electrical components and of elements related to the potential failure of the electrical components, such as structural support elements. Examples of components can include, but are not limited to, transformers, fuses, electrical wires, and associated structures such as utility poles, cross arms, insulators, and lightning arrestors.

[0036] Visual indicators related to component failure that may be present in the images 305a, 305b may include defects in the component itself, in any supporting structure (e.g., a utility pole that may begin to lean over time), such as rust (as illustrated in FIGS. 1 and 2), cracks, holes, deformations, etc., or combinations thereof. Indicators related to component failure that may be present in the thermal image may include, for example, higher than normal operating temperatures, i.e., hot spots, on the component. The images may be encoded in any suitable format, including, but not limited to, joint photographic expert group (JPEG), Tag Image File Format (TIFF), or a lossless format such as RAW.

[0037] In some implementations, the feature acquisition engine 310 can acquire additional feature data. For example, the additional feature data may include a grid map, component features, and operating environment features. Operating environment features may include, but are not limited to, the number and timing of blackouts, brownouts, lightning strikes, and blown fuses, as well as weather and environmental conditions (e.g., temperature, humidity, vegetation levels). Component features may include, but are not limited to, manufacturer, model, age, rating, thermal constants, winding type, service history, and load metrics (maximum load, average load, time under maximum load, etc.). A grid map may include, for example, the components present in the grid, their interconnection patterns, and the distances between the elements.

[0038] The feature data may further include metadata describing the feature data, such as a timestamp of the feature data (e.g., the date and time the image was captured), a timestamp of when the feature data was obtained, a location (e.g., the location of the image capture device and / or the object captured in the image, as provided by GPS or other means), the provider of the feature data, and an asset identifier (e.g., provided by the person capturing the image of the asset).

[0039] The feature acquisition engine 310 can acquire feature data using various techniques. In some implementations, the feature acquisition engine 310 retrieves feature data from data repositories such as databases and file systems. The feature acquisition engine 310 can collect feature data at regular intervals (e.g., daily, weekly, monthly, etc.) or upon receiving an indication that the data has changed. In some implementations, the feature acquisition engine 310 can include an application programming interface (API) that can provide feature data to the feature acquisition engine 310. For example, the API can be a web services API.

[0040] The image identification engine 320 can accept an image of an electrical component and determine whether one or more other images depict the same electrical component. The image identification engine 320 can include an object recognition machine learning model, such as a convolutional neural network (CNN) or a Barlow Twin model, that is configured to identify objects in the image.

[0041] In some implementations, the image identification engine 320 can evaluate metadata associated with characteristics of the electrical components. For example, if the metadata includes the location of the assets and the image identification engine 320 determines that the locations of two assets are different, the image identification engine 320 can determine that the images represent different electrical components. Similarly, if the metadata includes the asset identifiers of the assets and the image identification engine 320 determines that the asset identifiers of the two assets are different, the image identification engine 320 can determine that the images represent different electrical components.

[0042] The evaluation engine 330 can accept feature data (described above) and evaluate one or more machine learning models to generate predictions regarding electrical component failure. Example predictions of failure prediction machine learning models can include, but are not limited to, the likelihood that a component will fail over a single time period, the likelihood that a component will fail over each of multiple time periods, the mean time to failure, a distribution of failure probabilities, and the most likely period in which a component will fail.

[0043] The evaluation engine 330 can include one or more machine learning models. In some implementations, the evaluation engine 330 includes a failure prediction neural network 334 configured to accept inputs and generate predictions, such as predictions of the types listed above. In some implementations, the evaluation engine 330 includes one failure prediction neural network 334 that generates one or more types of predictions. In some implementations, the evaluation engine 330 includes multiple failure prediction neural networks 334, each generating one or more types of predictions.

[0044] As described above, the input may include images of the asset at multiple time periods. Additionally, the input characteristics may further include, but are not limited to, a grid map, component characteristics, and operating environment characteristics. Operating environment characteristics may include, but are not limited to, the number and timing of blackouts, brownouts, lightning strikes, and blown fuses, as well as weather conditions (e.g., temperature and humidity). Component characteristics may include, but are not limited to, manufacturer, model, age, rating, thermal constants, winding type, and load metrics (e.g., maximum load, average load, time under maximum load, etc.). A grid map may include, for example, the components present in the grid, their interconnection patterns, and the distances between the elements.

[0045] In some implementations, the evaluation engine 330 includes a defect detection machine learning model 332 and one or more failure prediction machine learning models 334. To determine which defects, if any, exist on the component, the system can process inputs including one or more images of the component using the defect detection machine learning model 332. The defect detection machine learning model 332 can be a neural network, and in some implementations, the defect detection machine learning model 332 is a recurrent neural network (e.g., a long short-term memory (LSTM) model) or another type of sequential machine learning model. Recurrent models differ from feedforward models in that they can process sequences of data, such as images (or output from processing images) of the component over multiple time periods.

[0046] The system can provide inputs (including images) to the defect detection machine learning model 332, which can generate outputs including encodings of the images. The encodings can include an indication of the presence and type of defect. The system can process images of the component taken at different times using the defect detection machine learning model 332 and use one or more outputs as inputs to the failure prediction machine learning model 334. The system can then process inputs including the output of the defect detection machine learning model and other feature data (described above) using a machine learning model configured to generate a prediction describing the likelihood of failure.

[0047] In some implementations, the defect detection machine learning model is a component of the failure prediction machine learning model 334. For example, defect detection may be performed by one or more hidden layers in the failure prediction machine learning model 334, and outputs from those layers may be used by other layers of the failure prediction machine learning model.

[0048] The prediction serving engine 340 can provide one or more predictions generated by the rating engine 330. In some implementations, the prediction serving engine 340 can generate user interface presentation data 345 that, when rendered by a client device, causes the client device to display the predictions. In some implementations, the prediction serving engine 340 can transmit one or more predictions to a network-connected device, including a storage device and a database.

[0049] 3B is a diagram of an example system 350 for predicting electrical component failures. System 350 is similar to system 300 of FIG. 3A, but is capable of processing inputs including different types of sensor measurement data.

[0050] System 350 may include feature acquisition engine 310, image identification engine 320, audio feature acquisition engine 371, audio identification engine 371, rating engine 380, and prediction-providing engine 340. Engines 361, 371, and 380 may be provided as one or more computer-executable software modules, hardware modules, or a combination thereof. For example, one or more of engines 361, 371, and 380 may be implemented as blocks of software code having instructions that cause one or more processors of system 350 to perform the operations described herein. Additionally or alternatively, one or more of engines 361, 371, and 380 may be implemented in electronic circuitry, such as, for example, a programmable logic circuit, a field programmable logic array (FPGA), or an application-specific integrated circuit (ASIC).

[0051] The audio feature acquisition engine 361 is similar to the feature acquisition engine 310 and can acquire audio feature data related to component failures. The audio feature data can include, but is not limited to, images 306a, 306b of the electrical component and of elements related to the potential failure of the electrical component, such as structural support elements.

[0052] Audio indicators related to component failures may be present in audio recording 306a or 306b and may include defects in the component itself or in any supporting structure, abnormal operating sounds such as buzzing (e.g., rattles from loose connections), or combinations thereof. The audio recording may be encoded in any suitable format, including, but not limited to, a spectrogram or other audio format.

[0053] For example, audio recording 306b may include audio features that indicate that the operation of the component is louder or unusual compared to normal operation or compared to the audio features of audio recording 306a.

[0054] In some implementations, the audio feature acquisition engine 361 can acquire additional feature data as described with reference to the feature acquisition engine 310. The feature data can further include metadata describing the feature data, such as a timestamp of the feature data (e.g., the date and time the audio recording was captured), a timestamp of when the feature data was acquired, a location (e.g., the location of the audio recording capture device and / or of an object captured in the audio recording, as provided by GPS or other means), a provider of the feature data, an asset identifier (e.g., provided by the person capturing the audio recording of the asset), etc.

[0055] The audio feature acquisition engine 361 may acquire the feature data using various techniques described with reference to the feature acquisition engine 310 .

[0056] The audio identification engine 371 is similar to the image identification engine 320 and can accept an audio recording of an electrical component and determine whether one or more other audio recordings represent the same electrical component. The audio identification engine 371 can include a machine learning model configured to identify sounds made by the electrical component in the audio recording.

[0057] In some implementations, the audio identification engine 371 can evaluate metadata associated with characteristics of the electrical components. For example, if the metadata includes locations where audio recordings were captured, the audio identification engine 371 can determine that the locations of the audio recordings differ by more than a threshold distance, and the image identification engine 371 can determine that the audio recordings capture different electrical components. Similarly, if the metadata includes asset identifiers for assets, and the audio identification engine 371 determines that the asset identifiers for the two assets differ, the audio identification engine 371 can determine that the images represent different electrical components.

[0058] Evaluation engine 380 is similar to evaluation engine 330 but can include additional machine learning models. For example, evaluation engine 380 can include a failure prediction neural network configured to accept inputs and generate predictions. In some implementations, evaluation engine 380 can include separate failure prediction neural networks, such as failure prediction neural network 334 and failure prediction neural network 384, configured to generate predictions for different types of inputs.

[0059] As described above, input to the failure prediction neural network 334 can include images of the asset over multiple time periods. Input to the separate failure prediction neural network 384 can include audio recordings of the asset over multiple time periods. Additionally, input features can further include, but are not limited to, a grid map, component features, and operating environment features. Operating environment features can include, but are not limited to, the number and timing of blackouts, brownouts, lightning strikes, and blown fuses, as well as weather conditions (e.g., temperature and humidity). Component features can include, but are not limited to, manufacturer, model, age, rating, thermal constants, winding type, and load metrics (e.g., maximum load, average load, time under maximum load, etc.). A grid map can include, for example, the components present in the grid, their interconnection patterns, and the distances between elements.

[0060] In some implementations, evaluation engine 380 includes one or more defect detection machine learning models, such as defect detection machine learning model 332 and defect detection machine learning model 382, ​​and one or more failure prediction machine learning models, such as 334 and 384. To determine which defects, if any, exist on the component, the system can use defect detection machine learning model 332 to process inputs including one or more images of the component. To determine which defects, if any, exist on the component, the system can use defect detection machine learning model 382 to process inputs including one or more audio recordings of the component. Defect detection machine learning model 382 may be a neural network, and in some implementations, defect detection machine learning model 382 is a recurrent neural network (e.g., a long short-term memory (LSTM) model) or another type of sequential machine learning model.

[0061] The system can provide inputs (including images or audio recordings) to a corresponding defect detection machine learning model 332 or defect detection machine learning model 382. The defect detection machine learning model 332 can generate an output including an encoding of the image. The defect detection machine learning model 382 can generate an output including an encoding of the audio recording. The encoding can include an indication of the presence and type of defect. The system can process images of the component taken at different times using the defect detection machine learning model 332 and use one or more outputs as inputs to the failure prediction machine learning model 334. The system can process audio recordings of the component taken at different times using the defect detection machine learning model 382 and use one or more outputs as inputs to the failure prediction machine learning model 384. The system can then process the input including the output of the defect detection machine learning model 332 and other feature data (described above) using a machine learning model configured to generate a first prediction describing the likelihood of failure. The system can then process the input including the output of the defect detection machine learning model 382 and other feature data (described above) using a machine learning model configured to generate a second product describing the likelihood of failure. The system can determine a final prediction based on a weighted combination of the first prediction and the second prediction.

[0062] In some implementations, the defect detection machine learning model is a component of the failure prediction machine learning model 334 or the failure prediction machine learning model 384, as described above.

[0063] 4 is a flow diagram of an exemplary process for predicting electrical component failure. For convenience, process 300 will be described as being performed by a system for predicting electrical component failure, e.g., a system for predicting failure of electrical component 300 of FIG. 3, suitably programmed to perform the process. The operations of process 400 may also be performed as instructions stored on one or more computer-readable media, which may be non-transitory, and execution of the instructions by one or more data processing devices may cause the one or more data processing devices to perform the operations of process 400. One or more other components described herein may perform the operations of process 400.

[0064] The system obtains (410) a first sensor measurement of a component of the power distribution grid taken at a particular time. The sensor measurement may include, for example, an image or an audio recording. The sensor measurements, including the first sensor measurement, may be obtained from a variety of sources. For example, an owner of the component may capture images at periodic intervals. In another example, the images may be obtained from another party, for example, a vehicle including a camera, such as an autonomous car, a drone, a photo sharing website (if the photo owner approves such use), etc.

[0065] The system identifies (420) a second sensor measurement of the component taken at a later time. The system can process the first sensor measurement and each sensor measurement in the set of second sensor measurements using a machine learning model configured to determine whether an electrical component in the first sensor measurement is also present in the second sensor measurement. For example, if the sensor measurements are images, the system can use an object detection machine learning model configured to determine whether an electrical component depicted in the first image is also present in the second image. For each of one or more second images (drawn from the set), the system can use the machine learning model to determine a predicted likelihood that the component is present in the second image. If the system determines that the predicted likelihood meets a threshold, the system determines that the second image contains the component. In some implementations, the system can process the machine learning model using the first image and all second images in the set.

[0066] In some implementations, the system can use metadata from a first sensor measurement and each sensor measurement in a second set of sensor measurements to determine whether an electrical component in the first sensor measurement is also present in the second sensor measurement. For example, the system can use metadata from a first image and each image in a second set of images to determine whether an electrical component depicted in the first image is also present in the second image. For example, the system can compare location data (e.g., GPS readings) of the first image with location data of each image in the second set of images. If the image locations are the same or within a threshold distance, the system can determine that the component is depicted in both images. The threshold distance can be predefined or calculated based on the geographic distribution of similar assets within a geographic region. For example, a larger threshold distance may be used for more rural areas with fewer transformers per unit area, while a smaller threshold distance may be used for urban areas with more transformers per unit area.

[0067] The machine learning model can obtain a second set of sensor measurements using the technique of operation 410 or a similar technique. Additionally, in some implementations, once the sensor measurements obtained in operation 410 have been evaluated using process 400, the sensor measurements can be retained for future use in operation 420.

[0068] In some implementations, the system is provided with a first sensor reading and a second sensor reading of the component, such that the second sensor reading is identified when the sensor readings are provided. For example, a user can call an API provided by the system to provide the first and second sensor readings.

[0069] The system optionally acquires additional characteristic data related to the electrical component failure 430. The additional characteristic data may include grid maps, component characteristics, and operating environment characteristics, as described above.

[0070] The system can obtain the additional feature data using various means. The system can retrieve data from information sources using APIs provided by the data sources. The system can retrieve data from various databases using Structure Query Language (SQL) operations. The system can retrieve data from file systems using traditional file system operations. The system can provide an API, and users of the system (which can be computing devices) can call the API to provide data.

[0071] The system processes (440) an input including at least the first sensor reading and the second sensor reading using one or more machine learning models, the one or more machine learning models configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor reading compared to the first sensor reading, the time interval being a period of time after the second time.

[0072] To determine which, if any, defects exist on the component, the system can process inputs including sensor measurements using a defect detection machine learning model. The system can provide two or more sensor measurements of the component to the defect detection machine learning model, and the defect detection machine learning model can determine an output including an encoding of the sensor measurements. The encoding can include an indication of the presence and type of defect. The system can use the defect detection machine learning model to process sensor measurements of the component taken at different times and use the multiple outputs as inputs to a failure prediction machine learning model.

[0073] To determine the likelihood of failure, the system can process inputs including output of the defect prediction machine learning model for two or more images of the component with a failure prediction machine learning model configured to generate predictions related to failure of the component over a period of time. The inputs can further include additional feature data, as described above.

[0074] Failure prediction machine learning models can be evaluated in response to various triggers. For example, the models can be evaluated whenever new data arrives (e.g., images of components), at periodic intervals, and when a user requests an evaluation (e.g., during a maintenance planning exercise).

[0075] In some implementations, the system can process inputs including a first sensor measurement and other feature data (e.g., component characteristics and operating environment characteristics) without a second sensor measurement. In such implementations, the system can use one or more machine learning models configured to generate predictions representing the likelihood that a component will experience a certain type of failure within a certain time interval. Such machine learning models can be trained using backpropagation on examples, where each example includes a component's sensor measurement, other feature data, and a result. The other feature data can include component characteristics and operating environment characteristics. A result can represent a failure if the component fails within the time interval or a success if the component does not fail during the interval. Note that component characteristics can enable the machine learning model to learn which components fail under similar circumstances. For example, components of the same make and model are likely to fail under similar circumstances, and such failures will be present in the training data, allowing the machine learning model to learn failure patterns. Additionally, components of the same type (e.g., transformers) may follow similar failure patterns even if the patterns vary somewhat due to differences in make and model. Such an approach can provide early failure prediction before a second image is available. Note that a "new" asset may be an asset that has been installed on the distribution grid for some time, but is newly introduced into a system for predicting electrical component failures.

[0076] In some implementations, the system can process input including a first sensor measurement and other feature data (e.g., features of the component and features of the operating environment) using one or more machine learning models configured to generate a prediction representing a recommended time for capturing one or more subsequent sensor measurements of the component. The machine learning models can be trained with examples including the sensor measurements, the other feature data, and a label. The label can represent a recommended duration before the next sensor measurement of the component is taken.

[0077] To configure the model, the system can train the failure prediction machine learning model using training examples that include feature values ​​and results. The results can indicate whether a component has failed during a given period of time. For example, a value of "1" can indicate a failure, and a value of "0" can indicate no failure. The feature values ​​can include two or more images of the component, a grid map, features of the component, and features of the operating environment, as described above.

[0078] In some implementations, the first sensor measurement and the second sensor measurement may be images, and the system may further acquire a first acoustic recording of the component. For example, the acoustic recording may be taken at a location near the component such that the audio recording includes any sounds made by the component, such as operating sounds. The first acoustic recording may be taken at a specific time when the first image is taken. For example, the first acoustic recording may be taken within a predetermined time window, before or after the specific time when the first image is taken. For example, the first acoustic recording may be taken seconds, minutes, hours, or days before or after the specific time when the first image is taken. The system may further identify a second acoustic recording of the component taken at a time after the second image is taken. For example, the second acoustic recording may be taken within a predetermined time window, before or after the time after the second image is taken. For example, the second acoustic recording may be taken seconds, minutes, hours, or days before or after the time after the second image is taken. The system can process the first audio recording and each audio recording in the set of second audio recordings using a machine learning model configured to determine whether an electrical component in the first audio recording is also present in the second audio recording.

[0079] In these implementations, the system can process an input including at least a first image and a second image using one or more machine learning models configured to generate an image-based prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as represented in the second image compared to the first image, the time interval being a period of time after a second time. The system can process a second input including at least a first acoustic recording and a second acoustic recording using one or more machine learning models configured to generate an audio-recording-based second prediction representing a likelihood that the component will experience a type of failure during the time interval based on one or more changes in one or more characteristics of the component as represented in the second acoustic recording compared to the first acoustic recording.

[0080] In some implementations, the first sensor measurement and the second sensor measurement may be optical images, and the system may further acquire a first thermal image of the component. The first thermal image may be taken at a specific time when the first optical image was taken. For example, the first thermal image may be taken within a predetermined time window, before or after the specific time when the first optical image was taken. For example, the first thermal image may be taken seconds, minutes, hours, or days before or after the specific time when the first optical image was taken. The system may further identify a second thermal image of the component taken at a time after the second optical image was taken. For example, the second thermal image may be taken within a predetermined time window, before or after the time after the second optical image was taken. For example, the second thermal image may be taken seconds, minutes, hours, or days before or after the time after the second optical image was taken. The system can process the first thermal image and each thermal image in the set of second thermal images using a machine learning model configured to determine whether an electrical component in the first thermal image is also present in the second thermal image.

[0081] In these implementations, the system can process an input including at least a first optical image and a second optical image using one or more machine learning models configured to generate an optical-image-based prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as represented in the second optical image compared to the first optical image, the time interval being a period of time after a second time. The system can process a second input including at least a first thermal image and a second thermal image using one or more machine learning models configured to generate a thermal-image-based second prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as represented in the second thermal image compared to the first thermal image. The second input can also include characteristics of the operating environment, such as, for example, a temperature in an environment near the component.

[0082] In these implementations, the system can determine the data indicative of the prediction based on a weighted combination of the prediction and the second prediction. For example, the system can multiply the prediction and the second prediction by a predetermined weight and add the weighted prediction to the weighted second prediction to determine the final prediction.

[0083] The system provides data indicative of the prediction for presentation by a display 450. The system can provide the presentation data by transmitting the data over a network to a client device or by storing the presentation data in a data store (e.g., a file system or a database).

[0084] In implementations where the system acquires sensor measurements that are images and audio recordings, the system can provide data indicative of a final prediction based on a weighted combination of a prediction based on the images and a second prediction based on the audio recordings. In implementations where the system acquires optical images and thermal images, the system can provide data indicative of a final prediction based on a weighted combination of a prediction based on the optical images and a second prediction based on the thermal images.

[0085] FIG. 5 is an illustration of a component defect that may be detectable in a thermal image over a period of time. FIG. 5 depicts an insulator 500 for two time periods, 1990 and 1995, where areas of different temperatures, or hot spots 510, grow over time. For example, in 1990, the insulator 500 has no hot spots. In 1995, the insulator 510 has one small hot spot 510. The hot spot 510 may indicate tracking or degradation on the surface of the insulator 500, which may adversely affect the functionality of the insulator 500. The hot spot 510 may be detectable or present in a thermal image of the insulator 500.

[0086] Both the presence of defects (in this example, hot spots indicating tracking) and the rate of change of the defects can be used to predict component failure. In Figure 5, hot spots 510 grow over time, which can predict failure, for example, if the number or area of ​​hot spots exceeds a threshold, and the component is no longer able or unlikely to function properly.

[0087] 5 illustrates tracking as an example, a wide variety of defects are possible. Examples of defects detectable in a thermal image may include thermal qualities such as missing or damaged insulation, hot spots during operation, or the operating temperature of a component, among others.

[0088] A system that considers the thermal history of a component, or the thermal quality of a component at different times, can utilize predictive signals of failure or non-failure based on the thermal history. For example, a component exposed to or operating at a higher temperature in its environment may wear out faster than a component exposed to or operating at a lower temperature. A component exposed to a higher temperature for a longer period of time may wear out faster than a component exposed to a higher temperature for a shorter period of time. A component exposed to a higher rate of temperature change may wear out faster than a component exposed to a slower rate of temperature change.

[0089] 6 is a block diagram of an exemplary computer system 600 that can be used to perform the operations described above. System 600 includes a processor 610, a memory 620, a storage device 630, and an input / output device 640. Each of the components 610, 620, 630, and 640 can be interconnected using, for example, a system bus 650. Processor 610 can process instructions for execution within system 600. In one implementation, processor 610 is a single-threaded processor. In another implementation, processor 610 is a multi-threaded processor. Processor 610 can process instructions stored in memory 620 or on storage device 630.

[0090] The memory 620 stores information within the system 600. In one implementation, the memory 620 is a computer-readable medium. In one implementation, the memory 620 is a volatile memory unit. In another implementation, the memory 620 is a non-volatile memory unit.

[0091] The storage device 630 can provide mass storage for the system 600. In one implementation, the storage device 630 is a computer-readable medium. In various different implementations, the storage device 630 may include, for example, a hard disk device, an optical disk device, a storage device shared over a network by multiple computing devices (e.g., a cloud storage device), or some other mass storage device.

[0092] The input / output device(s) 640 provide input / output operations for the system 600. In one implementation, the input / output device(s) 640 may include one or more of a network interface device, e.g., an Ethernet card, a serial communication device, e.g., an RS-252 port, and / or a wireless interface device, e.g., an 802.11 card. In another implementation, the input / output device(s) may include a driver device configured to receive input data and send output data to other input / output devices, e.g., a keyboard, a printer, and a display device 660. However, other implementations, such as a mobile computing device, a mobile communication device, a set-top box, a television client device, etc., may also be used.

[0093] Although an exemplary processing system is depicted in FIG. 6, implementations of the subject matter and functional operations described herein can be realized in other types of digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed herein and their structural equivalents, or in one or more combinations of these.

[0094] Embodiments of the subject matter and functional operations described herein can be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed herein and their structural equivalents. Embodiments of the subject matter described herein can be implemented as one or more modules of computer program instructions encoded on a computer-readable medium for execution by or controlling the operation of a data processing apparatus. The computer-readable medium may be an article of manufacture such as a hard drive in a computer system, an optical disk sold through retail channels, or an embedded system. The computer-readable medium may be obtained separately and encoded with one or more modules of computer program instructions, such as by distribution of one or more modules of computer program instructions over a wired or wireless network. The computer-readable medium may be a machine-readable storage device, a machine-readable storage substrate, a memory device, or a combination of one or more of these.

[0095] The term "data processing apparatus" encompasses all apparatus, devices, and machines for processing data, including, by way of example, a programmable processor, a computer, or multiple processors or computers. In addition to hardware, an apparatus may include code that creates an execution environment for the computer program in question, such as code that constitutes processor firmware, a protocol stack, a database management system, an operating system, an execution environment, or one or more combinations thereof. Additionally, an apparatus may employ a variety of different computing model infrastructures, such as web services, distributed computing, and grid computing infrastructures.

[0096] A computer program (also known as a program, software, software application, script, or code) can be written in any suitable form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any suitable form, including as a stand-alone program or included as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file in a file system. A program can be stored within a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subprograms, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communications network.

[0097] The processes and logic flows described herein may be performed by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows may also be performed by, and apparatus may be implemented as, special purpose logic circuitry, such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0098] Processors suitable for executing computer programs include, by way of example, dedicated microprocessors. Generally, a processor will receive instructions and data from a read-only memory or a random-access memory, or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include one or more mass storage devices, such as magnetic, magneto-optical, or optical disks, for storing data, or be operatively coupled to receive data from, transfer data to, or both. However, a computer need not have such devices. Furthermore, a computer can be incorporated into another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device (e.g., a universal serial bus (USB) flash drive), to name just a few. Suitable devices for storing computer program instructions and data include, by way of example, semiconductor memory devices such as EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), and flash memory devices, magnetic disks, e.g., internal hard disks or removable disks, magneto-optical disks, and all forms of non-volatile memory, including CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0099] To provide for interaction with a user, embodiments of the subject matter described herein can be implemented on a computing device capable of providing information to a user. The information can be provided to the user in any form of sensory format, including visual, auditory, tactile, or a combination thereof. The computing device can be coupled to a display device, such as an LCD (liquid crystal display) display device, an OLED (organic light emitting diode) display device, another monitor, a head-mounted display device, etc., to display information to the user. The computing device can be coupled to an input device. Input devices can include a touchscreen, a keyboard, and a pointing device, such as a mouse or trackball, through which a user can provide input to the computing device. Other types of devices can be used to provide interaction with a user as well; for example, feedback provided to the user can be any suitable form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback, and input from the user can be received in any suitable form, including acoustic, speech, or tactile input.

[0100] A computing system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communications network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. Embodiments of the subject matter described herein may be implemented in a computing system that includes back-end components, such as data servers, or middleware components, e.g., application servers, or front-end components, such as client computers having graphical user interfaces or web browsers through which users can interact with implementations of the subject matter described herein, or any combination of one or more such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication, e.g., a communications network. Examples of communications networks include local area networks (LANs) and wide area networks (WANs), internetworks (e.g., the Internet), and peer-to-peer networks (e.g., ad-hoc peer-to-peer networks).

[0101] In addition to the embodiments described above, the following embodiments are also innovative.

[0102] Embodiment 1 is a method for predicting a power grid asset failure, comprising: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; processing an input including a first sensor measurement and a second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period after a second time; and providing data indicative of the prediction for presentation by a display.

[0103] Embodiment 2 is the distribution grid asset failure prediction method of embodiment 1, wherein the machine learning models include a fault detection machine learning model and a failure prediction machine learning model.

[0104] Embodiment 3 is the distribution grid asset failure prediction method according to embodiment 1 or 2, wherein the machine learning model comprises a failure prediction machine learning model.

[0105] Embodiment 4 is the distribution grid asset failure prediction method of embodiment 3, wherein the failure prediction machine learning model includes a fault detection hidden layer.

[0106] Embodiment 5 is the distribution grid asset failure prediction method of any one of embodiments 1 to 4, wherein the prediction includes one or more of the likelihood that the component will fail over a single time period, the likelihood that the component will fail over each of a plurality of time periods, the mean time to failure, a distribution of failure probabilities, or the time period in which the component is most likely to fail.

[0107] Embodiment 6 is the distribution grid asset failure prediction method according to any one of embodiments 1 to 5, wherein the component characteristics include one or more of bulging, tilting, loose fasteners, missing fasteners, cracks, burn marks, rust, oil leaks, missing insulation, damaged insulation, operating noise, or thermal quality.

[0108] A seventh embodiment is the power distribution grid asset failure prediction method according to any one of the first to sixth embodiments, wherein the machine learning model is a recurrent neural network.

[0109] Embodiment 8 is the distribution grid asset fault prediction method according to embodiment 7, wherein the recurrent neural network is a long short-term memory machine learning model or a cross-attention based transformer model.

[0110] A ninth embodiment is the method for predicting a power grid asset failure according to any one of the first to eighth embodiments, wherein the input further includes characteristics of the components and characteristics of the operating environment of the components.

[0111] Embodiment 10 is the distribution grid asset failure prediction method of embodiment 9, wherein the characteristics of the operating environment include a set of temperature values ​​measured at or around the location of the component.

[0112] An eleventh embodiment is the method for predicting failure of a power grid asset according to any one of the first to tenth embodiments, wherein the sensor measurements are acoustic recordings of the components.

[0113] A twelfth embodiment is the method for predicting failure of a power distribution grid asset according to any one of the first to eleventh embodiments, wherein the sensor measurements are images of the components.

[0114] Embodiment 13 is a method for detecting a component in a sensor, the method comprising: obtaining a first acoustic recording of a component of the electrical grid taken at a first time; identifying a second acoustic recording of the component taken at a second time; and processing a second input including the first acoustic record and the second acoustic record using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of failure during the time interval based on one or more changes in one or more characteristics of the component as represented in the second acoustic record compared to the first acoustic record; 13. The method of any one of embodiments 1 to 12, further comprising: determining the data based on a weighted combination of the prediction and the second prediction.

[0115] Embodiment 14 is a method for detecting a change in a sensor measurement value, the method comprising: obtaining a first thermal image of a component of the electrical distribution grid taken at a first time; identifying a second thermal image of the component taken at a second time; processing a second input including the first thermal image and the second thermal image using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as depicted in the second thermal image compared to the first thermal image; 14. The method of any one of embodiments 1 to 13, further comprising: determining the data based on a weighted combination of the prediction and the second prediction.

[0116] Embodiment 15 is 15. The method of any one of embodiments 1 to 14, further comprising: processing an input including the first sensor measurement and characteristics of the operating environment using a machine learning model configured to generate a prediction representing a recommended time for capturing one or more subsequent sensor measurements of the component.

[0117] Embodiment 16 is directed to one or more computers and instructions, the instructions, when executed by the one or more computers, causing the one or more computers to: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; processing an input including a first sensor measurement and a second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period after a second time; and one or more storage devices that store instructions that cause the system to perform operations including: providing data indicative of the prediction for presentation by a display.

[0118] Embodiment 17 is the system described in embodiment 16, wherein the machine learning models include a defect detection machine learning model and a failure prediction machine learning model.

[0119] Embodiment 18 is the system described in embodiment 16 or 17, wherein the machine learning model includes a failure prediction machine learning model.

[0120] Embodiment 19 is the system described in embodiment 18, wherein the failure prediction machine learning model includes a defect detection hidden layer.

[0121] When executed by one or more computers, embodiment 20 causes the one or more computers to: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; processing an input including a first sensor measurement and a second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period after a second time; and providing data indicative of the prediction for presentation by a display.

[0122] While this specification contains many implementation details, these should not be construed as limiting the scope of what is or may be claimed, but rather as describing features unique to particular embodiments of the disclosed subject matter. Certain features described herein in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Furthermore, while features may be described above as acting in a particular combination and may even initially be claimed as such, one or more features from a claimed combination may, in some cases, be deleted from the combination, and the claimed combination may be directed to a subcombination or a variation of the subcombination. Thus, unless expressly specified otherwise or unless the knowledge of one of ordinary skill in the art clearly dictates otherwise, any feature of the above-described embodiments can be combined with any other feature of the above-described embodiments.

[0123] Similarly, although operations are depicted in the figures in a particular order, this should not be understood as requiring such operations to be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desired results. In certain situations, multitasking and / or parallel processing may be advantageous. Furthermore, the separation of various system components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems may generally be integrated together in a single software product or packaged in multiple software products.

[0124] Thus, while specific embodiments of the present invention have been described, other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results.

Claims

1. 1. A method for predicting electrical grid asset failures, comprising: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; and processing an input including the first sensor measurement and the second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as represented in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period of time after the second time; providing data indicative of said prediction for presentation by a display; A method for predicting power grid asset failures, comprising:

2. The method of claim 1 , wherein the machine learning models include a fault detection machine learning model and a failure prediction machine learning model.

3. The method of claim 1 , wherein the machine learning model comprises a failure prediction machine learning model.

4. The method of claim 3 , wherein the failure prediction machine learning model includes a fault detection hidden layer.

5. 2. The distribution grid asset failure prediction method of claim 1, wherein the prediction includes one or more of the likelihood that the component will fail over a single time period, the likelihood that the component will fail over each of a plurality of time periods, a mean time to failure, a distribution of failure probabilities, or a time period when the component is most likely to fail.

6. 2. The method of claim 1, wherein the component characteristics include one or more of bulging, tilting, loose fasteners, missing fasteners, cracks, burn marks, rust, oil leaks, missing insulation, damaged insulation, operating noise, or thermal quality.

7. The method of claim 1 , wherein the machine learning model is a recurrent neural network.

8. 8. The method of claim 7, wherein the recurrent neural network is a long short-term memory machine learning model or a cross-attention based transformer model.

9. The method of claim 1 , wherein the input further comprises characteristics of the component and characteristics of the component's operating environment.

10. The method of claim 9 , wherein the characteristics of the operating environment include a set of temperature values ​​measured at or around the location of the component.

11. The method of claim 1 , wherein the sensor measurements are acoustic recordings of the component.

12. The method of claim 1 , wherein the sensor measurements are images of the component.

13. the sensor measurements are images of the component, and the method comprises: obtaining a first acoustic recording of the component of the electrical grid taken at the first time; identifying a second acoustic recording of the component taken at the second time; and processing a second input including the first acoustic record and the second acoustic record using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of failure during the time interval based on one or more changes in one or more characteristics of the component as represented in the second acoustic record compared to the first acoustic record; 2. The method of claim 1, further comprising: determining the data based on a weighted combination of the prediction and the second prediction.

14. the sensor measurements are optical images of the component, and the method comprises: obtaining a first thermal image of the component of the electrical grid taken at the first time; identifying a second thermal image of the component taken at the second time; and processing a second input including the first thermal image and the second thermal image using a second machine learning model configured to generate a second prediction representing a likelihood that the component will experience a type of failure during the time interval based on one or more changes in one or more characteristics of the component as depicted in the second thermal image compared to the first thermal image; 2. The method of claim 1, further comprising: determining the data based on a weighted combination of the prediction and the second prediction.

15. 2. The method of claim 1, further comprising: processing input including the first sensor measurement and characteristics of the operating environment using a machine learning model configured to generate a prediction representing a recommended time for capturing one or more subsequent sensor measurements of the component.

16. one or more computers; and instructions, which when executed by the one or more computers, cause the one or more computers to: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; and processing an input including the first sensor measurement and the second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period of time after the second time; and one or more storage devices that store instructions that cause the system to perform operations including: providing data indicative of the prediction for presentation by a display.

17. The system of claim 16 , wherein the machine learning models include a defect detection machine learning model and a failure prediction machine learning model.

18. The system of claim 16 , wherein the machine learning model comprises a failure prediction machine learning model.

19. The system of claim 18 , wherein the failure prediction machine learning model includes a defect detection hidden layer.

20. When executed by one or more computers, the one or more computers: obtaining a first sensor measurement of a component of the electrical distribution grid taken at a first time; identifying a second sensor measurement of the component taken at a second time, the second time being after the first time; and processing an input including the first sensor measurement and the second sensor measurement using a machine learning model configured to generate a prediction representing a likelihood that the component will experience a type of failure during a time interval based on one or more changes in one or more characteristics of the component as reflected in the second sensor measurement compared to the first sensor measurement, wherein the time interval is a period of time after the second time; and providing data indicative of the prediction for presentation by a display.

Images