Thermal predictive modeling of physical assets

Abstract

A method is provided that includes receiving a time constant and a training regression model determined during a training phase, receiving a real-time measured current used by an asset; receiving real-time measured temperatures measured at the basic monitoring points; predicting the temperature of a predicted point in real time by applying a trained regression model using the real-time measured current, a previously predicted temperature of the predicted point, a time lapse since the previously predicted temperature was predicted, and a time constant, where the predicted point is selectable to include both the same and different predicted points as the base monitoring point; comparing the predicted temperature of the predicted point subset with the current received temperature of the basic monitoring point corresponding to the predicted point subset; correcting the predicted temperature of the selected predicted point using the comparison result; and outputting the predicted temperature in real time.

Description

[0001] Cross - Reference to Related Applications

[0002] This application claims the priority and benefit of U.S. Non - Provisional Patent Application Serial No. 18 / 090,650, filed on December 29, 2022, which is incorporated herein by reference in its entirety. Technical Field

[0003] This disclosure relates to the health monitoring of physical assets, and more particularly, to applying machine learning to thermal prediction modeling of industrial physical assets. Background Art

[0004] Thermal monitoring systems can identify thermal changes in equipment located in environments such as industrial environments, vehicles, appliances, data processing centers, or other mechanical environments. Thermal changes can be indicators of potential safety, functionality, or process problems and / or the need for equipment maintenance intervention. Effective heat monitoring can be used to detect problems before they develop into safety risks or affect the functionality or productivity of the monitored equipment.

[0005] Thermal monitoring systems use sensors placed on the thermally conductive surfaces of physical assets. The sensors are placed at discrete locations that are accessible on the asset where the sensors are to be placed or within the field of view of the sensors. The sensors cannot monitor asset locations that are physically inaccessible. Thus, real - time thermal monitoring using sensors is limited to monitoring discrete, physically accessible locations. Potential critical information related to locations outside the discrete locations accessible to the sensors, including inaccessible locations, is not available to or used by the thermal monitoring system. Therefore, if a safety alarm system, maintenance system, and / or optimization system uses data from the thermal monitoring system to make real - time decisions, those decisions can be made based on incomplete data. While traditional methods and systems are generally considered satisfactory for their intended purposes, there is still a need in the art for a method of real - time thermal monitoring of locations other than the discrete locations accessible to sensors, including continuous real - time thermal monitoring of accessible and / or inaccessible thermally conductive surfaces. Summary of the Invention

[0006] The objectives and advantages of the illustrated embodiments described below will be set forth in and become apparent from the following description. The additional advantages of the illustrated embodiments will be realized and attained by means of the devices, systems, and methods particularly pointed out in the written description, its claims, and the drawings. ‎

[0007] To achieve these and other advantages, and in accordance with the purpose of the illustrated embodiments, in one aspect, a method for predicting temperature in an asset is disclosed. The method includes receiving a time constant and at least one trained regression model determined during a training phase that applies machine learning to simulated multi-dimensional simulation points of a simulated asset and temperatures associated with the corresponding simulation points; receiving real-time measured current used by the asset; and receiving real-time measured temperature measured at a base monitoring point. The method further includes predicting the temperature of a prediction point in real time by applying at least one trained regression model and using the real-time measured current, a previously predicted temperature of the prediction point, the time elapsed since the previous predicted temperature was predicted, and the time constant, where the prediction point is selectable to include a prediction point that is the same as the base monitoring point and a prediction point that is different from the base monitoring point. The method further includes comparing the predicted temperature of a subset of prediction points with the currently received temperature of the base monitoring point corresponding to the subset of prediction points; using the comparison result to correct the predicted temperature of the selected prediction point; and outputting the predicted temperature in real time.

[0008] In one or more embodiments, the method may further include using the predicted temperature to update in real time an augmented reality visualization of the asset.

[0009] In one or more embodiments, the method may further include determining whether a difference between the predicted temperature of a subset of prediction points and the received temperature at the corresponding base monitoring point exceeds a threshold; and causing an action affecting the asset to be performed in response to a determination that the difference exceeds the threshold.

[0010] In one or more embodiments, the time constant of the method may be associated with a corresponding clustering of the simulation points, and at least one regression model may be determined from the clustering of the simulation data.

[0011] In one or more embodiments, at least one regression model may include a steady-state regression model using polynomial regression and a transient regression model using exponential regression. The method may include predicting the temperature at a prediction point by predicting the steady-state temperature at the prediction point by applying the steady-state regression model. The method may further include predicting the transient temperature at the prediction point by applying the transient regression model using the predicted steady-state temperature at the prediction point, the real-time measured current, the previously predicted transient temperature of the prediction point, the time elapsed, and the time constant. The predicted temperature of the real-time prediction point may include predicting the transient temperature.

[0012] In one or more embodiments, the simulation may be a digital twin.

[0013] In one or more embodiments, the simulation may include two or more steady-state simulations using different simulation parameters, and the method may further include repeating the following during a training phase until a steady-state prediction is determined to be acceptable: for the two or more steady-state simulations, extracting steady-state simulation points in the simulation points and the temperature associated with each steady-state simulation point; for each of the two or more steady-state simulations, applying a clustering algorithm to the extracted steady-state simulation points and their corresponding associated temperatures to form a plurality of steady-state clusters; applying a steady-state regression model to each steady-state cluster to represent the relationship between the temperature associated with the corresponding steady-state simulation point and the simulation parameters; generating a steady-state prediction by applying the steady-state regression model to the selected simulation parameters for predicting the steady-state temperature of the steady-state simulation points at the selected simulation parameters; determining a steady-state difference between the predicted steady-state temperature of the steady-state simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein when the steady-state difference is below a steady-state threshold, the steady-state prediction is determined to be acceptable; and adjusting the selected simulation parameters for reuse in the next repetition, if any, to attempt to reduce the steady-state difference.

[0014] In one or more embodiments, the simulation may include a transient simulation, and the method may include, during a training phase, for a plurality of spaced time steps, extracting transient simulation points in the simulation points and the temperature associated with each transient simulation point; and applying a clustering algorithm to the extracted transient simulation points and their temperatures over the plurality of spaced time steps to form a plurality of transient clusters. The method may further include repeating the following until a transient prediction is determined to be acceptable: applying a transient regression model to each transient cluster using the nearest time constant associated with each transient cluster; once the steady-state prediction is determined to be acceptable, generating a transient prediction for predicting the most recent temperature associated with the corresponding transient cluster by applying the transient regression model using the steady-state prediction, the previously predicted transient temperature of the corresponding transient cluster obtained at an earlier simulation time, the amount of simulation time elapsed since the earlier simulation time, and the nearest time constant of the corresponding transient cluster; determining a transient difference between the predicted transient temperature of the transient simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein when the transient difference is below a transient threshold, the transient prediction is determined to be acceptable; and adjusting the time constant for reuse in the next repetition, if any, to attempt to reduce the transient difference.

[0015] In one or more embodiments, the method may further include using at least one of the transient prediction and the steady-state prediction to real-time update an augmented reality visualization of the asset.

[0016] In one or more embodiments, the method may further include obtaining the measured temperatures at a plurality of corresponding monitoring points; and continuously updating the measured temperatures with the measurements obtained at a subset of the monitoring points for use in determining the transient difference.

[0017] According to another aspect of the present disclosure, a method for training at least one model to predict temperature in an asset is provided. The method includes repeating the following until a steady-state prediction is determined to be acceptable: for two or more steady-state simulations using different respective simulation parameters, extracting steady-state simulation points and the temperature associated with each steady-state simulation point; for each of the two or more steady-state simulations, applying a clustering algorithm to the extracted steady-state simulation points and their respective associated temperatures to form a plurality of steady-state clusters; applying a steady-state regression model to each steady-state cluster to represent the relationship between the temperature associated with the respective steady-state simulation point and the simulation parameters; generating a steady-state prediction by applying the steady-state regression model to the selected simulation parameters for predicting the steady-state temperature of the steady-state simulation points at the selected simulation parameters; determining a steady-state difference between the predicted steady-state temperature of the steady-state simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein the steady-state prediction is determined to be acceptable when the steady-state difference is below a steady-state threshold; and adjusting the selected simulation parameters for reuse in the next repetition, if any, to attempt to reduce the steady-state difference.

[0018] In one or more embodiments, the method may further include simulations including transient simulations, and the method may further include, during a training phase: for a plurality of spaced time steps, extracting transient simulation points in the simulation points and the temperature associated with each transient simulation point; and applying a clustering algorithm to the extracted transient simulation points and their temperatures over the plurality of spaced time steps to form a plurality of transient clusters. The method may further include repeating the following until a transient prediction is determined to be acceptable: applying a transient regression model to each transient cluster using the nearest time constant associated with each transient cluster; once the steady-state prediction is determined to be acceptable, generating a transient prediction for predicting the most recent temperature associated with the corresponding transient cluster by applying the transient regression model using the steady-state prediction, the previously predicted transient temperature of the corresponding transient cluster obtained at an earlier simulation time, the amount of simulation time elapsed since the earlier simulation time, and the nearest time constant of the corresponding transient cluster; determining a transient difference between the predicted transient temperature of the transient simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein the transient prediction is determined to be acceptable when the transient difference is below a transient threshold; and adjusting the time constant for reuse in the next repetition, if any, to attempt to reduce the transient difference.

[0019] In one or more embodiments, the method may further include using at least one of the transient prediction and the steady-state prediction to update an augmented reality visualization of the asset in real time.

[0020] In other aspects of the present disclosure, one or more computer systems for performing the disclosed methods are provided. In still other aspects of the present disclosure, one or more non-transitory computer-readable storage media and one or more computer programs embedded therein are provided, which, when executed by a computer system, cause the computer system to perform the corresponding disclosed methods.

[0021] These and other features of the systems and methods of the present disclosure will become more apparent to those skilled in the art from the following detailed description of the preferred embodiments in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] A more detailed description of the present disclosure, briefly outlined above, can be obtained by reference to various embodiments, some of which are illustrated in the accompanying drawings. Although the drawings illustrate selected embodiments of the present disclosure, these drawings should not be considered as limiting its scope, as the present disclosure may permit other equally effective embodiments.

[0023] Figure 1 is a schematic system diagram of a thermal prediction system that receives measurement data from sensors and sends prediction data to a maintenance system and / or an augmented reality system according to an embodiment of the present disclosure;

[0024] Figure 2 shows a block diagram of an example thermal prediction system according to an aspect of the present disclosure; ‎

[0025] Figure 3A shows a flowchart of an example method performed to train a thermal predictor of a thermal prediction system according to an embodiment of the present disclosure;

[0026] Figure 3B shows, during asset operation according to an embodiment of the present disclosure, by Figure 3A a flowchart of an example method performed by the thermal predictor;

[0027] Figure 3C shows, according to an embodiment of the present disclosure, the execution of Figure 3B a flowchart of an example method of an operation for predicting temperature data shown; and

[0028] Figure 4 shows, according to an embodiment of the present disclosure, an example computing system that can be used to implement a thermal predictor as shown in Figure 1 and 2 shown. ‎

[0029] Where possible, the same reference numerals are used to denote the same elements common to the drawings. However, elements disclosed in one embodiment may be beneficially used in other embodiments without specific recitation. DETAILED DESCRIPTION

[0030] Reference will now be made to the accompanying drawings, where like reference numerals represent similar structural features or aspects of the present subject matter disclosure. For purposes of explanation and illustration and not limitation, a block diagram of an exemplary embodiment of a thermal monitoring system 100 in accordance with the present disclosure is shown in Figure 1 and is generally designated by reference numeral 100. Methods related to using thermal measurements and machine learning to predict the thermal condition of a physical asset, including for inaccessible regions of the physical asset; three-dimensional visualization of measuring and predicting thermal conditions; and Figures 2-4 provides maintenance, safety alerts, and optimization processes using predicted and measured thermal conditions for a thermal monitoring system 100 or aspects thereof in accordance with the present disclosure, as will be described.‎

[0031] Now referring to Figure 1 , the thermal monitoring system 100 includes a thermal predictor 102 that receives measurement data from a plurality of sensors 122 disposed in accessible regions of a thermally conductive surface of a physical asset 10, such as a physical asset of an industrial system, vehicle, instrument, data processing center, or other mechanical environment. The physical asset 10 can be, for example, an industrial device, motor, conduit, processing or data storage device, etc.

[0032] The sensors 122 can be, for example, infrared imagers or thermometers (or thermocouples) / temperature sensors. The sensors 122 can be placed on the thermally conductive surface of the physical asset at discrete locations or positioned such that discrete portions of the thermally conductive surface are within the field of view of the sensors. The measurement data output by the sensors 122 indicates the thermal condition of discrete, accessible locations of the physical asset.

[0033] Some regions of the physical asset 10 are inaccessible for detecting thermal conditions such that sensors cannot be mounted or placed on the thermally conductive surface of that region or the thermally conductive surface of that region cannot be seen. Examples of inaccessible regions include, for example, internal portions of the physical asset 10, rear regions of the physical asset 10 (e.g., when the asset 10 is fixed in place and its rear region is mounted against an obstacle, such as a wall or insulator or near a movable contact).

[0034] The thermal predictor 102 outputs thermal prediction data representative of the thermal conditions of the accessible and inaccessible regions. The thermal prediction data includes a prediction of a continuous thermal condition that includes a high concentration of data points that are several orders of magnitude greater than the discrete data points from which the sensors 122 obtain actual measurement data.

[0035] The maintenance system 124 receives thermal prediction data, where the thermal prediction data can be used to cause actions to be performed to maintain the physical asset 10, protect the safety of the asset 10 and / or its environment, and / or optimize the operation of the asset 10. Decisions regarding maintenance, safety, and optimization related to the physical asset 10 can include updating maintenance schedules, performing actions to cause the execution of repairs or replacements, outputting control signals to generate alerts, and / or controlling components related to the physical asset when a safety issue occurs and / or controlling components related to the physical asset to optimize the operation of the asset.

[0036] The augmented reality system 126 receives thermal prediction data, where the thermal prediction data can be used to provide augmented reality visualizations in two or three dimensions of predicted thermal conditions in accessible and inaccessible areas, thereby providing a digital twin. In this way, a user can view predicted thermal conditions in inaccessible or invisible areas of the physical asset.

[0037] Reference Figure 2 , the thermal predictor 102 includes a simulation engine 110 that generates a virtual model 112 based on a simulation of the physical asset, a machine learning (ML) engine 114, and a comparison engine 116, and may also include a user interface (UI) 118. The comparison engine 116 compares the results of measurement data from the sensors 122 with the predictions output by the ML engine 114, and outputs thermal prediction data and / or alerts to the maintenance system 124 and / or the augmented reality system 126 (such as a digital twin). The simulation engine 110 receives simulation parameters from a set of simulation parameters 130 and a geometric model of the physical asset. The term "receive" with respect to data is intended to be interpreted broadly, such as pulling data, receiving data passively, accessing data from a remote memory, storing data locally, and accessing, reading data, receiving data through a transmission, or otherwise obtaining data.

[0038] The machine learning engine 114 applies two phases: a clustering phase and a regression phase. For example, in the clustering phase, the k-means clustering algorithm is applied to sample data to cluster the sample data into n groups with substantially equal variances, minimizing a value called inertia or within-cluster sum of squares. The k-means algorithm divides the sample data into disjoint clusters, and each cluster is described by the mean of the data samples in that cluster. The mean of each cluster is called the cluster centroid; the cluster centroid is calculated and may not actually exist in the data samples. In this case, the sample data is formed by multi-dimensional points of the asset steady-state simulation (meaning points in two dimensions (2D) or three dimensions (3D)).

[0039] During the regression phase, a regression model can be derived, such as a linear regression model, like ordinary least squares (OLS), which minimizes the sum of the squares of the differences between the output of a function of the observed dependent and independent variables in the input dataset. Geometrically, this is regarded as the sum of the squares of the distances parallel to the dependent variable axis between each data point in the input dataset and the corresponding point on the regression surface. The smaller the difference, the better the linear regression model fits the data. Not limited to a specific type of regression, this example type of linear regression model includes steady-state polynomial regression and transient exponential regression. In this case, the input dataset is the clusters formed during the clustering phase, and the independent variable can be current or another system parameter, such as ambient temperature, electrical / thermal contact resistance, material, conductor, architecture, natural / forced convection, etc.

[0040] The simulation parameters capture the system parameters of asset 10. The simulation parameters can include, for example, current, ambient temperature, electrical / thermal contact resistance, material, conductor, architecture, natural / forced convection, etc. The geometric model can be a two-dimensional (2D) or three-dimensional (3D) computer-aided design (CAD) model or scan.

[0041] The simulation engine 110 operates on the geometric model to generate a virtual model 112 using cooling simulation software for electronic components. The cooling simulation software can use, for example, a computational fluid dynamics (CFD) solver for electronic thermal management. The cooling simulation software can predict, for example, the airflow, temperature (also known as the thermal condition), and heat transfer in integrated circuit (IC) packages, power circuit boards (PCBs), electronic components / enclosures, and power electronic devices. Examples of cooling simulation software for electronic components include ANSYS Icepak™, Flowtherm™, Comsol™, etc.

[0042] The virtual model can be, for example, a digital twin, a CFD model, a finite element model, or a CAD model.

[0043] The virtual model 112 simulates a simulated thermal condition continuum associated with a physical asset, including regions accessible by the sensors 122 and regions inaccessible by the sensors 122. As the virtual model 112 is trained, the modeling of the inaccessible regions evolves, where the training uses a comparison of predicted data and real-time measurement data. The thermal condition continuum is modeled in real time, where the continuum can include a large number of temperature points (e.g., more than one million) distributed within the domain of the geometric model at any location within the domain, while only one or a few actual thermal measurements are actually sensed by the sensors 122. Real time refers to the response time for outputting results (e.g., during training, updating the model, generating predictions during operation, etc.). The response time is a function of the model size and the available computing power. In one or more embodiments, the response time is on the order of minutes or seconds. In one or more embodiments, the response time is less than one minute. In one or more embodiments, the response time is 3 - 30 seconds. In one or more embodiments, the response time is 3 - 10 seconds.

[0044] The simulation engine 110 outputs simulation data, and the virtual model 112 uses this simulation data to generate steady-state simulation data using simulations of at least two different currents when the simulated physical asset is in equilibrium, and to generate transient simulation data using simulations of the physical asset as it changes over time. The steady-state simulation data provides boundary conditions that can be used to make predictions.

[0045] A digital twin is an executable virtual model designed to accurately represent a physical asset. A digital twin is a simulation in which a physical replica of the physical asset is fully simulated. A digital twin provides more than just a virtual model using computer-aided design and / or engineering (CAD-CAE). A digital twin can be used to quickly transfer data sets between the physical asset and the digital twin, which enables users to view the operation of the physical asset in real time. In an embodiment, augmented reality can be used with the digital twin. A digital twin uses evolving data, thus providing an accurate description of the physical asset as it changes over time, rather than providing a snapshot of the behavior of an object at a particular moment. A digital twin can use a time scale on which the physical asset and its behavior are prone to significant changes.

[0046] As an overview of the method used, the simulation data output by the virtual model 112 is processed together with the processed steady-state simulation data and transient simulation data to cluster the data respectively, and to perform regression on the steady-state simulation data and transient simulation data of each cluster. Using the regression determined for the steady state, the steady-state temperature at any current is predicted in real time. The transient temperature is predicted in real time. The prediction of the transient temperature uses the regression determined for the transient at different currents and the prediction of the steady state. The regression uses a time constant to define the rate of change of temperature over time. The time constant is updated as a function of the comparison for predicting the transient.

[0047] When the measurement data output from sensor 122 changes, the sensor output is compared with the prediction of the simulated data output based on virtual model 112. The comparison result between the steady-state measurement data and the steady-state prediction is used to adjust the simulation parameters of virtual model 112 to better fit the measurement data output by sensor 122. The comparison result between the transient measurement data and the transient prediction is used to adjust the time constant, which affects the exponential regression used for transient temperature prediction.

[0048] Examples of the processes performed by ML engine 114 and comparison engine 116 are now described in more detail according to one or more embodiments of the present disclosure. ML engine 114 processes the simulated data output by virtual model 112, processing the steady-state simulated data and the transient simulated data separately. The steady-state simulated data is processed by extracting all the points of each simulated current. Each point includes a location corresponding to the simulated physical asset 10 and the simulated temperature. The extracted points are clustered into classes with points having similar simulated temperatures. Clustering algorithms such as the K-means algorithm (e.g., implemented by sklearn.cluster™), similarity algorithms, spectral algorithms, agglomerative algorithms, mean shift algorithms, etc. are used to perform the clustering. The specific clustering algorithm or implementation is not limited.

[0049] The simulated temperature can be adjusted for different simulation parameters, such as current, etc. For example, the temperature at the nominal current and a fraction of the nominal current can be obtained (or by using different simulation parameters). For example, polynomial regression can be used to adjust the simulated temperature for different currents. Some example regression algorithms include multivariate linear regression, ordinary least squares (OLS), gradient descent, etc., and the specific regression algorithm or implementation is not limited. Based on these adjustments (e.g., using polynomial regression), steady-state predictions of the temperature at one or more selected currents can be made. The steady-state predictions for various currents can be used by ML engine 114 to process transient predictions, and / or can be further provided to comparison engine 116. The transient simulated data is processed by extracting all the points at different time steps, where the time steps are separated by equal or unequal time intervals. Each point includes a location corresponding to the simulated physical asset 10 and the simulated temperature. The extracted points are clustered into classes with points having similar simulated temperatures using a clustering algorithm.

[0050] Exponential regression uses an adjustable time constant for the cluster that defines the evolution of the simulated temperature, where the exponential regression is a physical model that expresses the transient behavior of the temperature. Based on the results of the exponential regression, the real-time temperature can be predicted and output as prediction data (using the adjusted time constant). A steady-state prediction for the selected current can be obtained. For example, the selected current can include the rated current (meaning the maximum current for which the asset is designed to meet the specifications) and a lower current that is a selected percentage lower than the rated current. Transient predictions can be obtained, including the final temperature predicted for the selected current and the time elapsed since the last transient prediction. Transient prediction data for various currents can be provided to the comparison engine 116. The results of the comparison can be used to adjust the time constant.

[0051] The comparison engine 116 receives actual measurement data from the sensors 122. The measurements can be related to prompted tests, such as periodically or in response to an event. The actual measurement data can be used as real data for updating the simulation parameters used by the simulation engine 112, and thus updating the virtual model 112. In particular, the comparison engine 116 can receive real-time actual measurement data (also referred to as real-time measurement data) obtained by operating the asset 10 at a specific current and predicted steady-state data for the specific current from multiple sensors 122. The comparison engine 116 can compare the measurement data from the performed test with the predicted steady-state data. The difference represents the prediction error. Alternatively, during operation, the difference can indicate a change in condition, which may be very different from the simulation parameters.

[0052] In one or more embodiments, as shown and described with reference to Figure 2 When the result of the comparison indicates that there is a difference between the measurement data and the predicted steady-state data that requires correction (e.g., the difference (meaning the prediction error) exceeds a predetermined threshold (e.g., 5%, not limited to a specific threshold)), the simulation parameters used to simulate the steady state of the asset 10 can be adjusted, thereby generating updated steady-state simulation data and updated boundary conditions. The adjustment can be manual. The simulation parameters to be adjusted are, for example, thermal resistance, electrical contact resistance, surface emissivity, air gap modeling, etc. In one or more embodiments, the simulation parameters are fixed throughout the simulation and thus are not adjusted.

[0053] The measurement results can also be used as real data for updating one or more ML parameters applied by the ML engine 114. In particular, the comparison engine 116 can update the time constant calculated and applied by the ML engine 114. The comparison engine 116 can receive actual measurement data obtained by operating the asset 10 at a specific current and predicted transient data for the specific current. The actual measurement data can be received from multiple sensors 122 at time intervals for performing tests to obtain measurement values.

[0054] The comparison engine 116 can compare the measurement data with the predicted transient data. In a first correction performed to correct the transient simulation data used to simulate the transient of the asset 10, the time constant is adjusted to mitigate the difference between the measurement data and the predicted transient data. The first correction is performed using a fixed algorithm that uses a regression model. The fixed algorithm is used regardless of the measurement data or predicted transient data values.

[0055] In addition, measurements are continuously obtained from the sensors 122 located at the basic monitoring points for monitoring the integrity of the asset within its environment and the safety associated with the asset, such as for monitoring the integrity of the electrical connection (which may loosen over time or be affected by torque or other forces) and / or monitoring the risk of overheating.

[0056] The maintenance system 124 receives and monitors the prediction data, including the predicted steady-state data and the predicted transient data. The maintenance system 124 can detect anomalies and make determinations, decisions, and recommendations related to the maintenance, safety, and optimization of the asset 10. These determinations and decisions can control the processes for maximizing maintenance, safety, and / or optimization. The augmented reality system 126 receives the prediction data and uses it to provide a two-dimensional or three-dimensional visualization of the predicted thermal conditions in the accessible and inaccessible areas of the asset 10. The thermal prediction data represents the thermal conditions in the accessible and inaccessible areas. The thermal prediction data includes a prediction of a continuum of thermal conditions formed by highly concentrated data points, where the number of data points forming the continuum is several orders of magnitude larger than the discrete data points from which the sensors 122 obtain actual measurements.

[0057] The maintenance system 124 receives the thermal prediction data provided by the virtual model 112 and the boundary conditions of the set of boundary conditions 132 to make decisions regarding the maintenance of the physical asset. The boundary conditions can be updated when the steady-state simulation data is updated. The decisions regarding maintenance can include, for example, decisions to request or modify one or more of the orders for maintenance or safety tasks, which can include generating or modifying a maintenance schedule, decisions to perform repair or replacement tasks, outputting a control signal to generate an alarm when a safety issue arises and / or controlling the components associated with the physical asset, and / or controlling the components associated with the physical asset to optimize the operation of the asset. The augmented reality system 126 receives the thermal prediction data from the virtual model 112 and provides a two-dimensional or three-dimensional augmented reality visualization of the predicted thermal conditions in the accessible and inaccessible areas. In this way, the user can view the predicted thermal conditions in the inaccessible or invisible areas of the physical asset.

[0058] The input to the simulation engine 110 includes boundary conditions, geometric data, information about materials, and simulation parameters from simulation data. The boundary conditions provided by the simulation data are used to generate a virtual model 112, which is used to generate prediction data that is compared in real time with measurement data (ground truth data) to update the simulation parameters used to generate the virtual model 112. This real-time adjustment is performed in the transient regime, which helps train the virtual model 112 over time. This training of the virtual model 112 enables improved modeling of accessible and inaccessible regions of the asset 10 over time. Additionally, the prediction data is compared with the measurement data (ground truth data) in real time and is used to recompute the time constants used by the ML engine 114 to generate transient prediction data. This helps the ML engine 114 with ML training over time to generate transient simulation data.

[0059] The UI 118 can receive requests (e.g., queries) from the user device 140, the maintenance system 124, and / or the augmented reality system 126. The requests can be submitted by a user or can be submitted automatically, e.g., periodically or in response to a condition.

[0060] Referring to the architecture of the thermal predictor 102 and its associated memory, the thermal predictor 102 includes a central processing unit (CPU), random access memory (RAM), and a storage medium, which can be connected by a bus and are used to further support the processing of data, such as Figure 4 shown and described. Programmable instructions can be stored in the storage medium and executed by the CPU to cause the CPU to perform the operations described herein. The thermal predictor 102 can be implemented as a physical or virtual device. Whether implemented as a physical or virtual device, the thermal predictor 102 uses local or remote hardware processing devices that execute software instructions, which enables the performance of the disclosed functions.

[0061] Each of the ML engine 114, the virtual model 112, and the UI 118 can be accessed by the thermal predictor 102 and can be integrated with or external to the thermal predictor 102. Additionally, each of the ML engine 114, the virtual model 112, and the UI 118 can be implemented using software, hardware, and / or firmware.

[0062] The external network 142 can include one or more WANs, such as the Internet, which can be used to provide communication between the user device 140 and the thermal predictor 102.

[0063] The user device 140 can be a computing device, such as a server, a laptop device, a network element (such as a router, a switch, and a firewall), an embedded computer device embedded in other devices, such as an appliance, a tool, a vehicle, or a consumer electronic product, a mobile device, such as a laptop device, a smartphone, a cellular phone, and a tablet computer. The user device 140 can operate as a client in a client / server exchange, such as requesting services from the thermal predictor 102. The user device 140 can be included in or associated with the maintenance system 124 or the augmented reality system 126.‎

[0064] The set of simulation parameters 130 can store data structures used by the thermal predictor 102. The data structures can be stored in a memory integrated with the thermal predictor 102 or a permanent storage device (such as a file system), or in a database system external to the thermal predictor 102. For example, the set of simulation parameters 130 can be stored in a storage device, which includes a computer system readable medium in the form of volatile or non-volatile memory or a storage medium, such as random access memory (RAM), a cache, a magnetic disk, an optical disk, etc. The storage device can be accessed by the thermal predictor 102 and can be integrated with or external to the thermal predictor 102.‎

[0065] The communication between the thermal predictor 102 and the sensor 122, the maintenance system 124, the augmented reality system 126, the set of simulation parameters 130, and / or the user device 140 can be via wired and / or wireless communication links and / or can be via the network 142, and the network 142 can be a local area network (LAN), a protected network, a wide area network (WAN), such as the Internet.

[0066] Figures 3A-3C An exemplary and non-limiting flowchart is shown, which shows a method related to thermal prediction modeling of industrial physical assets according to certain illustrated embodiments. Before turning to the description of Figures 3A-3C it should be noted that Figures 3A-3C the flowcharts in

[0067] show examples of performing the operation boxes in a specific order, as indicated by the lines connecting the boxes, but the various boxes shown in these flowcharts can be performed in different orders or different combinations or sub-combinations. It should be understood that in some embodiments, some of the boxes described below can be combined into a single box. In some embodiments, one or more additional boxes can be included. In some embodiments, one or more boxes can be omitted.‎

[0067] Refer to Figure 3A and the flowchart 300 shows a method for training a thermal predictor (such as Figure 2An example method for performing a training phase of thermal prediction modeling of an industrial physical asset by the predictor 102 shown. Flowchart 300 includes a steady-state side 340 that processes operations for generating a steady-state regression model and a transient side 342 that processes operations for generating a transient regression model.

[0068] Blocks 302, 304, and 322 are executed by the simulation part 350 of the thermal predictor. Blocks 306, 308, 310, 312, 316, 324, 326, and 328 are executed by the ML part 352 of the thermal predictor. Blocks 314, 318, 320, and 330, 332, and 334 are executed by the comparison and measurement part 354 of the thermal monitoring system. The comparison is performed by a comparison engine, such as Figure 2 the comparison engine 116 shown. The measurement is performed by a sensor, such as Figure 2 the sensor 122 shown. In Figure 3A this, the outputs of blocks 304 and 322 each represent a part of the virtual model 112.

[0069] At block 302, a geometric model of the physical asset and initial simulation parameters for simulating the physical asset are generated or received. The geometric model can be generated by a CAD application operating on a user input file that includes data about the physical asset. The simulation parameters can be input by the user through a user interface. The simulation parameters are inputs to a simulation engine (such as Figure 2 the simulation 110 shown), which can be selected as any simulation parameters that are realistic for the physical asset.

[0070] Blocks 304, 306, 308a, 310, 312, 314, and 316 form a steady-state loop for processing steady-state simulations, where block 320 provides measurements from a large number of test monitoring points used by both the steady-state loop and the transient loop. The initial simulation parameters are used to generate a first steady-state simulation when the steady-state loop is executed for the first time. At block 316, the simulation parameters are adjusted for subsequent iterations of the steady-state loop. The adjustment at block 316 can be performed manually. Each iteration of the steady-state loop produces a different steady-state simulation.

[0071] Focusing on the processing of steady-state simulations, at block 304, a steady-state simulation program is executed that operates on the geometric model using cooling simulation software and the simulation parameters to produce a steady-state virtual model. The virtual model is an executable software model that provides a steady-state simulation of the asset. The steady-state virtual model defines multi-dimensional (2D or 3D) steady-state simulation points and their associated temperatures, where the steady-state simulation points correspond to a 2D or 3D representation of the asset when operating in a steady state. The steady-state simulation points can be numerous, for example, thousands or millions of points.

[0072] As indicated by the arrows from blocks 314 and 316, block 304 is updated and the updated output of block 304 provides the steady-state portion of the virtual model. Steady-state simulations are provided for each of two or more different sets of simulation parameters for the asset, such as a nominal current and one or more lower currents, each of which is a selected corresponding percentage below the nominal current, ambient temperature, etc. Still focusing on the processing of the steady-state simulations, at block 306, for each steady-state simulation, steady-state simulation points (e.g., all steady-state simulation points, not limited thereto) and their associated temperatures are extracted. In some cases, millions of points can be extracted. At block 308A, a clustering algorithm is applied to group all the points extracted from all iterations performed so far at block 306 into steady-state clusters such that points with similar temperatures are clustered together. The clustering algorithm can be, for example, the K-means clustering algorithm, such as provided in sklearn.cluster. Some of the extracted outliers or points not relevant to the simulation may not be assigned to the steady-state clusters.

[0073] At block 310, polynomial regression is applied to each steady-state cluster. In this way, a steady-state regression model is developed that determines how the simulated temperature relates to the simulation parameters for the steady-state simulation iterations using the corresponding different simulation parameters for the simulated temperature associated with the corresponding steady-state simulation points.

[0074] At block 312, steady-state predictions are made using the steady-state regression model to predict the temperature for any selected set of simulation parameters that define the simulation points. Block 312 uses the steady-state regression model and interpolation and / or extrapolation to predict the temperature while training the virtual model, such as virtual model 112 of thermal predictor 102. For example, when performing a real test, the set of simulation parameters can be selected to be the same as the set of real parameters that define the test monitoring points used at block 320. Temperature predictions can be made for many corresponding simulation points, referred to as steady-state prediction data. The simulation points can be set at the same and different locations as the test monitoring points used at block 320.

[0075] Figure 3A Arrows from block 312 to block 328 are shown for providing the steady-state temperature predictions to block 328. In one or more embodiments, the steady-state temperature predictions are provided to block 328 after it is determined at block 314 that the comparison result is satisfactory, such as below a predetermined threshold. This is indicated by the dashed arrow from block 314 to block 312 and the dashed arrow from block 312 to block 328. The present disclosure is not limited to the specific criteria that allow the transfer of the steady-state temperature predictions to block 328, as other criteria can alternatively be used.

[0076] At block 320, a real test is performed to obtain measurement data from sensors 122 set at a large number of test monitoring points. The test monitoring points used can include as many available test monitoring points as possible. Such a large number of test monitoring points are especially used in the learning phase. The test performed at block 320 can be performed only once or as needed. Block 320 is not included in the loop.

[0077] At block 314, the real-time measurement data is compared with the steady-state prediction data. When a test monitoring point has a corresponding simulation point, this comparison can determine the difference between the predicted temperature of the simulation point and the measured temperature of the monitoring point. When a monitoring point has no corresponding simulation point, this comparison can determine the difference between the extrapolated predicted temperature of the interpolated simulation point and the measured temperature of the monitoring point. Determine whether the difference (also known as the steady-state prediction error) between the real-time measurement data and the steady-state prediction data of the test monitoring point exceeds a predetermined steady-state threshold.

[0078] At block 316, the simulation parameters are adjusted to reduce the steady-state prediction error. A user interface, such as Figure 2 the UI 118 shown, can be used to manually adjust the simulation parameters. The next iteration of the steady-state loop is performed using the adjusted simulation parameters, starting again from block 304. For each iteration of the steady-state loop, the ML engine re-runs its two phases, including re-running the clustering and the steady-state regression model. The steady-state loop can be repeated until the steady-state prediction error is below the steady-state threshold. In this way, the steady-state simulation generated for the second iteration of the steady-state loop provides updated steady-state simulation data and updated boundary conditions as a function of the real-time measurement.

[0079] In addition, when the steady-state prediction error is low enough (e.g., the steady-state prediction error does not exceed the steady-state threshold), at block 318, the training phase ends, and the steady-state regression model trained at block 310 and the transient regression model trained at block 326 are ready to perform predictions during operation, as Figure 3B shown. In addition, at block 318, the steady-state prediction data is provided to a maintenance system and / or an augmented reality system, such as Figure 2 the maintenance system 124 and the augmented reality system 126 shown.

[0080] The maintenance system 124 can use the steady-state prediction data to generate a maintenance schedule for the physical asset, which can include, for example, detecting anomalies, making determinations, and outputting one or more of recommendations (such as instructions and / or alerts) related to maintenance, safety, and optimization of the physical asset.

[0081] The augmented reality system 126 can use the steady state prediction data to provide a two - dimensional or three - dimensional visualization of the predicted thermal conditions in the accessible and inaccessible regions of the physical asset. In one or more embodiments, providing the steady state prediction data to the maintenance system 124 and the augmented reality system 126 as described is the end - use of the steady state prediction data, after which the steady state model is used to make predictions.

[0082] Now turning to the boxes 322, 324, 326, 328, 330, 332, and 334 for transient simulation processing, the transient simulation processing can be executed in parallel or serially with the steady state simulation processing. There are two loops on the transient side 342. The first loop is the transient loop that returns at box 334, and the second loop is the small test loop that returns at box 332. The small test loop corresponds to each real - time loop during the actual test. The transient loop corresponds to the iterations between actual tests.

[0083] For each time step of a plurality of intervals of time steps and each cluster of each time step, the initial time constant is used when the transient loop is executed for the first time. At box 334, the time constant is adjusted for each iteration of the transient loop. At box 322, a transient simulation program is executed, which operates on the geometric model using cooling simulation software to produce a transient virtual model. When the transient simulation program is executed, the transient portion of the virtual model output at box 322 provides a transient asset simulation that varies dynamically over time. Time can be divided into time steps. The transient virtual model defines multi - dimensional (2D or 3D) transient simulation points and their associated temperatures, where the transient simulation points correspond to a 2D or 3D representation of the asset when operating transiently. The transient simulation points can be numerous, for example, thousands or millions of points.

[0084] At box 324, all transient simulation points (e.g., all transient simulation points, but not limited to this) of the transient simulation, and their associated temperatures at each time step, are extracted. At box 308B, a clustering algorithm (the same clustering algorithm used at box 308A) is applied to group all transient simulation points on time steps with similar temperatures (meaning clustering multiple or all time steps, with no limitation on this), thereby producing one or more transient clusters. Some of the extracted outliers or points not relevant to the simulation are not assigned to the transient clusters. The initial time constant is inherently associated with each transient cluster, and can be determined for each transient cluster using a transient regression model.

[0085] Starting the transient loop, at box 326, transient regression is applied to each of the one or more transient clusters using the most recent time constant of each transient cluster. The initial time constant is used for the first iteration of the transient loop, and the adjusted time constant (which is adjusted at box 334) is used for the corresponding transient clusters in subsequent iterations. Transient regression can use exponential regression with a single term or an exponential sum.

[0086] At block 328, once a steady state prediction is provided (as indicated by the dashed arrow from block 312 to block 328, e.g., after it is determined at block 314 that a predetermined criterion is met), a new transient temperature is predicted for the corresponding simulation point. The prediction process applies a transient regression model in real time (which means it is predicted as the actual temperature at that time, as simulated). The predicted transient temperature for each cluster is output as transient prediction data. Based on the most recent adjustment of the relevant time constant, the real-time transient temperature is predicted as a function of the steady state temperature prediction determined at block 312 at the simulation time, the previous (which can be the most recent, but is not limited to) predicted transient temperature of the relevant cluster obtained at an earlier simulation time (for the first iteration, it can be a default value, e.g., the ambient temperature), the amount of simulation time elapsed since the earlier simulation time, and the time constant of the corresponding transient cluster. This prediction can apply Newton's law of cooling or a variant thereof as shown in Equation (1) below. Note that the present disclosure is not limited to Equation (1), and other equations applying Newton's law of cooling or its variants can be used.

[0087] Equation (1)

[0088] Where, T i+1 is the predicted temperature of the transient cluster, T i is the last temperature of the transient cluster, T amb is the ambient temperature, T ss is the steady state temperature, t is the time elapsed since obtaining T i started, and τ is the time constant of the transient cluster. In the example shown, an exponential sum is used and multiple time constants are provided.

[0089] While performing the steady state and transient simulations, during the training phase, the sensor 122 set at as many test monitoring points as possible can be used to perform block 320 once to obtain temperature measurement data from the sensor 122. At block 330, the evolution of the measurement data is compared with the time constant associated with the corresponding transient cluster. If the evolution difference between the measurement data and the corresponding time constant exceeds a predetermined transient threshold for comparison, the method continues at block 334 to recalculate the time constant based on the evolution of the measurement data, and then starts the iteration of the transient loop at block 326 using the adjusted time constant. The adjustment of the time constant can be determined using a formula based on Newton's law of cooling. If the transient prediction error is below the predetermined transient threshold, the method continues at block 318.

[0090] The inner loop including blocks 330 and 332 is continuously executed until block 318 is executed. At block 332, measurement data is received from sensor 122 disposed at the basic monitoring point. The basic monitoring points are much fewer than the test monitoring points used at block 320. For example, in one or more embodiments, compared to about 5 - 10 monitoring points used at block 320, at block 332, 1 or 2 basic monitoring points are used per power phase (e.g., three power phases, but not limited to a specific number of power phases). The number of monitoring points is provided as an example and is not intended to limit the present disclosure to a specific number of monitoring points at block 220 or 332. Thus, each time block 330 is executed, the comparison uses the updated measurement of the basic monitoring point most recently provided at block 332. Block 332 uses a formula to correct the predicted temperature. The formula can use any error calculation, such as percentage error calculation.

[0091] At block 334, the result of the comparison performed at block 330 is used to adjust the time constant at block 326. The adjusted time constant determined at block 334 is provided to block 326, and the transient loop is repeated, e.g., until the transient prediction error determined at block 330 is low enough (e.g., not exceeding the transient threshold), which indicates that the exponential regression model is sufficiently trained. Once the regression model trained at block 310 and the exponential regression model trained at block 326 are sufficiently trained, they can be used for real - time temperature prediction during asset operation.

[0092] Reference Figure 3B , flowchart 370 shows an example method for real - time processing of steady - state and transient prediction data during an operating phase in which an asset (e.g., Figure 1 the asset 10 shown) is operating. The method can be executed by a maintenance and / or augmented reality system, such as Figure 1 the maintenance system 124 and / or augmented reality system 126 shown. The method begins at blocks 372 and 374. At block 372, an initial measured temperature is received from the asset. At block 374, at least one regression model (e.g., the steady - state regression model trained at block 310 and the transient regression model trained at block 326), and the corresponding cluster time constants adjusted at block 326 are received.

[0093] Block 376 begins a prediction loop, which includes blocks 376, 378, 380, 382, 384, and 386. The prediction loop can be iteratively executed.

[0094] At block 376, real-time measured per-phase electric current and temperature are received from sensors disposed at the basic monitoring points. At block 378, temperature data is predicted in real-time, e.g., for selectable prediction points, where the prediction points can be selected to include prediction points located at the same location as the basic monitoring points and prediction points at different locations from the basic monitoring points, e.g., including locations of inaccessible assets. In other words, the prediction points can be selected to include prediction points the same as and different from the basic monitoring points.

[0095] The temperature prediction for the prediction points is performed by applying at least one trained regression model using the real-time received current measurements, the previously predicted temperature of the prediction points (using the initial measured temperature for the first execution of block 378), the amount of time elapsed since the previous predicted temperature was predicted, and the time constant received at block 374. As shown in example equation (1), the time constant is used as the exponential term in calculating the predicted temperature.

[0096] At block 380, the predicted temperatures of a subset of the prediction points are compared with the real-time temperature measurements of the corresponding basic monitoring points received at block 376. Block 380 reflects block 330 of the training phase. At block 382, it is determined whether the difference between the predicted temperature data and the measured temperature data compared at block 380 is significant.

[0097] If it is determined at block 382 that the difference is significant, the method continues at block 388. Such a significant difference may indicate that the asset is not operating as expected and that action needs to be taken to avoid problems or to find the cause of the significant difference. At block 388, one or more actions are caused to be performed to maintain the physical asset, protect the safety of the asset and / or its environment, and / or optimize the operation of the asset. These actions can be physical actions and / or computational actions affecting the asset itself. In one or more embodiments, the action affects the operation of the asset or an action performed on the asset. In one or more embodiments, the action causes the execution of a repair or replacement operation. In one or more embodiments, the action updates the maintenance schedule of the asset, thereby causing the execution of a repair or replacement operation. In one or more embodiments, the action includes outputting a control signal to control a component associated with the physical asset, e.g., disabling the component to address a safety issue or to optimize the operation of the asset. In one or more embodiments, the action causes the generation of an alert, e.g., when a safety issue occurs.

[0098] If it is determined at block 382 that the difference is not significant, the method continues at block 384. At block 384, the predicted temperature of the prediction point is corrected using the result of the comparison performed at block 380. The result of the comparison can include, for example, the difference between the predicted temperature of the prediction point and the received temperature of the corresponding base monitoring point. Thus, correcting the predicted temperature based on the result of the comparison performed at block 380 includes correcting the predicted temperature of the prediction point based on the difference between the predicted temperature of the prediction point and the temperature of the corresponding base monitoring point received. For example, the difference can be used as a calibration form. Block 384 corrects the predicted temperature using a formula. The formula can use any error calculation, such as a percentage error calculation. Block 384 reflects block 332 of the training phase.

[0099] At block 386, the predicted temperature is displayed. A timestamp of the predicted temperature is provided for the next iteration of the prediction loop for use by block 378 during the next iteration for prediction.

[0100] Reference Figure 3C , shows a flowchart of an example method of performing Figure 3B block 278. At least one regression model includes a steady-state regression model using polynomial regression and a transient regression model using exponential regression. The method begins at block 390, where steady-state temperature prediction is performed using the steady-state regression model to predict the steady-state temperature at the prediction point. Block 390 reflects block 312 of the training phase. The method continues at block 392, where the transient regression model is applied using the steady-state prediction (once acceptable), the last predicted transient temperature (which can be the initial measured temperature of the first iteration), the time elapsed since predicting the last predicted transient temperature, and the corresponding time constant. The predicted temperature of the prediction point provided in real time by block 378 includes the predicted transient temperature. Block 392 reflects block 328 of the training phase.

[0101] Above, various embodiments have been referenced. However, the scope of the present disclosure is not limited to the specifically described embodiments. Instead, any combination of the described features and elements, whether or not related to different embodiments, is considered to implement and practice the contemplated embodiments. Further, although an embodiment may achieve advantages over other possible solutions or the prior art, whether a given embodiment achieves a particular advantage does not limit the scope of the present disclosure. Thus, the foregoing aspects, features, embodiments, and advantages are merely illustrative and are not considered elements or limitations of the appended claims unless expressly recited in the claims.

[0102] The various embodiments disclosed herein may be implemented as a system, a method, or a computer program product. Accordingly, aspects may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.) or an embodiment combining software and hardware aspects, which are generally referred to herein as “circuits,” “modules,” or “systems.” Additionally, aspects may take the form of a computer program product embodied in one or more computer-readable media having computer-readable program code thereon.

[0103] Any combination of one or more computer-readable media may be utilized. The computer-readable media may be a non-transitory computer-readable media. A non-transitory computer-readable media may be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the non-transitory computer-readable media may include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. The program code embodied on the computer-readable media may be transmitted using any appropriate medium, including but not limited to wireless, wireline, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

[0104] The computer program code for performing operations of aspects of the present disclosure may be written in any combination of one or more programming languages. Additionally, such computer program code may be executed using a single computer system or by multiple computer systems in communication with each other (such as using a local area network (LAN), a wide area network (WAN), the Internet, etc.). Although the various features have been described above with reference to flowcharts and / or block diagrams, those of ordinary skill in the art will understand that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented by computer logic (such as computer program instructions, hardware logic, combinations of both, etc.). Generally, the computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device. Additionally, execution of such computer program instructions by the processor produces a machine capable of performing the functions or acts specified in the blocks of the flowcharts and / or block diagrams.

[0105] Figure 1 The illustrated embodiment of the thermal predictor 102 may be implemented or executed by one or more computer systems. For example, the thermal predictor 102 may use, for instance Figure 4implemented by the computer system of the illustrated example computer system 400. In various embodiments, the computer system 400 can be a server, mainframe computer system, workstation, network computer, desktop computer, laptop computer, etc., and / or include one or more of a field programmable gate array (FPGA), application specific integrated circuit (ASIC), microcontroller, microprocessor, etc.

[0106] The computer system 400 is only an example of a suitable system and is not intended to impose any limitation on the scope of use or functionality of the embodiments of the present disclosure described herein. In any case, the computer system 400 is capable of implementing and / or performing any of the functions set forth above.

[0107] The computer system 400 can be described in the general context of a computer system, such as executable instructions like program modules executed by the computer system. Generally, program modules can include routines, programs, objects, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. The computer system 400 can be implemented in a distributed data processing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed data processing environment, program modules can be located in local and remote computer system storage media including memory storage devices.

[0108] The computer system 400 is shown in the form of a general-purpose computing device. The computer system 400 includes one or more processors 402, a memory 404, an input / output (I / O) interface (I / F) 406 that can communicate with internal components such as a user interface 410, and an optional external component 408.

[0109] The thermal predictor 102 can be configured to process large amounts of data. The computer system used to implement the thermal predictor 102 can be implemented, for example, using a multi-processor, big data architecture, or one or more cloud-based computer systems.

[0110] The processor 402 can include, for example, a single-core or multi-core processor, a programmable logic device (PLD), a microprocessor, a DSP, a microcontroller, an FPGA, an ASIC, and / or other discrete or integrated logic circuits with similar processing capabilities.

[0111] The processor 402 and the memory 404 may include components provided in, for example, an FPGA, an ASIC, a microcontroller, or a microprocessor. The memory 404 may include, for example, volatile and non-volatile memories for temporarily or long-term storing data and for storing programmable instructions executable by the processor 402. The memory 404 may be a removable (e.g., portable) memory for storing program instructions. The I / O I / F 406 may include interfaces and / or conductors to couple to one or more internal components and / or external components 408.

[0112] By way of example and not limitation, the program instructions may include program modules for Figure 1 the ML engine 114, the virtual model 112, and the UI 118 shown, as well as an operating system, one or more application programs, other program modules, and program data. Each of the operating system, one or more application programs, other program modules, and program data or some combination thereof may include an implementation of the network environment. The program modules generally perform the functions and / or methods of the embodiments of the present disclosure described herein.

[0113] These computer program instructions may also be stored in a computer-readable medium, which can direct a computer, other programmable data processing apparatus, or other devices to operate in a particular manner, such that the instructions stored in the computer-readable medium produce a manufacture including instructions implementing the functions / actions specified in one or more boxes of the flowchart and / or block diagram.

[0114] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operations to be performed on the computer, other programmable apparatus, or other devices, thereby producing a computer-implemented process. When executed on a computer or other programmable apparatus, the instructions provide a process for implementing the disclosed functions / actions, including those specified in one or more boxes of the block diagram.

[0115] Embodiments of the processing component of the thermal predictor 102 may be implemented or executed by one or more computer systems 400. Each computer system 400 or multiple instances thereof may be included within the thermal monitoring system 100. The computer system 400 may be provided as an embedded device or include an embedded device. Portions of the computer system 400 may be provided externally, e.g., via virtual, centralized, and / or cloud-based computers.

[0116] The computer system 400 is only one example of a suitable system and is not intended to impose any limitation on the scope of use or functionality of the embodiments of the present disclosure described herein. In any event, the computer system 400 is capable of implementing and / or executing any of the functions set forth above.

[0117] The computer system 400 can be described in the overall context of a computer system, such as executable instructions like program modules executed by the computer system. Generally, program modules can include routines, programs, objects, components, logics, data structures, etc., which perform specific tasks or implement specific abstract data types.

[0118] The term "comprising" or "including" shall be construed as specifying the presence of the stated features, integers, operations or components, but not precluding the presence of one or more other features, integers, operations or components or combinations thereof.

[0119] Those of ordinary skill in the art understand that any numerical value disclosed herein can be an exact value or a value within a certain range. Additionally, any approximate terms (such as "about", "approximately", "around") used in this disclosure can represent the value within a range. For example, in some embodiments, the range can be within (plus or minus) 20%, or 10%, or 5%, or 2%, or within any other suitable percentage or number understood by those of ordinary skill in the art (such as for known tolerances or error ranges).

[0120] Unless the context clearly dictates otherwise, the articles "a", "an", and "the" used in this specification and the appended claims refer to one or more than one (i.e., at least one) of the grammatical objects of the article. For example, "an element" means one element or more than one element.

[0121] The phrase "and / or" used in the specification and claims should be understood to mean "either or both" of the elements so conjoined, i.e., elements that are present conjunctively in some cases and disjunctively in others. Multiple elements listed with "and / or" should be construed in the same way, i.e., "one or more" of the elements so conjoined. Other elements may optionally be present in addition to the elements specifically identified by the "and / or" clause, whether or not related to those specifically identified. Thus, as a non-limiting example, in one embodiment, when used in conjunction with open-ended language such as "comprising", a reference to "A and / or B" can refer only to A (optionally including elements other than B); in another embodiment, only to B (optionally including elements other than A); in yet another embodiment, to both A and B (optionally including other elements), etc.

[0122] As used in the specification and claims, "or" shall be understood to have the same meaning as "and / or" as defined above. For example, when separating items in a list, "or" or "and / or" shall be interpreted inclusively, i.e., including at least one, but also including more than one of a plurality or series of elements, and optionally, additional unlisted items. Only terms that explicitly state the contrary, such as "only one" or "exactly one", or when used in a claim, "comprising" will refer to including exactly one element of a plurality or series of elements. In general, the term "or" as used herein shall only be interpreted as representing an exclusive alternative (i.e., "one or the other, but not both") when preceded by an exclusive term, such as "either", "one", "only one" or "exactly one".

[0123] The techniques described herein are exemplary and should not be construed as implying any particular limitation to certain illustrated embodiments. It should be understood that those skilled in the art can devise various substitutions, combinations, and modifications. For example, operations associated with the processes described herein can be performed in any order, unless the operations themselves are otherwise specified or indicated. The present disclosure is intended to embrace all such substitutions, modifications, and variations that fall within the scope of the appended claims.

[0124] Above, various embodiments have been referenced. However, the scope of the present disclosure is not limited to the specifically described embodiments. Instead, any combination of the described features and elements, whether or not associated with different embodiments, is considered to implement and practice the contemplated embodiments. Additionally, although an embodiment may achieve advantages over other possible solutions or the prior art, whether a given embodiment achieves a particular advantage does not limit the scope of the present disclosure. Accordingly, the foregoing aspects, features, embodiments, and advantages are merely illustrative and are not considered to be elements or limitations of the appended claims, unless explicitly recited in the claims.

[0125] The various embodiments disclosed herein may be implemented as a system, method, or computer program product. Accordingly, aspects may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects, which are generally referred to herein as "circuitry", "module", or "system". Additionally, aspects may take the form of a computer program product embodied in one or more computer-readable media having computer-readable program code thereon.

[0126] Any combination of one or more computer-readable media can be utilized. The computer-readable media can be non-transitory computer-readable media. Non-transitory computer-readable media can be, by way of example and not limitation, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of non-transitory computer-readable media can include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. Program code embodied on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wireline, fiber optic cable, RF, etc., or any suitable combination of the foregoing.

[0127] The computer program code for performing the operations of aspects of the present disclosure can be written in any combination of one or more programming languages. Additionally, such computer program code can be executed using a single computer system or by multiple computer systems in communication with each other (such as using a local area network (LAN), a wide area network (WAN), the Internet, etc.).

[0128] The flowcharts and block diagrams in the figures illustrate the architectures, functions, and / or operations of possible implementations of various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram can represent a module, segment, or portion of code, which includes one or more executable instructions for implementing the specified logical function. It should also be noted that, in some alternative implementations, the functions noted in the blocks may not occur in the order noted in the figures. For example, two blocks shown in succession may in fact be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can be implemented by a system based on dedicated hardware for performing the specified functions or actions, or by a combination of dedicated hardware and computer instructions.

[0129] It should be understood that the foregoing description is intended to be illustrative, not restrictive. After reading and understanding the foregoing description, many other example embodiments will be apparent. Although the present disclosure describes specific examples, it should be recognized that the systems and methods of the present disclosure are not limited to the examples described herein, but can be implemented with modifications within the scope of the appended claims. Accordingly, the specification and drawings are to be regarded as illustrative rather than restrictive. Therefore, the scope of the present disclosure should be determined with reference to the appended claims and the full scope of equivalents of these claims.

Claims

1. A method, comprising: Receiving a time constant and at least one trained regression model determined during a training phase that applies machine learning to simulated multi-dimensional simulation points of an asset and temperatures associated with the respective simulation points; Receiving real-time measured current used by the asset; Receiving real-time measured temperature at a base monitoring point; Real-time predicting the temperature of a prediction point by applying at least one trained regression model and using the real-time measured current, a previous predicted temperature of the prediction point, the time elapsed since the previous predicted temperature was predicted, and the time constant, wherein the prediction point is selectable to include the same prediction point as the base monitoring point and prediction points different from the base monitoring point; Comparing the predicted temperature of a subset of prediction points with the currently received temperature of the base monitoring point corresponding to the subset of prediction points; Using the comparison result to correct the predicted temperature of the selected prediction point; And Real-time outputting the predicted temperature.

2. The method according to claim 1, further comprising using the predicted temperature to real-time update an augmented reality visualization of the asset.

3. The method according to claim 1, further comprising: Determining whether a difference between the predicted temperature of the subset of prediction points and the received temperature at the corresponding base monitoring point exceeds a threshold; And Causing an action affecting the asset in response to a determination that the difference exceeds the threshold.

4. The method according to claim 1, wherein The time constant is associated with a respective clustering of the simulation points, and the at least one regression model is determined from clusters of simulation data.

5. The method according to claim 1, wherein The at least one regression model includes a steady-state regression model using polynomial regression and a transient regression model using exponential regression, and predicting the temperature at the prediction point includes: Predicting the steady-state temperature at the prediction point by applying the steady-state regression model; and Predicting the transient temperature at the prediction point by applying the transient regression model using the predicted steady-state temperature at the prediction point, the real-time measured current, a previous predicted transient temperature of the prediction point, the time elapsed, and the time constant, wherein the predicted temperature of the real-time prediction point includes the predicted transient temperature.

6. The method according to claim 1, wherein The simulation is a digital twin.

7. The method according to claim 1, wherein The simulation includes two or more steady-state simulations using different simulation parameters, and the method further comprises repeating the following during the training phase until the steady-state prediction is determined to be acceptable: For two or more steady-state simulations, extracting steady-state simulation points from the simulation points and temperatures associated with each steady-state simulation point; For each of the two or more steady-state simulations, applying a clustering algorithm to the extracted steady-state simulation points and their respective associated temperatures to form a plurality of steady-state clusters; Applying the steady-state regression model to each steady-state cluster to represent the relationship between the temperature associated with the respective steady-state simulation point and the simulation parameters; Generating a steady-state prediction by applying the steady-state regression model to the selected simulation parameters for predicting the steady-state temperature of the steady-state simulation points at the selected simulation parameters; Determining a steady-state difference between the predicted steady-state temperature of the steady-state simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein the steady-state prediction is determined to be acceptable when the steady-state difference is below a steady-state threshold; And Adjust the selected simulation parameters for reuse in a next iteration, if any, to attempt to reduce the steady state error.

8. The method according to claim 7, wherein The simulation includes a transient simulation and the method further includes, during a training phase: For a plurality of spaced time steps, extract transient simulation points from the simulation points and the temperature associated with each transient simulation point; Apply a clustering algorithm to the transient simulation points and their temperatures extracted over the plurality of spaced time steps to form a plurality of transient clusters; and Repeat the following until the transient prediction is determined to be acceptable: Apply a transient regression model to each transient cluster using the nearest time constant associated with each transient cluster; Once the steady state prediction is determined to be acceptable, generate a transient prediction for predicting a most recent temperature associated with a corresponding transient cluster by applying the transient regression model using the steady state prediction, a previous predicted transient temperature of the corresponding transient cluster obtained at an earlier simulation time, an amount of simulation time elapsed since the earlier simulation time, and the nearest time constant of the corresponding transient cluster; Determine a transient error between the predicted transient temperature of the transient simulation points and the measured temperature at a plurality of corresponding monitoring points of the asset, wherein the transient prediction is determined to be acceptable when the transient error is below a transient threshold; And Adjust the time constant for reuse in a next iteration, if any, to attempt to reduce the transient error.

9. The method of claim 8, further comprising using at least one of the transient prediction and the steady state prediction to update a virtual reality visualization of the asset in real time.

10. The method of claim 8, further comprising: Obtain the measured temperature at the plurality of corresponding monitoring points; And Continuously update the measured temperature with the measurements obtained at a subset of the monitoring points for use in determining the transient error.

11. A thermal monitoring system for predicting temperature, the system comprising: A memory configured to store instructions; ‎ At least one processing device disposed at a location and in communication with the memory, wherein the at least one processing device is configured, when executing the instructions, to: Receive a time constant and at least one training regression model determined during a training phase that applies machine learning to simulated multi-dimensional simulation points of an asset and the temperature associated with the corresponding simulation points; Receive a real-time measured current used by the asset; Receive a real-time measured temperature measured at a base monitoring point; Predict in real time the temperature of a prediction point by applying at least one training regression model and using the real-time measured current, a previous predicted temperature of the prediction point, an amount of time elapsed since the previous predicted temperature was predicted, and the time constant, wherein the prediction point is selectable to include the same prediction points as the base monitoring point and prediction points different from the base monitoring point; Compare the predicted temperature of a subset of the prediction points with the currently received temperature of the base monitoring points corresponding to the subset of the prediction points; Use the comparison result to correct the predicted temperature of the selected prediction points; And Output the predicted temperature in real time.

12. The thermal monitoring system according to claim 11, wherein, When executing the instructions, the at least one processing device is further configured to update a virtual reality visualization of the asset in real time using the predicted temperature.

13. The thermal monitoring system according to claim 11, wherein, When executing the instructions, the at least one processing device is further configured to: Determine whether the difference between the predicted temperature of the predicted point subset and the received temperature at the corresponding basic monitoring point exceeds a threshold; and Cause an action affecting the asset to be executed in response to a determination that the difference exceeds the threshold.

14. The thermal monitoring system according to claim 11, wherein, The time constant is associated with a respective clustering of the simulation points, and the at least one regression model is determined from the clustering of the simulation data.

15. The thermal monitoring system according to claim 11, wherein, The at least one regression model includes a steady-state regression model using polynomial regression and a transient regression model using exponential regression, and predicting the temperature at the predicted point includes: Predicting the steady-state temperature at the predicted point by applying the steady-state regression model; and Predicting the transient temperature at the predicted point by applying the transient regression model using the predicted steady-state temperature at the predicted point, the real-time measured current, the previously predicted transient temperature of the predicted point, the time elapsed, and the time constant, wherein the predicted temperature of the real-time predicted point includes the predicted transient temperature.

16. The thermal monitoring system according to claim 11, wherein, The simulation includes two or more steady-state simulations using different simulation parameters, and wherein during the training phase, the at least one processing device is further configured, when executing the instructions, to repeat the following until the steady-state prediction is determined to be acceptable: For two or more steady-state simulations, extract the steady-state simulation points among the simulation points and the temperatures associated with each steady-state simulation point; For each of the two or more steady-state simulations, apply a clustering algorithm to the extracted steady-state simulation points and their corresponding associated temperatures to form a plurality of steady-state clusters; Apply the steady-state regression model to each steady-state cluster to represent the relationship between the temperature associated with the respective steady-state simulation point and the simulation parameters; Generate a steady-state prediction by applying the steady-state regression model to the selected simulation parameters for predicting the steady-state temperature of the steady-state simulation points at the selected simulation parameters; Determine the steady-state difference between the predicted steady-state temperature of the steady-state simulation points and the measured temperatures at a plurality of corresponding monitoring points of the asset, wherein when the steady-state difference is below the steady-state threshold, the steady-state prediction is determined to be acceptable; And Adjust the selected simulation parameters for reuse in the next iteration, if any, to attempt to reduce the steady-state difference.

17. The thermal monitoring system according to claim 11, wherein, The simulation includes a transient simulation, and wherein during the training phase, the at least one processing device is further configured, when executing the instructions, to: For a plurality of spaced time steps, extract the transient simulation points among the simulation points and the temperatures associated with each transient simulation point; Apply a clustering algorithm to the extracted transient simulation points and their temperatures over a plurality of spaced time steps to form a plurality of transient clusters; and Repeat the following until the transient prediction is determined to be acceptable: Apply the transient regression model to each transient cluster using the nearest time constant associated with each transient cluster; Once the steady-state prediction is determined to be acceptable, generate a transient prediction for predicting the latest temperature associated with the respective transient cluster by applying the transient regression model using the steady-state prediction, the previously predicted transient temperature of the corresponding transient cluster obtained at an earlier simulation time, the amount of simulation time elapsed since the earlier simulation time, and the nearest time constant of the corresponding transient cluster. Determine the transient difference between the predicted transient temperature at the transient simulation points and the measured temperatures at multiple corresponding monitoring points of the asset, where the transient prediction is determined to be acceptable when the transient difference is below a transient threshold; and Adjust the time constant for reuse in the next iteration, if any, to attempt to reduce the transient difference.

18. A method of training at least one model for predicting temperature in an asset, the method comprising: Repeating the following until the steady-state prediction is determined to be acceptable: For two or more steady-state simulations using different respective simulation parameters, extract the steady-state simulation points and the temperatures associated with each steady-state simulation point; For each of the two or more steady-state simulations, apply a clustering algorithm to the extracted steady-state simulation points and their corresponding associated temperatures to form a plurality of steady-state clusters; Apply a steady-state regression model to each steady-state cluster to represent the relationship between the temperature associated with the corresponding steady-state simulation point and the simulation parameters; Generate a steady-state prediction by applying the steady-state regression model to the selected simulation parameters for predicting the steady-state temperature of the steady-state simulation points at the selected simulation parameters; Determine the steady-state difference between the predicted steady-state temperature of the steady-state simulation points and the measured temperatures at multiple corresponding monitoring points of the asset, where the steady-state prediction is determined to be acceptable when the steady-state difference is below a steady-state threshold; and Adjust the selected simulation parameters for reuse in the next iteration, if any, to attempt to reduce the steady-state difference.

19. The method according to claim 18, wherein, The simulation includes transient simulation, and the method further comprises, during the training phase: For multiple spaced time steps, extract the transient simulation points in the simulation points and the temperatures associated with each transient simulation point; Apply a clustering algorithm to the extracted transient simulation points and their temperatures over multiple spaced time steps to form a plurality of transient clusters; and Repeating the following until the transient prediction is determined to be acceptable: Apply a transient regression model to each transient cluster using the nearest time constant associated with each transient cluster; Once the steady-state prediction is determined to be acceptable, generate a transient prediction for predicting the latest temperature associated with the corresponding transient cluster by applying the transient regression model using the steady-state prediction, the previous predicted transient temperature of the corresponding transient cluster obtained at an earlier simulation time, the amount of simulation time elapsed since the earlier simulation time, and the nearest time constant of the corresponding transient cluster; Determine the transient difference between the predicted transient temperature at the transient simulation points and the measured temperatures at multiple corresponding monitoring points of the asset, where the transient prediction is determined to be acceptable when the transient difference is below a transient threshold; and Adjust the time constant for reuse in the next iteration, if any, to attempt to reduce the transient difference.

20. The method according to claim 19, further comprising using at least one of the transient prediction and the steady-state prediction to update in real time the augmented reality visualization of the asset.

Images