Systems and methods for determining component predicted life using machine learning models

Abstract

The present disclosure provides computer-implemented methods and systems for determining component predicted life. The method includes obtaining, by a computing system comprising one or more computing devices, an image of a component. The method also includes obtaining, by the computing system, a stress map of the component. The method further includes processing, by the computing system, the image and the stress map of the component with a machine learning model to generate, as an output of the machine learning model, a predicted life of the component.

Description

Technical Field

[0001] This disclosure relates in its entirety to systems and methods for determining the predicted lifespan of components by utilizing artificial intelligence and / or machine learning. Background Technology

[0002] In various applications, consistent and accurate prediction of component lifespan is often required. Such predictions can reduce damage caused by component failure and improve efficiency by allowing for improved plans to remove components from service.

[0003] One application requiring such consistent and accurate predictions is one where components are subjected to numerous extreme conditions, such as high temperatures, high pressures, and enormous stress loads. Over time, individual components of a device may suffer from creep, deformation, fatigue cracking, etc., which can shorten the service life of the components. Such problems may apply to, for example, some turbines, such as gas turbine systems. During turbine operation, various components within the turbine, particularly within the turbine section (collectively referred to as turbine components), such as turbine blades, may undergo creep due to high temperatures and stress. For turbine blades, creep can cause partial or complete blade elongation, causing the blade tip to contact a fixed structure, such as the turbine housing, and potentially leading to unwanted vibrations and / or performance degradation during operation. Furthermore, excessive creep can lead to creep rupture and resulting component breakage, which can result in unexpected power outages and damage to other components in the system.

[0004] Therefore, creep in components such as turbine parts can be monitored. One way to monitor creep is to place strain sensors on the component and analyze these sensors at various intervals to monitor the deformation associated with creep strain. This positioning can be both time-consuming and expensive, resulting in an inefficient deformation monitoring process.

[0005] Therefore, improved systems and methods for predicting component life are desired. For example, systems and methods that can consistently and accurately predict creep and thus allow for predictive life planning based on such predictions would be advantageous. Summary of the Invention

[0006] The aspects and advantages of the systems and methods according to this disclosure will be set forth in part in the following description, or may be apparent from the description, or may be learned by practice of the technique.

[0007] According to one embodiment, a computer-implemented method for determining the predicted lifetime of a component is provided. The method includes obtaining an image of the component by a computing system including one or more computing devices. The method also includes obtaining a stress map of the component by the computing system. Furthermore, the method further includes processing the image and stress map of the component by the computing system using a machine learning model to generate a predicted lifetime of the component as the output of the machine learning model.

[0008] According to another embodiment, a computing system is provided. The computing system includes one or more processors. The computing system also includes one or more non-transitory computer-readable media that co-store instructions, which, when executed by the one or more processors, cause the computing system to perform operations including: obtaining an image of a component; obtaining a stress map of the component; and processing the image and stress map of the component using a machine learning model to generate a predicted lifetime of the component as the output of the machine learning model.

[0009] These and other features, aspects, and advantages of the systems and methods of the present invention will become more readily understood with reference to the following description and the appended claims. The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. Attached Figure Description

[0010] This specification sets forth, with reference to the accompanying drawings, a complete and practicable disclosure of the systems and methods of the present invention, which are applicable to those skilled in the art, including the best mode of manufacturing and using the systems and methods of the present invention, wherein:

[0011] Figure 1 This is a schematic diagram of a system for obtaining images of components according to an embodiment of this disclosure;

[0012] Figure 2 This is a process flowchart for generating a predicted lifetime of a component according to an embodiment of this disclosure;

[0013] Figure 3 Examples of using machine learning and / or computer vision to enhance images according to embodiments of this disclosure are illustrated;

[0014] Figure 4 Examples of using machine learning and / or computer vision to modify images according to embodiments of this disclosure are illustrated;

[0015] Figure 5 Videos illustrating components according to embodiments of this disclosure;

[0016] Figure 6 Several images of components according to embodiments of the present disclosure are illustrated;

[0017] Figure 7 An example is illustrated of the processing of a component image by cropping an image according to an embodiment of the present disclosure;

[0018] Figure 8 Examples of processing of component images by at least one of pixel analysis, computer vision, and / or machine learning models according to embodiments of the present disclosure are illustrated.

[0019] Figure 9 A schematic diagram illustrating the grain structure detected on a component according to an embodiment of the present disclosure is shown;

[0020] Figures 10(A), 10(B), 10(C) and 10(D) illustrate a comparison between the detected grain structure and stress map of a component according to an embodiment of the present disclosure;

[0021] Figure 11 A computing system according to an embodiment of this disclosure is illustrated; and

[0022] Figure 12 This is a flowchart illustrating a method for determining the predicted lifespan of a component according to an embodiment of the present disclosure. Detailed Implementation

[0023] Reference will now be made in detail to embodiments of the systems and methods of the present invention, one or more examples of which are illustrated in the accompanying drawings. Each example is provided by way of explanation of the inventive technique and not as a limitation thereof. Indeed, it will be apparent to those skilled in the art that modifications and variations may be made to the inventive technique without departing from the scope or spirit of the technique protected by the claims. For example, features illustrated or described as part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, this disclosure is intended to cover such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0024] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any specific implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other specific implementations. Furthermore, unless otherwise specifically stated, all embodiments described herein should be considered exemplary.

[0025] The detailed description uses numbers and letters to refer to features in the drawings. Similar or analogous names in the drawings and description have been used to refer to similar or analogous parts of the invention. As used herein, the terms "first," "second," and "third" are used interchangeably to distinguish one part from another and are not intended to indicate the location or importance of the various parts.

[0026] As used herein, the terms “upstream” (or “upward”) and “downstream” (or “downward”) refer to the relative directions of fluid flow within a fluid passage. For example, “upstream” refers to the direction from which fluid flows, and “downstream” refers to the direction towards which fluid flows. However, as used herein, the terms “upstream” and “downstream” can also refer to electric current. The term “radial” refers to a relative direction substantially perpendicular to the axial centerline of a particular component, the term “axial” refers to a relative direction substantially parallel and / or coaxially aligned with the axial centerline of a particular component, and the term “circumferential” refers to a relative direction extending around the axial centerline of a particular component.

[0027] Terms with approximate meanings (such as “about,” “approximately,” “generally,” and “substantially”) are not limited to specified precise values. In at least some cases, approximate language may correspond to the precision of the instrument used to measure the value, or the precision of the method or machine used to construct or manufacture the component and / or system. For example, approximate language may refer to tolerances of 1%, 2%, 4%, 5%, 10%, 15%, or 20% of the individual value, the range of values, and / or the end values ​​of the defined range of values. When used in the context of angles or directions, such terms include those greater than or less than ten degrees of said angle or direction. For example, “generally vertical” includes directions within ten degrees of vertical along any direction (e.g., clockwise or counterclockwise).

[0028] Unless otherwise stated herein, the terms “connection,” “fixed,” “attached to,” etc., refer to direct connection, fixation, or attachment, as well as indirect connection, fixation, or attachment via one or more intermediate components or features. As used herein, the terms “comprising,” “including,” “having,” or any other variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, article of manufacture, or apparatus that includes a list of features is not necessarily limited to those features, but may include other features not expressly listed or inherent to such process, method, article of manufacture, or apparatus. Furthermore, unless expressly stated to the contrary, “or” means inclusive or as well as non-exclusive or. For example, condition A or B is satisfied by any one of the following: A is true (or exists) and B is false (or does not exist); A is false (or does not exist) and B is true (or exists); and both A and B are true (or exist).

[0029] Throughout this specification and claims, unless the context or language otherwise indicates otherwise, scope limitations are combined and interchanged, such scopes are identified and include all subscopes contained herein. For example, all scopes disclosed herein include end values, and end values ​​can be combined independently of each other.

[0030] This disclosure relates throughout to a method for predicting the life of a component (e.g., the creep life of a component). Specifically, the component may be a component of a gas turbine engine, such as a compressor component (e.g., compressor blades or stator blades) or a turbine component (e.g., turbine blades or fixed nozzles). The method may include two approaches. In a first approach (i.e., a manual approach), a computing system may receive or acquire an image (such as a picture, scan, or video) of the component. The computing system may also receive a stress map of the component. The stress map can be obtained through finite element analysis (FEA) of the component. Based on the stress map and the image, the system can determine or identify and crop the region of interest. The system may detect contours, filter contours representing actual grains on the component, and overlay the grains with the stress map. Based on the comparison of the grains and the stress map, the computerized system may extract one or more features and / or predict the remaining creep life of the component. Alternatively or additionally, in a second approach, the image and stress map may be provided to a machine learning model, which may then generate a prediction of the remaining creep life based on the input. Reference is now made to... Figure 1 A component 10 is provided. Component 10 (and more specifically, the substrate of the entire component 10) may include various types of components for a wide range of applications, such as components for high-temperature applications (e.g., components including nickel-based or cobalt-based superalloys, austenitic steels, etc.). In some embodiments, component 10 may include industrial gas turbine or steam turbine components, such as combustion components or hot gas path components. In some embodiments, component 10 may include turbine blades, compressor blades, vanes, shrouds, rotors, or transition elements. In other embodiments, component 10 may include any other components of a turbine, such as any other components for gas turbines, steam turbines, etc. In some embodiments, the component may include non-turbine components, including but not limited to automotive components (e.g., cars, trucks, etc.), aerospace components (e.g., aircraft, helicopters, space shuttles, aluminum parts, etc.), locomotive or rail components (e.g., trains, train tracks, etc.), structural, infrastructure, or civil engineering components (e.g., bridges, buildings, construction equipment, etc.) and / or power plant or chemical processing components (e.g., piping for high-temperature applications).

[0031] In an exemplary embodiment, component 10 is an isometric or oriented solidification component. For example, component 10 may be a cast component, and after casting, the melt in the mold may advantageously be isometric or oriented solidification.

[0032] In some embodiments, component 10 may be a new component (e.g., not yet installed in a machine or not yet having undergone any operating cycles). However, in exemplary embodiments, component 10 may have already undergone at least one operating cycle. For example, in an embodiment where component 10 is a turbine component, component 10 may have already undergone at least one operating cycle within a gas turbine. That is, during operation of the gas turbine, the turbine component may be exposed to high-temperature combustion gases, operational vibrations, and mechanical forces, resulting in performance degradation, warping, and / or creep of the turbine component over time. As used herein, "operating cycle" may refer to one or more components that have been operated within a machine (such as a gas turbine) for a certain amount of time. Thus, a component that has not yet undergone any operating cycle has never been used in operation of a machine (such as a gas turbine). The operating cycle of a gas turbine may be between about 0 hours and about 50,000 hours (or such as between about 500 hours and about 50,000 hours, or such as between about 1,000 hours and about 50,000 hours, or such as between about 1,000 hours and about 50,000 hours).

[0033] Figure 1 The coordinate system is also illustrated. The coordinate system includes an X-axis 50, a Y-axis 52, and a Z-axis 54, all of which are orthogonal to each other and defined by reference component 10.

[0034] Figure 1 A computing system 100 is further illustrated, which may include, for example, a data acquisition system 102 and a user computing system 104. The data acquisition system 102 typically acquires data about component 10, and the computing system 104 typically analyzes that data and performs various calculations and other functions as discussed herein. In particular, the computing system 100 according to this disclosure provides accurate and efficient predictions of the lifespan of component 10, as discussed herein.

[0035] It should be noted that the various subsystems in the computing system 100 (such as the data acquisition system 102, the user computing system 104, and other suitable subsystems) can be linked together as discussed herein, or they can be separate, discrete systems.

[0036] According to one embodiment, the data acquisition system 102 may include an imaging device 106 for acquiring one or more images of component 10. Such images may be in the form of discrete images (e.g., photographs) or videos, including multiple video frame images extracted from a video. For example, the imaging device 106 may include a lens assembly 110 and an image capture device 112. The lens assembly 110 typically magnifies the image viewed by the lens assembly 110 for processing by the image capture device 112. In some embodiments, the lens assembly 110 may be, for example, a suitable camera lens, telescope lens, etc., and may include one or more spaced-apart lenses to provide the desired magnification. The image capture device 112 may typically be in communication with the lens assembly 110 for receiving and processing light from the lens assembly 110 to generate an image. In an exemplary embodiment, for example, the image capture device 112 may be a camera sensor that receives and processes light from a camera lens to generate an image, such as a digital image, as generally understood. In some embodiments, image capture device 112 (and generally device 106) may also be connected to computing system 104 via, for example, a suitable wired or wireless connection, for storing and analyzing images generally derived from image capture device 112 and device 106. In some embodiments, user computing system 104 may operate imaging device 106 to perform various disclosed steps. In other embodiments, imaging device 106 may be a standalone device operated independently by a user and linked to user computing system 104, or it may be a separate, discrete system.

[0037] Additionally or alternatively, the data acquisition system 102 may also include a three-dimensional data acquisition device 108 for inspecting the outer surface 11 of the component 10. The device 108 according to this disclosure typically utilizes surface metrology techniques to obtain direct measurements of the component 10 along three axes. In particular, in an exemplary embodiment, non-contact surface metrology techniques may be utilized. Generally, any suitable three-dimensional data acquisition device 108 can be used, which utilizes surface metrology techniques to obtain direct measurements in three dimensions (such as blue light scanning). In an exemplary embodiment, the device 108 is a non-contact device utilizing non-contact surface metrology techniques.

[0038] According to one embodiment, in some exemplary embodiments, device 108 is a laser scanner that generates laser scan images. The laser scanner typically includes a laser 120 that emits light in the form of a laser beam toward an object (such as component 10 in these embodiments). The light is then detected by sensor 124 of device 108. For example, in some embodiments, the light is then reflected from the surface it contacts, and the light is received by sensor 124 of device 108. The round-trip time of the light to sensor 124 is used to determine measurements along the respective axes. These devices are generally referred to as time-of-flight devices. In other embodiments, sensor 124 detects the light on the surface it contacts and determines the measurements based on the relative position of the light in the field of view of sensor 124. These devices are generally referred to as triangulation devices. Then, as mentioned, X-axis, Y-axis, and Z-axis data points are calculated based on the detected light.

[0039] In some embodiments, the light emitted by laser 120 is emitted in a frequency band that is only wide enough to be reflected from a portion of the object to be measured. In these embodiments, a robotic arm (as discussed herein) or other suitable mechanism for moving laser 120 can be used to move laser 120 and the emitted radio frequency band as needed until the light has been reflected from the entire object to be measured.

[0040] In other embodiments, other suitable surface metrology devices may be utilized. For example, in some embodiments, device 108 may be an X-ray scanner that provides images in the form of X-rays. In some embodiments, device 108 may be or include a high-resolution crystal orientation system that provides images in the form of Laue diffraction patterns, such as Laue orientation images. In some embodiments, device 108 may be a scanning electron microscope that provides images in the form of electron channel contrast images. In some embodiments, device 108 may be a three-dimensional scanner that provides images in the form of three-dimensional scanning geometry.

[0041] In some embodiments, the data acquisition system 102 may include a robotic arm 130. The robotic arm 130 may support and facilitate movement of other components of the data acquisition system 102 relative to the components to obtain images of the components. For example, imaging device 106 and data acquisition device 108 (or components thereof, such as a light source) may be mounted to the robotic arm 130. In an exemplary embodiment, the mobile robotic arm 130 may position the data acquisition system 102 or its components (such as a light source) relative to component 10. In some embodiments, other components, such as imaging device 106, may remain stationary, while components such as a light source are movable. In an exemplary embodiment, the robotic arm 130 is a six-degree-of-freedom arm 130, which provides movement along and about axes 50, 52, 54.

[0042] In some embodiments, the user computing system 104 can operate the data acquisition system 102 to perform various disclosed steps. In other embodiments, the data acquisition system 102 may be a standalone device operated independently by a user and linked to the user computing system 104, or it may be a separate, discrete system. For example, in some embodiments, the user can manually acquire images and upload them to the user computing system 104.

[0043] Now for reference Figure 2 A flowchart illustrating a process for generating a predicted lifetime 522 for component 10 according to an embodiment of the present disclosure is provided. Process 500 can be implemented by the computational system 100 described above (and further described in detail below). Specifically, process 500 illustrates two separate paths for generating the predicted lifetime 522, e.g., a multi-step path 501 and an end-to-end path 503. Both paths 501 and 503 can be performed by the computational system 100, but the end-to-end path 503 can utilize a machine learning model 520 that simplifies the multi-step path 501 (e.g., in an end-to-end manner). Data utilized and / or generated by the multi-step path 501 (such as input 502 and output predicted lifetime 522) can be provided for teaching or training the machine learning model 520. In some embodiments, the machine learning model 520 can be trained independently, e.g., without reference to data generated via the multi-step path 501. Both paths 501 and 503 can be automatic, but the machine learning model can replace certain steps in the multi-step path 501. Machine learning model 520 can be one (or a combination thereof): a multi-input neural network (e.g., a multimodal neural network); a convolutional neural network (CNN) with tabular data fusion (which is an optional extension of CNN); a deep embedded clustering (DEC); an autoencoder; a residual network (ResNet); and / or a capsule network (CapsNet). Machine learning model 520 can be able to directly predict creep life based on a combination of tabular data (e.g., operating conditions), images, and / or stress maps.

[0044] Process 500 may include one or more inputs 502, including images 504, stress maps 506, and / or operational data 508. As shown, image 504 may be at least one of photographs 510A, 510B, video 512, laser scan images, X-ray scan images, Laue orientation images, electron channel contrast images, or three-dimensional scan geometry 514. Photographs 510A and 510B may be at least one of photograph 510A of the outer surface of the component or photograph 510B of the inner surface of the component. That is, in some embodiments, photograph 510B may be a photograph of a metallurgical sample made of the same or similar material as the component. In other embodiments, photograph 510B may be a cross-section of another component. For example, if the component analyzed by process 500 is a gas turbine airfoil (e.g., for predicting life), one of the inputs 502 may be an image of a cross-section (or metallurgical sample) of another gas turbine airfoil.

[0045] Process 500 may include dividing video 512 into one or more (such as multiple) frames at step 516. That is, video 512 may include multiple frames that can be separated and / or selected (e.g., by computing system 100) to generate multiple photographs. Similarly, at step 518, an image of the three-dimensional scan geometry 514 may be generated, for example, by computing system 100. It should be noted that such images may typically be provided by data acquisition system 102 and computing system 100, or may be provided independently of data acquisition system 102 and computing system 100. In many embodiments, process 500 may also include creating a single “mean” image at step 524. The mean image may be generated from multiple frames generated from video 512 in step 516 and / or from multiple images generated from three-dimensional scan 514. Creating the mean image at step 524 may include collecting images, initializing a summed image, summing pixel values, calculating the mean, converting the image format, and saving the mean image. Creating a mean image can include extracting the mean, median, minimum, maximum, weighted average, and any other aggregation of pixel intensities.

[0046] In many embodiments, process 500 may include representing a three-dimensional scan 514 of component 10 as a collection (or multiple slices) at step 526. Each of the multiple slices may be an image illustrating a different cross-sectional portion of the scanned component. Thus, when the multiple slices are stacked together, they collectively form the scanned component. The multiple slices may be generated by computing system 100, or may be generated separately (e.g., by a separate computing system) and provided to computing system 100 for analysis. In other embodiments, the three-dimensional scan may be a point cloud dataset generated using a lidar sensor. In other embodiments, the three-dimensional scan may be a wave-based detection dataset. In yet another embodiment, the three-dimensional scan may be an original CAD drawing.

[0047] Stress diagram 506 may be generated, for example, via finite element analysis (“FEA”) of component 10, which may be performed, for example, by computing system 100 or performed separately (e.g., via a separate computing system) and provided to computing system 100. Examples of suitable FEA software for such analysis include, for example, ANSYS, Simulia, Nastran, etc. Stress diagram 506 may indicate areas of high and low stress in component 10, which may be based at least in part on operating data associated with component 10 (such as operating hours, average temperature of component exposure, or other operating data).

[0048] Operational data 508 may be associated with component 10 and may include time data and temperature data. Time data may indicate the amount of time spent by component 10 during operation. For example, in embodiments where the component is a gas turbine component, time data may indicate the amount of time the gas turbine component spends inside the operating gas turbine. Temperature data may indicate multiple temperatures (or average temperatures) exposed to by the component during operation. For example, in embodiments where the component is a gas turbine component, temperature data may be the temperatures exposed to by the gas turbine component during operation, which may be multiple temperatures or average temperatures over a period of time. In many embodiments, temperature data may describe a temperature range, temperature increments over operating time, and / or a set of temperature increments within a given interval to identify rapid temperature changes.

[0049] In many embodiments, process 500 (e.g., multi-step path 501) may include identifying regions of interest in component 10 at step 527 based on a stress map 506 of component 10. For example, stress map 506 may indicate locations on component 10 that experience greater stress relative to other locations. These areas of increased stress may be more susceptible to material creep. In various specific embodiments, process 500 may also include cropping the region of interest in the image identified based on stress map 506 at step 528. For example, the region of interest illustrated in image 504 identified by stress map 506 may be cropped (e.g., by computing system 100). Cropping image 504 may include removing portions of image 504 that are unnecessary for further processing (e.g., areas of component 10 illustrated by image 504 that do not correspond to the areas of increased stress identified by stress map 506). In many specific embodiments, the cropped image may be provided to machine learning model 520. The cropped image may rely on the region indicated by stress map 506, which may be generated by a finite element analysis (FEA) system.

[0050] In various implementations, process 500 may also include applying computer vision (CV) and machine learning (ML) techniques at step 530 to enhance, modify, and / or transform image 504. For example, Figure 3 and Figure 4 Examples of how to use ML and CV (e.g., by computing system 100) to enhance or modify images are illustrated. Figure 3 In this process, the blurred image 600 can be processed by the computing system 100 using computer vision (CV) and / or machine learning (ML) to generate an enhanced image 602. Specifically, the computing system 100 may include a machine learning model configured to receive the blurred image 600 and generate the enhanced image 602. That is, the computing system can utilize a machine learning model and / or computer vision to process the blurred image 600 to generate the enhanced image 602. The enhanced image 602 may be sharper than the blurred image 600 and have more defined lines (e.g., have a higher pixel count). Figure 4 In this process, the grayscale image 700 can be processed by the computing system 100 using CV and / or ML to generate a black-and-white image 702 representing grain boundaries. (Return to Reference) Figure 2 In many specific implementations, the enhanced image 602 and / or the black and white image 702 can be provided to the machine learning model 520.

[0051] Examples of ML models that can be used by the computing system 100 to enhance or modify images may include, but are not limited to: Convolutional Neural Networks (CNNs); Deep Image Priors (DIPs); Autoencoders and Denoising Autoencoders (DAEs); Generative Adversarial Networks (GANs); Image Super-Resolution Models; Recurrent Neural Networks (RNNs); and Long Short-Term Memory (LSTMs). Examples of CV techniques that can be used by the computing system 100 to enhance or modify images may include, but are not limited to: Histogram Equalization and Adaptive Histogram Equalization; Image Binarization; Smoothing methods, such as Gaussian Smoothing; Unsharpening Masking; Sharpening Filters; Median Filtering; and Blob Analysis and Removal. Examples of suitable computer vision software for such analyses include, for example, the OpenCV Python library. The OpenCV Python library includes a wide range of computer vision techniques, including image processing, video analysis, object detection, feature detection, and matching.

[0052] Histogram equalization can be used to redistribute pixel intensity values ​​so that pixels span the entire range of possible values, thereby enhancing the visibility of features in an image. Image binarization can be used to convert grayscale or color images into binary images, which may consist of only black and white. Gaussian smoothing may involve convolving an image with a Gaussian function to produce a weighted average of the surrounding pixel values, resulting in a smoother image. Blob analysis and removal can be used to remove "noise" or "blobs" from an image, making it clearer.

[0053] Return to reference Figure 2 In many embodiments, process 500 may further include detecting the contours of image 504 at step 532 and filtering these contours at step 534 to represent the true grains of component 10. Furthermore, in step 536, process 500 may include overlaying (or superimposing) the grains onto stress map 506. In step 538, process 500 may further include extracting grain characteristics (e.g., feature extraction). Steps 532 through 538 are discussed in more detail below. Based at least in part on the relationship between the grains and stress map identified in step 536, process 500 may include determining a predicted lifetime 522 for component 10. In many embodiments, process 500 may include determining the predicted lifetime 522 of the component based at least in part on temperature data and exposure time data (e.g., time exposed to peak temperature).

[0054] In step 538, feature extraction may include extracting features (e.g., counts, averages, medians, maximums, minimums, etc.) associated with individual grains or rapid aggregates of grains of component 10. These features may include, for example, the following characteristics of the grains, which can be extracted from an image of component 10: relative orientation of stress at a given point relative to grain boundaries; number of grains; area; perimeter; centroid; bounding box; aspect ratio; range; density; equivalent diameter; orientation; major axis length; minor axis length; eccentricity; convex hull; convexity defects; torque; and profile length.

[0055] In an exemplary embodiment, process 500 (e.g., end-to-end path 503) may include a machine learning model 520, which may be stored in computing system 100 and may be used by computing system 100 to generate a predicted lifetime 522 for component 10. That is, computing system 100 may use machine learning model 520 to process input 502 to generate a predicted lifetime 522 for component 10 as the output of machine learning model 520.

[0056] It should be understood that the predicted lifetime 522 generated using the multi-step path 501 can be used to train the machine learning model 520. Thus, the predicted lifetime 522 generated using the multi-step path 501 can be a multi-step predicted lifetime 522 used to train the machine learning model 520. Alternatively, the machine learning model 520 can be trained based on manually labeled samples (e.g., images of manually labeled parts and / or manually labeled images of parts).

[0057] The predicted lifetime 522 can be at least one of a binary output 540, an integer output 542, and / or a probability density function 544. The binary output 542 can be a single output indicating whether component 10 can survive another running cycle (e.g., yes or no). The integer output 542 can be the number of running cycles that component 10 can survive (e.g., another cycle, two more cycles, three more cycles, etc.). The probability density function 544 can indicate the number of hours component 10 can survive during operation. More specifically, the probability density function 544 can describe the likelihood of a continuous random variable taking a specific value. Specifically, the cumulative distribution function (CDF) output can be used as the predicted lifetime 522, describing the probability that a random variable (X) will take a value less than or equal to a specific value (x). For example, the CDF as the predicted lifetime 522 can indicate the probability of reaching X hours (e.g., expressed as a percentage).

[0058] Machine learning models suitable for generating predicted lifetimes of 522 may include, but are not limited to: logistic regression; parametric survival models such as general log-linear models (GLL); decision trees; random forests; gradient boosting machines (GBM); XGBoost (extreme gradient boosting); LightGBM (lightweight gradient boosting); CatBoost; support vector machines (SVM); K-nearest neighbors (KNN); Naive Bayes; generalized linear models and / or artificial neural networks (ANN).

[0059] In many implementations, process 500 may include an optional step 546 whereby a subject matter expert (SME) conducts an evaluation. During step 546, the SME may evaluate the predicted lifetime, component, and grain-stress diagram overlay to ensure accurate predictions. The SME may also rely on data to make decisions independently, without any automated prediction tools.

[0060] Figure 5 An exemplary embodiment of a plurality of images 504 is illustrated, wherein the plurality of images are in the form of a video comprising a plurality of video frames. Figure 6 Multiple images 504 are illustrated, which may be from... Figure 5 Multiple video frames, multiple photographs, or multiple images obtained from another suitable embodiment of the data acquisition system 102 discussed above. It should be noted that such images may typically be provided by the data acquisition system 102 and the computing system 100, or typically independently of the data acquisition system 102 and the computing system 100.

[0061] The computing system 100 (such as the data acquisition system 102 and / or its user computing system 104) may be able to process one or more images 504 of the component 10. Such processing may detect one or more grain structures on the component 10.

[0062] The grain structure according to this disclosure may be or include one or more of the following: grain, grain boundary, grain or grain boundary size (e.g., length, etc.), grain shape factor, aggregate grain size (such as minimum, maximum, median, etc.), grain or grain boundary orientation, and / or grain boundary triple point.

[0063] Computer vision can be used to perform cropping of image 504 and / or pixel analysis of image 504. For example, in some embodiments, such processing may include cropping one or more images 504. Cropping allows focusing on specific areas of interest on part 10, such as portions of the part that are particularly susceptible to high temperatures and high stresses. Computer vision and cropping can also be used to remove unwanted background that would otherwise consume processing power. Figure 7 Image 504 and its cropped portion 202 are shown as an example.

[0064] Additionally or alternatively, in some embodiments, such processing may include performing computer vision techniques and / or pixel analysis of image 504. This analysis is typically an analysis that distinguishes a reference object (e.g., grain structure) from a background (e.g., part surface and background) based on differences in color depth (i.e., differences in color or grayscale). The analysis may be performed on each individual pixel or group of pixels defining image 504. For pixel analysis, the number of bits per pixel of the image (i.e., 128, 256, etc.) may be divided into two or more groups (e.g., groups including lighter color depths and groups including darker color depths). Each group is classified as either a reference object portion or a background portion. For example, color depth analysis may classify pixels or groups of pixels with darker or lighter color depths as representing a reference object (i.e., a surface feature relative to a part, or a part relative to the background), and may classify other groups of pixels or groups of pixels with darker or lighter color depths as representing the background (i.e., a part relative to a surface feature, or a background relative to a part). It is worth noting that the different divisions within the shallower and darker groups can be used to distinguish surface features from parts, as well as parts from the background.

[0065] Figure 8 An example of a pixel analysis implementation is illustrated, in which image contrast is gradually increased through pixel analysis to detect grain structure.

[0066] Figure 9 Multiple detected grain structures 210 on component 10 are illustrated. Notably, in the exemplary embodiment, the detected grain structures can be detected with reference to axes 50, 52, 54. Understanding the location (e.g., orientation, position, and / or shape) of the detected grain structures is crucial for creep and lifetime prediction according to this disclosure.

[0067] Figure 11 A block diagram depicts an exemplary computing system 100 performing one or more of the various functions and steps described herein according to embodiments of this disclosure. The computing system 100 may include multiple computing systems communicating via a network 1680. The computing system 100 may include one or more user computing systems 104 and one or more data acquisition systems 102.

[0068] User computing system 104 may include one or more computing devices. Computing devices may include mobile computing devices (e.g., smartphones or tablets), laptop computing devices, desktop computing devices, wearable computing devices (e.g., smartwatches, smart jackets, smart glasses, smart backpacks, etc.), smart appliances (e.g., smart thermostats, smart refrigerators, smart washing machines, smart dryers, etc.), embedded computing devices, monitoring computing devices (e.g., drones), or any other type of computing device.

[0069] User computing system 104 may include one or more processors 1612 that can be used to perform one or more operations. The one or more processors 1612 may include any suitable processing device (e.g., processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and may be one or more processors operatively connected. The one or more processors 1612 may perform operations serially and / or in parallel. The one or more processors 1612 may be dedicated to a particular computing device and / or may be used by multiple devices to perform processing tasks.

[0070] User computing system 104 may include memory 1614 for storing data 1616 and / or instructions 1618. Memory 1614 may include one or more non-transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, disks, etc., and combinations thereof. Data 1616 may include user data, application data, operating system data, etc. Data 1616 may include text data, image data, audio data, statistical data, latently encoded data, etc. Instructions 1618 may include instructions that, when executed by one or more processors 1612, cause user computing device 104 to perform operations.

[0071] In some embodiments, the user computing system 104 may store and utilize one or more machine learning models 1620 (such as the machine learning model 520 described above or another model). The one or more machine learning models 1620 may include computer vision models, such as object detection models. The one or more machine learning models 1620 may also include detection models, natural language processing models, segmentation models, classification models, augmentation models, generative models, discriminative models, and / or one or more other model types. In some embodiments, the one or more machine learning models 1620 may include various machine learning models, such as neural networks (e.g., deep neural networks) or other types of machine learning models, including nonlinear and / or linear models. Neural networks may include convolutional neural networks, feedforward neural networks, recurrent neural networks (e.g., long short-term memory recurrent neural networks), or other forms of neural networks. The one or more machine learning models 1620 may include one or more transformer models.

[0072] User computing system 104 may include one or more user input components 1622, one or more user interfaces 1624, and / or one or more user sensors 1626. One or more user input components 1622 may be configured to receive user input and / or environmental input. For example, one or more user input components 1622 may include touch-sensitive components (e.g., touch-sensitive displays or touchpads) that are sensitive to user input objects (e.g., fingers or styluses). Touch-sensitive components can be used to implement a virtual keyboard. Other exemplary user input components include a computer mouse, remote control, controller, microphone, conventional keyboard, or other devices through which a user can provide user input. In some implementations, one or more user input components 1622 may include one or more gesture processing engines to determine touch gestures, audio gestures, and / or body postures. One or more user interfaces 1624 may be configured to acquire and / or display data. One or more user interfaces 1624 may be associated with an operating system, one or more applications, one or more network platforms, and / or one or more devices. One or more sensors 1626 may include one or more image sensors, one or more infrared sensors, one or more light detection and ranging (lidar) sensors, one or more audio sensors, one or more touch sensors, one or more acoustic navigation and ranging (sonar) sensors, and / or one or more thermal sensors.

[0073] For example, as discussed, Figure 10(A) illustrates an exemplary grain structure 210 within and generally perpendicular to the stress direction 222. Figure 10(C) also illustrates an exemplary grain structure 210 within and generally perpendicular to the stress direction 222. Such a generally perpendicular grain structure 210 will increase the probability of creep, thereby shortening the predicted lifespan of component 10. Figure 10(B) illustrates an exemplary grain structure 210 within and generally parallel to the stress direction 222. Figure 10(D) also illustrates an exemplary grain structure 210 within and generally parallel to the stress direction 222. Such a generally parallel grain structure 210 will reduce, not increase, or increase the probability of creep relatively less than, for example, in the perpendicular case, thereby increasing or not shortening the predicted lifespan of component 10.

[0074] It should be understood that this disclosure is not limited to comparisons with stress directions only along axes 50, 52, and / or 54, or to predicted lifetime determinations based solely on generally parallel or generally perpendicular grain structures. Rather, such comparisons and determinations are exemplary embodiments, and this disclosure covers stress directions in any suitable orientation and predicted lifetime determinations based on grain structures having any suitable orientation relative to the stress directions. The inventors have discovered the ability to detect macroscopic structures on real three-dimensional parts using nondestructive characterization techniques; and the ability to correlate the extracted grain structures with stress maps to perform computer-based automated reasoning for the creep lifetime of a particular part. This disclosure advantageously facilitates more accurate and consistent lifetime predictions and creep probability determinations on this basis.

[0075] In some implementations, additional variables may be used to determine the predicted lifespan of component 10. For example, such additional variables may be used to determine the creep probability. Examples of such variables may include, for instance, the material type of the component, the service time of component 10, and / or the service temperature of component 10.

[0076] The predicted lifetime can be adjusted using the value or magnitude of grain structure or one or more such additional variables. For example, relatively large grain boundaries (relative to, for example, a predetermined threshold for component 10) can increase the predicted lifetime, while relatively small grain structures (relative to, for example, a threshold) can decrease the predicted lifetime. Similarly, relatively short usage time or relatively low operating temperature (relative to, for example, a predetermined threshold for component 10) can increase the predicted lifetime, while relatively long usage time or relatively high operating temperature (relative to, for example, a threshold) can decrease the predicted lifetime. This increase and / or decrease can be made relative to a baseline predicted lifetime of the component, which can be predetermined or determined using the calculation system 100.

[0077] The values ​​or sizes of such additional variables and thresholds can be determined by the computing system 100, or they can be determined independently and provided to the computing system 100 for determining steps.

[0078] The determined predicted lifespan can advantageously be output from the computing system 100 so that the user of the computing system can receive the determined predicted lifespan. The predicted lifespan determination can advantageously provide a relatively accurate lifespan estimate, thus allowing the user to extend the actual service life of component 10 within permissible limits based on the determined predicted lifespan, while reducing concerns about the risk of failure. The predicted lifespan determination can further advantageously allow the user to shorten the actual service life of component 10 based on the determined predicted lifespan, and remove component 10 from service before the risk of failure, thereby reducing the problem of unexpected power outages.

[0079] Figure 11A block diagram depicts an exemplary computing system 100 performing one or more of the various functions and steps described herein according to embodiments of this disclosure. The computing system 100 may include multiple computing systems communicating via a network 1680. The computing system 100 may include one or more user computing systems 104 and one or more data acquisition systems 102.

[0080] User computing system 104 may include one or more computing devices. Computing devices may include mobile computing devices (e.g., smartphones or tablets), laptop computing devices, desktop computing devices, wearable computing devices (e.g., smartwatches, smart jackets, smart glasses, smart backpacks, etc.), smart appliances (e.g., smart thermostats, smart refrigerators, smart washing machines, smart dryers, etc.), embedded computing devices, monitoring computing devices (e.g., drones), or any other type of computing device.

[0081] User computing system 104 may include one or more processors 1612 that can be used to perform one or more operations. The one or more processors 1612 may include any suitable processing device (e.g., processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and may be one or more processors operatively connected. The one or more processors 1612 may perform operations serially and / or in parallel. The one or more processors 1612 may be dedicated to a particular computing device and / or may be used by multiple devices to perform processing tasks.

[0082] User computing system 104 may include memory 1614 for storing data 1616 and / or instructions 1618. Memory 1614 may include one or more non-transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, disks, etc., and combinations thereof. Data 1616 may include user data, application data, operating system data, etc. Data 1616 may include text data, image data, audio data, statistical data, latently encoded data, etc. Instructions 1618 may include instructions that, when executed by one or more processors 1612, cause user computing device 104 to perform operations.

[0083] In some embodiments, the user computing system 104 may store and utilize one or more machine learning models 1620 (such as the machine learning model 520 described above or another model). The one or more machine learning models 1620 may include computer vision models, such as object detection models. The one or more machine learning models 1620 may also include detection models, natural language processing models, segmentation models, classification models, augmentation models, generative models, discriminative models, and / or one or more other model types. In some embodiments, the one or more machine learning models 1620 may include various machine learning models, such as neural networks (e.g., deep neural networks) or other types of machine learning models, including nonlinear and / or linear models. Neural networks may include convolutional neural networks, feedforward neural networks, recurrent neural networks (e.g., long short-term memory recurrent neural networks), or other forms of neural networks. The one or more machine learning models 1620 may include one or more transformer models.

[0084] User computing system 104 may include one or more user input components 1622, one or more user interfaces 1624, and / or one or more user sensors 1626. One or more user input components 1622 may be configured to receive user input and / or environmental input. For example, one or more user input components 1622 may include touch-sensitive components (e.g., touch-sensitive displays or touchpads) that are sensitive to user input objects (e.g., fingers or styluses). Touch-sensitive components can be used to implement a virtual keyboard. Other exemplary user input components include a computer mouse, remote control, controller, microphone, conventional keyboard, or other devices through which a user can provide user input. In some implementations, one or more user input components 1622 may include one or more gesture processing engines to determine touch gestures, audio gestures, and / or body postures. One or more user interfaces 1624 may be configured to acquire and / or display data. One or more user interfaces 1624 may be associated with an operating system, one or more applications, one or more network platforms, and / or one or more devices. One or more sensors 1626 may include one or more image sensors, one or more infrared sensors, one or more light detection and ranging (lidar) sensors, one or more audio sensors, one or more touch sensors, one or more acoustic navigation and ranging (sonar) sensors, and / or one or more thermal sensors.

[0085] User computing system 104 can be communicatively connected to data acquisition system 102 via network 1680, which may include the Internet (e.g., Ethernet and / or WiFi), Bluetooth and / or direct cabling).

[0086] The data acquisition system 102 may include one or more computing devices. The computing devices may include mobile computing devices (e.g., smartphones or tablets), laptop computing devices, desktop computing devices, wearable computing devices (e.g., smartwatches, smart jackets, smart glasses, smart backpacks, etc.), smart appliances (e.g., smart thermostats, smart refrigerators, smart washing machines, smart dryers, etc.), embedded computing devices, monitoring computing devices (e.g., drones), or any other type of computing device.

[0087] The data acquisition system 102 may include one or more processors 1712 that can be used to perform one or more operations. The one or more processors 1712 may include any suitable processing device (e.g., processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and may be one or more processors operatively connected. The one or more processors 1712 may perform operations serially and / or in parallel. The one or more processors 1712 may be dedicated to a particular computing device and / or may be used by multiple devices to perform processing tasks.

[0088] The data acquisition system 102 may include a memory 1714 capable of storing data 1716 and / or instructions 1718. The memory 1714 may include one or more non-transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, disks, etc., and combinations thereof. The data 1716 may include user data, application data, operating system data, etc. The data 1716 may include text data, image data, audio data, statistical data, latently encoded data, etc. The instructions 1718 may include instructions that, when executed by one or more processors 1712, cause the user computing device 104 to perform operations.

[0089] In some implementations, the data acquisition system 102 may store and utilize one or more machine learning models 1720. The one or more machine learning models 1720 may include computer vision models, such as object detection models. The one or more machine learning models 1720 may also include detection models, natural language processing models, segmentation models, classification models, augmentation models, generative models, discriminative models, and / or one or more other model types. In some implementations, the one or more machine learning models 1720 may include various machine learning models, such as neural networks (e.g., deep neural networks) or other types of machine learning models, including nonlinear and / or linear models. Neural networks may include convolutional neural networks, feedforward neural networks, recurrent neural networks (e.g., long short-term memory recurrent neural networks), or other forms of neural networks. The one or more machine learning models 1620 may include one or more transformer models.

[0090] The data acquisition system 102 may include one or more user input components 1722, one or more user interfaces 1724, and / or one or more user sensors 1726. The one or more user input components 1722 may be configured to receive user input and / or environmental input. For example, the one or more user input components 1722 may include touch-sensitive components (e.g., touch-sensitive displays or touchpads) that are sensitive to user input objects (e.g., fingers or styluses). Touch-sensitive components can be used to implement a virtual keyboard. Other exemplary user input components include a computer mouse, remote control, controller, microphone, conventional keyboard, or other devices through which a user can provide user input. In some implementations, the one or more user input components 1722 may include one or more gesture processing engines to determine touch gestures, audio gestures, and / or body postures. The one or more user interfaces 1724 may be configured to acquire and / or display data. The one or more user interfaces 1724 may be associated with an operating system, one or more applications, one or more network platforms, and / or one or more devices. One or more sensors 1726 may include one or more image sensors, one or more infrared sensors, one or more light detection and ranging (lidar) sensors, one or more audio sensors, one or more touch sensors, one or more acoustic navigation and ranging (sonar) sensors, and / or one or more thermal sensors.

[0091] The data acquisition system 102 can be communicatively connected to the user computing system 104 via a network 1680, which may include the Internet (e.g., Ethernet and / or WiFi), Bluetooth and / or direct cabling).

[0092] User computing system 104 and / or data acquisition system 104 can train models 1620 and / or 1720 via interaction with training computing system 150, which is communicatively coupled to network 1680. Training computing system 150 may be separate from data acquisition system 102, or may be part of data acquisition system 102.

[0093] The training computing system 150 includes one or more processors 152 and memory 154. The one or more processors 152 can be any suitable processing device (e.g., processor core, graphics processing unit (GPU), tensor processing unit (TPU), microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and can be one or more processors operatively connected. The memory 154 can include one or more non-transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, flash memory devices, disks, etc., and combinations thereof. The memory 154 can store data 156 and instructions 158, which are executed by the processor 152 to cause the training computing system 150 to perform operations. In some specific implementations, the training computing system 150 includes one or more server computing devices, or is otherwise implemented by one or more server computing devices.

[0094] The training computation system 150 may include a model trainer 160 that uses various training or learning techniques, such as error backpropagation, to train machine learning models 1620 and / or 1720 stored at the user computation system 104 and / or data acquisition system 102. For example, a loss function can be used to update one or more parameters of the model through backpropagation (e.g., gradients based on the loss function). Various loss functions can be used, such as mean squared error, likelihood loss, cross-entropy loss, hinge loss, and / or various other loss functions. Gradient descent techniques can be used to iteratively update parameters over multiple training iterations.

[0095] In some implementations, performing backpropagation of errors may include performing truncated backpropagation over time. The model trainer 160 may perform various generalization techniques (e.g., weight decay, dropout, etc.) to improve the generalization ability of the model being trained.

[0096] Specifically, model trainer 160 can train machine learning models 1620 and / or 1720 based on a set of training data (or training dataset) 162. Training data 162 can include training images (e.g., training image data) of multiple training components. For example, if the component is a gas turbine component (such as an airfoil), the training component can also be a gas turbine component (e.g., an airfoil), and machine learning models 1620 and / or 1720 can be trained based on training images of multiple training components. Training images can include pre-operation images and post-operation images of the training components. For example, a pre-operation image can be an image of the training component before it undergoes an operating cycle. A post-operation image can be an image of the training component after it has undergone one or more operating cycles.

[0097] In an exemplary implementation, machine learning models 1620 and 1720 are trained based on training data 162, which includes training images of multiple training components, training stress maps of multiple training components, and multi-step predicted lifetimes generated by the multi-step path 501 described above. For example, training images and training stress maps (and training operation data) are collected or obtained from multiple training components (which may be components of the same type as component 10). The training images, training stress maps, and / or training operation data can be found in the above-mentioned references. Figure 2 The multi-step path 501 is executed to generate multi-step predicted lifetimes for the training components. In this way, machine learning models 1620 and 1720 (such as machine learning model 520 described above) can be provided with the same training input data as the training input data provided to the multi-step path 501, and generate predicted lifetimes for the components. These predicted lifetimes can then be compared with the multi-step predicted lifetimes to determine the accuracy of the machine learning models 1620 and 1720.

[0098] In other words, return to reference Figure 2 During the training of machine learning model 520, training images, training stress maps, and / or training operation data can be provided to both multi-step path 501 and machine learning model 520 via an end-to-end path. Multi-step path 501 can output a multi-step predicted lifetime, and the end-to-end path (e.g., via machine learning model 520) can output a predicted lifetime, which can be compared to the multi-step predicted lifetime of multi-step path 501. This process can be repeated multiple times to train machine learning model 520 and / or increase the accuracy of machine learning model 520.

[0099] Model trainer 160 includes computer logic for providing the desired functionality. Model trainer 160 may be implemented in hardware, firmware, and / or software that controls a general-purpose processor. For example, in some embodiments, model trainer 160 includes a program file stored on a storage device, loaded into memory, and executed by one or more processors. In other embodiments, model trainer 160 includes one or more sets of computer-executable instructions stored in a tangible computer-readable storage medium, such as RAM, hard disk, or optical or magnetic media.

[0100] Network 1680 may include any type of communication network (e.g., a local area network (e.g., an intranet), a wide area network (e.g., the Internet), or some combination thereof) and may include any number of wired or wireless links. Communication over Network 1680 may be performed using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), and / or protection schemes (e.g., VPN, Secure HTTP, SSL), via any type of wired and / or wireless connection.

[0101] The system and method disclosed herein can be implemented using computing system 100. The system and method disclosed herein can also be implemented using other computing systems, including other system configurations.

[0102] See now Figure 12 A flowchart illustrating a method 1100 for determining the predicted lifespan of component 10 is provided according to an embodiment of this subject. One or more steps of such a method can be performed by, for example, a computing system 100 discussed herein. Generally, this document will refer to the foregoing references. Figures 1 to 1 Method 1100 is described using the computing system 100, component 10, and process 500 described in section 0. However, those skilled in the art will understand that the disclosed method 1100 can generally be used with any suitable system configuration. Furthermore, although... Figure 12 The steps are depicted in a particular order for illustrative and discussion purposes, but unless otherwise specified in the claims, the methods discussed herein are not limited to any particular order or arrangement. Those skilled in the art will understand using the disclosure provided herein that the steps of the methods disclosed herein may be omitted, rearranged, combined, and / or modified in various ways without departing from the scope of this disclosure. Dashed boxes indicate optional steps of method 1100.

[0103] In some implementations, the method may include using a component during one or more operating cycles of the machine. For example, in an exemplary embodiment, the component may be a gas turbine component. One or more operating cycles may include operating the gas turbine for a set period of time (e.g., between about 10,000 hours and about 35,000 hours) with the gas turbine component installed. The method may include removing the gas turbine component from the gas turbine after the operating cycle.

[0104] For example, method 1100 may include, at (1102), preparing a component, such as for imaging the component. Such preparation may include, for example, surface preparation of the component (e.g., acid etching, electrochemical treatment or other chemical treatment or mechanical polishing, etc.) and / or other suitable processes that help improve the visibility of the grain structure of the component surface.

[0105] In many specific implementations, method 1100 may include, at (1104), obtaining an image of a component by a computing system including one or more computing devices. This image may be derived from the above reference. Figure 1 The data is acquired by the data acquisition system 102. The image may be at least one of photographs, videos, laser scan images, X-ray scan images, Laue orientation images, electron channel contrast images, or three-dimensional scan geometry.

[0106] In some specific implementations, method 1100 may include, at (1106), obtaining a stress diagram of the component by a computational system. The stress diagram may be generated, for example, via a finite element analysis (“FEA”) of the component, which may be performed, for example, by the computational system or performed separately (e.g., via a separate computational system) and provided to the computational system. Examples of suitable FEA software for such analysis include, for example, ANSYS, Simulia, Nastran, etc. The stress diagram may indicate areas of high and low stress in the component, which may be based at least in part on operational data associated with the component (such as operating hours, average temperature of component exposure, or other operational data).

[0107] In various embodiments, method 1100 may include, at (1108), obtaining operational data associated with a component. This operational data may be associated with the component and may include time data and temperature data. The time data may indicate the amount of time the component spends in operation. For example, in an embodiment where the component is a gas turbine component, the time data may indicate the amount of time the gas turbine component spends inside a running gas turbine. The temperature data may indicate multiple temperatures (or average temperatures) exposed to the component during operation. For example, in an embodiment where the component is a gas turbine component, the temperature data may be the temperatures exposed to by the gas turbine component during operation, which may be multiple temperatures or average temperatures over a period of time. Steps 1104, 1106, and 1108 may be performed in parallel (e.g., simultaneously) and / or in an alternative order different from the illustrated order.

[0108] In an exemplary implementation, method 1100 may include, at (1110), processing an image and stress map of a component using a machine learning model by a computing system to generate a predicted lifetime of the component as the output of the machine learning model. In some implementations, the machine learning model may utilize operating parameters (such as operating time data and temperature data) to generate the predicted lifetime. In some implementations, the method may include processing an image, stress map, and operating data using a machine learning model by a computing system to generate a predicted lifetime of the component. The predicted lifetime may be at least one of a binary output, an integer output, and / or a probability density function. The binary output may be a single output indicating whether the component can survive another operating cycle (e.g., yes or no). The integer output may be the number of operating cycles the component can survive (e.g., another cycle, two more cycles, three more cycles, etc.). The probability density function may indicate the number of hours the component can survive during operation. The probability density function may describe the probability that the component will survive within a specific number of operating hours. More specifically, the probability density function may describe the probability that a continuous random variable takes a specific value. Specifically, the cumulative distribution function (CDF) output can be used as a predictor of lifetime, which describes the probability that a random variable (X) will take a value less than or equal to a specific value (x). For example, the CDF as a predictor of lifetime can indicate the probability of reaching X hours (e.g., expressed as a percentage).

[0109] In many embodiments, the machine learning model (e.g., machine learning model 520) may be a first machine learning model, and the system may also include a second machine learning model. The second machine learning model may be configured to modify, enhance, or transform an image (e.g., before being used in the first machine learning model to process and generate a predicted lifetime). The computational system may utilize the second machine learning model to process the image to generate an enhanced image. In such specific embodiments, the method may further include the computational system using the first machine learning model to process the enhanced image and stress map of the part to generate a predicted lifetime of the part as the output of the first machine learning model.

[0110] In many implementations, the method may include, for example, training a machine learning model using a model trainer that can train the model based on a set of training data (or training dataset). The training data may include training images of multiple training components. For example, if the component is a gas turbine component (such as an airfoil), the training component may also be a gas turbine component (e.g., an airfoil), and the machine learning model may be trained based on training images of multiple training components. The training images may include pre-operation images and post-operation images of the training component. For example, a pre-operation image may be an image of the training component before it undergoes an operating cycle. A post-operation image may be an image of the training component after it has undergone one or more operating cycles.

[0111] In an exemplary implementation, the machine learning model is trained based on training data, which includes training images of multiple training components, training stress maps of multiple training components, and multi-step predicted lifetimes. For example, training images and training stress maps (and training operation data) are collected or obtained from multiple training components (which may be components of the same type as component 10). The training images, training stress maps, and / or training operation data can be referenced above. Figure 2 The multi-step path 501 is executed to generate multi-step predicted lifetimes for training components. In this way, a machine learning model can be provided with the same training input data as the training input data provided to the multi-step path 501, and a predicted lifetime for the component can be generated. This predicted lifetime can then be compared with the multi-step predicted lifetime to determine the accuracy of the machine learning model.

[0112] In many implementations, training the machine learning model may also include adjusting or modifying the weights or parameters of the machine learning model based on a comparison between the predicted lifetime of a component and the predicted lifetime of a multi-step process. This process may be repeated until the error (or difference) between the output generated from the machine learning model and the multi-step output is minimized or completely eliminated.

[0113] For example, to generate a multi-step predicted lifetime for each of a plurality of training components, the method may further include processing training images of the training components to detect grain structures on the training components; comparing the detected grain structures with corresponding training stress maps of the training components; and determining the multi-step predicted lifetime of the training components based on the location of the detected grain structures and the training stress maps. In such embodiments, processing the training images may include performing pixel analysis of the training images using computer vision. Additionally, in such embodiments, the comparison step may include overlaying the training stress map onto the processed training image to compare the location of the detected grain structures with the stress direction of the training stress map. Furthermore, in such embodiments, the determination step may include determining a creep probability based on the location of the detected grain structures and the stress direction of the training stress map. Based on the creep probability, the method may include determining the multi-step predicted lifetime of the training components.

[0114] In an exemplary embodiment, the determining step is further based on one of the following: the material type of the component, the usage time of the training component, and / or the operating temperature of the training component, as discussed herein. The training component may be a component of the same type as component 10. For example, both the training component and component 10 may be gas turbine components, such as turbine components or compressor components. A turbine component may be a turbine airfoil, turbine shroud, turbine power nozzle, or other turbine component. A compressor component may be a compressor rotor blade, compressor stator blade, or other compressor component.

[0115] This written description uses examples to disclose the invention, including the best mode, and also enables any person skilled in the art to practice the invention, including making and using any device or system and performing any combination of methods. The patentable scope of the invention is defined by the claims and may include other examples that would occur to a person skilled in the art. Such other examples are intended to be within the scope of the claims if they include structural elements that are not different from the literal language of the claims, or if they include equivalent structural elements that are not substantially different from the literal language of the claims.

[0116] Other aspects of the invention are provided by the subject matter of the following provisions:

[0117] A computer-implemented method for determining the predicted lifetime of a component, the method comprising: obtaining an image of the component by a computing system including one or more computing devices; obtaining a stress map of the component by the computing system, the stress map indicating regions of high stress and low stress in the component; and processing the image and the stress map of the component by the computing system using a machine learning model to generate a predicted lifetime of the component as the output of the machine learning model.

[0118] The method according to any of the foregoing clauses, wherein the component has undergone at least one operating cycle.

[0119] The component described in any of the foregoing clauses is a new component.

[0120] The method according to any of the foregoing provisions, wherein the predicted lifetime is at least one of a binary output of whether the component can survive another operating cycle, an integer output of the number of operating cycles in which the component can survive, or a probability density function.

[0121] The method according to any of the foregoing clauses, wherein the image is one of a photograph, a video frame, a laser scan image, an X-ray scan image, a Laue orientation image, an electron channel contrast image, or a three-dimensional scan geometry.

[0122] The method according to any of the foregoing provisions further includes: obtaining operational data associated with the component by the computing system, the operational data including time data and temperature data; and processing the image, the stress map, and the operational data by the computing system using the machine learning model to generate a predicted lifespan of the component as the output of the machine learning model.

[0123] The method according to any of the foregoing provisions, wherein the machine learning model is trained on a training dataset comprising training images of multiple training components.

[0124] The method according to any of the foregoing provisions, wherein the machine learning model is a first machine learning model, and wherein the method further comprises: processing the image using a second machine learning model to generate an enhanced image; and having the computing system process the enhanced image and the stress map of the component using the first machine learning model to generate the predicted lifetime of the component as the output of the machine learning model.

[0125] According to any of the preceding clauses, the machine learning model is trained based on a training dataset, the training dataset including training images of a plurality of training components, training stress maps of the plurality of training components, and multi-step predicted lifetimes of the plurality of training components.

[0126] According to the method described in any of the foregoing clauses, the multi-step predicted lifetime of each of the plurality of training components is generated by: processing the training image of the training component to detect grain structure on the training component; comparing the detected grain structure with the corresponding training stress map of the training component; and determining the multi-step predicted lifetime of the training component based on the location of the detected grain structure and the training stress map.

[0127] The method according to any of the foregoing provisions, wherein processing the training image includes performing pixel analysis of the training image by utilizing computer vision.

[0128] According to any of the foregoing provisions, the comparison step includes overlaying the training stress map onto a processed training image to compare the location of the detected grain structure with the stress direction of the training stress map.

[0129] According to any of the foregoing provisions of the method, the determining step includes determining a creep probability based on the location of the detected grain structure and the stress direction of the training stress map, and determining the multi-step predicted lifetime of the training component based on the creep probability.

[0130] According to any of the preceding clauses, the machine learning model is trained based on a training dataset, the training dataset including training images of a plurality of training components, training stress maps of the plurality of training components, and multi-step predicted lifetimes of the plurality of training components, wherein the multi-step predicted lifetime of each of the plurality of training components is generated by: processing the training images of the training components to detect grain structures on the training components; comparing the detected grain structures with the corresponding training stress maps of the training components; and determining the multi-step predicted lifetime of the training components based on the location of the detected grain structures and the training stress maps.

[0131] A computational system for determining the predicted lifetime of a component, the system comprising: one or more processors; and one or more non-transitory computer-readable media that co-store instructions, which, when executed by the one or more processors, cause the computational system to perform operations including: obtaining an image of the component; obtaining a stress map of the component, the stress map indicating regions of high and low stress in the component; and processing the image and the stress map of the component using a machine learning model to generate a predicted lifetime of the component as the output of the machine learning model.

[0132] According to any of the foregoing provisions, the predicted lifetime is at least one of the following: a binary output of whether the component can survive another operating cycle, an integer output of the number of operating cycles in which the component can survive, or a probability density function.

[0133] According to any of the foregoing provisions, the image is one of a photograph, a video frame, a laser scan image, an X-ray scan image, a Laue orientation image, an electron channel contrast image, or a three-dimensional scan geometry.

[0134] According to any of the foregoing provisions, the operation further includes: obtaining operational data associated with the component by the computing system, the operational data including time data and temperature data; and processing the image, the stress map, and the operational data by the computing system using the machine learning model to generate a predicted lifespan of the component as the output of the machine learning model.

[0135] According to any of the foregoing provisions, the machine learning model is trained on a training dataset of training images comprising multiple training components.

[0136] The system according to any of the foregoing provisions, wherein the machine learning model is a first machine learning model, and wherein the system further comprises: processing the image using a second machine learning model to generate an enhanced image; and processing the enhanced image and the stress map of the component using the first machine learning model by the computing system to generate the predicted lifetime of the component as the output of the machine learning model.

[0137] According to any of the foregoing provisions, the system wherein the machine learning model is trained based on a training dataset comprising training images of a plurality of training components, training stress maps of the plurality of training components, and multi-step predicted lifetimes of the plurality of training components, wherein the multi-step predicted lifetime of each of the plurality of training components is generated by: processing the training images of the training components to detect grain structures on the training components; comparing the detected grain structures with the corresponding training stress maps of the training components; and determining the multi-step predicted lifetime of the training components based on the location of the detected grain structures and the training stress maps.

Claims

1. A computer-implemented method for determining the predicted lifespan of a component, the method comprising: An image (504) of component (10) is obtained by a computing system (100) including one or more computing devices; The calculation system (100) obtains a stress map (506) of the component (10), the stress map (506) indicating the areas of high stress and low stress in the component (10); The computing system (100) processes the image (504) and stress map (506) of the component (10) using a machine learning model (520) to generate a predicted lifetime (522) of the component (10) as the output of the machine learning model (520).

2. The method according to claim 1, wherein, The component (10) has undergone at least one operating cycle.

3. The method according to claim 1, wherein, The component (10) is a new component.

4. The method according to any of the preceding claims, wherein, The predicted lifetime (522) is at least one of the following: a binary output (540) indicating whether the component (10) can survive another operating cycle, an integer output (542) indicating the number of operating cycles in which the component (10) can survive, or a probability density function (544).

5. The method according to any of the preceding claims, wherein, The image (504) is one of a photograph (510A, 510B), a video frame, a laser scan image, an X-ray scan image, a Laue orientation image, an electron channel contrast image, or a three-dimensional scan geometry.

6. The method according to any preceding claim, further comprising: The computing system (100) obtains operational data (508) associated with the component (10), the operational data (508) including time data and temperature data; and The computing system (100) processes the image (504), the stress map (506), and the operation data (508) using the machine learning model (520) to generate a predicted lifetime (522) of the component (10) as the output of the machine learning model (520).

7. The method according to any of the preceding claims, wherein, The machine learning model (520) is a first machine learning model (520), and the method further includes: The image is processed using a second machine learning model (504) to generate an enhanced image (602); and The computing system (100) processes the enhanced image (602) and the stress map (506) of the component (10) using the first machine learning model (520) to generate the predicted lifetime (522) of the component (10) as the output of the machine learning model (520).

8. The method according to any of the preceding claims, wherein, The machine learning model (520) is trained on a training dataset (162) that includes training images of multiple training components.

9. The method according to claim 8, wherein, The training dataset (162) also includes training stress maps of the plurality of training components and multi-step predicted lifetimes of the plurality of training components.

10. The method according to claim 9, wherein, The multi-step predicted lifetime for each of the plurality of training components is generated by the following operation: The training image of the training component is processed to detect the grain structure on the training component; The detected grain structure is compared with the corresponding training stress map of the training component; as well as The multi-step predicted lifetime of the training component is determined based on the location of the detected grain structure and the training stress map.

11. The method according to claim 10, wherein, Processing the training images includes performing pixel analysis of the training images using computer vision.

12. The method according to claim 10 or 11, wherein, The comparison step includes overlaying the training stress map onto the processed training image to compare the location of the detected grain structure with the stress direction of the training stress map.

13. The method according to any one of claims 10 to 12, wherein, The determination step includes determining the creep probability based on the location of the detected grain structure and the stress direction of the training stress map, and determining the multi-step predicted lifetime of the training component based on the creep probability.

14. A calculation system (100) for determining the predicted lifespan of a component, the system (100) comprising: One or more processors (1612); as well as One or more non-transitory computer-readable media (1614) storing instructions that, when executed by the one or more processors (1612), cause the computing system (100) to perform operations including: Obtain an image (504) of the component (10); A stress map (506) of the component (10) is obtained, the stress map (506) indicating areas of high stress and low stress in the component (10); The image (504) and stress map (506) of the component (10) are processed using a machine learning model (520) to generate a predicted lifetime (522) of the component (10) as the output of the machine learning model.

15. The system according to claim 14, wherein, The predicted lifetime (522) is at least one of the following: a binary output (540) indicating whether the component (10) can survive another operating cycle, an integer output (542) indicating the number of operating cycles in which the component (10) can survive, or a probability density function (544).

Images