System and method for assessing asset health

Abstract

Methods and systems for assessing the health of an asset are provided. For example, a system is provided. The system may include a processor and a memory including instructions that, when executed by the processor, cause the processor to perform operations consistent with identifying defects in a component of the asset. The operations may include acquiring, from an inspection system, a plurality of images obtained by the inspection system inspecting the component of the asset. The operations may include identifying a subset of images from the plurality of images based on an image processing technique encoded and included as part of the instructions. The subset of images represents defects in the component of the asset, and the image processing technique is selected from the group consisting of an automatic distress sorting technique, a structural similarity technique, a mean subtraction filtering technique, and a Hessian norm calculation technique.

Description

Technical Field

[0001] The present disclosure relates to asset inspection. More specifically, the present disclosure relates to systems and methods for assessing the health or performance of an asset or one or more of its subcomponents. Background Art

[0002] In many industrial applications, routine inspections of assets can help extend the useful life of those assets, as inspections can reveal damaged parts that need to be replaced or repaired. For example, in aviation applications, engines are routinely inspected to monitor their overall health and performance. When inspecting an engine, a borescope-based inspection (BSI) can be performed to view various engine subcomponents; BSI typically involves capturing a sequence of images, each image being a frame, which can be analyzed to characterize one or more aspects of the subcomponent depicted in that frame.

[0003] In one exemplary use case, a typical BSI system may capture a video using a probe that is inserted into the engine in order to reach the subcomponent of interest; then, through trial and error, a trained BSI system operator may select a good view of the subcomponent based on the video or video frames. The operator may then move to the next subcomponent by actuating the probe to another location within the engine. These typical steps in the BSI method result in the capture of many images, only a few of which show the best views of the subcomponent defects. Typically, an expert makes a decision on the condition of the subcomponent based on these selected few frames. Therefore, the quality of the inspection depends largely on being able to properly locate these frames of interest from the many captured frames; a process that is inherently difficult and subjective and depends on the technician's skills. As a result, current inspection methods are not only inefficient, but also prone to errors. Summary of the invention

[0004] The embodiments described herein help solve or mitigate the above-mentioned problems and other problems known in the art. For example, in one embodiment, a system for identifying defects in a component of an asset is provided. The system includes a processor and a memory including instructions that, when executed by the processor, cause the processor to perform operations consistent with identifying defects. For example, the operations may include acquiring from an inspection system a plurality of images obtained by the inspection system inspecting a component of the asset. The operations may include identifying a subset of images from the plurality of images based on an image processing technique that is encoded and included as part of the instructions. The subset of images represents defects in a component of the asset, and the image processing technique is selected from the group consisting of an automatic distress sorting technique, a structural similarity technique, a mean subtraction filtering technique, and a Hessian norm calculation technique.

[0005] In another embodiment, a method for identifying defects in a component of an asset is provided. The method includes acquiring, by a defect identification system, from an inspection system a plurality of images obtained by inspecting the component of the asset by the inspection system. The method also includes identifying, by the defect identification system, a subset of images from the plurality of images based on an image processing technique. The subset of images represents defects in the component of the asset, and the image processing technique is selected from the group consisting of an automatic distress sorting technique, a structural similarity technique, a mean subtraction filtering technique, and a Hessian norm calculation technique.

[0006] Additional features, modes of operation, advantages, and other aspects of various embodiments are described below with reference to the accompanying drawings. Note that the present disclosure is not limited to the specific embodiments described herein. These embodiments are presented for illustrative purposes only. Based on the teachings provided, other embodiments or modifications of the disclosed embodiments will be apparent to those skilled in the relevant art. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The illustrative embodiments may take the form of various components and arrangements of components. Illustrative embodiments are shown in the accompanying drawings, in which like reference numerals may indicate corresponding or similar parts throughout all the drawings. The drawings are for the purpose of illustrating the embodiments only and should not be construed as limiting the present disclosure. Given the following description of the enabling drawings, the novel aspects of the present disclosure will become apparent to those of ordinary skill in the relevant art.

[0008] Figure 1 A system according to several aspects described herein is shown.

[0009] Figure 2 Methods according to several aspects described herein are shown.

[0010] Figure 3 Methods according to several aspects described herein are shown.

[0011] Figure 4 Methods according to several aspects described herein are shown.

[0012] Figure 5 A system according to several aspects described herein is shown. DETAILED DESCRIPTION

[0013] Although illustrative embodiments are described herein for specific applications, it should be understood that the present disclosure is not limited thereto. Those skilled in the art and those who have access to the teachings provided herein will recognize additional applications, modifications and embodiments within the scope of the present invention and additional fields in which the present disclosure will be largely useful.

[0014] For example, one or more embodiments described herein can be a system that automatically determines the best frame from a set of frames acquired by an inspection system. In one embodiment, the best frame (e.g., one or more frames of interest) is identified by first preprocessing the image sequence using a mean subtraction filter and then monitoring the Hessian norm of the image information. The "best" frame can be selected based on one or more criteria, for example, the results of one or more preprocessing operations described above. In another embodiment, the exemplary system can be configured to select frames from a video based on the maximum exposure of different components; the selection can be achieved using structural similarity methods, etc. In this way, the embodiment eliminates subjectivity from the inspection process, thereby providing consistency and accuracy.

[0015] In an exemplary embodiment, the distress ranking method does not need to be performed on all captured frames, which allows the health assessment of the engine to be performed approximately 100 times faster than typical inspection methods, in addition to improving the accuracy of the inspection. The embodiment also provides a training data set for the distress ranking algorithm, if the operator decides to use such an algorithm when evaluating engine health and performance. In this way, the embodiment can reduce training time in addition to increasing reliability.

[0016] Embodiments confer several advantages. For example, one or more embodiments described herein can automatically estimate the extent of defects recorded during inspection. Furthermore, embodiments can automatically provide additional information statistics about defects, thereby providing a means for continuous time and / or continuous valued defect metrics. Thus, the embodiments facilitate redefining the ranking procedure, and therefore, they reduce errors in current processes that cannot be discretized by a step size of <1 (continuous).

[0017] Embodiments also allow tracking of distress progression, so they help predict parts replacement and optimize planned maintenance cycles or on-time delivery of assets. In addition, embodiments allow for a continuous distress prioritization method (DRM) and when combined with operational parameters, embodiments can help establish distress trends and predictions.

[0018] Additionally, in one or more exemplary embodiments, the acquired image may be rendered onto a computer-aided design (CAD) model to enable an inspector to assess actual distress relative to the designed (or manufactured) sub-component being inspected. In the alternative, the exemplary system may be configured to automatically (i.e., without user intervention) assess actual distress relative to the designed or manufactured sub-component being inspected using the synthetic image (i.e., the image rendered onto the CAD model).

[0019] In general, embodiments include an intelligent and dynamic component health monitoring system that combines programmable inspection hardware and embedded image processing techniques to characterize part failures and automatically provide recommendations taking into account historical and physics-based observations. In addition, in general, embodiments help reduce the time required for engine evaluations, and they improve the reliability of evaluations. In addition, embodiments help enhance the reliability of automated distress ranking algorithms, and they also help reduce the time required to prepare data sets for training automated distress ranking algorithms.

[0020] Embodiments include a dynamic recommendation system that enables component distress prioritization through seamless integration of advanced data acquisition, analytical techniques, and physics-based considerations. They may allow for keyframe extraction, which provides engineers with a new, reduced, but accurate data set that can be used to build more effective analysis based on engine health. Their embodiments also provide standardized and consistent images that can be stored for later use. Embodiments also provide improved predictive models due to built-in quantification of the number of connected areas and the distribution of defective areas. Embodiments also allow design, operations, and maintenance data to be automatically unified with image analysis and physics-based data, resulting in enhanced predictive analysis. The latter functionality reduces the tedious manual work of typical inspection procedures by leveraging and customizing deep learning methods for analysis.

[0021] Embodiments allow real-time streaming of image acquisition, which results in fast, scalable, efficient and lightweight image or video processing. This can allow for various downstream analyses, thereby improving component assessment. In general, the performance of embodiments and the quality of inspection do not depend on the skills and knowledge of the inspection engineer.

[0022] The exemplary embodiment may also be configured to identify key frames of interest based on at least five steps. For example, summary information (e.g., the Hamming norm) of each frame may be used to make the decision. A structural similarity method may be used to provide additional information, which is independent of absolute pixel values ​​and works using internal dependencies of pixel values. This method is insensitive to external structural changes of the component (e.g., cracks, discoloration, etc.).

[0023] Embodiments may also help quantify defects. This is accomplished through an algorithm that measures pixels marked as defects through a recognition machine learning model. Different statistical measures, such as the number of connected regions, the distribution of these regions, and the most affected regions, may be provided to improve downstream analysis.

[0024] In addition to the technical points of novelty and technical advantages described above, embodiments also have several commercial advantages that address currently unmet needs in the industry. For example, embodiments provide faster inspections, thereby avoiding spending more time finding the best view of a component in a video obtained through inspection. This will reduce defect omissions due to operator negligence, which can be detrimental because further damage may occur if a faulty component is missed during the inspection process. Therefore, embodiments reduce maintenance costs and improve the time on wing of aircraft engines.

[0025] Embodiments may utilize image recognition algorithms such as pattern recognition algorithms and may determine a frame to be a frame of interest if certain conditions are met while processing the image using the algorithm. For example, such a criterion may be the mean squared error (MSE). However, in general, exemplary methods or systems may be independent of absolute pixel values ​​and may work using internal dependencies of pixel values ​​because pixel dependencies carry important information about the structure of objects in the visual scene. In embodiments, CNN may be used as a deep learning technique to directly train a deep neural network that may directly quantify damage to a component, rather than implementing component assessment in two steps (detection and quantification).

[0026] Having generally described the features and several advantages of the embodiments, it is now possible to describe Figure 1 Discuss exemplary embodiments. Figure 1 The exemplary system 100 depicted in FIG. 1 includes a plurality of components and subsystems configured to perform an inspection of an asset 111 and detect one or more defects of a component 101 of the asset 111 without user intervention. The system 100 includes an evaluation system 108 configured to determine from a video 102 (or a collection of images 104a, 104c, and 104n) and a reference frame 104 (or image) provided by an inspection system 103. The inspection system 103 may be, for example, but not limited to, a borescope inspection system.

[0027] Inspection system 103 is configured to acquire video 102, specifically, to inspect component 101, via a probe inserted into asset 111. In one exemplary use case, the probe may be actuated such that video 102 includes several views of component 101 and / or several views of multiple components similar to component 101. Inspection system 103 may be further configured to provide reference frames 104, which may be a set of frames representing known non-defective portions of component 101. For example, reference frames 104 may have been saved in a memory of inspection system 103 when asset 111 or component 101 was first commissioned, or it may be an image of a similar component in a pristine state.

[0028] The system 100 also includes an evaluation system 108 configured to acquire the video 102 and the reference frame 104 from the inspection system 103. The evaluation system 108 is further configured to determine from the set of images (i.e., frames) from the video 102 to select key frames of interest, i.e., a subset of images 110 that respectively correspond to defects of the component 101. The defects here may be deformations, cracks, or any other undesirable structural changes caused by long-term use, which may increase the risk of failure of the component 101 during the operation of the asset 111.

[0029] Thereafter, the operation and various aspects of the system 100 are described in terms of an inspection of one or more aviation-related components. As explained herein, a video is a sequence of images (i.e., frames), and the quality of information that can be determined from the sequence of images depends on many factors. For example, for a borescope video of an aircraft engine, some factors may be (non-exhaustive) the operation of the camera, lighting conditions, accessibility to the components of interest, capture angles, capture locations, specific movement of the probe, and camera specifications. Although these factors may result in poor quality of the video 102, the system 100 is still configured to identify key frames of interest.

[0030] As a non-limiting example, the system 100 is described in the context of inspecting one or more blades of an engine. In a typical inspection, the probe of the inspection system 103 is introduced into the engine through a borescope port near the blade and supported in a convenient position. The blade group is then slowly rotated from the outside (usually manually) to rotate the blade about its axis. Captured in the video is one blade that starts out of focus from the background, is initially partially visible, then slowly comes into full focus, and then moves out as the next blade comes into view. Therefore, there is a position / time of maximum visibility of the blade. The latter situation can provide the best view for correctly identifying the blade.

[0031] The system 100 is configured to identify those key frames corresponding to the best view of the blade among all frames captured by the borescope. This identification is performed as follows according to an exemplary method 200 that may be performed by the evaluation system 108. The exemplary method 200 is as follows: Figure 2 As shown. The method 200 starts at step 202 and is characterized by step 204, which includes applying a mean subtraction filter on each frame to normalize the illumination / illumination of the component (i.e., in this example, the blade in the field of view of the camera of the borescope). Other filters such as Gaussian smoothing can also be used in combination or instead of the mean subtraction filter.

[0032] The method 200 is characterized by step 206, which includes converting each acquired frame (or each frame in the survey) to grayscale to further standardize all frame sequences being analyzed. The method 200 also includes generating a metric that summarizes the information of each frame (step 208). For example, but not limited to, the Hamming norm can be used to summarize specific statistical information. The method 200 further includes (at step 210) generating a time map between the summarized information and the group of frames being examined. In one embodiment, the evaluation system 108 is configured to determine whether the time map exhibits a periodic trend, such as, but not limited to, a sine curve. The method 200 includes identifying and / or collecting all frames corresponding to the maximum position in the detected periodic trend (step 212). The frames corresponding to these maximum values ​​are the frames with the majority of the image showing the blade. In other words, the frames corresponding to the maximum values ​​of each cycle of the sequence Hamming norm are the desired key frames. These are the frames with the maximum desired information. Similarly, the frames corresponding to the minimum values ​​of each cycle have less information about the blade, but have more information about the background, which can also be the inspection value in some embodiments. Method 200 may then include publishing the collected frames (step 214 ) and end at step 216 .

[0033] In another embodiment, the evaluation system 108 can be configured to overcome the problem of identifying the frame with the maximum exposure based on the structural similarity method. In this embodiment, the structural similarity method is independent of the absolute pixel values ​​and works using the internal dependence of the pixel values, so the evaluation system 108 is able to identify frames with the maximum exposure of different parts and is insensitive to changes in the external structure of the parts (such as cracks, discoloration, etc.). In this method, pixels that are spatially close in the image will have strong internal dependence and therefore carry important information about the structure of the object in the visual scene. The exemplary method 200 can therefore include a threshold adjustment based on the video quality to filter the right frame, as well as an internal threshold adjustment to avoid marking multiple frames of the same part.

[0034] In yet another embodiment, the system 100 is configured to quantify defects that have been detected on a component in a manner that will improve the current damage analysis model. For example, a machine learning algorithm for defect identification can be used on an input image to mark defect areas. In the above example of blade inspection, key frames detected by the evaluation system 108 and then processed by the defect detection module of the evaluation system 108 mark pixels in the original image that correspond to defects in the blade.

[0035] For example, these defects come from a pre-trained set of spalling, oxidation, cracking, material removal, or others. The evaluation system 108 uses this set as input to perform the method 300 ( Figure 3), thereby quantifying these defects. The method 300 begins at step 302 and includes separating the input into binary images corresponding to each defect pattern of interest from the training set (step 304). The method 300 further includes performing the same separation for each frame acquired from the inspection (step 306). In the binary image, areas that are defects are pixels with a value of +1, while unaffected areas are pixels with a value of 0 (step 308). It should be noted that binary values ​​are assigned to defects and non-defective areas by convention and are therefore not limiting.

[0036] The method 300 includes finding the number of connected regions (310). A connected region is a set of pixels that have a value of +1 and that touch each other, so that within a connected region, one can traverse from any starting pixel to any other pixel in the region without skipping any pixels. Depending on the component and the defect, there may be any number of such connected regions. The method 300 then provides an area distribution of connected regions for each defect in each image (312). The method 300 also provides a cumulative sum of these connected regions. These are then used to better represent the condition of the components in the engine. For example, differences in distribution between different blades represent different health levels for each blade. In this way, the method 300 helps improve analytical models used to predict life and thereby repair or remove components because defects can be quantified and therefore classified using the exemplary method 300 and system 100. The method 300 then ends at step 314.

[0037] In yet another embodiment, the system 100 may be configured to perform the method 400 ( Figure 4 ). The method 400 includes annotating (marking regions representing detected damaged areas in a visual image or frame) damage patterns in a visual form, such as an image or video frame format. The method 400 includes utilizing image processing techniques to determine a set of features including, but not limited to: extent of damage to the part (area or length), nature (specific to the damage pattern: example: extended vs. localized), such as geometric attributes of shape, texture (smooth, abraded, patterned, or granular), form (continuous or wavy), color (gradient variations and shaded patterns), orientation relative to image edges, and severity metrics defined for each damage pattern.

[0038] In general, method 400 includes providing annotations to visual input that help classify, i.e., distinguish, each of the aforementioned damage modes. These annotated features of the image or frame are processed at the pixel level to extract relevant metrics for each of the aforementioned damage modes. Specifically, method 400 begins at step 402 and includes extracting specific channel pixel values ​​(e.g., RGB) to perform channel segmentation (step 404). Method 400 also includes identifying color gradients across image or frame regions based on, for example, pixel values ​​(step 406).

[0039] The method 400 further includes determining the intelligence encoded in the channel as a heuristic method related to a specific damage pattern based on the segmented or isolated areas that match the heuristic conditions (step 408). These may include: cropping, masking, smoothing, adding noise, blending, geometric transformations or convolution. Then, the method 400 includes (step 410) generating metrics of the damage pattern from the above identified areas, such as shape, form, geometric properties, texture and color. In addition to the above operations, in order to quantify the extent of damage, the method 400 may include operations such as maximum continuous area identification of large isolated areas within the image (step 412). The method 400 also includes using information from these individual maximum continuous areas identified for each image to summarize metrics including total pixel area, area percentage, maximum continuous pixel area (step 414). These metrics provide partition (location) and component information based on specific criteria. The method 400 also includes generating a set of rules and knowledge base database modules that maintain necessary information for different components and damage patterns and input requirements. The method 400 further includes generating one or more reports or outputs for each of the damage patterns and locations within the frames or images using the quantitative information consistent with the above-mentioned metrics (step 416), and the method 400 ends at step 418. The format of the one or more outputs or reports can be a specified format that conforms to a downstream system (e.g., analysis, diagnostics, safety, operations, maintenance planning).

[0040] Figure 5 A system 500 is shown according to an exemplary embodiment. The system 500 can be configured to implement one or more of the methods for defect identification described above. The system 500 includes a dedicated processor 514 that is configured to perform tasks specific to assessing the health and / or performance of an asset. The processor 514 has a specific structure imparted by instructions stored in the memory 502 and / or by instructions 518, which can be retrieved by the processor 514 from a storage medium 520. The storage medium 520 can be co-located with the processor 514, or can be located elsewhere and communicatively coupled to the processor 514 via a communication interface 516.

[0041] System 500 may be a stand-alone programmable system or a programmable module in a larger system, which may itself be centralized or distributed in various locations or computing infrastructures, such as cloud-based computing infrastructures. Processor 514 may include one or more hardware and / or software components configured to acquire, decode, execute, store, analyze, distribute, evaluate, and / or classify information. In addition, processor 514 may include an input / output module (I / O module 512) that may be configured to ingest data related to a single asset or a group of assets. Processor 514 may include one or more processing devices or cores (not shown). In some embodiments, processor 514 may be a plurality of processors, each having one or more cores. Processor 514 may be configured to execute instructions retrieved from memory 502 (i.e., from one of storage block 504, storage block 506, storage block 508, and storage block 510).

[0042] Without loss of generality, storage 520 and / or memory 502 may include volatile or nonvolatile, magnetic, semiconductor, tape, optical, removable, non-removable, read-only, random access, or any type of non-transitory computer-readable computer media. Storage media 520 may be configured to record data processed, recorded, or collected during operation of processor 514. The data may be time-stamped, location-stamped, cataloged, indexed, or organized in a variety of ways consistent with data storage practices. Storage 520 and / or memory 502 may include programs and / or other information that processor 514 may use to perform tasks consistent with the processes and / or methods described herein.

[0043] For example, but not limited to, processor 514 may be constructed from instructions from memory blocks 506, memory blocks 508, and memory blocks 510 to perform operations that result in identification of a subset of images 507 representing one or more defects 513 of component 501 from asset 511. Processor 514 may execute the aforementioned image processing instructions 515 from memory blocks 506, 508, and 510, which will cause processor 514 to perform certain operations associated with monitoring the health and / or performance of components of an engine. The operations may include acquiring from inspection system 500 a plurality of images 505 acquired from an inspection of component 501 of asset 511 by inspection system 503. The operations may include identifying a subset of images 507 from the plurality of images 505 based on image processing techniques encoded and included as part of the instructions in memory blocks 506, 508, and 510. The subset of images 507 represents defects in component 501 of asset 511. The image processing technique is selected from the group consisting of an automatic distress sorting technique, a structural similarity technique, a mean subtraction filtering technique, and a Hessian norm calculation technique.

[0044] Note that although the embodiments have been described in the context of aerospace applications and using the BSI method, they may be used in many industrial applications where inspections are performed without necessarily using BSI. As such, those skilled in the relevant art will appreciate that various adaptations and modifications of the above-described embodiments may be constructed without departing from the scope and spirit of the present disclosure. Therefore, it should be understood that within the scope of the appended claims, the present disclosure may be practiced in a manner other than as specifically described herein.

[0045] Further aspects of the invention are provided by the subject matter of the following clauses:

[0046] 1. A system for identifying defects in a component of an asset, the system comprising: a processor; a memory containing instructions that, when executed by the processor, cause the processor to perform operations comprising: acquiring from an inspection system a plurality of images obtained by the inspection system inspecting the component of the asset; identifying a subset of images from the plurality of images based on an image processing technique encoded and included as part of the instructions, wherein the subset of images represents the defects in the component of the asset, and wherein the image processing technique is selected from the group consisting of an automatic distress sorting technique, a structural similarity technique, a mean subtraction filtering technique, and a Hessian norm calculation technique.

[0047] 2. A system according to any of the preceding clauses, wherein the operation further comprises: obtaining a CAD model of the part and rendering a specified image from the subset of images onto the CAD model.

[0048] 3. A system according to any preceding clause, wherein the operation further comprises: obtaining the CAD model from a database communicatively connected to the system.

[0049] 4. A system according to any preceding clause, wherein the operation further comprises: identifying one or more areas of the specified image representing the defect.

[0050] 5. A system according to any preceding clause, wherein the asset is an engine.

[0051] 6. A system according to any preceding clause, wherein the inspection system comprises a borescope inspection system.

[0052] 7. A system according to any of the preceding clauses, wherein the operations further comprise: extracting the plurality of images from a video.

[0053] 8. A system according to any preceding clause, wherein the plurality of images comprises a plurality of views of the component.

[0054] 9. The system of any preceding clause, the operations further comprising providing a metric associated with the defect based on a specified image from the subset of images.

[0055] 10. A system according to any preceding clause, wherein the metric is independent of absolute pixel values ​​of a specified image from the subset of images.

[0056] 11. A system according to any preceding clause, wherein the metric is based on summary information of a specified image from a subset of the images.

[0057] 12. A system according to any preceding clause, wherein the summary information comprises a Hamming norm.

[0058] 13. A system according to any preceding clause, wherein the operation further comprises: providing a data set comprising metrics generated from a subset of the images based on the image processing technique.

[0059] 14. A system according to any of the preceding clauses, wherein the operation further comprises: training a neural network or a deep learning system based on the data set.

[0060] 15. A system according to any preceding clause, wherein the operations further comprise providing another data set based on the results of the training, the another data set representing the predicted performance of the component.

[0061] 16. A method for identifying defects in a component of an asset, the method comprising: acquiring, through a defect recognition system, a plurality of images obtained by inspecting the component of the asset through the inspection system; identifying, through the defect recognition system based on image processing technology, a subset of images from the plurality of images, wherein the subset of images represents the defects in the component of the asset, and wherein the image processing technology is selected from the group consisting of automatic distress sorting technology, structural similarity technology, mean subtraction filtering technology and Hessian norm calculation technology.

[0062] 17. A method according to any preceding clause, further comprising: obtaining a CAD model of the component and rendering a specified image from the subset of images onto the CAD model.

[0063] 18. The method of any preceding clause, further comprising: obtaining the CAD model from a database communicatively coupled to the system.

[0064] 19. The method according to any preceding clause, further comprising: extracting the plurality of images from a video.

[0065] 20. The method of any preceding clause, further comprising providing a metric associated with the defect based on a specified image of the subset of images.

Claims

1. A system for identifying defects in a component of an asset, characterized in that The system comprises: processor; a memory containing instructions that, when executed by the processor, cause the processor to perform operations including: acquiring, from an inspection system, a plurality of images obtained by the inspection system inspecting the component of the asset; identifying a subset of images from the plurality of images based on an image processing technique encoded and included as part of the instructions, providing a metric associated with the defect based on a specified image from the subset of images, wherein the metric is independent of absolute pixel values ​​of the specified image from the subset of images, the metric is based on summary information of the specified image from the subset of images, the summary information comprising a Hamming norm; wherein the subset of the images represents the defect in the component of the asset, and Wherein, the image processing technology is selected from the group consisting of automatic distress sorting technology, structural similarity technology, mean subtraction filtering technology and Hessian norm calculation technology.

2. The system according to claim 1, characterized in that in, The operations further include obtaining a CAD model of the part and rendering a specified image from the subset of images onto the CAD model.

3. The system according to claim 2, characterized in that in, The operations further include obtaining the CAD model from a database communicatively coupled to the system.

4. The system according to claim 2, characterized in that in, The operations further include identifying one or more regions of the designated image that represent the defect.

5. The system according to claim 1, characterized in that in, The asset is an engine.

6. The system according to claim 1, characterized in that in, The inspection system includes a borescope inspection system.

7. The system according to claim 1, characterized in that in, The operations further include extracting the plurality of images from a video.

8. The system according to claim 1, characterized in that in, The plurality of images includes a plurality of views of the component.

9. The system according to claim 1, characterized in that in, The operations further include providing a data set including metrics generated from a subset of the images based on the image processing technique.

10. The system according to claim 9, characterized in that in, The operations further include: training a neural network or a deep learning system based on the data set.

11. The system according to claim 10, characterized in that in, The operations further include providing another second data set based on results of the training, the another second data set representing predicted performance of the component.

12. A method for identifying defects in a component of an asset, characterized in that The method comprises: acquiring, by a defect identification system, from an inspection system, a plurality of images obtained by inspecting the component of the asset by the inspection system; The defect recognition system identifies a subset of images from the plurality of images based on image processing technology, providing a metric associated with the defect based on a specified image from the subset of images, wherein the metric is independent of absolute pixel values ​​of the specified image from the subset of images, the metric is based on summary information of the specified image from the subset of images, the summary information comprising a Hamming norm; wherein the subset of the images represents the defect in the component of the asset, and Wherein, the image processing technology is selected from the group consisting of automatic distress sorting technology, structural similarity technology, mean subtraction filtering technology and Hessian norm calculation technology.

13. The method according to claim 12, characterized in that Further including: A CAD model of the part is obtained, and a specified image from the subset of images is rendered onto the CAD model.

14. The method according to claim 13, characterized in that Further including: The CAD model is retrieved from a database communicatively coupled to the system.

15. The method according to claim 12, characterized in that Further including: The plurality of images are extracted from a video.

16. The method according to claim 12, characterized in that Further including: A metric associated with the defect is provided based on a specified image of the subset of images.

Images