Machine learning training based on photorealist training sets

Abstract

Disclosed herein are methods and systems for generating a synthetic training set for training a machine learning (ML) model to detect machine failures, comprising receiving visual media depicting one or more faults associated with one or more failure modes of one or more elements of a machine, generating one or more photorealistic images imaging the one or more faults by processing the visual media in correlation with realistic image data of the one or more elements, and outputting the one or more photorealistic images for training one or more ML models to estimate presence of the one or more failure modes in one or more machine items comprising the one or more elements based on analysis of real-world image data depicting the one or more elements.

Description

[0001] MACHINE LEARNING TRAINING BASED ON PHOTOREALIST

[0002] TRAINING SETS

[0003] RELATED APPLICATION / S

[0004] This application claims the benefit of priority of U.S. Provisional Patent Application No. 63 / 533,668, filed on August 20, 2023, the contents of which are incorporated herein by reference in their entirety.

[0005] BACKGROUND

[0006] The present invention, in some embodiments thereof, relates to detecting failure modes of mechanical elements using optical sensors, and, more specifically, but not exclusively, to generating photorealistic training sets for training Machine Learning (ML) models to detect failure modes of mechanical elements using optical sensors.

[0007] Machine maintenance conducted to monitor health of machinery assets and detect potential failure modes is essential for practically any mechanical machinery, system, platform, assembly and / or the like in a plurality of industries and market segments ranging from industrial and agricultural machinery, through automotive to defense and military applications to name just a few.

[0008] Efficient, effective, and reliable machine maintenance must be conducted in order to comply with regulatory requirements, reduce risk for damage, injury and / or death, meet production schedules, minimize costly downtime, and / or the like.

[0009] Machine maintenance may include, for example, inspection, testing, repair, replacement, alignment, and / or the like of mechanical elements, for example, element, parts, modules, assemblies, elements, and / or the like which may be carried out in regularly scheduled service, routine checkups, scheduled and / or emergency repairs and / or the like.

[0010] However such periodic and / or scheduled maintenance may present major challenges in terms of downtime costs, maintenance resources (equipment, personnel, etc.), maintenance time, and / or the like since they are typically done offline during downtime while the inspected and maintained machinery is not operating in its designated capacity.

[0011] Major resources are therefore invested to develop and deploy efficient health monitoring and predictive failure detection means which may automatically inspect machine elements while the machine is fully operational in order to detect, estimate, and / or predict presence and / or development of failure modes in the mechanical elements. SUMMARY

[0012] According to a first aspect of the present invention there is provided a method of generating a synthetic training set for training a machine learning (ML) model to detect machine failures, comprising using one or more processors for: receiving visual media depicting one or more faults associated with one or more failure modes of one or more elements of a machine, generating one or more photorealistic images imaging the one or more faults by processing the visual media in correlation with realistic image data of the one or more elements, and outputting the one or more photorealistic images for training one or more ML models to estimate presence of the one or more failure modes in one or more machine items comprising the one or more elements based on analysis of real-world image data depicting the one or more elements.

[0013] According to a second aspect of the present invention there is provided a system for generating a synthetic training set for training a machine learning (ML) model to detect machine failures, comprising one or more processors configured to execute a code. The code comprising: code instructors to receive visual media depicting one or more faults associated with one or more failure modes of one or more elements of a machine, code instructors to generate one or more photorealistic images imaging the one or more faults by processing the visual media in correlation with realistic image data of the one or more elements, and code instructors to output the one or more photorealistic images for training one or more ML models to estimate presence of the one or more failure modes in one or more machine items comprising the one or more elements based on analysis of real-world image data depicting the one or more elements.

[0014] According to a third aspect of the present invention there is provided a method of generating a synthetic training set for testing a failure detection system, comprising using at least one processor for: receiving at least one 3D graphical representation depicting at least one fault associated with at least one failure mode of at least one element of a machine, generating at least one photorealistic image imaging the at least one fault by processing the at least one 3D graphical representation in correlation with realistic image data of the at least one element, and outputting the at least one photorealistic image for testing at least one failure detection system employing at least one ML model trained to estimate presence of the at least one failure mode in at least one machine item comprising the at least one element based on analysis of real-world image data depicting the at least one element.

[0015] According to a fourth aspect of the present invention there is provided a method of detecting machine failures , comprising using one or more processors for: receiving one or more image depicting, at least partially, one or more elements of one or more machines, applying one or more trained ML models to estimate presence of one or more failure modes of one or more elements of the one or more machines based on detection of one or more faults associated with the one or more failure modes in the one or more images, and outputting the estimation of presence of the one or more failure modes of the one or more elements. Wherein the one or more ML models are trained using one or more photorealistic images imaging the one or more faults. The one or more photorealistic images are generated by processing visual media depicting the one or more faults in correlation with realistic image data of the one or more elements.

[0016] According to a fifth aspect of the present invention there is provided a method of adjusting imaging parameters of failure detection system based on analysis of photorealistic images, comprising using one or more processors for: receiving one or more photorealistic image imaging the one or more elements generated by processing visual media in correlation with realistic image data of the one or more elements, generating adjustment instructions based on analysis of the one or more photorealistic images for adjusting one or more image capturing parameters of a failure detection system adapted to detect one or more faults associated with one or more failure modes of the one or more elements based on image data of the machine and / or part thereof, and outputting the adjustment instructions to the failure detection system.

[0017] In a further implementation form of the first, second, third, and / or fifth aspects, the one or more digital design files are members of a group comprising: a 2D design file, a 3D design model, and / or the like.

[0018] In a further implementation form of the first, second, third, and / or fifth aspects, at least part of the realistic image data of the one or more elements is captured by one or more image sensors deployed to capture image data of the one or more machine items and / or part thereof.

[0019] In a further implementation form of the first, second, third, and / or fifth aspects, the one or more image sensors are adapted to capture image data of the one or more machine items and / or part thereof while both the one or more image sensors and the one or more machine items and / or part thereof are moving.

[0020] In a further implementation form of the first, second, third, and / or fifth aspects, the one or more image sensors are adapted to capture image data of the one or more machine items and / or part thereof while the one or more image sensors are static with respect to the one or more elements.

[0021] In an optional implementation form of the first, second, third, and / or fifth aspects, at least part of the realistic image data captured by the one or more image sensors is processed to reduce one or more degrading effects. In an optional implementation form of the first, second, third, and / or fifth aspects, one or more positioning parameters of the one or more image sensors are adjusted to at least reduce the one or more degrading effects in captured images of the one or more machine items and / or part thereof.

[0022] In a further implementation form of the first, second, third, and / or fifth aspects, at least part of the realistic image data is generated using one or more generative ML models adapted to generate synthetic 3D images of the one or more elements based on one or more digital design files relating to the one or more elements.

[0023] In a further implementation form of the first, second, third, and / or fifth aspects, the one or more photorealistic images are generated using one or more generative ML models.

[0024] In an optional implementation form of the first, second, and / or third aspects, the visual media is generated to depict a trend of the one or more faults comprising development of the one or more faults into one or more failures, and generating one or more photorealistic image sequences imaging the trend of the one or more faults.

[0025] In a further implementation form of the first, second, and / or third aspects, the visual media is a plurality of images depicting development of the one or more faults into one or more failures.

[0026] In a further implementation form of the first, second, and / or third aspects, the visual media depicting the one or more faults are generated by adjusting one or more digital design files relating to the one or more elements adjusted to inject the one or more failure modes into design of the one or more elements.

[0027] In a further implementation form of the first, second, and / or third aspects, the visual media is generated based on one or more 3D animation depicting the one or more fault. The one or more 3D animations are created based on the one or more digital design files adjusted to inject the one or more failure modes.

[0028] In a further implementation form of the first, second, and / or third aspects, the one or more faults are visible in the one or more elements in the one or more of the photorealistic images.

[0029] In a further implementation form of the first, second, and / or third aspects, the one or more failure modes induces one or more faults in one or more another elements of the machine. The one or more faults are visible in one or more another elements of the machine in the one or more photorealistic images.

[0030] In a further implementation form of the first, and / or second aspects, the estimation of the one or more trained ML models comprises a probability score indicative of a probability of presence of the one or more failure modes in the one or more machine items. In an optional implementation form of the first, and / or second aspects, the estimation of the one or more trained ML models further comprises a severe estimation indicative of a level of severeness of the one or more failure modes in the one or more machine items.

[0031] In an optional implementation form of the first, and / or second aspects, one or more of the photorealistic images are used for validating and / or testing the one or more ML models.

[0032] In a further implementation form of the third aspect, the one or more photorealistic images are injected into the one or more failure detection systems via one or more image sensors adapted to capture the real- world image data depicting the at one or more elements.

[0033] In a further implementation form of the fourth aspect, the one or more trained ML models are adapted to compute a probability score indicative of a confidence level of the estimation.

[0034] In a further implementation form of the fourth aspect, the one or more failure modes are estimated to exist in the one or more machine based on a comparison of the probability score with a certain threshold.

[0035] In a further implementation form of the fifth aspect, the one or more image capturing parameters comprise one or more imaging parameters of one or more image sensors used by the failure detection system to capture the image data. The one or more imaging parameters are members of a group comprising: a positioning parameter, an inherent parameter, an operational parameter, and / or the like.

[0036] In a further implementation form of the fifth aspect, the one or more image capturing parameters comprise one or more illumination parameters relating to one or more light sources used by the failure detection system to illuminate the machine and / or part thereof.

[0037] Other systems, methods, features, and advantages of the present disclosure will be or become apparent to one with skill in the art upon examination of the following drawings and detailed description. It is intended that all such additional systems, methods, features, and advantages be included within this description, be within the scope of the present disclosure, and be protected by the accompanying claims.

[0038] Unless otherwise defined, all technical and / or scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention pertains. Although methods and materials similar or equivalent to those described herein can be used in the practice or testing of embodiments of the invention, exemplary methods and / or materials are described below. In case of conflict, the patent specification, including definitions, will control. In addition, the materials, methods, and examples are illustrative only and are not intended to be necessarily limiting. Implementation of the method and / or system of embodiments of the invention can involve performing or completing selected tasks manually, automatically, or a combination thereof. Moreover, according to actual instrumentation and equipment of embodiments of the method and / or system of the invention, several selected tasks could be implemented by hardware, by software or by firmware or by a combination thereof using an operating system.

[0039] For example, hardware for performing selected tasks according to embodiments of the invention could be implemented as a chip or a circuit. As software, selected tasks according to embodiments of the invention could be implemented as a plurality of software instructions being executed by a computer using any suitable operating system. In an exemplary embodiment of the invention, one or more tasks according to exemplary embodiments of method and / or system as described herein are performed by a data processor, such as a computing platform for executing a plurality of instructions. Optionally, the data processor includes a volatile memory for storing instructions and / or data and / or a non-volatile storage, for example, a magnetic hard-disk and / or removable media, for storing instructions and / or data. Optionally, a network connection is provided as well. A display and / or a user input device such as a keyboard or mouse are optionally provided as well.

[0040] BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0041] Some embodiments of the invention are herein described, by way of example only, with reference to the accompanying drawings. With specific reference now to the drawings in detail, it is stressed that the particulars shown are by way of example and for purposes of illustrative discussion of embodiments of the invention. In this regard, the description taken with the drawings makes apparent to those skilled in the art how embodiments of the invention may be practiced.

[0042] In the drawings:

[0043] FIG. 1 A is a flowchart of an exemplary process of generating photorealistic images for use as training samples for training ML models to detect machine failures, according to some embodiments of the present invention;

[0044] FIG. IB is a flowchart of an exemplary process for modeling and analyzing trend-based fault development, according to some embodiments of the present invention;

[0045] FIG. 2 is a schematic illustration of an exemplary system for generating photorealistic images for use as training samples for training ML models to detect machine failures, according to some embodiments of the present invention; FIG. 3 presents a photorealist image of a first exemplary mechanical element generated by processing visual media depicting the element in correlation with image data of the element, according to some embodiments of the present invention;

[0046] FIG. 4 presents a photorealist image of a second exemplary mechanical element generated by processing a visual media depicting the element in correlation with image data of the element and its exemplary background, according to some embodiments of the present invention;

[0047] FIG. 5A and FIG. 5B present photorealist images of an exemplary fault associated with a failure mode of an exemplary element of a machine where the failure mode is injected in a digital design file of the element used to create a visual media depicting the element which is used to generate the photorealist image, according to some embodiments of the present invention;

[0048] FIG. 6 is a flowchart of an exemplary process of detecting machine failures using ML models trained using photorealistic images, according to some embodiments of the present invention;

[0049] FIG. 7 is a schematic illustration of an exemplary system for detecting machine failures using ML models trained using photorealistic images, according to some embodiments of the present invention;

[0050] FIG. 8 is a flowchart of a process of adjusting image capturing parameters of a failure detection system based on analysis of photorealistic images, according to some embodiments of the present invention; and

[0051] FIG. 9 is a schematic illustrating of an exemplary system for generating instructions for adjusting image capturing parameters of a failure detection system based on analysis of photorealistic images, according to some embodiments of the present invention.

[0052] DETAILED DESCRIPTION

[0053] The present invention, in some embodiments thereof, relates to detecting failure modes of mechanical elements using optical sensors, and, more specifically, but not exclusively, to generating photorealistic training sets for training Machine Learning (ML) models to detect failure modes of mechanical elements using optical sensors.

[0054] Applying machine learning (ML) for predictive maintenance based on visual inspection may be highly efficient since ML based models, for example, a Neural Network (NN), a classifier, a statistical a classifier, a Support Vector Machine (SVM), and / or the like may be trained and learned to identify visual failure mode patterns which may exist and / or develop in machinery mechanical elements, components, parts, modules, assemblies, and / or the like, collectively designated elements here in after. However, in order to produce high performing ML models, the ML models must be trained using large training sets (datasets) comprising a plurality of training samples, specifically image data samples reflecting failure modes which are typical, common, and / or estimated to mechanical elements targeted for visual inspection and failure detection.

[0055] As used herein, according to some embodiments, the term “fault” refers to an anomaly or undesired effect or process in the mechanical element and / or component and / or associated elements (such as a machine that the mechanical element is a part of) that may or may not develop into a failure but requires follow-up, to analyze whether any elements should be repaired or replaced. According to some embodiments, the fault may include, among others, a change in length (increase or decrease), structural deformation, surface deformation, a crack, crack propagation, a defect, inflation, bending, wear, corrosion, a change in color, a change in appearance and the like, or any combination thereof.

[0056] As used herein, according to some embodiments of the invention, the term "failure" refers to any problem that may cause the mechanical element and / or associated elements to not operate as intended. In some cases a failure may disable usage of the mechanical element and / or other associated element(s) or even pose a danger to the associated element or user.

[0057] As used herein, according to some embodiments of the invention, the term “failure mode” is to be widely construed to cover any manner in which a fault or failure may occur, such as a tear (partial or complete), a detachment of a mechanical element, a movement or dislocation of an element, a change in shape (e.g. length), a structural deformation, surface deformation, a crack, crack propagation, a defect, inflation, bending, wear, corrosion, a change in color, a change in appearance, and the like, or any combination thereof. It is appreciated that an element may be subject to a plurality of failure modes, related to different characteristics or functionalities thereof.

[0058] Some failure modes may be common to different element types, while others may be more specific to one or more element types. For example, tear failure modes may be relevant to a pulley belt, bending failure modes may be relevant to a pulley axle and a corrosion failure mode may be relevant to a chain.

[0059] Such failure modes may include, for example, mechanical failure modes (e.g., cracks, breaks, tears, erosion, deformations, etc.), chemically induced failure modes (e.g., erosion, corrosion, oxidation, rust, etc.), dynamic failure modes (e.g., movement deviation, movement inaccuracy, etc.) which may be traced to one or more root causes, for example, a design error, a manufacturing defect, a wear condition, an exposure to one or more mechanical, physical, and / or environmental forces, loads, impacts, and / or effects, material fatigue, and / or the like. For example, a failure mode of a mechanical element, according to embodiments of the present invention, may be a crack in the element. Depending on the length of the crack, a fault could be a slight crack, such as a 0.1 mm crack, and a failure could would be a severe crack, such as a 2 cm. crack.

[0060] For example, a failure mode of a mechanical element, according to embodiments of the present invention, may be corrosion in the element. The corrosion fault depending on the area size, color, depth, such as an aluminum plate with 0.1X0.1 mm2area with powdery-grey appearance and with depth of 0.02 mm. A failure could be each of the parameters or a combination of them such as a 10X10 mm2affected area.

[0061] For example, a failure mode of a fluid carrier, according to embodiments of the present invention, may be leakage in the element. The leakage fault may relate to the amount, or rate of liquid that dripped. For example, a leakage of 1 drop per 10 minutes can be considered as a fault and a failure would be 1 drop per 10 seconds.

[0062] As used herein, according to some embodiments of the invention, the term “trend” is to be widely construed to cover any behavior over time of a fault, or a failure mode, when or under what circumstances the fault will turn into a failure. The trend is optionally associated with additional circumstances such as environmental conditions, usage characteristics of the machine, characteristics of a user of a machine, or the like.

[0063] According to some embodiments of the present invention, there are provided methods, systems, devices and computer software programs for generating training samples for training ML models to detect faults associated with failure modes of one or more (target) mechanical elements of one or more machines.

[0064] Specifically, the training samples may comprise photorealistic images which realistically, i.e., accurately, genuinely, truthfully, etc., image and / or reflect the failure modes, or more accurately, image (depict) one or more faults associated with the failure modes. The faults may comprise visual effects, deviations, impacts, and / or the like which may be visible in the photorealistic images, for example, in the target elements in which the failure mode(s) is present, in one or more other elements which are in relation, specifically mechanical relation, with the target element(s), in dynamic behavior of the machine and / or part thereof, and / or the like.

[0065] The photorealistic images may be generated by processing visual media, such as one or more two dimension (2D)) representations and / or (three dimension (3D)) representations of the mechanical element, the machine, and / or part thereof, in correlation with realistic image data of the machine, the element, and / or part thereof, for example, images, video sequences, image maps, and / or the like captured by image sensors (e.g., camera, video camera, Infrared sensor, thermal camera, etc.) deployed to monitor the element, the machine, and / or part thereof.

[0066] The realistic image data may optionally be captured in close proximity to the monitored element(s) such that it may not depict background of the machine which is not part of the machine. Moreover, the realistic image data may optionally be subject to and / or injected with one or more visual degrading effects, for example, reflection, glare, mist, and / or the like.

[0067] The visual media includes 2D representation(s) and / or 3D representation(s), for example, a 3D model, a 3D animation, a 3D simulation , images, video, and / or the like may be or may be created based on one or more digital design files of the element, for example, a two dimension (2D) design file, a 3D design file, and / or the like such as, for example, a Computer Aided Design (CAD) file, and / or the like in which the failure mode(s) is injected. As such the 2D representation(s) and / or the 3D representation(s) may show the fault(s) associated with the injected failure mode(s), optimally including development of the fault(s) into the failure(s) thus showing the trend of failure(s) of the injected failure mode(s).

[0068] One or more generative ML models may be used for processing the visual media depicting the machine and / or part thereof in correlation with at least part of the realistic image data of the element, the machine, the background, and / or the like. The generative ML model(s) may be trained to synthesize, fuse and / or superimpose the failure mode injected visual media with the realistic image data and / or part thereof to produce the photorealistic image(s) depicting the injected failure mode(s).

[0069] Moreover, one or more photorealistic images may be created based on visual media which do not show any faults, i.e., visual media generated based on digital design file(s) relating to target mechanical element(s) which are not adjusted to inject any failure modes. Additional photorealistic images may be then created based on visual media which do show one or more faults associated with one or more failure modes injected into the digital design file(s) used to generate these visual media.

[0070] Optionally, the trend of one or more failure modes may be defined by applying one or more environmental (e.g., temperature, pressure, humidity, precipitation, radiation, chemicals, etc.) and / or operational conditions (heat, air pressure, acceleration, speed, etc.) to the visual media. The visual media applied with such environmental and / or operational conditions may exhibit (show, image, simulate, etc.) the trend of one or more faults associated with failure modes of one or more mechanical elements which may develop into failures due to impact, and / or effect of the applied condition(s) as may happen to corresponding real-world items of the mechanical elements subject to such conditions. The photorealistic images may be included as training samples in training sets used for training the ML model(s) to detect, estimate, and / or predict existence, presence and / or development of the failure modes thus significantly increasing failure detection performance of the ML model(s).

[0071] Generating photorealistic training sets reflecting failure modes of mechanical elements for training ML model(s) to detect, estimate, and / or predict this failure modes in machine items comprising these elements may present major benefits and advantages compared to currently existing failure detection systems and methods.

[0072] First, while imagery data of machine elements targeted for monitoring and failure detection (e.g., images, video sequences, mapping images, etc.) may be captured and available in abundance, images imaging, depicting and / or reflecting failure modes of the target elements may be very difficult to collect. This is because failure modes may be significantly scarce and rare which makes it extremely hard and typically impossible to track and record their associated faults. Moreover, even if imagery data of faults associated with one or more failure modes may be captured, such imagery data may be limited to failure modes which actually developed and occurred in real- world machines. As such, collecting training samples which reflect failure modes is highly limited and typically mostly useless for training ML models to detect failure modes.

[0073] These limitations may be overcome by generating the photorealistic training sets reflecting failure modes to produce and use useful training data for training ML model(s) to detect, estimate, and / or predict presence and / or development of failure modes of mechanical elements.

[0074] Since the faults are injected in digital design file(s) relating to the target element(s) rather than to the element(s) themselves, faults associated with a plurality of failure modes, practically any failure mode, of the target element(s) may be defined and injected to the digital design file(s). These diverse failure modes, specifically their associated faults, may be obviously reflected in the visual media generated based on the adjusted digital design file(s). The faults associated with these diverse failure modes may be thus imaged in the photorealistic images generated using generative ML model(s) adapted to process (e.g., synthesize, fuse, superimpose, etc.) the imagery data of the machine elements, which is available in abundance, in correlation with the visual media reflecting the associated faults. Moreover since realistic image data of the machine and its mechanical elements is highly available, the generative ML model(s) may generate a large, rich and / or diversified training set comprising a plurality of photorealist images reflecting faults associated with one or more of the failure modes in a plurality of viewpoints, under different conditions, and / or the like. Trained with the large, rich, and diversified training set, performance of the failure detection ML model(s), for example, accuracy, reliability, consistency, robustness, and / or the like may be significantly increased.

[0075] Moreover, one or more trends of failures, i.e., development of faults associated with one or more failure modes into failures may be emulated and / or simulated in the visual media by injecting into the digital design file(s) one or more faults which evolve over time. Synthesized photorealistic images generated based on such visual media processed in correlation with realistic image data may therefore image the evolution, development, and / or progress of the injected fault(s) into potential failures, e.g. the trend of the failure mode.

[0076] Therefore, the failure detection ML model(s) trained to detect, estimate and / or predict faults and / or failure modes, optionally while developing (trend), may be highly efficient for Structural Health Monitoring (SHM), Predictive Health Management (PHM), Condition Based Maintenance (CBM), and / or the like to provide accurate analysis of health of the machines and their elements which may be used to accurately and efficiently schedule maintenance sessions when actually needed and preventing unnecessary maintenance activities. Efficient machine maintenance may significantly reduce maintenance costs, resources, and / or time and may also reduce downtime of the machines thus increasing their productivity and service. Moreover, efficient machine maintenance may significantly increase longevity and / or life span of the elements, the machines, and / or part thereof.

[0077] Furthermore, realistically simulating trends of one or more failure modes which may develop due to impact and / or effect of environmental and / or operation conditions the machine and / or part thereof may enable the trained failure detection ML model(s) to learn the impacts and / or effects of these conditions in the formation, development, criticality of faults and / or failures thus further improving performance of the trained failure detection ML model(s).

[0078] In addition, generating photorealistic images which show background of the monitored machine and / or part thereof may increase diversity of the photorealistic images showing background elements which used as training samples may enable the ML model(s) to further adapt and learn to distinguish the mechanical elements from their background thus further improving performance of the failure detection ML model(s).

[0079] Moreover, training the ML model(s) with photorealistic images training samples which contain visual degrading effects may significantly increase the ML model(s) failure detection performance in the presence of such degrading effects.

[0080] According to some embodiments of the present invention, there are provided methods and systems for adjusting or providing instructions to adjust one or more parameters of one or more imaging elements used by a failure detection system to monitor one or more machines in order to detect failure modes in one or more elements of the machine based on analysis of one or more of photorealistic images in order to improve quality of image data depicting the machine and / or part thereof.

[0081] Optionally, at least some of the photorealistic images used to adjust image capturing parameters of the failure detection system may not depict faults but rather show the target mechanical element(s) having no faults, i.e., not injected with any failure mode.

[0082] For example, according to some embodiments, one or more processors are provided to analyze photorealistic image data and provide instructions for adjustment of one or more imaging parameters of images sensors used by the failure detection system. These imaging parameters may include, for example, one or more positioning parameters of the image sensor(s) (e.g., location, position, orientation, viewpoint, etc.), one or more inherent and / or operational parameters of the image sensor(s) (e.g., shutter speed, contrast, zoom, dynamic range, etc.), and / or the like.

[0083] In another example, one or more processors may analyze the photorealistic image data to provide instructions for one or more illumination parameters of one or more light sources adapted to illuminate one or more monitored elements of one or more machines. For example, instructions may include instructions to adjust of illumination intensity, light spectrum, lighting mode, and / or the like based on analysis of images captured to depict the monitored element(s) in order to improve quality of subsequently captured images.

[0084] According to some embodiments of the invention, one or more processors may be used for real time monitoring of fault development. According to some embodiments, the system may input an actual image taken in which a fault is identified. One or more processors may then be used to emulate or simulate one or more trends of failures, i.e., development of the identified fault into failure. The trend may be emulated by injecting into the digital design file(s) one or more faults identified which evolve over time. Synthesized photorealistic images generated based on such visual media processed in correlation with realistic image data may therefore image the evolution, development, and / or progress of the injected fault(s) into potential failures, e.g. the trend of the failure mode.

[0085] Thus, the failure detection ML model(s) may be used to detect, estimate and / or predict development of a fault into failure (trend). This may enable a Structural Health Monitoring (SHM), Predictive Health Management (PHM), Condition Based Maintenance (CBM), and / or the like to provide accurate analysis of health of the machines and their elements and may be used to accurately and efficiently schedule maintenance sessions when actually needed. Furthermore, realistically simulating trends of one or more failure modes may be affected based on environmental conditions inputted. Optionally, the image is provided along with environmental and operational conditions of the machine in which it operates in order to provide accurate prediction of the trend.

[0086] Before explaining at least one embodiment of the invention in detail, it is to be understood that the invention is not necessarily limited in its application to the details of construction and the arrangement of the components and / or methods set forth in the following description and / or illustrated in the drawings and / or the Examples. The invention is capable of other embodiments or of being practiced or carried out in various ways.

[0087] The present invention may be a system, a method, and / or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.

[0088] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0089] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0090] Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages.

[0091] The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.

[0092] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0093] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0094] Referring now to the figures, FIG. 1A is a flowchart of an exemplary process of generating photorealistic images for use as training samples for training ML models to detect machine failures, according to some embodiments of the present invention.

[0095] An exemplary process 100 may be executed to generate one or more training samples for training one or more ML models to detect faults associated with failure modes of one or more (target) mechanical elements of one or more machines.

[0096] In particular, the training samples may comprise photorealistic images generated by processing one or more 3D representations of the element in correlation with realistic image data of the element and / or part thereof. The 3D representation(s), for example, a 3D model, a 3D animation, a 3D simulation, and / or the like may be created based on one or more digital design files of the element, for example, a 2D design file, a 3D design file, and / or the like created using one or more design tools, for example, a CAD file.

[0097] Specifically, the 3D representations of the element(s) are generated based on digital design file(s) adjusted to inject one or more faults of the target element(s) such that the produced photorealistic images depict one or more faults associated with one or more failure mode(s) of the depicted elements.

[0098] The photorealistic images may be used as training samples for training one or more ML models to estimate presence of the failure mode(s) in one or more machine items comprising the target element(s).

[0099] Reference is also made to FIG. 2, which is a schematic illustration of an exemplary system for generating photorealistic images for use as training samples for training ML models to detect machine failures, according to some embodiments of the present invention.

[0100] An exemplary training data generation system 200 may execute the exemplary process 100 for generating photorealistic images used as training samples for training one or more ML model to detect faults associated with failure modes of one or more elements of one or more machines.

[0101] The training data generation system 200, for example, a server, a processing node, a cluster of processing nodes, and / or the like may comprise an Input / Output (I / O) interface 210, a processor(s) 212 for executing the process 100, and a storage 214 for storing data and / or code (program store). The I / O interface 210 may include one or more wired and / or wireless I / O interfaces, for example, a Universal Serial Bus (USB) port, a serial port, a Bluetooth (BT) interface, a Radio Frequency (RF) interface, and / or the like for communicating and / or attaching to one or more external devices and / or attachable devices. The I / O interface may further include one or more network adapters, interfaces, ports, and / or links for connecting to one or more wired and / or wireless networks, for example, a Local Area Network (LAN), a Wireless LAN (WLAN, e.g. WiFi), a Wide Area Network (WAN), a Metropolitan Area Network (MAN), a cellular network, the internet and / or the like.

[0102] Via the TO interface 210, the training data generation system 200 may receive data, for example, realistic image data of one or more elements of one or more machines targeted for monitoring and inspection in order to estimate presence or development of one or more faults associated with one or more failure modes of the target element(s). In another, the training data generation system 200 may receive, via the I / O interface 210, one or more digital design files relating to the target element(s). For example, the training data generation system 200 may retrieve data, for example, realistic image data, digital design files, and / or the like from one or more attachable storage devices, for example, an USB storage media, and / or the like attached and / or connected to the I / O interface 210. In another example, the training data generation system 200 may communicate, over one or more networks via the I / O interface 210, with one or more remote networked resources, for example, a server, a storage device, a database, a cloud service, and / or the like to receive data, for example, realistic image data, digital design files, and / or the like.

[0103] In another example, via the TO interface 210, the training data generation system 200 may receive realistic image data captured by one or more image sensors adapted to capture image data of one or more machines and / or part thereof.

[0104] Moreover, via the I / O interface 210, the training data generation system 200 may output data, for example, one or more synthetic training sets comprising one or more photorealistic images which may be used for training one or more ML models 230, for example, a Neural Network (NN), a Deep Neural Network (DNN), a Support Vector Machine (SVM), a classifier, a statistical classifier, and / or the like. For example, the training data generation system 200 may store data, for example, one or more training sets in a USB storage media, and / or the like attached and / or connected to the I / O interface 210. The USB storage media may be then attached to one or more systems, devices, and / or platforms which may retrieve the stored training dataset(s) and use them for training one or more ML models 230. In another example, the training data generation system 200 may transmit one or more training sets over one or more networks via the I / O interface 210, with one or more remote networked resources which may use the received training set(s) for training one or more ML models 230.

[0105] The processor(s) 212, homogenous or heterogeneous, may include one or more processing nodes and / or cores arranged for parallel processing, as clusters and / or as one or more multi core processor(s).

[0106] The storage 214 may include one or more non-transitory persistent storage devices, for example, a ROM, a Flash array, a Solid State Drive (SSD), a hard drive (HDD), and / or the like. The storage 214 may also include one or more volatile devices, for example, a RAM component, a cache, and / or the like. The storage 214 may further comprise one or more network storage devices, for example, a storage server, a Network Accessible Storage (NAS), a network drive, a database server and / or the like accessible through the I / O interface 210.

[0107] The processor(s) 212 may execute one or more software modules such as, for example, a process, a script, an application, an agent, a utility, a tool, an Operating System (OS), and / or the like each comprising a plurality of program instructions stored in a non-transitory medium (program store) such as the storage 214 and executed by one or more processors such as the processor(s) 212.

[0108] The processor(s) 212 may optionally further, integrate, utilize and / or facilitate one or more hardware elements (modules) integrated and / or utilized in the training system 200, for example, a circuit, a component, an Integrated Circuit (IC), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signals Processor (DSP), a Graphic Processing Unit (GPU), an Artificial Intelligence (Al) accelerator and / or the like.

[0109] The processor(s) 212 may therefore execute one or more functional modules implemented using one or more software modules, one or more of the hardware modules and / or combination thereof. For example, the processor(s) 212 may execute a training data generator 220 for executing the process 100 to generate one or more training sets comprising one or more photorealistic images which may be used for training one or more ML models 230. In another example, the processor(s) 212 may execute one or more CAD applications 222, specifically mechanical CAD applications such as, for example, Solidworks, Siemens NX, Fusion 360, Dassault CATIA, and / or the like. The CAD application(s) 222 may be executed to create, adjust, manipulate and / or process one or more digital data files relating to one or more (target) elements of one or more machine(s) targeted for monitoring and inspection in order to inject one or more failure modes into the target element(s).

[0110] It should be noted, that each of functional modules executed by the processor(s) 212, for example, the training data generator 220, the CAD application(s) 222, and / or the like may be executed by the processor(s) 212 such that any one or more processors of the processor(s) 212 may execute one or more of the functional modules and / or part thereof or optionally not participate in execution of any of the functional modules.

[0111] Optionally, the training data generation system 200, specifically, the training data generator 220 and the CAD application(s) 222 may be utilized by one or more cloud computing services, platforms and / or infrastructures such as, for example, Infrastructure as a Service (laaS), Platform as a Service (PaaS), Software as a Service (SaaS) and / or the like provided by one or more vendors, for example, Google Cloud, Microsoft Azure, Amazon Web Service (AWS) and Elastic Compute Cloud (EC2), IBM Cloud, and / or the like.

[0112] For brevity, the process 100 is described for generating training data samples relating to a single failure mode of a single element of a single machine. This, however, should not be construed as limiting since, as may be apparent to a person skilled in the art, the process 100 may be duplicated, expanded and / or scaled for generating training data samples relating to a plurality of failure modes in each of one or more elements of one or more machines.

[0113] The terms “element” and “machine” as used herein may relate to the element (component) and the machine as expressed in design files while corresponding terms “machine item” and “element item” or “element of machine item” may relate to real-world physical machine and elements produced according to the design files.

[0114] As shown at 102, the process 100 starts with training data generator 220 one or more digital design files, for example, a CAD file and / or the like relating to a (target) element of a machine.

[0115] The digital design file(s) may comprise, for example, one or more two-dimension (2D) design files each comprising one or more design drawings of the element. In another example, the digital design file(s) may comprise one or more 3D design models of the element.

[0116] As described herein before, the training data generator 220 may receive the digital design file(s), for example, from one or more remote networked devices, systems, and / or services via the I / O interface 210. In another example, the training data generator 220 may retrieve the digital design file(s) from one or more attachable storage devices attached to the I / O interface 210.

[0117] As shown at 104, one or more of the digital design file(s) of the element may be adjusted (e.g., manipulated, processed, etc.) or even created to inject one or more failure modes of the element into the element, specifically into a design of the element.

[0118] The failure modes may be traced to one or more root causes, for example, a design error, a manufacturing defect, a wear condition, an exposure to one or more mechanical, electrical, physical, and / or environmental forces, loads, impacts, and / or effects, material fatigue, and / or the like. The failure modes may comprise one or more failure modes typical, common, and / or estimated for the mechanical element. For example, one or more failure modes may relate to one or more mechanical failure modes, for example, a crack, a break, a deformation, and / or the like. In another example, one or more failure modes may relate to one or more chemically induced failure modes, for example, erosion, corrosion, oxidation, rust, and / or the like.

[0119] Moreover, one or more fault associated with failure modes injected into the design of the element may may develop over time comprising a trend of the failure mode. For example, a fault may be injected into a model of the element and simulated over a certain time period during which the trend of failure may be seen as one or more injected faults associated with the failure mode may develop into one or more failures. Optionally, the trend of development depends on environmental and / or operational conditions provided.

[0120] The digital design file(s) may be adjusted automatically, for example, by the training data generator 220 using the CAD application(s) 222. In another example, the digital design file(s) may be adjusted manually by one or more users, for example, an engineer, a mechanical designer, and / or the like using the CAD application(s) 222.

[0121] Alternatively, rather than receiving the digital design file(s) and adjusting them as described in steps 102 and 104 respectively, the training data generator 220 may receive, in step 102, one or more digital design files which are already adjusted to inject one or more failure modes into the design of the element. In such case, the training data generator 220 may skip step 104.

[0122] Optionally, one or more of the digital design file(s) received in step 102 may comprise one or more adjusted files in which one or more failure modes are already injected in the design of the element, and the training data generator 220 may further adjust the received adjusted digital design file(s) to inject one or more additional failure modes into the design of the element.

[0123] As shown at 106, one or more 2D or 3D animations depicting the element and / or part thereof may be created (generated) based on the adjusted digital design file(s). Optionally, the 3D animation(s) may depict the machine and / or part thereof comprising the element.

[0124] The 2D or 3D animation(s) may comprise a sequence of a plurality of images depicting the element and / or part thereof, typically while moving according to their designated operation in the machine.

[0125] The 3D animation(s) may be created automatically, for example, by the training data generator 220 using the CAD application(s) 222. In another example, the 3D animation(s) may be created manually by one or more users using the CAD application(s) 222. Alternatively, and / or additionally, the training data generator 220 may receive, for example, in step 102, one or more 3D animations created based on one or more of the digital design file(s) adjusted to inject the failure mode(s) into the element. In such case, steps 104 and 106 of the process 100 may be skipped.

[0126] It should be noted that creating 3D animations based on digital design files may be performed using one or more methods known in the art.

[0127] The 2D or 3D animation(s) which are created using the adjusted digital design files may depict one or more faults associated with one or more of the failure mode(s) injected into the design of the element, interchangeably designated failure injected element. Moreover, the 2D or 3D animation(s) may depict one or more trend of failure modes, in which one or more faults may develop, progress, and / or evolve into failures of the associated failure modes. For example, a certain 2D or 3D animation may depict a trend of a certain failure mode of a certain element in which a certain fault, for example, a small crack (e.g., 0.01 mm) gradually develops into an associated failure, for example, a major crack (e.g., 2 mm).

[0128] Optionally, the trend of one or more failure modes may be defined by applying one or more environmental and / or operational conditions to the 2D or 3D animation(s). The environmental conditions may include, for example, temperature, pressure (height, depth), humidity, precipitation, exposure to radiation, chemicals, and / or the like. The operational conditions may include, for example, machine heat, air pressure, acceleration, speed, and / or the like. One or more of the environmental and / or operational conditions may be obtained, for example, from sensory data captured by one or more sensors, for example, a temperature sensor, a pressure sensor, a humidity sensor, a vibration sensor, a sound sensor, and / or the like deployed in one or more machine items comprising mechanical element items which are at least partially similar to the target element, i.e., share one or more structural, dynamic, and / or design features. In another example, one or more of the environmental and / or operational conditions may be obtained from one or more simulations conducted to simulate the mechanical element, and / or an at least partially similar element in one or more operational environments.

[0129] When the mechanical element is subject to the environmental and / or operational condition(s), these conditions may affect, for example, cause, increase, enhance, and / or escalate one or more faults associated with failure mode(s) injected to the element and / or the trend of the failure mode(s), i.e., development of the fault(s) into failure(s). As such, the 3D animation(s) applied with these conditions may simulate, and exhibit one or more of these faults which may develop into failure due to impact, and / or effect of the applied condition(s).

[0130] For example, reference is now made to FIG. IB wherein an exemplary process for modeling and analyzing trend-based fault development is illustrated. This embodiment allows for simulating how faults associated with injected failure modes progress over time, with the progression trend dependent on environmental and operational conditions.

[0131] As shown in 151, visual media such as a digital design file or set if images of a target mechanical element is received. In 152, one or more failure modes are injected into the design by adjusting the digital design file. This creates an initial fault condition in the modeled element.

[0132] In step 153, a set of environmental and operational conditions are defined. These may include factors such as temperature, humidity, vibration levels, load forces, operating speed, duty cycle, etc. Multiple condition sets may be created to model different scenarios.

[0133] In 154 a series of 3D representations is generated, showing the progression of the injected fault over time under the defined conditions. This may be accomplished through physics-based simulation, finite element analysis, or other suitable modeling techniques. The same initial fault may develop differently depending on the applied conditions. For example, fault progression under nominal conditions, with a crack slowly growing over 1000 operating hours may be different from accelerated progression under high temperature and vibration, with critical failure occurring after only 500 hours and / or arrested crack growth when operating speed is reduced.

[0134] This capability enables predictive trend analysis for different operating scenarios. As shown in 155, photorealistic images may be generated from the 3D representations to create synthetic datasets for training ML models to recognize various fault progression patterns.

[0135] The generation of the photorealistic images may be implemented using a combination of physics-based simulation software and machine learning algorithms. For example, finite element analysis (FEA) tools such as ANSYS or COMSOL Multiphysics may be used to model the structural behavior of the element under various conditions and fault scenarios. To generate the temporal progression of faults, a time-stepping approach may be employed. At each time step, the simulation updates the state of the element based on the applied loads, environmental conditions, and current fault state.

[0136] Machine learning techniques, particularly deep learning models such as convolutional neural networks (CNNs) or graph neural networks (GNNs), may be utilized in the simulation process. These models can be trained on a dataset of simulated fault progressions to learn the underlying patterns and physics. Once trained, they can quickly predict fault progression for new scenarios, allowing for real-time analysis.

[0137] For generating photorealistic images from the simulation data, advanced rendering techniques such as ray tracing or path tracing may be used. These can be implemented using graphics libraries like OpenGL or specialized rendering engines like V-Ray or Arnold. In some embodiments of the present invention, the process is used for real-time industrial monitoring and the system accepts actual image(s) of equipment with a detected fault, along with current environmental / operational data, as inputs (156). The trained ML model then projects the expected fault progression trend based on the specific operating conditions (157).

[0138] This prognostic capability may allow for optimized maintenance planning. As indicated in 158, the system may output maintenance instructions and timelines tailored to the projected degradation rate. Alternatively, it may suggest modified operating parameters (e.g. reduced speed or load) to extend the safe operating window until scheduled maintenance can occur.

[0139] By enabling condition-based predictive maintenance, equipment uptime and availability may be improved while unnecessary interventions are reduced. The ability to model fault progression under varying conditions also supports what-if analysis for optimizing operating procedures and environments.

[0140] The faults depicted in the 3D animation(s) may be visible in one or more modes, manners, and / or effects, whether directly in the element in which the failure mode(s) is injected, and or in one or more other elements which may be in relation, for example, mechanical relation, induced relation, and / or the like with the failure injected element.

[0141] For example, one or more failure modes injected into the digital design file(s) of the element may be associated with corresponding faults directly visible in the element which is depicted at least partially in the 3D animation(s). For example, assuming a certain failure mode comprises a crack in the element, an associated fault may comprise a corresponding crack visible in the element in the 3D animation(s). Another failure mode may be corrosion in the element. The corrosion fault depending on the area size, color, depth, such as an aluminum plate with 0.1X0.1 mm2 area with powdery-grey appearance and with depth of 0.02 mm. A failure could be each of the parameters or a combination of them such as a 10X10 mm2 affected area. Another failure mode is a leakage in the element. The leakage fault may relate to the amount, or rate of liquid that dripped. For example, a leakage of 1 drop per 10 minutes can be considered as a fault and a failure would be 1 drop per 10 seconds

[0142] In another example, one or more failure modes injected into the digital design file(s) of the element may be associated with corresponding faults directly visible one or more other elements which are at least partially depicted 3D animation(s). For example, assuming a certain failure mode comprises a deformation of the element, an associated fault may comprise a deformation, alteration, and / or the like visible in one or more other element depicted at least partially in the 3D animation(s) which are affected by the certain failure mode of the element. In another example, one or more failure modes injected into the digital design file(s) of the element may be associated with faults which are visible in the 3D animation(s) through altered operation and / or altered dynamic behavior of one or more elements of the machine which, whether the failure injected element itself and / or one or more other elements which are in relation, specifically mechanical relation with the failure injected element. For example, assuming a certain failure mode comprises a deformation, corrosion, and / or material fatigue of a bolt element securing another element, for example, an element having a certain dynamic mode in the machine, for example, a rotor, a push-pull rod, and / or the like. In such case, one or more faults associated with the certain failure mode may comprise an altered dynamic behavior of the other element, for example, a rotor spinning around an axis deviating from a normal (designed) axis, a push-pull rod which leans to one side while moving, a landing gear deviating from its normal deployment process, and / or the like In another example, one or more failure modes injected into the digital design file(s) of the element may be associated with faults which are visible in the visual media through altered operation and / or altered dynamic behavior of one or more elements of the machine, whether the failure injected element itself and / or one or more other elements which are in relation, specifically mechanical relation with the failure injected element. It is important to clarify that the injected failure can be in the dynamic movement of the element itself, not just in its static properties.

[0143] Examples of injected failures in dynamic movement include a bearing with increased friction due to wear, causing a shaft to rotate more slowly or with irregular speed, a gear with chipped teeth, resulting in a jerky or inconsistent rotational movement, a hydraulic cylinder with a leaking seal, leading to non-smooth extension or retraction, a robotic arm joint with loosened components, causing vibration or wobble during movement, a conveyor belt with uneven wear, resulting in off- center tracking or irregular speed, a turbine blade with minor deformation, leading to imbalance and vibration during rotation, a crane cable with partial fraying, causing uneven lifting or swaying of loads, and an elevator with worn guide rails, resulting in subtle shaking or non-smooth vertical movement. These examples illustrate how injected failures can manifest in the dynamic behavior of machine elements, providing fault scenarios in the analysis.

[0144] Moreover, since one or more faults injected into the element may develop over time, i.e., trend of failure mode, one or more faults associated with such evolving (trending) failure modes may also develop over time accordingly and may be therefore characterized by a changing, trending, and / or transitional appearance which may change and typically escalate over time. As shown at 108, visual media, comprising a plurality of images depicting fault(s) associated with the failure mode(s) injected into the element, may be generated based on the visual media depicting the element and / or part thereof, specially depicting the fault(s).

[0145] In particular, the visual media may depict the machine and / or part thereof such that the fault(s), visible via one or more elements of the machine, is depicted in the visual media. Moreover, the visual media may comprise a plurality of images depicting the fault(s) during a certain time period, for example, several seconds (e.g., 1,2, 3, 5 seconds, etc.), a few minutes (e.g., 1, 3, 5, minutes, etc.), or even more, for example, 15 minutes, 30 minutes, 60 minutes, and / or the like.

[0146] Furthermore, the visual media may depict the element over significantly prolonged time periods, such as days, weeks or even months, and may thus depict development of one or more of the fault(s) into failure(s). This temporal progression enables the prediction of fault trends and allows for precise indication of when a fault is likely to develop into a failure.

[0147] By generating visual media that captures fault progression over various time scales, from seconds to months, the system enables accurate trend prediction. This predictive capability allows determining the optimal timing for maintenance interventions. Based on the projected fault progression, the system can provide tailored maintenance instructions, specifying not only what maintenance is required but also when it should be performed to prevent failure while minimizing unnecessary downtime.

[0148] This approach allows for proactive, condition-based maintenance planning, where interventions are scheduled based on the actual degradation rate of the element rather than fixed time intervals. By accurately predicting when a fault will develop into a failure, maintenance can be optimized to maximize equipment uptime and minimize the risk of unexpected breakdowns.

[0149] The 3D graphic presentation(s) may be generated, for example, by the training data generator 220 using one or more graphic engines, graphic accelerators, and / or the like. Alternatively, and / or additionally, the training data generator 220 may receive, for example, in step 102, one or more 3D graphic presentations generated based on respective 3D animation(s) to depict the fault(s) associated with failure mode(s) injected into the element. In such case, steps 104, 106 and 108 of the process 100 may be skipped.

[0150] It should be noted that generating visual media based on 3D animations may be performed using one or more methods known in the art.

[0151] Optionally, step 108 may be performed first before injecting any failure mode(s) into the digital design file(s) in order to generate photorealistic images which do not show fault(s) associated with such failure mode(s). The digital design file(s) may be then adjusted to inject one or more failure modes of the target mechanical element and one or more further (other) photorealistic images may be generated which do depict one or more faults associated with the injected failure mode(s).

[0152] As shown at 110, the training data generator 220 may receive realistic image data, for example, one or more 2D and / or 3D images, panoramic images and / or the like depicting the element, the machine, a background of the element, a background of the machine, and / or part thereof.

[0153] The received realistic image data may comprise realistic (real-world) images, frames, and / or of the target element. However, optionally, rather than realistic image data of the specific target element, the realistic image data may comprise realistic (real-world) images, frames, and / or the like of one or more elements similar to the target element which have similar, common, and / or resembling mechanical features shared with the element. In another example, the realistic image data may comprise realistic images of one or more elements not necessarily similar to the target element(s) but which share common environmental features with the target element(s), for example, common deployment location in machine(s), common background, and / or the like.

[0154] As described herein before, the training data generator 220 may communicate with one or more remote networked devices, systems, and / or services over one or more networks via the I / O interface 210 to receive the realistic image data. In another example, the training data generator 220 may retrieve the realistic image data from one or more attachable storage devices attached to the I / O interface 210. In another example, the training data generator 220 may communicate, via the I / O interface 210, with one or more image sensors deployed to monitor one or more machine items and / or comprising the element to receive the realistic image data.

[0155] According to some embodiments, at least part of the realistic image data relating to the element and / or the machine may comprise real-world realistic image data captured by one or more image sensors deployed to capture image data of one or more machine items comprising the element, i.e., real-world machines comprising the element produced according to the digital design file(s).

[0156] The image sensor(s), for example, a camera, a video camera, an Infrared camera, an Ultra Violet (UV) camera, a thermal camera, a stereoscopic camera, a Light Imaging, Detection and Ranging (LiDAR) sensor, and / or the like may be deployed in close proximity to the element such that it may effectively monitor the element and / or part thereof.

[0157] The image sensor(s) may be designed and produced to have a significantly small form factor, while typically having a wide view angle, such that the image sensor(s) may be easily deployed in small spaces in close proximity to monitored elements and / or parts of the machine, for example, between a few centimeters to approximately two meters, and preferably less than one meter.

[0158] One or more of the image sensor(s) may be adapted to capture realistic image data of the element, the machine item and / or part thereof while the machine item and / or part thereof are moving. For example, a certain image sensor deployed to capture realistic image data of a certain dynamic element, for example, a rotor, a piston, a push-pull rod, and / or the like of a certain machine, may be mechanically coupled to another element or to a static part of the certain dynamic element such that the certain image sensor may capture image data of the certain dynamic element while the certain dynamic element is moving and the certain image sensor is static and not moving.

[0159] However, one or more of the image sensor(s) may be adapted to capture realistic image data of the element, the machine and / or part thereof while both the image sensor(s) and the machine item and / or part thereof are moving. For example, a certain image sensor deployed to capture realistic image data of a certain dynamic element of a certain machine, may be mechanically coupled to the certain dynamic element. The image sensor(s) may be adapted to capture image data of the certain dynamic element while the certain dynamic element is moving such that both the certain dynamic element and the mechanically coupled certain image sensor are moving.

[0160] The realistic image data depicting the element, the machine, a background of the element, a background of the machine, and / or part thereof may comprise one or more visual degrading effects, for example, reflection, glare, mist, and / or the like. The sensor capturing the realistic image data may also capture images that depict actual faults present in the machine or element. These faults may be in various stages of development, from incipient to severe. For instance, the sensor might capture an image showing a small crack in a component, surface corrosion, unusual wear patterns, or even dynamic anomalies such as irregular movement or vibration

[0161] Optionally, the realistic image data captured by the image sensor(s) may be processed, for example, by the training data generator 220, to reduce and possibly remove one or more of the visual degrading effects. Optionally, the realistic image data captured by the image sensor(s) may be processed, for example, by the training data generator 220, to inject, and / or increase one or more degrading effects in the realistic image data in order to later produce one or more photorealistic images which contain degrading effects.

[0162] ML model(s) trained using the such versatile synthetic training dataset comprising photorealistic images which may comprise such visual degrading effects, either inherently present and / or injected synthetically through processing of the photorealistic image data, may therefore learn to detect faults in the presence as well as in absence of these visual degrading effects thus significantly increasing their failure detection performance. The training data generator 220 may employ one or more methods, techniques, algorithms, and / or tools to process the realistic image data and reduce and / or remove the degrading effects. For example, the training data generator 220 may apply one or more ML model(s) adapted and trained to process image data to reduce and possibly remove one or more of the degrading effects. Such ML model(s), for example, a neural network, a DNN, an SVM, and / or the like may be trained to identify one or more of the degrading effects in image data, specifically in image data of real- world items of the element and / or real-world data of machine items comprising the element. In another example, the training data generator 220 may use one or more image processing tools, and / or algorithms, for example, compare vision, visual filters, and / or the like to reduce and possibly remove one or more of the degrading effects.

[0163] Optionally, one or more parameters of one or more of the image sensors deployed to monitor one or more element items, one or more machines comprising the element and / or part thereof the machine, may be adjusted to improve quality of subsequently captured realistic image data, for example, reduce and potentially remove one or more of the degrading effects, improve focus and / or visibility of the monitored element(s), and / or the like. In particular, the parameters of one or more of the image sensor(s) may be adjusted based on analysis of one or more of the images captured by the image sensor(s).

[0164] For example, one or more positioning parameters of the image sensor(s), for example, location, position, orientation, viewpoint, and / or the like may be adjusted according to analysis of the realistic image data in order to improve the quality of subsequently captured realistic image data. In another example, one or more inherent and / or operational parameters of the image sensor(s), for example, shutter speed, contrast, zoom, dynamic range, and / or the like may be adjusted according to analysis of the realistic image data in order to improve the quality of subsequently captured realistic image data.

[0165] Moreover, optionally, based on analysis of the realistic image data, one or more illumination parameters, for example, intensity, light spectrum and / or wave length (e.g., visible light, infrared, ultraviolet,, etc.), illumination mode (e.g., wide angle lighting, focused light beam, strobe illumination, etc.), and / or the like may be adjusted when capturing subsequent realistic image data in order to further improve the quality of the subsequently captured realistic image data.

[0166] For example, assuming that based on analysis of realistic data depicting a certain element, it is identified that intensity of certain light source illuminating the element is low such that realistic image data capturing the element may have reduced quality. In such case, the intensity of the light source may be adjusted, for example, increased thus improving quality of subsequently captured realistic image data depicting the dynamic element. In another example, assuming that based on analysis of realistic data depicting a certain dynamic element which moves significantly fast, it is identified that a frequency of a strobe light projected to illuminate the dynamic element is inefficient for efficiently illuminating the dynamic element and thus the realistic image data capturing the dynamic element has reduced quality. In such case, the frequency of a strobe light may be adjusted, for example, to align with a movement frequency of the dynamic element which may improve the illumination conditions and produce improved quality realistic image data subsequently captured for the dynamic element.

[0167] According to some embodiments, at least part of the realistic image data relating to the element and / or the machine may comprise realistic image data generated using one or more generative ML models adapted to generate synthetic 3D images of the element. The generative ML model(s) may be adapted and trained to generate synthetic 3D images of the element based on one or more source data items, for example, one or more of the digital design files relating to the element and / or to the machine and / or the environment in which the machine operates, one or more real-world images of the element, and / or the like, and / or a combination thereof.

[0168] Using generative Al to produce realistic image data may be performed using one or more methods known in the art.

[0169] As shown at 112, the training data generator 220 may generate one or more photorealistic images imaging fault(s) associated with the failure mode(s) injected into the (design of) element.

[0170] To this end, the training data generator 220 may process the visual media depicting the machine and / or part thereof in correlation with at least part of the realistic image data of the element, the machine, the background, and / or the like. For example, the training data generator 220 may synthesize and / or fuse the visual media with the realistic image data and / or part thereof to produce the photorealistic image(s). In another example, the training data generator 220 may superimpose the realistic image data and / or part thereof over the visual media to produce the photorealistic image(s).

[0171] Optionally, the training data generator 220 may generate the photorealistic images using one or more generative ML models adapted and trained, using one or more methods known in the art, to process the visual media in correlation the realistic image data.

[0172] In order to achieve high quality, accuracy, and / or fidelity of the photorealistic images, the training data generator 220 may identify one or more reference points in the visual media and the realistic image data. The training data generator 220 may then align, and / or register the realistic image data with the visual media based on a common coordinate system defined according to corresponding reference points identified in the visual media and in the realistic image data. Moreover, the training data generator 220 may further adjust, process and / or manipulate the visual media and / or the realistic image data, for example, rotate, transform, increase and / or decrease (zoom in / out), and / or the like in order to bring them into a common scale and / or proportion.

[0173] Reference is now made to FIG. 3, which presents a photorealist image of a first exemplary mechanical element generated by processing a visual media of the element in correlation with image data of the element, according to some embodiments of the present invention. Reference is also made to FIG. 4, which presents a photorealistic image of a second exemplary mechanical element generated by processing a visual media of the element in correlation with image data of the element and its exemplary background, according to some embodiments of the present invention.

[0174] Illustration 300 shows an image extracted from a visual media of an exemplary element, for example, a push-pull rod 310. As described herein before, the visual media is generated based on one or more digital design files of the push-pull rod 310. As seen the push-pull rod 310 may comprise a bolt 312 which is displayed as modeled based on the digital design file(s)

[0175] Illustration 302 shows a photorealistic image presenting the push-pull rod 310 with the bolt 312 as it may actually look in real-world images. The photorealistic image 302 may be generated by a training data generator such as the training data generator 220 by processing the modeled push-pull rod 310 of image 300 and the bolt 312 in correlation with realistic image data of the bolt 312 or a similar element.

[0176] Illustration 400 shows a realistic image of part of a real-world aircraft wing, specifically a section of a steering mechanism of the aircraft comprising a push-pull rod. Illustration 402 shows an image extracted from a visual media of an exemplary element, for example, a push-pull rod 410. As described herein before, the visual media is generated based on one or more digital design files of the push-pull rod 410.

[0177] Illustration 404 shows a photorealistic image presenting an exemplary aircraft steering mechanism including the push-pull rod 410 as it may actually look in real-world images. The photorealistic image 404 may be generated by the training data generator 220 by processing the modeled push-pull rod 310 extracted from the 3D graphical presentation in correlation with the realistic image 400.

[0178] Reference is now made to FIG. 5 A and FIG. 5B, which present photorealistic images of an exemplary fault associated with a failure mode of an exemplary element of a machine where the failure mode is injected in a digital design file of the element used to create a visual media of the element which is used to generate the photorealist image, according to some embodiments of the present invention.

[0179] Illustration 500 shows an image extracted from a visual media of an exemplary element, for example, a metal plate 510 with one or more bolts 512 secured in holes in the plate 510. The visual media is generated based on one or more digital design files of the metal plate 510 and the bolts 512.

[0180] Moreover, image 500 further shows an exemplary fault 520, namely a crack associated with an exemplary failure mode of the metal plate 510, for example, a crack, a break, and / or the like. The failure mode may be injected in the modeled metal plate as seen in image 500 by adjusting one or more of the digital design files of the metal plate 510.

[0181] Illustration 502 shows a realistic image presenting the metal plate 510 and several bolts 512 depicting real- world plate and bolts.

[0182] Exemplary photorealistic image 504, 506 and 508 may be generated by the training data generator 220 by processing the modeled metal plate 510 and bolts 512 in correlation with the realistic image data of the metal plate 510 extracted from the image 502.. The training data generator 220 may apply one or more generative ML models to process the realistic image data of the metal plate 510 to insert the crack 520 which may typically not be present real-world metal plates 510 specifically not in the exact location and / or outline of the failure mode injected into the modeled metal plate 510 as shown in image 500.

[0183] As seen in the photorealistic 504 which may be captured at an early stage of the failure mode, the fault, i.e., the crack 520 is not visible since the fault may be in its initial development stage As seen in the photorealistic image 506, the crack 520 is developing and is visible as a small crack 520 in the plate 510. As seen in the photorealistic image 508, which may be captured while the fault is maturing, the crack 520 has significantly increased.

[0184] Reference is made once again to FIG. 1A.

[0185] As shown at 114, the training data generator 220 may provide, for example, output the photorealistic image(s) to one or more training systems, devices, and / or services adapted to train one or more ML models 230 using the photorealistic image(s) as training samples to detect one or more failures in one or more machine items comprising the (target) element.

[0186] For example, the training data generator 220 may transmit, via the I / O interface 210, the photorealistic image(s) to one or more remote networked training systems, devices, and / or services adapted to train the ML model(s) 230. In another example, the training data generator 220 may store the photorealistic image(s) in one or more attachable storage devices attached to the I / O interface 210 which may be then detached and used to transfer the photorealistic image(s) to one or more training systems, devices, and / or services.

[0187] Optionally, the ML model(s) 230 may be trained by the training data generation system 200 itself. In such case, the training data generator 220 may locally store the photorealistic image(s), for example, in the storage 214 from which it may be retrieved by one or more functional modules executed by the training data generation system 200 for training the ML model(s) 230.

[0188] As shown at 116, the ML model(s) 230 may be trained, for example, by a training system, using one or more training sets comprising the generated photorealistic image(s), typically among a plurality of additional training samples.

[0189] In particular, the ML model(s) 230 may be trained, using the photorealistic image(s) among the other training samples, to detect, estimate, and / or predict one or more failures in one or more machine items comprising the (target) element by estimating presence of the failure mode(s) in the target element. As such, using the photorealistic image(s), the ML model(s) 230 may be trained to detect and / or estimate presence of the failure mode(s) injected into the digital design file(s) relating to the target element as described in step 104.

[0190] For example, the ML model(s) 230 may be trained to compute a probability score indicative of a probability of presence of one or more of the failure modes in the machine item, specifically in one or more target elements included in the respective machine item. In another example, the ML model(s) 230 may be trained to estimate severeness of one or more faults detected in the machine item and may optionally compute a severeness score indicative a level (e.g., in a scale of 1-10, 10-100, etc.) of severeness of each of one or more detected faults. For example, severeness of one or more faults may be computed according to a level, phase, and / or stage of a trend of their associated failure mode, i.e., how far is the respective fault from becoming a failure and / or what and how operational and / or environmental conditions affect the trend.

[0191] The training samples including the photorealistic images may be used for training the ML model(s) 230 in one or more training sessions, supervised (with labeled samples), unsupervised (with unlabeled samples), semi- supervised, and / or combination thereof. During training, the training set(s) which comprise the including the photorealistic images may be further used for validating and / or testing the ML model(s) after trained.

[0192] For example, a first subset of the training set(s) may be used for training the ML model(s), a second subset of the training set(s) may be used for validating the ML model(s), and a third subset of the training set(s) may be used for testing the ML model(s). Moreover, the subsets each of the subsets may comprise a respective exclusive group of plurality of training samples such that each training sample may be included in only one of the subsets and the subsets may therefore not overlap. This may significantly reduce and potentially prevent overfitting of the trained ML model(s) to any specific representation of the element and / or fault(s) associated with failure mode(s) of the element.

[0193] For example, as part of validation, one or more thresholds may be defined for the output of the trained ML model(s) which may define a decision point for positive or negative estimation of presence and / or development the failure mode(s). As such one or more of the scores computed by the trained ML model(s), for example, the probability score, the severeness score, and / or the like, may be compared to the threshold(s) to determine whether the failure mode(s) is estimated or not. the validation subset, i.e., the second subset may be therefore optionally used to defined the threshold(s) according to one or more performance requirements of the ML model(s), for example, accuracy, precision, recall, and / or the like.

[0194] The third subset may be then used for testing the trained ML model(s) and its performance applied with the threshold(s) defined and / or computed in the validation phase.

[0195] As shown at 118, the trained ML model(s) 230 may be provided and / or output to one or more systems, services, and / or the like, for example, a health monitoring system, a maintenance monitoring system, a failure detection system, a predictive failure system, and / or the like adapted to monitor one or more machine items and track, predict and / or estimate potential failures in the machine items.

[0196] For example, the trained ML model(s) 230 may be transmitted by the training system, via one or more networks, to one or more failure detection systems, devices, and / or services adapted to detect failures in one or more machine items comprising the target element. In another example, the trained ML model(s) 230 may be stored by the training system in one or more attachable storage devices which may be used to transfer the trained ML model(s) 230 to one or more failure detection systems.

[0197] Reference is now made to FIG. 6, which is a flowchart of an exemplary process of detecting machine failures using ML models trained using photorealistic images, according to some embodiments of the present invention. Reference is also made to FIG. 7, which is a schematic illustration of an exemplary system for detecting machine failures using ML models trained using photorealistic images, according to some embodiments of the present invention.

[0198] An exemplary process 600 for detecting machine failures using one or more trained ML models such as the ML model 230 may be executed by an exemplary failure detection system 700, for example, a health monitoring system, a maintenance monitoring system, a failure detection system, a predictive failure system, and / or the like adapted to monitor one or more machine items in order to track, predict and / or estimate failures in one or more elements (items) 704 of the machine items.

[0199] The failure detection system 700, for example, a server, a processing node, a cluster of processing nodes, and / or the like may comprise an I / O interface 710 such as I / O interface 210, a processor(s) 712 such as the processor(s) 212 for executing the process 600, and a storage 714 such as the storage 214 for storing data and / or code (program store).

[0200] Via the I / O interface 710, the failure detection system 700 may receive one or more images, video sequences, thermal images, and / or the like, collectively designated images, of one or more elements of one or more machine items captured by one or more image sensors 702, for example, a camera, a video camera, an Infrared camera, a thermal camera, a stereoscopic camera, and / or the like.

[0201] The processor(s) 712 may execute one or more software modules such as, for example, a process, a script, an application, an agent, a utility, a tool, an OS, and / or the like each comprising a plurality of program instructions stored in a non-transitory medium (program store) such as the storage 714 and executed by one or more processors such as the processor(s) 712. The processor(s) 712 may optionally further, integrate, utilize and / or facilitate one or more hardware elements (modules) integrated and / or utilized in the failure detection system 700, for example, a circuit, a component, an IC, an ASIC, a FPGA, a DSP, a GPU, an Al accelerator and / or the like.

[0202] The processor(s) 712 may therefore execute one or more functional modules implemented using one or more software modules, one or more of the hardware modules and / or combination thereof. For example, the processor(s) 712 may execute a failure detector 720 for executing the process 600 to detect, estimate, and / or predict one or more failures in one or more elements 704 of one or more machine items. It should be noted, that each of functional modules executed by the processor(s) 712, for example, the failure detector 720and / or the like may be executed such that any one or more processors of the processor(s) 712 may execute one or more of the functional modules and / or part thereof or optionally not participate in execution of any of the functional modules.

[0203] Optionally, the failure detection system 700, specifically, the failure detector 720 may be utilized by one or more cloud computing services, platforms and / or infrastructures such as, for example, laaS, PaaS, SaaS and / or the like provided by one or more vendors, for example, Google Cloud, Microsoft Azure, Amazon AWS and EC2, IBM Cloud, and / or the like that may communicate with the image sensor(s) 702 via one or more networks to receive the images of the machine item(S) and / or part thereof. Optionally, one or more light sources 706 may be deployed to illuminate one or more of the element 704. The light source(s) 706 may be adapted to project light according to one or more illumination parameters which may be optionally adapted dynamically. The illumination parameters may include, for example, one or more illumination modes, and / or techniques, for example, continuous light, periodic light, probe lighting, focused light beam, wide angle lighting, flood lighting, and / or the like. In another example, the illumination parameters may include a spectral region (wave length), i.e., spectrum of the projected light, for example, visible light, Infrared, UV, and / or the like. In another example, the illumination parameters may include one or more illumination attributes, for example, an intensity a projection angle, and / or the like.

[0204] One or more of the light source(s) 706 may be optionally synchronized with one or more of the image sensor(s) 702 such that the light source(s) 706 may project light while, typically starting at least slightly before, the image sensor(s) 702 capture images of the element(s) 704.

[0205] For brevity, the process 600 is described for monitoring and detecting one or more faults associated with one or more failure modes of a single element 704 of a single machine. This, however, should not be construed as limiting since, as may be apparent to a person skilled in the art, the process 600 may be duplicated, expanded and / or scaled for monitoring and detecting faults associated with failure modes of a plurality of elements 704 of a single machine.

[0206] As shown at 602, the process 600 starts with the failure detector 720 receiving one or more images depicting at least partially one or more element of a machine including a target element 704 having one or more potential failure modes which are targeted for detection, estimation, and / or prediction of the machine's health.

[0207] The images may be captured by one or more image sensors 702 deployed to monitor one or more elements of the machine. In particular, the image sensor(s) 702 may be deployed in close proximity to the element 704 such that it may effectively monitor the element 704 and / or part thereof.

[0208] To this end, the image sensor(s) 702 may optionally have a significantly small form factor, and further optionally a wide view angle, such that the image sensor(s) 702 may be deployed in small spaces in close proximity to monitored element(s) 704 and / or parts of the machine, for example, a steering mechanism, a landing gear compartment, a rotor base, a piston, and / or the like.

[0209] Since the monitored element 704 may be dynamic, meaning that it may move during operation of the machine, the image sensor(s) 702 may be adapted to capture one or more images of the element 704, the machine item and / or part thereof while the machine item and / or part thereof are moving. For example, a certain image sensor 702 deployed to capture images of a certain dynamic element 704, for example, a push-pull rod, a rotor, a piston, and / or the like of a certain machine item, may be mechanically coupled to another element and / or part of the machine item, and / or outside the machine which is static such that the certain image sensor 702 may capture images of the certain dynamic element 704 while the certain dynamic element 704 is moving and the certain image sensor 702 is static.

[0210] In another example, one or more of the image sensor(s) 702 deployed to monitor the element item(s) 704 may be adapted to capture images of the element 704, the machine and / or part thereof while both the image sensor(s) 702 and the element 704 are moving. For example, a certain image sensor 702 deployed to capture images data of a certain dynamic element 704 of a certain machine, may be mechanically coupled to the certain dynamic element 704 and adapted to capture images of the certain dynamic element 704 while both the certain dynamic element 704 and mechanically coupled certain image sensor 702 are moving.

[0211] Optionally, the one or more of the images of the element(s) 704 captured by the image sensor(s) 702 may be processed, for example, by the failure detector 720, to reduce and possibly remove one or more degrading effects, for example, reflection, glare, mist, and / or the like in attempt to enhance quality, reliability, and / or robustness of the realistic image data. The failure detector 720 may employ one or more methods, techniques, algorithms, and / or tools to process the realistic image data and reduce and / or remove the degrading effects, for example, one or more ML model(s) adapted and trained to process image data to reduce and possibly remove one or more of the degrading effects, image processing tools, algorithms, and / or the like.

[0212] As shown at 604, the failure detector 720 may apply one or more of the trained ML model(s) 230 to analyze the received image(s) depicting the element 704, the machine, and / or part thereof.

[0213] The ML model(s) 230 trained to detect, estimate, and / or predict presence of one or more failure modes of the element 704 based on detection of one or more of the faults associated with failure mode(s) may therefore determine, estimate, and / or predict whether one or more of the associated fault(s) are detected in the received images and predict accordingly whether the one or more of the failure modes exist and / or develop in the element 704.

[0214] As shown at 606, which is a conditional step, the failure detector 720 may proceed execution of the process 600 according to an output of the ML model(s) 230. Optionally, if a fault is detected, the failure detector 720 may send the image depicting the fault, along with current environmental and / or operational conditions, to a specialized predictive model. This model, taking into account the provided environmental and / or operational conditions, will analyze the fault and accurately predict its trend of development into a potential failure. Based on this prediction, the system can then generate and provide tailored maintenance instructions specific to the detected fault and its projected progression and / or suggested changes in operational conditions (such as limiting speed, adjusting operational height, etc.) that could be implemented until proper maintenance can be performed. This predictive capability enables proactive maintenance planning and operational adjustments, potentially extending the safe operational window of the equipment while minimizing the risk of unexpected failures. The failure detector 720 may then proceed with the process based on these predictive insights and recommendations.

[0215] In case the outcome indicates the ML model(s) 230 positively estimate that one or more of the failure mode(s) of the element 704 exists and / or develops, the process 600 may branch to 608. Otherwise, the failure detector 720 may take no action.

[0216] For example, the ML model(s) 230 may compute and output a probability score (e.g., in range of 0.00 to 1.00) computed to indicate the probability of presence and / or development of one or more of the failure mode(s) of the element 704. In such case, the failure detector 720 may compare the computed probability score to a certain threshold which may be set, for example predefined, dynamically defined, and / or a combination thereof to define a decision point for positive or negative estimation of presence and / or development the failure mode(s). The certain threshold may be set, as known in the art, according to one or more performance criteria defined for the failure detector 720, for example, accuracy, precision, recall, F-score, and / or the like.

[0217] Therefore, responsive to the probability score computed by the ML model(s) 230 being equal or exceeding the certain threshold, the failure detector 720 may branch to 608. Otherwise, responsive to the probability score not exceeding the certain threshold, the failure detector 720 may optionally take no further action.

[0218] In another example, the ML model(s) 230 may compute and output a severeness score, for example, in a scale of 1 to 10, 1 to 100, and / or the like indicative a level of severeness of each detected failure mode. As described herein before, severeness may be estimated based on a level, phase, and / or stage of one or more faults associated with the failure mode to become, induce, and / or turn into one or more failures. Responsive to the severeness score computed by the ML model(s) 230 being equal or exceeding another certain threshold, the failure detector 720 may branch to 608. Otherwise, responsive to the probability score not exceeding the another certain threshold, the failure detector 720 may optionally take no further action.

[0219] Optionally, the failure detector 720 may compare an aggregated score aggregating multiple scores, for example, the probability score, the severeness score, and / or the like to another (e.g., third) threshold and decide accordingly how to proceed, i.e., in case the aggregated score is equal or exceeds the third threshold, the failure detector 720 may branch to 608 and otherwise it may take no action.

[0220] As shown at 608, the failure detector 720 may output the estimation of presence and / or development of the failure mode(s) of the element 704 in the machine.

[0221] In particular, since the trained ML model(s) 230 estimate the failure mode(s) of the element 704 exists and / or develops, the failure detector 720 may output a detection, i.e., positive estimation, that the failure mode(s) are detected in the machine item.

[0222] The failure detector 720 may therefore output one or more indications, messages and / or alerts indicative that one or more of the failure modes of the element 704 are detected in the machine item. For example, the failure detector 720 may transmit, via one or more networks, and / or interfaces of the I / O interface 710, one or more failure detection messages to one or more systems, devices, and / or services, for example, a health monitoring system, a maintenance system, and / or the like. In another example, the failure detector 720 may output one or more alerts, for example, visual, audible, and / or tactile alerts presented to one or more users, operators, and / or the like that the failure mode(s) are detected.

[0223] Optionally, the failure detector 720 may output one or more maintenance schedules for conducting one or more maintenance, repair, and / or part replacement activities. For example, based on the severeness score computed for one or more failure modes detected for the element 704, the failure detector 720 may estimate a trend of the respective failure mode(s) and estimate a time period until detected fault(s) associated with the respective failure mode(s) become failures. The failure detector 720 may then generate one or more maintenance schedules accordingly.

[0224] Typically, as shown at 610, the process 600 may be an iterative process comprising a plurality of iterations. The failure detector 720 may therefore continuously, and / or periodically branch back to 602 to receive and analyze new images depicting the element 704, the machine, and / or part thereof in order to check for presence and / or development of the failure mode(s) of the element 704.

[0225] Optionally, the failure detector 720 may output one or more indications, messages and / or alerts indicative that no failure modes are estimated for the element 704. For example, the failure detector 720 may transmit, via one or more networks, and / or interfaces of the I / O interface 710, one or more no failure detection messages to one or more systems, devices, and / or services, for example, a health monitoring system, a maintenance system, and / or the like. In another example, the failure detector 720 may output one or more alerts, for example, visual, audible, and / or tactile alerts presented to one or more users, operators, and / or the like. Optionally, one or more of the photorealistic image(s) generated by a training data generator such as the training data generator 220 may be used for testing one or more failure detection systems such as the failure detection system 700 employing one or more trained ML models 230. In particular, the photorealistic image(s) depicting one or more faults associated with one or more failure modes of one or more mechanical elements such as the element 704 may be injected to the failure detection system 700 via one or more of the image sensor(s) 702 of the failure detection system 700. For example, the photorealistic image(s) may be injected via the communication interface connecting the image sensor(s) 702 to the failure detection system 700. In another example, the image sensor(s) 702 may be operated to image the photorealistic image(s) generated by the training data generator 220.

[0226] According to some embodiments of the present disclosure, there are provided methods, systems, and computer programs for optimizing image capturing parameters for the image sensor(s) 702 deployed and adapted to capture images of the monitored element(s).

[0227] Specifically, the image capturing parameters may be optimized based on analysis of one or more of the photorealistic images created as described in process 100 to improve quality of images of the monitored element(s) captured by the image sensor(s) 702, for example, reduce and potentially remove one or more of the degrading effects, improve focus and / or visibility of the monitored element(s), and / or the like.

[0228] Reference is now made to FIG. 8, which is a flowchart of a process of generating instructions for adjusting image capturing parameters of a failure detection system based on analysis of photorealistic images, according to some embodiments of the present invention.

[0229] Reference is also made to FIG. 9, which is a schematic illustrating of an exemplary system for generating instructions for adjusting image capturing parameters of a failure detection system based on analysis of photorealistic images, according to some embodiments of the present invention.

[0230] An exemplary process 800 may be executed for computing and / or generating instructions for adjusting one or more image capturing parameters of one or more failure detection systems such as the failure detection system 700, for example, imaging parameters of one or more image sensors such as the image sensor 702, illumination parameters of one or more light sources such as the light source 706, and / or the like.

[0231] The process 800 may be executed by an exemplary adjustment instructions generation system 900, for example, a server, a processing node, a cluster of processing nodes, and / or the like comprising an I / O interface 910 such as RO interface 210, a processor(s) 912 such as the processor(s) 212 for executing the process 800, and a storage 914 such as the storage 214 for storing data and / or code (program store).

[0232] The processor(s) 912 may execute one or more functional modules implemented using one or more software modules, one or more of the hardware modules and / or combination thereof. For example, the processor(s) 912 may execute an image capture optimizer 920 for executing the process 800 to optimize image capturing parameters of the failure detection system 700.

[0233] Moreover, as described here in before the training data generator 220 may be adjusted to further generate one or more photorealistic images which do not depict faults in the machine and / or part thereof. Such photorealistic images imaging no faults may be used by the for computing

[0234] For brevity, the process 800 is described for adjusting image capturing parameters relating to imaging of a single element of a single machine, for example, the element 704. This, however, should not be construed as limiting since, as may be apparent to a person skilled in the art, the process 800 may be duplicated, expanded and / or scaled for adjusting image capturing parameters relating to imaging of a plurality of elements of a plurality of machines.

[0235] As shown at 802, one or more digital design files relating to a (target) element, for example, the element 704 may be received, for example, by the adjusted training data generator, as described in step 102 of the process 100.

[0236] As shown at 804, one or more 3D animations depicting the element and / or part thereof may be created (generated), for example, by the adjusted training data generator, based on the digital design file(s) as described in step 106 of the process 100.

[0237] However, contrary to process 100, one or more of the 3D animation(s) may be created based on digital design file(s) which are not be adjusted to inject faults associated with failure modes of the target element, the 3D animation(s) comprising a sequence of a plurality of images depicting the element and / or part thereof may therefore show the element with no faults.

[0238] Optionally, one or more 3D animations may be created based on both adjusted and not adjusted digital design file(s) such that one or more 3D animations may depict one or more faults associated with one or more failures injected in the adjusted digital design file(s) while one or more other 3D animations may depict the element with no faults.

[0239] As shown at 806, visual media, comprising a plurality of images depicting the element and / or part thereof, may be generated, for example, by the adjusted training data generator, based on the 3D animation(s) as described in step 108 of the process 100.

[0240] Since one or more of the 3D graphical presentation(s) may be generated based on 3D animation(s) depicting the element with no injected faults, such 3D graphical presentation(s) may also show the element with no faults. Optionally, visual media may be generated based on 3D animation(s) created based on the adjusted digital design file(s) as described in step 108 such that the fault(s), visible via one or more elements of the machine, is depicted in the 3D graphical presentation(s).

[0241] As shown at 808, realistic image depicting the element, the machine, a background of the element, a background of the machine, and / or part thereof, may be received, for example, by the adjusted training data generator, as described in step 110 of the process 100.

[0242] As shown at 810, one or more photorealistic images imaging the element and / or part thereof may be generated, for example, by the adjusted training data generator, as described in step 112 of the process 100.

[0243] Since the photorealistic image(s) are generated based on visual media showing the element with no faults, such realistic images may also depict the element with no faults. Optionally, one or more photorealistic images may be generated based on visual media depicting one or more faults associated with failure mode(s) injected to the element, such photorealistic images may depict this fault(s).

[0244] As shown at 812, one or more of the photorealistic image(s) may be analyzed, for example, by the image capture optimizer 920, by the adjusted training data generator, and / or the like to evaluate visibility of the element in the photorealistic image(s).

[0245] For example, one or more photorealistic image may be analyzed to determine whether the target element and / or part thereof is sufficiently visible in one or more aspects, for example, whether the element is in frame, whether the element is illuminated, whether a fault associated with a failure mode of the element can be visible, and / or the like. In another example, one or more photorealistic image may be analyzed to determine whether the target element and / or part thereof is subject to reflections, sun glare and / or the like which may affect visibility of the element.

[0246] As shown at 814, the image capture optimizer 920 may generate adjustment instructions for adjusting one or more image capturing parameters of the failure detection system 700 based on the analysis of the photorealistic image(s) to improve visibility of the monitored element 704 in images captured by the image sensor(s) 702.

[0247] The image capturing parameters may comprise one or more imaging parameters of one or more of the image sensor(s) 702. Such imaging parameters may include, for example, positioning parameters, for example, location, position, orientation, viewpoint, and / or the like. In another example, the imaging parameters may include one or more inherent and / or operational parameters of the image sensor(s) such as, for example, shutter speed, contrast, zoom, dynamic range, and / or the like. For example, assuming that, based on analysis of one or more photorealistic images, the image capture optimizer 920 determines that due to reflections, either sunlight reflections and / or reflections of light projected by one or more of the light sources 706, visibility of at least part of the element 704 is poor. In such case, the image capture optimizer 920 may generate adjustment instructions for adjusting one or more positioning parameters of a certain image sensor 702, for example, an orientation, a viewpoint, and / or the like to reduce and potentially prevent the reflection thus increasing visibility of the element 704 in images captured by the image senor 702.

[0248] In another example, assuming that, based on analysis of one or more photorealistic images, the image capture optimizer 920 determines that the element 704 is highly dynamic and moves, for example, rotates at high frequency. In such case, the image capture optimizer 920 may generate adjustment instructions for adjusting one or more operational parameters of a certain image sensor 702, for example, increase a shutter speed and / or the like to configure the image senor 702 to capture high quality image data of the rotating element 704.

[0249] In another example, the image capturing parameters may comprise one or more illumination parameters of one or more of the light sources 706 adapted to illuminate the monitored element 704. The illumination parameters may include, for example, an intensity, a light spectrum and / or wave length (e.g., visible light, infrared, ultraviolet,, etc.), an illumination mode (e.g., wide angle lighting, focused light beam, strobe illumination, etc.), and / or the like.

[0250] For example, assuming that, based on analysis of one or more photorealistic images, the image capture optimizer 920 determines that a certain section of the element 704 is not exposed to external light and is thus poorly visible. In such case, the image capture optimizer 920 may generate adjustment instructions for adjusting one or more illumination parameters of a certain light source 706, for example, intensity of light projected on the certain section of the element 704 to increase the projected light level and improve visibility of the illuminated section in image data captured by the image sensor(s) 702.

[0251] In another example, assuming that, based on analysis of one or more photorealistic images, the image capture optimizer 920 determines that the element 704 rotates at a high frequency and reflects, at this high frequency, light projected on it. In such case, the image capture optimizer 920 may generate adjustment instructions for adjusting one or more illumination parameters of a certain light source 706, for example, a probe light frequency to align with the frequency of the dynamic element 704 in order to reduce reflections and improve visibility of the element 704 in image data captured by the image sensor(s) 702.

[0252] As shown at 816, the image capture optimizer 920 may output the adjustment instructions, for example, to one or more failure detection system 700 which may apply the adjustment instructions and adjust accordingly one or more of its image capturing parameters, for example, an imaging parameter, an illumination parameter, a combination thereof, and / or the like.

[0253] For example, the image capture optimizer 920 may transmit the adjustment instructions to one or more failure detection systems 700 over one or more networks via the I / O interface 910. In another example, assuming the image capture optimizer 920 is executed by each of one or more failure detection systems 700, specifically by their respective processor(s) 712, the image capture optimizer 920 may communicate with the failure detector 720 to provide the adjustment instructions via one or more interfaces available in the processing environment of the processor(s) 712, for example, a system call, an OS routine, an Application Programming Interface (API) of the failure detector 720, and / or the like.

[0254] The descriptions of the various embodiments of the present invention have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

[0255] It is expected that during the life of a patent maturing from this application many relevant systems, methods and computer programs will be developed and the scope of the terms ML models, neural networks, CAD applications, generation of 3D animations, 3D simulations and 3D representations are intended to include all such new technologies a priori.

[0256] As used herein the term “about” refers to + 10 %.

[0257] The terms "comprises", "comprising", "includes", "including", “having” and their conjugates mean "including but not limited to". This term encompasses the terms "consisting of" and "consisting essentially of".

[0258] The phrase "consisting essentially of" means that the composition or method may include additional ingredients and / or steps, but only if the additional ingredients and / or steps do not materially alter the basic and novel characteristics of the claimed composition or method.

[0259] As used herein, the singular form "a", "an" and "the" include plural references unless the context clearly dictates otherwise. For example, the term "a compound" or "at least one compound" may include a plurality of compounds, including mixtures thereof.

[0260] The word “exemplary” is used herein to mean “serving as an example, an instance or an illustration”. Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments and / or to exclude the incorporation of features from other embodiments.

[0261] The word “optionally” is used herein to mean “is provided in some embodiments and not provided in other embodiments”. Any particular embodiment of the invention may include a plurality of “optional” features unless such features conflict.

[0262] Throughout this application, various embodiments of this invention may be presented in a range format. It should be understood that the description in range format is merely for convenience and brevity and should not be construed as an inflexible limitation on the scope of the invention. Accordingly, the description of a range should be considered to have specifically disclosed all the possible subranges as well as individual numerical values within that range. For example, description of a range such as from 1 to 6 should be considered to have specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6 etc., as well as individual numbers within that range, for example, 1, 2, 3, 4, 5, and 6. This applies regardless of the breadth of the range.

[0263] Whenever a numerical range is indicated herein, it is meant to include any cited numeral (fractional or integral) within the indicated range. The phrases “ranging / ranges between” a first indicate number and a second indicate number and “ranging / ranges from” a first indicate number “to” a second indicate number are used herein interchangeably and are meant to include the first and second indicated numbers and all the fractional and integral numerals there between.

[0264] It is appreciated that certain features of the invention, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable sub-combination or as suitable in any other described embodiment of the invention. Certain features described in the context of various embodiments are not to be considered essential features of those embodiments, unless the embodiment is inoperative without those elements.

[0265] Although the invention has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art. Accordingly, it is intended to embrace all such alternatives, modifications and variations that fall within the spirit and broad scope of the appended claims.

[0266] It is the intent of the applicant(s) that all publications, patents and patent applications referred to in this specification are to be incorporated in their entirety by reference into the specification, as if each individual publication, patent or patent application was specifically and individually noted when referenced that it is to be incorporated herein by reference. In addition, citation or identification of any reference in this application shall not be construed as an admission that such reference is available as prior art to the present invention. To the extent that section headings are used, they should not be construed as necessarily limiting. In addition, any priority document(s) of this application is / are hereby incorporated herein by reference in its / their entirety.

Claims

WHAT IS CLAIMED IS:

1. A method of generating a synthetic training set for training a machine learning (ML) model to detect machine failures, comprising: using at least one processor for: receiving at least one visual media depicting at least one fault associated with at least one failure mode of at least one element of a machine; generating at least one photorealistic image imaging the at least one fault by processing the at least one visual media in correlation with realistic image data of the at least one element; and outputting the at least one photorealistic image for training at least one ML model to estimate presence of the at least one failure mode in at least one machine item comprising the at least one element based on analysis of real-world image data depicting the at least one element.

2. The method of claim 1, further comprising: generating the at least one visual media to depict a trend of the at least one fault comprising development of the at least one fault into at least one failure, and generating at least one photorealistic image sequence imaging the trend of the at least one fault.

3. The method of claim 1, wherein the at least one visual media comprises a plurality of images depicting development of the at least one fault into at least one failure.

4. The method of claim 1, wherein the at least one visual media depicting the at least one fault is generated by adjusting at least one digital design file relating to the at least one element adjusted to inject the at least one failure mode into design of the at least one element.

5. The method of claim 4, wherein at least one visual media is generated based on at least one 3D animation depicting the at least one fault, the at least one 3D animation is created based on the at least one digital design file adjusted to inject at least one failure mode.

6. The method of claim 4, wherein the at least one digital design file is a member of a group comprising: a 2D design file, and a 3D design model.

7. The method of claim 1, wherein at least part of the realistic image data of the at least one element is captured by at least one image sensor deployed to capture image data of the at least one machine item and / or part thereof.

8. The method of claim 7, wherein the at least one image sensor is adapted to capture image data of the at least one machine item and / or part thereof while both the at least one image sensor and the at least one machine item and / or part thereof are moving.

9. The method of claim 7, wherein the at least one image sensor is adapted to capture image data of the at least one machine item and / or part thereof while the at least one image sensor is static with respect to the at least one element.

10. The method of claim 7, further comprising processing at least part of the realistic image data captured by the at least one image sensor to reduce at least one degrading effect.

11. The method of claim 10, further comprising adjusting at least one positioning parameter of the at least one image sensor to at least reduce the at least one degrading effect in subsequently captured realistic image data of the at least one machine item and / or part thereof.

12. The method of claim 1, wherein at least part of the realistic image data is generated using at least one generative ML model adapted to generate synthetic 3D images of the at least one element based on at least one digital design file relating to the at least one element.

13. The method of claim 1, wherein the at least one photorealistic image is generated using at least one generative ML model.

14. The method of claim 1, wherein the at least one fault is visible in the at least one element in the at least one of the photorealistic image.

15. The method of claim 1, wherein the at least one failure mode induces at least one another fault in at least one another element of the machine, the at least one another fault is visible in at least one another element of the machine in the at least one photorealistic image.

16. The method of claim 1 , wherein the estimation of the at least one trained ML model comprises a probability score indicative of a probability of presence of the at least one failure mode in the at least one machine item.

17. The method of claim 1, wherein the estimation of the at least one trained ML model further comprises a severe estimation indicative of a level of severeness of the at least one failure mode in the at least one machine item.

18. The method of claim 1, wherein the at least one photorealistic image is further used for validating and / or testing the at least one ML model.

19. The method of claim 1, wherein the at least one fault associated with the at least one failure mode comprises an alteration in dynamic movement of the at least one element, and wherein the at least one visual media and the at least one photorealistic image depict the alteration in dynamic movement.

20. The method of claim 1, further comprising: generating a temporal progression of the at least one fault over an extended time period; predicting, based on the temporal progression, when the at least one fault will develop into a failure; and outputting maintenance instructions based on the prediction.

21. The method of claim 1, further comprising: receiving current environmental and operational conditions associated with the at least one machine item; predicting, using the at least one ML model, a trend of fault development based on the current environmental and operational conditions; and outputting suggested changes in operational conditions to extend a safe operational window until maintenance can be performed.

22. The method of claim 1, wherein generating the at least one photorealistic image comprises: using physics-based simulation software to model structural behavior of the at least one element under various conditions and fault scenarios; and applying machine learning techniques to accelerate the simulation process.

23. The method of claim 1 , wherein the realistic image data comprises images captured by at least one sensor depicting actual faults present in the at least one machine item.

24. The method of claim 1 , wherein the at least one visual media depicts the at least one fault over time periods ranging from seconds to months, enabling prediction of fault trends across various time scales.

25. A system for generating a synthetic training set for training a machine learning (ML) model to detect machine failures, comprising: at least one processor configured to execute a code, the code comprising: code instructors to receive at least one visual media depicting at least one fault associated with at least one failure mode associated with at least one fault of at least one element of a machine; code instructors to generate at least one photorealistic image imaging the at least one fault by processing the at least one visual media in correlation with realistic image data of the at least one element; and code instructors to output the at least one photorealistic image for training at least one ML model to estimate presence of the at least one failure mode in at least one machine item comprising the at least one element based on analysis of real-world image data depicting the at least one element.

26. A method of generating a synthetic training set for testing a failure detection system, comprising: using at least one processor for: receiving at least one visual media depicting at least one fault associated with at least one failure mode of at least one element of a machine; generating at least one photorealistic image imaging the at least one fault by processing the at least one visual media in correlation with realistic image data of the at least one element; and outputting the at least one photorealistic image for testing at least one failure detection system employing at least one ML model trained to estimate presence of the at least one failure mode in at least one machine item comprising the at least one element based on analysis of real-world image data depicting the at least one element.

27. The method of claim 26, wherein the at least one photorealistic image is injected into the at least one failure detection system via at least one image sensor adapted to capture the real- world image data depicting the at least one element.

28. A method of detecting machine failures , comprising: using at least one processor for: receiving at least one image depicting, at least partially, at least one element of at least one machine; applying at least one trained ML model to estimate presence of at least one failure mode of at least one element of the at least one machine based on detection of at least one fault associated with the at least one failure mode in the at least one image; and outputting the estimation of presence of the at least one failure mode of the at least one element; wherein the at least one ML model is trained using at least one photorealistic image imaging the at least one fault, the at least one photorealistic image is generated by processing at least one visual media depicting the at least one fault in correlation with realistic image data of the at least one element.

29. The method of claim 28, wherein the at least one trained ML model is adapted to compute a probability score indicative of a confidence level of the estimation.

30. The method of claim 28, wherein the at least one failure mode is estimated to exist in the at least one machine based on a comparison of the probability score with a certain threshold.

31. A method of adjusting imaging parameters of failure detection system based on analysis of photorealistic images, comprising: using at least one processor for: receiving at least one photorealistic image imaging the at least one element generated by processing at least one visual media in correlation with realistic image data of the at least one element; generating adjustment instructions based on analysis of the at least one photorealistic image for adjusting at least one image capturing parameter of a failure detection systemadapted to detect at least one fault associated with at least one failure mode of the at least one element based on image data of the machine and / or part thereof; and outputting the adjustment instructions to the failure detection system.

32. The method of claim 31 , wherein the at least one image capturing parameter comprises at least one imaging parameter of at least one image sensor used by the failure detection system to capture the image data, the at least one imaging parameters is a member of a group comprising: a positioning parameter, an inherent parameter, and an operational parameter.

33. The method of claim 31 , wherein the at least one image capturing parameter comprises at least one illumination parameter relating to at least one light source used by the failure detection system to illuminate the machine and / or part thereof.

34. The method of claim 31, wherein at least one visual media is generated based on at least one 3D animation created based on at least one digital design file relating to the at least one element.

35. The method of claim 31, wherein the at least one digital design file is a member of a group comprising: a 2D design file, and a 3D design model.

36. The method of claim 31, wherein at least part of the realistic image data of the at least one element is captured by at least one image sensor deployed to capture image data of at least one machine item and / or part thereof.

37. The method of claim 31, wherein at least part of the realistic image data is generated using at least one generative ML model adapted to generate synthetic 3D images of the at least one element based on at least one digital design file relating to the at least one element.

38. The method of claim 31, wherein the at least one photorealistic image is generated using at least one generative ML model.

Images