Method for inspecting a component of a turbomachine, and corresponding device for inspecting such a component

Abstract

The present invention relates to a method for inspecting a component of a turbomachine (1), in particular an aircraft engine, said method comprising the steps of: providing a sensor for detecting the component (20); moving the sensor along a sensor path relative to the component (35, 36); using the sensor to detect at least one region of the component (25, 26) that lies on the sensor path; wherein the sensor is moved in a robot-guided manner along the sensor path using control signals, and wherein the control signals are generated (55) based on AI.

Description

[0001] Method for examining a component of a fluid machine and corresponding device for examining such a component

[0002] DESCRIPTION

[0003] Technical area

[0004] The present invention relates to a method for examining a component of a turbomachine, in particular an aircraft engine.

[0005] State of the art

[0006] An examination or inspection of components of a turbomachine, particularly an aircraft engine, can be carried out during new production, for example, during a process or output inspection. It can also be performed on a component that has already been in use, for example, to examine the component for abnormalities that have arisen during operation, particularly damaged areas. These can result, for example, from particle impacts or wear during operation; they can be cracks or chipping, for example. Regardless of these details, component examination can be particularly important due to the possible consequences of failure of a component of a turbomachine or aircraft engine, regardless of whether it is a new production or a revision.

[0007] Description of the invention

[0008] The present invention is based on the technical problem of providing an advantageous method for examining a component of a turbomachine, in particular an aircraft engine.

[0009] This is achieved with the method according to claim 1. A sensor is provided (step i) and moved robot-guided with control signals along a sensor path relative to the component (step ii), wherein at least one area of ​​the component is detected by the sensor (step iii). A special feature here is that these control signals are created using AI, i.e. an AI algorithm determines the control signals and thus the course of the sensor path. An AI algorithm is therefore not only used, for example, when evaluating the sensor data, but already during the data acquisition process. To illustrate, for example, a camera can be moved along the sensor bike using AI-supported control signals, and (initial) images can be created for examining the component. Compared to an alternative approach in which the camera is moved under human guidance and then only the images are subjected to AI analysis, this can, for example,This can be advantageous in terms of reproducibility, as it can prevent deviations from one operator to another.

[0010] As discussed in detail below, in this case, not only the sensor path ("at which points is the component measured?") but also the sensor selection ("which sensor is used for the measurement?") can be AI-based. Overall, this can reduce both artifacts, i.e., incorrectly identified anomalies / defects, and so-called duplicates, i.e., multiple counts made due to different recordings of the same anomaly. The latter can occur particularly with the combination of human-guided recording and AI analysis described above.

[0011] Further preferred embodiments can be found in the dependent claims and the entire disclosure, which is at least implicitly directed to all categories of claims.

[0012] As already mentioned, the subject matter can in particular relate to a multi-stage method with several sequential AI-based steps, namely, in addition to the sensor path, the sensor selection and / or the evaluation can also be carried out AI-based. A method in which the component is captured with a camera in a first pass and, based on these images, a sensor is then selected or a sensor path for a detailed measurement is created in a second pass can be particularly preferred. The steps i) to iii) according to the main claim can be carried out both in the first pass (cf. claims 2 / 3) and in the second pass (cf. claim 5 ff.). Furthermore, a combination is also possible, i.e., in particular, the preparatory recording can already be created AI-based in preparation for the AI-based sensor selection and, in addition, the sensor path for the AI-selected sensor can also be created AI-based.Overall, several Kl algorithms can be used sequentially or linked together in the component analysis.

[0013] In general, “AI-based” means using an AI algorithm, i.e. artificial intelligence (AI). Unsupervised learning methods can be used here, for example, to examine data for patterns without any known target values. This can be used to distinguish anomalies (outliers) from normal variance (e.g., background noise), and the data can then be clustered according to their features, i.e., sorted by commonalities. Alternatively or additionally, supervised learning methods can be used, which can be trained and / or monitored, for example. This can, for example, make it possible to draw on existing human experience during the examination or inspection. The algorithm can, for example,trained using recordings (in the case of the camera) or measurement results (in the case of other sensors), in which anomalies are previously identified based on human experience, which is also referred to as “labeling”.

[0014] In the case of the exemplary embodiment, the individual algorithms are numbered consecutively for conceptual differentiation ("first Kl algorithm" or "KI 1", etc.), but this should not generally preclude a (partial) integration of the algorithms within a common control / evaluation program. With regard to the exemplary embodiment, step ii) according to the main claim or steps ii) and iii) according to the main claim can be based on control signals from KI 1 or KI 4. Regardless of these details, the sensor is moved or guided by a robot. In principle, the sensor is always robot-guided. However, it is also possible for the robot to guide the component past the sensor(s). The latter is particularly suitable for new part inspection.

[0015] In other words, the sensor can be provided as an effector on a robot arm as a manipulator, with the latter being controlled by the control signals. Preferably, the sensor is moved by a robot, whereby the component can be at rest when viewed in a fixed coordinate system. The "sensor path" describes the path along which the sensor is moved relative to the component. In the case of a camera, and especially during an initial measurement, recordings can be made essentially along the entire sensor path; for example, a video of the component surface can be recorded with the movement along the sensor path.

[0016] After selecting a specific sensor or during a follow-up measurement, the sensor can be moved along the sensor path toward the anomaly. The sensor measurement can then be initiated on-site, specifically at the anomaly, for example. In summary, a sensor measurement does not have to be taken along the (entire) sensor path; the sensor path can also simply determine the distance the sensor moves toward the anomaly.

[0017] According to a preferred embodiment, the sensor is a camera, and a sequence of images is created along the sensor path, preferably as a video recording. This can generally also relate to the second or subsequent measurement, e.g., with a special camera (for close-ups or with particular wavelength sensitivity), but in particular it is directed at the first measurement. The images of the component taken with the camera are evaluated on a K1-based basis to generate the control signals, i.e., to define a measurement strategy. These control signals can relate either to the sensor path ("camera path") of the camera itself and / or to a sensor then used in the subsequent measurement. The video or sequence recording can, for example, facilitate a coherent recording of the component, thus, for example, reducing the probability of duplicates as mentioned above.

[0018] According to a preferred embodiment, which relates to the aforementioned variant ("camera"), the further sensor or camera path is dynamically adjusted based on the AI-based evaluation. This allows the sensor path to be adjusted during step ii) or iii), for example, so that a conspicuous area is recorded in greater detail (from a shorter distance / with greater zoom) and / or captured at a different angle. Figuratively speaking, the AI-based adjustment allows for a "closer look" as needed during image acquisition. This can also be done, for example, based on an initial classification of the abnormality, i.e., an initial categorization of the abnormality or damaged area. The corresponding AI algorithm can, for example, be trained as a supervised / earwrig algorithm such that for certain classes, e.g., "chipping," "crack," "delamination," "material change / discoloration," etc., different information is stored for suitable shots (e.g. regarding the perspective, such as “top view” vs. “oblique view”).

[0019] According to a preferred embodiment, which is generally intended to be disclosed independently of the AI-based control signal generation, the sensor is selected from a plurality of sensors in step i) based on AI. Possible sensors can be, for example, an optical stripe projection sensor, a UV light sensor, an eddy current test sensor, a tactile sensor, a white light interferometer sensor, and a fluorescence sensor, with at least two, preferably at least three, of the aforementioned sensors then being available for selection in step i). In the case of the exemplary embodiment, the sensor selection is carried out by AI 4, which can then simultaneously determine the sensor path ("which sensor?" and "which anomaly is approached with it?").

[0020] AI-based sensor selection is preferably carried out in coordination with an anomaly that is identified in an image of the component using AI. This image is preferably part of a sequence or video (see above) that is created during a previous acquisition of the component. An "anomaly" can, for example, generally be a deviation from the appearance of the surrounding surface, i.e., an optical deviation (e.g., in shape and / or color). To reduce the likelihood of misclassification, the image(s) can be subjected to image processing, for example, to filter out shadows, reflections, etc. Alternatively or additionally, the position of the anomaly can also be determined, i.e., its coordinates can be determined in a computer model (e.g., in a 3D CAD model), which can simplify the assignment of holes, etc.

[0021] According to a preferred embodiment, the abnormality is assigned to a (provisional) class for sensor selection, i.e., assigned to a possible damage class (e.g., crack, chip, etc.) based on AI. Different sensors can then be assigned to the different preliminary classes, with which the respective damage pattern can be specifically verified. Generally, but especially in the case of an ambiguous classification, multiple sensors can be used. In addition to the sensor selected in step i), the abnormality can therefore be measured with another sensor.

[0022] In general, AI-based sensor guidance (sensor path) or sensor selection is preferably linked to an AI-based evaluation of the sensor measurement. In a preferred embodiment, the anomaly is (additionally) classified after the second or subsequent measurement, i.e., based on the data obtained with the specifically selected sensor. This classification assigns the anomaly to a verified class that, for example, has a lower probability of misclassification compared to the preliminary class. Preferably, the preliminary and verified classifications are compared with each other, from which, for example, training data for the preliminary classification and / or the sensor selection or verified classification can be generated.

[0023] The (verified) classification can be output as a result ("What type of error is involved?"), for example, in conjunction with a coordinate position. In addition to this label, i.e., the tag of the anomaly, details of the classification process (e.g., sensor selection and measurement results) can also be stored, thus documenting the classification.

[0024] In a preferred embodiment, an AI-based recommendation is issued, e.g., for a maximum permissible time interval (e.g., in operating hours) until the next overhaul. Alternatively, a component can be removed for rework or replacement, e.g., if the abnormality in terms of size, shape, or location lies outside predefined limits. Rework control signals can then be generated from the measurements taken for a component overhaul; see below for details.

[0025] According to a preferred embodiment, as already mentioned, the AI-based sensor selection and guidance is combined with prior AI-based camera guidance. Then, both the camera control signals (AI 1 in the exemplary embodiment) and the control signals for the sensor path (AI 4 exemplary embodiment) are generated AI-based.

[0026] The images captured with the camera can preferably be subjected to image processing and / or comparison (localization) with a computer model, see also the above comments. This can, for example, reduce misclassifications during the initial detection of the anomaly, thus correspondingly reducing the subsequent effort. Comparison with a computer model can also be advantageous for later revision, for example, by allowing the anomaly to be localized based on the determined coordinates.

[0027] In general, according to a preferred embodiment, the control signals and / or measurement results are used to generate AI-based rework control signals. These signals can then be used to move a tool relative to the component along a tool path in a robot-guided manner, allowing the tool to rework the abnormality. This can, for example, be a material-removing and / or material-depositing operation, such as back-grinding in conjunction with subsequent (re)coating, etc. Independent of these details, not only the camera or sensor path is generated AI-based, but also the tool path.

[0028] According to a preferred embodiment, the component is examined in the assembled state, i.e., it is mounted in a component group, preferably in the complete engine. This can be removed from the aircraft or even still mounted on it. Although a conventional industrial robot can generally be used here (e.g., an articulated arm), for example, to examine the fan, the sensor is then preferably fed and moved using a robotic borescope—colloquially also referred to as a continuum robot. This can be done, for example, via the gas duct or specific borescope openings.

[0029] The application further relates to a device for examining a component of a turbomachine, which in addition to the robot and sensor has a computer unit. The robot can in particular be provided in the form of a robot borescope, i.e., be designed for measuring on the mounted engine. Colloquially, a robot borescope is understood to mean a continuum robot. The computer unit is configured to move the sensor along the sensor path; see the above disclosure for further details. Alternatively or additionally, the computer unit is configured to select the sensor from a plurality of sensors; see above. In general, the method steps described in this disclosure should always be interpreted to mean that a computer unit is configured to carry out the method step (e.g., KL calculation) and / or to initiate the method step (e.g.,Movement by means of the robot arm / detection with the sensor).

[0030] Short description of the drawings

[0031] In the following, the invention is explained in more detail using exemplary embodiments, whereby the individual features may also be essential to the invention in other combinations. In detail,

[0032] Figure 1 shows a turbomachine, namely an aircraft engine in a schematic longitudinal section;

[0033] Figure 2 is a schematic illustration of a robot-guided sensor moved relative to the component;

[0034] Figure 3 follows the situation. Figure 2 shows another sensor that is moved by a robot;

[0035] Figure 4 is a flow chart summarizing some process steps.

[0036] Preferred embodiment of the invention

[0037] Fig. 1 shows a turbomachine 1, specifically an aircraft engine 3, in an axial section. The section plane thus includes a longitudinal axis 2, around which the rotors rotate during operation. Functionally, the turbomachine 1 is divided into a compressor 1a, a combustion chamber 1b, and a turbine 1c. During operation, air drawn in the compressor 1a is compressed and combusted in the downstream combustion chamber 1b with added fuel, particularly kerosene. The resulting combustion or hot gas is expanded in the turbine 1c, driving its rotors, which is also partially converted into rotation of the compressor 1a and the fan.

[0038] Such an engine contains a large number of components which must meet different requirements depending on their location and purpose. This subject matter focuses on the examination of such components, both during new production and, in particular, during overhaul. This is explained below using as an example a rotor blade or a rotor blade element located in the engine section of a so-called rotor. Of course, the method can also be applied to stators and their components or to engine components in general; it can be of particular interest for components arranged in the gas duct or which delimit it. Fig. 2 shows a rotor blade 21 as the examined component 20, of which only one rotor blade element 21a is accessible in the assembled state. For this reason, its rotor blade root 21b is shown in dashed lines.In Figure 2, the rotor blade 21a is examined with a sensor 25, namely initially a camera 26. Its detection field 28 is directed onto the component 20 or rotor blade 21a, wherein the camera 26 is moved by a robot 30 along a sensor path 35 or camera path 36.

[0039] The corresponding control signals 45 are generated on a KI basis, namely by a computer unit 50, shown schematically here, based on a sequence 55 of images 55.1-55.3 recorded with the camera 26 along the sensor or camera path 35, 36. The sensor or camera path 35, 36 can also be dynamically adjusted; for example, in the event of a detected abnormality 60, the subsequent sensor or camera path 35.1, 36.1 can be adjusted so that the abnormality 60 is recorded again, for example, from a different direction and / or at a different angle.

[0040] Fig. 3 illustrates a subsequent examination step, wherein a sensor 25 is again moved by the robot 30 along a sensor path 35, in this case specifically toward the abnormality 60. Previously, the sensor 25 was selected based on AI from a plurality of sensors 25.1-25.3. The abnormality 60 was classified based on the images 55.1-55.3, and a sensor 25 matching the class, i.e., the damage pattern (e.g., crack, chipping, etc.), was selected; in the example shown in Figure 3, this was a tactile sensor 27. The sensor path 35, i.e., the control signals 45, are also based on an AI-based evaluation of the images 55.1-55.3. Following the final measurement with the tactile sensor 27, the previously performed preliminary classification can be verified.

[0041] Fig. 4 summarizes some process steps and elements, particularly the AI-based evaluation, in a flowchart. First, a first, second, third, fourth, and fifth AI, namely AI 1 - AI 5, are shown. AI 1 generates the control signals 45, by means of which the robot 30, in particular a robot borescope, guides the camera 26 along the sensor or camera path, see Figure 2. The resulting sequence 55 of images is evaluated by AI 2, which thus identifies possible anomalies, particularly damaged areas. There may also be partial integration or feedback between AI 1 and AI 2, for example, for the dynamic adaptation of the sensor path (see Figure 2).

[0042] During the processing of the images by AI 1, image processing and / or registration can also be performed. This means, for example, that shadows, reflections, etc. can be filtered out, and the position can be determined (e.g., identification of the blade or its position on the blade). If AI 2 determines the fundamental presence of a possible anomaly, it is classified by AI 3 and assigned to a provisional class 70. With this (provisional) classification into a class (e.g., crack, chip, etc.), a probability can also be determined with which the anomaly or defect falls into the corresponding class.

[0043] Based on this (preliminary) classification, AI 4 then selects the sensor for the subsequent measurement (see the description of Figure 3). AI 4 can also simultaneously specify the sensor path, i.e., control signals 45 and, in conjunction with the sensor selection, a measurement strategy are specified. This can, for example, also include the position from which the best result can be expected with the selected sensor. Furthermore, in cases of doubt, alternative measurements can be suggested to identify exclusion criteria. The measurement results 75 are evaluated by AI 5, which can, in particular, assign the anomaly to a verified class 80. A comparison with the preliminary class 70 is also possible. Furthermore, the reasons for the classification can, for example, be documented and / or further instructions can be issued.

[0044] Considering the process steps functionally, a sensor is first provided 91, in the first step predefined as a camera, and in the second step through AI-based sensor selection. The sensor is then moved along the sensor path 92, and the component is detected 93, in the case of the camera 26 by taking images, for example, by recording a video. In the case of the second measurement, the type of detection also depends on the specific sensor type selected; it can be, for example, a tactile or optical or electrical measurement (e.g., eddy current sensor, tactile sensor, or white light interferometer, etc.).

[0045] Training 101 of AI 1 can be supported by human experience from the examination or inspection area, which can, for example, affect both initially suitable sensor paths and strategies for behavior in the event of anomalies. Training 102 of AI 2 can, for example, be carried out using indexed images, which also applies to training 103 of AI 3. Training 104 of AI 4 can, in turn, be supported by experience from the inspection area. For training 105 of AI 5, existing catalogs of acceptance criteria can be used, i.e., criteria catalogs stored in a database.

[0046] LIST OF REFERENCE SYMBOLS

[0047] Turbomachine 1

[0048] Compressor la

[0049] Combustion chamber 1b

[0050] Turbine 1c

[0051] Component 20

[0052] Blade 21

[0053] Blade 21a

[0054] Blade root 21b

[0055] Sensor 25,

[0056] Plural 25.1-25.3

[0057] Camera 26

[0058] Tactile Sensor 27

[0059] Recording field 28

[0060] Robot 30

[0061] Robot Borescope 31

[0062] Sensor path 35

[0063] Camera path 36

[0064] Control signals 45

[0065] Computer unit 50

[0066] Sequence 55

[0067] Recordings 55.1-55.3

[0068] abnormality 60

[0069] AI AI 1-AI 5.

[0070] Provisional Class 70

[0071] Verified Class 80

[0072] Before see 91

[0073] Move 92

[0074] Capture 93

[0075] Training AI 1 101

[0076] Teaching AI 2 102 Teaching AI 3 103

[0077] Training AI 4 104

[0078] Training AI 5 105

Claims

CLAIMS 1. A method for examining a component (20) of a turbomachine (1), in particular an aircraft engine (3), comprising the steps of: i) providing (91) a sensor (25) for detecting the component (20); ii) moving (92) the sensor (25) along a sensor path (35) relative to the component (20); iii) using the sensor (25) to detect (93) at least one region of the component (20) that lies on the sensor path (35); wherein in step ii) the sensor (25) is moved along the sensor path (35) in a robot-guided manner using control signals (45), and wherein the control signals (45) are generated using AI.

2. The method according to claim 1, wherein the sensor (25) is a camera (26) with which a sequence (55) of images (55.1-55.3) is created along the sensor path (35), wherein the images (55.1-55.3) are evaluated on an AI-based basis to create the control signals (45).

3. The method according to claim 2, wherein the sensor path (35) is dynamically adapted on the basis of the AI-based evaluation of the recordings (55.1-55.3), i.e. the further sensor path (35) is adapted during step ii) on the basis of the AI-based evaluation of the recordings (55.1-55.3).

4. The method according to claim 1, wherein in step i) the sensor (25) is selected AI-based from a plurality of sensors (25.1-25.3).

5. The method according to claim 4, wherein the AI-based selection of the sensor (25) is carried out on the basis of an anomaly (60) which is determined AI-based in a recording (55.1-55.3) of the component (20).

6. The method according to claim 5, wherein the abnormality (60) is classified based on AI, i.e. is assigned to a preliminary class (70), wherein the Selection of the sensor (25) is made based on the assigned preliminary class (70).

7. Method according to 6, in which, if the AI-based classification does not result in a clear assignment to a preliminary class (70), the abnormality (60) is additionally measured with a further sensor (25).

8. The method according to one of claims 5 to 7, wherein, based on the measurement with the sensor (25) selected in step i) based on AI, the abnormality (60) is classified based on AI, i.e. is assigned to a verified class (80).

9. Method according to claim 8 in conjunction with claim 6 or 7, in which the classifications are compared, i.e. the preliminary (70) and the verified (80) class are compared with each other.

10. The method according to claim 8 or 9, wherein, based on the verified class (80), an AI-based recommendation for a next revision interval or for a revision of the component (20) is issued.

11. Method according to one of claims 5 to 10, in which the image (55.1-55.3) of the component (20), in which the abnormality (60) is then determined based on AI, is previously recorded of the component (20) with a camera (26), which is moved robot-guided with camera control signals along a camera path (36) relative to the component (20) and records a sequence (55) of images (55.1-55.3), the camera control signals being created based on AI.

12. Method according to claim 11, wherein the recording (55.1-55.3) or recordings (55.1-55.3) in the course of determining the abnormality (60) of a Image processing and / or a comparison with a computer model of the component (20) for coordinate determination.

13. Method according to one of the preceding claims, in which, on the basis of the control signals (45) from step ii) and / or from the detection (93) of data obtained with the sensor (25) in step iii), K1-based reworking control signals are generated for the robot-guided movement of a reworking tool along a tool path.

14. Method according to one of the preceding claims, in which the component (20) is inspected in an assembled state and the sensor (25) is fed and moved with a robotic borescope (31).

15. Device for examining a component (20) of a turbomachine (1), in particular an aircraft engine (3), comprising a robot (30), in particular a robot borescope (31); a sensor (25); a computer unit (50); wherein the computer unit is configured to move the sensor (25) along a sensor path (35) relative to the component (20) using control signals (45) generated on an AI basis; and / or to select the sensor (25) on an AI basis from a plurality of sensors (25.1-25.3).

Images