Non-destructive testing of a mechanical part using eddy currents

Abstract

A non-destructive eddy current testing method for a mechanical part, comprising the steps of: - obtaining (E20) a control image of the mechanical part, - detecting (E22) edge effects in the control image by extracting features from a predetermined number of first and last lines of the control image and applying a learned classification model to classify the predetermined number of first and last lines of the control image into lines with edge effects and lines without edge effects, - setting (E22) the pixel values ​​of the lines classified as lines with edge effects to zero, to form a control image without edge effects, - determining (E23) indications in the control image without edge effects that may indicate a defect in the mechanical part, - detecting (E24) indications corresponding to a defect in the mechanical part among the determined indications. Figure for the abstract: Figure 4

Description

Title of the invention: Non-destructive testing of a mechanical part using eddy currents technical field

[0001] The present invention relates to the non-destructive testing by eddy currents of a mechanical part, in particular an aeronautical part, to detect possible surface defects and / or defects close to the surface of the metal part. Prior state of the art

[0002] Non-destructive testing by eddy currents is a method used to detect surface and near-surface defects of a part made of an electroconductive material.

[0003] Eddy currents are alternating magnetic fields that are created when an alternating electric current passes through one or more coils in a probe assembly. When the probe is placed near the part to be inspected, the alternating magnetic field induces eddy currents in the part. Discontinuities or variations in the properties of the inspected part modify the flow of eddy currents and are detected by the probe, thus enabling the detection of defects such as cracks and corrosion.

[0004] Eddy current control is based on the physical phenomenon of electromagnetic induction. In an eddy current probe, an alternating current flows through a coil of conducting wire and generates an oscillating magnetic field. If the probe and its magnetic field are placed near a conductive material, such as a metal part, circular flows of electrons called "eddy currents" begin to move through the metal. These eddy currents flowing through the metal, in turn, generate their own magnetic field, which interacts, by mutual inductance, with the coil and its field.

[0005] Surface or underlying defects in the inspected part interrupt or alter the amplitude and distribution of eddy currents and the resulting magnetic field. The eddy current distribution is also modified in the event of geometric variation or non-homogeneity in the structure of the inspected material. The movement of electrons in the coil is then altered, which changes the electrical impedance of the coil. Eddy current inspection devices record the changes in impedance amplitude and phase angle, and the results can be used by a qualified operator to detect defects in the part.

[0006] Eddy current testing of a mechanical part results in a C-scan type inspection image showing the part's structure in a planar view. This graphical representation allows for the characterization of the part's material properties or structural defects.

[0007] The interpretation of the inspection images is lengthy and requires an inspector holding specific certification and trained in this application. All inspection images of parts inspected by eddy currents are thus examined, which increases the inspection time.

[0008] There is therefore a need to develop an automated tool for processing inspection images to efficiently and quickly identify non-conforming parts. Description of the invention

[0009] The invention aims to solve the problems of the prior art by providing a method for non-destructive testing of a mechanical part using an eddy current testing device, the method comprising the following steps:

[0010] - obtaining by the eddy current control device an image of mechanical part inspection,

[0011] - detection of edge effects in the control image, by feature extraction in a predetermined number of first and last lines of the control image and application of a learned classification model, to classify the predetermined number of first and last lines of the control image into lines with edge effect and lines without edge effect,

[0012] - resetting the pixel values ​​of lines classified as lines with edge effect, to form a control image without side effects,

[0013] - determination of indications in the control image without edge effect, which can indicate a defect in the mechanical part,

[0014] - detection of indications corresponding to a defect in the mechanical part among the specific indications.

[0015] Thanks to the invention, it is possible to process control images automatically, quickly and repeatably to identify non-conforming parts.

[0016] Furthermore, the inventors observed that existing eddy current testing methods do not automatically take into account the edge effects that occur when the test is performed near the edges of a part. However, defect identification is significantly impaired by these edge effects. Indeed, an unidentified edge effect can lead to a failure to detect a defect, and a falsely identified edge effect can result in a false alarm during defect detection.

[0017] Thanks to the invention, the edge effects that occur when inspection is carried out near the edges of a part are taken into account. This makes it possible to improve the identification of defects in the inspected parts.

[0018] According to a preferred characteristic, the features that are extracted from the predetermined number of first and last lines of the control image are chosen from:

[0019] - the average distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image,

[0020] - the standard deviation of the distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image,

[0021] - the percentage of pixels whose value exceeds a predetermined threshold in each the predetermined number of first and last lines of the control image, and

[0022] - the maximum distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image.

[0023] According to a preferred characteristic, the classification model is chosen from: a support vector machine, called SVM, k nearest neighbors, called k-NN, a neural network or a decision tree.

[0024] According to a preferred feature, a group of pixels whose value exceeds a predetermined threshold and which are adjacent either horizontally, vertically, or diagonally to the control image without edge effect is an indication in the control image without edge effect, which may indicate a defect in the mechanical part.

[0025] According to a preferred feature, an indication in the control image without edge effect, which may indicate a defect in the mechanical part, is detected as corresponding to a defect in the mechanical part when the pixels of the pixel group corresponding to the indication are distributed over at least a predetermined number of consecutive and contiguous lines of the control image.

[0026] According to a preferred feature, the non-destructive testing method further includes calibration of the testing device prior to eddy current testing of the mechanical part by the testing device.

[0027] According to a preferred feature, the non-destructive testing method further includes a calibration of the testing device after the eddy current testing of the mechanical part by the testing device.

[0028] The invention also relates to a non-destructive testing device for a mechanical part comprising means adapted for:

[0029] - obtain a control image of the mechanical part,

[0030] - detect edge effects in the control image, by extraction of characteristics in a predetermined number of first and last lines of the control image and application of a learned classification model, to classify the predetermined number of first and last lines of the control image into lines with edge effect and lines without edge effect,

[0031] - set the pixel values ​​of lines classified as lines with edge effect to zero, to form a control image without side effects,

[0032] - determine indications in the control image without edge effect, which can indicate a defect in the mechanical part,

[0033] - detect indications corresponding to a defect in the mechanical part among the determined indications.

[0034] The device has advantages similar to those previously presented.

[0035] In a particular embodiment, the steps of the process according to the invention are implemented by computer program instructions.

[0036] Consequently, the invention also relates to a computer program on an information medium, this program being capable of being implemented in a computer, this program comprising instructions adapted to the implementation of the steps of a process as described above. Brief description of the drawings

[0037] Other features and advantages will become apparent from the following description of a preferred embodiment given by way of non-limiting example, described with reference to the figures in which:

[0038] [Fig. 1] illustrates a non-destructive testing device using eddy currents according to an embodiment of the invention.

[0039] [Fig.2] illustrates a non-destructive testing method using eddy currents according to an embodiment of the invention.

[0040] [Fig.3] illustrates a pre-calibration of the non-destructive testing device, included in the process of [Fig.2].

[0041] [Fig.4] illustrates a non-destructive testing step of a mechanical part by eddy currents, included in the process of [Fig.2].

[0042] Identical, similar or equivalent parts of the different figures bear the same numerical references so as to facilitate the transition from one figure to another.

[0043] The different parts represented in the figures are not necessarily shown on a uniform scale, in order to make the figures more legible.

[0044] The different possibilities (variants and embodiments) should be understood as not being mutually exclusive and can be combined with each other.

[0045] Detailed explanation of specific implementation methods

[0046] According to a preferred embodiment shown in [Fig.1], a non-destructive testing device for a mechanical part by eddy currents includes a base 1 for receiving the mechanical part to be tested 2. The mechanical part to be tested is a part made of an electrically conductive material, for example a metallic part.

[0047] For example, the mechanical part 2 is an aeronautical part, in particular a turbine disc.

[0048] The mechanical part to be inspected 2 has at least one examination area 20 defined on the surface of the mechanical part 2. The examination area extends over all or part of the surface of the mechanical part.

[0049] The non-destructive testing device also includes an eddy current probe 3 mounted on a probe movement system 4. The movement system 4 is motorized and adapted to impart movement to the probe 3 so that it traverses the examination area of ​​the surface of the mechanical part to be inspected 2.

[0050] The non-destructive testing device includes a control module 5 adapted to control the movement system 4 and the operation of the eddy current probe 3. The probe 3 thus travels across the inspection area while an alternating current flows through it. Induced current lines are generated on the surface of the part being inspected 2. The raw signal analyzed during the eddy current testing corresponds to the probe's impedance. This impedance varies in the event of a defect in the part being inspected.

[0051] The control module 5 receives the data resulting from the control of a mechanical part 2. Alternatively, a specific data processing module receives the data resulting from the control of a mechanical part 2.

[0052] The control module 5 has the general structure of a computer. In particular, it includes a processor 100 executing a computer program implementing the method according to the invention, a memory 101, an input interface 102 and an output interface 103. The control module includes a human-machine interface and can be associated with a display screen.

[0053] These different elements are conventionally connected by a bus 105.

[0054] The processor 100 executes the processes described below. These processes are carried out in the form of computer program code instructions which are stored in memory 101 before being executed by the processor 100.

[0055] The output interface 103 is connected to the eddy current probe 3 and its displacement system 4 so as to transmit to them the operating instructions defined by the processor.

[0056] The input interface 102 is connected to an output of the eddy current probe 3 and is intended to receive the quantities measured by the probe.

[0057] The data resulting from the control of a mechanical part consists of amplitude and phase data associated with probe position data. All of this data is represented in the form of a control image, or C-scan map, which is stored in memory 101.

[0058] Amplitude values ​​are expressed as a percentage of the control image height. Phase values ​​are expressed in degrees.

[0059] The operation of the non-destructive testing device for a mechanical part by eddy currents is represented in [Fig.2], in the form of a process comprising steps E1 to E4 implemented in the control module.

[0060] Step E1 is a pre-calibration of the non-destructive testing device. Step E1 is detailed in [Fig. 3] in the form of a flowchart comprising steps E10 to E14. The pre-calibration is a preparation of the non-destructive testing device aimed at ensuring that the device is correctly calibrated to detect and measure any defects in the mechanical part to be inspected.

[0061] In step E10, a reference part is placed on the base of the non-destructive testing device. The reference part is made of the same material, has the same shape and size as the part to be tested, and contains at least one known artificial defect, or reference defect. In the following, the reference part is considered to have a single defect. The reference defect is, for example, machined by electrical discharge machining (EDM).

[0062] The control module 5 controls the movement system 4 and the operation of the eddy current probe 3 to perform eddy current testing of the reference part. A given inspection area, including the reference defect of the reference part, is scanned by the probe 3. The data resulting from the inspection of a mechanical part are amplitude and phase data associated with the probe's position data. With respect to the reference part, the amplitude and phase data associated with the probe's position data are known a priori, since the reference defect is known.

[0063] The sensitivity of the measurement chain is adjusted so as to obtain an amplitude on standard fault of 70% of the control image height for a phase of 90°.

[0064] The set of amplitude and phase data associated with the probe position data forms a control image which is stored.

[0065] The next step El 1 is a thresholding of the calibration control image formed in the previous step. The thresholding involves comparing the amplitude value of each pixel of the control image with a predefined threshold and forming a binary image in which the value 1 is assigned to pixels whose amplitude is greater than or equal to the threshold and the value 0 is assigned to the other pixels.

[0066] Thus, for a control image c of dimension N x M, y) being the amplitude value in percentage of a pixel of row 1 and column j, the thresholding uses a threshold, a binary image c' of dimension N x M is formed.

[0067] The amplitude value c(i, j) is compared to the threshold s, and if it is greater than or equal to the thresholds, the binary value c'(i, j) of the pixel in row i and column j of the binary image c' receives the value 1, and otherwise receives the value 0:

[0068] sic(i,J) >s otherwise

[0069] The next step E12 is a processing of the binary image c' to determine regions containing pixels of value 1, called regions of interest.

[0070] A region of interest is defined as any group of pixels with a value of 1 and adjacent either horizontally, vertically, or diagonally in the binary image.

[0071] A region of interest in the binary image corresponds to a region of interest in the control image. A region of interest in the control image is defined as any group of pixels with an amplitude value greater than the threshold s, and adjacent either horizontally, vertically, or diagonally in the control image.

[0072] It should be noted that the calibration control image contains only one known defect to be detected. Step E12 must therefore result in a single region of interest corresponding to the reference defect. This is assumed to be the case in the following.

[0073] The next step E13 is an analysis of the region of interest. The region in the calibration control image that corresponds to the region of interest determined in the previous step in the binary image is considered. The maximum amplitude in the considered region and the phase corresponding to the maximum amplitude are determined.

[0074] The next step E14 is a comparison of the result of the previous step with an expected result consisting of the amplitude and phase data associated with the probe position data which are known a priori.

[0075] If the result of the measurements carried out is sufficiently close to the expected result, then the pre-calibration of the non-destructive testing device is complete.

[0076] Otherwise, the sensitivity of the measurement chain is modified and steps E10 to E14 are repeated.

[0077] The pre-calibration of the non-destructive testing device is completed when the comparison carried out in step E14 indicates that the measurement result is within a predetermined acceptable range relative to the expected result.

[0078] The calibration step El is followed by step E2 in which the part to be checked is processed.

[0079] Step E2 is detailed in [Fig.4], in the form of a flowchart comprising steps E20 to E24.

[0080] At step E20, the part to be inspected 2 is placed on the base of the non-destructive testing device.

[0081] The control module 5 controls the movement system 4 and the operation of the eddy current probe 3 to perform the eddy current control of the part to be controlled 2. The given examination area 20 of the part to be controlled is traversed by the probe 3. The data resulting from the control of the part to be controlled are amplitude and phase data associated with position data of the probe.

[0082] The set of amplitude and phase data associated with the probe position data forms a control image which is stored.

[0083] The next step E21 is a thresholding of the control image for the part to be controlled. The thresholding involves comparing the amplitude value of each pixel of the control image with a predefined threshold and forming a binary image in which the value 1 is assigned to pixels whose amplitude is greater than or equal to the threshold and the value 0 is assigned to the other pixels.

[0084] This step uses the same threshold as for step El 1 of thresholding the calibration control image.

[0085] Thus, for a control image c of dimension Ar x M, j] being the amplitude value in percentage of a pixel of row * and column j, the thresholding uses the thresholds, a binary image C' of dimension N x M is formed.

[0086] The amplitude value C(i, j) is compared to the threshold s, and if it is greater than or equal to the threshold s, the binary value C'(i, j) of the pixel in row i and column j of the binary image C' receives the value 1, and otherwise receives the value 0:

[0087] , f 1 sic(i,j)>S l 0 otherwise

[0088] The result of step E21 is the binary image C' for the part to be controlled.

[0089] The next step E22 is edge effect detection in the binary image formed in the previous step, by feature extractions and application of a learned classification model. It should be noted here that we are working on the binary image, but the edge effect is a phenomenon present in the control image.

[0090] Edge effect in an eddy current inspection image is a phenomenon that occurs when the inspection is performed near the edges of the part being inspected. This can create interference in the inspection image data, as the eddy current fields are disturbed near the edges, which can lead to falsely high or falsely low results.

[0091] According to the invention, a predetermined number of lines at the beginning and end of the control image of a specific area are analyzed. For example, only the first two and last two lines of the control image of a specific area are to analyze. Indeed, only the beginning and end of the inspection of an area of ​​the part can be altered by this edge effect.

[0092] Lines altered by an edge effect are detected in order to exclude them from further processing.

[0093] For this purpose, an annotated database is used, containing "edge effect" lines and "non-edge effect" lines. This database is constructed from inspection images of mechanical parts, these images having been previously obtained with the inspection device. The annotations on the lines of these inspection images designate, on the one hand, lines with edge effect and, on the other hand, lines without edge effect. The annotations are made by a certified inspector who marks the lines as having an edge effect or not.

[0094] Each control image used to construct the annotated database corresponds to a binary image constructed by thresholding the pixel amplitude values, as previously explained.

[0095] We are working here on annotated binary lines.

[0096] Features are extracted from the annotated binary lines of this database, namely, for a given line: - The average distance between peaks of the given line:

[0097] __LV^Yn „ V °ù P — {P* P^ • • •, Pn] is the set of peaks of the ^d “ (n-1) ' Pi) given line and Pi represents the position of each peak. Note that the peaks correspond to pixels with a value of 1. - The standard deviation of the distance between peaks of the given line:

[0098] H w_j \2 , where d, represents each individual distance between two cr = y (1) i ^dJ consecutive peaks, that is: = pM-p and P represents the average of these distances; - The percentage of pixels with a value of 1, corresponding to the pixels in the control image whose amplitude is greater than the threshold s, or pixels outside the criteria:

[0099] m = , where H is the number of pixels with a value of 1 and T is the total number of pixels in the given line; - The maximum distance between two peaks:

[0100] n -maJn JP°urï = (l,

[0101] These characteristics make it possible to identify lines which have edge effects.

[0102] A specific percentage, for example 70%, of the annotated rows in the database is used to train a classification model. It is possible to use for the classification model for example a support vector machine, called SVM, k nearest neighbors, called k-NN, a neural network or a decision tree.

[0103] The classification model and hyperparameters used to control the learning process of the classification model are chosen using cross-validation and grid search to optimize the performance of the classification model. Hyperparameters are parameters external to the model that are defined before the learning process. They influence how the model learns from the data. In the context of a classification model, hyperparameters can include elements such as the learning rate, the number of iterations, the size of the layers in a neural network, or the regularization parameter.

[0104] The learned classification model is then tested on the other annotated rows of the database which were not used to learn the classification model, i.e. for example 30% of the annotated rows.

[0105] Thus, edge detection in the control image obtained by eddy currents is performed on the first two and last two lines of the control image, by extracting features from the first two and last two lines of the binary image formed in step E21 and corresponding to the control image, and using the learned classification model. The extracted features are those described above: the average distance between peaks of each of the first two and last two lines of the binary image, the standard deviation of the distance between peaks of each of these lines, the percentage of out-of-criteria pixels, and the maximum distance between two peaks.

[0106] This allows the two classes to be identified: "lines with edge effect" and "lines without edge effect" in the control image being processed.

[0107] Lines with edge effects are not to be considered in the following. The amplitude values ​​of the pixels of lines detected with edge effects are therefore set to zero.

[0108] The results are a control image and an associated binary image in which the edge effects have been removed.

[0109] The next step E23 is the identification of regions of interest in the binary image. As before, a region of interest is defined as any group of pixels with a value of 1 that are adjacent either horizontally, vertically, or diagonally in the binary image. Each region of interest constitutes an indication, which is information that may indicate a defect in the inspected part.

[0110] A region of interest in the binary image corresponds to a region of interest in the control image. A region of interest in the control image is defined as any group of pixels with an amplitude value greater than the threshold s, and adjacent either horizontally, vertically, or diagonally in the control image.

[0111] The next step E24 is the classification of the regions of interest identified in the previous step into acceptable indication or unacceptable indication.

[0112] An unacceptable indication corresponds to a defect in the mechanical part.

[0113] An unacceptable indication corresponds to a region of interest containing pixels of value 1 and adjacent either horizontally, vertically, or diagonally in the binary image, these pixels being distributed over at least a predetermined number, for example three, consecutive and contiguous lines of the binary image.

[0114] Conversely, an acceptable indication does not correspond to a defect.

[0115] An acceptable indication also corresponds to a region of interest comprising pixels of value 1 and adjacent either horizontally, or vertically, or diagonally in the binary image, but these pixels are not distributed over at least the predetermined number, for example three, consecutive and contiguous lines of the binary image.

[0116] Step E2 of the mechanical part inspection is followed by step E3, which is a post-calibration of the inspection device. The purpose of the post-calibration is to verify that the device has remained accurate and reliable throughout the measurements performed.

[0117] Post-calibration is similar to the pre-calibration described above. Post-calibration confirms that no malfunction of the device occurred during the mechanical part inspection process that could compromise the validity of the results.

[0118] It is assumed in the following that the result of the post-calibration indicates that the control device has remained correctly calibrated. Otherwise, the entire process must be restarted from the beginning.

[0119] Step E3 is followed by step E4, which is a classification of the controlled mechanical part in step E2, based on the result of the control step.

[0120] If the part inspection result includes at least one unacceptable indication, then the part is classified as having a defect. The position and size of each defect identified by the unacceptable indication(s) are determined from the results of the unacceptable indication determination.

[0121] Data processing also makes it possible to characterize each indication by determining its maximum and minimum amplitudes as well as the phase associated with these amplitudes. The phases of the indication signals are compared to the phase of the reference fault signal measured during the pre-calibration step E1 in order to determine data on the nature of the indication.

[0122] Conversely, if the result of the part inspection does not include any unacceptable indications, then the part is declared sound.

[0123] In all cases, the result is stored, for example by generating a CSV type file which indicates whether the part is sound or defective.

[0124] Thanks to this information, only the defective parts will be analyzed subsequently by an operator. The sound parts do not require inspection by an operator.

Claims

Demands

1. A method for non-destructive testing of a mechanical part by means of an eddy current testing device, the method comprising the steps of: - obtaining (E20) by the eddy current control device a control image of the mechanical part, - detection (E22) of edge effect in the control image, by extracting features from a predetermined number of first and last lines of the control image and applying a learned classification model, to classify the predetermined number of first and last lines of the control image into lines with edge effect and lines without edge effect, - Zeroing (E22) the pixel values ​​of the lines classified as lines with edge effect, to form a control image without edge effect, - determination (E23) of indications in the control image without edge effect, which may indicate a defect in the mechanical part, - detection (E24) of indications corresponding to a defect in the mechanical part among the determined indications.

2. A non-destructive testing method according to claim 1, wherein the features extracted from the predetermined number of first and last lines of the test image are selected from: - the average distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image, - the standard deviation of the distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image, - the percentage of pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image, and - the maximum distance between pixels whose value exceeds a predetermined threshold in each of the predetermined number of first and last lines of the control image.

3. A non-destructive testing method according to claim 1 or 2, wherein the classification model is chosen from: a support vector machine, known as SVM, k nearest neighbors, known as k-NN, a neural network or a decision tree.

4. A non-destructive testing method according to any one of claims 1 to 3, wherein a group of pixels whose value exceeds a predetermined threshold and which are adjacent either horizontally, vertically, or diagonally to the control image without edge effect is an indication in the control image without edge effect, which may indicate a defect in the mechanical part.

5. A non-destructive testing method according to claim 4, wherein an indication in the control image without edge effect, which may indicate a defect in the mechanical part, is detected as corresponding to a defect in the mechanical part when the pixels of the pixel group corresponding to the indication are distributed over at least a predetermined number of consecutive and contiguous lines of the control image.

6. Non-destructive testing method according to any one of claims 1 to 5, further comprising a calibration (El) of the testing device prior to the eddy current testing of the mechanical part by the testing device.

7. Non-destructive testing method according to claim 6, further comprising a calibration (E3) of the testing device after eddy current testing of the mechanical part by the testing device.

8. A non-destructive testing device for a mechanical part comprising means adapted for: - obtaining a control image of the mechanical part, - detecting edge effects in the control image, by extracting features from a predetermined number of first and last lines of the control image and applying a learned classification model to classify the predetermined number of first and last lines. - divide the control image into lines with edge effect and lines without edge effect, - set the pixel values ​​of the lines classified as lines with edge effect to zero, to form a control image without edge effect, - determine indications in the control image without edge effect, which may indicate a defect in the mechanical part, - detect indications corresponding to a defect in the mechanical part among the determined indications.

9. A computer program comprising instructions for carrying out the steps of the process according to claim 1 when said program is executed by a computer.

10. Computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the process according to claim 1.

Images