ORDER FOR DETERMINING THE SURFACE TEXTURE OF COMPONENT SURFACES

Description

[0001] The invention relates to an arrangement for determining the surface quality of component surfaces, in particular defects, especially cracks.

[0002] To date, it is known to conduct a penetrant test according to DIN EN 4179 and NAS 410. This involves using a so-called fluorescent crack detection oil. The surface is wetted with this oil and then partially removed by cleaning with a liquid. Components of a material that can be excited to fluorescence when irradiated with electromagnetic radiation of at least one wavelength remain in areas where depressions, particularly cracks, are present. If the surface is irradiated with electromagnetic radiation suitable for fluorescence excitation, visual detection is achieved, which can be exploited by the testing personnel. This may lead to subjective errors on the part of the testing personnel.

[0003] Another established testing method involves coupling sound waves into a component's surface using a transducer, and detecting reflected sound waves within the component. This requires a high level of equipment complexity and can only be used on surfaces large enough to accommodate the transducer. Furthermore, it is correspondingly time-consuming.

[0004] When testing with X-rays, the high cost is a disadvantage and radiation protection requirements must be met.

[0005] For example, JP 2008 180618 A discloses possibilities for the optical detection of surface defects.

[0006] WO 2015 / 133287 A1 relates to a device, a method and a program for classifying surface textures.

[0007] A device and a method for determining the thickness of layers on surfaces are described in WO 2013 / 035726 A1.

[0008] It is therefore an object of the invention to provide possibilities for the non-destructive determination of the surface quality of component surfaces, which lead to results with sufficient accuracy in a significantly shorter time and can thus even be used for inline measurement in manufacturing processes.

[0009] According to the invention, this object is achieved by an arrangement having the features of claim 1. Advantageous embodiments and further developments of the invention can be realized with features defined in subordinate claims.

[0010] The inventive arrangement for determining the surface quality of components comprises a plurality of detectors designed for the spatially resolved spectral analysis of electromagnetic radiation within a wavelength interval. These detectors are arranged in a row or a row and column arrangement. The detectors are connected to an electronic evaluation unit and arranged such that electromagnetic radiation emitted by a broadband radiation source impinges on the detectors either after being reflected from the surface of a component, a layer formed on the surface of a component, and / or after passing through a component transparent to the electromagnetic radiation.The irradiation is carried out in such a way that a homogeneous intensity of electromagnetic radiation is maintained on a surface from which the electromagnetic radiation is reflected or transmitted. The surface to be detected simultaneously should therefore be irradiated with a homogeneous intensity. In broadband irradiation, electromagnetic waves within a wavelength interval are directed onto the respective component surface to be examined.

[0011] The electronic evaluation unit is designed in such a way that the measurement signals recorded by the detectors with spatial and wavelength resolution can be assigned to a specific wavelength range within a wavelength interval and to an individual location point within a predeterminable sub-area of ​​the irradiated surface.

[0012] According to the invention, the electronic evaluation unit uses the measurement signals recorded for individual locations to create a multidimensional image of the respective irradiated surface. This image can be displayed, for example, on a screen. Using intensity analysis, an evaluation can then be performed to determine the presence of defects. This makes it possible to achieve an evaluation independent of inspection personnel. It is even possible to quantify defects and / or objectively determine the quality of the respective surface finish.

[0013] In addition, a three-dimensional data structure consisting of one wavelength-resolved and two spatially resolved dimensions (hypercube) is formed with the totality of all intensities recorded at locations of the respective irradiated area.

[0014] Subsequently, a data reduction can be carried out for all wavelength-resolved measurement signals of the sub-area, in which meaningful features are selected and the resulting feature set is subjected to a regression procedure using a regression model stored in an electronic memory, which was determined on the basis of feature sets that were previously determined on samples whose surface properties were determined using a different measuring method that preferably has a higher measurement accuracy, whereby a statement about the surface properties of the relevant sub-area of ​​the irradiated surface can be derived, wherein at least one sub-area is used to evaluate the surface properties.

[0015] A surface finish of components examined using the invention should belong to the same class of surface finishes, which should have a comparable structure or composition, as those previously examined using another measuring method.

[0016] The functional relationship between adequately selected sets of surface properties determined using another measurement method can be determined, for example, using linear or non-linear regression, a partial least square algorithm (PLS), a neural network, a combination of at least two of these methods, or other regression methods (regression model).

[0017] The electronic memory should therefore contain the regression model created using the data sets acquired with the inventive system from samples that were subjected to data reduction and feature extraction, and whose surface properties were determined using a different measurement method. Using a regression model stored in the electronic memory, the surface properties of a similarly examined sample of the same sample class with a similar structure can be determined from the data sets acquired with the inventive system.

[0018] The electronic evaluation unit performs data reduction and feature extraction from the data sets acquired with the inventive arrangement. This can involve evaluating the spectral information first, followed by the spatial information. A reverse sequence or any combination of more than two individual steps for data reduction and feature extraction is also possible.

[0019] Data reduction and feature extraction can be performed using principal component analysis (PCA), parameterization of texture information, determination of statistical parameters, such as averaging and / or standard deviation, and combinations thereof.

[0020] When using principal component analysis, the n intensities of the wavelengths (spectra) of all location points are transformed by a coordinate transformation into a new orthogonal coordinate system - the principal component space - in which the original data have maximum variance, and where n represents the number of measured wavelengths.

[0021] The coordinate transformation is calculated by determining the n Eigenvectors (principal components) and the corresponding n eigenvalues ​​of the covariance matrix of the data set of the measured partial area. The larger the n-th eigenvalue, the more the corresponding nth principal component describes the original variance, i.e. the eigenvector with the largest eigenvalue is the 1st principal component of the data set and describes the majority of the original variance of the data set. The eigenvector with the lowest eigenvalue is the nth principal component of the data set and does not describe any relevant properties of the data set. Only a certain number of principal components are taken into account - often the first three to six, which already ensure a sufficient description - for example > 95% - of the original variance of the data set. The number of principal components to be considered - and thus the dimensionality of the principal component space - can be selected based on criteria, for example based on the proportion of the total variance or a scree test.

[0022] The coordinates of the spectra in the newly spanned principal component space are the so-called score values, which sufficiently characterize the corresponding location point.

[0023] With the electronic evaluation unit, data reduction and feature extraction can advantageously be carried out in such a way that the data acquired according to the invention are evaluated in the same way as the data from the samples, based on whose known surface properties the regression model was created. Based on the data set acquired according to the invention, the intensities of the electromagnetic radiation recorded with spatial and wavelength resolution, the surface properties of the component surface are then determined using the regression model.

[0024] The evaluation of the data acquired according to the invention can preferably be carried out as follows: Variant a) Using a principal component analysis of the spectral information of all locations of the partial area, the first six principal components, including the score values, are preferably determined. Since the data acquired according to the invention represent all locations of an examined area, the areal distribution of the score values ​​can be specified for each principal component. The different areal distributions of the score values ​​are quantified using various statistical parameters that are determined based on all score values ​​of a principal component of the respective partial area. In particular, these are the variance, the interquantile distance, or the mean absolute deviation. This procedure can be applied to all principal components.The surface texture is determined by the electronic evaluation unit on the basis of a regression model, such as a partial least square regression model (PLS), whereby the parameters may have been subjected to further prior feature extraction, preferably via another principal component analysis (PCA).

[0025] The regression model used (here PLS) was previously determined using samples of the same sample class with a comparable structure, whose surface properties were determined using a different measuring method, whereby all steps of feature extraction were carried out analogously to the method described above.

[0026] Variant b) Using a principal component analysis of the spectral information of all recorded location points of the respective examined sub-area, one or more principal components, in particular the score values ​​of the location points per principal component, are calculated. These score values ​​can be specified as distributed across the sample area. The location points can be described by various parameters of the score values, in addition to the actual value and in particular by three times the standard deviation from the mean value of all score values ​​of the respective sub-area. If this parameter at a location deviates by a certain threshold value from the mean value of this parameter of all location points, it is classified as a defect. If such a location point is located next to a location point already classified as defective, groups of location points classified as defective (defect areas) can be identified in this way.For further feature extraction and thus for determining the "surface texture" feature, the defect areas are divided into different groups. The grouping of the defect areas determined according to the invention can be carried out by determining parameters that provide spatial information about the defect area, in particular the area, extent in the direction of both spatial coordinates, center point, eccentricity, equivalent diameter, circumference, filling of the enclosing polygon area, and ratio of the area to the enclosing convex surface. Using a further classification method, advantageously a cluster analysis, the parameter sets describing the defects can be divided into specific defect types. A classification into four defect types has proven advantageous. If an assignment to specific, previously determined defect types is to be made, the use of a supervised classification method, e.g.A support vector method or discriminant analysis is advantageous. From the number of defects per defect type detected on the examined surface, the surface texture of the partial surface can be determined using a regression model, particularly a linear regression model. The regression model used was previously determined using samples of the same sample class with a comparable structure, whose surface textures were determined using a different measurement method. All feature extraction steps were performed analogously to the method described above.

[0027] The sizes of the examined sub-areas can be adjusted depending on the component surface being examined and can be set from micrometers to meters; the spatial resolution can also be varied.

[0028] At least 30, preferably at least 100 detectors should be arranged in a row.

[0029] At least one, preferably at least 50 detectors should be arranged in a column.

[0030] An arrangement according to the invention can be used as a hyperspectral camera with suitable radiation-shaping optical elements and electronic evaluation electronics.

[0031] The surface should be irradiated at an angle of at least 0° to <90° relative to the normal to the surface onto which the electromagnetic radiation impinges. When irradiating at least one component transparent to electromagnetic radiation, the angle to the sample normal should preferably be maintained at a minimum of 0°, i.e., the radiation should be directed as perpendicularly as possible to this surface in order to minimize the reflected portion. Irradiation and detection can also be performed with a variable angle of incidence of the electromagnetic radiation. As already stated, angles of incidence in the range of 0° to a maximum of 89° can be selected. Detection can also be limited to linearly polarized electromagnetic radiation. In this case, an advantageous orientation of one or more polarization planes before and / or after the irradiated surface can be selected.

[0032] Particularly when using detectors that can only measure one line simultaneously with spectral and spatial resolution, the detectors and the component surfaces can be moved relative to one another along at least one axis and preferably at an advantageous distance from one another. With statically fixed detectors and radiation source, a component surface can be moved along an axis. This can be achieved with a correspondingly movable table on which a sample is arranged that can be moved in an x- and, if necessary, also in a y-direction. However, unwinding from roll to roll is also possible if the respective sample is formed from one or more flexibly deformable material(s), for example in the form of a film.

[0033] The radiation source can contain elements that shape the electromagnetic radiation. In a simple design, the radiation source can be combined with a microscope. However, the radiation source can also be arranged in a hollow body, from which the electromagnetic radiation emerges diffusely and can be directed onto the surface to be irradiated. The hollow body can be a sphere or cylinder. It should be possible to irradiate a surface to be detected simultaneously and homogeneously. For a radiation source with beam-shaping optical elements, the wavelength range used should be considered when selecting the optical elements used for beam shaping.

[0034] Preferably, a diaphragm preventing the incidence of scattered electromagnetic radiation can be arranged in front of the detectors in the beam path of the electromagnetic radiation.

[0035] The radiation source can emit electromagnetic radiation whose wavelength range begins in the UV range and ends in the IR range. Radiation from the wavelength range of visible light up to the NIR range, i.e. from 250 nm to 2500 nm, is particularly preferred. However, irradiation within a narrow wavelength range, preferably in the UV range, can also be selected to excite fluorescence radiation. Within a used wavelength range, it should be possible to use as many wavelengths as possible for irradiation. The limits should be determined solely by the sensitivity range of the detectors used in terms of their sensitivity / measurement accuracy of the intensities they can detect for the respective wavelengths and the optical properties of the beam guidance components.Preference should be given to using the spectral ranges that show the greatest variance between the spectra of the respective component surfaces and have the smallest possible determination error.

[0036] At least one element with which a targeted selection of the polarization of the electromagnetic radiation can be achieved can also be present or integrated in the beam path.

[0037] The detectors used, the electronic evaluation unit, and possibly also the radiation source, can constitute a so-called hyperspectral imaging system, which can be used in the arrangement according to the invention. This allows not only spectral but also spatial information to be obtained for the respective detected portion of the sample.

[0038] In order to increase the contrast and / or the maximum achievable differences in measured intensities, in particular between surface areas that are at least almost free of defects and those in which defects are present, the procedure according to DIN EN 4179 can be used. In this case, after the respective component surface has been coated with a suspension which contains fluorescent components when irradiated with electromagnetic radiation with at least one predeterminable wavelength, and after the surface has been cleaned, an irradiation with electromagnetic radiation with which fluorescence can be excited, a visual evaluation can be carried out, preferably on an image of the intensity measurement values ​​of the recorded location points.The irradiation can be monochromatic with the fluorescence-exciting wavelength or broadband irradiation, in which the respective fluorescence-exciting wavelength is part of the wavelength spectrum of the electromagnetic radiation used.

[0039] It can be advantageous to reduce the data collected during the detection of the intensities determined at the individual locations for the individual wavelengths of a recorded wavelength range (spectrum) during the actual analysis. This allows information relevant to determining the surface texture to be separated from information irrelevant, significantly reducing the electronic computational effort and thus the required time. The use of highly complex, cost-intensive electronic computing technology is unnecessary.

[0040] The invention allows the creation of a statistical model on the basis of which the surface quality of the component under investigation can be predicted in the corresponding underlying sub-area. For this purpose, the data set can be acquired using one of several optical detectors, preferably arranged in series, which enable wavelength- and spatially resolved intensities.

[0041] The surface texture value is influenced by deviations from the ideal condition of the component under investigation. These deviations can include cracks, defects, particles, layer thickness variations, material changes, etc. These artifacts also cause light interacting with the surface of the component to be reflected, scattered, or transmitted differently (spectrally, intensity). The totality of a sufficient number of individual wavelength-resolved intensities (spectra) from different locations of a detected partial area of ​​a component surface can thus represent information about the surface texture in the observed surface area.

[0042] To determine the surface texture, the respective surface texture of the component under investigation can be determined on the basis of a measured set of features using a regression model previously established with sets of features with known surface textures measured on samples of the same sample class with a comparable structure.

[0043] The measured feature sets must be subjected to data reduction and thus feature extraction. The wavelength spectra detected at the individual locations of the respective subarea can be subjected to spectral feature extraction, for example, cluster analysis or principal component analysis. The parameters used in the regression model can therefore be the number and distribution of clusters, the score values ​​of the principal components, or their distribution.

[0044] Furthermore, image compression methods can be used for texture evaluation for feature extraction. These methods involve subjecting the intensities of individual wavelengths detected at all locations within the respective subarea, or the sum or average sum of the intensities of multiple wavelengths, or the parameters determined by spectral feature reduction and / or combinations thereof to image compression. To describe the texture information of the dataset, at least one parameter should be determined using a wavelet transform or another image compression method, such as Taylor polynomials, Fourier and cosine transforms, discrete cosine transform, or the grayscale matrix method.

[0045] Based on a calibration model (regression model) created using measured data sets from samples of the same sample class with comparable structures and known surface properties, the surface properties of partial surfaces can be predicted. The measured data sets are treated for calibration and prediction using identical feature reduction steps. Examples of regression models used include multiple linear regression analysis (MLRA), principal component regression (PCR), partial least squares regression (PLS), or a neural network.

[0046] Prerequisites for sufficient determination accuracy are homogeneous illumination of the area used for detection, so that the superposition of intensity fluctuations caused by the sample with lateral fluctuations in the illumination intensity can be avoided by implementing a laterally homogeneous light field. For small sample areas, microscope optics can be advantageously used. The arrangement according to the invention can be adapted to the respective area of ​​the component surface to be detected by using different optics, working distances, and magnifications.

[0047] The invention allows the surface texture of components to be determined within a few milliseconds to minutes. A non-contact, non-destructive measurement is possible. In many cases, the components do not require any additional treatment or other preparation. It is preferably used in quality control and inline production monitoring, eliminating the need for sampling and separate determination.

[0048] The invention will be explained in more detail below by way of example. Example 1

[0049] A 0.5 cm thick sheet of a Ti alloy measuring 8 cm x 5 cm was used for the investigation. The aim of the investigation is automated crack and defect detection. Engravings of varying widths and depths were laser-etched into this sample, simulating a defect in the Ti component. The sample was then treated with a fluorescent oil. After application, the oil was removed by intensive washing with water. Residues of the oil remain in the crack structures of the sample. The thus prepared sample was diffusely irradiated with a fluorescent light source (365 nm emission maximum). A total of 1000 x 190 optical detectors were arranged above the sample in a row-column arrangement, allowing the entire sample surface to be recorded with wavelength and spatial resolution.The optical detectors were used to capture wavelength-resolved measurements of the intensities of the radiation reflected from the sample at individual locations. A total of 190 wavelengths in the range from 400 nm to 1000 nm were considered. The sample was moved perpendicular to the array of detectors to capture the entire sample surface. The resulting spectral image of the sample can also be referred to as a hypercube.

[0050] The resulting hypercube of the sample under investigation was then processed and evaluated.

[0051] The processing of the spectra was carried out using the following steps: Mean smoothing (11 points) Determination of the baseline intensity at a wavelength where no fluorescence is measurable (675 nm) Subtraction of the baseline intensity Determination of the intensity at the emission wavelength of the fluorescent oil (510 nm)

[0052] The fluorescence intensity is proportional to the amount of fluorescent oil on the sample surface. In the defect areas of the sample surface, the fluorescent oil accumulates, resulting in a larger amount of oil and thus increased emission. Defect areas can be automatically detected based on the fluorescence intensity. For this detection and further classification, characteristic features of the defect areas can be determined. For example, the shape and size of the defect areas as well as their fluorescence intensity and distribution can be used. The characteristics thus determined enable the defects to be classified into defect classes previously determined on other (reference) samples. For classification, discriminant analysis, support vector methods, or neural networks can be used, for example. Example 2

[0053] For the tests, metal foils were coated with varying amounts of a precursor solution using a slot die. Subsequent annealing of the substrates resulted in a thin layer of the precursor solution being applied to the metal foils. This layer can contain various defects, for example, caused by contamination of the metal foil or precursor, or by problems during the coating process, resulting in defective areas that pose problems for further processing. The coated metal foil can be produced continuously using a roll-to-roll process. The goal of the test is the automatic detection of defect areas on the continuously produced samples.

[0054] For this purpose, the coated metal foils were diffusely irradiated with halogen lamps. A total of 1000 x 190 optical detectors were arranged above each sample in a row-column arrangement, allowing the entire sample surface to be recorded with wavelength and spatial resolution. The optical detectors were used to record the intensities of the radiation reflected by the sample at individual locations with wavelength resolution. A total of 190 wavelengths in the range 400 nm to 1000 nm were considered. The sample was moved perpendicular to the row of detectors to capture the entire sample surface. The resulting spectral image of the sample can also be referred to as a hypercube.

[0055] The tests were first conducted on samples produced under normal manufacturing conditions and whose quality was considered sufficient for further processing. The spectral images of the samples examined in this way were then subjected to the following processing steps: Mean smoothing (11 points) Conversion of reflection to absorption spectra Creation of a principal component model from all selected hypercubes Calculation of the residuals (R) and the T 2< value for each spectrum of the hypercubes ∘ Residuals: variability of the respective spectrum not explained by the principal component model ∘ T 2< value for each spectrum: ∘ T 2 = ∑ p = 1 P s p 2 σ sp 2 with PNumber of principal components used S p Score value of the respective principal component σ sp Eigenvalue of the corresponding principal component Calculation of a threshold value for R and T 2< . Spectra with values ​​above this threshold are considered defects

[0056] Subsequently, new, unknown samples can be examined using the same measurement system. The resulting hypercubes are analyzed using the previously determined principal component model. Areas deviating from the model (defects) can be identified by R and T 2< values ​​above the previously defined threshold. The total number of defects can be used as a measure of the quality of the produced coating. If the number of defects is too large, the corresponding sample area can be discarded or the coating process adjusted.

Claims

1. A system for determining the surface quality of component surfaces, in particular defects, wherein a plurality of detectors, which are designed for the spatially resolved spectral analysis of electromagnetic radiation within a wavelength interval, are arranged in a row or in a row and column arrangement, and the detectors are connected to an electronic evaluation unit and arranged so that electromagnetic radiation emitted by a broadband radiation source impinges on the detectors either after reflection on the surface of a component, a layer formed on the surface of a component, and / or after penetration of a component that allows the electromagnetic radiation to pass, the irradiation taking place in such a way that a laterally and temporally homogeneous intensity of the electromagnetic radiation is maintained on a surface by which the electromagnetic radiation is reflected or through the surface of which the same is transmitted, and the electronic evaluation unit being designed so that the measurement signals of the detectors detected in a spatially and wavelength resolved manner within a wavelength interval are detectable for individual location points that are situated on a predefinable surface of the component; a three-dimensional data structure, composed one wavelength-resolved dimension and two spatially resolved dimensions, being formed from the entirety of all intensities detected in a wavelength-resolved manner at location points of the particular irradiated surface, and a multidimensional image of the particular irradiated surface being creatable from the measurement signals detected for the individual location points.

2. The system according to claim 1, characterized in that measurement signals detected at a plurality of positions can each be assigned to a sub-region of the detected surface, and a data reduction can be carried out for all measurement signals of the sub-regions of the detected surface which are detected in a wavelength-resolved manner, during which meaningful features are selected and based on these selected features, information about a corresponding surface quality can be attained using a regression model which is stored in an electronic memory and ascertained in advance and which, based on the feature sets obtained in analogous fashion whose surface qualities have been ascertained.

3. The system according to any one of the preceding claims, characterized in that the electronic evaluation unit is designed so that a data reduction can be achieved by way of feature extraction using principal component analysis, extraction of texture information, averaging, determination of the standard deviation and / or combinations thereof.

4. The system according to any one of the preceding claims, characterized in that the irradiation of the surface takes place at an angle at least in the range of 0° to < 90° with respect to the normal of the particular component surface.

5. The system according to any one of the preceding claims, characterized in that the detection and evaluation can be carried out using at least one polarizer having at least one defined known polarization plane with respect to the plane of incidence or the combinations thereof.

6. The system according to any one of the preceding claims, characterized in that, during the data reduction and feature extraction, the electronic evaluation unit initially evaluates the spectral information and thereafter evaluates the location information.

7. The system according to any one of the preceding claims, characterized in that the detectors and the component can be moved along at least one axis relative and preferably at a constant distance with respect to one another.

8. The system according to any one of the preceding claims, characterized in that the radiation source comprises the optical elements that shape the electromagnetic radiation or a radiation source that emits electromagnetic radiation diffusely onto the surface, which is in particular arranged within a hollow body, and particularly preferably that a diaphragm avoiding the incidence of scattered electromagnetic radiation is arranged in the optical path of the electromagnetic radiation upstream from the detectors.

9. The system according to any one of the preceding claims, characterized in that a visual evaluation of the respective component surface can be carried out after having been coated with a suspension which contains constituent parts that fluoresce when irradiated with electromagnetic radiation having at least a predefinable wavelength and after having subsequently been cleaned, an irradiation with electromagnetic radiation by way of which fluorescence can be excited.

10. The system according to any one of the preceding claims, characterized in that at least 30 detectors are arranged in a row and at least 50 detectors are arranged in a column.

11. The system according to any one of the preceding claims, characterized in that the arrangement of detectors in rows and columns including optical elements and evaluation electronics is formed with a hyper-spectral camera.