Method for measuring or inspecting reflective surfaces by means of deflectometry

Abstract

The invention relates to a method for measuring or inspecting reflective surfaces by means of deflectometry, wherein the optical patterns (3) used in the deflectometry are automatically calculated beforehand as a function of the target geometry of the surface (4) to be measured and in the course of the calculation, using a loss function, are optimized in respect of features of the surface (4) that are to be determined. The calculation is carried out using a differentiable ray tracing method. The proposed method makes it possible to inspect and / or measure primarily relatively complex surface geometries with high accuracy.

Description

[0001] Method for measuring or inspecting reflective surfaces using deflectometry

[0002] Technical application area

[0003] The present invention relates to a method for measuring or inspecting reflective surfaces by means of deflectometry, in which different optical patterns are displayed one after the other on a screen, which are recorded via a reflection on a surface to be measured or inspected with a spatially resolving optical detector and then evaluated.

[0004] Deflectometry is an established measurement technique for measuring or inspecting reflective surfaces, such as painted car panels, mirrors, or optical components. Typically, several optical patterns are displayed one after the other on a screen, viewed by a camera via the reflection in the surface to be measured. By decoding the patterns, a relationship can be established between camera pixels and screen pixels. Thus, the local inclination of the surface can be reconstructed by triangulation and used for surface measurement or to detect material defects.

[0005] State of the art

[0006] Currently, striped sinusoidal patterns with different spatial frequencies are used as optical patterns in deflectometry. These allow good results to be achieved when measuring surfaces with simple geometries, i.e., low curvature and without complex structures. However, with more complex object geometries, the reconstruction quality and defect detection accuracy decrease.

[0007] From S. Kammei, "Automated optimization of measurement setups for the inspection of specular surfaces", Proc. SPIE 4567, Machine Vision and Three-Dimensional Imaging Systems for Inspection and Metrology II, pages 199 to 206, a method is known in which the geometric parameters of a deflectometric arrangement, for example the camera pose, the screen orientation, etc., are optimized for the respective object to be measured using a simulation. The optimization is carried out using CAD data of the object to be measured and a model of the imaging function of the optical system using a ray tracing method. The generation of object-adapted optical patterns is not discussed in this publication.

[0008] The object of the present invention is to provide a method for measuring or inspecting reflective surfaces by means of deflectometry, which enables the measurement or inspection of more complex surface geometries with higher accuracy than is achieved when using strip-shaped sinusoidal patterns.

[0009] Description of the invention

[0010] The object is achieved by the method according to claim 1. Advantageous embodiments of the method are the subject of the dependent

[0011] Patent claims or can be taken from the following description and the example of implementation.

[0012] In the proposed method for measuring or inspecting reflective surfaces using deflectometry, different optical patterns are displayed one after the other on a screen in a known manner, recorded via a reflection on a surface to be measured or inspected using a spatially resolved optical detector, in particular a camera, and then evaluated. Depending on the application, the evaluation is carried out, for example, to reconstruct the surface profile or the spatially resolved inclination of the surface and / or to detect material defects on the surface. The proposed method is characterized in that the optical patterns used are automatically calculated in advance as a function of a target geometry of the surface to be measured or inspected and are optimized during the calculation using a cost function for specific surface features to be determined.The optical patterns are calculated based on the target geometry of the surface to be measured or inspected and a deflectometric arrangement specified for performing the deflectometry using a differentiable ray tracing method with the gradient descent technique. Suitable available ray tracers, for example the Mitsuba 3 ray tracer (see e.g. https: / / mitsuba.readthedocs.io / en / latest / ), can be used for this purpose. The deflectometric arrangement includes the position and orientation of the screen and camera as well as the surface to be measured or inspected or the object with this surface. A sequence of optical patterns is thus calculated or generated in advance that is specifically adapted to the concrete target geometry of the surface to be measured or inspected in order to enable optimal results, particularly in terms of measurement uncertainty.The method is based on modern techniques of differentiated ray tracing from computer graphics. Ray tracing methods are capable of generating photorealistic images of synthetic scenes. By using a ray tracing method that supports automatic differentiation, a quality measure can be defined for each generated image, for example, the measurement uncertainty, which can then be used to optimize the optical patterns.

[0013] By specifically adapting the optical patterns used in deflectometry to the surfaces to be measured or inspected and the respective objective of the measurement or inspection of these surfaces, test objects or surfaces with more complex surface geometries can be measured or inspected with greater accuracy than is the case when using the strip-shaped sinusoidal patterns used previously.

[0014] In one possible embodiment of the proposed method, in which the geometry of the surface to be measured or inspected is to be determined, the deviation between a geometry reconstructed using the patterns and the target geometry is minimized when calculating the optical patterns. In a further embodiment, in which material defects in the surface to be measured or inspected are to be detected, the detection sensitivity or detection accuracy of defects synthetically introduced into the target geometry and / or a target reflectance distribution of the surface is maximized when calculating the optical patterns.

[0015] The method can be used advantageously in all applications of deflectometry technology where more complex surfaces, i.e. surfaces with greater curvature or a special structure, are to be measured or inspected.

[0016] Short description of the drawings

[0017] The proposed method is explained again below using an example in conjunction with the drawings.

[0018] Fig. 1 is a schematic representation of a deflectometric arrangement and a framework for calculating the optical patterns used in deflectometry;

[0019] Fig. 2 is a diagram of a conventional optical pattern for deflectometry; and Fig. 3 is an example of an optical pattern for deflectometry used in the proposed method.

[0020] Ways to implement the invention

[0021] In Figure 1, on the left-hand side, a typical deflectometric arrangement is shown schematically. It comprises a screen 1 and a camera 2 which are arranged in such a way that an optical pattern 3 displayed on the screen is recorded by the camera 2 via a reflection on the surface 4 to be measured. By comparing the pattern 3 displayed on the screen with the pattern recorded by the camera 2 via the reflection, the surface shape is reconstructed in a deflectometric reconstruction unit 5. Alternatively or additionally, the pattern modulated by the inspected surface can be used in a defect detection network 6 to detect defects in the surface 4 to be measured. In principle, any suitable machine learning method can be used for this, for example also classic methods based on feature extraction and subsequent classification, e.g.Support Vector Machines etc. ä .

[0022] In this example, a complete deflectometric measurement setup, modeled using differentiated ray tracing, is combined with a reconstruction method formulated using a differentiated method in a common framework. For the surface to be measured, a test specimen geometry given in this example by a CAD model, the displayed screen patterns are then optimized with respect to measurement uncertainty using gradient descent. There are two optimization options, for which two optimization functions, li and I2, are used below.

[0023] Optimization functional li describes the deviation between the reconstructed surface geometry and the specified target geometry (given by the 3D model of the test object), also referred to below as ground truth. Optimization functional I2 describes how well synthetically introduced material defects were detected by a corresponding method (e.g. deep neural network). The optimization is carried out by differentiable inverse rendering. The reconstruction is carried out using known deflectometric reconstruction methods. Both optimization functionals can be derived with respect to the pattern displayed on the screen or the displayed pattern sequence. The screen patterns can then be successively updated by gradient descent such that the corresponding optimization functional is minimized, i.e. the reconstruction accuracy is optimized for li and the defect detection accuracy is optimized for I2.This is also indicated schematically in Figure 1 .

[0024] While with the optimization functional I2 any pattern is possible, i.e. all pixels of the displayed pattern are released as optimization parameters, with regard to the optimization functional li the patterns must still allow a deflectometric reconstruction, i.e. a determination of the correspondence from camera pixel to screen pixel. For this purpose, for example, a parameterization of the screen patterns can be used in which deformed wavefronts are permitted in the optical patterns instead of the conventional planar wavefronts. Figure 2 shows an example of a conventional striped sinusoidal pattern, and Figure 3 shows an example of an optical pattern generated according to the proposed method in which deformed wavefronts are used. Different approaches to modeling the deformation can be chosen, e.g.a polynomial of a certain degree or a Fourier series (composition of cos and sin patterns) with a fixed number of coefficients. This severely restricts the pattern space, and optimization is limited to determining the coefficients of the chosen model.

[0025] The proposed method requires a ray tracing method that supports automatic differentiation, also referred to in the present patent application as a differentiable ray tracing method. Such methods are available; for example, the Mitsuba 3 ray tracer can be used. The target geometry of the object to be measured / inspected, for example as a CAD model, is provided as input data for calculating the optical patterns. Furthermore, the bidirectional reflectance distribution function of the surface of the object to be measured / inspected may be required. The calculation also requires the parameters of the deflectometric arrangement, i.e. the position and orientation of the screen and optical detector or camera, parameters of the optical detector (e.g. optical parameters such as focal length, pixel size, etc.), as well as corresponding parameters of the screen (resolution, dimensions).If necessary, positions / size / properties of synthetic defects can also be specified if I2 is to be optimized.

[0026] The proposed method allows optical patterns to be generated and used as screen patterns that are optimal for the specific object or surface to be measured or inspected with respect to a problem-specifically selectable cost function, such as measurement uncertainty. This allows the desired features or properties to be captured with greater accuracy than is the case with the striped sinusoidal patterns used previously.

[0027] Reference symbol list

[0028] 1 Screen 2 Camera

[0029] 3 optical pattern

[0030] 4 Surface to be measured

[0031] 5 def lectometric reconstruction unit

[0032] 6 Defect detection network

Claims

Patent claims 1. A method for measuring or inspecting reflective surfaces by means of deflectometry, in which different optical patterns (3) are displayed one after the other on a screen (1), which optical patterns are recorded by a spatially resolving optical detector (2) via a reflection on a surface (4) to be measured or inspected and are then evaluated, characterized in that the optical patterns (3) are calculated automatically in advance as a function of a target geometry of the surface (4) to be measured or inspected and are optimized during the calculation with a cost function for features of the surface (4) to be determined, wherein the calculation of the optical patterns (3) is carried out on the basis of the target geometry of the surface (4) and a deflectometric arrangement predetermined for carrying out the deflectometry using a differentiable ray tracing method with the technique of gradient descent.

2. Method according to claim 1, characterized in that in order to determine a geometry of the surface (4) as a feature to be determined in the calculation of the optical patterns (3), a deviation between a reconstructed when using the patterns geometry and the target geometry is minimized.

3. Method according to claim 1 or 2, characterized in that for the determination of material defects in the surface (4) as a feature to be determined in the calculation of the optical patterns (3), a detection sensitivity or detection accuracy of defects synthetically introduced into the desired geometry and / or a desired reflectance distribution is maximized.

4. Method according to one of claims 1 to 3, characterized in that the desired geometry of the surface (4) is provided as a CAD model.

Images