Structured Surfaces under the Magnifying Glass
Texture Analysis on Components with a Complex Distribution of Surface Structures
In quality inspection of surfaces, the adaptive texture analysis is continuing to gain importance, because defects are detectable relatively to dominant structure of the surface. To use the adaptive texture analysis on components with different surface structures that have a complex geometric distribution, it is necessary to create masks that overlay the associated surface structure correctly. Fraunhofer IPA developed a concept that performs the creation of masks semi-automatic, which allows a fast adjustment of such an inspection.
Small defects like impact points or scratches on the surface can compromise the functionality of a component and therefore make is useless or ruin it esthetically. The industrial demand for 100% control is rising in order to detect those kinds of defects robust and reliably.
Many manufacture processes like milling using a lathe machine lead to specific patterns on the component. For an automatic control of these components the image processing must be able to detect defects on surfaces with unknown patterns.If there are several different surface structures on one component, which are distributed in a complex manner (fig. 1), masks are necessary for an inspection. Fraunhofer IPA developed a concept that performs this process semi-automatically. Possible applications for this concept can be found on all components with different surface structures and/or materials. It can be used for a functional inspection as well as an esthetic inspection.
See and Inspect Like Humans
The inspiration for defect detection on surface structures is the human ability to recognize mistakes and irregularities fast, even though the kind of surface is unknown. The inspection concept, which was developed at Fraunhofer IPA is based on this ability. Under the condition that the defect only covers a small area of the overall surface, the defect can be extracted as a noticeable deviation from the background.
To describe surface structure models are used. With different models, different characteristics of surface structures can be described. In an analysis process a chosen structure model is used and trained with the data of the surface structure.
Given that the defect only covers a small area of the overall surface structure a model is trained that describes the characteristics of the background dismissing local distortions. When a classifier is trained with this model and used on the image, it is ensured, that the evaluation of the surface structure is based on the dominant structure of the surface. Opposed to a conventional inspection, defects aren’t deviations from a prescribed structure, but distortions in the dominant structure of the image. This way deviation in the surface structure can be tolerated, because the system is adapted for every component. This allows applications for a wide range of materials.
With the help of the adaptive test method, surface defects on a very varying surface can be tested without the operator having to make any adjustments to the parameterization of the inspection method. The high-performance algorithm enables a fast and reliable inspection and is therefore applicable in inspection systems with high throughput.
Surface Inspection with Variable Image Acquisition
In addition to the application on flat surfaces in which the image acquisition is realized by matrix cameras and suitable bright-field illumination, Fraunhofer IPA developed an inspection concept for the surface inspection on rotationally symmetrical components. The image acquisition is done with a bright field line camera. A robot (or by another handling system) places the component on a turntable located directly in front of the camera. By rotating the component, the surface is captured completely by the camera, which allows a 100% inspection of the surface area. Thus even small imperfections, which make the workpiece unusable, can be detected on rotationally symmetrical components.
This test system was developed based on the image processing package EMSIS, which was also developed at the Fraunhofer IPA and includes, in addition to surface inspection, further image processing tools (inspection of internal and external threads, dimensional measurement of lengths, diameters and angles, etc.).
Components with Geometrically Complex Distributed Surface Structures
The appearance of different surface structures on a test part is a major challenge in the automated surface inspection. A geometrically complex distribution of different surface structures requires the creation and adaptation of masks. A mask must be adapted in such a way that it maps only the area of the corresponding surface structure. The creation and adaptation of such masks is usually very time-consuming and often has to be carried out manually by the user. The Fraunhofer IPA has developed a concept that automates the creation of these masks.
In this method, all textures recognizable on the test part are initially determined and offered to the operator. The operator selects a texture relevant to the inspection task. The software then creates the corresponding mask, which covers the respective surface structure area. This procedure is repeated for all desired surface structures, whereby a suitable mask (fig. 2) is obtained for each surface structure.
The automated segmentation of the image into different areas of the structure allows fast creation of the masks while preventing small areas corresponding to only a few pixels of a particular texture to be added to this mask. An expensive manual fine correction of the masks is therefore obsolete.
After completing the creation process, the masks are saved and applied to the surface structures of the respective component in the inspection process. This allows individual parameterization for the different surface structures (fig. 3). Using a template-matching algorithm, or the automated generation of component-related coordinate systems, ensures that the masks and the component are superimposed in the inspection process. This makes the method robust against positioning inaccuracies in the process.
For the mathematical description of surface structures, textural models are required, which can describe the individual properties of the present surface structure adequately and distinguishably from other surface structures. For example, a textural model which reliably describes surfaces with a stochastic gray-value distribution is not necessarily suitable for describing the properties of a structured surface structure with periodic patterns in a sufficiently distinguishable manner. The test systems described above offer the possibility to use a wide variety of textural models in order to be able to automatically test a wide range of surface structures.
In addition to the choice of a suitable textural model, the choice of a suitable classifier is essential for the evaluation of surface structures. Only through the harmonious interplay of textural model and classifier a robust and automated assessment of the surface structure is possible. The Fraunhofer IPA has a broad background knowledge in this area due to various research and investigations.
It is possible to develop tailor-made conceptions and implementations of sophisticated solutions for automated testing of various surface structures. Automated mask production with downstream texture analysis has already been applied in the testing of plastic housings of electrical appliances and is suitable for all applications in which various textures occur on components which cannot be reproduced with the same textural model.