استریو های فوتومتریک پویا برای کنترل کیفیت خط از کاشی و سرامیک

The rapid and automated detection of manufacturing flaws is becoming increasingly important in order to maintain competitive advantage in many production environments. In the case of natural and ornamental materials, the presence of both surface colouration and surface topography is often such that manual inspection, along with many conventional imaging techniques, fails to isolate physical or structural defects in the presence of complex and random patterns. In this paper the concepts of photometric stereo are adapted and extended for application in manufacturing environments. A case study on the high speed inspection of ceramic tiles is presented for the analysis of surfaces at production line rates of up to 30 m/min. This new technique, for the first time, demonstrates a genuine and commercially attractive potential for the practical automated quality control of complex surfaces. A commercial system, based on this research, is currently being developed.

Machine vision often involves the acquisition, digitisation and analysis of images for the purpose of data capture and process control. When data resulting from these processes is translated into information equivalent to that derived from human visual perception, it can often provide a versatile means for improving efficiency, product quality and consistency in manufacturing industries. Testimonials in published literature are numerous [1] and [25]. Practical machine vision inspection systems consist of a combination of computer software, hardware, cameras and lighting, working together to capture and analyse image data. Machine vision is a broad discipline which encompasses aspects of image acquisition, computer vision, image processing and pattern recognition as well as aspects of data extraction and or environment interaction. Often the overall process involves a conversion of physical quantities into digital data and subsequently into meaningful information. Needless to say, the term meaningful information is representative of a specific requirement and therefore is nearly always subjective and application dependant. The output of these systems is increasingly being used to automate manufacturing inspection, feature detection and control. For example, inspecting date codes on pharmaceuticals delivers meaningful information in the form of a production or expiry date for batch classification. Awcock and Thomas [2] describe machine vision in terms of seven stages, these are shown in Fig. 1. The process is considered to be logical, practical and serves well as a template for the design of machine vision systems.1.1. Surface inspection In surface inspection applications, surfaces can be analysed using a range of techniques including vision based methods. Surface analysis is sometimes referred to as the study of surface metrology or surface texture. Swonger [3] identifies vision based 3D metrology as a “niche” market area. His paper highlights the advantages of an automated approach over human inspection where hazardous environments, subjectivity, inconsistency, limited attention span, speed of operation and accuracy all affect and limit human operation. Photometric stereo (PS) emerged in 1978 [4] and has developed into a robust technique for object shape recovery using a simple methodology. However, it has largely remained dormant in its range of application. The classical application of photometric stereo (CPS) has been based on the interaction of collimated, uniform lighting with stationary continuous Lambertian surfaces. Although the technique is widespread, in the main it has remained essentially a laboratory tool, which is capable of recovering discrete surface orientation information for the regeneration of surface form. Within this domain of reverse engineering the emphasis has often been in obtaining surface height information or surface profiling. This type of output requirement, e.g. a CAD model, is sometimes regarded as a complication to the technique as the recovered gradient data must be integrated and therefore is prone to the effects of noise. Weiner filters have been successfully used to suppress spurious noise providing a mechanism for smoothing of the information for reliable profiling. Applying these types of noise reduction transforms is time and processor consuming and seem only beneficial to applications where regions of interest are relatively small. In an industrial application of static PS, Hansson and Johansson [5] have used this approach to analyse surface profiles for ink coverage analysis. Smith et al. [6] innovated the application range by employing PS for surface inspection purposes without generating height information. This new realm exploited the properties of ‘bump maps’ [7] whereby gradient information was used directly to analyse topography. Here, the surface data is held as an array of surface gradient values. Furthermore, along with the recovery of 3D topography, the isolated reflectance (surface colouring) or 2D information could be analysed simultaneously yet independently. Thus emphasising the appropriateness of applying PS to examine complex surfaces. That is, those surfaces that contain superimposed variable surface colouring on variable surface geometry. It was further shown how existing image processing algorithms could now be applied directly to the transformed 3D information for purposes of segmentation, convolution, morphology, thresholding and classification [8]. Despite the accomplished nature of the classical process, namely its ability to directly resolve both 3D data and 2D data, there is little evidence that the technique has had any impact on manufacturing quality control. 1.2. A requirement for dynamic surface analysis The manufacturing marketplace has become highly competitive in recent years. Every major organisation strives to gain competitive advantage by being innovative, faster, flexible and more customer focused. Embracing quality standards and total quality management (TQM) concepts have pushed philosophies and practices to incorporate continuous inspection and continuous improvement techniques, thus ensuring that the finished product has successfully passed each and every stage of the manufacturing and inspection process. CPS is effective for analysing static surfaces, as time has no influence on the location of surface features with respect to a fixed vision system. A more difficult situation occurs when surfaces or objects are subject to some form of movement. This dynamism could quite simply be a predefined linear motion, like mass-produced rapidly moving objects travelling along a conveyor or it could be a random omni-directional motion, like the internal organs of a living being or the image shake experienced in handheld video cameras. Inherently, high speed movement of surfaces often limits the number of inspection techniques that can be applied to them. As such resulting inspection systems are highly specialised and are a trade-off between capability and computational complexity. The application of CPS to moving surfaces has not been possible because of the apparent requirement for strict light and object structuring. Consequentially, the technology has not been able to adequately cross into the domain of industrial application, where, part motion, non-idealised lighting and environmental conditions have restricted its efficacy. There exists a wide range of processes that could benefit from a comprehensive and flexible inspection system capable of on line operation. The manufacturing of timber products, leather, fabric, woven textiles, sheet materials, tiles and any other web production generally lacks implementation of modern TQM philosophies apparent in many other industries. The qualities and attributes exhibited by CPS are the same as those needed for a dynamic surface analysis regime. This paper presents an evolution of photometric stereo aimed at industrial inspection tasks. It focuses on a scheme that addresses the limitations of the conventional approach with a view of adapting the methodology for a range of dynamic applications.

An adaptation of photometric stereo has been presented for the on line analysis of difficult moving surfaces containing complex mixed two- and three-dimensional textures. The new method, spatially multiplexed photometric stereo, was based on projecting light onto separate regions of the inspection plane. Typical example applications are surprisingly common and include the processing of natural materials such as stone, timber, leather and fabrics, where stochastic 2D coloured patterns may be concomitant with 3D topographic textures. Problems also arise when defects such as cracks and scratches may be coincident with colouring defects. Examples include any process where surface topography and reflectance must be controlled and include painted, plated or printed finishes on metal or plastic components, in fact any process where two- and three-dimensional features are simultaneously present. Unfortunately, conventional imaging techniques are unable to reliably distinguish between these kinds of features, often leading to unsatisfactory performance through feature misinterpretation. Photometric stereo has been shown to offer a direct, implicit technique for recovering superimposed surface reflectance and geometrical information in isolation. However, limiting assumption forming the basis of CPS is that differing separate directional illumination configurations must be present in each of the acquired views. While this is readily achieved in the static scenario, either by moving the light sources or by switching on and off differing lights between image acquisition stages, in terms of practical application allowing potential for rapid or uncontrolled relative movement between the camera and the surface, this assumption is found to be the most restrictive. This coupled with the ability to simultaneously acquire images under different lighting configurations that forms the basis of realising the dynamic photometric stereo method. The new method combines the adeptness of photometric stereo for textural analysis with new theoretical mechanisms and practical hardware capable of imaging dynamic surfaces, providing a flexible and generic scanning system for the inspection of moving complex surfaces. Consideration of the geometries and behaviour of industrial lighting was taken into account resulting in equations for irradiance modelling of the line lights. A case study was used to demonstrate a potential area of application. The case study was used to practically simulate the on line inspection of ceramic biscuit tiles. 3D topographic ground truth data was generated for a selection of test tiles with a high precision 3D surface scanning CMM. This data was converted into bump map format and compared directly with data obtained using DPS. The largest average absolute error obtained was ∼2°, which is equivalent to 9 μm per 0.25 mm × 0.25 mm square patch (resolution), at a tile travel speed of 30 m/min. This provides a simple quantitative measure of the accuracy of the new system. It was highlighted how the resulting data from the process could be used to provide interaction with the manufacturing process in order to reduce wastage and improve manufacturing quality. A prototype system is currently being commercially developed.