Seeing is believing 0 In the machine vision age Advanced imaging systems allied to artificial intelligence could provide powder metallurgy with an array of measurement tools to realise the productivity benefits of precise process control... M ACHINE VISION can be usefully applied to nearly all aspects of powder metallurgy. Applications include measurement and analysis of the pow- ders employed, monitoring and control of the manufac- turing process, and inspection of the final product. As the capabilities of the available technology increase and the costs decrease, manufacturers are becoming increasingly aware of the productivity benefits that can be obtained through precise process control and the utilization of accu- rate visually derived measurement data. The characteristics of machine vision that enable these benefits to be attained, include the following: - The technology is flexible and the applications multitudinous. - Machine vision is a non-contact measurement technique. - Measurements can be taken quickly and repeatably. - High-resolution cameras and advanced image pro- cessing techniques allow high measurement accuracies. - Machine vision can be used to gather data in envi- ronments or situations where it is difficult or impossible to perform manual measurements (e.g. in-situ shape mon- itoring during sintering). Advances in Machine Vision The application of machine vision requires the auto- mated detection and analysis of visual scene information. This process includes three essential elements: Figure I. Micrograph of irregular water-atomised iron particle. l Image capture l Image processing l Pattern recognition Capturing a good quality image is highly dependent upon the nature of the scene illumination. Advances are occurring in this area, with the emergence of new types of light sources such as high power diodes and solid-state- lasers offering cool stable illumination at a specific fre- quency. There have also been developments in associated delivery systems, such as optic fibres able to provide intense illumination precisely at the desired location. The ability to closely control the nature and configura- tion of the illumination has enabled the development of powerful new techniques, such as photometric stereo (PS), which usefully allows the separation of albedo and surface topographic information [M. L. Smith 20001 dur- ing surface inspection tasks. Specific devices are also available for focusing light into the required form within a scene; for example, a cylindrical lens may be used to produce a thin line of light. Such “structured” light is proving to be very useful in metrology, inspection and reverse engineering operations. Image processing Once an image has been captured, it is necessary to process it in order to extract useful information. Numerous commercial vision system software packages are available, providing extensive functionality. For the non-specialist it may be desirable to perform image-pro- cessing tasks within a graphical user interface (GUI) environment, with image processing functions being available from pull-down menus. Initial processing can involve the application of various convolution filters to remove noise, enhance features of interest, and possibly transforms for identification of geo- metric elements, such as lines or periodic functions. In machine vision a binary image is often found to be useful, and so a manual or automatic thresholding function may be employed. Advances in machine vision are occurring both in the range of image processing functions being made available by software vendors and the research into image processing being performed in company research laboratories and universities. This research often involves a more detailed manipu- lation of pixel data, for example, using computer lan- guages based in ANSI C. A good example is the applica- tion of photometric stereo for texture analysis, for which typically three images are obtained, each showing a 20 MPR Julv/August 2002 0026.0657/02/B-see front matter 0 2002 Elsevier Science Ltd. All rights reserved.