The development of an automated image shearing microscope for measuring the lengths of the gaps in magnetic recording heads is discussed. A model is developed for a manual measurement system and this is used to identify sources of measurement error. The architecture of an automated measurement system is described in detail. Algorithms for automating the measurement process are presented and the model is extended to analyse their performance. These algorithms include a method of finding a clean site, automatic focusing and automatic measurement based on image shearing. The effects of system parameters such as illumination bandwidth, focus error and transducer resolution are investigated and it is shown that gap lengths as narrow as 0.5um may be measured reliably. Experimental performance tests show that the system can measure gap lengths to better than 0.05μm (3a) reproducibility in under 4 seconds.
The objective of this paper is to describe the design of a pair of sophisticated robot fingers which enable to sense three axis of forces and three axis of torques information using piezo-resistive strain gauges as sensing elements with frequency output. The fingers may be mounted onto a servo controlled robot gripper and interfaced with robot controller to serve as active compliance device for small parts assembly tasks. This newly direct force/torque sensor with frequency output employing RC oscillation principle has demonstrated great advantages with signal conditioning and processing relative to the conventional voltage output procedures. A description of the test results is also presented.
New sensors have been developed that inspect the quality of transparent materials, specifically silicon solar cells. These sensors inspect for surface, as well as, subsurface cracks and induced stresses. The main reason for development of these sensors is their ability to quantize the stresses. In the area of large solar array production, which was introduced by the Solar Array Flight Experiment (SAFE) flown on the space shuttle in 1984, these sensors are under development to automate the inspection and measurement of specific solar cell array parameters during various stages of their production. The primary parameter is the electrical interconnection's bond quality to the individual solar cells. The measured stress magnitudes at these points are used to determine the quality, specifically the strength of the bond. Preliminary results have demonstrated this correlation on a limited sampling, and utilization of these sensors in production has begun. Other application areas include the manufacture (production) of solar cells (arrays) consisting of other types of materials, potentially gallium arsenide, and the manufacture of microelectronics and other semiconductor devices.
Hardware has been designed for high speed acquisition of the boundaries of objects in binary images. The boundaries, described using chain codes, are encoded from line scanned images and only one line of image data is stored at one time. High speed (5-10MHz.) and low memory requirements enable the hardware to process large images (2048 pixels wide) required for some applications of image processing to automatic inspection e.g. the inspection of bare printed circuit boards (PCB's). The operating principles of the hardware have been validated in a software simulation which itself is a useful image processing tool.
Drawing on experience gained in the design and construction of vision based workstations for measurement and inspection, utilized in industrial process and quality control, the requirements for a vision analysis system are explored. A description of an architecture for a vision system implementing these requirements is presented.
Visual inspection is a common feature of manufacturing for which machines are currently illsuited. A major problem has been in bringing enough knowledge to bear on the camera input, fast enough, to be of practical value. This paper describes a machine architecture for such expert vision research in the domain of pattern inspection: in particular, the inspection of printed wiring patterns. The machine includes a powerful model of image-to-symbol transduction and a Lisp multiprocessor environment.
In addition to reviewing the major alignment and inspection procedures in producing miniaturized electronic components, this paper describes a 3D automated vision system used in the inspection of TAB (Tape Automated Bonding) products. This multi-sensor system utilizing four solid-state cameras is aliased IBIS Intelligent Bump-tape Inspection System. Besides the capabilities to measure the major planar geometric features, the key features of this system include (1) structured-light 31) vision for etch depth measurement, (2) data-base driven inspection, (3) close coupling of microsensing and micropositioning, (4) sequential usage of multiple resolution cameras and computer controlled illumination for optimal micromeasurements, and ( 5 ) statistical analysis of measured data for process trend monitoring.
SAIL ( Simple Assembly Inspection Language) is a C-language based high level command translator that allows programming of automatic visual inspection tasks for industrial assemblies. It allows the writing of short and simple programs that perform visual inspection of assemblies. Their input is an image file of any dimension and grey level scale, and the output a (possibly multiclass) classification of the assembly. The advantages of SAIL are the ease of programming, since virtually no previous knowledge of a programming language is needed, the power of the inspection algorithms available and, most importantly, its expandability.
This paper describes selected multiprocessor architectures for very high speed automated inspection systems. Traditional image processor architectures, oriented toward sequential processing, lack the throughput potential for use in high-end applications. Therefore, non-Von Neumann multiprocessor architectures are emerging as candidates for future automated inspection systems that constantly challenge the state of the art in image processing. Four processor concepts are compared and contrasted with respect to their applicability in high performance inspection systems. Each has advantages, so the selection of an appropriate architecture should be based primarily on system requirements, such as sensor data rate and format, resolution, image size, and behavior of algorithms.
I. Introduction: Digital image inspection and measurement systems, in recent years, have made significant progress toward small size, light weight, and low cost, as computer technology and specialized processing hardware have made rapid advances. This paper is a report of a developmental effort on an image processor based on IBM PC XT/AT for semiconductor manufacturing industry where extensive inspection and measurement take place [1-3].
A computer vision program was developed to recognize the precise position of surface mount components. It directs a robot to load surface mount components on PC boards. This paper describes software algorithms and required hardware. Calibration between the robot and the vision system and the precision of the vision system are also presented.
An approach to the design of a new, low cost, general purpose machine vision engine is discussed. Some of the problems and issues in the specifications of the product, are presented. The advantages and disadvantages of commercially available image understanding approaches are discussed. The approach taken to the design of a general purpose Machine Vision engine is rationalized.
This paper mainly deals with methods for high-speed automatic visual inspection and measurement. First, the algorithmic aspects of pattern recognition are briefly reviewed. Second, the different elements of a visual inspection and measurement system are discussed. Third, special attention is paid to image processing operations for visual inspection and measurement which can be executed in real time or pseudo real time. They include run length encoding, the calculation of projections, windowing, edge detection, texture analysis and the calculation of spatial moments. A hardware implementation of each of those operations exists or is being developed at our laboratories. Finally, we emphasize the importance of artificial intelligence techniques for visual inspection tasks where complexity instead of speed is the limiting factor.
The science of automatic inspection and measurement has assumed greatly increased importance in today's competitive business environment and it will assume still greater importance with the coming of tomorrow's Factory of the Future (FOF). Even today, the terms Computer Aided Manufacturing, the "CAM" in CAD/CAM, and Computer Integrated Manufacturing (CIM) have become common in the popular jargon. Whether computer aided or computer integrated, the modern manufacturing operation requires automated inspection and measurement for two fundamental reasons: (1) economic pressures dictate that waste and inefficiency be minimized, and (2) today's customer base requires not only competitive prices but also competitive quality. Therefore, automated inspection and measurement is needed to provide cost minimization while also providing quality maximization. These dual requirements are discussed in terms of turnkey systems supplied by independent vendors to the manufacturer. Specific examples are presented of the pitfalls that are sometimes encountered. We illustrate the process in terms of actual turnkey on-line inspection and measurement systems that have been developed for industrial use on fixed assembly lines, as well as for use in flexible assembly operations. The systems to be discussed contain image acquisition subsystems, real time data processing, optical non-contact gauging subsystems, and the associated mechanical, optical and computer hardware to achieve total automated operation. As such, these systems address the problems of integrated system design for automated inspection and gauging. The systems described have the capability to inspect and gauge a variety of products, ranging from spur gears to rigid, dimensionally accurate bodies, to metal webs, to transparent/transluscent paper, plastic and woven materials, and to the testing and inspection of optical systems.
The new method of multireflected autocollimation allows to perform angle measurements with greater accuracy than the accuracy of the measuring device itself (theodolite, autocollimators, etc.). The article contents are: description of the method's principle and its theoretical analysis; photograph of experimental angular measurements with theodolite T16 - using this method, an accuracy of + 0°0'5" was achieved by direct reading (less than 000'1" by estimated reading); photograph of the set-up which was used for reception tests of "Microcontrole" Mirror Mount SL20A; possible method application principle layout for increasing measurement accuracy of devices and for calibration of angular measurement devices.
Many automatic visual inspection applications involve checking circular holes. In this paper four circularity measures based on shape compactness, distribution of radius values, distribution of chord lengths between random border points and the Hough transform are comparatively studied. Properties of the various approaches are investigated over a range of image resolutions which are considered to be pertinent to industrial inspection tasks.
Because it is nearly impossible with todays technology to make a processor which is optimized for all possible image processing tasks one tries to realize different processors each of them being a compromise between technology, processing speed and class of operations that must be performed optimaly. For our Image' Computer  a new processor, called the Transformation Processor, is optimized for a specific class of operations.
This paper presents a study on the automatic second optical inspection of ICs. The emphasis lays on the image processing algorithms. A distinction is made between "primary feature inspection" and "morphologic feature inspection". Both reference and non-reference methods are considered. In each case, inspection in one pass is possible. Some results are given and discussed.
Visual inspection during the high-speed manufacturing of web materials often yields inefficient results due to the delay between detection of process imposed defects and the implementation of corrective action. The defect detection and classification process, coupled with defect source determination, can delay process alteration by minutes, hours and sometimes days, yielding wasted materials valued in the thousands of dollars. This paper describes recent activities in the use of image acquisition and analysis techniques to detect and classify defects produced on high-speed web materials at video frame rates. The parameters of the sensor systems are described for optimizing the detection of the defects that alter the surface topography or texture of various materials. Additionally, real time image analysis techniques are described for processing and/or analyzing an image and detecting the presence of surface defects in the background noise. System implementation, including practical limitations and trade-offs relating to resolution, sampling rates, etc., is discussed.
Two automated laser scanning systems have been developed for the rapid inspection of flexible and rigid magnetic recording disks. Surface defect characterization is performed by means of the detection of light scattering pulses produced by a wide range of surface or near-surface physical defects. These imperfections are highly correlated with magnetic defects associated with data drop-outs. In addition, surface irregularities can result in transient air bearing discontinuities which cause read/write data errors and head disk interference. The inspection systems described in this paper include complete robotic disk handling subsystems with computer control for all inspection parameters. The automated inspection process is performed within a self-contained clean-room environment to preclude disk contamination. Disks are classified and physically segregated into several "quality" groupings in accordance with user-selected criteria combining the number and size of defects. Complete disk inspection of flexible media is performed in 3 to 5 seconds, and in 30 to 45 seconds for rigid media, depending on operating parameters. The minimum detectable defect size for rigid disks is of the order of one micrometer, and 5 to 15 micrometers for flexible disks.
Westinghouse Hanford Company has designed and is constructing a nuclear fuel fabrication process line for the Department of Energy. This process line includes a pellet surface inspection system that remotely inspects the cylindrical surface of nuclear fuel pellets for surface spots, flaws, or discoloration. The pellets are inspected on a 100 percent basis after pellet sintering. A feeder will deliver the pellets directly to a fiber optic inspection head. The inspection head will view one pellet surface at a time. The surface image of the pellet will be imaged to a closed-circuit color television camera (CCTV). The output signal of the CCTV will be input to a digital imaging processor that stores approximately 25 pellet images at a time. A human operator will visually examine the images of the pellet surfaces on a high resolution monitor and accept or reject the pellets based on visual standards. The operator will use a digitizing tablet to record the location of rejected pellets, which will then be automatically removed from the product stream. The system is expandable to automated disposition of the pellet surface image.
The paper describes a system developed for the automatic performance testing of self-illuminating display panels used, for example, in automobile dashboard instrumentation. The system employs a high resolution TV camera to view the displays; the image being analysed in a microcomputer system using special-purpose software. The change from moving needle instruments to electronic displays, which employ vacuum fluorescent, LCD and LED technologies, forces a new testing philosophy on manufacturers. The system performs 64 tests on the panels in approximately 45 seconds, which is considerably faster than a human operator could achieve.
The metal cutting operation has been affected very heavily by the ongoing development of the computing power of the microprocessors. This report addresses to one such application in "semiautonomous machining". The prime interest of this study is to obtain maximum metal removal rate while maintaining the surface finish of the workpieces at the desirable level. Characteristically these two targets introduce a trade-off and as a result an optimal operating condition. At an experimental sight a lathe is retrofitted with 3 actuators to control the spindle speed, the depth of cut and the feed rate of the turning process. A sensory device, dynamometer, is used to monitor the cutting forces in 3-D at the tip of the tool. These force readings are processed via a computer, in-situ, to update the operating conditions, i.e. cutting speed, chip thickness and feed rate. There is a complex statistical analysis to forecast the trend of the dynamic characteristics which inturn yields an "anticipatory control" action. Fundamental principles of DDS (Dynamic Data Systems) and Corresponding Time Series Analysis are used off-line to describe the nature of the dynamics. Once the discrete dynamics is obtained off-line, the real time optimal control becomes one of lesser complexity but is not treated in this text. Two different packages are used to compare the convergence in the statistical part of the study. A table of comparison is prepared. An analytical model is developed to configure the effects of various parameters in the cutting mechanism. The actual data registries of force vs. the analytical model comparison is made. The future development points are discussed in the direction of introducing autonomous machining capabilities to the tools.