Search

Modeling and simulation have always been highly useful techniques to use when designing, analyzing, or diagnosing an engineered system. Development of an appropriate model, in terms of accuracy, granularity, and complexity, has typically been the burden of the designer, analyst, or troubleshooter. It would naturally be advantageous if a model could be developed automatically, with the user supplying only some final minor refinements. The fact that most modern systems are designed in a computer-aided design (CAD) environment makes this a realistic prospect, because much of the data necessary for the model is already in electronic form. This article outlines a system called the automated knowledge generator (AKG), which embodies techniques that automatically create a model of an engineered system directly from its CAD representation, and describes the extensive testing process followed in order to validate its performance.

Advanced courses in the areas of Machine Perception and Machine Intelligence are not easy to prepare or conduct due to the dynamic research nature of the field. In engineering, our interest is in the construction of intelligent machines in the form of robotic devices or through computer simulation models. Studies in this area are further complicated by the broad range of disciplines that must be accessed in order to provide adequate coverage. This discussion focuses on a set of computer simulation projects that have been assigned at various times during offerings of an advanced graduate level course entitled Machine Perception. The format of the projects are presented with discussion of student results. The projects were novel in that they require advanced techniques of machine vision and are open-ended in that there are no specific solutions. The problems represent the mainstream of machine perception research without requiring more effort than is possible in a one semester course. Students responded with algorithms developed in a variety of computer languages and on various machines.

Signal reconstruction from the measurements of its Fourier transform magnitude remains an important and difficult problem that occurs in different areas in signal processing. Among all the approaches developed to solve this problem, the iterative transform algorithms are currently the most efficient. However, these algorithms suffer from major drawbacks such as stagnation, slow convergence, and high computational cost that limit their practical application. In this brief, we introduce a wavelet adaptation of the general iterative algorithm where the problem is decomposed into different resolution levels and the image is reconstructed following a coarse-to-fine strategy. We show that the proposed approach can significantly improve the performance of the existing algorithms while dramatically reducing their computational complexity.