Bruce Edward Stuckman

A global search method for optimizing nonlinear systems

The theory and implementation of a global search method of optimization in n dimensions, inspired by Kushners method in one dimension, are presented. This method is meant to address optimization problems where the function has many extrema, where it may or may not be differentiable, and where it is important to reduce the number of evaluations of the function at the expense of increased computation. Comparisons are made to the performance of other global optimization techniques on a set of standard differentiable test functions. A new class of discrete-valued test functions is introduced, and the performance of the method is determined on a randomly generated set of these functions. Overall, this method has the power of other Bayesian/sampling techniques without the need for a separate local optimization technique for improved convergence. This makes it possible for the search to operate on unknown functions that may contain one or more discrete components. >

Comparison of global search methods for design optimization using simulation

A methodology for the application of global search methods for optimizing the results of a computer simulation is presented. Specific global optimization methods including simulated annealing, genetic algorithms, and Bayesian techniques are discussed in terms of their strengths and weaknesses as applied to this methodology. In particular, the effects of simulation time, constraints, dimensionality, and computational complexity are considered as they relate to the choice of algorithms. Simulated annealing and genetic algorithms perform similarly, yet differ in many ways from the class of Bayesian algorithms. Bayesian algorithms spend additional computation time in modeling all past values of the unknown function in an effort to minimize the number of evaluations of the function. These methods would be the algorithms of choice for determining the optimal design via simulation, provided the number of design variables is less than 10 and the time required to run a single simulation is large compared with the time it takes the algorithm to determine the next point.<<ETX>>

A comparison of Bayesian/sampling global optimization techniques

A survey of current global optimization techniques for continuous variables is presented, inspired by recent publications of computer coding of several popular Bayesian/sampling methods. The methods of C.D. Perttunen (1990), B.E. Stuckman (1988), J.B. Mockus (1989), A. Zilinskas (1980), and V.K. Shaltenis and G. Dzemyda (1982) are compared with a clustering algorithm, a simulated annealing algorithm, and the Monte Carlo method. Results are given for these methods based upon the experimental rate of convergence on a series of standard test functions. A new test function is presented which has a global solution within an area which is small in comparison with the search space. >

Multicriteria optimization of simulation models

The authors suggest a framework for the multicriteria optimization of simulation models by first discussing the unique difficulties of this problem area along with important problem characteristics, and then discussing the way that these problem characteristics would affect the choice of a particular technique. The problem of manufacturing system optimization is addressed. Various techniques, along with their advantages and disadvantages, are discussed and categorized according to the timing of the articulation of the required preference (tradeoff) information with respect to the optimization.<<ETX>>

The rank transformation applied to a multivariate method of global optimization

The incorporation of the rank transformation with an existing multivariate method of global optimization is presented. By transforming the objective function evaluations into ranks and applying Stuckmans multi-univariate method of global optimization to the ranked data, the resulting method gains the same benefits as the nonparametric method. The method is applied to a standard set of test functions. The application of the rank transformation is shown to significantly reduce the number of function evaluations needed for convergence within a specified tolerance.<<ETX>>

A solid state infrared device for detecting the presence of car in a driver's blind spot

Many automobile accidents are caused each year by a driver failing to see that there is a car in his blind spot. A method for detecting the presence of an object in a drivers blind spot using an active infrared sensor is presented. The details of an experimental implementation are presented. Due to safety considerations, more extensive testing is needed to develop the detector into a commercially viable product.<<ETX>>

A multidimensional Bayesian global search method which incorporates knowledge of an upper bound

The authors present a method of incorporating a priori information into an N-dimensional Bayesian method of global optimization. Many real-world optimization problems are of such variety that the objective function is not convex. Therefore, a global algorithm which considers the entire search space is necessary. A priori knowledge of a bound on the objective function can be incorporated into the search algorithm. The algorithm restricts its sampling to areas most likely to contain the global maximum of the function. Past work on 1D algorithms has provided a means of incorporating this information into a Bayesian method of global optimization to hasten convergence. Numerical results of this method on a standard test function are presented.<<ETX>>

Stochastic modeling of calibration drift in electrical meters

The drift in the calibration bias in an instrument can be modeled as a stochastic process, specifically, as the Wiener process. The probability distribution of this Wiener model can be found conditioned upon calibration at some time t/sub 1/. Bounds on the drift of the bias after calibration can be found as a function of time based upon an estimate of the parameter alpha , of the Wiener process. This parameter can be easily estimated based upon the collection of data calibration bias as a function of time. The statistical bounds on the calibration drift allow an instrument user to make his or her own choices as to the calibration interval based on the desired accuracy of the measurement.<<ETX>>

The normal score transformation applied to a multi-univariate method of global optimization

Nonparametric global optimization methods have been developed that determine the location of their next guess based on the rank-transformed objective function evaluations rather than the actual function values themselves. Another commonly-used transformation in nonparametric statistics is the normal score transformation. This paper applies the normal score transformation to the multi-univariate method of global optimization. The benefits of the new method are shown by its performance on a standard set of global optimization test problems. The normal score transformation yields a method that gives equivalent searches for any monotonic transformation of the objective function.

Electronic measurement of fluid level using acoustic sensors

Many fluid tanks are, in effect, a sealed box which, if forced by an acoustic source will resonate at a fundamental frequency dependent upon the volume of the box as well as its shape and the thickness and stiffness of its walls. The introduction of fluid to the bottom of the tank will change the effective volume of the tank as seen by the acoustic source. This will change the acoustic properties of the tank by an amount which corresponds to the amount of fluid in the tank. Three methods of fluid volume measurement based on this principle are presented. Results of these three methods are compared. The advantages of these methods and applications are discussed.<<ETX>>