Michael H. Thursby

Composite damage assessment employing an optical neural network processor and an embedded fiber-optic sensor array

This paper discusses a novel approach for composite damage assessment with potential for DoD, NASA, and commercial applications. We have analyzed and modeled a two-dimensional composite damage assessment system for real-time monitoring and determination of damage location in a composite structure. The system combines two techniques: a fiberoptic strain sensor array and an optical neural network processor. A two-dimensional fiberoptic sensor array embedded in the composite structure during the manufacturing process can be used to detect changes in the mechanical strain distribution caused by subsequent damage to the structure. The optical processor, a pre-trained Kohonen neural network, has the capability to indicate the location of the damage due to its positionally associative architecture. Because of the parallel, all optical architecture of the system, the system has the advantages of having high resolution, a simple architecture, and almost instantaneous processor output. Results of the modeling and simulation predict a highly robust system with accurate determination of damage location. We are currently beginning work on a breadboard demonstration model of the system.

Fiber-Optic Sensor And Smart Structures Research At Florida Institute Of Technology

This paper discusses the fundamental issues being investigated by Florida Institute of Technology (F.I.T.) to implement the technology of smart structural systems for DoD, NASA, and commercial applications. Embedded sensors and actuators controlled by processors can provide a modification of the mechanical characteristics of composite structures to produce smart structures1-3. Recent advances in material science have spurred the development and use of composite materials in a wide range of applications from rotocraft blades and advanced tactical fighter aircraft to undersea and aerospace structures. Along with the advantages of an increased strength-to-weight ratio, the use of these materials has raised a number of questions related to understanding their failure mechanisms. Also, being able to predict structural failures far enough in advance to prevent them and to provide real-time structural health and damage monitoring has become a realistic possibility. Unfortunately, conventional sensors, actuators, and digital processors, although highly developed and well proven for other systems, may not be best suited for most smart structure applications. Our research has concentrated on few-mode and polarimetric single-fiber strain sensors4-7 and optically activated shape memory alloy (SMA) actuators controlled by artificial neural processors. We have constructed and characterized both few-mode and polarimetric sensors for a variety of fiber types, including standard single-mode, high-birefringence polarization preserving, and low-birefringence polarization insensitive fibers. We have investigated signal processing techniques for these sensors and have demonstrated active phase tracking for the high- and low-birefringence polarimetric sensors through the incorporation into the system of an electrooptic modulator designed and fabricated at F.I.T.. We have also started the design and testing of neural network architectures for processing the sensor signal outputs to calculate strain magnitude and actuator control signals for simple structures.

Smart structures and fiber optic sensor research at Florida Institute of Technology: 1990

This paper discusses several novel concepts being investigated in the Center for Fiberoptic Sensor Systems and Smart Structures at Florida Institute of Technology associated with fiberoptic sensors, actuators and processor technology, and efforts to integrate these components into distributed smart systems. Projects include: a polarimetric sensor with active phase tracking test set, a combination polarimetric/two mode sensor, an N-mode sensor with neural processor, damage assessment using embedded fiber-optic arrays and a neural processor, a pulsed interferometric sensor, neural network-processed polarimetric sensor signals, and optically-energized shape-memory alloy actuators.

Smart electromagnetic structures: a neural network antenna

Artificial neural networks(ANN5) and their ability to model and control dynamical systems for smart structures including sensors actuators and plants are being considered in our lab. Both linear and non-linear systems have been successfully modeled. We are presently working on two diverse regimes smart mechanical systems and smart electromagnetic systems. In order to better understand neural controllers as used in the smart electromagnetic structures we have directed our study of ANNs toward understanding the ability of the network to approximate system responses. We are training networks to mimic the desired output of the system. The damped sinusoid was chosen as the model and was approximated using a Jordan-like 8 iterative network. The results to date indicate that the ANNs can easily mimic these systemsthe question is whether the mechanism that the network applies can be related to the mechanisms that we understand for classical analysis. Sensor preprocessing represents a significant element in the smart material and structure concept. We are looking at certain network architectures as sensor preprocessors. Results from both these areas will be presented in this paper.

Neural network processing of fiber optic sensors and sensor arrays

For sophisticated smart structures where sensing and actuation is distributed over large areas or consists of dozens to thousands of discrete elements, the processing task is computationally intensive. Artificial neural networks offer an opportunity to implement a massively parallel architecture with near real time processing speed and the ability to learn the desired response. This overview of applied neural network processing projects at Florida Institute of Technology includes: processing polarimetric and N-mode strain sensor signals, damage assessment using embedded sensor arrays, and development of electrooptic neural networks

Phase control of a micropatch antenna element

We will describe an effort that to date has produced a working prototype of a micropatch antenna incorporating a single-dollar- per-bit phase shifter. In this paper we will describe a method of controlling a micropatch antenna to provide phase only variation of the antenna characteristics using a similar device to that used for the frequency control experiments. We have successfully varied the phase of the antenna element without significantly changing the operating frequency. This work has led us to pursue further the design and fabrication of an array of such phase adjustable element to test the hypothesis that such phase- controlled micropatch elements can be used to fabricate a low- cost phased-array antenna.

Neural Control of Smart Electromagnetic Structures

Artificial neural networks (ANNs) and their ability to model and control dynamical systems for smart structures, including sensors, actuators, and plants, are directly applicable to the smart electromagnetic structures (SEMS) concept. The application of neural networks to the area of controls is being reported frequently. The ability of a structure to adapt to impinging electromagnetic (EM) energy will allow the structure to change its reflection characteristics and thus to change its radar signature. By embedding a control element in the structure of a single microstrip patch element, its electrical characteristics can be changed. If such an element can be controlled by a closed loop system the patch antenna element can be made to adjust its operating characteristics through the control algorithm. If the control algorithm can be implemented in a neural network, the system can be made to change its characteristics in response to the stimulus. This change can be used to alter the antennas performance in real time. As part of our research, a model of the patch neural network antenna system is being developed and this analytical model, as well as experimental models of the antenna are being tested and compared. The neural network antenna model and prototypes are being taught to adapt to the magnitude and phase response of microstrip patch antennas to incoming signals. The response characteristics and speed are reported in this paper. We demonstrate that the patch can be given autonomous adaptive capabilities using neural networks. An array of such smart patches could be assembled to create an even more adaptable antenna system.

Processing Of One-Fiber Interferometric Fiber-Optic Sensor Signals

Recent research conducted in the Fiber-Optic Sensor Systems Laboratory at Florida Institute of Technology in fiber-optic strain sensors for use in smart structures has concentrated primarily on one-fiber interferometers. Sensors using this technique are of particular interest because they are rugged and provide reasonable strain sensitivities, typically a thousand times that of microbend sensors, while requiring only a single optical fiber. As in two-fiber interferometers, some method must be used to convert the sensor signal output to an absolute strain value. One technique is to provide active phase tracking to keep the sensor signal output at phase quadrature for maximum sensitivity and to eliminate fringe count uncertainty. This paper contains a description of a technique for active phase tracking in polarimetric sensors composed of high- and low-birefringence fiber using an electro-optic polarization modulator. On-going development of artificial neural processors for processing fiber-optic sensor signals from both polarimetric and few mode sensors, as well as generating control signals for actuators in smart structures, is also discussed.

Smart Structures Incorporating Artificial Neural Networks, Fiber-Optic Sensors, And Solid-State Actuators

Neural networks have been shown to be useful in sensor pre-processing and dynamic system modeling and control, as well as various other signal analysis areas. Their application to the problem of developing a real-time smart structure is further indicated by the extremely fast post-learning speed of the neural network. Fiber-optic sensors also possess very high response rates, and sensitivities and are well suited to the detection of structural changes. The integration of these two technologies is a natural step toward a high-speed smart structure. A smart structure that is applicable to airfoil control surfaces as well as artificial human joints is being modeled and built at Florida Institute of Technology. The actuators (muscles) in our model are replaced with shape-memory metal strands. The joint-position sensors of the neuro-muscular system are modeled by fiber-optic sensors. The control (neural) circuits are replaced by an artificial neural network. We are investigating the suitability of each of these subsystems to the problem. The appropriate neural-network architecture, as well as the sensors and actuators, is under investigation, and a prototype system is being fabricated and tested. Results of the modeling, design, and performance will be discussed.

Robust linear quadratic regulation using neural network

Using an Artificial Neural Network (ANN) trained with the Least Mean Square (LMS) algorithm we have designed a robust linear quadratic regulator for a range of plant uncertainty. Since there is a trade-off between performance and robustness in the conventional design techniques, we propose a design technique to provide the best mix of robustness and performance. Our approach is to provide different control strategies for different levels of uncertainty. We describe how to measure these uncertainties. We will compare our multiple strategies results with those of conventional techniques e.g. H(infinity ) control theory. A Lyapunov equation is used to define stability in all cases.