Sesh Commuri

CMAC neural networks for control of nonlinear dynamical systems: structure, stability and passivity

The cerebellar model articulation computer (CMAC) neural network (NN) has advantages over fully connected NNs due to its increased structure. While its advantages over conventional control techniques have been recognized in the literature, it has primarily been used in system identification and pattern recognition, but not in control applications. This paper attempts to provide a comprehensive treatment of CMAC NNs in closed-loop control applications. Novel weight-update laws are derived that guarantee the stability of the closed-loop system. The passivity properties of the CMAC under the specified tuning laws are examined and the relationship between passivity and closed-loop stability is derived.

CMAC neural network control of robot manipulators

A comprehensive treatment of the cerebellar model articulation controller (CMAC) neural network (NN) for the control of robot manipulators is presented. The structure and localized learning properties of CMAC NN is exploited to design efficient controllers for nonlinear systems belonging to a given useful class. Continuous- and discrete-time implementation of these controllers is systematically examined. Novel weight update schemes are derived and the closed-loop stability of the controller and the system is rigorously proved. These weight update schemes are shown to be nonstandard modifications of adaptive techniques prevalent in the literature. Finally, the validity of these techniques is demonstrated through numerical studies. ©1997 by John Wiley & Sons, Inc.

Feedback linearization using CMAC neural networks

The objective of this paper is to achieve tracking control of a class of unknown nonlinear dynamical systems using a Cerebellar Model Articulation Controller (CMAC). With mild assumptions on the state-feedback linearizable nonlinear systems, using this CMAC the uniform boundedness of the closed-loop signals is presented and that the controller achieves tracking. In fact, the CMAC system designed is a universal CMAC that can applied for any system in the given class of systems. New passivity properties of CMAC systems are examined and the relationship between passivity and closed-loop stability is derived. The utility of the CMAC NN in controlling a nonlinear system with unknown dynamics is demonstrated through numerical examples.

Neural Network–Based Intelligent Compaction Analyzer for Estimating Compaction Quality of Hot Asphalt Mixes

Continuous real-time estimating of compaction quality during the construction of a hot mix asphalt (HMA) pavement is addressed in this paper. The densification of asphalt pavements during construction usually is accomplished by using vibratory compactors. During compaction, the compactor and the asphalt mat form a coupled system whose dynamics are influenced by the changing stiffness of the mat. The measured vibrations of the compactor along with process parameters such as lift thickness, mix type, mix temperature, and compaction pressure can be used to predict the asphalt mat density. Contrary to existing techniques in the literature in which a model is developed to fit experimental data and to predict mat density, a neural network-based approach is adopted that is model-free and uses pattern-recognition techniques to estimate density. The neural network is designed to read the entire frequency spectrum of roller vibrations and to classify these vibrations into different levels. The intelligent asphalt c...

Coverage Strategies in Wireless Sensor Networks

An energy efficient cover of a region using Wireless Sensor Networks (WSNs) is addressed in this paper. Sensor nodes in a WSN are characterized by limited power and computational capabilities, and are expected to function for extended periods of time with minimal human intervention. The life span of such networks depends on the efficient use of the available power for sensing and communication. In this paper, the coverage problem in a three dimensional space is rigorously analyzed and the minimum number of sensor nodes and their placement for complete coverage is determined. Also, given a random distribution of sensor nodes, the problem of selecting a minimum subset of sensor nodes for complete coverage is addressed. A computationally efficient algorithm is developed and implemented in a distributed fashion. Numerical simulations show that the optimized sensor network has better energy efficiency compared to the standard random deployment of sensor nodes. It is demonstrated that the optimized WSN continues to offer better coverage of the region even when the sensor nodes start to fail over time. A localized “self healing” algorithm is implemented that wakes up the inactive neighbors of a failing sensor node. Using the “flooding algorithm” for querying the network, it is shown that the optimized WSN with integrated self healing far outweighs the performance that is obtained by standard random deployment. For the first time, a “measure of optimality” is defined that will enable the comparison of different implementations of a WSN from an energy efficiency stand point.

A Novel Neural Network-Based Asphalt Compaction Analyzer

Achieving the desired density during field compaction of asphalt mixes is critical to meeting the design specifications of an asphalt pavement. Existing techniques measure the density of asphalt mixes at a discrete number of points. As such, the process is cumbersome, time consuming, and is not indicative of the overall compaction achieved unless large amounts of data is collected and analyzed. In this paper, the concept of a novel neural network-based asphalt compaction analyzer capable of predicting the density continuously, in real time, during the construction of the pavement is presented. The concept is verified using laboratory data from an asphalt vibratory compactor (AVC). The compaction analyzer is based on the hypothesis that a vibratory compactor and the hot mix asphalt (HMA) mat form a coupled system having unique vibration properties. The measured vibrations of the compactor along with the process parameters such as lift thickness, mix type, mix temperature, and compaction pressure can be used to predict the density of the asphalt mat. Vibration data obtained during compaction of asphalt mixes in the laboratory is used to design and train the neural network (NN). The trained NN is then used to continuously predict the degree of compaction in real time. The proposed approach is validated through compaction studies in the laboratory. Preliminary field studies demonstrate the capability of the analyzer in predicting the density of an asphalt pavement during construction.

Inclusion of aggregate angularity, texture, and form in estimating dynamic modulus of asphalt mixes

The present study was undertaken to develop a model that utilizes aggregate shape parameters (i.e. angularity, texture and form) in estimating the dynamic modulus of asphalt mixes. Dynamic modulus tests were conducted on 20 different mixes comprised of different aggregate sources, sizes, asphalt binder, and air void levels. The coarse and fine aggregates were recovered from each mix, and their shape parameters were measured using an automated aggregate image measurement system (AIMS). A nonlinear regression model was developed to estimate the dynamic modulus of the mix in terms of its aggregate gradation, aggregate shape parameters, viscosity of asphalt binder, and volumetric properties. The correlation coefficient (R 2) for the developed model was found to be 0.95 and 0.92 on logarithmic and arithmetic scales, respectively, with a mean average relative error (MARE) of 21.9%. The performance of this model was compared with the widely accepted Witczak model that does not use the shape parameters of the aggregates. The MARE for the Witczak model was estimated significantly higher than the developed model. Results show that the dynamic modulus of the mix increases with an increase in the angularity and texture of aggregates and that the inclusion of shape parameters can enhance the prediction capability of a model.

Optimal 3-dimensional sensor deployment strategy

Recent advances in computing technology have led to the production of a new class of computing device: the wireless, battery powered, smart sensor. A Wireless Sensor Network (WSN) is an ad-hoc network composed of densely-populated tiny electronic sensing devices, distributed over an area or volume. Deployment of a sensor network can be in a random fashion or can be planted manually. These networks promise a maintenance free, fault-tolerant platform for gathering different kinds of data. In this paper, contrary to existing techniques, the deployment problem in a three dimensional space is rigorously analyzed. The problem of determining the minimum number of sensor nodes that guarantee complete coverage is studied which will serve as an optimality measure for the coverage problem.

Quality Assurance of Hot Mix Asphalt Pavements Using the Intelligent Asphalt Compaction Analyzer

Adequate compaction of asphalt pavements during their construction is essential to the long-term performance of the pavement. Current quality control techniques determine the quality at a limited number of points and are not indicative of the overall quality of the pavement. In this paper, the intelligent asphalt compaction analyzer (IACA) is used to estimate the density of an asphalt pavement during its construction and thereby determine the overall quality of compaction. A comparison of these estimates with the percent within limits (PWL) calculations based on roadway cores demonstrates that the IACA can be effectively used as a nondestructive quality assurance (QA) tool. Further, since the IACA continuously estimates the density of the asphalt in real time, inadequate compaction can be addressed during the construction, thereby improving the overall quality of pavement. Thus, the IACA can also serve as a valuable quality control (QC) tool during the construction.

The 3-Dimensional Wireless Sensor Network Coverage Problem

The rapid development of wireless technologies and embedded sensing devices has made it possible for people to monitor, control and interact with the physical world via wireless sensor networks. While holding promise in a wide variety of applications, sensor networks are driven by extremely limited energy resources. This necessitates the energy aware design and operation of the sensor network. One of the fundamental issues in sensor networks is the coverage problem. In this paper, contrary to existing techniques, the coverage problem in a three dimensional space is rigorously analyzed. Given a distribution of sensor nodes, a new distributed algorithm to choose a subset of working nodes for full coverage is derived. A computationally efficient algorithm to check for complete coverage is also proposed