Pengxiang Liu

Web-controlled Wireless Network Sensors for structural health monitoring

Wireless network sensors are being implemented for applications in transportation, manufacturing, security, and structural health monitoring. This paper describes an approach for data acquisition for damage detection in structures. The proposed Web-Controlled Wireless Network Sensors (WCWNS) is the integration of wireless network sensors and a web interface that allows easy remote access and operation from user-friendly HTML screens. The WCWNS is highly flexible in terms of functions and applications. Algorithms and tools for data analysis can be directly installed on and executed from the web server. This means WCWNS will have unlimited capabilities in performing data analysis. Data can be analyzed for damage detection either on site distributed amongst the intelligent sensors or off site either in the web server or at an end users location after downloading from the web server. This feature allows for a variety of health monitoring algorithms to be investigated by researchers of all backgrounds and abilities. In addition, both short-range and long-range communications devices handle data exchange and communications in WCWNS. The system can be setup to operate efficiently in any topological arrangement. Short-range communications devices facilitate fast and low-power local data transfer, while long-range communications devices support high quality long-distance data exchange. The proposed system is demonstrated on an experimental setup.

Active control of smart structures with optimal actuator and sensor locations

Sensors and actuators used in active control of smart structures have to be located appropriately in order to ensure maximum control and measurement effectiveness. Many placement techniques are based on the structure itself and overlook the effects of the applied control law. The optimal locations determined from open-loop system can not guarantee the best performance of the closed-loop system because the performance is closely related with the design requirements and applied controller. In this paper, we presented a method of obtaining the optimal locations of actuators/sensors by combining the open-loop and closed-loop optimal criterions. First, for open-loop system, location indices of the controlled modes are calculated on the basis of modal controllability and observability. The controlled modes are weighted based on the controller design requirements. To reduce the spill-over effect of uncontrolled modes, the location index values of uncontrolled modes are added as penalty terms. Locations with high index values are chosen as candidate locations of actuator/sensor for the next determining step on the closed-loop system. Three control techniques, optimal H2, H(infinity ) norms and optimal pole-placement, are utilized for two different control objectives, disturbance rejection and damping property enhancement. Linear matrix inequality (LMI) techniques are utilized to formulate the control problems and synthesize the controllers. For each candidate location of actuator/sensor, a controller is designed and the obtained performance is taken as location index. By solving the location problem in two steps, we reduced the computational burden and ensured good control performance of the closed-loop system. The proposed method is tested on a clamped plate with piezoelectric actuators and sensors.

Distributed computing and sensing for structural health monitoring systems

Structural health monitoring involves automated evaluation of the condition of the structural system based on measurements acquired from the structure during natural or controlled excitation. The data acquisition and the ensuring computations involved in the health monitoring process can quickly become prohibitively expensive with the increase in size of the structure under investigation. In this paper, we propose a distributed sensing and computation architecture for health monitoring of large structures. This architecture involves a central processing unit that communicates with several data communication and processing clusters paced on the structure by wireless means. With this architecture the computation and acquisition requirements on the central processing unit can be reduced. Two different hardware implementation of this architecture one involving RF communication links and the other utilizing commercial wireless cellular phone network are developed. A simple health monitoring experiment that uses neural network based pattern classification is carried out to show effectiveness of the architecture.

Design of Reduced-Order Robust Controllers for Smart Structural Systems

This paper presents a method of designing reducedorder robust controllers for smart structural systems. We consider two kinds of uncertainties in the structural systems, norm bounded unstructured uncertainty and the variations of structural parameters. The smart structure with lead zirconate titanate (PZT) sensors and actuators is represented by a high order analytical model. Model error due to the uncertainties in structural parameters is modeled by parametric uncertainty in natural frequencies and damping ratios. By generalizing balanced truncation method for uncertain system, the high order model with uncertainties is reduced to a low order model with preserved parametric uncertainty. Based on linear matrix inequalities (LMIs) and Popov criterion, a robust controller is designed to suppress the lowfrequency vibrations caused by external disturbances and to guarantee the stability and performance of the closed-loop system under uncertainties. The order of the controller is further reduced by a frequency weighted model reduction method using LMI. The procedure was experimentally tested on a smart structural test article.

Reduced order controllers for smart structural systems

The application of robust control theory, H2 and H(infinity ) controls, on the active control of smart structural systems results in higher order controllers. The implementations of these controllers need complex hardware and more power and is difficult to embed in the structure. In this paper, we present two lower-order controller design methods, indirect and direct methods. For indirect method, using linear matrix inequalities (LMIs), the full order strictly proper controller with output limitations is first synthesized for given control objectives, bounded H(infinity ) or H2 system norm. The full order controller is then reduced to a lower-order controller by a LMI based model reduction method for minimal H(infinity ) norm reduction error. For direct method, the fixed-order controller synthesis conditions are represented by LMIs with an additional nonconvex rank constraint. To utilize efficient computational tool for numerical solutions of convex LMIs, we relax the rank condition to a convex optimization. Although this relaxation can not fully solve the rank condition, in most cases, it gives a controller with lower-order. The proposed methods are tested on an experimental three-mass structure with PZT sensors and actuators. These two methods are compared based on the simulation and experimental results.

Damage detection using modal strain energy and laser vibrometer measurements

Structural health monitoring for complex systems can contribute significantly to reduced life cycle costs. Many damage detection algorithms have been proposed in the literature for investigating the structural integrity of systems. Changes in modal strain energy have been used to detect the location and extent of damage in structures. In the previous studies, the stiffness matrix is analytically derived and assumed constant even after damage. This paper reports a study on the sensitivity of the modal strain energy method to the stiffness matrix and its accuracy in detecting the location and extent of damage. The modal strain energies for each element of the undamaged structure are computed for each mode using the original analytical matrix and measured modal data. Modal data from the damaged case is used to update the stiffness matrix by a simplified matrix update scheme. This updated matrix is used to correct the elemental matrices for the damaged system. Two case studies are presented in this work. The first is an experimental and analytical model of a cantilever beam and the second, a truss model of the European Space Agency. In the first case three identical aluminum cantilever beams are used. Damage is simulated on two of them by milling 1-inch long slots at two different locations on the beams. Modal data are obtained from experiment using Scanning Laser Vibrometer (SLV) and STAR software to extract the mode shape vectors from the experimental results. These are also compared with finite element simulations of the beams. The second case is an analytical example in which damage is simulated by reducing the area of one of the truss elements hypothetically by 50%. Results from these studies show a slight improved accuracy in determining the location of damage using an updated elemental stiffness matrix. For experimental results however, modal strain energy change method does not give an accurate location of the damages. There is need for further analysis of the application of modal strain energy techniques to damage identification. The results also demonstrate the potential of SLV as a structural health- monitoring tool.

Structural health monitoring using parameter identification methods

A structural health monitoring method for determination of damages in structural system is developed using state variable model. A time-domain identification method, the subspace system identification algorithm, is first applied to get a state-space model of the structure. The identified state-space model is then transformed to two special realization forms, for determination of the equation of motion of multiple- degrees-freedom of the structure. The parameters of equation of motion, mass and stiffness matrices or damage indices are used to determine the location and extent of the damage. This method is also extended for the health monitoring of substructural system. Unlike the health monitoring of the whole structure, the health monitoring of substructure uses localized parameter identification which only involves the measurement of substructure parameters. Using this method, the number of unknown parameters and the computational requirement for each identification can be significantly reduced, hence the accuracy of estimation can be improved. Illustrative cases studies using both numerical and experimental structures are presented.

Decentralized controller design for smart structural systems

In this paper, we show how to design decentralized controllers for smart structural systems to achieve multiple objectives associated with performances defined by H∞ or H2 norm, robust stability for model uncertainty, and limited actuator authority. The decentralized controller is designed in a sequential procedure for each control loop by solving synthesis conditions represented in terms of linear matrix inequalities (LMIs). The simultaneous design of decentralized controller, formulated as a non-convex minimization problem, is also discussed. The proposed design method was experimentally tested on a three-mass structure with three control loops.

Lower-order robust controller design for smart structures

The low order controller has many advantages such as simple hardware implementation and high reliability and is very important for the successful integration of controllers with smart structures. Designing a controller with robustness to different uncertainties ins mart structure always leads to a high order controller. In this paper, two low order controller design methods are proposed. One method is to design a low order controller based on the reduced plant model. The model error between the full order model and reduced order model is considered as an additive uncertainty in the controller design to reduce the spill-over effect. Another method, controller reduction, is to find a low order controller by reducing the full order controller. The effect of the controller reduction on the system performance is taken into account by selecting a maximum allowable controller reduction error for preserving the performance. The full order controller can be synthesized to provide optimal performance or maximum allowable controller reduction error. Linear matrix inequalities (LMIs) are utilized in those methods to design the low order controllers. The variations of structural parameters, natural frequencies and damping ratios are considered in the controller design as parametric uncertainties.

Model reduction and robust control of smart structures with parametric uncertainties

The actively controlled smart structures should have robust stability and robust performance when structural parameters vary in a reasonable range. In this paper, the parametric uncertainty is defined in a framework of quadratic inequality constraint. This representation of uncertainty can effectively reduce the conservatism. A generalized balanced truncation method for continuous uncertain system represented by linear fractional transformation is presented. The model reduction method can keep the uncertainty information of the full order system in the reduced order model for robust controller design. Linear matrix inequality (LMI) conditions are given for designing a robust output feedback controller to assign the poles of closed-loop uncertain system in a constrained conic sector subregion under the input limits and unmodeled dynamics. The proposed method is demonstrated using an experimental smart structure system.