Principal Engineer, Dr M Khan, MAESP, MAIENG
Civil engineering projects and the construction of buildings and plants contribute a fair percentage of Southern African GNP. International trends to hold contractors or owners responsible for structural guarantees over prolonged periods of time, so called Design Build and Operate contracts, are changing the nature in which structures will be built in the next century.
Artificial Intelligence (AI) techniques in structural / construction engineering have already found potential applications. AI has the potential to help design engineers develop vital structural design infrastructure.
Less obvious is the need for development of acceptable sensor technology for applications in structural health monitoring, or even truly Intelligent Structures (IS) that can adapt to changing conditions, such as seismic activity. The evolution of sensor systems suitable for civil engineering projects or plant construction, as well as control mechanisms capable of great flexibility or adaptation, is discussed in this paper. The paper is intended to be a brief technology overview and is by nature a tutorial.
The Intelligent System concept
The age of the Intelligent or Smart Structure is the natural successor to modern man’s progression from the Stone Age through to the present Synthetic Materials Age . The nineteen nineties has seen the intelligent or smart structure development emerge as a truly multidisciplinary field. A Smart structure is defined [ 2 ] as “a nonbiological structure that has definite designed in purpose and will, and some means to achieve that purpose.” It is in the achievement of these objectives that electronics and information technology will find an increasing role to play in the development of adequate sensor, control and actuation mechanisms.
AI in civil engineering
The need for structural infrastructure to assist in design becomes vital if one considers the many uncertainties and variable conditions that design engineers have to deal with. Artificial intelligence techniques involving “expert systems” can play an important role in the design phase of large civil structures. [ 3 ] Design case studies and databases can for instance be extremely valuable since structural life cycles are often measured in centuries or decades. Design engineers could therefore apply components and concepts from one design to another quite easily. The fact that some knowledge is standardised as codes of accepted practice lends itself quite naturally to incorporation in AI databases. AI software for civil engineering is still in its embryonic phase and should be exploited more fully. A knowledge based framework that can assist in the often difficult task of finite-element structural modelling have been proposed. [ 4 ]
The continuous development of automated or semi-automated techniques for knowledge acquisition would allow engineers greater flexibility in experimenting with AI Learning tools.
Photonic sensors for civil engineering
Photonic sensors for civil engineering structures is yet another exiting and logical application of this new and far superior class of sensor technology. Introducing sensors into a large structure was not previously viable due to:
- Extensive interconnection requirements using twisted wire cables and coaxial cables.
- Prohibitive electrical power requirements
- Expensive signal conditioning
- Sensor mountings in a potentially hostile environment (concrete)
What is photonic sensor technology all about?
These sensors operate by measuring parametric changes in light between a light source and a detector due to external physical factors. Modern photonic sensors generally require only a fibre optic cable interconnection and can be embedded in a structure. With the revolutionary changes that took place in the last decade in fibre-optic cable technology, as well as research into viable point and distributed photonic sensors, these techniques are set to dominate instrumentation in the next century.
Figure 1: Photonic sensors using only fibre-optic cable embedded in a structure
Sensors for bridges and buildings
Adding sensors to a bridge or building is not a simple task. Monitoring systems often have to be designed in at the conception of a project in order to meet specific requirements. Photonic sensors meet the structural requirements for civil engineering in being able to:
- be embedded in the structure itself
- uses no electrical cables , only fibre optic sense and transmission lines
- requires low operating power
- have high noise immunity
These factors make photonic sensors ideal for civil engineering applications. The essential sensing requirements for civil structures can be summarised as: 
- the need to sense external loads on a structure
- the need to sense movement relative to a fixed position
- the need to measure the effects of external loads on a structure
- the need to measure stresses/loads within a structure
- the need to measure chemical stability of the materials that make up the structural components.
Construction loads, both during and after construction may also be important to measure and may increase site safety as well as providing valuable data for improved construction techniques. Latent responses such as the response of bridges to heavy truck loads, peak hour traffic, wind and earthquakes, are vital to measure and could alert owners to possible remedial action. The cost saving in maintaining a structure that has the capability of damage assessment and health monitoring is self evident.
Control of intelligent civil structures
The smart materials age is changing the way in which structures are viewed today. A structure that in general can sense, and not only sound an alarm but also adopt a temporary or permanent course of action to restore things back to “normal”, is the class of structure that one can classify as “intelligent.” Intelligent civil structures require robustness, both in terms of the sensor technology as well as the proposed control system. Control theory applied to a civil structure is again relatively new, and is of course implicitly linked to adequate sensor technology.
The theory of control
Control theory have found a wide spectrum of applications which are of major technological importance. A control system generally have the components outlined below:
Figure 2: Closed loop control system
Disturbances acting on the structure causes conditions that are sensed by the photonic sensors. The sensor response is fed back to the controller, programmed or designed with a control law that determines what actuator response is needed to keep the disturbances from degrading the structure. In classical control theory, the stability of a negative feedback closed loop system, outlined in figure 2, is well established.
In order to design a control system, a reasonably accurate model of the structure and its dynamics must be attained. Generally, feedback design of large systems without model approximations is unrealistic. It is usually unnecessary because the input/output behaviour of large systems can usually be approximated with lower order models. The model order here refer to the order of the differential equations governing the dynamic behaviour of a structure.
Implementing a classical control methodology to structure would typically follow the steps outlined in figure 3.
Figure 3: Steps in the design of a control system
Neural networks for control
Obviously the need to achieve fast, accurate and stable control is the aim of all control systems design. The term neural network (more properly : artificial neural network) have a specific meaning in that it refers to any architecture that has a massively parallel interconnection of basic processors. A neural network or processor is generally made up of “neurones” , which forms the basic computational element of the processing system. Neural nets (NN’s) are not, properly speaking, just an extension of artificial intelligence (AI). A neural network can be trained and on the basis of this training make generalisations and decisions. This is not the realm or objective of AI systems.
A general neural network is illustrated below in figure 4.
Figure 4: A three input neural network
From the above illustration, one can see that NN’s are ideally suited for complex non-linear systems with multiple inputs and various degrees of uncertainty in the approximate model of the system to be controlled.
The main advantages of neural networks in the control of intelligent civil structures are:
- the ability to accept and process multiple sensor inputs
- the ability to adapt
- the ability to provide a truly “model free” controller
Neural networks already found applications in simple manufacturing processes such as regulation in the manufacture of rebars for reinforced concrete construction.[6 ] In another application , damage assessment, employing photonic sensors and neural network processors have also been demonstrated.  Thus far, neural network generalisation techniques have found more applications in modelling, design and estimation procedures for civil engineering , than in prospective control applications.
The development of effective AI techniques can enhance the development of complex structures that are intended to be classified as “intelligent.” In order to generally develop a large intelligent structure, the need for an appropriate sensor technology, such as that provided by photonic sensors, would be ideally suited for this type of application. Applying control theory to a structure is essential if degradation is not only intended to be monitored but acted upon by the structure itself. Finding a systems model for controller development could however be demanding.
Present applications of neural networks to the control processes of systems are dependent on a fair amount of heuristics. The advantages of using NN’s in the control of complex systems are however not to be treated lightly. Together with photonic sensors, it is the next step in the evolution of an acceptable and optimal control architecture for complex systems such as a structure.
 Kekana M and Sun B, “A Technical review of Smart Materials and Structures. Part 1: An Introduction,” Submitted to Journal of SA Mechanical Engineers, June 1997.
 Spillman WB, “Smart Structures,” Optical Engineering, Vol 36, No 7, July 1997.
 Garret JH, Smith FC, “AI Applications in Structural/Construction Engineering,” IEEE Expert, August 1996, pp 20-22
 Turkiyyah GM, Fenves SJ, “Knowledge-Based Assistance for Finite- Element Modelling,” IEEE Expert, June 1996, pp23 -32
 Culshaw B, Michie C, Gardiner and McGown, “Smart Structures and Applications in Civil Engineering,” Proceedings of the IEEE, Vol 84, No.1 January 1996, pp78 -86
 Dunston PS, Ranjithan S, “Neural network Model for the Automated Control of Springback in Rebars,” IEEE Expert, August 1996, pp45-49.
 Grossman B, Gao X and Thursby M, “Composite damage assessment employing a neural network processor and an embedded fibre-optic sensor array,” SPIE Vol.1588, 1991, pp64-69
 Teh CI et al, “Prediction of Pile Capacity using Neural Networks,” Journal of Computing in Civil Engineering, April 1997, pp129 -132.
[9 ] Mukerjee A and Despande JM, “modelling initial design process using Artificial Neural Networks,” Journal of Computing in Civil Engineering, Vol 9, No 3, July 1995.