Structural health monitoring denotes the ability to collect data about critical engineering structural elements using various sensors and to detect and interpret adverse “changes” in a structure in order to reduce life-cycle costs and improve reliability. The process of implementing and maintaining a structural health monitoring system consists of operational evaluation, data processing, damage detection and life prediction of structures. This process involves the observation of a structure over a period of time using continuous or periodic monitoring of spaced measurements, the extraction of features from these measurements, and the analysis of these features to determine the current state of health of the system. Such health monitoring systems are common for bridge structures and many examples are citied in (Maalej et al., 2002). The phenomenon of damage in structures includes localized softening or cracks in a certain neighborhood of a structural component due to high operational loads, or the presence of flaws due to manufacturing defects. Damage detection component of health monitoring system are useful for non-destructive evaluations that are typically employed in agile manufacturing systems for quality control and structures, such as turbine blades, suspension bridges, skyscrapers, aircraft structures, and various structures deployed in space for which structural integrity is of paramount concern (Figure 1). With the increasing demand for safety and reliability of aerospace, mechanical and civilian structures damage detection techniques become critical to reliable prediction of damage in these structural systems. Most currently used damage detection methods are manual such as tap test, visual or specially localized measurement techniques (Doherty, 1997). These techniques require that the location of the damage have to be on the surface of the structure. In addition, location of the damage has to be known a priori and these locations have to be readily accessible. This makes current maintenance procedure of large structural systems very time consuming and expensive due to its heavy reliance on human labor.
TopIntroduction
Structural health monitoring denotes the ability to collect data about critical engineering structural elements using various sensors and to detect and interpret adverse “changes” in a structure in order to reduce life-cycle costs and improve reliability. The process of implementing and maintaining a structural health monitoring system consists of operational evaluation, data processing, damage detection and life prediction of structures. This process involves the observation of a structure over a period of time using continuous or periodic monitoring of spaced measurements, the extraction of features from these measurements, and the analysis of these features to determine the current state of health of the system. Such health monitoring systems are common for bridge structures and many examples are citied in (Maalej et al., 2002).
The phenomenon of damage in structures includes localized softening or cracks in a certain neighborhood of a structural component due to high operational loads, or the presence of flaws due to manufacturing defects. Damage detection component of health monitoring system are useful for non-destructive evaluations that are typically employed in agile manufacturing systems for quality control and structures, such as turbine blades, suspension bridges, skyscrapers, aircraft structures, and various structures deployed in space for which structural integrity is of paramount concern (Figure 1). With the increasing demand for safety and reliability of aerospace, mechanical and civilian structures damage detection techniques become critical to reliable prediction of damage in these structural systems.
Figure 1. Examples of engineering structures that require structural health monitoring systems
Most currently used damage detection methods are manual such as tap test, visual or specially localized measurement techniques (Doherty, 1997). These techniques require that the location of the damage have to be on the surface of the structure. In addition, location of the damage has to be known a priori and these locations have to be readily accessible. This makes current maintenance procedure of large structural systems very time consuming and expensive due to its heavy reliance on human labor.
TopBackground
The damage in structures and structural systems is defined through comparison between two different states of the system, where the first one is the initial undamaged state, and the second one is damaged state. Emerging continuous monitoring of an instrumented structural system often results in the accumulation of a large amount of data that need to be processed, analyzed and interpreted for damage detection. However, the rate of accumulating such data sets far outstrips the ability to analyze them manually. As a result, there is a need to develop an intelligent data processing component that can significantly improve current damage detection systems. Since damage in changes in the properties of the structure or quantities derived from these properties, the process of health monitoring eventually reduces to a form of data mining problem. Design of data mining techniques that can enable efficient, real-time and robust (without false alarm) prediction of damage presents one of key challenging technological opportunity.
In recent years, various data mining techniques such as artificial neural networks (ANNs) (Anderson et al., 2003; Lazarevic et al., 2004; Ni et al., 2002; Sandhu et al., 2001; Yun & Bahng, 2000; Zhao, Ivan & DeWolf, 1998;), support vector machines (SVMs) (Mita & Hagiwara 2003), decision trees (Sandhu et al., 2001) have been successfully applied to structural damage detection problems. This success can be attributed to numerous disciplines integrated with data mining such as pattern recognition, machine learning and statistics. In addition, it is well known that data mining techniques can effectively handle noisy, partially incomplete and faulty data, which is particularly useful, since in damage detection applications, measured data are expected to be incomplete, noisy and corrupted.