Measurement system design for damage detection using model-free data interpretation methods

Designing a measurement system is usually performed using only engineering judgment and experience. Thus, it may result in either a large amount of redundant data or insufficient data leading to ambiguous results. This paper presents a systematic approach to design measurement systems for damage detection using model-free data-interpretation methods. The approach provides decision support for two tasks: (1) determining the appropriate number of sensors and (2) placing these sensors in the most informative locations. These tasks correspond to two steps. The first step is to evaluate the performance of measurement systems in terms of the increasing numbers of sensors. Engineers can then determine the number of sensors to be used at the point where the improvement of the performance is modest with the addition of more sensors. The second step is to configure measurement systems in order to place the sensors at the most informative locations. This step involves evaluating the performance of measurement configurations using three criteria: minimizing the number of non-detectable scenarios, minimizing the average time to detection and maximizing the average damage detectability. The result obtained from three-objective optimization is a set of non-dominated solutions. Thus, to choose the best compromise solution, a multi-criteria decision-making strategy is employed. A railway truss bridge in Zangenberg (Germany) is used to illustrate the applicability of the approach. Measurement systems are designed to monitor the bridge where measurement data are interpreted using two model-free (non-physics-based) methods: moving-principal-component analysis and robust-regression analysis. The configurations that are found do not correspond to configurations that would be selected using engineering experience only. Results demonstrate that the proposed approach is able to provide engineers with decision support in determining the appropriate number of sensors and identifying the most informative locations that lead to more rational measurement system designs.