Multiple criteria-based sensor optimization for Structural Health Monitoring of civil engineering structures

The primary factor governing the accuracy of the vibration test is the sensor placement. Positioning of sensors for conducting the modal analysis test must be done with utmost care. Conducting a trial test on massive structures like dams, bridges, high-rise buildings, etc. is generally challenging and expensive. With the availability of advanced finite element software packages, it is possible to simulate models and obtain satisfactory and reliable results. Therefore, a pretest planning of sensor positions can be done, based on results obtained from finite element analysis. There are only limited number of earlier studies in the field of optimization of sensors for Structural Health Monitoring of civil engineering structures. In the practice of Vibration-based Structural Health Monitoring of structures, in order to detect both bending and twisting modes, the common practice is to fix the sensors on both sides of the structure and at equal spacing. When the sensors are placed in this fashion, there might be sharing of information between adjacent sensors, thus unnecessarily escalating the instrumentation cost. Sensor optimization helps in reducing the cost of instrumentation and maximizing the individuality in the information obtained from the sensors. The current study utilizes multiple criteria-based sensor optimization for Structural Health Monitoring of a bridge. The optimization strategy utilizes two criteria: (i) Triaxial Effective Independence and Threshold of Redundancy, in order to ensure the individuality in the information gathered from the sensors. Based on the above criteria, it was found that for a redundancy value of 0.60, 27 accelerometers were required for observing 12 modes. As the optimization strategy involves a least square estimation, it was observed that when the sensors were increased beyond 25, the condition number of Fisher Information Matrix (FIM) was stabilized to 1.5 and the mean of trace of estimation error variance minimized to a value of 0.17. The modal information extracted from modal test conducted based on the optimized layout of sensors was in good agreement with the analytically obtained results. A maximum variation of 8.55% for modal frequency and 10.13% for damping ratio was observed. In addition to these two criteria, the study also uses an information entropy-based criterion to fine tune the number and position of sensors in the final sensor set. The proposed method is also compared with two other state-of-the-art optimization methods—Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Information Entropy Index (IEI) was used to compare the layout obtained from proposed method and state-of-the-art method. The IEI values for the layout based on the proposed method were 22–40% lower than those obtained using GA and PSO.