Damage detection based on acceleration data using artificial immune system

Nowadays, Structural Health Monitoring (SHM) is essential in order to prevent damages occurrence in civil structures. This is a particularly important issue as the number of aged structures is increasing. Damage detection algorithms are often based on changes in the modal properties like natural frequencies, modal shapes and modal damping. In this paper, damage detection is completed by using Artificial Immune System (AIS) theory directly on acceleration data. Inspired from the biological immune system, AIS is composed of several models like negative selection which has a great potential for this study. The negative selection process relies on the fact that T-cells, after their maturation, are sensitive to non self cells and can not detect self cells. Acceleration data were provided by using the numerical model of a 3-story frame structure. Damages were introduced, at particular times, by reduction of story's stiffness. Based on these acceleration data, undamaged data (equivalent to self data) and damaged data (equivalent to non self data) can be obtained and represented in the Hamming shape-space with a binary representation. From the undamaged encoded data, detectors (equivalent to T-cells) are derived and are able to detect damaged encoded data really efficiently by using the r-contiguous bits matching rule. Indeed, more than 95% of detection can be reached when efficient combinations of parameters are used. According to the number of detected data, the localization of damages can even be determined by using the differences between story's relative accelerations. Thus, the difference which presents the highest detection rate, generally up to 89%, is directly linked to the location of damage.