Application of Artificial Neural Networks for Membrane Damage Detection in a Bending-Active Construction

Bending-active building constructions consist of a membrane shell, supported by a flexible bearer frame. Top chord of the frame, made of high-strength low-modulus struts, takes on arch-like shape when assembling and installing the construction in situ. The membrane, protecting the building from environmental impacts, contributes to load carrying capacity enhancement of the bearer frame. Reliability of a bending-active structure in the whole directly depends on the membrane serviceability. The membrane, made of architectural fabrics, may get damaged in form of local slits. The slits, being accidental, emerge both in loaded and not loaded areas. They are initially hidden from view, but tend to quickly increase due to stress concentration. Thus, detection of membrane damage in the early stage is an important problem to be solved for reliability enhancement of the structure. Membrane slit detection is proposed to be implemented by means of Artificial Neural Networks (ANNs). The input layer of the networks is fed with theoretical indications of gauging equipment embedded in the structural frame, while the output layer produces the slit location. Supervised learning approach and gradient descent technique are used to obtain weight coefficients of the network. The present work contributes to structural damage detection in the early stage, allowing maintenance personnel taking corrective measures in due time.