Keywords: lifetime prognosis, model-free prediction, accelerated life testing, acceleration function, neural network, textile reinforcement.

In this paper a model-free concept for the lifetime prediction of novel materials in civil engineering is presented. Due to its capabilities and features, this concept is of particular interest in view of the economic goal and trend of shortening the time for the introduction of novel materials into engineering practice. The model-free prediction concept has been developed to investigate the long-term behaviour of textile-reinforced concrete. This novel building material is developed for strengthening and repair of aged reinforced concrete structures, in particular. The extensive investigations are conducted within the Collaborative Research Centre 528 in the Department of Civil Engineering at Technical University Dresden, Germany.

The lifetime of textile-reinforced concrete depends on a variety of environmental influences such as applied stresses, temperatures and weathering. Experimental investigations over periods relevant to structural lifetime under natural conditions are, however, not realizable due to temporal and financial restrictions. Further, a clear separation of long-term effects with respect to individual environmental influences and the formulation of an associated mechanical model are problematic due to the limited amount of experiments and not yet complete physical-mechanical insight. In order to obtain an economical solution procedure, two effective components are combined to develop the novel prediction method: accelerated life testing and neural networks. The concept of accelerated life testing is applied to conclude from short-term tests to long-term behaviour. The test duration is shortened by applying defined stresses with an increased intensity. The lifetime of the specimens is measured for different stress levels. For each stress level, a random sample is obtained for the lifetime, which characterises a respective probability distribution. Based on this information, a relationship has to be found between the stress level and the associated probability distribution of the lifetime of the specimens to infer a probability distribution that belongs to the natural stress level associated with the serviceability conditions of the real structure. This relationship is denoted as acceleration function a(.). Commonly, a differential equation can be formulated and solved to find the acceleration function - provided that the physical phenomenon behind the experiment is known.

In the case of textile reinforced concrete, however, the physical insight to formulate a specific differential equation is not available. Thus, a model-free concept based on neural networks is employed to extract and to evaluate the complex information from the experiments. The empirical distribution functions for the lifetime under the individual stress levels are fed to a neural network to analyse the stress-lifetime dependencies and to approximate the acceleration function a(.). This corresponds to the solution of a differential equation with the aid of a neural network. Finally, the prediction of the probability distribution for the lifetime of the specimens under natural stress is generated with the neural network.

The developed model-free prediction method is demonstrated by way of two examples. First, a numerical example is investigated for validation purposes. Then, an application to the lifetime prediction of textile-reinforced concrete specimens under tension load is presented.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £62 +P&P)