A neural network model for predicting fatigue life of carbon steel, copper alloy and aluminium alloy under constant amplitude loading
Loading...
Date
2005-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand)
Abstract
Fatigue failure has been intensively studied for a more than a century. Fatigue in metals often progresses by the initiation of a single crack and its intermittent propagation until catastrophic failure, which occurs with or without warning. Fatigue testing of materials is a time consuming and expensive process. If every fatigue condition is to be investigated, this results in many different experiments involving many variables. The problem is compounded by the proliferation of new materials requiring evaluation and increasingly demanding applications for these materials. This derives the need for more accurate fatigue life prediction. A potential solution to this requirement is offered by artificial neural networks (ANNs). ANNs are an alternative to conventional programmed computing and are based on the operation of the brains. Now ANN has been used successively to many engineering applications. ANNs also offers a means of handling many multi-variable parameters for which an exact analytical model does not exist or would be difficult to develop. ANNs also provide a compact method of considering large amounts of data and simple means of assessing the liking outcome of a complex problem with a specified set of conditions. The analysis of fatigue life data requires all these capabilities. In the present investigation an artificial neural network (ANN) model has been developed to predict the fatigue life for a given probability of failure for carbon steel, aluminium alloys and copper alloys. Architecture of the of the ANN model consists of two hidden layers and –24-6-; -24-6-; -18-6-; -30-6-; hidden nodes for 0.41 and 0.19carbon steel, plain carbon steel, aluminium alloys, copper alloys respectively with an output target. This has been established through a series of trials, which allows convergence in a shortest training time. The inputs for training the ANN model includes chemical composition, mechanical properties, applied stress and probability of failure. The predicted fatigue life is found to be well within a scatter band of 2 with regression coefficient r 0.987. The average rms error is 0.00004, 0.00005, 0.00004, 0.00004 for 0.41 and 0.19 carbon steel, plain carbon steel, aluminum alloys, copper alloys respectively. The predicted results are also compared with the other statistical model available in the literature.
Description
Keywords
null