Predicting Strain Rate during Ir Imaging of Tensile Deformation Using Mlp Based Ann

During tensile testing, a part of the mechanical work done on the specimen is transformed into heat energy. Using thermal imaging, it is possible to detect and measure the variation of temperature and relate it to the deformation behaviour of the material. It is now well established that Artificial Neural Networks(ANN) can be used to solve complex non-linear classification and prediction problems. In thermal NDE, ANNs have been used in a very limited way, mainly for defect analysis and classification. In this paper, we establish the feasibility of using multi-layered perceptron based ANN for predicting the strain rate during tensile deformation of nuclear grade Type 316 Stainless Steel with prediction errors less than 2.7%. This approach provides feasibility to interpolate and determine the stress or temperature based on minimum experiments and can also be applied to industrial components. Introduction : It is well known that during tensile testing, a part of the mechanical work done on the specimen is transformed into heat energy. The maximum temperature and the rate of temperature rise is related to the nature of the material, test conditions and the deformation behaviour of the material during loading. Using infrared (IR) imaging, it is possible to detect the variations in temperature and consequently predict the deformation behaviour of the material. Thus, IR imaging can be viewed as an adjunct to conventional mechanical test techniques. Apart from the basic advantage of non contact measurement, the technique offers itself to the detection of transient exothermic or endothermic changes that cannot be normally observed through conventional testing practices. This kind of work, though initiated by Wilburn [1] and Y.Huang [2], is limited. In all the investigations pursued so far, experiments have been on measurement of temperature and relating it to the mechanical properties. At the authors’ laboratory, for the first time both IR imaging and acoustic emission (AE) have been successfully used to characterise the various stages of tensile deformation of a nuclear grade AISI type 316 stainless steel (SS) [3,4]. Recent years have seen extensive applications of ANN to solve complex problems, particularly where the analytical relationship between the input and output values of a system are unknown. ANNs are powerful, robust and adaptive tools for detection and classification of features under changing signature or environmental conditions. They are also known for their processing speed, high classification accuracy, low-sensitivity to noise and flexible thresholding capability. These networks can be trained from a set of observed examples (called the training set) and then applied on other similar data not contained in the training set. In the field of thermal NDE, the reported applications of ANN are only a few. ANNs have been primarily used for defect feature extraction and classification [5-12] and in one case for depth estimation [13]. The application of ANN for analysis of the thermal measurements obtained from tensile tests and the concept of predicting the temperature or stress, based on experimental interpolations has not been attempted so far. Such an approach provides opportunities to interpolate and determine the stress or temperature based on minimum experiments. In this paper, we explore the feasibility of this new idea of using multi-layered, error-back-propagating, feed-forward artificial neural network (MLP-ANN) for predicting the temperature and strain rate during tensile deformation of nuclear grade AISI Type 316 SS. General Architecture of the MLP : The multilayered perceptron (MLP) is one of the most versatile artificial neural networks and is popular for data classification and prediction applications. The basic architecture of the MLP neural network is shown in Figure 1. The first layer, known as the input layer,