The Artificial Neural Networks (ANN) is a computing system inspired by the human brain of biological neural networks. In biological systems, the human learns to do a task by plotting examples in the mind. For example, recognition of voices, images. ANN is also used for machine translation, social network, speech recognition, and computer vision. The structure of ANN consists of a collection of connected units or nodes called artificial neurons like analogous to biological neurons in human brain. A connection between analogous to synapse transmit a signal from one to another. The Nodes or units receives the signal can analyze and process it and then send this signal to another layer of connected neurons. In general, ANN works on a connection between artificial neurons, which is a real number. The result of an ANN is calculated by a non-linear function of the sum of its inputs. The connection has a weight factor to adjust a learning ability. If the weight increases, then the strength of the signal increases and if the weight decreases, then the strength of the signal decreases. It has threshold value in some cases for rectifying the signal for the further signal to be sent (Wikipedia/Artificial_neural_network). The ANN is working on different layers and they perform different kinds of transformations on their inputs.
The main components of an artificial neural network are Neurons (Units or nodes), connection (weights), non-linear function (Propagation function), and learning rule. As shown in Figure 3 that the Neurons are representing a Circle. The weights are on the connection between all neurons. The propagation function applies in a hidden layer. The learning rule will use for the modification of the parameters of the neural network, to optimize the desired result as an output from the given input. The process only modifies the weights and threshold of the neurons within the system.
Figure 3 The three-layered feed forward system for the Artificial neural network
ANN also used in Automated detection and fault diagnosis of machine conditions like generalization or classification problems based on learning pattern from input data of empirical data modeling. Though, the traditional neural network methods have borders on simplification giving upsurge to models that can overfit the data. The deficit is due to the optimization algorithms used in ANN for selection of parameters and the statistical measures used to select the model. (Samanta, 2003). For real-world operation, ANN requires too much input datasets for training and theoretically, backpropagation is a critical part of most ANN.