Applications of artificial neural networks in machining processes: a comprehensive review

In the present era of artificial intelligence and machine learning, artificial neural networks (ANNs) have appeared as one of the potent tools in modeling the complex nonlinear relations between the input and output parameters in many of the machining processes, and helping the process engineers in predicting the tentative response values for given sets of input parameters. This paper conducts a systematic and content-wise analysis of a considerable number of research articles available in some of the reputed scholarly databases dealing with application of ANNs as effective predictive tools in three main machining operations (turning, milling and drilling) with an aim to extract the relevant information with respect to types of the ANN, their corresponding learning algorithms and activation (transfer) functions, optimal architecture, and statistical metrics employed to evaluate their prediction performance. It is revealed that the past researchers have maximally applied those ANN models in turning (42.07%), followed by milling (34.48%) and drilling operations (23.45%). In those machining operations, cutting speed, feed rate and depth of cut have been treated as the most important input parameters, and surface roughness as the predominant predicted response. Among different ANN models, feed-forward neural networks (94.44%) have been the most preferred choice among the researchers mainly due to their simple topology and availability of well-structured experimental datasets. On the other hand, Levenberg–Marquardt (58.3%), Sigmoid (31.6%) and mean squared error (47.2%) are identified as the most favored learning algorithm, activation function and statistical measure, respectively. This review paper would act as a ready reference to the process engineers in providing the optimal architecture of the ANNs, thus relieving them from additional computational effort. Finally, advantages and limitations of ANNs are summarized along with future research directions.