Comparison of regression model with multi-layer perceptron model while optimising cutting force using genetic algorithm

Abstract

Cutting force is an important measurement in machining to predict the life a tool and to estimate the power required. Standard mathematical models can be used to minimise the cutting forces (Fz). A comparative study is made in modelling the cutting force (Fz) through L18 and ANN models, while turning of AISI 1040 steel with tungsten carbide cutting insert. The input parameters that are considered are volume concentration, MQL flow rate, speed, feed and DOC The experiments were carried out using L18 Taguchi design process, and the analysis was made using SPSS to determine the model adequacy and also the influencing parameters effecting cutting force. Multi-Layer Perceptron (MLP) which is a class of feedforward artificial neural network was adopted to develop the mathematical prediction models. The predictive capabilities of L18 and ANN models were further compared in terms of their mean absolute percentage error. The results concluded that the ANN model is better in predicting the response with 3.78% mean absolute percentage error where as L18 model has an average percentage error of 7.58%. It was observed that the cutting force was reduced through ANN method by 7.814% when compared to L18 model. Both the models were used to further optimise the cutting force Fz through genetic algorithm. The results showed that the ANN model predicted optimal cutting force already, and the usage of genetic algorithm as a post processing step did not improve it any further.