Measurement, modelling and optimization of the average temperature at the tool work interface for turning of AISI 1040 steel using ANN-GA methodology

Abstract

Due to the mechanical properties and machinability of AISI 1040 steel, it has a broad range of industrial applicability. The present work investigates the dependence of the average temperature generated at the tool work interface on machining parameters such as cutting speed, feed rate and depth of cut. The machining parameters were varied at three equally spaced levels. Experiments were carried out following combinations in Taguchi's L 27 orthogonal array. A simple and cost-effective tool work thermocouple was devised and calibrated to obtain the tool work interface temperature. A 3-5-1 feed forward artificial neural network (ANN) model trained using the Levenberg-Marquardt (LM) back propagation algorithm resulted in efficient modelling of the complex relationship between the average cutting temperature and machining parameters (R-value of 0.99). The trained network was optimized using genetic algorithm (GA) to predict optimal turning parameters. This methodology has been termed as ANN-GA method. A significant reduction in the average cutting temperature could be realized due to optimization using ANN-GA method. Analysis of variance revealed highest contribution from the cutting speed and its square term in controlling the cutting temperature at tool work interface.

The use of machine learning and metaheuristic algorithm for wear performance optimization of AISI 1040 steel and investigation of corrosion resistance

“…33 The final algorithm and model used in the present work has been selected finally after several trials and literature review. 14,15,19,31 The feed-forward network is made up of neurons that are divided into three layers: input, hidden, and output, with weights connecting them. During the training stage of the learning process, these weights are modified.…”

“…During the training stage of the learning process, these weights are modified. The output O j from a j th neuron at any layer may be calculated as 14 :…”

“…Finally, this objective function may be coupled along with GA to obtain optimal parametric combination. This is the proposed ANN-GA 14,19 methodology and has been represented by the flow chart as shown in Figure 3. MATLAB is used to develop the ANN model.…”

“…An 'adaptive feasible' mutation function was also considered. After multiple trial runs and literature review 9,14,19,31 all of these parameters were carefully chosen. Figure 6 shows the variation of wear rate (fitness values) with generations for regression-based GA optimization.…”

“…Support vector regression (SVR) and ANN are two ML techniques that are used in predictive modelling. Researchers have demonstrated the use of experimental design and ML techniques in the prediction and optimization of machining of carbon steel 14 as well as tool steels. [15][16][17][18] When these models are combined with GA, they can be utilised to generate optimal response parameters.…”