Dimensional analysis and ANN simulation of chip-tool interface temperature during turning SS304

OBRABOTKAMETALLOV MATERIAL SCIENCE Том 23 № 3 2021 EQUIPMEN . INSTRUM TS Vol. No. 4 2021 Arti fi cial neural network Arti fi cial neural network ( ANN ) is a computational technique that can model relationships between input and output parameters. There are different types of ANN , however, the most used is the multilayer perceptron ( MLP ). A typical MLP architecture is shown in Fig. 5. MLP is characterized by three different layers, namely the input layer, the hidden layer, and the output layer, which consist of an interconnected group of arti fi cial neurons. Each neuron in a layer is connected to all the neurons in adjacent layers. The number of neurons present in the input layer and output layer is equal to the number of input variables and corresponding output values. The number of hidden layers and the neurons in those layers is user- de fi ned. To predict output with higher accuracy, training or learning of the developed network is essential. The procedure used to perform the learning process is called a learning algorithm, the function of which is to modify the synaptic weights of the network in an orderly fashion to attain the desired output. There are various algorithms to train a neural network. One of the most preferred training algorithms is the er- ror backpropagation algorithm. For a typical ANN algorithm, let x 1 , x 2 ,… x 3 be an input data, y 1 , y 2 ,… y n be the desired output, and o 1 , o 2 … o k be the output obtained from the output layer of the network when x 1 , x 2 ,… x 3 is presented at the input layer. At the fi rst step, the weights and thresholds are initialized. Then, the output of each neuron f ( wi ) is calculated from the input data and initialized weights which lead to the fi nal output prediction of the network. Then, the error at i th output node ( o i –y i ) is calculated. Further, the weights between the hidden layer and output layer are modi fi ed based on an error at each output node. And weights in the previous layers are modi fi ed by back-propagating errors calculated at output layer nodes [20]. This process is repeated for a set of input and output of training data. The training stops when the output of the neural network is suf fi ciently close to the desired output for each set. ANN model is developed to predict the chip-tool interface temperature considering the input parameters as the tool type, cutting speed, and feed using MATLAB Toolbox . The ANN architecture has three layers Fig. 5 . Typical ANN architecture Ta b l e 6 The chip-tool interface temperature with different models and tools Expt. no. Uncoated tool TiAlN coated tool TiN / TiAlN coated tool SM DA ANN SM DA ANN SM DA ANN 1 834 838 837 973 963 941 996 996 987 2 889 895 936 1,021 1,017 1,027 1,047 1,049 1,045 3 926 918 942 1,054 1,055 1,041 1,081 1,082 1,049 4 955 965 939 1,085 1,103 1,099 1,104 1,114 1,098 5 1,019 1,026 1,037 1,140 1,137 1,169 1,161 1,161 1,172 6 1,061 1,061 1,038 1,176 1,183 1,217 1,199 1,203 1,195 7 1,056 1,034 1,078 1,176 1,182 1,188 1,191 1,191 1,195 8 1,126 1,114 1,119 1,235 1,229 1,210 1,252 1,245 1,254 9 1,173 1,189 1,178 1,275 1,288 1,261 1,293 1,301 1,275