Optimization and prediction of the hardness behaviour of LM4 + Si3N4 composites using RSM and ANN: A comparative study

Abstract

In the present work, LM4 + Si3N4 (1, 2, and 3 wt.%) composites were fabricated using the two-stage stir casting method. Precipitation hardening treatment was carried out on the cast composites and hardness results were compared with as-cast specimens. Microstructural analysis was performed using Scanning Electron Microscope (SEM) images to validate the existence and homogenous distribution of reinforcement in the matrix. LM4 + 3 wt.% Si3N4 composite with multistage solution heat treatment (MSHT) and aging at 100°C showed higher hardness viz., 124% improvement when compared to as-cast LM4 due to the uniform distribution of Si3N4 and precipitation of metastable phases during the heat treatment process. The microhardness values of the fabricated composites was investigated using Artificial Neural Network (ANN) and Response Surface Methodology (RSM). Both RSM and ANN models predicted hardness values close to experimental values with minimum error, and the prominence of aging temperature in the improvement of hardness was observed. The data obtained illustrate that the proposed regression model can accurately predict hardness values within the constraints of the factors under consideration. Based on the error values it can be concluded that the ANN model can deliver results with higher accuracy than the RSM model.

Comparison of RSM, ANN, and ANFIS in predicting the weld strength of Laser transmission welds of oak wood powder-reinforced polypropylene absorbent part.

“…ANFIS generally emulates given data by dividing the data set (input-output data pairs) into training and testing data, in which the training data determines the initial premise parameters for membership functions [20]. In this study, ANFIS predicts weld strength from experimental data.…”

“…Sugeno was used to model the response (weld strength). The ANFIS training method begins with defining the membership functions, after which the neural network processes all learning data, changes the input parameters to the input-output connection, and finally, the neural network learns [20].…”

Prediction of weld strength of LTW of Natural plant-based fiber reinforced Polypropylene using RSM, ANN, and ANFIS - a comparative study

“…Figure 10: model training, validation, and testing MSE performance (a), Regression plot of the ANN model (b).The hidden layer had the tan-sigmoid transfer function while the output layer consisted of the linear transfer function. The regression plot of the chosen model is shown in Figure(10) which consists of an R-value of (0.95068) training, (0.99603) validation, (0.96855) testing, and an overall R-value of (0.96318) for the recommended model to give the ideal match throughout all the data set.Trainlm training function updates weight and biases values based on the Levernberg-Marquartdt which is expressed as[18],[19] …”