Predicting the mechanical properties of polypropylene and recycled polypropylene by the application of sigmoid and ReLU functions in neural networks

• An ANN model was developed to predict mechanical properties of PP/r-PP composites. • The model uses ReLU and Sigmoid activation functions for performance comparison. • Accurately predicts tensile strength, modulus, and elongation at break. • 2–5–5–5 ANN specified with layer-wise equations and loss function. • Rationale clarified for selecting ReLU and Sigmoid activation functions in this application. This research analyzed artificial neural network (ANN) models for predicting the mechanical properties of polypropylene (PP) composite plastics blended with recycled polypropylene (r-PP), comparing the efficacy of ReLU and Sigmoid activation functions under similar training conditions. The input variables consist of the composite plastic blending ratio and silane. The predicted results included 5 properties: tensile strength (σ T ), Young's modulus (E), elongation at break (e), impact resistance (F), and hardness (H shore D ). A dataset including 30 samples was divided into training, validation, and testing sets in a 70/15/15 ratio, using normalization and early stopping methods. The model's efficacy was determined using MSE, RMSE, and R². The test results indicated that the model using the Sigmoid function achieved better accuracy, with R² values of 0.95, 0.97, and 0.99 for σ T , E, and H shore D , respectively. The MSE was around 0.8, while the RMSE was below 1 unit, with R² values of 0.82 and 0.93 for σ T and E, respectively. In the conclusion, the choice of activation function profoundly influences the ability to predict of the ANN and influences the design of network models for improving the prediction efficiency of PP:r-PP material properties in the industrial field.