Neural network prediction of friction coefficients of rosemary leaves

Abstract

AbstractIn the current study, the static friction coefficient (SFC) and dynamic friction coefficient (DFC) of rosemary leaves were predicted. In this regard, a wavelet‐based neural network (WNN) was proposed as a tool for mapping of the inputs including moisture content (3.7–23 d. b.%), contact surface (aluminum, rubber, glass, galvanized steel, and plywood), and sliding velocity (1.25–16.5 cm/s) and output data (friction coefficients of rosemary leaves) which were obtained as a result of laboratory activities. Also, 80% and 20% of the total data set were used for training and testing the WNN, respectively, while the statistical parameters for test data set of the proposed NN (MAPE = 6.1310%, RMSE = 0.0468, R2 = 0.956) were compared with those obtained from conventional NNs, such as multilayer perceptron (MLP) (MAPE = 4.1787%, RMSE = 0.0248, R2 = 0.968) and radial basis function (RBF) (MAPE = 10.0311%, RMSE = 0.0591, R2 = 0.831). The results indicated that the accuracy of the MLP NN was slightly better than that of the proposed WNN; however, the learning time of WNN was much less than the MLP NN. The high capability of the proposed NN in the prediction of the SFC and DFC of the rosemary leaves was demonstrated by our findings.Practical applicationsData on the frictional characteristics of agricultural products may be useful for the design of specific processing equipment, and laboratory measurements may also provide some interesting results that can guide designers. This study deals with the static friction coefficient (SFC) and dynamic friction coefficient (DFC) prediction of rosemary leaves with a view of performing predictive tools useful in the design of postharvest equipment. Rosemary has been widely cultivated as an aromatic and medicinal plant. The extracted essential oil from rosemary has several applications due to its chemical composition with beneficial properties. The prediction of friction coefficients is performed by a wavelet‐based neural network (WNN). The proposed NN has a multiresolution nature that enables accurate modeling and also fast training. The results showed that the proposed NN is a powerful tool to predict the friction coefficients of rosemary leaves.