PM—Power and Machinery: Nozzle Wear Rate Prediction using Regression and Neural Network

Nozzle wear can cause overapplication of chemicals, which results in increased production costs, and may also cause environmental damage due to excess chemicals contaminating both ground and surface water. The objectives of this research were to predict the nozzle wear rates for four fan nozzles using regression and neural network techniques and drawing comparisons between the two techniques from the prediction results. Both regression and neural network methods were applied for modelling nozzles made of four different materials, namely, brass, nickel-coated brass, plastic and stainless steel. The model relates wear rates to usage time at three different operating spray pressures: 138, 256 and 552 kPa Dummy variables were used to represent different nozzle types. One hundred and sixty-eight observations were used to build a regression model. Forty-four observations, chosen from within the sample data, were used for model validation. One of the United States Department of Agriculture, Agricultural Research Serviceʹs (USDA-ARS) data set, considered to be out-of-sample data was also used for model validation. Log–log linear model was used to build the regression model by using Statistical Analysis System (SAS). A neural network software, NeuroShell Easy Predictor, was chosen to develop a neural network model. Model adequacy was established both by visual inspection and statistical techniques. Part of the actual data, which was not used for model building, was applied to the model for its validation. A comparison of the regression model and neural network model shows that the regression and the neural network models perform equally well for the predictions made with both sets of validation data.