Development of an ANN Integrated Streamlit Web App for Estimating Soil Compaction Beneath Agricultural Tyre
摘要
In this study, multiple linear regression (MLR) and artificial neural network (ANN) models were developed to estimate soil compaction with the operation of a powered wheel fitted with bias-ply tyre in different soil conditions i.e., soft, medium and hard soils. Input parameters of the both the models were drawbar pull, inflation pressure, wheel load, soil condition, and cone index before (CIbefore) the tyre whereas, cone index after (CIafter) operating the tyre was the output parameter. The tyre was tested at varying normal loads (1000 and 1400 kg) and inflation pressures (82.74 to 137.90 kPa). The maximum soil compaction was observed at normal load of 1400 kg and 137.90 kPa inflation pressure in all soil conditions. Soil compaction was increased with the increase in pull, inflation pressure and wheel load in all soil conditions. Under a normal load of 1000 kg, average soil compaction increased by 228.76%, 56.67%, and 10.37% in soft, medium, and hard soils, respectively as compared to 300.30%, 71.99%, and 21.59% for the same soil conditions under the normal load of 1400 kg. The correlation heatmap revealed that among all other parameters, pull had the strongest correlation with CIafter with 13.6–28 powered wheel. The ANN and MLR models were trained and tested on a dataset of 303 data points for the 13.6–28 bias-ply tyre, with a split ratio of 80:20, respectively. The ANN model having 5-8-6-1 architecture outperformed the MLR model with R2 values 0.954, 0.936 0.928 for training, testing and validation, respectively. Sensitivity analysis of developed ANN model revealed that the drawbar pull was given highest importance among all the input parameters. The developed ANN model was also deployed in Streamlit web application interface in which user can enter the input parameters and application will show the predicted value of CIafter and amount of soil compaction incurred due to it.