Design, Fabrication and Machine Learning-Based Performance Prediction of a Self-Propelled Lawn Mower

There is a growing demand for low-cost and locally produced lawn-mowing machines that can cut vegetation to a satisfactory quality and efficiency. The study presents the production and performance prediction of a self-propelled petrol engine-driven lawn mower. The machine consists of a locally fabricated welded structural frame, top mounting plate, power transmission train, a three-blade rotary cutting assembly, a four-wheel running gear and an operator handle. The principal dimensions and operating requirements were used for fabrication. The performance of the mower was determined by machine learning techniques using a simulated dataset ($n = 180$ runs) generated from the governing engineering relationships (forward speed, cutting height, blade speed, grass height, grass density, grass moisture and terrain slope as inputs) with injected random variation added to simulate real measurement noise. Linear Regression, K-Nearest Neighbours, Decision Tree, Random Forest and Extra Trees regressors were trained on an $80:20$ split and compared using $R^2$, MAE, RMSE and MAPE for effective field capacity ($\text{EFC}$). Linear Regression gave the best test-set fit ($R^2 = 0.97$, $\text{RMSE} = 0.005\text{ ha h}^{-1}$), with forward speed dominating feature importance ($\approx 93\%$), consistent with the near-linear capacity relationship built into the simulation; K-Nearest Neighbours performed worst ($R^2 < 0$), illustrating the sensitivity of distance-based methods to unscaled inputs. The exercise validates the end-to-end modelling pipeline and is intended to be re-run on genuine field data in a follow-up study.