Sensor Fusion with NARX Neural Network to Predict the Mass Flow in a Sugarcane Harvester

Area of Science:

• Agricultural Engineering
• Sensor Fusion
• Precision Agriculture

Background:

• Accurate real-time mass flow measurement is crucial for sugarcane harvester automation and crop monitoring.
• Integrating data from multiple sensors enhances prediction reliability and accuracy compared to single-sensor systems.

Purpose of the Study:

• To evaluate the impact of fusing data from different sensors on sugarcane mass flow prediction accuracy.
• To compare the performance of neural network models against traditional regression techniques for mass flow estimation.

Main Methods:

• Instrumenting a sugarcane harvester with multiple sensors.
• Developing and comparing Nonlinear Autoregressive Networks with Exogenous Input (NARX) and Multiple Linear Regression (MLR) models.
• Utilizing sensor data fusion for self-calibration and mass flow prediction.

Main Results:

• The NARX model demonstrated superior performance over MLR in predicting mass flow.
• Multi-sensor data fusion with NARX achieved an RMSE of 0.3 kg s⁻¹ and a MAPE of 0.7%.
• Sensor fusion significantly improved prediction accuracy compared to single-sensor approaches.

Conclusions:

• The fusion of multiple sensor signals in sugarcane harvesters substantially enhances mass flow prediction accuracy.
• This multi-sensor approach shows potential as a replacement for conventional yield monitors, offering high-density, accurate data.
• The developed system provides a robust solution for real-time crop monitoring and harvester automation.