Machine Learning Model for Predicting Transient Jerk Behavior in Knuckle Boom Loaders

The knuckle boom loader machine experiences a significant issue where rapidly retracting the joystick to the neutral (0) position causes the machine's legs to lift momentarily. This unintended lifting occurs when the joystick is moved back quickly, while gradual movements do not trigger this effect. Addressing this problem is crucial for enhancing both safety and operational efficiency. The primary objective of this project is to develop a machine learning model to predict jerk based on joystick movements. This prediction will aid in creating a model predictive controller (MPC) that suggests optimal joystick positions, thereby reducing unintended lifting. To simulate the knuckle boom loader's response to joystick inputs, a high-fidelity Simulink model developed using Simscape Multibody was utilized. Data were collected through a Design of Experiments (DOE), logging key parameters such as head side pressure, jerk, jerk rate, and lift levels across various joystick positions. The input features for the machine learning model included time, joystick position, and delta position (the change in position). The developed machine learning model effectively predicts jerk based on joystick input patterns, achieving an R² score of 0.98. This model leverages detailed data captured from the Simulink model to provide valuable insights into the operational behavior of the knuckle boom loader. This predictive model will inform the MPC, guiding optimal joystick movements to prevent unintended lifting and enhance overall safety. In future developments, the predicted jerk from the machine learning model will serve as input for the MPC, suggesting the most effective joystick movements to minimize both lift and jerk. This advancement is expected to significantly improve safety and efficiency in the operation of knuckle boom loaders.