Excavators are ubiquitous machines used in construction and agriculture globally. Many fluid power machines, including excavators, are operated directly by humans. For this reason, the efficacy of the communication channels between the human and machine have a high impact on system performance. While current excavator controls have been mastered by experts in the field, novel methods of control can improve operator performance and accelerate the learning process for novice operators.
The current state of the art in excavator control uses 2 dual degree of freedom joysticks to control the flow to each of the excavator joints. There is a steep learning curve associated with this interface due to the large cognitive load it places on the operator. A coordinated rate control scheme was developed to alleviate the need for the operator to mentally perform the inverse kinematics of the linkage, and implemented using the standard joysticks to ensure compatibility with current state of the art hardware. In a 20 person experiment, subjects using coordinated rate control consistently removed more soil/time and soil/fuel than subjects using the conventional control.
A novel method of task identification was developed to determine which phase of a trenching cycle the excavator is in at each time step. A supervised Artificial Neural Network with eight inputs, the four joystick velocity inputs and four joint positions, is used to classify the data into one of the three phases of a trenching cycle: dig, unload, and return. The ability to segment data enabled further analysis of the controllers within each phase, and can potentially be used to change the hydraulic priorities real time or to augment the operator input to achieve an optimal command.