6.1 Conclusions

One of the vehicle design goals was to enhance the overall fidelity of the neutral buoyancy simulation. To accomplish this goal, Ranger NBV would need to have precise knowledge of its own motion within the neutral buoyancy environment, along with the ability to accurately direct its motion. To help accomplish this goal, Ranger was designed to incorporate high quality sensors. The analog sensor signals were digitized locally to reduce noise. Actuator precision was also greatly increased over previous designs by allowing the thruster system to control each thruster in a closed loop velocity mode. This allowed the static thruster performance to be accurately characterized. The advanced buoyancy compensation system has allowed Ranger NBV to balance itself increasing the accuracy and repeatability of the buoyancy compensation. This system now enables the highest level of precision buoyancy control ever available at the SSL thus increasing the fidelity of the neutral buoyancy simulation.
The vehicle was outfitted with a network of specialized computers to individually handle specific tasks. These individual tasks could then be orchestrated by the main computer into the total vehicle behavior. This approach has allowed each individual processor to be isolated to a specific which has greatly simplified anomaly resolution. This has also made the vehicle wiring harnesses more organized by keeping the bulk of the wires near the sensors or actuators and their co-located CPUs, and only returning power and communications. This method has also helped to provide cleaner sensor data by not requiring analog signals to travel over long wires through a noisy electronic environment.
Operator interfaces have been iteratively designed by repeated testing and refinement. This approach has developed the interfaces to the point where a single trained pilot can easily select and control trajectory modes, choose controllers and settings, and collect data, while flying the vehicle. Even advanced features like bias estimation and automatic buoyancy compensation have been developed and tested to the point that these modes may be selected at the touch of a button, which then start and run seamlessly in the background. Displays show sensor data in real time, and enable rapid anomaly identification and resolution. Advanced 3D displays combine the advanced rendering capabilities of the control station computers with the precision state estimate to form clear graphic representations of desired and estimated vehicle attitude. This display is used for both real time data display and analysis of recorded data. This system has allowed the relative motion of the desired and estimated vehicle attitudes to be clearly understood which has given great insight into controller performance. This level of understanding has allowed the effects of various parts of the control law to be clearly understood, and modified to handle inaccuracies in system model or state estimation. Additionally, the ability to observe the attitude estimate with respect to the desired attitude in real time has allowed each test session to be more productive due to the fact that problems could be identified and solutions implemented while the test was still in progress. This is very helpful when working with a real vehicle as opposed to a simulation. Test time is not as readily available with a real vehicle as it is with a simulation, so it is helpful to make test time as productive as possible.
Models of the vehicle buoyancy, drag, and attitude dynamics were developed, and their parameters refined to match the vehicle characteristics. A method of vehicle state estimation was developed, and the effects of noise and bias were formulated. Filters were developed to reduce the effects of sensor noise, and magnetometer and angular rate sensor bias estimation algorithms were developed and implemented. Several control laws were implemented. Each control law uses an advanced quaternion based PD algorithm to handle errors. The principal advantage of formulating the PD controller in quaternion space is that the moment calculated is in the direction of the most direct rotation from the estimated to the desired attitude. This is true even for very large slew maneuvers. The last two controllers also incorporate the vehicle dynamic model in order to compute the torque necessary to autonomously follow the desired trajectory. In these controllers, the PD portion is used to handle errors due to model or estimator inaccuracies. The final controller implemented includes a law for refining the vehicle dynamic model parameter estimates.
A hardware in the loop vehicle flight simulator was developed that allowed initial testing and refinement of control algorithms before they were used on the actual vehicle. This simulation allowed the effects of noise and other inaccuracies on the controller to be studied and reduced. The simulation was implemented to run within the software modules on the vehicle. With this configuration, the pilot uses the same interfaces to operate the controller as are used with the actual vehicle. Data is displayed and recorded in the same way as during actual vehicle testing. Additionally, the effects of interprocess, and interprocessor communications along with the effect of the various loop rates are included in the simulator performance. The simulation matches the actual vehicle performance very closely.
An automatic buoyancy compensation algorithm was developed and implemented that used the responses of the vehicle attitude controller to direct the motion of the buoyancy compensators to reduce the buoyancy moment. This buoyancy compensation algorithm is actually four closed loop control systems working together. Thrusters operate in a closed loop velocity mode which are commanded by the selected closed loop attitude control law. The average desired moment from the attitude controller is used as the error metric for the closed loop automatic balancing algorithm. The balancing algorithm then specifies the desired position of each RBC to the closed loop BCS controller. The combined system can determine the neutral positions the RBCs with an accuracy of about ±1/8 inch.
Each of the vehicle controllers were tuned, and algorithms were developed and implemented to help reduce the effects of sensor noise, along with model and estimator inaccuracies.
In Section 5.2.1, a gain modification strategy was introduced that "softened" the response of the controller as the angular increased. This gain modification strategy reduced regulation errors by a factor of 2.3.
In Section 5.3.1, a strategy for reducing the effect of noise in the nonlinear compensation controller was implemented. The strategy reduces the gain [lambda] within the noisiest term of the nonlinear controller. This strategy reduces the average angular tracking error by a factor of 3.3.
Testing demonstrates that the estimated nonlinear controller improves PD tracking by about a factor of 2 (4° vs. 2° at 9°/sec)
When the adaptive controller asymptotically reduced the average tracking error to 1.8°, producing a factor of 7.7 reduction of average angular tracking error over a quaternion based PD controller using the same gains (Figure 5-37). The average commanded moment required to track the specified trajectory also decreased by a factor of 1.2.