Denis Tudor - Swiss Federal Institute of Technology of Lausanne (EPFL)
Fedele Tegarelli - Swiss Federal Institute of Technology of Lausanne (EPFL)
Simone Rametti - Swiss Federal Institute of Technology of Lausanne (EPFL)
Mario Paolone - Swiss Federal Institute of Technology of Lausanne (EPFL)
SpaceX organized the first Hyperloop Pod Competition in 2015 with goals of supporting the development of functional hyperloop pods and encouraging innovation of this new transportation platform. The idea behind the competition is to challenge student teams to design and realize functioning high-speed pods. SpaceX held the first two competitions in January 2017 and August 2017. The EPFLoop team participated in the 2018 competition for the first time.
The selection process includes four main stages: (1) the preliminary design, (2) the full pod design, (3) the design week, and (4) the final competition among the top three teams. SpaceX engineers carefully evaluate each of these stages using security and innovation key performance indicators. For stage 4, engineers judge the selected teams solely on one criterion: maximum speed with successful deceleration.
History of the Team
Several decades ago, Professor Marcel Jufer of the École Polytechnique Fédérale de Lausanne (EPFL) proposed the Swissmetro, a futuristic Swiss national transportation system, which was the first complete hyperloop concept. Even today, this legacy motivates many of the students at EPFL. Inspired by this experience, and relying on the technical background of Swissmetro, the EPFL decided to form a team of students to participate in the 2018 SpaceX Hyperloop Pod Competition. Denis Tudor assembled the team to support the advance of hyperloop technology and help shape the fifth mode of transportation. This team is called EPFLoop and consists of inspiring professors and passionate students with diverse backgrounds and engineering skills, all committed to put their hearts and souls into this futuristic competition.
We equipped the EPFLoop pod with a fault-tolerant avionic, which is tailored to control the specific hardware available on the pod. As a function of the operating principles of the various pod sub-components (energy storage, traction, pressure vessels, braking system, and more), we defined a suitable software hierarchy. Hierarchical-dependent software routines run simultaneously with time-deterministic constraints, each one characterized by a specific task. In this respect, each avionics’ control must be robust with respect to the loss of any sensor or actuator. Every actuator has been embedded with suitable feedback, so the system can have a unique actuation. Timeouts and heartbeat check methods have been implemented to detect sensor faults. With respect to this last requirement, the avionic of the pod integrates a navigation system that can estimate the position of the pod along the hyperloop tube. It exploits data coming from three different sensors fed into a state estimation algorithm based on the linear weighted least square method. The state estimation algorithm is suitably integrated with a bad data identification process relying on the largest normalized residual test.
With respect to these software requirements, NI offers the best reliable and computational-performing embedded control units. Using NI products was one of our key advantages with this project. The FPGA and processor integrated in the cRIO-9042 controller delivered an extremely efficient and robust computation platform. We included an NI-9215 C Series voltage input module, two NI-9401 C Series digital modules, and an NI-9853 C Series CAN interface module to command all the other subsystems of the pod. We used LabVIEW Real-Time software, with all the necessary tools it provides, to program the CompactRIO. Moreover, we used a ground station, composed of a standard laptop running LabVIEW, to command the pod remotely. The wireless communication between the pod and the ground station relies on the TCP/IP protocol with the help of shared variables in LabVIEW. This solution offered a data throughput below 1.5 Mb/s with a maximum time latency of 10 ms.
Figure 1 shows the structure of the EPFLoop pod composed by the following main subsystems:
- Chassis providing the mechanical housing for each pod’s subsystem
- Propulsion system and drivetrain composed of a voltage source inverter and a three-phase permanent magnet synchronous motor
- Front and rear stability systems, which mechanically constrain the pod laterally and transversely to the rail
- Main and lateral pressure vessels (top of the figure)
- High-voltage batteries (embedded in the two lateral pressure vessels)
- Braking systems
Figure 2 shows the architecture of the pod control system. The cRIO-9042 manages more than 20 different devices. We designed the system to rely on CAN communication protocol, at least to communicate with the most critical devices and sensors, in view of its efficiency to support distributed real-time controls with a high level of security.
We designed the pod to be used in two different driving modes (see Figure 3). In autopilot mode, as a function of target mission parameters, the pod navigation system completes the mission autonomously. In manual mode, the ground station operator can manually sense and control all the pod’s subsystems. We have used the manual driving mode extensively to quantify the drivetrain performance and pod’s kinematics. We have developed the autopilot driving mode to steer the pod at high speeds. We needed this last driving mode to take care of all the necessary tasks to be completed to finish a mission and, in parallel, to take care of all the possible failures that could happen before, during, or after a run.
The most critical part of the navigation system implemented in the cRIO-9042 was the algorithm responsible for correctly estimating the position of the pod in the vacuum tube. The algorithm uses three different inputs: two contrast sensors and the pod speed (provided by the encoder of the motor). The algorithm estimates the position of the pod using a state estimation process relying on the weighted least square method suitably integrated with a normalized residual test to identify, and reject, bad measurements. Figure 4 shows a test done to verify the capability of the algorithm to correctly estimate the position of the pod in presence of bad measurements. Disturbances were intentionally injected on the two contrast sensors, while the only uncorrupted measurement was the speed provided by the pod motor encoder. As it can be seen, the navigation system can identify and reject the two bad measurements and rely solely on the pod speed measurement. The linear state estimator has a time deterministic delay (less than 1 ms) and delivers a reliable output with a relative error of less than 3 percent.
The EPFLoop pod finished in third place out of 18 finalists selected among 5,000 student teams that participated to the 2018 SpaceX Hyperloop Pod Competition. During the testing week, the solutions developed by the EPFLoop team received extremely positive feedback from SpaceX engineers. The team used LabVIEW and the CompactRIO hardware platform to rapidly develop a sophisticated and reliable software to validate the pod performance and consistently steer it during the various runs done during the testing week.
During the testing week, thanks to the flexible architecture of the pod, the team passed all the tests required by SpaceX and finished in first place of that phase of the competition.
Swiss Federal Institute of Technology of Lausanne (EPFL)