UAS NMIMS 2017 Journal Paper for 15 th AUVSI Student UAS Competition. Mukesh Patel School of Technology Management and Engineering

Transcription

1 UAS NMIMS 2017 Journal Paper for 15 th AUVSI Student UAS Competition Mukesh Patel School of Technology Management and Engineering Figure 1 SkyKing 17 ABSTRACT This journal paper documents the design and development process of Team UAS NMIMS s Unmanned Aerial System for the AUVSI Student Unmanned Aerial System Competition The team has focused on improving and surpassing the capabilities of the system presented last year. This year, Team UAS NMIMS presents SkyKing 17, an Aerial Mapping and Aerial Surveillance Fixed Wing craft SkyKing 17, is an electric, twin propeller UAV with dual motor configurations capable of long ctr5range, stable, silent and fast cruise flight. This paper gives a detailed description of the team s objectives in building the system. The paper also describes the UAS flight structure and concludes with a description of the testing that has been performed to ensure that each piece of the system is safe and reliable. 1

2 INDEX 1. SYSTEMS ENGINEERING APPROACH... 3 MISSION REQUIREMENT ANALYSIS... 3 DESIGN RATIONALE... 3 PROGRAMMATIC RISKS AND MITIGATION METHODS UAS DESIGN... 5 AIRFRAME... 5 Overall Design... 6 Wing... 6 Propulsion System... 6 AUTOPILOT... 5 Air System... 6 Ground Control System... 7 IMAGING SYSTEM... 7 OBJECT DETECTION, CLARIFICATION AND LOCALIZATION... 9 Target Detection System... 9 Target Identification COMMUNICATION LINK INTEROPERABILITY AIR DELIVERY CYBER SECURITY TEST AND EVALUATION MISSION OPERATION PAYLOAD SYSTEM PERFORMANCE INDIVIDUAL COMPONENT AND SOFTWARE TESTING SAFETY CONSIDERATIONS MISSION AND OPERATION RISKS & MITIGATIONS SAFETY RISKS & MITIGATIONS CONCLUSION

3 1. System Engineering Approach 1.1 Mission Requirement Analysis The AUVSI Student UAS competition 2017 stimulates a real life mission which includes various tasks. UAS NMIMS s experience from SUAS competition 2016 has helped to identify a set of objectives which, if achieved, would allow for the completion of most of the tasks. The team has prioritized tasks using the following factors: Value: The estimated grade in competition for completing the task. Complexity: The knowledge of the system on which the team has been working. Development: The amount of time required to construct and test the system for the task. The tasks are divided in three groups: Will Accomplish: Tasks that will be successfully completed by the team Will Attempt: Tasks for which an effort will be made to reach the threshold Will Not Attempt: Tasks that will not be attempted by the team The team decided on: Mission task Autonomous Flight Obstacle Avoidance Object Detection, Classification, Localization Air delivery Mission Sub-Task Autonomous Flight Waypoints navigation Waypoint accuracy GCS Display Static Obstacle Dynamic Search Area & Off-Axis Classification Localization / Geotagging Autonomy Delivery accuracy Payload Table 1 Expected Tasks Will Accomplish Will Accomplish Will Attempt Will Accomplish Will Not Attempt Will Not Attempt Will Accomplish Will Attempt Will Accomplish Will Attempt Will Attempt Will Accomplish 1.2 Design Rationale A thorough examination of 2016 s systems combined with 2017 s mission has yielded 4 major development points which were prioritized to achieve the more tasks: 3

4 Increase the payload capacity. Upgrade the imaging system to new and powerful one. Enhance the durability of aircraft when flying in rough winds Develop and test new system software. Focusing these points several changes were made in comparison to last year: Durability and Reliability: Extended wing span to increase the payload with better equipment and wind tolerance. New camera: A more efficient camera installed which suits our requirement. The team has decided to go with the Sony A5100 paired with the 16-50mm lens. Safety: As per feedback from last year more emphasis was placed on safety by maintaining a checklist-through all the design and development stages. Improve Interoperability System: The team failed to achieve the threshold value last year. The interoperability system was thus modified to meet the requirements of 2017 mission. FPV: The team has decided to use the First person view for off axis task. System / Component Camera Autopilot Board GPS OnBoard Computer (OBC) DATA LINK Frequency Options D3300 D5300 Sony nex5r Sony A5100 Sony a600 Pixhawk pixhawk 2 APM ublox M8 UBlox 7m Raspberry Pi Odroid XU4 Our Selection Sony A5100 Support gphoto2 Light weight Cheaper High resolution Pixhawk More stable Support available Support Interop server ublox M8 More GPS Count More Accurate Refreshing rate is far better Odroid XU4 Easy integration More powerful 5.8 GHz 5.8 GHz 2.4GHz No interference 433 MHz More Bandwidth 900 MHz Supporting equipment Available in Country Table 2 Summary of system rationale 4

5 1.3 Programmatic Risks and Mitigation During the beginning of the development process, we made a programmatic risk assessment. Being our second year in the competition, we based our assumptions on the experience from the last year. The table below represents the major risks we found, the analysis of their likelihood, their impact and corresponding mitigation strategies. Risk Likelihood Impact Mitigation Strategy Design and/or manufacturing delays Damage to the airframe during testing Moderate Low High High Insufficient Crew training High Medium Loss of Source Codes Low Medium Table 3 Mitigation Reuse 2016 system as a backup vehicle for team practice 2016 system is kept as a backup for training Extensive flight sessions planned to ensure flight team capability All the source code has been backed up on the Cloud 5

6 2. Unmanned Aerial System (UAS) Design 2.1 Airframe Overall Design In preparation for SUAS competition 2017, the team has chosen the previous year model with extended wings. This selection was made keeping in mind the competition tasks parameters, and requirements like Propulsion System, Aircraft size, Payload size, Aircraft Material, Budget Limitations. Figure 2 Airframe specification It is a twin propelled electric system, made of EPO foam and reinforced carbon fibre sheets providing flight stability and sufficiently large payload to carry all on board systems. The aircraft is associated with hand launch system and uses a controlled belly landing system. Sr.No. Characteristic Specification 1 Wingspan 2200mm 2 Fuselage Length 1220mm 3 Material EPO 4 Autopilot Pixhawk 5 Take Off Hand Launch 6 Landing Auto & Stabilized 7 Camera Sony A Sensors Airspeed Sensor, Battery Current Sensor 9 GPS with Compass Neo M8N Gps 10 Telemetry(Data Link) RFD Lipo Battery Lipo- 4s, 10000mah (15c) 12 Brushless Motor 2 x 1000 Kv 13 ESC 2 x 80A 14 Remote Control Taranis X9D Plus (Mode2)(2.4Ghz) 15 FPV with Mobius 5.8Ghz Rx/Tx Table 4 Aircraft Specifications 6

7 2.1.2 Wing The SkyKing 17 wing features a hollow structure made of EPO foam sheet 1.8 mm thick in a layered structure of Gr/Ep and balsa wood. Both wings are attached to the fuselage through two 3 mm thick carbon fiber tubes and aligned with a torque screw. The current wing presents an improvement in comparison to the 2016 s Syclla wing. The new wing, with extensions of 200mm in each provides better flight stability with reliable take-off and landing. Figure 3 Wing specification Propulsion System SkyKing 17 is a twin electrical motor configuration with one motor on each side. The motors are rated at a maximum current draw of 50A each. The speed controllers used are rated at 80A. The motors are rated for a thrust of 2.8 kg with a 10 X 6 propeller, thereby providing a total thrust of 5.2 kg with a twin motor configuration. High thrust allows for smooth and easy hand take-off. This configuration provides the system with better thrust and stability in highly windy conditions. Twin motor also consume less power as compared to single motor in extreme conditions such as high winds. This configuration also help us quick mapping as it increases the speed of the aircraft. 2.1 Autopilot Air System The team chooses to continue with Pixhawk autopilot system, as it is well tested, reliable, and cost efficient at the same time. The system is easily acquirable so that allows quick replacements to be made in a case of system malfunction or damage. The Px4 provides superior CPU performance and higher memory capacity. And since it is an open source, it allows to customize functionality. Furthermore, the Pixhawk Px4 supports various redundant sensors which increases the safety and reliability of the vehicle. The autopilot provides the following key features: 7

8 Abundant connectivity options for additional peripherals (UART, I2C, CAN) Integrated backup system for in-flight recovery and manual override with dedicated processor and stand-alone power supply (fixed-wing use) Backup system integrates mixing, providing consistent autopilot and manual override mixing modes (fixed wing use) Redundant power supply inputs and automatic failover External safety switch Multi-coloured LED main visual indicator High-power, multi-tone piezo audio indicator microsd card for high-rates logging over extended periods of time Ground Control System The ground control system s main objective is to control, supervise and communicate with the flight platform at any stage of the flight. The GCS includes two processing units independent of each other Navigation Unit and Data processing. Both units are placed side by side and use independent protocols to communicate with the UAV. Each system has been designed to allow a user-friendly and highly efficient operation. Navigation Unit The navigation unit is used to communicate with aircraft during flight. It contain combination of telemetry devices and computers. The RFD900+, which operates in the 900 MHz band, has been selected for this link due to its demonstrated performance at 60 km without requiring high gain antennas. The telemetry link is used to communicate with the FC and relays telemetry data to the interoperability server. The team has chosen to use the Mission Planner Software as it can quickly create complete surveys in seconds. Data Processing Unit The data processing unit is used to acquire captured images through 5.8 GHz data link. It contains 6 different computers which performs data processing. The targets identified are sent to the interoperability server via a local network as well as manually through secondary storage device. 2.3 Imaging System Primary Camera and Imagery System The primary camera employed during the primary task was one of the most important instruments mounted on the drone. A modular with high image quality camera and high shutter speed was a must. Various camera options, of both point-and-shoot and DSLR categories, were considered. It was important that the camera was lean enough to fit in the payload area of the vehicle. 8

9 Camera Placement. The camera is mounted closest to the center of gravity of the vehicle with its lens facing the camera slit in the floor of the fuselage. The camera slit is covered with a servo controlled flap to protect the lens when not in use. Camera Details Figure 4 Camera Slit Assessment of previous year s camera Sony NEX-5R performance had the following problems: Quality was not good enough for decoding the images clicked. Low resolution images. Small field of view with 18mm focal length (27 mm 1.5 equivalent). Slow Frame rate (< 1FPS) during performance. The following table gives the details of Sony A5100 Camera and its competitors: PARAMETERS Sony A5100 SONY NEX 5R NIKON D3300 IMAGE SENSOR 24.3MP 16.1MP 24.2MP LIGHT SENSITIVITY (ISO) EXPOSURE COMPENSATION ISO ISO ISO , ±3 EV range, in 1/3 EV steps ±3 EV range, in 1/3 EV steps FOCAL LENGTH 16-50mm mm 35mm STILL IMAGE RESOLUTION L: 6000 x 3376 (20 M), M: 4240 x 2400 (10 M), S: 3008 x 1688 (5.1 M) 2448 x 1376, 2448 x 1624, 3568 x 2000, 3568 x 2368, 4912 x 2760, 4912 x 3264 BATTERY 400 shots 320 shots 700 shots DIMENTIONS (W) x 62.8 (H) x 35.7 (D) mm ±5 EV range, in 1/3 EV steps 2992 x 2000, 4496 x 3000, 6000 x x 2.3 in 4.9 x 3.9 in Price 35,000 26,000 53,000 Weight (with battery) 283 g 218 g 672 g Table 5 Camera Comparison 9

10 Considering the problems with D3300 and NEX-5R we decided to use Sony A5100 camera. In Comparison with Sony NEX-5R and Nikon D3300, The Sony Alpha 5100 has High true resolution of 24.3 MP. Great colour depth of 23.8 bits. Large field of view with 16mm focal length (24mm 1.5 equivalent). In total, 174 cross type focus points. Small in size and thin. Very cheap. First Person View Camera (Secondary Camera) The team has installed a First Person View camera system in the drone this time. We preferred this system as last year the camera could not capture sufficient amount of images required for the search area task. So now, FPV provides first person view with real time video on the GCS through which crew members will be able to identify some images and characters during the flight time. The system also records the live video of the flight which can be viewed afterwards, helping in identifying the images clicked by the primary camera during the post processing session. 2.4 Object clarification, detection and localization Target Detection System In order to provide proper image processing and communication, Team has decided to use ODROID XU4 which runs the Ubuntu operating system. The system block diagram is shown below: Block Diagram 1 - Imagery System 10

11 2.4.2 Target Identification Due to the constrained OBC processing power and limited wireless communication capabilities, the team decided to use a Sony A5100 along with an Odroid Xu4 and a bullet M5. The image captured by the camera will be sent to the GCS with help of an Odroid and a bullet M5. Images captured will have a unique characteristic that a large portion in images will be uniform and uncluttered. Our algorithm will be able to detect the required Region of Interest (ROI). The algorithm is used on images where the regions of interest are small, in comparison to the background. Smoothing Of Image Pre-processing of the image is required to remove noise and texture from the image. The small variations caused by the texture of the background such as grass can add to the difficulty of finding contrast and hence, smoothing the image will make it easier to identify objects. This includes the following two steps:- 1. Gaussian Smoothening 2. Mean Shift Filtering Region of interest Potential target identification on the image will be done by using MSER (Maximally Stable Extraction Region) running on a sub computer of the data processing unit at the ground control system. This algorithm works well for finding text regions because of the consistent color and contrast of stable intensity profile. There can be multiple objects of interest scattered throughout the image, which will not have a fixed shape and size which means that certain other regions will also be detected having same intensity profiles but will be removed based on geometric properties followed by Stroke Width Variation. The camera s angle of view, the speed and the altitude at which our UAV will fly during mission, ensured that there will be enough overlap between two consecutive captures to cover the entire ground area. These MSER crop images will now be sent to various sub computers to look for some errors, just in case some region is missed. 11

12 Normalization Furthermore, the final image generated contains information about the location of visually salient objects. The steps for extracting the target using this information are done by Blob Detection Figure 5 Detection Neural network We are using Neural Network to classify the number and Alphabets from the image. Once the Image characteristics is detected it will be directed to the judge s computer by the image processing team. 2.5 Communication link Last year s communication system consisted of 2.4 GHz RC transmitter and 915 MHz telemetry data. In addition to that an Imagery link has been introduced this year which works on 5.8 GHz band. The imagery link does not interferes with the telemetry link. The datalink consists of two Bullet-M5s from Ubiquiti Networks. We chose the 5.8GHz frequency range since 2.4GHz would conflict with the manual RC control. 5.8 GHz was also selected to reduce the image acquisition time. 2.4 GHz Frequency band: Used by our manual RC transmitter for controlling the aircraft. 5.8 Ghz Frequency band: The data link used to communicate with Onboard Computer ODROID MHz Frequency band: Used for communicating with pixhawk. 2.6 Interoperability Pre-defined python scripts for the interoperability task perform login, retrieval of obstacle data, and posting of UAS telemetry data. It consists of an HTTP GET and POST requests for same. The telemetry data, obtained from Mission Planner is pushed to the interoperability server. The processes are executed in parallel using multiple threads to maintain the transmission and reception rate at the mean threshold value. 2.7 Air Delivery In order to accomplish the Air Drop successfully, the team has decided to perform the task with its secondary plane, Scylla 2k16. Last year the team achieved 100% accuracy in the task with the payload being 0.2m away from the Bull s eye. 12

13 Figure 6 Air Drop in Action The payload mechanism consists of a separate payload bay attached below the nose of the aircraft. This structure with an open bottom was fabricated using thick, strong polystyrene blocks reinforced with carbon fiber sheets. The bungee cord attached to the payload bay to hold the payload in its place is manually controlled by a servo. During the air drop task the safety pilot will manually trigger the servo which will release the bungee and hence causing the payload to drop. 13

14 The reason for using secondary plane for this task was to reduce time delay. While performing the said task last year, the team lost 10 minutes from the total flight time in calibrating and replacing the batteries. 2.8 CYBER SECURITY UAVs are highly exposed technical system due to the unique configuration such as open state of the sensors at all times, wireless network etc. Cyber security aspect plays a major role in terms of system safety. Authentication: It is used by a server when the server needs to know exactly who is accessing their system. Authorization: It is a process by which a server determines if the client has the permission to use a resource or access a file. Authorization is usually coupled with authentication so that the server has some understanding of who the client, the access, actually is. Encryption: Encryption involves the process of transforming data so that it is unreadable by anyone who does not have a decryption key. Cyber security attack feasibilities that can be categorized into three groups: 1) Hardware Attack In a scenario where the attacker has access to the UAV s autopilot components directly, we secure the following modules of our system. a) Ground Station In order to protect our system, we have used an access control mechanism (MAC).The team will try to isolate the system s network to minimize threats. Only authenticated team members system will be allowed to connect to the main system. MAC spoofing will be prevented by using a password authentication that will be known only to the team members. b) Internal UAV Figure 7 Hook mechanism All the internal components are tamper resistant. Locks, key, stiffing tie are used to add tamper protection. In case of tampering of internal component, it can be easily detected visually. 14

15 2) Wireless Attack In this scenario, the attacker carries out attacks on the system via the wireless communication channels. Radio Link Protection Encryption and authentication during the communications (Control and Payload) shall be used. For using encryption, the keys will be shared in advance between the ground station and the airborne system. 3) Sensor Spoofing In this scenario, the attacker passes bogus data through the on-board sensors like GPS. a) Waypoint Attack It is an attack that changes the list of waypoint in the autopilot s flight plan. b) GPS Spoofing It is an attack that corrupts the GPS coordinates. Spoofing detection based on lock loss has two disadvantages: (i) (ii) Strong attackers can achieve a seamless satellite-lock takeover, Lock loss can occur due to natural causes (e. g. signal loss in a tunnel). GPS Spoofing Countermeasure We propose a countermeasure against GPS spoofing attacks that does not rely on the signal analysis or on the lock loss of signal. These GPS receivers can be deployed in a static, known formation. The basic idea of the countermeasure is that if the GPS receivers can exchange their individual GPS locations, they can check if their calculated locations preserve their physical formation (within certain error bounds). In the case that the calculated GPS locations do not match the known information, an attack must be detected and a warning message should be displayed. Even if only two GPS receivers are used, this countermeasure can detect any attacker that is only using a single antenna. In case of a single-antenna attack both GPS receivers would report the same location (with small time offsets). General Measures The following are the general measures that our team takes while conducting the pre-flight assessments: 1) Hiding the SSID of router : This prevents the attacker from finding out the exact name of the hosted network which will be used for mission tasks. 2) Physically remove the default password: Many devices have a default login password and the URL for the administrator dashboard written over it. The team removes all such information as a safety measure. 3) Change of Default password and Login address: The team changes all the default virtual address as per team and all the details are known only to team members. 4) Dual Level Authentication: The Team prefers in maintaining a two layer authentication system for logging in. 15

16 3. Test and Evaluation Results Team UAS NMIMS and development members of the team has conducted over 350 minutes of 40 plus test flights, 28 of which had complete autonomous take-off and landing. All this was conducted within the span of three months with numerous subsystem tests in team s test center The following sections interpret the tests performed and their results as an evidence for mission performance. 3.1 Mission Operations Autonomous Flight Autonomous flight has been one of the most challenging tasks according to the flight records because the possibility of failure is high, and failure of this segment of the mission has caused the most severe of the damages. A total of 27 autonomous take-offs and landings have been successfully attempted, of which the accuracy of waypoint capturing during the last 4 flights was recorded to have a slight deflection of 5 meters. In order to achieve this, we have decided to make some changes in the Mission Planner s parameters for accuracy such as waypoint_radius,tkoff_dly,thr_max,trigger_distance etc. The team is successfully able to display all the details as per the requirement. Air delivery We are implementing a manual air delivery system with a drop switch. During the flight tests, we have evaluated the airdrop subsystem in many scenarios and flight conditions. We managed to achieve approximately 60-70% accuracy in dropping the payload using our system. We realize that our system needs to improve in terms of preventing the payload from breaking. We have currently achieved only 10% success rate in this field. The damage incurred to the bottle varies with the maximum recorded to be as high as 60%. Interoperability The interoperability rate has been tested different time with consistent data rates of 4-5 Hz. There were initial problems such as cloning of the judge server, but later it was resolved by team. Data dropouts have occurred rarely. Our interop system is capable of meeting the mission requirement. Mission task Mission Sub-Task Expected Performance Autonomous Flight 100% Accuracy Autonomous Flight Waypoint navigation 100% Accuracy Waypoint accuracy 100% Accuracy GCS Display 100% Accuracy Search Area & Off-Axis. 60% achievable Object Detection, Classification 70% achievable Classification, Localization Localization / Geotagging 80% achievable Autonomy Manual Air delivery Delivery accuracy 70% accurate Interoperability Will meet requirement Table 6 Mission Analysis 16

17 3.2 Payload System Performance Our payload system performance has proved to be successful in all test flights, as we have been able to fly with consistent accuracy and communication with ground station. We have almost doubled our payload capacity as compared to last year competition. The center of gravity is maintained such that our aircraft in flight is always stable. Scylla 2k16 also got a recognition for the same in 14 th AUVSI Competition. This year also, the team made endless efforts and is successful in reaching their objective of stable flight during the mission performance. 3.3 Individual Component and Software Testing The team has flown SkyKing 17 for almost 350 minutes. The individual component testing was performed before every flight along with calibration. The set of test values were recorded and cross checked to specific component data sheet. Our software testing system consists of six distributed systems which enables compartmentalization of software bugs, so that most of the system and other ground station computers are unaffected at the time of failure. 17

18 4. Safety Considerations Safety is an essential factor to be considered in order to deploy a fully functional autonomous flight. Continuous risk assessments were carried out throughout the process of developing and flying in order to minimize possible harm of personnel and property. The team prioritized the safety aspect of mission performance and after understanding the kind of issues that may arise and these issues were mitigated with each subsequent flight. 4.1 Mission and Operation Risk Mitigation The team made sure to test all the systems at extremes to ensure that the aircraft could safely take-off, perform the mission and land in any competing condition. All team members were trained to use a fire extinguisher in case of a LiPo fire or fire caused by overheated components. Team members have their separate roles to ensure safety with each department having its own checklist. Pre-launch: Launch: Post Launch: Check the CG of the system Check the battery for damage and ensure they are charged Check the functioning of flight control servos Check manual control of the system Safe hand launch Check the damage Check Data link Throttle is off Proper Image capturing Remove the battery Spotter checks the line of flight Throttle check Autopilot malfunction protocol Ensure GPS check Stabilizers test Check air speed Ensure waypoints and boundaries Ensuring clear air space Turn off transmitter Ensure healthy autopilot link Table 7 Safety 18

19 4.2 Safety and Risk Mitigation Risk Likelihood Impact Mitigation Strategy Power failure Moderate High Different BEC to control the power. Three different BEC for RFD, CAMERA and SERVO Pixhawk power failure Low High An extra device using Zener diode for power detention Pre-flight failure Low Moderate Establish safe activity protocol, Go through the checklist before take-off Accidental change of mode Moderate High Double Safety switch for switching between manual and autonomous flight to avoid an accidental change in mode Hand launch Safety Start of motor can lead to physical injury low High A safety switch for transmission of mode. A 10 second delay in starting the motor for safety precaution in auto mode. Servo failure low Moderate Pre check will let us know if there is any failure in servo Loss of telemetry link Moderate Low Incorrect connection Low Moderate Use of a failsafe with a return to launch till control is regained. Receiver placement to minimize interference Periodic maintenance to identify frayed wires, chafed insulation, or loose connectors Table 8 Risk Mitigation 19

20 5. Conclusion This journal paper demonstrates UAS NMIMS team s flight readiness for the 2017 AUVSI Student UAS Competition. This year involved an extensive research to improve aerodynamics, software, imagery system and interoperability. The enhancements were made based on research results as well as our past experience with previous configuration. The entire system was then evaluated during system tests to ensure the UAS is reliable and that meets all desired mission objectives. TEAM MEMBERS 1. Sharad Gupta 2. Shruti Golchha 3. Sachin Sharma 4. Abhishek Khambhatha 5. Mrunmayee Inamke 6. Akash Agrawal 7. Ojas Shirekar 20