vicente2830/fault-tolerant-control

Project for fault tolerant control

Flight Simulator for Fault-Tolerant-Control Algorithms

Installation

git clone [email protected]:fdcl-ftc/fault-tolerant-control.git
cd fault-tolerant-control
pip install -e .

In Python 3.8>, use the module as follows.

import ftc

Usage

Register a Controller Class

The controller must be defined as a class in somewhere in a directory in ftc.controllers.

For example, to define a sliding mode controller, one must create a directory ftc.controllers.mysmc, and a module file smc.py in that directory. The directory may contain several .py files for readability. In any cases, the main controller class, let's say MYSMC, must be registered in ftc.controllers.__init__.py with an identifyable name (id) and a full entry point (entry_point) to that class as follows.

register(
    id="My-SMC-v1",
    entry_point="ftc.controllers.mysmc.smc:MYSMC",
)

As you may notice, this registration process is borrowed from the OpenAI Gym module.

Make a Controller Instance

After the controller is registered, one can make an instance of the controller class by ftc.make method. The env keyword in the ftc.make method would be used to initialize each controller.

controller = ftc.make("My-SMC-v1", env=env)

Controller Class API

Initialize

Controllers often require some information of the plant and/or the environement a priori. For example, linearized state-space system matrices is necessary to design an LQR. For ease of integrated simulation, each controller class must be initialized with the argument env as follows.

class MyController:
    def __init__(self, env):
        ...

get_control Method

Each controller class must have the following get_control method which takes the current time and the main env object of the class fym.BaseEnv and returns the rotor force inputs in (N, 1) dimension ndarray along with some dict-type information.

class MyController:
    def get_control(self, t, env):
        rfs = np.zeros((4, 1))
        info = {}
        return rfs, info

If the controller class also inherits fym.BaseEnv or fym.BaseSystem, the derivative must be defined in the get_control method.

class MyController(fym.BaseEnv):
    def __init__(self):
        super().__init__()
        self.adaptive_part = fym.BaseSystem()

    def get_control(self, t, env):
        W = self.adaptive_part.state

        self.adaptive_part.dot = ...

Env Class API

env.plant Object

The passed env object to the get_control method has a plant object of class fym.BaseEnv, which is the main plant to be controlled. Therefore, you may access the state vectors of the plant.

    def get_control(self, t, env):
        pos, vel, R, omega = env.plant.observe_list()
        rfs = - pos ... 

env.get_ref Method

Also, env object has a get_ref method which must be used in each controller class to get the reference profile. The get_ref method takes several arguments. The first argument is always the current time, and the subsequent arguments are the keys that you want to get.

    def get_control(self, t, env):
        posd, posd_dot, posd_ddot = env.get_ref(t, "posd", "posd_dot", "posd_ddot")

env.get_Lambda Method

For active fault tolerant control, one may require fault information of each rotor. The env object provides get_Lambda method that takes the current time as an input and returns Lambda (N, N) dimensional ndarray which denotes the healthy of each rotor, i.e., 1 - LoE.

    def get_control(self, t, env):
        Lambda = env.get_Lambda(t)

Note that this Lambda is the estimated Lambda, and therefore if one wants a time-delayed estimation, the get_Lambda method must be modified. For example, if plant.get_Lambda returns the true Lambda at the queried time, then the env.get_Lambda method should be

class Env(fym.BaseEnv):
    def get_Lambda(self, t):
        return self.plant.get_Lambda(t - 0.02)

Acknowledgment

• This research was supported by Unmanned Vehicles Core Technology Research and Development Program through the National Research Foundation of Korea (NRF) and Unmanned Vehicle Advanced Research Center (UVARC) funded by the Ministry of Science and ICT, the Republic of Korea (2020M3C1C1A01083162).
• 본 연구는 과학기술정보통신부의 재원으로 한국연구재단, 무인이동체원천기술개발사업단의 지원을 받아 무인이동체 원천기술개발사업을 통해 수행되었음 (2020M3C1C1A01083162).