FlyByWireCpp/FlyByWire.hpp
Jérôme cfc7631d16 First iteration of PID controller.
Added default Filter1 constructor.
Fixed Vote test.
2019-11-30 18:24:03 +01:00

391 lines
15 KiB
C++

#ifndef DEF_FlyByWire
#define DEF_FlyByWire
#include <cmath>
#include <complex>
#define HUGE_VALUE_REAL Real(1e20)
// =============================================================================
// General fly-by-wire helper functions
// =============================================================================
#ifndef MIN
#define MIN(a,b) (((a) < (b)) ? (a) : (b))
#endif
#ifndef MAX
#define MAX(a,b) (((a) > (b)) ? (a) : (b))
#endif
namespace FlyByWire
{
typedef double Real;
typedef std::complex<Real> Complex;
/// Computes the heading error using the shortest route from hdg to tgt, both expressed in radians.
Real Hdg_err(Real tgt, Real hdg);
/// Computes the heading error using the shortest route from hdg to tgt, both expressed in degrees.
Real Hdg_err_deg(Real tgt, Real hdg);
/// Computes the median of the triplet (a, b, c). Useful to reject an invalid measurement from a triplet of sensors, or to mix 3 control laws in a continuous fashion (C0 continuity).
/// Internally, it is implemented as a sorting network. The middle value is the median of the triplet.
Real Vote(Real a, Real b, Real c);
/// Copies the input as long as it's not within +/- a from the origin. The output is then 0.
Real Deadzone(Real x, Real a);
/// Saturation function. Ensures that a <= x <= b.
Real Sat1(Real x, Real a, Real b);
/// Limits the rate of the signal x(t) : [dx/dt](t) to the interval [dy_dt_min, dy_dt_max].
/// Use like a recursive filter.
///
/// \variable x_n : Input value at step n
/// \variable y_n_1 : Output value at step n-1
/// \variable dy_dt_min : Min derivative
/// \variable dy_dt_max : Max derivative
/// \variable dt : Time step
/// \return y[n] rate-limited.
Real Ratelim(Real x_n, Real y_n_1, Real dy_dt_min, Real dy_dt_max, Real dt);
/// Rate limiter implemented as a self-contained recursive filter.
class RateLimiter
{
public:
/// Creates the RateLimiter object with the specified limits, and initial value. By default, y0 = 0.
RateLimiter(Real dy_dt_min, Real dy_dt_max, Real dt, Real y_ = Real(0.));
/// Applies the rate limiter filter to the input x[n] (parameter x_n) and returns the result y[n].
Real Filter(Real x_n);
public:
Real dy_dt_min; //!< Min rate.
Real dy_dt_max; //!< Max rate.
Real dt; //!< Time step.
Real y_n_1; //!< Previous filter value.
};
/// Heaviside function. 1 when x >= 0, 0 otherwise.
Real Heaviside(Real x);
// =============================================================================
// IRR Filters from the continuous Laplace transfer function.
// =============================================================================
/// 1st order integrator implemented using the trapezoïdal method.
/// G(s) = 1/s
/// The integrator can also be limited using the lower_bound and upper_bound parameters.
/// By default, the integrator is not bounded.
///
/// The class is to be used like so :
///
/// flt = Integrator1(Ts)
///
/// while control/filtering loop:
///
/// y_n = flt.Filter(x_n)
///
class Integrator1
{
public:
/// Creates an integrator object. By default, y0 = 0, and the limits are +/- HUGE_VALUE_REAL.
Integrator1(Real Ts_ = Real(1.), Real y_ = Real(0.), Real lower_bound_ = -HUGE_VALUE_REAL, Real upper_bound_ = HUGE_VALUE_REAL);
/// Sets the state of the filter so that it can be initialized at any value.
/// By default, the states are initialized to 0.
void SetState(Real y_ = Real(0.), Real x_ = Real(0.));
/// Computes the next output value of the filter y[n] from the next input value x[n].
Real Filter(Real x_n);
public:
Real Ts;//!< Time step.
Real lower_bound;//!< Lower bound.
Real upper_bound;//!< Upper bound.
Real x_n_1;//!< x[n-1].
Real y_n_1;//!< y[n-1].
};
/// First order discretetized filter. The coefficients are automatically computed from the continuous-time transfer function :
///
/// Y b1 s + b0
/// G(s) = --- = ---------
/// X a1 s + a0
///
/// The Laplace transfer function is first discretized using the Tustin method (sampling period Ts).
///
/// Y -2 b1 + b0 Ts + (2 b1 + b0 Ts) z
/// G(z) = --- = --------------------------------
/// X -2 a1 + a0 Ts + (2 a1 + a0 Ts) z
///
/// The coefficients of the recurrence equation are then computed :
///
/// 2 b1 + b0 Ts
/// C(x[n]) = ------------
/// 2 a1 + a0 Ts
///
///
/// -2 b1 + b0 Ts
/// C(x[n-1]) = ------------
/// 2 a1 + a0 Ts
///
///
/// 2 a1 - a0 Ts
/// C(y[n-1]) = ------------
/// 2 a1 + a0 Ts
///
/// The class is to be used like so :
///
/// flt = filter1(b1, b0, a1, a0, Ts)
///
///
/// while control/filtering loop:
///
/// y_n = flt.Filter(x_n)
///
///
/// The necessary previous values y[n-1] and x[n-1] are automatically stored within the object.
class Filter1
{
public:
/// Creates pass-through Filter1 object.
Filter1();
/// Creates a Filter1 object from the continuous TF's coefficients.
Filter1(Real b1, Real b0, Real a1, Real a0, Real Ts, Real y0 = Real(0.));
/// Computes the filter coefficients from the continuous TF's coefficients.
void ComputeCoeffsFromContinuousTF(Real b1, Real b0, Real a1, Real a0, Real Ts);
/// Sets the state of the filter so that it can be initialized at any value.
/// By default, the states are initialized to 0.
void SetState(Real y_ = Real(0.), Real x_ = Real(0.));
/// Computes the next output value of the filter y[n] from the next input value x[n].
Real Filter(Real x_n);
/// First order discretized filter that computes the derivative of the input signal.
/// tau controls how fast the filter approximates the derivative of the input.
/// A good default value is 0.01.
///
/// s
/// G(s) = ---------
/// tau*s + 1
///
static Filter1 FilteredDerivative(Real Ts, Real tau = 0.01);
public:
Real cx_n; //!< C(x[n])
Real cx_n_1; //!< C(x[n-1])
Real cy_n_1; //!< C(y[n-1])
Real x_n_1; //!< x[n-1]
Real y_n_1; //!< y[n-1]
};
/// Second order discretetized filter. The coefficients are automatically computed from the continuous-time transfer function :
///
/// Y b2 s^2 + b1 s + b0
/// G(s) = --- = ------------------
/// X a2 s^2 + a1 s + a0
///
/// The Laplace transfer function is first discretized using the Tustin method (sampling period Ts).
///
/// Y 4 b2 - 2 b1 Ts + b0 Ts^2 + (-8 b2 + 2 b0 Ts^2) z + (4 b2 + 2 b1 Ts + b0 Ts^2) z^2
/// G(z) = --- = ---------------------------------------------------------------------------------
/// X 4 a2 - 2 a1 Ts + a0 Ts^2 + (-8 a2 + 2 a0 Ts^2) z + (4 a2 + 2 a1 Ts + a0 Ts^2) z^2
///
/// The coefficients of the recurrence equation are then computed :
///
/// 4 b2 + 2 b1 Ts + b0 Ts^2
/// C(x[n]) = ------------------------
/// 4 a2 + 2 a1 Ts + a0 Ts^2
///
///
/// -8 b2 + 2 b0 Ts^2
/// C(x[n-1]) = ------------------------
/// 4 a2 + 2 a1 Ts + a0 Ts^2
///
///
/// 4 b2 - 2 b1 Ts + b0 Ts^2
/// C(x[n-2]) = ------------------------
/// 4 a2 + 2 a1 Ts + a0 Ts^2
///
///
/// -8 a2 + 2 a0 Ts^2
/// C(y[n-1]) = ------------------------
/// 4 a2 + 2 a1 Ts + a0 Ts^2
///
///
/// 4 a2 - 2 a1 Ts + a0 Ts^2
/// C(y[n-2]) = ------------------------
/// 4 a2 + 2 a1 Ts + a0 Ts^2
///
/// The class is to be used like so :
///
/// flt = filter2(b2, b1, b0, a2, a1, a0, Ts)
///
/// while control/filtering loop:
///
/// y_n = flt.Filter(x_n)
///
/// The necessary previous values y[n-1], y[n-2], x[n-1], and x[n-2] are automatically stored within the object.
class Filter2
{
public:
/// Creates a Filter2 object from the continuous TF's coefficients.
Filter2(Real b2, Real b1, Real b0, Real a2, Real a1, Real a0, Real Ts, Real y0 = Real(0.));
/// Computes the filter coefficients from the continuous TF's coefficients.
void ComputeCoeffsFromContinuousTF(Real b2, Real b1, Real b0, Real a2, Real a1, Real a0, Real Ts);
/// Sets the state of the filter so that it can be initialized at any value.
/// By default, the states are initialized to 0.
void SetState(Real y_ = Real(0.), Real x_ = Real(0.));
/// Computes the next output value of the filter y[n] from the next input value x[n].
Real Filter(Real x_n);
/// Creates the transfer function from its poles, zeros, and a gain.
///
/// a (s - z1) (s - z2)
/// G(s) = -------------------
/// (s - p1) (s - p2)
///
static Filter2 FromPolesAndZeros(Real a, Complex p1, Complex p2, Complex z1, Complex z2, Real Ts);
/// Creates the transfer function from its poles, and the static gain a.
/// The poles can be real, or complex conjugate.
///
/// a p1 p2
/// G(s) = -----------------
/// (s - p1) (s - p2)
///
static Filter2 FromPoles(Real a, Complex p1, Complex p2, Real Ts);
/// Second order discretetized low-pass biquadratic filter.
///
/// The filter is parametrized using the cutoff frequency omega in rad/s, and the quality factor Q.
/// For no resonnance peak, use Q < 1/2
///
/// w^2
/// G(s) = -----------------
/// s^2 + w/Q*s + w^2
///
static Filter2 Lowpass(Real omega, Real Q, Real Ts);
/// Second order discretetized high-pass biquadratic filter.
/// The filter is parametrized using the cutoff frequency omega in rad/s, and the quality factor Q.
/// For no resonnance peak, use Q < 1/2
///
/// s^2
/// G(s) = -----------------
/// s^2 + w/Q*s + w^2
///
static Filter2 Highpass(Real omega, Real Q, Real Ts);
/// Second order discretetized band-pass biquadratic filter.
/// The filter is parametrized using the cutoff frequency omega in rad/s, and the quality factor Q.
/// For no resonnance peak, use Q < 1/2
///
/// w/Q*s
/// G(s) = -----------------
/// s^2 + w/Q*s + w^2
///
static Filter2 Bandpass(Real omega, Real Q, Real Ts);
/// Second order discretetized band-stop biquadratic filter.
/// The filter is parametrized using the cutoff frequency omega in rad/s, and the quality factor Q.
/// For no resonnance peak, use Q < 1/2
///
/// s^2 + w^2
/// G(s) = -----------------
/// s^2 + w/Q*s + w^2
///
static Filter2 Bandstop(Real omega, Real Q, Real Ts);
public:
Real cx_n; //!< C(x[n])
Real cx_n_1; //!< C(x[n-1])
Real cx_n_2; //!< C(x[n-2])
Real cy_n_1; //!< C(y[n-1])
Real cy_n_2; //!< C(y[n-2])
Real x_n_1; //!< x[n-1]
Real x_n_2; //!< y[n-2]
Real y_n_1; //!< x[n-1]
Real y_n_2; //!< y[n-2]
};
// =============================================================================
// PID controller
// =============================================================================
/// PID controller in its parallel form :
/// u = Kp*e + Ki*int(e*dt, t) + Kd*de/dt
///
/// The derivative of the error is computed using a 1st order discretized filtered derivative filter.
/// The integral of the error is computed using the trapezoïdal rule.
/// The integrator can be limited using integrator_lower_bound and integrator_upper_bound.
/// The integration can be frozen using the boolean freeze_integration (true to freeze it).
/// The output can be limited using output_lower_bound and output_upper_bound.
///
/// An automatic anti-windup functionnality is provided : the user can choose a value a past which
/// the integrator is automatically frozen in order to prevent wind-ups.
/// For instance, if margin = 0.05 and the limits are [-1 ; 1], the integrator will be frozen
/// in the intervals [-1 ; -0.95] and [0.95 ; 1]
class PID
{
public:
/// Initializes the PID controller with the right coefficients.
/// \param Kp_ : Proportionnal gain
/// \param Ki_ : Integral gain
/// \param Kd_ : Derivative gain
/// \param Ts_ : Time step
/// \param output_lower_bound_ : Lower bound of the controller output
/// \param output_upper_bound_ : Upper bound of the controller output
/// \param tau_filtered_derivative_ : Time constant of the filtered derivative
/// \param integrator_lower_bound_ : Lower bound of the integrator
/// \param integrator_upper_bound_ : Upper bound of the integrator
/// \param auto_anti_windup_ : State boolean to activate or deactivate the auto-anti-windup feature
/// \param auto_anti_windup_margin_ : Absolute margin for the auto-anti-windup feature
PID(Real Kp_, Real Ki_, Real Kd_, Real Ts_, Real output_lower_bound_ = -HUGE_VALUE_REAL, Real output_upper_bound_ = HUGE_VALUE_REAL, Real tau_filtered_derivative = Real(0.01), Real integrator_lower_bound_ = -HUGE_VALUE_REAL, Real integrator_upper_bound_ = HUGE_VALUE_REAL, bool auto_anti_windup_ = true, Real auto_anti_windup_margin_ = Real(1e-3));
/// Computes the next controller output value of the PID u[n] from the next error value e[n].
Real Filter(Real e_n);
/// Sets the state of the integration freezing feature. True to freeze the integration.
void SetFreezeIntegration(bool freeze_integration_);
/// Sets the time constant of the filtered derivative.
void SetFilteredDerivativeTimeConstant(Real tau_filtered_derivative);
/// Sets the integrator limits.
void SetIntegratorLimits(Real integrator_lower_bound, Real integrator_upper_bound);
/// Sets the current integrator value.
void SetIntegratorValue(Real integrator_value);
/// Computes whether the integrator should be frozen or not.
bool ComputeFreezeIntegrator();
public:
Real Kp; //!< Proportionnal gain
Real Ki; //!< Integral gain
Real Kd; //!< Derivative gain
Real Ts; //!< Time step
Real output_lower_bound; //!< Lower bound of the controller output
Real output_upper_bound; //!< Upper bound of the controller output
Real auto_anti_windup_margin; //!< Margin for the auto-anti-windup feature
Real u_n_1; //!< Previous controller output
bool auto_anti_windup; //!< State boolean to activate the auto-anti-windup feature
bool freeze_integration; //!< State boolean to freeze the integration of the error
Integrator1 integrator; //!< Discrete 1st-order integrator
Filter1 differentiator; //!< Discrete 1st-order filtered derivative
};
}
#endif