ardupilot/libraries/AC_PID/AC_PID.cpp

425 lines
13 KiB
C++

/// @file AC_PID.cpp
/// @brief Generic PID algorithm
#include <AP_Math/AP_Math.h>
#include "AC_PID.h"
#define AC_PID_DEFAULT_NOTCH_ATTENUATION 40
const AP_Param::GroupInfo AC_PID::var_info[] = {
// @Param: P
// @DisplayName: PID Proportional Gain
// @Description: P Gain which produces an output value that is proportional to the current error value
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("P", 0, AC_PID, _kp, default_kp),
// @Param: I
// @DisplayName: PID Integral Gain
// @Description: I Gain which produces an output that is proportional to both the magnitude and the duration of the error
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("I", 1, AC_PID, _ki, default_ki),
// @Param: D
// @DisplayName: PID Derivative Gain
// @Description: D Gain which produces an output that is proportional to the rate of change of the error
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("D", 2, AC_PID, _kd, default_kd),
// 3 was for uint16 IMAX
// @Param: FF
// @DisplayName: FF FeedForward Gain
// @Description: FF Gain which produces an output value that is proportional to the demanded input
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("FF", 4, AC_PID, _kff, default_kff),
// @Param: IMAX
// @DisplayName: PID Integral Maximum
// @Description: The maximum/minimum value that the I term can output
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("IMAX", 5, AC_PID, _kimax, default_kimax),
// 6 was for float FILT
// 7 is for float ILMI and FF
// index 8 was for AFF
// @Param: FLTT
// @DisplayName: PID Target filter frequency in Hz
// @Description: Target filter frequency in Hz
// @Units: Hz
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("FLTT", 9, AC_PID, _filt_T_hz, default_filt_T_hz),
// @Param: FLTE
// @DisplayName: PID Error filter frequency in Hz
// @Description: Error filter frequency in Hz
// @Units: Hz
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("FLTE", 10, AC_PID, _filt_E_hz, default_filt_E_hz),
// @Param: FLTD
// @DisplayName: PID Derivative term filter frequency in Hz
// @Description: Derivative filter frequency in Hz
// @Units: Hz
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("FLTD", 11, AC_PID, _filt_D_hz, default_filt_D_hz),
// @Param: SMAX
// @DisplayName: Slew rate limit
// @Description: Sets an upper limit on the slew rate produced by the combined P and D gains. If the amplitude of the control action produced by the rate feedback exceeds this value, then the D+P gain is reduced to respect the limit. This limits the amplitude of high frequency oscillations caused by an excessive gain. The limit should be set to no more than 25% of the actuators maximum slew rate to allow for load effects. Note: The gain will not be reduced to less than 10% of the nominal value. A value of zero will disable this feature.
// @Range: 0 200
// @Increment: 0.5
// @User: Advanced
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("SMAX", 12, AC_PID, _slew_rate_max, default_slew_rate_max),
// @Param: PDMX
// @DisplayName: PD sum maximum
// @Description: The maximum/minimum value that the sum of the P and D term can output
// @User: Advanced
AP_GROUPINFO("PDMX", 13, AC_PID, _kpdmax, 0),
// @Param: D_FF
// @DisplayName: PID Derivative FeedForward Gain
// @Description: FF D Gain which produces an output that is proportional to the rate of change of the target
// @Range: 0 0.02
// @Increment: 0.0001
// @User: Advanced
AP_GROUPINFO_FLAGS_DEFAULT_POINTER("D_FF", 14, AC_PID, _kdff, default_kdff),
#if AP_FILTER_ENABLED
// @Param: NTF
// @DisplayName: PID Target notch filter index
// @Description: PID Target notch filter index
// @Range: 1 8
// @User: Advanced
AP_GROUPINFO("NTF", 15, AC_PID, _notch_T_filter, 0),
// @Param: NEF
// @DisplayName: PID Error notch filter index
// @Description: PID Error notch filter index
// @Range: 1 8
// @User: Advanced
AP_GROUPINFO("NEF", 16, AC_PID, _notch_E_filter, 0),
#endif
AP_GROUPEND
};
// Constructor
AC_PID::AC_PID(float initial_p, float initial_i, float initial_d, float initial_ff, float initial_imax, float initial_filt_T_hz, float initial_filt_E_hz, float initial_filt_D_hz,
float initial_srmax, float initial_srtau, float initial_dff) :
default_kp(initial_p),
default_ki(initial_i),
default_kd(initial_d),
default_kff(initial_ff),
default_kimax(initial_imax),
default_filt_T_hz(initial_filt_T_hz),
default_filt_E_hz(initial_filt_E_hz),
default_filt_D_hz(initial_filt_D_hz),
default_slew_rate_max(initial_srmax),
default_kdff(initial_dff)
{
// load parameter values from eeprom
AP_Param::setup_object_defaults(this, var_info);
// this param is not in the table, so its default is no loaded in the call above
_slew_rate_tau.set(initial_srtau);
// reset input filter to first value received
_flags._reset_filter = true;
memset(&_pid_info, 0, sizeof(_pid_info));
// slew limit scaler allows for plane to use degrees/sec slew
// limit
_slew_limit_scale = 1;
}
// filt_T_hz - set target filter hz
void AC_PID::set_filt_T_hz(float hz)
{
_filt_T_hz.set(fabsf(hz));
}
// filt_E_hz - set error filter hz
void AC_PID::set_filt_E_hz(float hz)
{
_filt_E_hz.set(fabsf(hz));
}
// filt_D_hz - set derivative filter hz
void AC_PID::set_filt_D_hz(float hz)
{
_filt_D_hz.set(fabsf(hz));
}
// slew_limit - set slew limit
void AC_PID::set_slew_limit(float smax)
{
_slew_rate_max.set(fabsf(smax));
}
void AC_PID::set_notch_sample_rate(float sample_rate)
{
#if AP_FILTER_ENABLED
if (_notch_T_filter == 0 && _notch_E_filter == 0) {
return;
}
if (_notch_T_filter != 0) {
if (_target_notch == nullptr) {
_target_notch = NEW_NOTHROW NotchFilterFloat();
}
AP_Filter* filter = AP::filters().get_filter(_notch_T_filter);
if (filter != nullptr && !filter->setup_notch_filter(*_target_notch, sample_rate)) {
delete _target_notch;
_target_notch = nullptr;
_notch_T_filter.set(0);
}
}
if (_notch_E_filter != 0) {
if (_error_notch == nullptr) {
_error_notch = NEW_NOTHROW NotchFilterFloat();
}
AP_Filter* filter = AP::filters().get_filter(_notch_E_filter);
if (filter != nullptr && !filter->setup_notch_filter(*_error_notch, sample_rate)) {
delete _error_notch;
_error_notch = nullptr;
_notch_E_filter.set(0);
}
}
#endif
}
// update_all - set target and measured inputs to PID controller and calculate outputs
// target and error are filtered
// the derivative is then calculated and filtered
// the integral is then updated based on the setting of the limit flag
float AC_PID::update_all(float target, float measurement, float dt, bool limit, float boost)
{
// don't process inf or NaN
if (!isfinite(target) || !isfinite(measurement)) {
return 0.0f;
}
// reset input filter to value received
_pid_info.reset = _flags._reset_filter;
if (_flags._reset_filter) {
_flags._reset_filter = false;
// Reset target filter
_target = target;
#if AP_FILTER_ENABLED
if (_target_notch != nullptr) {
_target_notch->reset();
_target = _target_notch->apply(_target);
}
#endif
// Calculate error and reset error filter
_error = _target - measurement;
#if AP_FILTER_ENABLED
if (_error_notch != nullptr) {
_error_notch->reset();
_error = _error_notch->apply(_error);
}
#endif
// Zero derivatives
_derivative = 0.0f;
_target_derivative = 0.0f;
} else {
// Apply target filters
const float target_last = _target;
#if AP_FILTER_ENABLED
// apply notch filters before FTLD/FLTE to avoid shot noise
if (_target_notch != nullptr) {
target = _target_notch->apply(target);
}
#endif
_target += get_filt_T_alpha(dt) * (target - _target);
// Calculate error and apply error filter
const float error_last = _error;
float error = _target - measurement;
#if AP_FILTER_ENABLED
if (_error_notch != nullptr) {
error = _error_notch->apply(error);
}
#endif
_error += get_filt_E_alpha(dt) * (error - _error);
// calculate and filter derivative
if (is_positive(dt)) {
float derivative = (_error - error_last) / dt;
_derivative += get_filt_D_alpha(dt) * (derivative - _derivative);
_target_derivative = (_target - target_last) / dt;
}
}
// update I term
update_i(dt, limit);
float P_out = (_error * _kp);
float D_out = (_derivative * _kd);
// calculate slew limit modifier for P+D
_pid_info.Dmod = _slew_limiter.modifier((_pid_info.P + _pid_info.D) * _slew_limit_scale, dt);
_pid_info.slew_rate = _slew_limiter.get_slew_rate();
P_out *= _pid_info.Dmod;
D_out *= _pid_info.Dmod;
// boost output if required
P_out *= boost;
D_out *= boost;
_pid_info.PD_limit = false;
// Apply PD sum limit if enabled
if (is_positive(_kpdmax)) {
const float PD_sum_abs = fabsf(P_out + D_out);
if (PD_sum_abs > _kpdmax) {
const float PD_scale = _kpdmax / PD_sum_abs;
P_out *= PD_scale;
D_out *= PD_scale;
_pid_info.PD_limit = true;
}
}
_pid_info.target = _target;
_pid_info.actual = measurement;
_pid_info.error = _error;
_pid_info.P = P_out;
_pid_info.D = D_out;
_pid_info.FF = _target * _kff;
_pid_info.DFF = _target_derivative * _kdff;
return P_out + D_out + _integrator;
}
// update_error - set error input to PID controller and calculate outputs
// target is set to zero and error is set and filtered
// the derivative then is calculated and filtered
// the integral is then updated based on the setting of the limit flag
// Target and Measured must be set manually for logging purposes.
// todo: remove function when it is no longer used.
float AC_PID::update_error(float error, float dt, bool limit)
{
// don't process inf or NaN
if (!isfinite(error)) {
return 0.0f;
}
// Reuse update all code path, zero target and pass negative error as measurement
// Passing as measurement bypasses any target filtering to maintain behaviour
// Negate as update all calculates error as target - measurement
_target = 0.0;
const float output = update_all(0.0, -error, dt, limit);
// Make sure logged target and actual are still 0 to maintain behaviour
_pid_info.target = 0.0;
_pid_info.actual = 0.0;
return output;
}
// update_i - update the integral
// If the limit flag is set the integral is only allowed to shrink
void AC_PID::update_i(float dt, bool limit)
{
if (!is_zero(_ki) && is_positive(dt)) {
// Ensure that integrator can only be reduced if the output is saturated
if (!limit || ((is_positive(_integrator) && is_negative(_error)) || (is_negative(_integrator) && is_positive(_error)))) {
_integrator += ((float)_error * _ki) * dt;
_integrator = constrain_float(_integrator, -_kimax, _kimax);
}
} else {
_integrator = 0.0f;
}
_pid_info.I = _integrator;
_pid_info.limit = limit;
// Set I set flag for logging and clear
_pid_info.I_term_set = _flags._I_set;
_flags._I_set = false;
}
float AC_PID::get_p() const
{
return _pid_info.P;
}
float AC_PID::get_i() const
{
return _integrator;
}
float AC_PID::get_d() const
{
return _pid_info.D;
}
float AC_PID::get_ff() const
{
return _pid_info.FF + _pid_info.DFF;
}
void AC_PID::reset_I()
{
_flags._I_set = true;
_integrator = 0.0;
}
// load original gains from eeprom, used by autotune to restore gains after tuning
void AC_PID::load_gains()
{
_kp.load();
_ki.load();
_kd.load();
_kff.load();
_filt_T_hz.load();
_filt_E_hz.load();
_filt_D_hz.load();
}
// save original gains to eeprom, used by autotune to save gains before tuning
void AC_PID::save_gains()
{
_kp.save();
_ki.save();
_kd.save();
_kff.save();
_filt_T_hz.save();
_filt_E_hz.save();
_filt_D_hz.save();
}
// get_filt_T_alpha - get the target filter alpha
float AC_PID::get_filt_T_alpha(float dt) const
{
return calc_lowpass_alpha_dt(dt, _filt_T_hz);
}
// get_filt_E_alpha - get the error filter alpha
float AC_PID::get_filt_E_alpha(float dt) const
{
return calc_lowpass_alpha_dt(dt, _filt_E_hz);
}
// get_filt_D_alpha - get the derivative filter alpha
float AC_PID::get_filt_D_alpha(float dt) const
{
return calc_lowpass_alpha_dt(dt, _filt_D_hz);
}
void AC_PID::set_integrator(float integrator)
{
_flags._I_set = true;
_integrator = constrain_float(integrator, -_kimax, _kimax);
}
void AC_PID::relax_integrator(float integrator, float dt, float time_constant)
{
integrator = constrain_float(integrator, -_kimax, _kimax);
if (is_positive(dt)) {
_flags._I_set = true;
_integrator = _integrator + (integrator - _integrator) * (dt / (dt + time_constant));
}
}