/// @file AC_PID.cpp /// @brief Generic PID algorithm #include #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::filt_T_hz(float hz) { _filt_T_hz.set(fabsf(hz)); } // filt_E_hz - set error filter hz void AC_PID::filt_E_hz(float hz) { _filt_E_hz.set(fabsf(hz)); } // filt_D_hz - set derivative filter hz void AC_PID::filt_D_hz(float hz) { _filt_D_hz.set(fabsf(hz)); } // slew_limit - set slew limit void AC_PID::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; _target = target; _error = _target - measurement; _derivative = 0.0f; _target_derivative = 0.0f; #if AP_FILTER_ENABLED if (_target_notch != nullptr) { _target_notch->reset(); _target = _target_notch->apply(_target); } if (_error_notch != nullptr) { _error_notch->reset(); _error = _error_notch->apply(_error); } #endif } else { float error_last = _error; float target_last = _target; float error = _target - measurement; #if AP_FILTER_ENABLED // apply notch filters before FTLD/FLTE to avoid shot noise if (_target_notch != nullptr) { target = _target_notch->apply(target); } if (_error_notch != nullptr) { error = _error_notch->apply(error); } #endif _target += get_filt_T_alpha(dt) * (target - _target); _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(); } /// Overload the function call operator to permit easy initialisation void AC_PID::operator()(float p_val, float i_val, float d_val, float ff_val, float imax_val, float input_filt_T_hz, float input_filt_E_hz, float input_filt_D_hz, float dff_val) { _kp.set(p_val); _ki.set(i_val); _kd.set(d_val); _kff.set(ff_val); _kimax.set(fabsf(imax_val)); _filt_T_hz.set(input_filt_T_hz); _filt_E_hz.set(input_filt_E_hz); _filt_D_hz.set(input_filt_D_hz); _kdff.set(dff_val); } // 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)); } }