ardupilot/libraries/AP_Soaring/SpeedToFly.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

58 lines
2.2 KiB
C++
Raw Permalink Normal View History

/* SpeedToFly class by Samuel Tabor, 2021.
Calculates the optimal speed to fly given drag polar, expected climb rate in next thermal and
horizontal and vertical air movement between thermals.
*/
#include "SpeedToFly.h"
void SpeedToFly::update(float Wx, float Wz, float Wexp, float CLmin, float CLmax)
{
// The solution to the speed-to-fly problem does not have a closed form solution. A Newton
// method with some additional checks will converge to an acceptable level within 3-4 iterations.
// However, to keep the computation constant per function call we just do a single iteration using
// the previous approximation as a starting point.
// This gives good accuracy as the inputs don't change rapidly. It would also be possible to store
// the inputs and converge the solution over 3-4 function calls, but this real-time iteration
// approach gives better accuracy in tests as well as simpler code.
Wz *= -1.0f; // Sink defined positive.
float sqrtfk = sqrtf(_polarParams.K);
float minSink = (sqrtfk/sqrtf(CLmax)*(_polarParams.CD0 + _polarParams.B*CLmax*CLmax))/CLmax;
if (!is_positive(minSink+Wz+Wexp)) {
// Special case. If lift is greater than min sink speed, fly at min sink
// speed.
_CL_estimate = CLmax;
return;
}
float CD0 = _polarParams.CD0;
float B = _polarParams.B;
float Wxp = Wx/sqrtfk;
float WZ = (Wz + Wexp)/sqrtfk;
// Guess starting location.
float CL = _CL_estimate>0 ? _CL_estimate : 0.5f*(CLmax+CLmin);
float t0 = powf(CL,1.5f);
float t1 = CD0 + B*CL*CL + t0*WZ;
float t2 = 1.5f*sqrtf(CL)*WZ + 2.0f*B*CL;
float Jd = (1.5f*sqrtf(CL)*Wxp + 1.0f)/t1 - (t2*(CL + t0*Wxp))/(t1*t1);
float Jdd = 2.0f*t2*t2*(CL + t0*Wxp)/powf(t1,3) - (2.0f*t2*(1.5f*sqrtf(CL)*Wxp + 1.0f))/(t1*t1) - ((2.0f*B + 0.75f*WZ/sqrtf(CL))*(CL + t0*Wxp))/(t1*t1) + 0.75f*Wxp/(sqrtf(CL)*t1);
// Check we're heading to a maximum, not a minimum!!
if (is_positive(Jdd)) {
// Alternate mode, go uphill.
CL = CL + 0.1 * (Jd>0.0f ? 1.0f : -1.0f);
} else {
// Newton should work.
CL = CL - Jd/Jdd;
}
_CL_estimate = CL;
_CL_estimate = constrain_float(_CL_estimate, CLmin, CLmax);
}