meggert
4aca7a91e3
All checks were successful
EIVE/eive-obsw/pipeline/pr-v3.0.0-dev This commit looks good
209 lines
9.8 KiB
C++
209 lines
9.8 KiB
C++
#include "PtgCtrl.h"
|
|
|
|
#include <fsfw/globalfunctions/constants.h>
|
|
#include <fsfw/globalfunctions/math/MatrixOperations.h>
|
|
#include <fsfw/globalfunctions/math/QuaternionOperations.h>
|
|
#include <fsfw/globalfunctions/math/VectorOperations.h>
|
|
#include <fsfw/globalfunctions/sign.h>
|
|
|
|
PtgCtrl::PtgCtrl(AcsParameters *acsParameters_) { acsParameters = acsParameters_; }
|
|
|
|
PtgCtrl::~PtgCtrl() {}
|
|
|
|
void PtgCtrl::ptgLaw(AcsParameters::PointingLawParameters *pointingLawParameters,
|
|
const double *errorQuat, const double *deltaRate, const double *rwPseudoInv,
|
|
double *torqueRws) {
|
|
//------------------------------------------------------------------------------------------------
|
|
// Compute gain matrix K and P matrix
|
|
//------------------------------------------------------------------------------------------------
|
|
double om = pointingLawParameters->om;
|
|
double zeta = pointingLawParameters->zeta;
|
|
double qErrorMin = pointingLawParameters->qiMin;
|
|
double omMax = pointingLawParameters->omMax;
|
|
|
|
double qError[3] = {errorQuat[0], errorQuat[1], errorQuat[2]};
|
|
|
|
double cInt = 2 * om * zeta;
|
|
double kInt = 2 * pow(om, 2);
|
|
|
|
double qErrorLaw[3] = {0, 0, 0};
|
|
|
|
for (int i = 0; i < 3; i++) {
|
|
if (std::abs(qError[i]) < qErrorMin) {
|
|
qErrorLaw[i] = qErrorMin;
|
|
} else {
|
|
qErrorLaw[i] = std::abs(qError[i]);
|
|
}
|
|
}
|
|
double qErrorLawNorm = VectorOperations<double>::norm(qErrorLaw, 3);
|
|
|
|
double gain1 = cInt * omMax / qErrorLawNorm;
|
|
double gainVector[3] = {0, 0, 0};
|
|
VectorOperations<double>::mulScalar(qErrorLaw, gain1, gainVector, 3);
|
|
|
|
double gainMatrixDiagonal[3][3] = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
|
|
double gainMatrix[3][3] = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
|
|
gainMatrixDiagonal[0][0] = gainVector[0];
|
|
gainMatrixDiagonal[1][1] = gainVector[1];
|
|
gainMatrixDiagonal[2][2] = gainVector[2];
|
|
MatrixOperations<double>::multiply(*gainMatrixDiagonal,
|
|
*(acsParameters->inertiaEIVE.inertiaMatrixDeployed),
|
|
*gainMatrix, 3, 3, 3);
|
|
|
|
// Inverse of gainMatrix
|
|
double gainMatrixInverse[3][3] = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
|
|
gainMatrixInverse[0][0] = 1 / gainMatrix[0][0];
|
|
gainMatrixInverse[1][1] = 1 / gainMatrix[1][1];
|
|
gainMatrixInverse[2][2] = 1 / gainMatrix[2][2];
|
|
|
|
double pMatrix[3][3] = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
|
|
MatrixOperations<double>::multiply(
|
|
*gainMatrixInverse, *(acsParameters->inertiaEIVE.inertiaMatrixDeployed), *pMatrix, 3, 3, 3);
|
|
MatrixOperations<double>::multiplyScalar(*pMatrix, kInt, *pMatrix, 3, 3);
|
|
|
|
//------------------------------------------------------------------------------------------------
|
|
// Torque Calculations for the reaction wheels
|
|
//------------------------------------------------------------------------------------------------
|
|
|
|
double pError[3] = {0, 0, 0};
|
|
MatrixOperations<double>::multiply(*pMatrix, qError, pError, 3, 3, 1);
|
|
double pErrorSign[3] = {0, 0, 0};
|
|
|
|
for (int i = 0; i < 3; i++) {
|
|
if (std::abs(pError[i]) > 1) {
|
|
pErrorSign[i] = sign(pError[i]);
|
|
} else {
|
|
pErrorSign[i] = pError[i];
|
|
}
|
|
}
|
|
// torque for quaternion error
|
|
double torqueQuat[3] = {0, 0, 0};
|
|
MatrixOperations<double>::multiply(*gainMatrix, pErrorSign, torqueQuat, 3, 3, 1);
|
|
VectorOperations<double>::mulScalar(torqueQuat, -1, torqueQuat, 3);
|
|
|
|
// torque for rate error
|
|
double torqueRate[3] = {0, 0, 0};
|
|
MatrixOperations<double>::multiply(*(acsParameters->inertiaEIVE.inertiaMatrixDeployed), deltaRate,
|
|
torqueRate, 3, 3, 1);
|
|
VectorOperations<double>::mulScalar(torqueRate, cInt, torqueRate, 3);
|
|
VectorOperations<double>::mulScalar(torqueRate, -1, torqueRate, 3);
|
|
|
|
// final commanded Torque for every reaction wheel
|
|
double torque[3] = {0, 0, 0};
|
|
VectorOperations<double>::add(torqueRate, torqueQuat, torque, 3);
|
|
MatrixOperations<double>::multiply(rwPseudoInv, torque, torqueRws, 4, 3, 1);
|
|
VectorOperations<double>::mulScalar(torqueRws, -1, torqueRws, 4);
|
|
}
|
|
|
|
void PtgCtrl::ptgNullspace(AcsParameters::PointingLawParameters *pointingLawParameters,
|
|
const int32_t speedRw0, const int32_t speedRw1, const int32_t speedRw2,
|
|
const int32_t speedRw3, double *rwTrqNs) {
|
|
// concentrate RW speeds as vector and convert to double
|
|
double speedRws[4] = {static_cast<double>(speedRw0), static_cast<double>(speedRw1),
|
|
static_cast<double>(speedRw2), static_cast<double>(speedRw3)};
|
|
|
|
// calculate RPM offset utilizing the nullspace
|
|
double rpmOffset[4] = {0, 0, 0, 0};
|
|
double rpmOffsetSpeed = pointingLawParameters->nullspaceSpeed / 10 * RPM_TO_RAD_PER_SEC;
|
|
VectorOperations<double>::mulScalar(acsParameters->rwMatrices.nullspaceVector, rpmOffsetSpeed,
|
|
rpmOffset, 4);
|
|
|
|
// calculate resulting angular momentum
|
|
double rwAngMomentum[4] = {0, 0, 0, 0}, diffRwSpeed[4] = {0, 0, 0, 0};
|
|
VectorOperations<double>::subtract(speedRws, rpmOffset, diffRwSpeed, 4);
|
|
VectorOperations<double>::mulScalar(diffRwSpeed, 1e-1, diffRwSpeed, 4);
|
|
VectorOperations<double>::mulScalar(diffRwSpeed, acsParameters->rwHandlingParameters.inertiaWheel,
|
|
rwAngMomentum, 4);
|
|
|
|
// calculate resulting torque
|
|
double nullspaceMatrix[4][4] = {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}, {0, 0, 0}};
|
|
MatrixOperations<double>::multiply(acsParameters->rwMatrices.nullspaceVector,
|
|
acsParameters->rwMatrices.nullspaceVector, *nullspaceMatrix, 4,
|
|
1, 4);
|
|
MatrixOperations<double>::multiply(*nullspaceMatrix, rwAngMomentum, rwTrqNs, 4, 4, 1);
|
|
VectorOperations<double>::mulScalar(rwTrqNs, -1 * pointingLawParameters->gainNullspace, rwTrqNs,
|
|
4);
|
|
}
|
|
|
|
void PtgCtrl::ptgDesaturation(AcsParameters::PointingLawParameters *pointingLawParameters,
|
|
const double *magFieldB, const bool magFieldBValid,
|
|
const double *satRate, const int32_t speedRw0, const int32_t speedRw1,
|
|
const int32_t speedRw2, const int32_t speedRw3, double *mgtDpDes) {
|
|
if (not magFieldBValid or not pointingLawParameters->desatOn) {
|
|
return;
|
|
}
|
|
|
|
// concentrate RW speeds as vector and convert to double
|
|
double speedRws[4] = {static_cast<double>(speedRw0), static_cast<double>(speedRw1),
|
|
static_cast<double>(speedRw2), static_cast<double>(speedRw3)};
|
|
|
|
// convert magFieldB from uT to T
|
|
double magFieldBT[3] = {0, 0, 0};
|
|
VectorOperations<double>::mulScalar(magFieldB, 1e-6, magFieldBT, 3);
|
|
|
|
// calculate angular momentum of the satellite
|
|
double angMomentumSat[3] = {0, 0, 0};
|
|
MatrixOperations<double>::multiply(*(acsParameters->inertiaEIVE.inertiaMatrixDeployed), satRate,
|
|
angMomentumSat, 3, 3, 1);
|
|
|
|
// calculate angular momentum of the reaction wheels with respect to the nullspace RW speed
|
|
// relocate RW speed zero to nullspace RW speed
|
|
double refSpeedRws[4] = {0, 0, 0, 0};
|
|
VectorOperations<double>::mulScalar(acsParameters->rwMatrices.nullspaceVector,
|
|
pointingLawParameters->nullspaceSpeed, refSpeedRws, 4);
|
|
VectorOperations<double>::subtract(speedRws, refSpeedRws, speedRws, 4);
|
|
// convert speed from 10 RPM to 1 RPM
|
|
VectorOperations<double>::mulScalar(speedRws, 1e-1, speedRws, 4);
|
|
// convert to rad/s
|
|
VectorOperations<double>::mulScalar(speedRws, RPM_TO_RAD_PER_SEC, speedRws, 4);
|
|
// calculate angular momentum of each RW
|
|
double angMomentumRwU[4] = {0, 0, 0, 0};
|
|
VectorOperations<double>::mulScalar(speedRws, acsParameters->rwHandlingParameters.inertiaWheel,
|
|
angMomentumRwU, 4);
|
|
// convert RW angular momentum to body RF
|
|
double angMomentumRw[3] = {0, 0, 0};
|
|
MatrixOperations<double>::multiply(*(acsParameters->rwMatrices.alignmentMatrix), angMomentumRwU,
|
|
angMomentumRw, 3, 4, 1);
|
|
|
|
// calculate total angular momentum
|
|
double angMomentumTotal[3] = {0, 0, 0};
|
|
VectorOperations<double>::add(angMomentumSat, angMomentumRw, angMomentumTotal, 3);
|
|
|
|
// calculating momentum error
|
|
double deltaAngMomentum[3] = {0, 0, 0};
|
|
VectorOperations<double>::subtract(angMomentumTotal, pointingLawParameters->desatMomentumRef,
|
|
deltaAngMomentum, 3);
|
|
|
|
// resulting magnetic dipole command
|
|
double crossAngMomentumMagField[3] = {0, 0, 0};
|
|
VectorOperations<double>::cross(deltaAngMomentum, magFieldBT, crossAngMomentumMagField);
|
|
double factor =
|
|
pointingLawParameters->deSatGainFactor / VectorOperations<double>::norm(magFieldBT, 3);
|
|
VectorOperations<double>::mulScalar(crossAngMomentumMagField, factor, mgtDpDes, 3);
|
|
}
|
|
|
|
void PtgCtrl::rwAntistiction(ACS::SensorValues *sensorValues, int32_t *rwCmdSpeeds) {
|
|
bool rwAvailable[4] = {
|
|
(sensorValues->rw1Set.state.value && sensorValues->rw1Set.state.isValid()),
|
|
(sensorValues->rw2Set.state.value && sensorValues->rw2Set.state.isValid()),
|
|
(sensorValues->rw3Set.state.value && sensorValues->rw3Set.state.isValid()),
|
|
(sensorValues->rw4Set.state.value && sensorValues->rw4Set.state.isValid())};
|
|
int32_t currRwSpeed[4] = {
|
|
sensorValues->rw1Set.currSpeed.value, sensorValues->rw2Set.currSpeed.value,
|
|
sensorValues->rw3Set.currSpeed.value, sensorValues->rw4Set.currSpeed.value};
|
|
for (uint8_t i = 0; i < 4; i++) {
|
|
if (rwAvailable[i]) {
|
|
if (rwCmdSpeeds[i] != 0) {
|
|
if (rwCmdSpeeds[i] > -acsParameters->rwHandlingParameters.stictionSpeed &&
|
|
rwCmdSpeeds[i] < acsParameters->rwHandlingParameters.stictionSpeed) {
|
|
if (rwCmdSpeeds[i] > currRwSpeed[i]) {
|
|
rwCmdSpeeds[i] = acsParameters->rwHandlingParameters.stictionSpeed;
|
|
} else if (rwCmdSpeeds[i] < currRwSpeed[i]) {
|
|
rwCmdSpeeds[i] = -acsParameters->rwHandlingParameters.stictionSpeed;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|