1
0
Fork 0

Removed dos line endings.

This commit is contained in:
curt 2002-09-03 13:27:45 +00:00
parent 7cdcfcafb8
commit e9dba2caa1
2 changed files with 204 additions and 204 deletions

View file

@ -1,93 +1,93 @@
//#include <ansi_c.h>
//#include <math.h>
//#include <stdio.h>
//#include <stdlib.h>
#include "uiuc_ice_rates.h"
///////////////////////////////////////////////////////////////////////
// Calculates shed rate depending on current aero loads, eta, temp, and freezing fraction
// Code by Leia Blumenthal
//
// 13 Feb 02 - Created basic program with dummy variables and a constant shed rate (no dependency)
//
// Inputs:
// aero_load - aerodynamic load
// eta
// T - Temperature in Farenheit
// ff - freezing fraction
//
// Output:
// rate - %eta shed/time
//
// Right now this is just a constant shed rate until we learn more...
double shed(double aero_load, double eta, double T, double ff, double time_step)
{
double rate, eta_new;
if (eta <= 0.0)
rate = 0.0;
else
rate = 0.2;
eta_new = eta-rate*eta*time_step;
if (eta_new <= 0.0)
eta_new = 0.0;
return(eta_new);
}
///////////////////////////////////////////////////////////////////////////////////////////////////
// Currently a simple linear approximation based on temperature and eta, but for next version,
// should have so that it calculates sublimation rate depending on current temp,pressure,
// dewpoint, radiation, and eta
//
// Code by Leia Blumenthal
// 12 Feb 02 - Created basic program with linear rate for values when sublimation will occur
// 16 May 02 - Modified so that outputs new eta as opposed to rate
// Inputs:
// T - temperature and must be input in Farenheit
// P - pressure
// Tdew - Dew point Temperature
// rad - radiation
// time_step- increment since last run
//
// Intermediate:
// rate - sublimation rate (% eta change/time)
//
// Output:
// eta_new- eta after sublimation has occurred
//
// This takes a simple approximation that the rate of sublimation will decrease
// linearly with temperature increase.
//
// This code should be run every time step to every couple time steps
//
// If eta is less than zero, than there should be no sublimation
double sublimation(double T, double eta, double time_step)
{
double rate, eta_new;
if (eta <= 0.0) rate = 0;
else{
// According to the Smithsonian Meteorological tables sublimation occurs
// between -40 deg F < T < 32 deg F and between pressures of 0 atm < P < 0.00592 atm
if (T < -40) rate = 0;
else if (T >= -40 && T < 32)
{
// For a simple linear approximation, assume largest value is a rate of .2% per sec
rate = 0.0028 * T + 0.0889;
}
else if (T >= 32) rate = 0;
}
eta_new = eta-rate*eta*time_step;
if (eta_new <= 0.0)
eta_new = 0.0;
return(eta_new);
}
//#include <ansi_c.h>
//#include <math.h>
//#include <stdio.h>
//#include <stdlib.h>
#include "uiuc_ice_rates.h"
///////////////////////////////////////////////////////////////////////
// Calculates shed rate depending on current aero loads, eta, temp, and freezing fraction
// Code by Leia Blumenthal
//
// 13 Feb 02 - Created basic program with dummy variables and a constant shed rate (no dependency)
//
// Inputs:
// aero_load - aerodynamic load
// eta
// T - Temperature in Farenheit
// ff - freezing fraction
//
// Output:
// rate - %eta shed/time
//
// Right now this is just a constant shed rate until we learn more...
double shed(double aero_load, double eta, double T, double ff, double time_step)
{
double rate, eta_new;
if (eta <= 0.0)
rate = 0.0;
else
rate = 0.2;
eta_new = eta-rate*eta*time_step;
if (eta_new <= 0.0)
eta_new = 0.0;
return(eta_new);
}
///////////////////////////////////////////////////////////////////////////////////////////////////
// Currently a simple linear approximation based on temperature and eta, but for next version,
// should have so that it calculates sublimation rate depending on current temp,pressure,
// dewpoint, radiation, and eta
//
// Code by Leia Blumenthal
// 12 Feb 02 - Created basic program with linear rate for values when sublimation will occur
// 16 May 02 - Modified so that outputs new eta as opposed to rate
// Inputs:
// T - temperature and must be input in Farenheit
// P - pressure
// Tdew - Dew point Temperature
// rad - radiation
// time_step- increment since last run
//
// Intermediate:
// rate - sublimation rate (% eta change/time)
//
// Output:
// eta_new- eta after sublimation has occurred
//
// This takes a simple approximation that the rate of sublimation will decrease
// linearly with temperature increase.
//
// This code should be run every time step to every couple time steps
//
// If eta is less than zero, than there should be no sublimation
double sublimation(double T, double eta, double time_step)
{
double rate, eta_new;
if (eta <= 0.0) rate = 0;
else{
// According to the Smithsonian Meteorological tables sublimation occurs
// between -40 deg F < T < 32 deg F and between pressures of 0 atm < P < 0.00592 atm
if (T < -40) rate = 0;
else if (T >= -40 && T < 32)
{
// For a simple linear approximation, assume largest value is a rate of .2% per sec
rate = 0.0028 * T + 0.0889;
}
else if (T >= 32) rate = 0;
}
eta_new = eta-rate*eta*time_step;
if (eta_new <= 0.0)
eta_new = 0.0;
return(eta_new);
}

View file

@ -1,111 +1,111 @@
// SIS Twin Otter Iced aircraft Nonlinear model
// Version 020409
// read readme_020212.doc for information
#include "uiuc_iced_nonlin.h"
void Calc_Iced_Forces()
{
// alpha in deg
double alpha;
double de;
double eta_ref_wing = 0.08; // eta of iced data used for curve fit
double eta_ref_tail = 0.12;
double eta_wing;
//double delta_CL; // CL_clean - CL_iced;
//double delta_CD; // CD_clean - CD_iced;
//double delta_Cm; // CM_clean - CM_iced;
double delta_Cm_a; // (Cm_clean - Cm_iced) as a function of AoA;
double delta_Cm_de; // (Cm_clean - Cm_iced) as a function of de;
double delta_Ch_a;
double delta_Ch_e;
double KCL;
double KCD;
double KCm_alpha;
double KCm_de;
double KCh;
double CL_diff;
alpha = Alpha*RAD_TO_DEG;
de = elevator*RAD_TO_DEG;
// lift fits
if (alpha < 16)
{
delta_CL = (0.088449 + 0.004836*alpha - 0.0005459*alpha*alpha +
4.0859e-5*pow(alpha,3));
}
else
{
delta_CL = (-11.838 + 1.6861*alpha - 0.076707*alpha*alpha +
0.001142*pow(alpha,3));
}
KCL = -delta_CL/eta_ref_wing;
eta_wing = 0.5*(eta_wing_left + eta_wing_right);
delta_CL = eta_wing*KCL;
// drag fit
delta_CD = (-0.0089 + 0.001578*alpha - 0.00046253*pow(alpha,2) +
-4.7511e-5*pow(alpha,3) + 2.3384e-6*pow(alpha,4));
KCD = -delta_CD/eta_ref_wing;
delta_CD = eta_wing*KCD;
// pitching moment fit
delta_Cm_a = (-0.01892 - 0.0056476*alpha + 1.0205e-5*pow(alpha,2)
- 4.0692e-5*pow(alpha,3) + 1.7594e-6*pow(alpha,4));
delta_Cm_de = (-0.014928 - 0.0037783*alpha + 0.00039086*pow(de,2)
- 1.1304e-5*pow(de,3) - 1.8439e-6*pow(de,4));
delta_Cm = delta_Cm_a + delta_Cm_de;
KCm_alpha = delta_Cm_a/eta_ref_wing;
KCm_de = delta_Cm_de/eta_ref_tail;
delta_Cm = (0.75*eta_wing + 0.25*eta_tail)*KCm_alpha + (eta_tail)*KCm_de;
// hinge moment
if (alpha < 13)
{
delta_Ch_a = (-0.0012862 - 0.00022705*alpha + 1.5911e-5*pow(alpha,2)
+ 5.4536e-7*pow(alpha,3));
}
else
{
delta_Ch_a = 0;
}
delta_Ch_e = -0.0011851 - 0.00049924*de;
delta_Ch = -(delta_Ch_a + delta_Ch_e);
KCh = -delta_Ch/eta_ref_tail;
delta_Ch = eta_tail*KCh;
// rolling moment
CL_diff = (eta_wing_left - eta_wing_right)*KCL;
delta_Cl = CL_diff/4;
}
void add_ice_effects()
{
CL_clean = -1*CZ*cos(Alpha) + CX*sin(Alpha); //Check later
CD_clean = -1*CZ*sin(Alpha) - CX*cos(Alpha);
Cm_clean = Cm;
Cl_clean = Cl;
Ch_clean = Ch;
CL_iced = CL_clean + delta_CL;
CD_iced = CD_clean + delta_CD;
Cm_iced = Cm_clean + delta_Cm;
Cl_iced = Cl_clean + delta_Cl;
//Ch_iced = Ch_clean + delta_Ch;
CL = CL_iced;
CD = CD_iced;
Cm = Cm_iced;
Cl = Cl_iced;
//Ch = Ch_iced;
CZ = -1*CL*cos(Alpha) - CD*sin(Alpha);
CX = CL*sin(Alpha) - CD*cos(Alpha);
}
// SIS Twin Otter Iced aircraft Nonlinear model
// Version 020409
// read readme_020212.doc for information
#include "uiuc_iced_nonlin.h"
void Calc_Iced_Forces()
{
// alpha in deg
double alpha;
double de;
double eta_ref_wing = 0.08; // eta of iced data used for curve fit
double eta_ref_tail = 0.12;
double eta_wing;
//double delta_CL; // CL_clean - CL_iced;
//double delta_CD; // CD_clean - CD_iced;
//double delta_Cm; // CM_clean - CM_iced;
double delta_Cm_a; // (Cm_clean - Cm_iced) as a function of AoA;
double delta_Cm_de; // (Cm_clean - Cm_iced) as a function of de;
double delta_Ch_a;
double delta_Ch_e;
double KCL;
double KCD;
double KCm_alpha;
double KCm_de;
double KCh;
double CL_diff;
alpha = Alpha*RAD_TO_DEG;
de = elevator*RAD_TO_DEG;
// lift fits
if (alpha < 16)
{
delta_CL = (0.088449 + 0.004836*alpha - 0.0005459*alpha*alpha +
4.0859e-5*pow(alpha,3));
}
else
{
delta_CL = (-11.838 + 1.6861*alpha - 0.076707*alpha*alpha +
0.001142*pow(alpha,3));
}
KCL = -delta_CL/eta_ref_wing;
eta_wing = 0.5*(eta_wing_left + eta_wing_right);
delta_CL = eta_wing*KCL;
// drag fit
delta_CD = (-0.0089 + 0.001578*alpha - 0.00046253*pow(alpha,2) +
-4.7511e-5*pow(alpha,3) + 2.3384e-6*pow(alpha,4));
KCD = -delta_CD/eta_ref_wing;
delta_CD = eta_wing*KCD;
// pitching moment fit
delta_Cm_a = (-0.01892 - 0.0056476*alpha + 1.0205e-5*pow(alpha,2)
- 4.0692e-5*pow(alpha,3) + 1.7594e-6*pow(alpha,4));
delta_Cm_de = (-0.014928 - 0.0037783*alpha + 0.00039086*pow(de,2)
- 1.1304e-5*pow(de,3) - 1.8439e-6*pow(de,4));
delta_Cm = delta_Cm_a + delta_Cm_de;
KCm_alpha = delta_Cm_a/eta_ref_wing;
KCm_de = delta_Cm_de/eta_ref_tail;
delta_Cm = (0.75*eta_wing + 0.25*eta_tail)*KCm_alpha + (eta_tail)*KCm_de;
// hinge moment
if (alpha < 13)
{
delta_Ch_a = (-0.0012862 - 0.00022705*alpha + 1.5911e-5*pow(alpha,2)
+ 5.4536e-7*pow(alpha,3));
}
else
{
delta_Ch_a = 0;
}
delta_Ch_e = -0.0011851 - 0.00049924*de;
delta_Ch = -(delta_Ch_a + delta_Ch_e);
KCh = -delta_Ch/eta_ref_tail;
delta_Ch = eta_tail*KCh;
// rolling moment
CL_diff = (eta_wing_left - eta_wing_right)*KCL;
delta_Cl = CL_diff/4;
}
void add_ice_effects()
{
CL_clean = -1*CZ*cos(Alpha) + CX*sin(Alpha); //Check later
CD_clean = -1*CZ*sin(Alpha) - CX*cos(Alpha);
Cm_clean = Cm;
Cl_clean = Cl;
Ch_clean = Ch;
CL_iced = CL_clean + delta_CL;
CD_iced = CD_clean + delta_CD;
Cm_iced = Cm_clean + delta_Cm;
Cl_iced = Cl_clean + delta_Cl;
//Ch_iced = Ch_clean + delta_Ch;
CL = CL_iced;
CD = CD_iced;
Cm = Cm_iced;
Cl = Cl_iced;
//Ch = Ch_iced;
CZ = -1*CL*cos(Alpha) - CD*sin(Alpha);
CX = CL*sin(Alpha) - CD*cos(Alpha);
}