US 5,435,503 AGrant
Real Time Missile Guidance System
Issue Date:1995-07-25
•33 Claims
•7 Drawing Sheets
Abstract
A weak Hamiltonian finite element method is used for iterative computation of missile guidance acceleration commands for maximizing a missile's terminal velocity while satisfying control authority limits and terminal attitude constraints. The guidance acceleration commands include commands for controlling the angle of attack (.alpha.) and the bank angle (.phi.) of the missile. The angle of attack (.alpha.) and bank angle (.phi.) are related to a set of virtual control variables selected to avoid convergence problems when the angle of attack is approximately zero. The preferred control variables are .beta..sub.2 and .beta..sub.3 such that .beta..sub.2 =cos.phi.tan.alpha. and .beta..sub.3 =sin.phi.tan.alpha.. Iterative convergence is facilitated when control inequality constraint parameters are reached by adjusting iterative solutions between iterations toward satisfaction of the constraints. An approximation to an optimal trajectory is calculated at each guidance cycle during missile flight using data which are revised during each guidance cycle. The revised data include current position data for the target and the current position for the missile. The revised data are taken from the most reliable source currently available, such as on-board target-seeking radar when the target-seeking radar is locked onto the target, uplink data from ground or airborne tracking radar when an uplink is operational, or inertial guidance data. Extracted from the optimal trajectory is an optimal acceleration command for optimally controlling the angle of attack and bank angle of the missile.
Metadata
Assignee
- Loral Vought Systems Corp.
Inventors
- Malvin G. Johnson, Jr.
- Glenn T. Haskins
- Dewey H. Hodges
Application Information
Application Number:US 1136364
Filing Date:1993-08-27
Priority Date:1993-08-27
Art Unit:221
Classifications
IPC:
F41G 700F42B 1501
Field of Search:
2443643.11;3.15423;424.01
Patent Drawings (7 sheets)
Description
Background of the Invention
A. Field of the Invention
This invention relates to missile guidance, and more particularly to real-time computation of missile guidance commands for optimal guidance of the missile.
B. Description of the Related Art
1. General Theoretical Art
a. Calculus of Variations
The basic objective of the calculus of variations is to find the unknown function which will locate an extremum for a definite integral. Ordinary calculus locates the extremum for a function of one or more variables. For example:
is a velocity function. This function varies with the positions x,y,z, the thrust (thr) and the control variable u. To find the extremum of the velocity, the first variation of the velocity equation is used: ##EQU1## Following it has been shown that the aforementioned variational relation simplifies to the following set of equations: ##EQU2## The variables which solve these equations define an extremum for velocity.
As noted earlier, it is the extremum of a definite integral which is formally in the domain of the calculus of variations. Using another function for velocity provides the example: ##EQU3##
In lieu of an extremum for a function, an extremum for a definite integral is found. And in lieu of finding the appropriate variables, the appropriate functions (h) are found. It has been shown that the necessary and sufficient condition for the solution to this type of extremum problem can be the solution of the Euler-Lagrange equations: ##EQU4##
The Euler-Lagrange equations are obtained by setting the "first variation" of the integral in Equation 4 to zero. The first variation is found by varying each of the functions h.sub.i (x) by an amount .epsilon..sub.i u.sub.i (x) where .epsilon..sub.i is an arbitrary constant independent of x and h.sub.i (x), and u.sub.i (x) denotes any arbitrary function of x which is independent of .epsilon..sub.i. Then the integral I+.delta.I is defined by: ##EQU5##
Upon solving for 6I to obtain a power series expansion of .delta.I in terms of powers of .epsilon..sub.i, the "first variation" is the coefficient of the term for .epsilon..sub.i to the first power.
It is still necessary, for the problems of the type in optimal control, to solve a variational problem with auxiliary conditions. Now consider that the functions h.sub.i are not independent but restricted by some auxiliary conditions. Initially we considered velocity as a function of several variables x, y, z, thr and u. In solving the definite integral we allowed the integral of velocity to be several functions of one variable t. Now we will consider a system composed of several functions of several variables. The auxiliary system functions will be portrayed as f.sub.i, with the variables as x.sub.j and t, creating the set of equations: ##EQU6## with V[h.sub.i (x.sub.j,t)] as before. The auxiliary condition on the integrand creates a problem when the system functions are not known to be independent. As an expletive, if the system functions were independent, the independent equations could be expressed under separate integrals as follows: ##EQU7## From this set of equations would follow m+1 independent Euler-Lagrange equations. However, since it is not known that the functions are independent, knowledge of the other relationships is required. Using a Lagrange multiplier, the previously assumed independent equations may be expressed as dependent equations under the same integral as: ##EQU8## A solution set utilizing the Euler-Lagrange equations can be shown to be: ##EQU9##
An alternative methodology to the Euler-Lagrange technique for solution to the variational problem with auxiliary conditions is the Finite Element Method. This alternative methodology is a key element in making optimal missile guidance "viably eligible" for real-time processing.
b. Finite Element Method (FEM)
Given the integral: ##EQU10## The time intervals can be broken up into N elements with a normalized elemental time definition of: ##EQU11## As an example, the aforementioned equation using a trapezoidal integration scheme becomes: ##EQU12## and are of a similar nature to the forthcoming equations used in the weak Hamitonian method of optimal missile guidance. The finite discretization appears similar to trapezoidal integration because of the selection of the shape functions.
2. Application Specific Art
a. Missile Guidance
Missiles are rocket-motor-boosted vehicles that are designed to fly pilotless from a launch position to a desired target position. Various techniques have been used to guide a missile depending upon the launch and target positions and the operational characteristics of the missile.
A first class of techniques uses an external processor for performing computations to derive guidance commands for a missile. The external processor is located on the ground or in an airborne launch vehicle. The guidance commands are transmitted from the external processor to the missile via a radio frequency or laser uplink communication path. In response to the transmitted guidance commands, onboard instruments guide the missile.
A second class of techniques uses computations that are performed onboard the missile. Inputs for such onboard computations may come from instruments carried onboard the missile itself or from signals transmitted to the missile from external sources.
Generally, instruments carried in a missile for performing onboard guidance include inertial devices that sense body accelerations and rotational rates with respect to three orthogonal axes. These inertial measurements provide position, velocity, and attitude information to a programmed digital processor unit that computes the guidance commands for the missile.
Optimal guidance of a missile involves guiding the missile to optimize a preselected performance function (J) subject to certain constraints.
The real-time control of the missile involves the additional problem of determining control values as a function of time in order to guide the missile along a trajectory that will optimize the performance function J subject to certain inequality constraints on the control variables. An inequality constraint is a constraint that is an inequality. In other words, an inequality constraint is a prescribed condition that one quantity is less than, less than or equal to, greater than, or greater than or equal to another quantity. An inequality constraint function is a function that defines an inequality constraint. For a missile, the control variables include the angular acceleration of the missile, up to certain maximum values. For a solid-fueled rocket, the thrust generally is not controlled, although the thrust may be controlled for a liquid fueled missile, up to a certain maximum value.
b. Weak Hamiltonian Optimization Method (WHOA)
For solving the missile control problem, the dynamic state of the missile is represented by E state variables x.sub.i (t) satisfying E "state equations" that are the first-order differential equations: ##EQU13## The n state equations are specific formulations of the accessory conditions in equation (3) above, in which the time t has been selected as the independent variable. The state variables x.sub.j (t) typically include the missile's coordinates (x, y, z), angular orientation, linear velocity, angular velocity, and mass. The functions u.sub.1 (t), u.sub.2 (t), . . . , u.sub.r (t) are the control variables. The control problem can be expressed as the problem of finding the control variables that maximize or minimize the definite integral ##EQU14## In general, the final time t.sub.f will be an unknown.
Real-time control of the missile requires a numerical method of solving the equations subject to suitable boundary conditions. Due to limited data processing capability when a numerical solution is computed in the missile, a finite-element procedure is used in which the control variables are iteratively computed for a time t+.DELTA.t based on conditions existing at a time t. The computation should give a finite approximation of an optimal solution. Additionally, the computation should be "unconditionally stable."
A computational technique having these desirable properties is proposed in Hodges et al., "Weak Hamiltonian Finite Element Method for Optimal Control Problems," Journal of Guidance Control, and Dynamics, Vol. 14, No. 1, January-February 1991, American Institute of Aeronautics and Astronautics, pp. 148-156,incorporated herein by reference; and Bless et al., "Finite Element Solution of Optimal Control Problems with Inequality Constraints," Proceedings of the 1990 American Control Conference, San Diego, Calif., May 23-25, 1990, Volume 1, American Automatic Control Conference, pp. 242-247, incorporated herein by reference.
The weak Hamiltonian method is a combination of optimal control theory and finite element discretization with weakly coupled boundary conditions. Permitting a finite element solution to the first variation, the performance function (J) is formulated as: ##EQU15## where: L is an integrand component of the performance function;
.lambda. is a matrix of unknown Lagrangian multiplier functions for adjoining the system state equations;
T denotes the matrix transpose operation;
f is a matrix of system state equations;
x is a matrix of the state rates with respect to time;
.mu. is a matrix of unknown Lagrangian multiplier functions for adjoining the control constraints;
G is a matrix of control inequality constraint functions G(x,u,t) for control functions u(t) such that G.ltoreq.0;
K is a matrix of slack variables such that G+k.sup.2 =0; .PHI..ident..phi.+.upsilon..sup.T .PSI. where
.phi. is a performance function of the states and time, defined only at the initial and final times t.sub.o and t.sub.f ;
.upsilon. is a matrix of unknown discrete Lagrangian multipliers defined only at the initial and final times t.sub.o and t.sub.f for adjoining boundary condition constraints;
.PSI. is a matrix of boundary condition constraints imposed at the initial and final times t.sub.o and t.sub.f ;
.varies. is a matrix of unknown discrete Lagrangian boundary condition multipliers for natural coupling of finite element values to finite element nodes of the states at the initial and final times t.sub.o and t.sub.f ; and
x is a matrix of finite element nodes of the states at the initial and final times t.sub.o and t.sub.f, such that
x.vertline.t.sub.o is the limit of x(t) as t.fwdarw.t.sub.o ;
x.vertline.t.sub.f is the limit of x(t) as t.fwdarw.t.sub.f ;
x.vertline.t.sub.o is x (t.sub.o); and
x.vertline.t.sub.f is x (t.sub.f).
As described in Hodges et al. and Bless et al. cited above, the first variation is taken of the performance function (J) of Equation 16. The first variation has an x term, which is then integrated by parts. The boundary conditions are placed in weak form, and the first variation is set to zero, giving the following solution for a minimum or maximum to the performance index J of Equation 16: ##EQU16##
Equation 17 is the governing equation for the weak Hamiltonian method with control constraints. An important advantage of the weak Hamiltonian method is that time derivatives of the states and co-states do not appear in Equation 17. This facilitates the simplistic choice for the shape functions which leads to the simplified integration which looks like a trapezoidal integration scheme.
Equation 17 is solved by performing finite element discretization. The integral in Equation 17 is broken into N integrals, with each of the N integrals being an integral over one of the N finite elements, respectively from t=t.sub.o to t=t.sub.f. Constant shape functions are used for u, .delta.u, x, .lambda., .mu., and K, and linear shape functions are used for .delta.x , .delta..lambda., .delta..mu., and .delta.K. The integral over each element can therefore be evaluated by inspection, and the resulting integral in Equation 17 becomes a summation over N elements. Solutions are found for only discrete mid-point values for the controls u(t). The boundaries between the finite elements are referred to as nodes, which occur at t=t.sub.i. When there are no state discontinuities, the nodal values of the shape functions in the neighboring elements cancel each other in the summation. When there are state discontinuities, the nodal values of the shape functions in the neighboring elements do not necessarily cancel each other in the summation. This problem is conveniently solved by treating the state discontinuity as a pair of boundary conditions at the node, one boundary condition for the integral or summation for a time less than but approaching t.sub.i in the limit, and the other boundary condition for the integral or summation greater than but approaching t.sub.i in the limit. In other words, the time line and the performance integral in Equation 9 is broken into two different "phases" by the state discontinuity, and a respective discrete Lagrangian multiplier is introduced for adjoining each of the two boundary conditions into the performance function (J).
The summation over N elements becomes a system of simultaneous algebraic equations. In general, 2n of the 4n endpoint values for the states and corresponding Lagrangian multipliers (x.sub.o, x.sub.f, .lambda..sub.o, .lambda..sub.f) must be specified. The initial conditions (x.sub.o) are known in accordance with physical constraints. Also, .lambda..sub.f can be specified in terms of other unknowns, because the admissible variations of the states must be continuous at the initial and final times: ##EQU17## Therefore, the system of algebraic equations includes as many equations as unknowns, and it can be solved for the control variables.
Normally, the system of equations can be solved by expressing the Jacobian explicitly and using a Newton-Raphson solution procedure. The Newton-Raphson solution procedure is described, for example, on pages 43-44 of the CRC Handbook of Mathematical Tables, 1980, Chemical Rubber Company Press Inc., Boca Raton, Fla. Let the system of algebraic equations be designated as E.sub.i (X)=0, where X is a vector of unknowns. The Newton-Raphson solution procedure is an iterative method that starts with a trial solution X.sup.[0 ], and obtains successive approximations X.sup.[j+1] by solving the simultaneous linear equations ##EQU18##
Equation 21 can be written as a matrix equation
where [J] is the so-called Jacobian matrix having the elements that are the partial derivatives ##EQU19## evaluated at X.sub.k =X.sub.k.sup.[j]. Equation 22 can be solved by inverting the Jacobian matrix, to obtain the iterative equation:
A sparse matrix solver can be used as coded in Duff, I.S., Harwell Subroutine Library, Computer Sciences and Systems Division, Harwell Laboratory, Oxfordshire, England, February 1988, Chapter M. To solve for the solution at the first element, an initial guess can be made based on the initial conditions. An initial guess for successive elements can be based upon the solutions for a number of previous elements. In general, the iterations of the Newton-Raphson procedure converge rapidly, so that a large number of elements can be solved with a very efficient run-time on a computer.
Successful flight of a missile to its target and the missile's arrival at the target with a desired attitude and velocity depends upon the nature of the guidance procedure and its mechanization. Current guidance procedures and their mechanizations have failed to achieve the potential success suggested by optimal guidance and control theory. This results primarily from approximations and mathematical truncations imposed in order to enable completion of the guidance computational task in an acceptable time frame.
The weak Hamiltonian finite element method used for the basis of this invention failed to robustly converge upon an optimal solution when put into practice, as have all of the published schemes for real time optimal guidance. In particular the method as published has been subject to convergence problems when the control variables enter certain ranges. When the control values either enter around zero or close to a control boundary, the published method fails to converge upon a solution.
Summary of the Invention
This invention is a real-time missile guidance system and method that periodically computes guidance control commands. The guidance control commands generate a field of extrema for maximizing the missile's terminal velocity while satisfying control authority limits and terminal attitude constraints. The guidance control commands include commands for controlling the angle of attack (.alpha.) and the bank angle (.phi.) of the missile.
In accordance with a first aspect of the invention, at least two control variables, .beta..sub.2 and .beta..sub.3, are selected which specify the angle of attack (.alpha.) and the bank angle (.phi.), but which have definite and limited values when specifying an angle-of-attack (.alpha.) of approximately zero. The selection of .beta..sub.2 and .beta..sub.3 as virtual control variables is made in lieu of selecting the angle of attack (.alpha.) and bank angle (.phi.) as virtual control variables. The transformation between the control variables .beta..sub.2 and .beta..sub.3, and the angle of attack (.alpha.) and the bank angle (.phi.), avoids convergence difficulties when specifying an angle of attack (.alpha.) in the vicinity of zero.
In a preferred embodiment, the control variables are computed in accordance with a weak Hamiltonian method of optimizing the missile trajectory, and the control variables are selected to obtain a formulation that avoids negative eigenvalues, when the angle of attack is near zero in a partial derivative of the "Hamiltonian" matrix, corresponding to the performance function (J), with respect to the control variables. The preferred set of control variables includes .beta..sub.2 and .beta..sub.3 such that .beta..sub.2 =cos.phi.tan.alpha. and .beta..sub.3 =sin.phi.tan.alpha.. Because both .beta..sub.2 and .beta..sub.3 are proportional to tan.alpha., it is also desirable to define an intermediate variable .beta.=tan.alpha., and to express the aerodynamic tables, which are conventionally a function of alpha and Mach number, as a function of .beta. and Mach number.
In accordance with another aspect of the invention, the guidance commands are computed by an iterative Newton-Raphson procedure, and convergence is facilitated when control inequality constraint parameters are reached by modifying iterative solutions between iterations so that the iterative solutions satisfy the inequality constraints while facilitating the movement of the controls on to and off of the control boundaries. Moreover, when it is known that the solution must fall within certain limits, and when it is known that certain parameters have either positive or negative values, corresponding solution constraints can be defined for the purpose of accelerating convergence with this technique. A solution constraint, for example, can be imposed on the time of flight in order to accelerate convergence by guaranteeing positive time steps of at least a certain magnitude.
In accordance with a further aspect of the invention, an optimal trajectory is calculated at each computational cycle during missile flight using data that are revised during each computational cycle. The revised data include current position data for the target and the current position for the missile. Airborne initialization of each computational cycle is permitted to accommodate externally derived and uplinked data. The revised data are taken from the most reliable source currently available, such as on-board target-seeking radar when the target-seeking radar is locked onto the target, uplink data from ground or airborne tracking radar when an uplink is operational, or inertial guidance data.
Brief Description of the Drawings
Other objects and advantages of the invention will become apparent upon reading the following detailed description and upon reference to the drawings, in which:
FIG. 1 is a plan view of a tactical missile incorporating the present invention;
FIG. 2 is a schematic diagram illustrating the use of the tactical missile of FIG. 1 for air defense;
FIG. 3 is a block diagram of a guidance processor and its interconnection to various systems in the tactical missile of FIG. 1.;
FIG. 4 is a schematic diagram of the guidance system for the tactical missile of FIG. 1, including a dynamic model of the missile;
FIG. 5 is a time line illustrating the finite elements and nodes used in the derivation of the system of algebraic equations for optimizing a performance function (J) in accordance with the weak Hamiltonian finite-element method;
FIG. 6 is a schematic diagram illustrating body axes of the missile, a relative wind vector, and the angle of attack;
FIG. 7 is a schematic diagram illustrating the bank angle of the missile;
FIG. 8 is a flowchart illustrating the iterative control procedure of the present invention for generating guidance acceleration commands in real time during flight of the missile; and
FIG. 9 is a flow chart of a procedure for accelerating convergence when control inequality constraints are reached by modifying iterative solutions between iterations when the iterative solutions fall outside of the inequality constraint boundaries.
While the invention will be described in connection with a certain preferred embodiment, it is not intended to limit the invention to this particular embodiment. On the contrary, it is intended for the invention to include all modifications, alternatives, and equivalent arrangements as may be included within the scope of the appended claims.
Detailed Description of the Preferred Embodiment
Turning now to the drawings, there is shown in FIG. 1 a plan view of a tactical missile generally designated 20. Various internal components of the missile 20 are shown in phantom lines. For controlling the missile, the missile has a guidance processing unit 21 (GPU), which is a digital computer 21 that will be further described below with reference to FIG. 2. For determining the position and angular orientation of the missile, the missile has an inertial measurement unit 22 (IMU) including a respective gyro or angular rate sensor for sensing angular rotation of the missile about each of the three orthogonal (x,y,z) axes of the missile as shown in FIG. 7, and a respective accelerometer for sensing linear acceleration of the missile along each of the (x,y,z) axes of the missile. For adjusting the attitude or angular orientation of the missile, the missile has an attitude control system 23 (ACS) and an aerodynamic maneuvering system 24 (AMS). The aerodynamic maneuvering system 24 adjusts the angular orientation of fins 25 at the tail end of the missile, which will adjust the attitude of the missile when the missile is at low altitude. The attitude control system 23 includes lateral-thrusting rocket motors near the nose 26 of the missile, which can adjust the attitude of the missile even at high altitude. For seeking the target, the missile has a seeker 27 built into the nose of the missile. Depending on the desired target and the size of the missile 20, the seeker 27 may use radar, laser, infra-red, or optical imaging techniques for seeking a target.
FIG. 2 illustrates the use of the tactical missile 20 for seeking and intercepting a target 30, such as a ballistic missile. The ballistic missile 30 is initially detected and tracked by ground radar 31 or airborne radar 32. When the tactical missile 20 is launched, it is programmed with the coordinates of a predicted point of interception. To intercept the ballistic missile 30, however, it is necessary for the tactical missile 20 to be guided in real time so that the trajectory of the tactical missile intersects the trajectory of the ballistic missile. The seeker 27 of the tactical missile 20 must eventually acquire or lock onto the ballistic missile to ensure intersection. Because the seeker 27 has a limited range and limited field-of-view, successful acquisition by the seeker can be facilitated by initially guiding the missile in response to tracking data transmitted from the ground radar 21 or the airborne radar 32 before acquisition by the seeker. The tracking data are transmitted via laser or radio along an uplink path 33.
The tactical missile 20 should have a maximum velocity at the time of interception, and should have a predetermined attitude at interception in order to maximize damage to the ballistic missile 30. The predetermined attitude at interception is specified by a desired crossing angle .theta..sub.CR between the longitudinal axis 35 of the tactical missile 20 and the longitudinal axis 36 of the ballistic missile. The tactical missile 20 should impact the ballistic missile 30, for example, at a desired aim point 34 near the warhead of the ballistic missile. This desired aim point 34 is generally different from the radar centroid 35 as viewed by the seeker 27. In general, the field-of-view of the seeker 27 is directed along a line-of-sight that is the same as the velocity 38 of the tactical missile 20 relative to the tactical missile 30.
The present invention enables guidance of the missile to a terminal geometry condition where the crossing angle .theta..sub.CR can be specified with respect to two axes, while maximizing the missile's terminal velocity. For targets, such as airplanes, that are not approximately symmetrical about their longitudinal axis, the crossing angle .theta..sub.CR could be specified by two values, such as a total angle component .theta..sub.CR.alpha. giving the magnitude of the total angle between the between the longitudinal axis 35 of the tactical missile 20 and the longitudinal axis of the target, and an orientation component .theta.CR.sub..phi. specifying the crossing angle in a plane perpendicular to the longitudinal axis of the target and with respect to a lateral axis of the target. For a target such as an airplane, for example, the lateral axis of the target could be assumed to be a vertical axis with respect to the Earth.
Turning now to FIG. 3, there is shown a block diagram of the guidance processing unit 21 and its interconnection to other components and systems of the tactical missile (20 in FIG. 1). The guidance processing unit 21 is a high-speed digital processor having an internal floating-point math unit. It operates with external random access memory 41, external read-only memory 42, an external floating-point coprocessor 43, interrupt and reset control circuitry 44, synchronization and clock generator circuitry 45, and a multiplex bus controller 46. The multiplex bus controller controls a multiplex bus (MUX) for parallel input/output of the guidance processing unit. External interfaces to other missile subsystems are accomplished through both discrete signal interconnects and via the multiplex data bus. The multiplex bus interfaces the guidance processing unit to the radar seeker 27, the inertial measurement unit 22, the aerodynamic maneuvering system 24, and an uplink data receiver 47. The guidance processing unit 21 is interfaced to the attitude control system 23 via a serial link 48.
The guidance processing unit 21 is programmed to perform a number of time-shared functions. These functions include an alignment function 48, an autopilot function 49, a navigation function 50, and a guidance function 51. The alignment function 48 involves sensing and controlling the attitude and roll of the missile 20, as will be further described below with reference to FIG. 4. The autopilot function 49 maintains the velocity vector of the missile 20 along a specified direction by issuing angular acceleration commands to the alignment function. The navigation function 50 keeps track of the current position of the missile 20 and changes the specified heading of the missile along a specified trajectory. The guidance function 50 calculates in real time an optimal trajectory to a target based on data available during flight of the missile.
Turning now to FIG. 4, there is shown a schematic diagram that more particularly illustrates the relation of the guidance function 51 with the other functions and systems in the tactical missile. The guidance function 51 is responsive to time as indicated by the synchronization and clock generator circuits 45. The guidance function 51 is also responsive to a position error signal as indicated by the radar seeker 27 when the seeker 27 has acquired a definite lock on the target, or as indicated by the difference between the target position, as indicated by tracking data received by the uplink data receiver 47, and the position of the tactical missile, as maintained by the navigation function. The target position at desired intercept and the desired missile attitude at intercept are either pre-computed prior to launch and downloaded to the missile's guidance processing unit prior to launch or provided in-flight via an uplink message received by the uplink data receiver 47.
The navigation function 50 is initially programmed with the position of the launch site, and after launch, the navigation function updates the initial position by integration of linear acceleration signals from the inertial measurement unit 22. The navigation function may adjust this position computation with tracking data received by the uplink data receiver 47, to coincide with the position of the missile as determined by the ground or airborne tracking radar (31 and 32 in FIG. 2).
The guidance function is responsive to an attitude error signal that represents the difference between a desired terminal attitude at target interception, and the missile attitude as indicated by the alignment function. The missile attitude is conveniently represented by two attitude angles (.gamma., .PSI.). The first attitude angle .gamma., for example, represents a vertical angle from a horizontal inertial plane to the missile's velocity vector. The second attitude angle .PSI., for example, represents a horizontal angle measured from North to the projection of the missile's velocity vector on the horizontal plane.
The guidance function 51 is responsive to other performance criteria. The performance criteria may specify that the guidance function 51 is to maximize velocity at the point of interception with the target.
The guidance function 51 computes guidance acceleration commands in order to approximate an optimal trajectory. The acceleration commands, for example, specify linear acceleration in g's along each of body axes (b.sub.1, b.sub.2, b.sub.3) of the missile, as defined in FIG. 6 below. The autopilot 49 is responsive to these linear acceleration commands, and specifies changes to the bank angle (.phi.) and angle of attack (.alpha.) of the missile. The bank angle (.phi.) and angle of attack (.alpha.) are further discussed below with reference to FIGS. 6 and 7. The alignment function 48 operates the attitude control system 23 or the aerodynamic maneuvering system 24 to change the bank angle (.phi.) and angle of attack (.alpha.) of the missile in accordance with the commands from the autopilot 49. The actual change in attitude of the missile is sensed by the inertial measurement unit 22 and indicated to the alignment function, so that the alignment function may use feedback control to compensate for the missile's dynamic and aerodynamic response 52. The missile's dynamic and aerodynamic response 52 to the reactions of the attitude control system 23 or the aerodynamic maneuvering system 24 result in the missile flying to a new position with an altered attitude and velocity. Linear accelerometers (not shown) and gyros (not shown) sensing angular rates of change in the inertial measurement unit 22 provide measurement data of the missile's new state and enable feedback information for comparison with desired values and the initiation of the next computational cycle in the missile GPU.
The guidance function 51 is restricted by at least one control limit or inequality constraint. For aerodynamic stability, for example, the "angle of attack (.alpha.)" of the missile must not exceed a predetermined value, such as 20 degrees. In the system of FIG. 4, the guidance function does not receive as an input the actual angle of attack, nor does the guidance function directly specify as an output what the angle of attack (.alpha.) should be. Instead, the guidance function 51 has a mathematical model of the aerodynamic characteristics of the missile. The guidance function can therefore determine, for any potential trajectory, whether the angle of attack (.alpha.) will exceed its control limit. Consequently, the guidance function will not produce acceleration commands that would require the angle of attack (.alpha.) to exceed its control limit.
The guidance function 51 periodically computes guidance acceleration commands. An acceleration command for each of the three missile body axes is computed during each computational cycle. During each cycle, a solution vector X is computed iteratively by the Newton-Raphson solution procedure of Equation 23 above. The solution vector X represents a complete trajectory from the present position of the missile to the predicted position of the target at the time of interception of the target by the missile.
Preferably the system of equations E(X) in Equation 23 above are algebraic equations result from a discretization of Equation 17, above, that computes the integral in Equation 17 by a Finite Element Method. This system of algebraic equations will be listed below, but to understand the listing, it is necessary to first understand the notation employed in the equations. The equations will use the Einstein summation representation, wherein each permutation of the indices represents one equation in the system. The equations are written to represent a system for any number of nodes, and any number of phases, or groups, of consecutive nodes.
As shown in FIG. 5, the time interval from an initial time t.sub.o to a final time t.sub.f is broken into N discrete elements .DELTA.t.sub.1 to .DELTA.t.sub.N. The endpoints of the elements represent nodes at t.sub.o=t.sub.1 to t.sub.N+1 =t.sub.f. The value of a parameter (x) at a node will be represented as (x) The value of a parameter (x) at a mid-point between a pair of neighboring nodes will be represented as (x). Moreover, the parameter will be indexed as follows:
Examples:
With the above notation in mind, the preferred system of algebraic equations is: ##EQU20##
In accordance with an aspect of the present invention, the control variables (u.sub.k) include a set of control variables functionally related to the angle of attack (.alpha.) and bank angle (.phi.) in such a way that convergence problems are avoided when the angle of attack is approximately zero. When the angle of attack (.alpha.) and bank angle (.phi.) are used as two of the control variables (u.sub.k), there are convergence problems in the iterative solution procedure. These convergence problems arise when the angle of attack (.alpha.) is approximately zero, because the bank angle (.phi.) may assume any value when angle of attack (.alpha.) is zero. To solve these convergence problems, at least two control variables are used that are related to the angle of attack (.alpha.) and the bank angle (.beta.) by a predefined mathematical transformation of variables such that the two control variables assume appropriate values for typical flight characteristics of the missile. Moreover, because the guidance function does not directly sense the actual values of the angle of attack (.alpha.) or the bank angle (.beta.), nor does the guidance function directly control the angle of attack (.alpha.) or the bank angle (.beta.), the two control variables related to the angle of attack (.alpha.) and the bank angle (.beta.) are more precisely referred to as "virtual" control variables.
The angle of attack is shown in FIG. 6 in relation to the body axes (b.sub.1,b.sub.2, b.sub.3) of the tactical missile 20. The b.sub.1 body axis is the longitudinal axis of the missile 20. The angle of attack (.alpha.) is defined as the total angle between the longitudinal axis (b.sub.1) of the missile and the relative wind vector rw. The relative wind vector is the velocity of the wind relative to the missile. Therefore, when the velocity of the missile with respect to an inertial frame greatly exceeds the velocity of the wind with respect to an inertial frame, the relative wind vector will become approximately anti-parallel to the velocity vector of the missile, and the angle of attack (.alpha.) will become relatively small.
The bank angle (.phi.) is shown in FIG. 7. The bank angle (.phi.) is the angle of the relative wind vector rw with respect to the -b.sub.1 body axis of the missile 20.
The preferred set of control variables are .beta..sub.2 and .beta..sub.3 such that: ##EQU21##
The inverse transformation is: ##EQU22##
It is also convenient to define an intermediate variable .beta. such that: ##EQU23##
The control variables are used in the state equations of the missile, and in particular are related to forces in the equations of motion for the missile. The forces along the missile body axes are given by: ##EQU24## It is convenient to define a 3.times.3 transformation matrix [CwB] between the wind frame and the missile body frame, and a 3.times.3 transformation matrix [CwI] between the wind frame and the local inertial frame. Then the forces originating in the inertial frame are due to the mass (m) of the missile and the acceleration (g) due to gravity: ##EQU25## The forces in the wind frame are therefore given by: ##EQU26## The matrix [CwB] is a function f(.alpha.,100 ) of the angle of attack (.alpha.) and the bank angle (.phi.) according to:
Equation 24-7 and 24-8 above, however, can be satisfied with a negative eigenvalue for the optimality condition when .alpha..apprxeq.0. The eigenvalues of the optimality condition are eigenvalues of the matrix ##EQU27## A negative eigenvalue indicates that the solution converges upon a stationary point other than the desired extremum. Another formulation that has a negative eigenvalue is [CwB] =f(.alpha..sub.1,.alpha..sub.2), where .alpha..sub.1 is a "vertical" angle of attack, and .alpha..sub.2 is a "horizontal" angle of attack. Negative eigenvalues are avoided by the "tangent variable" formulation such that: ##EQU28##
A set of equations governing the motion of the missile 20 is needed for trajectory optimization. It is assumed that the missile 20, represented as a rigid body M, is moving in an inertial frame A subject to inertial, aerodynamic, and body forces. The inertial properties of M are idealized to be those of a particle M* of mass m. The gravity force at M* is mgk, where k is a unit vector directed vertically downward. The geometry of M is axisymmetric, and the aerodynamic forces acting at M* can be characterized in terms of a frame B in which the unit vector b.sub.1 is fixed, parallel to the axis of M, and in which the direction of V, the velocity of M* in A, is known. Thus, the aerodynamic forces consist of an axial force (T-D)b.sub.1 and a lift force Ln, where n is normal to b.sub.1 and parallel to the plane which contains b.sub.1 and V. The quantities D and L are known functions of v=.vertline.v.vertline. and .alpha., the angle between V and b.sub.1 ; T is a known function of time. (In the degenerate case for which V is parallel to b.sub.1, L=0. Note that the frame B, in which b.sub.1 is fixed, is not a body-fixed frame. The missile is free to spin about its axis in B.) Initial conditions affect only the initial position and velocity of M* in A.
First, we define a point 0 and a dextral triad a.sub.i, for i=1,2,3, all fixed in A. Letting p denote the position vector of M* relative to 0 and x.sub.i =p.multidot.a.sub.i, we can form the velocity of M* in A as
Denoting the unit vector tangent to the path of M* as w.sub.1 we can also write the velocity as
One could characterize the direction of V in terms of only two parameters, but we choose instead to introduce an intermediate frame W associated with the path of M* and in which V is fixed. A dextral triad w.sub.i, for i=1,2,3, is defined to be fixed in W, and we write the associated direction cosine matrix
in terms of Rodrigues parameters so that ##EQU29## where CC.sup.T= C.sup.T C=I, the 3.times.3 identity matrix, and ##EQU30## Since there are three .theta..sub.i and we only need two parameters, we will need to constrain .theta..sub.i. We choose to do so with a nonholonomic constraint on the angular velocity below.
Equating expressions for V from Eqs. (34-1) and (34-2), and using (34-3), we obtain three kinematical equations of the form
or, in matrix form
where ##EQU31##
Denoting the angular velocity of W in A by .omega., we let ##EQU32## where .omega..sub.1 =.omega..multidot.W.sub.i. Based on Eq. (26-4), one can show that ##EQU33## the inverse of which is given by ##EQU34##
We need the acceleration of M* , which can be written as the inertial time derivative of .upsilon.w.sub.1 so that ##EQU35##
Finally, we need to define the relationship between B and A to define the aerodynamic forces. We will actually work in terms of relating B to W, but first W must be well defined. We will set .omega..sub.1 =0, which nonholonomically constrains the set of coordinates .theta..sub.i. This way the orientation of frame W is governed only by the path of M*, and only the orientation of b.sub.1 and the lift force need to be defined relative to W.
As with pointing the velocity vector in A, we only need two parameters to accomplish this: the angle of attack .alpha. and the "bank" angle .phi. to orient the plane in W which contains the missile axis. Let B be tentatively oriented so that b.sub.i coincides with w.sub.i for i=1,2,3. Now, to get to the final orientation, we rotate B about b.sub.1 by an angle .phi.. Next, rotate B about b.sub.i by an angle .alpha.. This brings b.sub.1 into alignment with the axis of M during flight, and the lift is assumed to act along b.sub.2. Thus, ##EQU36## Now, to avoid convergence difficulties created by the vanishing of .alpha. in which .phi. then becomes undefined, we introduce two variables .beta..sub.2 and .beta..sub.3 such that
so that
Thus, ##EQU37## When combined with the expressions for the force and with the ()* operator notation introduced below, an orientation description such as embodied in Eq. (34-16) should be free of convergence difficulties when .alpha. and .beta. vanish.
This concludes the definition of all kinematical quantities needed to express the equations of motion.
The equations of motion can be written in accordance with Newton's second law
where, if k is taken as parallel to a.sub.3, the active force resultant acting at M* is given by
Introducing the column matrix of force components in the W system ##EQU38## where F.sub.i =F.multidot.w.sub.i so that ##EQU39## and taking the dot product of Eq. (35-1) with w.sub.1, and making use of Eqs. (34-12) and (35-4), one obtains three equations of motion of the form
Eqs. (35-5b) and (35-5c), along with the definition of .omega. and the constraint this .omega..sub.1 =0, ##EQU40##
The system equations can now be written as a system of seven first-order ordinary differential equations. The three equations embodied in Eq. (34-7) are already in a matrix form. Eq. (35-5a) is one scalar equation. Finally, Eqs. (34-11) and (35-6) can be written as one matrix equation. Thus, the system equations are ##EQU41## where F.sub.i is given by Eqs. (35-4). Control variables .beta..sub.2 and .beta..sub.3 can be used to determine the trajectory of M*.
Eqs. (35-4) appear to be quite complex. The reason for the complexity is, at least in part, related to the fact that L and D are functions of .alpha. while .beta. appears explicitly in the coefficients of L and D, as well as that of T. To circumvent this, L and D can be expressed in terms of .beta. instead. All explicit .beta. dependence can be grouped into implicit functions which can be splined or approximated as a series. One way to do this is to introduce
where ##EQU42## and where C.sub.N and C.sub.A are considered functions of .beta. and Mach number. Thus, the force components become ##EQU43## Now, upon substitution of the following even functions of .beta. ##EQU44## which can be conveniently represented as splines, Eq. (29-3) simplifies to ##EQU45##
In construction of the nonlinear algebraic equations and the associated Jacobian, we will ultimately need to form first and second derivatives of the force components with respect to the control variables B.sub.2 and .beta..sub.3. Such partial derivatives of any function of .beta. must be found from the chain rule, which involves derivatives of the function with respect to .beta. times one of the following partial derivatives: ##EQU46## which will have the effect of introducing indeterminate forms ##EQU47## in the vicinity of .beta.=0 every time there is differentiation with respect to one of the control variables.
The first derivative of any even function of .beta. will be odd, but its first derivative divided by .beta. will be even. Motivated by this observation, the following notation is introduced: ()*=()'/.beta. such that ()'=.differential.()/.differential..beta.. Now, ##EQU48## and so forth for higher derivatives ##EQU49## This way, if any of these functions, say Q, is represented as a power series ##EQU50## one can show that ##EQU51## Once these are calculated, we then can find
We note that if the expansion of Q is limited to cubic order about the origin, say a cubic spline, then Q*=2Q.sub.2 and Q**=0.
Of course, this operator does not change the results of finding derivatives. It just removes the myriad of indeterminate forms. Also, this operator greatly simplifies the expressions for the second derivatives. For example, ##EQU52## which are well behaved in the limit of small .beta.. In fact, F.sub.i and all their first and second derivatives with respect to .beta..sub.2 and .beta..sub.3 are also well behaved, as has been demonstrated by Mathematica 5 computations. For .beta.=0 it can be shown that ##EQU53## where the fact that Q*(0)=-1 is used.
If H is the Hamiltonian, and u is the column matrix which contains the control variables .beta..sub.2 and .beta..sub.3, then H.sub.uu for this problem becomes a diagonal matrix in the limit as .beta. approaches zero. Both diagonal elements are the same and are given by ##EQU54## It is therefore evident that H.sub.uu will be positive definite in the vicinity of .alpha.=.beta.=0 if .lambda..sub..upsilon. <0.
A specific solution depends on the mass, thrust, gravitational, atmospheric, and aerodynamic data. In the following data, for example, the thrust is presumed to have discontinuities at Stage1.sub.-- time seconds and Stage2.sub.-- time seconds after launch. Therefore, it is convenient to divide the performance integral J into three phases; a first phase from t=0 seconds to t=Stage1--time seconds; a second phase from t=Stagel.sub.-- time seconds to t=13 seconds; and a third phase from t=13 seconds to the time of intercept with the target.
Turning now to FIG. 8, there is shown a top-level flowchart of the guidance function (introduced as box 51 in FIG. 4). This flowchart includes a number of variable names and symbols. X.sup.j.sub.i denotes the solution vector for the jth cycle and the ith Newton-Raphson iteration. The solution vector X.sup.j.sub.i is composed of: values for the states, x; the costates, .lambda.; the Lagrangian multipliers .mu. for the control constraint equations; the slack variables K for the control constraint equations; the controls, u; the boundary condition multipliers .nu.; the time to the end-of-phase, t.sub.nph ; and the time to the end-of-flight, t.sub.go. An initial solution vector X.sup.FCC is down-loaded into the guidance processing unit (21 in FIG. 3) from a fire control computer (not shown) external to the missile. E(X.sup.j.sub.i) is the vector used in Equations 21 to 23 above and which contains the residuals from the equations when the equations are evaluated with the vector X.sup.j.sub.i.
The variable NC is the cycle counter. NC is initialized to zero, and NC is incremented every time that the WHOA optimal guidance routine is called. The variable `k` is a Newton-Raphson iteration counter, which is reinitialized before the Newton-Raphson iteration scheme is called each cycle. The variable `k.sub.max `, is the maximum value number of Newton-Raphson iterations allowed for convergence. ADG is the acceleration command vector for the missile. The acceleration command vector ADG is the output from the guidance procedure.
In the first step 61 of FIG. 8, execution branches to step 62 if NC is zero. In step 62, the solution vector X.sub.1.sup.1 is initialized with the solution vector X.sup.FCC down-loaded from the fire control computer. If NC is not zero, then step 63 is performed, which loads the initial solution vector X.sub.1.sup.NC with the last best solution vector X.sub.1.sup.NC-1 from the last cycle. Once the initial solution vector for this cycle is obtained in steps 63 or 64, the execution continues in step 64.
In step 64, the cycle counter NC is incremented, and the iteration counter `k` is initialized to 1. In step 65, the input boundary conditions are transformed to consistent units used by the program. These input boundary conditions include the missile position and velocity vector as well as the desired target intercept position and attitude. In step 66, execution branches to step 67 if it is desirable to re-size the Jacobian. An increase in the number of elements, for example, increases the precision of the solution, but also increases the computation time needed for computing a solution and obtaining convergence. The computation time is dependent on the particular phase of flight. Therefore, it is generally desirable to re-size the Jacobian if a flight phase transition occurs. In step 67, the solution vector X.sub.k.sup.NC is redefined based upon the re-sized Jacobian matrix. In step 68, the computation of the equation residuals E(X.sub.k.sup.NC) is done in vector form using the proposed solution vector X.sub.k.sup.NC. This is the initial step in the Newton-Raphson iteration loop including the following steps 68 to 75. During each iteration `k`, the Newton-Raphson iteration loop computes a new solution X.sub.k+1.sup.NC +1 as a predetermined solution function S(X.sub.k.sup.NC) of the last solution X.sub.k.sup.NC. The computation of the new solution, however, involves steps 71, 72 and 73 in addition to the initial step 68.
Once the equation residuals are defined, they are validated against convergence criteria, in step 69. The convergence criteria, for example, are specified by a vector of predefined error limits, and if any element of the vector of equation residuals E(X.sub.k.sup.NC) has a magnitude that exceeds it corresponding error limit, then the solution vector X.sub.k.sup.NC fails the convergence criteria. Otherwise, if each element of the vector of equation residuals E(X.sub.k.sup.NC) has a magnitude that is less than its corresponding error limit, then the solution vector X.sub.k.sup.NC passes the convergence criteria, and in step 77, the acceleration command requisite for optimal flight (AGD.sup.NC) is computed as a predetermined function g(X.sub.k.sup.NC) of the converged solution vector X.sub.k.sup.NC.
If the convergence criteria are not met, then in step 70, the Newton-Raphson iteration counter k is compared against k.sub.max. This step is used to control the computation time used by the procedure. If k is greater than k.sub.max, then an acceleration command vector is computed either based upon the last best acceleration command vector or by using a conventional proportional navigation scheme, depending upon the time-off-light. This is done in step 76.
If the Newton-Raphson iteration counter `k` does not exceed the limit `k.sub.max `, execution continues from step 70 to step 71. In step 71, the Jacobian matrix is updated by re-evaluating all of the non-zero elements of the Jacobian matrix [J] for the previously-computed solution vector X.sub.k.sup.NC. In step 72 the Jacobian matrix is inverted, and the inverted Jacobian matrix is multiplied by the previously-computed solution vector X.sub.k.sup.NC in order to compute the next delta value .DELTA.X on the solution vector X. Therefore, in step 73, the new solution vector X.sub.k+1.sup.NC is computed by subtracting the delta value .DELTA.X from the previously-computed solution vector X.sub.k.sup.NC In step 74 the Newton-Raphson counter, `k`, is incremented by one. Then, in step 75, the control constraint equations are checked using the control values (u) in the new solution vector X.sub.k+1.sup.NC. This step is essential to free up the controls (u) to move on and off of control boundaries. The Newton-Raphson iteration scheme is resumed again in step 68, until either the convergence criteria are passed in step 69 or the maximum number of iterations k.sub.max is exceeded in step 70.
Turning now to FIG. 9, there is shown a flowchart of a procedure for checking whether control limits are exceeded in each iterative solution X.sub.k.sup.NC, and adjusting the iterative solution when the controls move on or off a boundary or when a new time solution exceeds the bounds of realizability. The virtual angle of attack (.alpha.), for example, should be limited to less than a predetermined value such as 20.degree.. This could be done by imposing a constraint involving .beta..sub.2 and .beta..sub.3 such as tan.sup.2 (20.degree.)-.beta..sub.2.sup.2 -.beta..sub.3.sup.2 =G. The limit of 20.degree. represents a constraint boundary for the angle of attack. If the boundary is exceeded, then the constraint can be enforced by either setting the specified value for the angle of attack to 20.degree. or some value less than 20.degree.. Sometimes the constraint is enforced by setting the specified value for the angle of attack (.alpha.) to the limit of 20.degree., because the limit value might be optimum. This is a case where a control is set on the constraint boundary, in order to enforce the constraint. Sometimes the constraint is enforced by setting the specified value for the angle of attack (.alpha.) to a value less than 20.degree.. In this case, the control is said to be moved off of the boundary. The procedure of FIG. 9 may also check for solution constraints imposed for the express purpose of manifesting realizability thereby accelerating realizable convergence. For example, when it is known that the time step .DELTA.t must be greater than a certain predetermined amount, then a solution constraint can be imposed on .DELTA.t to accelerate convergence.
In general, the procedure in FIG. 9 enforces the constraints k.gtoreq.0, .mu.*K=0, G.ltoreq.0, t.sub.go .gtoreq.0 and .mu..gtoreq.0 in the iterative solution X.sub.k.sup.NC. Although a true solution to the system of algebraic equations being solved satisfies the constraints, an iterative solution does not necessarily satisfy the constraints because an iterative solution is only a guess at the true solution. Moreover, the procedure is performed on each respective set of K, G, u, and .mu. values.
In a first step 81 of FIG. 9, execution branches to step 82 when the computed time to the end of flight (tgo.sub.k) is not greater than zero. In step 82, the time to the end of flight tgo.sub.k for the present iteration `k` is set equal to the time to the end of flight for the previous iteration `k-1. When the computed time to the end of flight tgo.sub.k is found to be greater than zero in step 81, or after step 82, execution continues in step 83.
In step 83, execution branches depending on the logical value of the proposition ".mu.>0. AND. K=0". When this proposition is true, then execution branches to step 84. Execution branches in step 84 depending on whether G is greater than or equal to zero. When G is not greater than or equal to zero, an inequality constraint is violated, and execution branches from step 84 to step 85. In step 85, the value of the control (u) is moved off of the constraint boundary. The degree to which the new value of the control (u) differs from its limit at the control boundary is not critical, and the new value may be set to a value that differs from its limit by an amount proportional to the value of .mu.. Then, in step 86, the discrete Lagrangian multiplier .mu. is set to zero, and K is calculated based on the value of the control (u) and for .mu.=0, or example, such that K.sup.2 =-G(u). Then execution returns.
When G is greater than or equal to zero, none of the constraints are violated. However, in this case, execution branches from step 84 to step 86, because convergence can be accelerated in this case by setting .mu.=0 and calculating K for .mu.=0, for example, such that K.sup.2 =-G(u).
When it is not true that ".mu.0. AND. K=0", then execution branches from step 83 to step 87. Execution branches from step 87 depending on the logical value of the proposition "K<0. AND. .mu.=0". When this proposition is true, execution branches to step 88. In step 88, execution branches depending on whether G is greater than or equal to zero. When G is found to be less than zero, then execution branches to step 89. In step 89, K is calculated from the values of the other parameters, for example as K.sup.2 =-G(u), and execution returns. When G is found in step 88 to be greater than or equal to zero, then execution branches to step 90. In step 90, the control (u) is moved on the constraint boundary. Next, in step 91, K is set to zero and .mu. is calculated for the value of the control (u) and for K=0, for example, by applying Equation 24-7 above. Then execution returns.
When step 87 finds that "K<0. AND. .mu.=0" is not true, then execution branches to step 92. In step 92, execution branches depending on the logical value of the proposition "(K.noteq.0. AND. .mu..noteq.0) .OR. (K=0 .AND. .mu.=0)". If this proposition is false, then execution returns. Otherwise, execution branches to step 93.
In step 93, execution branches depending on whether G.gtoreq.0. When G is greater than or equal to zero, then execution branches to step 94. In step 94, K is set to zero, .mu. is calculated for K=0, for example by applying Equation 24-7 above, and execution returns. When G is less than zero, then execution branches to step 95. In step 95, .mu. is set equal to zero, and K is calculated for K=0, for example such that K.sup.2 =-G(u). Then execution returns.
Additional features and advantages of the present invention are illustrated in the program listed below in Appendix I to the specification. The entry point of the program is a subroutine Whoa.sub.-- flight. The subroutine Whoa.sub.-- flight receives, as an input parameter, a specified time after launch of the missile. The specified time is used to determine the thrust of the rocket motor and the weight of the missile, because the thrust and weight are assumed to be predetermined functions of time. The subroutine WHOA.sub.-- flight also receives, as input parameters, the present vehicle velocity, vehicle position, and the anticipated final azimuth, final elevation, and final position. The subroutine Whoa.sub.-- flight returns a parameter body.sub.-- guidance.sub.-- cmd(3), which specifies the acceleration in g's, along each of the three body axes (b.sub.1, b.sub.2, b.sub.3), that the missile should have to be guided along an optimal trajectory.
The subroutine Whoa.sub.-- flight first calls a subroutine Whoa.sub.-- init.sub.-- flight that sets up what needs to be initialized for computing the Hamiltonian in flight. Then the subroutine Whoa--flight calls a subroutine Whoa.sub.-- optimize to compute an optimal trajectory. If a converged solution is obtained, or if the time is less than 0.001 seconds since launch, then the subroutine Whoa.sub.-- flight calls a subroutine Inert.sub.-- guidance to compute a new guidance command. Otherwise, the previous guidance command is used. In any case, the subroutine Whoa.sub.-- flight is finished.
The first time that the subroutine Whoa.sub.-- init.sub.-- flight is called, it obtains an initial solution by calling a subroutine Initialize.sub.-- I.sub.-- O(sol), which creates aerodynamic data tables by calling a subroutine Bcuaero. Otherwise, the subroutine Whoa.sub.-- init.sub.-- flight uses the present vehicular states as boundary conditions. The subroutine Whoa.sub.-- init.sub.-- flight then calls a subroutine Find.sub.-- matrix.sub.-- size to determine the number of elements needed, and thus determine the size of the Jacobian matrix. If the size needs to be changed, then the subroutine Whoa.sub.-- init.sub.-- flight calls a subroutine Gen.sub.-- whoa.sub.-- init.sub.-- param in order to alter the solution vector to accommodate a new matrix size. Finally, the subroutine Whoa.sub.-- init.sub.-- flight determines an initial estimate for the time-of-flight (tgo).
The subroutine Whoa.sub.-- optimize first determines the number of non-zero elements in the Jacobian matrix. This information is used by low-level matrix manipulation routines (e.g., Factor.sub.-- matrix, Invert.sub.-- matrix) that are from the Harwell Subroutine Library, as cited above. The subroutine Whoa.sub.-- optimize then calls a subroutine Newton that performs the Newton-Raphson iterations. The subroutine Whoa.sub.-- optimize is then finished.
The subroutine Newton calls a subroutine Errjac to compute the error of a solution, to fill in the Jacobian matrix, and to invert the Jacobian matrix, which is actually done by the Invert.sub.-- matrix routine from the Harwell Subroutine Library. The subroutine Newton computes a new solution from the previous solution and the inverted Jacobian matrix. The subroutine Newton calls a subroutine Eval.sub.-- cc to check control boundaries and constrain the solution, and calls a subroutine Evaluate.sub.-- convergence to check the new solution. The subroutine Newton returns when a converged solution has been found, or after a predetermined number of iterations.
The subroutine Eval.sub.-- cc calls either a subroutine Check.sub.-- ccl to check a single control constraint equation, or a subroutine Check.sub.-- cc2 to check two control constraint equations. The subroutine Check.sub.-- ccl calls a subroutine Eval.sub.-- bndry.sub.-- cnd1 to evaluate the single boundary condition, and the subroutine Check.sub.-- cc2 calls a subroutine Eval.sub.-- bndry.sub.-- cn2.sub.-- to evaluate two boundary conditions. These subroutines are specific examples of programming for the flowchart of FIG. 9, as described above.
A large portion of the program in Appendix I is devoted to subroutines called by the subroutine Errjac and used to fill the Jacobian matrix. A low-level routine Get.sub.-- f.sub.-- g.sub.-- eqn.sub.-- der, for example, computes the derivatives for the elements of the Jacobian matrix, and also computes the derivative terms appearing in Equation 17.
In view of the above, there has been described a weak Hamiltonian finite element method for real-time iterative computation of missile guidance acceleration commands that define and optimal solution for maximizing the missile's terminal velocity while satisfying control authority limits and terminal attitude constraints. The guidance acceleration commands include commands for controlling the angle of attack (.alpha.) and the bank angle (.phi.) of the missile. In the virtual model of the missile contained in the guidance procedure, control variables .beta.2 and .beta..sub.2 are utilized in lieu of .alpha. and .phi. to avoid convergence problems when the angle of attack is approximately zero. This transformation avoids negative eigenvalues in the Hamiltonian corresponding to the performance function (J). Iterative convergence is permitted when control inequality constraint parameters are reached by modifying iterative solutions between iterations so that the iterative solutions satisfy the inequality constraints. This facilitates the movement of the controls on and off of constraint boundaries. Moreover, solution constraints can be imposed for the purpose of accelerating convergence by using this technique.
Preferably, an optimal trajectory is calculated at each guidance cycle during missile flight using data that are revised during each guidance cycle. The revised data include current position data for the target and the current position for the missile. The revised data are taken from the most reliable source currently available, such as on-board target-seeking radar when the target-seeking radar is locked onto the target, uplink data from ground or airborne tracking radar when an uplink is operational, or inertial guidance data. ##SPC1##
Claims
What is claimed is:
1. A method of real-time guidance and control of a missile by computing control commands in a data processor during flight of said missile to guide said missile along a trajectory that will optimize a predetermined performance function, said control commands adjusting an angle of attack (.alpha.) between a longitudinal axis of said missile and a relative wind vector, and a bank angle (.phi.) between a transverse axis of said missile and said relative wind vector, said method periodically performing computational cycles in a data processor to compute a set of values of said control commands during each cycle based on a current missile velocity vector, and a target intercept position relative to a current position of said missile, wherein the computations during each cycle include the steps of: a) computing new values for a set of control variables based on said performance function and said current missile velocity vector, and said target intercept position relative to said current missile position; and b) computing said control commands from said new values of said set of control variables; wherein said control variables include at least two control variables that together specify said angle of attack (.alpha.) and said bank angle (.phi.), but which have definite and limited values when specifying an angle of attach (60 ) of approximately zero.
2. The method as claimed in claim 1, wherein said at least two control variables include a first control variable (.beta..sub.2) that is a cosine function of the specified bank angle (.phi.), and a second control variable (.beta..sub.3) that is a sine function of the specified bank angle (.phi.), so that the ratio (.beta..sub.3 /.beta..sub.2) of said second control variable to said first control variable is proportional to the tangent of the specified bank angle (.phi.).
3. The method as claimed in claim 2, wherein said first control variable (.beta..sub.2) and said second control variable (.beta..sub.3) are related to the specified angle of attack (.alpha.) and the specified bank angle (.phi.) as .beta..sub.2 =cos.phi. tan.alpha. and .beta..sub.3 =sin.alpha. tan.alpha..
4. The method as claimed in claim 1, wherein said step (a) of computing new values for said set of control variables is performed iteratively and wherein a new iterative solution (X.sup.[k+1]) is obtained as a predetermined solution function (S(X.sup.[k])) of a previous iterative solution (X.sup.[k]) beginning with a predetermined trial solution (X.sup.[0]).
5. The method as claimed in claim 4, wherein said previous iterative solution (X.sup.[k]) computed from said predetermined solution function is checked for violation of an inequality constraint, and said previous iterative solution is adjusted toward satisfying said inequality constraint before being used in said solution function to compute said new iterative solution (X.sup.[k+1]).
6. The method as claimed in claim 5, wherein said solution function is predetermined to solve a system of algebraic equations including a summation over finite elements of time, and said inequality constraint specifies that a duration (.DELTA.t) of each of said finite elements of time is at least a predetermined positive value.
7. The method as claimed in claim 4, wherein said solution function is predetermined to solve a system of algebraic equations resulting from taking a first variation of an integral of said performance function along a projected trajectory of said missile to said target intercept position, setting said first variation to zero, and approximating an integral in said first variation as a summation over finite elements, wherein said integral of said performance function includes constraints adjoined to said integral of said performance function by Lagrangian multipliers.
8. The method as claimed in claim 7, wherein said performance function is selected to maximize a terminal velocity of said missile at said target intercept position, and said constraints include an angular orientation of said target intercept position.
9. The method as claimed in claim 4, wherein said solution function is a function of position of a moving target, and said solution function is updated between said computational cycles in response to variation in said position of said moving target.
10. A method of real-time guidance and control of a missile by computing control commands in a data processor during flight of said missile, said method including the steps of periodically performing computational cycles in a data processor to compute a set of values of said control commands during each cycle based on equations of motion for said missile and position of a target relative to position of said missile, wherein the computations during each cycle include an iterative computation of new values for control variables, wherein a new iterative solution (X.sup.[k+1]) is obtained as a predetermined solution function (S(X.sup.[k])) of a previous iterative solution (X.sup.[k]) beginning with a predetermined trial solution (X.sup.[0]), and wherein said previous iterative solution (X.sup.[k]) computed from said predetermined solution function is checked for violation of an inequality constraint, and said previous iterative solution is adjusted toward satisfying said inequality constraint before being used in said solution function to compute said new iterative solution (X.sup.[k+1]), and wherein said control commands are computed from said control variables.
11. The method as claimed in claim 10, wherein said previous iterative solution is adjusted toward satisfying said inequality constraint before being used in said solution function when said previous iterative solution is found to violate said inequality constraint by selectively moving a control variable either on or off of a control boundary of said inequality constraint so that said inequality constraint is satisfied.
12. The method as claimed in claim 10, wherein said solution function is predetermined to solve a system of algebraic equations including a summation over finite elements of time, and said inequality constraint specifies that a duration (.DELTA.t) of each of said finite elements of time is at least a predetermined positive value.
13. The method as claimed in claim 10, wherein said control commands adjust an angle of attack (.alpha.) between a longitudinal axis of said missile and a relative wind vector, and a bank angle (.phi.) between a transverse axis of said missile and said relative wind vector, and wherein said control variables specify said angle of attack (.alpha.) and said bank angle (.phi.) so that said control variables assume definite and limited values when specifying a value of approximately zero for said angle of attack (.alpha.).
14. The method as claimed in claim 13, wherein said control variables include a first control variable (.beta..sub.2) that is a cosine function of the specified bank angle (.phi.), and a second control variable (.beta..sub.3) that is a sine function of the specified bank angle (.phi.), so that the ratio (.beta..sub.3 /.beta..sub.2) of said second control variable to said first control variable is proportional to the tangent of the specified bank angle (.phi.).
15. The method as claimed in claim 14, wherein said first control variable (.beta..sub.2) and said second control variable (.beta..sub.3) are related to the specified angle of attack (.alpha.) and the specified bank angle (.phi.) as .beta..sub.2 =cos.phi. tan.alpha. and .beta..sup.3 =sin.phi. tan.alpha..
16. The method as claimed in claim 10, wherein said control commands guide said missile along a trajectory that will optimize a predetermined performance function, and said solution function is predetermined to solve a system of algebraic equations resulting from taking a first variation of an integral of said performance function along a projected trajectory of said missile, setting said first variation to zero, and approximating an integral in said first variation as a summation over finite elements, wherein said integral of said performance function includes inequality constraints adjoined to said integral of said performance function by Lagrangian multipliers.
17. The method as claimed in claim 16, wherein said integral of said performance function includes a terminal target constraint adjoined to said integral of said performance function by a discrete Lagrangian multiplier, said performance function is selected to maximize a terminal velocity of said missile upon interception with an intercept position of said target, and said terminal target constraint includes an angular orientation of said missile upon arrival at said intercept position of said target.
18. A method of real-time guidance and control of a missile by computing control commands in a data processor during flight of said missile to guide said missile along a trajectory that will optimize a predetermined performance function, said method including the steps of periodically performing computational cycles in a data processor to compute a set of values of said control commands during each cycle based on equations of motion for said missile and position of a target relative to position of said missile, wherein: the computations during each cycle include an iterative computation of new values for control variables, wherein a new iterative solution (X.sup.[k+1]) is obtained as a predetermined solution function (S(X.sup.[k])) of a previous iterative solution (X.sup.[k]) beginning with a predetermined trial solution (X.sup.[0]), said control commands are computed from said control variables; said solution function is predetermined to solve a system of algebraic equations resulting from taking a first variation of an integral of said performance function along a projected trajectory of said missile, setting said first variation to zero, and approximating an integral in said first variation as a summation over finite elements, wherein said integral of said performance function includes an integrand component (L) and an inequality constraint function (G) adjoined to said integral of said performance function by a Lagrangian multiplier (.mu.), and said control commands adjust an angle of attack (.alpha.) between a longitudinal axis of said missile and a relative wind vector, and a bank angle (.phi.) between a transverse axis of said missile and said relative wind vector, and wherein said control variables are selected to specify said angle of attack (.alpha.) and said bank angle (.phi.) so that said control variables assume definite and limited values when specifying a value of approximately zero for said angle of attack, and wherein said equations of motion are defined by state functions (f), and said control variables are selected to avoid negative eigenvalues in the matrix ##EQU55## where i and k are indexed over said control variables, j is indexed over said control variables, and k is indexed over said state functions and said inequality constraint function such that: wherein .lambda. is a matrix of unknown Lagrangian multiplier functions for adjoining the state functions (f), and T denotes a matrix transpose operation.
19. The method as claimed in claim 18, wherein said control variables include a first control variable (.beta..sub.2) that is a cosine function of the specified bank angle (.phi.), and a second control variable (.beta..sub.3) that is a sine function of the specified bank angle (.phi.), so that the ratio (.beta..sub.3 /.beta..sub.2) of said second control variable to said first control variable is proportional to the tangent of the specified bank angle (.phi.).
20. The method as claimed in claim 19, wherein said first control variable (.beta..sub.2) and said second control variable (.beta..sub.3) are related to the specified angle of attack (.alpha.) and the specified bank angle (.phi.) as .beta..sub.2 =cos.phi. tan.alpha. and .beta..sub.3 =sin.phi. tan.alpha..
21. The method as claimed in claim 18, wherein said previous iterative solution (X.sup.[k]) computed from said predetermined solution function is checked for violation of said inequality constraint function, and said previous iterative solution is adjusted toward satisfying said inequality constraint function before being used in said solution function to compute said new iterative solution (X.sup.[k+1]).
22. The method as claimed in claim 21, wherein said previous iterative solution is adjusted toward satisfying said inequality constraint function before being used in said solution function when said previous iterative solution is found to violate said inequality constraint function by selectively moving one of said control variables either on or off of a control boundary of said inequality constraint function so that said inequality constraint function is satisfied.
23. The method as claimed in claim 22, wherein a slack variable. (K) is used in adjoining said inequality constraint function (G) to said integral of said performance function, a new value for said slack variable (K) is computed before being used in said solution function when said control variable is moved off of said control boundary, and wherein a new value for said Lagrangian multiplier (.mu.) is computed before being used in said solution function when said control variable is moved off of said control boundary.
24. The method as claimed in claim 21, wherein said system of algebraic equations includes a summation over finite elements of time, said previous iterative solution includes a duration (.DELTA.t) of each of said finite elements of time, and when said duration (.DELTA.t) of said previous iterative solution is negative, said duration is set to a positive value before being used in said solution function to compute said new iterative solution.
25. The method as claimed in claim 18, wherein said integral of said performance function includes a terminal constraint adjoined to said integral of said performance function by a discrete Lagrangian multiplier, said performance function having been selected to maximize a terminal velocity of said missile upon interception with said target, and said terminal constraint includes an angular orientation of said missile upon said interception with said target.
26. The method as claimed in claim 18, wherein said solution function is a function of position of said target, and said solution function is updated between said computational cycles in response to variation in said position of said target.
27. A missile comprising an inertial measurement unit and a control system for maneuvering said missile in response to control commands to guide said missile along a trajectory that will optimize predetermined performance function, and a data processing unit connected to said control system to transmit said control commands to said control system, said data processor being programmed for periodically computing said control commands during computational cycles by computing a set of values of said control commands during each cycle based on equations of motion for said missile, position of a target relative to position of said missile, and attitude of said missile, said data processor being programmed to perform during each cycle an iterative computation of new values for control variables, wherein a new iterative solution (X.sup.[k+1]) is obtained as a predetermined solution function (S(X.sup.[k])) of a previous iterative solution (X.sup.[k]) beginning with a predetermined trial solution (X.sup.[0]), and said control commands are computed from said control variables, wherein said solution function is predetermined to solve a system of algebraic equations resulting from taking a first variation of an integral of said performance function along a projected trajectory of said missile, setting said first variation to zero, and approximating an integral in said first variation as a summation over finite elements, wherein said integral of said performance function includes an integrand component (L) and an inequality constraint function (G) adjoined to said integral of said performance function by a Lagrangian multiplier (.mu.), and at least one discrete Lagrangian multiplier adjoined to said integral of said performance function for specifying a terminal constraint on said attitude of said missile upon interception with said target, and wherein said control commands adjust an angle of attack (.alpha.) between a longitudinal axis of said missile and a relative wind vector, and a bank angle (.phi.) between a transverse axis of said missile and said relative wind vector, and wherein said control variables are selected to specify said angle of attack (.alpha.) and said bank angle (.phi.) so that said control variables assume definite and limited values when specifying a value of approximately zero for the angle of attack (.alpha.), and wherein said processor is programmed for checking said previous iterative solution X.sup.[k] computed from said predetermined solution function for violation of said inequality constraint function, and adjusting said previous iterative solution toward satisfying said inequality constraint function before being used in said solution function to compute said new iterative solution X.sup.[k+1].
28. The method as claimed in claim 27, wherein said control variables include a first control variable (.beta..sub.2) that is a cosine function of the specified bank angle (.phi.), and a second control variable (.beta..sub.3) that is a sine function of the specified bank angle (.phi.), so that the ratio (.beta..sub.3 /.beta..sub.2) of said second control variable to said first control variable is proportional to the tangent of the specified bank angle (.phi.).
29. The missile as claimed in claim 28, wherein said first control variable (.beta..sub.2) and said second control variable (.beta..sub.3) are related to the specified angle of attack (.alpha.) and the specified bank angle (.phi.) as .beta..sub.2 =cos.phi. tan.alpha. and .beta..sub.3 =sin.phi. tan.alpha..
30. The missile as claimed in claim 27, wherein said data processor is programmed to adjust said previous iterative solution toward satisfying said inequality constraint function before being used in said solution function when said previous iterative solution is found to violate said inequality constraint function by selectively moving a control variable either on or off of a control boundary of said inequality constraint function so that said inequality constraint function is satisfied.
31. The method as claimed in claim 30, wherein a slack variable (K) is used in adjoining said inequality constraint function (G) to said integral of said performance function, and said data processor is programmed to compute a new value for said slack variable (K) before being used in said solution function when said control variable is moved off of said control boundary, and to compute a new value for said Lagrangian multiplier (.mu.) before being used in said solution function when said control variable is moved on said control boundary.
32. The missile as claimed in claim 27, further comprising an uplink receiver connected to said data processor for providing uplinked tracking data indicating the position of said target relative to the position of said missile during flight of said missile for use in said computational cycles.
33. The missile as claimed in claim 27, wherein said performance function has been selected to maximize a terminal velocity of said missile upon interception with said position of said target.
Patent Citations (2)
Non-Patent Literature (6)
- Hodges et al., "Weak Hamiltonian Finite Element Method for Optimal Control Problems," Journal of Guidance Control, and Dynamics, vol. 14, No. 1, Jan.-Feb. 1991, American Institute Aeronautics and Astronautics, pp. 148-156.
- Bless et al., "Finite Element Solution of Optimal Control Problems with Inequality Constraints," Proceedings of the 1990 American Control Conference, San Diego, California, May 23-25, 1990, vol. 1, American Automatic Control Conference, pp. 242-247.
- CRC Handbook of Mathematical Tables, 1980, Chemical Rubber Company Press Inc., Boca Raton, Florida, pp. 43-44.
- McGraw-Hill Encyclopedia of Science & Technology, vol. 8, 1992, McGraw-Hill, Inc., New York, N.Y., pp. 307-310.
- Korn et al., Mathematical Handbook for Scientists and Engineers, 1961, McGraw-Hill, Inc., New York, N.Y., pp. 344-372, 471-480, 784-793.
- Robert R. Bless, "Time-Domain Finite Elements in Optimal Control With Application to Launch-Vehicle Guidance," NASA Contract Report 4376, National Aeronautics and Space Administration, 1991.