DE10033368A1 - Guiding structure for missiles - Google Patents

Abstract

The invention relates to a steering structure for guiding a missile provided with a sensor system to a target, wherein a measurement vector of observable quantities can be derived from the sensor system. The invention is based on the object of improving the probability of a missile being used against highly agile targets by using the "encounter kinematics", ie the movements of the missile and the target, to steer the missile. According to the invention, a model of both the missile and the sensor system and the target is displayed in a missile-side computer. The sensor system delivers a real measurement vector. The measurement vector results from the real relative kinematics of the missile and target. The real measurement vector is compared with an estimated measurement vector which results from the model of the missile, target and sensor system. The real and the estimated measurement vector will normally not match, especially since the target's motion states are unknown. Therefore an optimal filter, e.g. a Kalman filter, an extended Kalman filter or the state-dependent Riccati equation SDRE, acted upon by the deviations of the estimated measurement vector from the real measurement vector and modifies the missile target model, i.e. changes its parameters in the sense of reducing this deviation and aligning the measurement vectors to one another. Ultimately, this leads to a model that not only the motion parameters ...

Description

The invention relates to a steering structure for steering with a sensor system provided missile to a target, with a measurement vector of observable quantities can be derived.

A missile is intended for a target, e.g. B. an enemy aircraft to be destroyed be performed. For this purpose, the missile has a seeker head with or several sensors. The sensors detect a target. From the signals of this Sensors are derived from steering signals which guide the missile to the target.

Known missiles have z. B. an image-resolving infrared detector on a two-dimensional arrangement of detector elements. Through a imaging optical system becomes an object scene containing the target on the Infrared detector shown. The infrared detector thus observes a thermal image, in which z. B. the hot engines of an aircraft and their exhaust gas jet as intense Radiation source appear. Through image processing, the Infrared detector delivered "electronic image" the position of the target in the visual field determined. In known missiles, the seeker head is stabilized in space and by the Missile movements decoupled. The search head can also with the Infrared detector rigidly sitting in the missile, being also rigid with the cell of the Missile connected ("strap-down") inertial sensors the decoupling of the electronic image of the movement of the missile.  

The search head is usually tracked real or virtual to the target. From this The line of sight rotation rate is determined by tracking. The steering takes place at the "Proportional navigation" so that the lateral acceleration of the missile is proportional this line of sight rotation rate is.

There are also missiles with radar or laser sensors or sensors used in others Wavelength ranges work as infrared. There are also missiles with more than one Sensor. One then speaks of a sensor system.

The goal, e.g. B. an attacked aircraft, usually also has sensors, which recognize the attack by an approaching missile. The target then hits "Countermeasures" that seek to disrupt the approaching missile or distract. In particular, however, the attacked aircraft is made by suitable maneuvers seek to escape the approaching missile.

When the missile is directed into the target, the movement of the Missile itself determined and considered by an inertial measuring unit. The Movement quantities of the target, however, are unknown. The expected behavior of the Target, such as whether the target performs an escape maneuver and how this escape maneuver the missile does not "know". The missile only detects the line of sight to the Target, if necessary distance and change of distance and reacts to their changes.

The invention is based on the object, the hit probability when used to improve a missile against highly agile targets.

The invention is based in particular on the task for the control of Missile the "encounter kinematics", ie the movements of the missile and target consulted.

According to the invention this object is achieved in that

• a) in the missile a model of missile-target relative kinematics and Sensor system is shown,
• b) the model is modifiable by optimal filter means, which of the Deviation of a measurement vector resulting from the model from the real one Measurement vector are applied, with the modification an adaptation of the Model is aimed at reality, and
• c) the missile depending on one generated in the model, the relative Kinematics of missile and target representing relative kinematics vector steerable is.

According to the invention, a model of both the Missile and the sensor system and the target shown. The sensor system returns a real measurement vector. The measurement vector results from the real one Relative kinematics of missile and target. The real measurement vector is shown with a estimated measurement vector, which is derived from the model of missile, target and Sensor system results. The real and the estimated measurement vector are usually do not match, especially since the movement states of the target are not known are. Therefore an optimal filter, e.g. B. a Kalman filter, an extended Kalman filter or the state-dependent Riccati equation SDRE, from the deviations of the estimated measurement vector acted upon by the real measurement vector and modified Missile target model, so changes its parameters, in the sense, this To reduce deviation and to align the measurement vectors with each other. Leading finally to a model that is not just the motion parameters of the missile but also simulates the movement parameters of the target. There is in this model a vector of relative kinematics. Depending on this vector, the Guiding the missile. It can be shown in the simulation that this also applies to highly agile targets of the missiles despite escape maneuvers with significantly improved Hit probability can be led to the target.  

The relative kinematics vector is advantageously based on an optimal steering law the missile's autopilot.

A commanded lateral acceleration a _{MC is applied to} the autopilot. The commanded lateral acceleration depends on the relative kinematics vector after a function (). This function () represents the steering law and is selected so that there is an optimal control behavior.

The model of the missile and target can be improved in that the signals of an inertial measuring unit of the missile are also applied to the model. The movements of the missile in inertial space can be measured. The determination of the missile movement states in the relative kinematics x _{M is} improved by the connection of these measured variables to the model.

A further improvement of the modeling can be achieved by estimating means for estimating the target acceleration to which the measurement vector, image data from a image-resolution detector of the missile and data of the inertial measurement unit of the Missile are activated and by which the model can be corrected.

The estimating means are expediently a suitably trained neural network or a fuzzy neural network.

The target is no longer a point target in the distance range of interest here. On image-resolving detector "sees" a contour in the image of the target. With an airplane as The goal is lateral acceleration combined with a change in the contour. The Airplane must be in an inclined position and / or changes in position around the Go vertical axis. Taking these contour changes into account, if necessary other observed variables such as the distance to the lateral acceleration getting closed. A human pilot can do this based on his experience estimated. Accordingly, a neural network trained on it can do the same  except from the image data from the measurement vector and - taking into account the Own movement of the missile - from the signals of the inertial measuring unit of the Missile is affected.

Modeling can also be improved by estimating means for estimating the three-dimensional trajectory of the target and the type of target to which the Measurement vector, image data from an image-resolving detector of the missile and data the inertial measurement unit of the missile and through which the model is correctable.

Here, too, the estimation means are expediently a suitably trained neuronal Network, a fuzzy-neural network or, in the simplest case, a rule-based fuzzy Logic unit.

An observer can from the image of the target with knowledge of the measurands of the Measurement vector and of course taking into account the own movement due to its Experience statements about the three-dimensional path of the target as well as the target type do. This can also be a neural network trained on it or a rule-based fuzzy logic unit.

If so through "prior knowledge" certain statements about the model, in particular about the destination and its path, then it becomes easier, d. H. going it quickly and requires less computation, using the optimal filter Adapt the model to reality.

The model can be developed in such a way that

• a) that the model of missile-target relative kinematics a plurality of Contains hypotheses about the movement of the target
• b) the model in parallel for each of these hypotheses through an associated optimal filter can be modified in order to adapt the model to reality,  
• c) the estimated state vectors obtained and an output variable of the associated optimal filter is connected to an inference unit, the one or selects several state vectors,
• d) by the inference unit at least one selected state vector on one associated steering controller can be switched on, according to an associated optimal Steering Act works.

In many cases there is knowledge of the movement behavior of potential targets. If the target is an enemy aircraft, then only a limited number of come Aircraft types as a target. The flight behavior of such aircraft types is in the Usually known. In particular, it is known what optimal escape movements are executes such aircraft when approaching a missile and must execute it if it wants to escape the missile. This knowledge can be used to simplify the Modeling can be exploited by for the different target types, e.g. B. Aircraft types, hypotheses over which movements are based. With these Hypotheses are provided for parallel models of the missile-target relative kinematics. Each of these models is optimized by an associated optimal filter. The Optimal filters provide sizes that indicate how well the after a calculation step Adaptation of the model to reality has taken place. With an optimal filter that delivers Residual, d. H. the difference between the measurement vector resulting from the model and the actually measured measurement vector is a measure of the quality of this adaptation. This one from the Sizes supplied to optimal filters, e.g. B. residuals, and estimated value of the relative kinematics from the model are switched to an inference unit. The inference unit can be a neural network or a fuzzy neural network. The inference unit then dials in model closest to reality. The inference unit can also have several Select a model with appropriate weights, if several such models are around are equally well adapted to reality. The estimate of the relative kinematics of each of the selected models is with an associated optimal steering law on the The missile's autopilot.  

By using previous knowledge about the goal, the optimal filter needs less computational effort to model the missile and target to reality adapt. The adjustment is faster and more precise. Accordingly stands also the vector of the missile-target relative kinematics faster and more accurately Available. This corresponds to the way the human brain works: the brain a model of the real world is created based on the sensory impressions, and the human being acts accordingly. Thereby experiences from the subconscious similar situations are brought out and processed. A driver who realizes that image of a vehicle driving in front of him suddenly becomes larger on the retina, does not consider that the apparent growth of the vehicle on an approach caused by the reduced speed of this vehicle will, accordingly, to avoid bumping up its speed should also decrease and that you can do this by pressing a brake pedal can reach. Such considerations would take far too long and become one Cause rear-end collision. Rather, the driver pulls out of his subconscious Experience from comparable situations and applies the brakes without thinking. Thinking is only necessary to adapt the basic pattern to the specific one Situation. The aforementioned logical considerations correspond to the mode of operation of the Matched filter. The use of prior knowledge by estimators or hypotheses corresponds the subconscious.

Embodiments of the invention are below with reference to the associated drawings explained in more detail.

Fig. 1 is a block diagram of a guide structure for directing a sensor system provided with a missile to a target, wherein the adaptation of the model of the encounter situation takes place in the real world only through a matched filter.

FIG. 2 is a block diagram similar to FIG. 1, the target acceleration also being estimated using a neural network.

FIG. 3 is a block diagram similar to FIG. 1, with an additional estimation of the three-dimensional trajectory of the target using a neural network.

FIG. 4 is a block diagram of a steering structure with a plurality of target models and a plurality of optimal filters, each target model being based on a hypothesis about the target movement and the target model used for the steering being selected by means of a neural or fuzzy-neuronal network.

In Fig. 1, the real world is shown as a block diagram above a chain-dotted line 10 . Block 12 designates the flight controller of the target, an enemy aircraft to be attacked. The target dynamics 14 provides the change in a state vector _{T of} the target. A block 16 represents the autopilot of a missile. The autopilot 16 receives a commanded lateral acceleration a _MC at an input 18 in a manner to be described. At an input 20 , the autopilot 16 receives signals from an inertial measurement unit ("IMU"). The change _{M in} the state vector of the missile results from the missile dynamics. This change _{M in} the state vector of the missile acts on the inertial measurement unit 22 of the missile.

The changes in the state vectors _T and _M of the target and the missile result in a change in the relative kinematics, which is represented by a vector _R. This is shown schematically in FIG. 1 by a summing point 24 . In fact, the vector _{R is} not simply the difference between the state vectors _T and _M , but it is generally _R = f ( _T , _M ) plus noise.

The vector _{R of} the change in the relative kinematics is integrated, which is represented by block 26 . This returns the vector x _R. In block 26 , "s" is the Laplace transform variable. The relative kinematics identified by vector x _{R is} detected by sensors which form a sensor system 28 . These sensors can be passive sensors such as image-resolving infrared sensors, or active sensors such as radar or laser sensors. The sensor system delivers z. B. Signals according to the line of sight σ, the line of sight rotation rate, the distance R between the missile and the target and its change. These signals are subjected to signal processing 30 and provide a real measurement vector z .

The missile now contains a model 31 of the real world described above in a computer. In the present case, this is an algorithmically represented model. This model 31 is outlined in FIG. 1 by a dashed line. This model includes a model 32 of the missile and a model 34 of the target. The signals of the (real) inertial measurement unit 22 are applied to the model 32 of the missile. The model 32 of the missile provides an estimate _{M of} the change in the state vector of the missile. The model 34 of the target provides an estimate _{T of} the change in the state vector of the target. The two vectors _M and _T result in a vector _R which represents the change in the relative kinematics between the missile and the target as an estimated value in the model. In Fig. 1 this is again symbolically represented as a summing point 36 , although the vector _{R is} not simply the difference in the changes in the state vectors. Here too, _R = f ( _M , _T ) plus noise in general. The vector _R is integrated, as represented by block 38 . This gives an estimate _{R of} the vector of the relative kinematics. This estimate _R is applied to a sensor model 40 . The sensor model 40 is a model of the sensors and the associated signal processing. The sensor model 40 provides an estimate of the measurement vector.

Since the model generally does not initially correspond to reality, in particular because the behavior of the target is unknown, the estimated value of the measurement vector will initially not match the real measurement vector z . The difference between the two measurement vectors z and is formed at a summing point 42 . This difference is switched to an optimal filter 44 . This optimal filter 44 can, for. B. a Kalman filter, an extended Kalman filter or a solution of the state-dependent Riccati equation (SDRE). The parameters of the model are modified by the optimal filter in order to adapt the model 31 to reality. This is shown symbolically in FIG. 1 by the connection of a vector Δ _R from the optimal filter 44 to the summing point 36 . This corrects the vector _R , which leads to a corrected relative kinematics vector _R after integration. As a result, the estimated value of the measurement vector is adjusted to the real measurement vector z .

The relative kinematics vector _R thus resulting from the model is now connected to an optimal steering controller 46 , which applies a commanded lateral acceleration a _MC to the autopilot of the missile.

The steering structure of FIG. 2 essentially corresponds to the steering structure of FIG. 1. Corresponding parts are provided with the same reference symbols in both figures and are not described again in detail.

The steering structure of FIG. 2 additionally contains a neural network 50 . The measurement vector z is applied to this neural network 50 once. Furthermore, the neural network 50 receives image data from an image sensor or image-resolution detector. Finally, the signals of the inertial measuring unit are also connected to the neural network and take into account the own movement of the missile. The neural network 50 is trained in such a way that from these input variables it delivers estimates for the lateral acceleration of the target. The neural network 50 uses e.g. B. from the fact that a lateral acceleration of the aircraft forming the target is accompanied by a change in position of the aircraft in space and thus a characteristic change in the contour, which is "seen" from the missile. From such a change in contour, experience - in the form of learning processes of the neural network - can therefore be used to infer the size and direction of the lateral acceleration.

The knowledge of the target obtained in this way is taken into account when modeling the target. This is shown symbolically in FIG. 2 in that the acceleration a _{T of} the target is additionally applied to the "summing point" 36 via a loop 52 .

The steering structure of FIG. 3 also essentially corresponds to the steering structure of FIG. 1. Corresponding parts are provided with the same reference symbols in both figures and are not described again in detail.

In a manner similar to FIG. 2, the steering structure of FIG. 3 additionally contains a neural network 54 . The measurement vector z is applied to this neural network 54 as in FIG. 2. Furthermore, the neural network 54 likewise receives image data from an image sensor or image-resolution detector. Finally, the signals of the inertial measuring unit are also connected to the neural network 54 and take into account the intrinsic movement of the missile. However, the neural network 54 is trained in such a way that estimated values for the three-dimensional path of the target are obtained from these input variables. These estimates also serve to better model the relative kinematics. This is symbolized in FIG. 3 by the fact that the output of the neural network 54 is connected to the "summing point" 36 via a loop.

The neural network 54 also provides information about the type of target object from the image data that has been activated.

Another embodiment of the steering structure is shown in FIG. 4. The embodiment of FIG. 4 assumes that the number of possible targets is limited in many cases. So there is only a limited number of aircraft types that are used by an opponent. The flight characteristics of these aircraft are known. It is also known which optimal escape maneuvers a particular aircraft type can make in order to escape an approaching missile. This knowledge is realized in the steering structure of FIG. 4.

The steering structure of Fig. 4 corresponds to Fig. 1. Here, too, corresponding parts are provided with the same reference numerals, and not described again in detail in essential parts of the.

Instead of a model of the target, a plurality of models are provided in the steering structure of FIG. 4, of which each model is based on a hypothesis about the target movement. These hypotheses correspond to e.g. B. possible escape movements of one aircraft type or several aircraft types. These hypotheses about the target movement are shown in FIG. 4 by blocks 34.1 , 34.2 . , , 34 .n shown. Each of these hypotheses gives a vector _T. Each of the hypotheses 34.1 , 34.2 . , , 34 .n is an optimal filter 44.1 , 44.1 . , , 44 .n assigned. The model of the missile is of course the same in all cases. The result is a plurality of parallel models of the missile-target relative kinematics with associated optimal filters, which are adapted to reality in the manner described by Δ _R.

The better the hypothesis about the target movement corresponds to reality in a particular model of the missile-target relative dynamics, the smaller the "residual" with an optimal filter, ie the difference between the measured and the estimated value. An inference unit 58 receives, as inputs, the vectors _R from the various models and the residuals. The inference unit 58 then selects a vector _{R of} one of the models in accordance with the residuals. This is then given to an assigned optimal steering law, which results in a commanded lateral acceleration a _MC . It may be that after the residuals, more than one model approximates reality quite well. In this case, it may happen that the inference unit 58 selects more than one vector _R and the selected vectors and the associated steering laws are weighted overlaid. As indicated in FIG. 4, the inference unit 58 can contain a neural network or a fuzzy-neuronal network.

Claims (7)

1. Steering structure for guiding a missile provided with a sensor system to a target, wherein a measurement vector ( z ) of observable quantities can be derived from the sensor system, characterized in that

• a) a model ( 31 ) of missile-target relative kinematics and sensor system ( 28 ) is shown in the missile,
• b) the model can be modified by an optimal filter means ( 44 ) which is acted upon by the deviation of a measurement vector () resulting from the model from the real measurement vector ( z ), the modification adapting the model ( 31 ) to reality is sought, and
• c) the missile can be steered as a function of a relative kinematics vector ( _R ) generated in the model and representing the relative kinematics of the missile and the target.

2. Steering structure according to claim 1, characterized in that the relative kinematics vector ( _R ) is applied to the autopilot ( 16 ) of the missile according to an optimal steering law ( 46 ). 3. Guiding structure according to claim 1 or 2, characterized in that the signals of an inertial measuring unit ( 22 ) of the missile are also applied to the model ( 31 ). 4. Guiding structure according to one of claims 1 to 3, characterized by estimating means ( 50 ) for estimating the target acceleration to which the measurement vector, image data from an image-resolving detector of the missile and data of the inertial measurement unit ( 22 ) of the missile are applied and by which the model ( 31 ) can be corrected. 5. Guiding structure according to one of claims 1 to 4, characterized by estimating means ( 54 ) for estimating the three-dimensional path of the target, to which the measurement vector, image data from an image-resolving detector of the missile and data from the inertial measurement unit ( 22 ) of the missile are applied and by which the model ( 31 ) can be corrected. 6. Steering structure according to claim 4 or 5, characterized in that the estimating means ( 50 , 54 ) are a suitably trained neural network. 7. Steering structure according to one of claims 1 to 6, characterized in that

• a) that the model ( 31 ) of the missile-target relative kinematics contains a plurality of hypotheses ( 34.1 , 34.2... 34. n) about the movement of the target.
• b) the model ( 31 ) in parallel for each of these hypotheses ( 34.1 , 34.2... 34. n) through an associated optimal filter ( 44.1 , 44.2... 44. n) in the sense of adapting the model ( 31 ) to reality can be modified,
• c) the estimated state vectors ( _R ) obtained and an output variable () of the associated optimal filter ( 44.1 , 44.2... 44. n) are connected to an inference unit ( 58 ) which selects one or more state vectors ( _R ),
• d) at least one selected state vector ( _R ) can be connected to an associated steering controller ( 46.1 , 46.2 ... 46. n) by the inference unit ( 58 ), which operates according to an associated optimal steering law.