Controller and method of controlling an apparatus

These This invention relates to control methods and devices and in particular electronic or computerized control devices, predictive electronic filters incorporate.

Lots Types of control system are known and in consumer and industrial machines and operations all types implemented. These control systems are based exclusively on a closed loop in which a control variable is sensed, as opposed to a nominal or setting value, an error signal derive and applied in response to the error signal Correction to hopefully drive the error to zero. The stability of the control system but primarily, and filter functions are implicitly or explicitly in the feedback paths included to stability maintain. about the years have many additional feedback and "feed forward" mechanisms developed the performance of control systems, in particular in its reaction rate, to improve. However, the need for stability limits these approaches in a well known manner a, in particular when the Filter functions, taking into account variations and permissible calculated and implemented variations in a number of conditions Need to become.

Known Systems are in <patcit><text>US 6181975</text></patcit>. <patcit><text>EP 1030231</text></patcit>. <patcit><text>EP 0334698</text></patcit> and <patcit><text>US 3013721</text></patcit> disclosed.

It it was recognized that many control systems measurable "sturgeon" factors affected around be independently measured from the main control variable can be. Some previous attempts have been made to the equalizer the noise include within the control system. However, these are in their Applicability is limited, as it on analytical or empirical models of the controlled Process is based, which are not always available or, more importantly, between samples, or over time are not stable. It would be desirable, provide a control system, that the correlation between can watch various signals and automatically learn the Device as a result controlling better.

the The goal of all electronic filters is the desired signal called from all other signal components, noise to separate. Predictive adaptive filters are also known and exploit the fact, that a signal in contrast to the change of the additional noise usually changes slowly. This is due to the fact that the noise all frequencies includes while signals predominantly low frequencies include. At first, a kind of adaptive Remove the predefined setting, but this setting will then continuously adapt to the changing signal to strive to the noise in an optimal manner to eliminate. This is accomplished by comparing the temporarily incoming signal with the signal of the immediate past achieved by means of a built-in itself mechanism as an autocorrelation can be mapped. Such filters react on changes when the signal to accordingly adjust the filter settings. As a consequence such filters are able to produce a shape of the input signal with decreasing Prädiktionsverlässlichkeit to increase temporary (Predictive) intervals the immediate future to foresee (extrapolate).

A Class adaptive predictive Filters are the filters of the Kalman type, which are capable of using recursive estimation adapt to the characteristics of an input signal. Through a such adjustment mechanism these filters remain optimally its corresponding object matched. Due to the extent to the analog electronic signals are disturbed by noise, have such predictive adaptive filter a comprehensive application domain analogous in all areas electronic signal processing, for example, telecommunications, Broadcasting, Radar signal processing, and more, on. This filter is based However, on the self-similarity of the input signal (that is, they are optimized to particular characteristics of the expected Signal to respond). Their goal is to provide a maximum amount to retain on information from the input signal. Thus, the Output usually only an improved version of the original signal directly is derived from the input. Accordingly, the Kalman filter provides no solution the problems of complex control systems, although of itself Interest.

The Invention provides a control unit prepared for controlling a physical apparatus, the controller first Input means for receiving a primary input signal representing a measured state of the apparatus is, and signal processing means, on the primary Input signal responsive to a control signal for influencing the State to produce the device, so that they desired a State maintains, includes, wherein the control device at least one further input for a having signal additional Measurements of the device or its environment is, the Signal processing means correcting means for entrapping Corrections in said control signal in response to the additional Measurement signal and means for automatically conditioning the reaction of the correcting means in response to temporal cross-correlation including, between the additional Measurement signal and the control signal during operation of the control unit and Device is observed together, is observed.

the Conditioning agent may comprise means for adjusting a gain and a frequency response of a signal path in the correcting means include. Such an adjustment is similar to that in a Kalman filter is implemented.

In a preferred embodiment, However, the correcting means includes means for filtering the additional Signal by means of a plurality of filters, the various fixed have impulse responses and means for forming a weighted Sum of the differently filtered signals to the correction to be applied derive, includes wherein the conditioning means for Adjusting the weightings of the various signaling pathways in response to their respective observed correlation with the control signal includes.

The Inventors describe the corrective and the conditioning in this case, as a cross-modal predictive filter (CMP filter). The corrective and conditioning has in this case also similarities with the well-known "neutral Network "circuits , where between desired outputs and a set of inputs of a number of sample patterns, a cross correlation is learned. While the individual input signals coming from respective sensors However, as in the neutral network, they are in this case signals from the same sensor, but subject to different filter characteristics. By learning the pattern of correlation between these signals and the control signal, the novel controller effectively learns the temporary Correlation between the additional measurement signal and the control signal.

the Conditioning agent may be arranged to determine the correlation by Multiplying the signal of each path with a derivative of Control signal to observe and the product of the signals over time integrated to derive the weighting. Integration may lossy or not lossy, depending on if the conditioning will be kept in the absence of an incentive is or may be forfeited. Integration must not lossy be for the CMP filter is able to temporary changes of interference adapt.

the control unit a plurality of additional inputs exhibit, each with associated correcting means and conditioning means within the signal processing means.

In Embodiments, wherein the correcting means comprises a plurality of filters of different having impulse responses, wherein the correcting means for each additional Signal not all have the same number of filters, still the same must have set of impulse responses. The number of filter paths can are thus reduced if the designer of an idea has scope of the expected correlation.

the control unit may be arranged to generate multiple control signals based on at least some of the same input signals, wherein the signal processing means correcting means and conditioning includes generating each control signal.

The multiple correcting means share filter components.

In Embodiments, which are arranged to generate multiple control signals, can a first control signal to be connected to additional as an Measurement input to the correcting means for generating a second control signal to serve. A derivative or other transformation the application can reasonably be applied to each of the measurement inputs will.

The Invention further provides a method for controlling a physical Device willing to said method on a consistent basis Includes: <ul><li>- Receiving a primary input signal, representing a measured state of the apparatus;</li><li>- generating a control signal for influencing the state of the apparatus in response to the primare Input signal, so that it maintains a desired state; and</li><li>- receiving at least one further input for a signal representing additional represents measurements of the device or its environment;</li><li>- Including Corrections in said control signal in response to the additional Measurement signal; and</li><li>- automatic Conditioning the response of said correcting means in response to a temporary cross-correlation, between the additional Measurement signal and the control signal during operation of the control unit and Device is observed together, is observed.</li></ul>

Of the correcting step may comprise filtering the additional signal by a Multitude of filter functions, the different fixed impulse responses have, and forming a weighted sum of the different filtered signals to derive the correction to be applied include, wherein the conditioning step, setting the weights of the different signaling pathways in response to their respective observed Correlation with the control signal contains.

Of the Conditioning step can begin with a zero condition, or may with a set of conditions in another device were learned, begin.

The Invention further provides a control device of the kind set forth above ready, said correcting means by a stationary with a Reaction, but one that the by another control unit, is was learned stationary reaction during operation, transmitted, replaced. Whether this is practical, is the repeatability of the environment and the characteristics of the device depend under control.

A Another alternative is, as an initial condition Learned of a control unit in another control device transmit, the other controller capable the reaction of the correcting means during its further operation adjust.

SHORT DESCRIPTION THE DRAWINGS

The embodiments the invention will now by way of example only and with reference described to the accompanying drawings, in which:

<figref idrefs="S38">1</figref> on generalized block diagram of a conventional chemical treatment system, the typical control system;

<figref idrefs="S38">2</figref> schematically the values ​​of certain signals over time shows in a typical control system, responsive to a reactive manner to an external disturbance;

<figref idrefs="S39">3</figref> the chemical treatment plant from <figref idrefs="S38">1</figref> shows, The modified with a CMP filter based on a control unit is that the control of steam on based on both the input and the output parameters of the system goes into the plant, accomplished becomes;

<figref idrefs="S39">4</figref> the Signals in which <figref idrefs="S38">2</figref> correspond, in a control system using a cross-modal predictive filter shows the responds directly to the same external disturbance after a learning phase;

<figref idrefs="S40">5</figref> on schematic block diagram of a generalized process control system is that according to a system based on a CMP filter controller a generalized embodiment the present invention incorporated;

<figref idrefs="S41">6</figref> a multistage CMP filters within the control system of <figref idrefs="S40">5</figref> detailed shows;

<figref idrefs="S42">7</figref> response curves a set of resonators in a CMP filter of the in <figref idrefs="S41">6</figref> shown Type shows, when presented with a square pulse;

<figref idrefs="S40">8</figref> a Gain control circuit (GCC), consisting with each resonator in the CMP filter <figref idrefs="S41">6</figref> connected is showing in more detail;

<figref idrefs="S42">9</figref> a Relationship between a single output signal u_i of Resonator, a control signal v driving the process and its Derivative v 'under consideration time shows;

<figref idrefs="S43">10</figref> on schematic block diagram of an alternative form of CMP filter is, each resonator having a variable frequency response;

<figref idrefs="S44">11</figref> on schematic block circuit diagram for a robot control system is that a more specific embodiment, forms of the present invention;

<figref idrefs="S45">12</figref> a Simulation of the path of the robot under control of the circuit of <figref idrefs="S44">11</figref> is, if it encounters obstacles in an environment over a period; and

<figref idrefs="S45">13</figref> the Learning of the optimal resonator Torge Klobuk for a fault signal within the robot circuit while the period of time in <figref idrefs="S45">12</figref> is shown, illustrating.

DETAILED DESCRIPTION THE EMBODIMENTS

These The invention is not specifically of an embodiment for either analog or digital hardware or software limited. The following is valid in general for each implementation, which specific differences, if necessary, be highlighted. The mixture of analogue and digital circuits, and of hardware and software that is used in a specified application, familiar to many to the expert Factors. Such factors include the requisite for the system Bandwidth, the relative development costs of the various solutions and the expected volume of production.

The following descriptions of classical mechanical and chemical describe treatment plant control problems, such as the use kreuzmodaler predictive Filter can be used to provide an improved multi-modal control system provide that from experience the correlations between multiple inputs and the desired Corrections learned, in particular in the time domain.

In a typical mechanical system, the controller and a mechanical actuator includes the position of the actuator is change due to internally generated and external (disturbing) forces. In the classical control paradigm the actuator position is measured by a position sensor, from the the signal by means of the scheme to compensate for the internal forces, which determine the mechanical motion pattern of the actuator. If an external force (disturbance) compresses the position, the control unit generates a signal to compensate. The combined inertia of the actuator and control unit but induces a delay wherein the compensation, which is used to this disorder, the while of compensation (transition) to a -Zeitraums significant position error results, counteract.

The same problems occur mainly generally in industrial processes, For example, the chemical industry, on.

<figref idrefs="S38">1</figref> provides a typical process <figref>102</figref> represents the heat-driven by means of a Reaction, a starting material I in the product P transformed. As in <figref idrefs="S38">1</figref> shown, the starting material typically occurs in the process at a temperature T1. The optimum output is reached at T2. The complete system<figref>100</figref> includes Control systems for maintaining the various parameters on their optimum level to ensure process reliability. The temperature T2 is fed back as a signal FB and used as a prescribed point in a conventional scheme, to the flow velocity <figref>104</figref> about the Valve <figref>106</figref> of the steam S that heats the plant to control. The starting temperature T2 will change, however, if T1 due to a malfunction, which will lead to a suboptimal situation fluctuates. Known Method for overcoming these problems have a heuristic additive effect of T1 to the input side of the feedback loop. This can however only achieved if the relationship between the input and output variables a deterministic manner are known. There were complex developed systems that address these problems, but, as described later, have these disadvantages in terms of complexity, requirement known models and instability the loop.

<figref idrefs="S38">2</figref> provides a graphical representation that schematically values ​​requested, real and error values ​​(T_D, T2 and Err) while the transition period the chemical process control system to a step change (Disorder) shows in the input temperature T1. T_D provides the setting or the requested value for T2 is. One can by the response time of T2 to the variation in T1 triggered delay and the time required for the system, to the change by changing the flow rate FR of the steam S to T2 to the requested value T_D return, adapt, watching. While the transition period are some of the processing paramters not optimal in their SETTINGSpayments are, the quality of the heat-driven reaction and their resulting output be affected.

<figref idrefs="S39">3</figref> shows a modified control system, wherein a control device <figref>200</figref> not only the output signal T2, but also the "fault" signal T1 receives. The exact causal relationship between the variations of T1 and the output signal T2 is unknown when the system constructed and is set. The new control unit<figref>200</figref> learned but by an adaptive process to an initially unknown Change of T1 (disorder) with the delayed change from T2 to correlate. The exact implementation of which is here not described, but is at the bottom of related <figref idrefs="S40">5</figref> to <figref idrefs="S43">10</figref> described. As soon as the control unit the disturbance has adapted, at the moment in which a change in T1 is observed, the valve <figref>106</figref> adjusted by the signal FB to the flow rate FR on a präkompensierte to change settings. Thus, the system is <figref>100</figref> be able directly to the disorder to respond in T1, where it anticipates the resulting effect on T2, and consequently the error in T2 (and its negative effect weakened significantly in the process), as in <figref idrefs="S39">4</figref> shown.

at the simple example of <figref idrefs="S38">1</figref> and <figref idrefs="S39">3</figref> is the number of sizes, be controlled and measured, very small. with general Words can be constructed, the novel type of controller for each compare number of different input signals and by to learn an adaptive process, the output by observing the cross-correlation characteristics between the input and the output signal foresee. In particular, the control unit can from his observations learn which of the many input signals relevant to the foreknowledge the output signal is, and in what manner. embodiments the invention include a cross-modal predictive (CMP) filter is built into the control loop, which receives multiple inputs. After a learning period, the CMP filter an output event (output) by responding to the earliest occurring relevant input consistent the associated output precedes foresee.

The Construction of the novel controller, both in general is and in a specific example below with reference to <figref idrefs="S40">5</figref> to <figref idrefs="S45">13</figref> described. By applying the novel control unit first to the previous Example of a classical mechanical problem would CMP filter the external forces (Disturbances) measure and during the adaptive process temporarily with the (much later occurring) position change correlate. Once the filter has adapted to the disturbances, a position correction signal (a drag) are generated once an external force occurs, without having to wait until a delaying Position error is detected at the output. Thus, the system is be able immediately to the application of an external force to react, and the position error will be considerably attenuated.

<figref idrefs="S40">5</figref> shows a system based on a CMP filter controller <figref>200</figref> in a typical Configuration of a closed loop, that is with a Output demand value v, which in the operating position of the controlled Device, here by the generalized system <figref>100</figref> shown, is fed and set. A reference signal P which is the current actual state of the plant <figref>100</figref> represents, is of a desired (Which may be zero) (adjusted) value SV subtracted <figref>202</figref>. a Differences or error term X_R provide, the in the control unit fed back is where it is used to define the output demand value, whereby the loop "closed" is. To the multimodal Skills of the controller <figref>200</figref> evaluate, is also a number of external fault inputs X₁ ... X_N fed into the control unit. These will from sensors throughout the plant <figref>100</figref> derived and include Voltage-time functions of an arbitrary shape.

The investment <figref>100</figref> must according to the actual Application domain for example, a power position transformation, as in the earlier mechanical described example, or a steam heating device, as in the former Example of chemical Aufbreitungsanlage described specified will. The skilled artisan will readily recognize that the principles the new controller in a large Selection of "physical" devices and Systems, applicable from classic machines to economic systems, are.

<figref idrefs="S41">6</figref> shows a processing control system, a very simple on a CMP filter-based controller additionally to the conventional Error signal X_R used. The attachment<figref>600</figref> is shown with the conventional Result signal P and also two disorders X₁. X₂ emits. The number of faults is not limited to two. Although these are shown, in the widest sense as known from Conditioning come, they are for the purpose of conventional Ambient measurements of control systems, their influence on the process is not known exactly meant. The Subtraktionschaltung<figref>602</figref> receives the EinstellwertsiSV signal and forwards the conventional error signal X_R from. The feedback filter<figref>604</figref> (With the transfer function F = h_R × ρ_R) generates a Steuersignalkontribution u_R. the above the summing circuit <figref>606</figref> the control value generated v. This Away <figref>600</figref>. <figref>602</figref>. <figref>604</figref>Forms the proportional Standard term reference loop of a control system and ensures no further description. The summing circuit<figref>606</figref> allowed However, many additional Contributions to the tax base of conventional v together with the Term next u_R to determine each from a fault "channel" on one of the interference signals X₁, X₂ etc. responds.

Each fault channel includes a resonator <figref>611</figref>-<figref>615</figref> having fixed Transfer function h_i (Fixed impulse response) which accordingly a corresponding filtered signal u_i generated. Each signal u_i leads by a variable gain block <figref>621</figref>-<figref>625</figref> With ρ the gain_i. this to the (positive or negative) strength of contribution Channel in the summing circuit <figref>606</figref> define. Resonators<figref>611</figref>-<figref>615</figref> are se of a well-known form and include bandpass filter, as LRC-circuits (Inductor, Resistor, capacitor) in an analog circuit or as IIR (infinite impulse response) or FIR filters (filter be implemented with finite impulse response) in a digital circuit can. Each input disturbance X_i is at least one resonator performed. The number of channels, with each fault signal are connected, is not fixed and is determined by the familiar figure the input waveform and the desired response by the controller on it, certainly. In the illustrated example feeds the signal X₁ the M channels, while signal X₂ the N-channels fed.

On Differentiator <figref>630</figref>Which v receives the control output signal, and individual gain control circuits (GCC) <figref>631</figref>-<figref>635</figref>. which each reinforcement a respective gain block <figref>621</figref>-<figref>625</figref> control, to complete the CMP filter section of the in <figref idrefs="S41">6</figref> shown Control device. These GCC <figref>631</figref>-<figref>635</figref> put the learning mechanism of the CMP filter ready and be down with in reference to <figref idrefs="S40">8</figref> described in more detail.

<figref idrefs="S42">7</figref> shows, how the resonators Torre Actions (u₀, u₁, u₂ etc.) differ, using the example of a square pulse function as Input X. For each channel are the characteristics of its associated resonator <figref>611</figref> etc. selected a unique response to each input perturbation provide. Your natural frequencies logarithmically (for example, f₀, f₀/ 2, f₀/ 4, etc.) to proceed.

the reader should now realize that by providing an appropriate set to transfer functions h_i for each fault signal any desired Reaction in the summing circuit <figref>606</figref> by varying the relative gains ρ_i synthesized can be. This assumes, of course an infinite number of channels advance, which is not practical, and assumes that the appropriate can be identified reaction. Typically, however, would for a adequate approach less than ten resonators, possibly five or less, for each Input required. Furthermore, set the gain control circuits<figref>631</figref>-<figref>635</figref> by an adaptive process in the course of operation automatically the weights fixed to the optimal response without advanced Knowing the desired To achieve transfer function, as will now be described.

<figref idrefs="S40">8</figref> shows one of the gain control circuits <figref>631</figref>-<figref>635</figref> detail, which together form the "learning" mechanism of controller provide. Each unit<figref>631</figref>-<figref>635</figref> in this embodiment provides a Vervielfacherfunktion <figref>800</figref> and an integrator function <figref>802</figref> ready. The differentiator <figref>630</figref>Here clarity again shown, the derivative v 'of Manipulated variable v ready as they continuously through the summing circuit <figref>606</figref> output is. As is well known, is the derivative term an advanced 90 ° phase version the control signal ready and can theoretically as a predictor of this be considered signal. The differentiator<figref>630</figref> can by using well known technology in analogue form a differentiator or digitally by subtracting successive Samples of the signal or implemented by a more sophisticated FIR or IIR filter function will. The multiplier<figref>800</figref> multiplies the resonator output u_i, A damping factor μ (typically a small fraction, to avoid instability) and the derivative term v 'to each other, a measure of the Correlation between the resonator output and the derivative v 'derive. This correlation Δρ_i becomes continuously (either as an analogue signal or in a digital System is produced as a stream of discrete sample values) and over time by the integrator function <figref>802</figref> integrated, ρ to the actual gain_i of each gain block <figref>621</figref>-<figref>625</figref> etc. adjust. In a digital embodiment would Integrator feature a simple numeric battery include.

In Combination provide the components described for this purpose a control unit ready the (during the Operation in a new environment) the relationship between input and disturbances their effect "learned" in the process, so that the reaction through the process to disturbances anticipated and more by an anticipatory manner as may be offset by a purely reactive manner. The operation of the CMP filter can be obtained by the following mathematical Equations are described.

Of the V output of the CMP filter is given as follows: <st32:df xmlns:st32="http://lighthouseip.com/">ν (t) = ρ<st32:sub>R</st32:sub>u<st32:sub>R</st32:sub>(T) + Σ<st32:frac>N<st32:over>i = 1</st32:over></st32:frac> ρ<st32:sub>i</st32:sub>u<st32:sub>i</st32:sub>(T) (1)</st32:df>in which u is given as follows: <st32:df xmlns:st32="http://lighthouseip.com/">u<st32:sub>i</st32:sub>(T) = x<st32:sub>i</st32:sub>(T) ⊗ h<st32:sub>i</st32:sub>(T) (2)</st32:df>in which X is convolved with h. The functions h are the transfer functions the resonators <figref>611</figref> etc., which are given as follows: <img img-content="mf" img-format="tif" he="9" wi="79" file="00190001.tif" />wherein H (s) the resonator in Laplacedarstellung as usual with two complex / complex-conjugating Parameters p and p * describes, by p = a + ib and p * = a - ib given are with: <img img-content="mf" img-format="tif" he="9" wi="81" file="00200001.tif" />where f is the resonant frequency the relevant resonators <figref>611</figref>-<figref>615</figref> and Q their damping factor is. (The value of Q is about with the number of oscillations a resonator in response a δ-function input will carry out the same.) For the purpose of this application would preferably Q about are 1, for example in the range of 0.5-1.2. As discussed earlier, hanging the Frequency response of the response time from that in the control unit of a real application is required. If a CMP filter within a electronic control loop is used, f can be very high be (kilohertz to megahertz) when in a mechanical system is used, f will normally be in the range of 1-100 Hertz, when used in chemical control situations f can in the range of Millihertz or even lower.

Of the Process of changing the reinforcements, with each disturbance term X_i are connected, is a process of adaptation by a simple variety of neutral learning. Each gain fixing the variable gain block <figref>621</figref>-<figref>625</figref> becomes by its gain control circuit (GCC) <figref>631</figref>-<figref>635</figref> modified. All gain control circuits <figref>631</figref>-<figref>635</figref> are identical (but receive different inputs). On a continuous Base is ρ the gain setting_i each Gain control circuit <figref>631</figref>-<figref>635</figref>. ρ with the exception of the reference gain_R. by the addition of small (positive or negative) values ​​Δρ_i modified according to: <st32:df xmlns:st32="http://lighthouseip.com/">ρ<st32:sub>i</st32:sub> → ρ<st32:sub>i</st32:sub> + Δρ<st32:sub>i</st32:sub> 5 (a)</st32:df><st32:df xmlns:st32="http://lighthouseip.com/">Δρ<st32:sub>i</st32:sub>(T) = microIU<st32:sub>i</st32:sub>(T) ν '(t) 5 (b)</st32:df>where μ is a small Number is typically in the range from 0.000001 to 0.1, which an attenuation factor is, of all the variable gain block <figref>621</figref>-<figref>625</figref> applied is to rapid gain changes to prevent and v '(t) the temporary Derivative of v (t) is obtained by the differentiator <figref>630</figref> calculated is a signal in direct proportion and polarity to the rate of change produces its input signal. On this basis, v ', when the control word v falls, negative and according to equation (5b) are all Verstärkungsmodifizierer the gain block Δρ for a positive u_i be negative. why are the gain values fall, and the influence of the disturbance signals X_i will be reducing. However, this is only then when the other input to equation (5b), the Resonatorbegriff, is positive. The disturbance signal may be negative, which would eliminate the negative derivative signal, wherein a positive influence is produced. polarity the disturbance sensor signals and the signals within the CMP filter are assigned to the influencing control loop in a converging manner, otherwise would positive feedback occur, the instability cause the loop would.

<figref idrefs="S42">9</figref> provides Waveforms prepared the relationship between v, v 'and u_i in show a case in which the special function u_i Good is correlated with a peak in the derivative v '.

There the differentiator only a concept in which v is in operation, he does not "physically" in each gain control circuit <figref>631</figref>-<figref>635</figref> present to may be, and as a common means <figref>630</figref>, as in <figref idrefs="S41">6</figref> shown, exist to obtain a result in all Gain control circuits <figref>631</figref>-<figref>635</figref> the channels is fed. In alternative implementations it may be convenient, the differentiation step with the individual GCC functions to combine. In other cases , the derivative function already somewhere in the control system be implicit and an explicit step would be omitted.

Of the Learning process in a work based on a CMP filter closed Control system is a convergent process that stop by itself is (what infinite growth of the gains ρ prevented), which is at the optimum values ​​for the various disorders settles. The reason for this is that, while the process of adapting to one or more disorders X_i, To adjust the manipulated variable of the process is to the interference X_i counteract, and thus their weaken relative effect. As a result, the effect of the disturbance X_i gradually from the process removes, and / the gain value (s) ρ_i for each fault reaction u_i will settle at its optimum value. The adaptation will continue when a change in the disturbance occurs, which do not fully by the current Setting its associated gain value is compensated. The damping factor μ not (which for every disorder must be the same) is effectively the number of coincidences a that have to be observed, to constitute a definite correlation.

<figref idrefs="S43">10</figref> shows a block diagram of a modified CMP filter, whereby in place of of multiple fixed resonators, one or more resonators Torre action control circuits (RRCC) <figref>1000</figref> the frequency responses within resonators <figref>1010</figref>. <figref>1020</figref> With variable response for the disturbance X₁ set to. This is primary the equalizing including the resonance frequency, but in principle could the Q factor can also be adjusted. The fault channel for the disorder X₂, The variable resonators includes, the variable gain block <figref>623</figref> to <figref>625</figref> and the gain control circuits <figref>633</figref> to <figref>635</figref> are shown not as detailed. The rest of the components of the in<figref idrefs="S43">10</figref> shown CMP-filter as about the conventional block <figref>604</figref>, The summing <figref>606</figref> and the differentiator <figref>630</figref> can in the same way and Manner as for the CMP-filter <figref idrefs="S41">6</figref> be in operation and ensure therefore no further description.

the Using resonators with variable reactions <figref>1010</figref>. <figref>1020</figref> provides the advantage of reducing the number of resonators provided that are necessary to the interference X_i v to correlate with the process control concept. Instead of a large Amount, such as ten, requiring at resonators per disorder, may be Using preferably no more than two resonators with variable response per fault the same effect can be obtained, whereby a drawback is that in addition to the increased circuit complexity the learning process is slightly stretched, as the variable resonators also need to set, to the interference correspond to. It must be ensured also that stability of the control loop to preserve.

Each Resona Torre action control circuit (RRCC) <figref>1000</figref> is working A display device of the correlation success for each disorder: in the following manner the magnitude the weight concept each gain control circuit (GCC) <figref>631</figref>. <figref>632</figref>, These are in cooperation with the output from the resonator u_i each channel and the derivative of the process term v 'used to adjust the frequency response of each resonator <figref>1010</figref>. <figref>1020</figref> variable response to modify the order the weight transmission times on the optimal correlation of the disturbance in Referring to process term v maximize. If more than one resonator<figref>1010</figref>. <figref>1020</figref> per fault signal X_i is used, as in the example of <figref idrefs="S43">10</figref> shown, then evaluated the RRCC (<figref>1000</figref>) All the relevant signals, with the disorder are connected to the frequency response of all resonators with this particular disorder are connected to adjust. are Different reactions for the Resonators with each disorder are connected, selected, the correlation windows during the initial Periods to maximize learning, which could cause the window, when learning progresses and reaches optimum settings are.

<figref idrefs="S44">11</figref> presents a practical example of a control device for a moving robot <figref>1100</figref>. the functionality this invention to demonstrate. Used for Environmental Sensors the robot bump sensors <figref>1102 (FL)</figref>. <figref>1102 (FR)</figref>. <figref>1102 (BL)</figref><figref>1102 (BR)</figref>. one in each corner, and three visual range sensors, an advanced sensor <figref>1104 (F)</figref> and the other two in the front corner sighted sensors <figref>1106 (FL)</figref>. <figref>1106 (FR)</figref>, The suffixes of sensors L, R, F, B, used individually or in combination be featuring "left", "right", "forward" or "back".

In this example a more complex CMP filter based on a control unit adopted in the robot system, followed by two control systems (Φ), (S) are who do not live in isolation mode are, but close to each other are coupled. This involves a control loop of the drive control causes the output S (speed - within reach causes forwards / stop / backwards), and a control circuit of the steering operation, the output Φ (within reach the left / straight / right).

the Control system includes similar Components such as for the preceding examples: for any visual (range) disorder Corresponding banks of resonators <figref>1108</figref> etc. available (In this particular example are fixed responses available, could but be variable responses that use less weights) for every physical (unevenness) disorder Corresponding filters <figref>1110 (L)</figref>. <figref>1110 (R)</figref>. <figref>1110 (F)</figref>. <figref>1110 (B)</figref> With fixed responses, gain control circuits <figref>1112 (φ)</figref>. <figref>1112 (s)</figref> etc. and Summierungseinrichtungen <figref>1114 (φ)</figref>. <figref>1114 (s)</figref> available. The control unit also includes an additional Exhaust block f ' <figref>1116</figref> and a biasing means <figref>1118</figref>Whose operation described later is. The variable gain block<figref>621</figref> out <figref idrefs="S41">6</figref> exist where Ro botersteuergerät, Depicted as a large arrow of the bus Weighted outputs each resonator bank <figref>1108</figref> etc. crosses when the Summierungseinrichtungen <figref>1114 (φ)</figref>. <figref>1114 (s)</figref> entered will. Bandpass filter<figref>1110 (L)</figref>. <figref>1110 (R)</figref>. <figref>1110 (F)</figref>. <figref>1110 (B)</figref> are analogous to the block <figref>604</figref> with fixed reaction from <figref idrefs="S41">6</figref>. the X_R process. Resonators<figref>1108</figref> etc, GCC <figref>1112 (φ)</figref>. <figref>1112 (S)</figref> etc. and the gain block are analogous to the resonators <figref>611</figref> etc., GCC <figref>631</figref> etc. and variable gain block <figref>621</figref> etc. out <figref idrefs="S41">6</figref>The interference signals X₁, X₂ etc. process. Three representative channels with the reactions f, f / 2, f / N are shown for each disorder.

It it is readily apparent that the "circuit" shown entirely in digital form or entirely in analog form, in any mix of these, and in any mixture , as appropriate, implemented by hardware and software could.

While his overcoming initial from any obstacle, the robot <figref>1100</figref> initially its bump sensors <figref>1102</figref> use, to the control system using a feedback control and the steering, to overcome every obstacle on the to inform limits within which it is operating. The visual range sensors <figref>1104 (F)</figref>. <figref>1106 (FL)</figref>. <figref>1106 (FR)</figref> inform the control system potentially over an imminent collision, but the control system is initially not the connection between the visual stimuli and a Bumpsignal have learned that temporarily somewhat later occurs. Once the CMP filter has learned that a relationship exists between the two, the robot control system be able to control the way that S with the drive and especially the steering Φ by direct influence of the visual sensors <figref>1104 (F)</figref>. <figref>1106 (FL)</figref>. <figref>1106 (FR)</figref> connected have to be modified, wherein the robot <figref>1100</figref> of a Obstacle is diverted away before it hits it.

the Signal from each Bumpsensor <figref>1102 (FL)</figref>. <figref>1102 (FR)</figref>. <figref>1102 (BL)</figref>. <figref>1102 (BR)</figref> becomes supplied in both loops. The front bump sensors<figref>1102 (FL)</figref>. <figref>1102 (FR)</figref> are for the Control of conversion speed along as a negative signal in the speed (s) -Regelkreis coupled, and the rear bump sensors <figref>1102 (BL)</figref>. <figref>1102 (BR)</figref> are for the Controlling the speed of forward together as a positive coupled signal in the speed (s) -Regelkreis. An additional biasing means <figref>1118</figref> is fed into the same speed (s) -Regelkreis, to set a non-zero speed, so that a not be irritated robot at a constant speed by moved forward until it receives a signal from an environmental sensor <figref>1102 (FL)</figref>. <figref>1102 (FR)</figref>. <figref>1102 (BL)</figref>. <figref>1102 (BR)</figref>. <figref>1104 (F)</figref>. <figref>1106 (FL)</figref>. <figref>1106 (FR)</figref> receives. On the same manner as for the speed control are the right bump sensors <figref>1102 (FR)</figref>. <figref>1102 (BR)</figref> for the left Directional control together as a negative signal in the steering (Φ) -Regelkreis coupled, and the bump sensors <figref>1102 (FL)</figref>. <figref>1102 (BL)</figref> are for the right direction control together as a positive signal in the Steering (Φ) -Regelkreis coupled. A not being irritated robot will move straight, until it receives a signal from an environment sensor. It will be appreciated, that a robot should do a real job, additional Incentives obtained, to desired changes display in the speed and the direction which the control device in combination will affect the sensor inputs shown here. In the present application is the resonators for Bumpsignale assigned a relatively low Q, such as 0.6, to provide high damping and "overshoot" in their reaction (According to an unevenness of the robot should turn around and not before and back oscillate) to avoid.

The sighted forward, left and right range sensors <figref>1104 (F)</figref>. <figref>1106 (FL)</figref>. <figref>1106 (FR)</figref> be in their own associated banks resonators <figref>1108</figref> fed, their outputs in both Control loops s, dine Φ. After all is the control loop cross-coupling by taking each output control variable S and Φ (from the summation) achieved, and by feeding them into a bank of resonators dine in the other loop. In the case where the speed control the steering (Φ) control influenced, is its derivative f 'taken to the offset of the constant Geschwindigkeitsvorspannung <figref>1112</figref> to remove. Without control loop crosstalk would the CMP filter incapable be a change the speed with a change the direction and a change the direction with a change to set the speed in relationship. This ensures that the robot when it reverses, modifies its steering system in order a narrow bow with which he collided straight away from an obstacle, is to go through until the visual sensor indicates that he the Obstacle no longer sees, and to also ensure that the velocity is reduced when steering narrow arches are performed.

Ideally, the visual sensor is always able be, the robot obstacles deflect, and the bump sensors would therefore never be activated, but in reality, the robot is still the have potential to get into situations that it is not can be deflected out. In these cases the bump sensors are<figref>1102</figref> always still be required, but more as a secondary sensor, rather than the primary sensor, they were at the beginning of the learning process. Some typical situations could the be down in a darkened dead end or when reversing and abut an obstacle (there are no reversing visual sensors in this example exists, although the skilled reader will appreciate, that this achieved with the addition of other sensors and control maturities could be).

Around Obstacles to overcome, would the robot traditionally different parameters have to be taught the relate to the environment. The present invention needs to be simply configured with control loop parameters make sure that the control of the robot is smooth, such as resonators Torre actions tailored to the visual sensor outputs are. None of the parameters is ambient, but with the embodiment the robot connected. They are only at the point of the configuration the robot fixed and need no further adjustment. In contrast, a traditional robot would new parameters associated with every new environment in he must be in operation, need to be taught. The new controller would a Robots allow to be placed in different environments, without requiring any adjustment, and also assumes its reaction, when the mechanism of the robot wears or the nature of the environment changes.

Among with reference to <figref idrefs="S45">12</figref> you can see how the robot (Here on a computer simulated) a "maze" full of obstacles passes, wherein he starts at point "0" and "11000" (time steps arbitrary Value) ends. When the robot overcomes obstacles to face the CMP filter to a relationship between the visual feel, steering provide and speed. One can see that the number gradually collisions is reduced to the point at which the robot is effected, to traverse long distances without any collisions, especially of the upper left corner (time steps 7900-8900) to the end point. little one can steering corrections be seen, where the visual sensor an imminent collision has shown that causes that the steering committee in advance is established, wherein the robot is directed away from the obstacle. In the lower left corner, where the result of learning is most evident, the path which leads to the outside is with OUT in the back and, the later is located in time, is labeled RTN. On closer inspection one can see that the robot according to the in the upper left corner spent the period 7900-8900 is significantly more capable of steering without bumping into obstacles. This is due to the learning that while the control system of this Period has been reached, because the robot had come up against a large number of obstacles to able to be herauszulenken from the corner.

Among with reference to <figref idrefs="S45">13</figref> one can observe that the rate of change the weights of the resonator during the period 7900-8900 of Robot in the corner significantly higher than at any other time during his journey. This shows ρ a set of ten weights_i for ten resonators with the left range sensor <figref>1106 (FL)</figref> connected are if it affects the steering control variable Φ. Considering all sensors and the cross-correlation paths between Φ and S is recognized that the example circuit as a whole 5 × 50 = 50 resonators <figref>1108</figref> and individual 8 × 10 = 80 gain control circuits (GCC) <figref>1112 (φ)</figref>. <figref>1112 (s)</figref> use etc. can. Man can observe also that the weights continue to change slowly to step 9000, even if no collisions available. This is the learned mutual influence of outputs from exclusively the visual sensors, the weightings to their optimum Settings can be adjusted.

The Sensors that are used to obtain ambient feel, might well be well provided, being different sensor technologies, such as ultrasonic range sensors, visual image processing sensors, Radiation sensors, electromagnetic field sensors, radar sensors, etc. use, the method by which the robot's environment feels, not essential for the novel Invention.

The Methodology, which is used to obtain the robot control, could as well be used in other applications, such as a chemical plant where, for example, the disturbance signals of pressure and temperature are and the control concepts flow rate and thermal control are.

the The method by which the CMP filter based on a robot achieves optimized weights associated with each disturbance is not limited to a "learning" process. The "learning" can from another reached robot and the "learned" parameters as a passed initial condition to an "unskilled" robot have been, so that the "unskilled" robots overcome immediately a "maze" of obstacles can. The robot of the CMP-based filter may be capable further to assume the weights associated with each disorder are as time progresses to allow the robot to the changes such as a different environment or physical degradation align, or may fixed weights as a cost-effective solution at the expense of loss of adaptability to changing use environments or physical degradation of the robot.

Of the expert Readers will appreciate that numerous variations within the Principles of the apparatus described above are possible. More or less strategies can be tested, and variations of each can within the specified Rules are provided.

Accordingly, understands it found that the embodiments described herein for the purpose of understanding shown as an example set forth and are not intended the scope of the claimed restrict the invention.