CONTROL METHOD AND CONTROL DEVICE WITH ANOMALY DETECTION

Abstract

A method is provided for controlling a technical system by means of a Two Degree of Freedom controller, which allows for increased accuracy, higher robustness and better safety even in cases of changes of the technical system. The method comprises deriving an anomaly detection signal at an anomaly detection time point during control of the technical system. At least one control parameter parametrizing the Two Degree of Freedom controller is adapted depending on the anomaly detection signal and, from the anomaly detection time point onwards, the technical system is controlled by means of the Two Degree of Freedom controller parameterized by said at least one adapted control parameter.

Claims

1. A method for controlling an output of a technical system to a predefined setpoint by a Two Degree of Freedom controller, wherein the Two Degree of Freedom Controller comprises: a feedforward part that provides a first control output; and a feedback part that provides a second control output, wherein the Two Degree of Freedom controller is parametrized by a plurality of control parameters, and wherein the method comprises: computing, by the feedback part, the second control output from a control error representative of a deviation between the output of the technical system and the predefined setpoint; determining, from the first control output and the second control output, a sum-control-variable acting on the technical system to control the output to the setpoint; deriving, at an anomaly detection time point during control of the output of the technical system, an anomaly detection signal from the control error present at the anomaly detection time point; adapting at least one control parameter of the plurality of control parameters parametrizing the Two Degree of Freedom controller, based on the anomaly detection signal; and controlling, from the anomaly detection time point onwards, the output of the technical system by the Two Degree of Freedom controller parameterized by the at least one control parameter.

2. The method according to claim 1, wherein the feedforward path uses an inverse model of the technical system to compute the first control output from the setpoint.

3. The method according to claim 1, wherein: the first control output is computed outside of the Two Degree of Freedom controller, fed to the Two Degree of Freedom controller as an input, and processed forward by the feedforward path.

4. The method according to claim 1, further comprising computing the sum-control-variable from the first control output and the second control output by summing the first control output and the second control output.

5. The method according to claim 1, wherein: the anomaly detection signal is derived from the control error as an absolute value of the control error, or the anomaly detection signal is derived from the control error as a squared value of the control error, or the anomaly detection signal is derived from the control error as an L-norm of the control error.

6. The method according to claim 1, further comprising filtering, by a predefined filter, the anomaly detection signal before using the anomaly detection signal to adapt the at least one control parameter of the plurality of control parameters.

7. The method according to claim 1, wherein a linear or non-linear function is provided to describe a dependency between the anomaly detection signal and the at least one control parameter of the plurality of control parameters.

8. The method according to claim 1, wherein: a limitation element parameterized by at least one limitation parameter is provided in the Two Degree of Freedom controller to limit the first control output and/or to limit the second control output and/or to limit the sum-control-variable, the at least one limitation parameter is one of the plurality of control parameters that parametrize the Two Degree of Freedom controller, and the method further comprises adapting the at least one limitation parameter based on the anomaly detection signal.

9. The method according to claim 1, wherein: the Two Degree of Freedom controller is implemented as a discrete-time Two Degree of Freedom controller, the control outputs are ongoingly computed at equidistantly spaced discrete control time points, and the anomaly detection signal is ongoingly computed at equidistantly spaced discrete detection time points.

10. The method according to claim 9, wherein the discrete control time points and the detection time points coincide.

11. The method according to claim 1, wherein the adaptation of the at least one control parameter of the plurality of control parameters is carried out if an absolute value of the anomaly detection signal surpasses a predefined anomaly-threshold.

12. The method according to claim 11, wherein an absolute value of the at least one control parameter of the plurality of control parameters is reduced in order to adapt the at least one control parameter of the plurality of control parameters if the anomaly detection signal surpasses the predefined anomaly-threshold.

13. The method according to claim 1, wherein: a planar motor is controlled as the technical system, or a long stator linear motor is controlled as the technical system, or an industrial robot is controlled as the technical system, or a tool machine is controlled as the technical system.

14. The method according to claim 1, wherein the feedback part solely consists of a proportional controller, having no integral part.

15. A control system having a control unit on which a Two Degree of Freedom controller is implemented for controlling an output of a technical system to a predefined setpoint, the Two Degree of Freedom controller comprising: a feedforward part that provides a first control output; and a feedback part that provides a second control output, wherein: the Two Degree of Freedom controller is parametrized by a plurality of control parameters, the feedback part is configured to; compute the second control output from a control error representative of a deviation between the output of the technical system and the predefined setpoint, the Two Degree of Freedom controller is configured to: determine a sum-control-variable acting on the technical system to control the output to the setpoint from the first control output and the second control output, and the control unit is configured to: at an anomaly detection time point during control of the output of the technical system, derive an anomaly detection signal from the control error present at the anomaly detection time point; adapt at least one control parameter of the plurality of control parameters parametrizing the Two Degree of Freedom controller based on the anomaly detection signal; and from the anomaly detection time point onwards, control the output of the technical system by means of the Two Degree of Freedom controller parameterized by the at least one adapted control parameter.

16. The method according to claim 3, wherein the first control output is computed by an inverse model of the technical system.

17. The method according to claim 5, wherein: the absolute value of the control error is scaled, or the squared value of the control error is scaled, or the L-norm of the control error is scaled.

18. The method according to claim 6, wherein the predefined filter is selected from a group consisting of: a low-pass filter, a band-pass filter, a high-pass filter, and a notch filter.

19. The method according to claim 13, wherein the anomaly detection signal is indicative of a collision of the robot with a static mechanical object.

20. The control system according to claim 15, wherein the feedforward path uses an inverse model of the technical system to compute the first control output from the setpoint.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0019] The present disclosure is described below in greater detail with reference to FIGS. 1 to 7, which show schematic and non-limiting advantageous embodiments of the present disclosure by way of example.

[0020] FIG. 1 shows the known structure of a 2DoF controller.

[0021] FIG. 2 shows a measured 2DoF control signal.

[0022] FIGS. 3a and 3b show control error signals, an anomaly threshold and an anomaly detection signal obtainable with the present disclosure.

[0023] FIG. 4 shows a first implementation of a 2DoF controller.

[0024] FIG. 5 shows a second implementation of a 2DoF controller.

[0025] FIG. 6 shows a first implementation of the present disclosure based on an adapted limitation value.

[0026] FIG. 7 shows a second implementation of the present disclosure based on adapted limitation values.

DETAILED DESCRIPTION

[0027] In FIG. 1, the general and well-known structure of a Two Degree of Freedom (2DoF, 2DoF and Two Degree of Freedom are used synonymously hereafter) controller 2 for controlling a technical system 1 is shown. The Two Degree of Freedom controller 2 comprises a feedforward part 3 and a feedback part 4. As it is often impossible to achieve good setpoint tracking and fast disturbance rejection at the same time with ordinary feedback controllers having only one degree of freedom (for example classical P-, PI-, PID-controllers etc.), a Two Degree of Freedom (2DoF) controller 2 with two degrees of freedom as shown in FIG. 1 can be used. The technical system 1 is, for example, a mechanical system, driven by a sum-control-variable u, providing an actual position as an output y, which is, in some embodiments, measured by an encoder, for example. More specifically, a planar motor PM or a long stator linear motor LLM or an industrial robot or a delta robot or a tool machine may be controlled as a technical system 1. Depending on the specifics of the technical system 1, said sum-control-variable u may be fed to an actuator, for example an amplifier or a power amplifier or a frequency converter or a servo drive etc., which actuator eventually acts on the technical system 1 in accordance with the sum-control-variable u. A skilled person familiar with a technical system 1 to be controlled of course knows which actuator best fits a specific use case. However, a sum-control-variable u may also act directly on a technical system 1 to be controlled.

[0028] The aim of the Two Degree of Freedom controller shown is to control an output y of the shown technical system 1 to a predefined setpoint y.sub.set. As is well-known from control engineering, said feedforward part 3 and said feedback part 4 typically comprise parameters, such as a proportional gain k.sub.p in a proportional path or an integral gain ki of an integral path or gains of an Anti-Windup scheme, such as of an Anti-Windup scheme in the sense of Hanus (cf. Hanus, Raymond, Michel Kinnaert, and J-L. Henrotte. Conditioning technique, a general anti-windup and bumpless transfer method. Automatica 23.6 (1987): 729-739.) in the feedback part 4, or a feed-forward gain in the feedforward part 3, or a limitation value u.sub.max to limit at least one, potentially also more of the control signals computed in the controller 2.

[0029] As it is typically the case in control engineering practice, the feedback part 4 is designed to compute said second control output u.sub.FB from a control error e.sub.y representative of a deviation between the output y of the technical system 1 and the predefined setpoint y.sub.set, and the Two Degree of Freedom controller 2 is designed to eventually determine a sum-control-variable u acting on the technical system 1 to control the output y to the setpoint y.sub.set from the first control output u.sub.FF and the second control output u.sub.FB. As can be seen from FIG. 1, the sum-control-variable u is computed from the first control output u.sub.FF and the second control output u.sub.FB by summing the first control output u.sub.FF and the second control output u.sub.FB. In the case of a mechanical system as technical system 1, the sum-control-variable u may correspond to a torque or to a force.

[0030] The controller design is usually carried out by means of a (mathematical) model of the technical system 1 that comprises also certain state variables x.sub.i(i representing an index, as is well-known in control engineering). The model of the technical system 1 maps the input (control variable u) onto the output y. There can of course also be more than just one output variable y. State variables x.sub.i are typically stacked into state-vector x with dimension nd1, n describing the dimension of the system 1. The 2DoF controller 2 oftentimes additionally utilizes certain state variables x.sub.i of the technical system 1, such as an engine speed n in case of an engine as technical system 1 to be controlled. State variables x.sub.i can be measured by using appropriate sensors or encoders, or can be calculated by using a simulation model, or can be estimated with an observer for the state variable x.sub.i from other known (for example measured) variables of the technical system 1, such as from an output y.

[0031] As is also well-known from the prior art, the feedforward part 3 may encompass an inverse of the technical system 1 to be controlled, namely an inverse of the model of the technical system 1. By means of an inverse model of the technical system 1, it becomes possible to compute a control variable u.sub.FF from a desired setpoint y.sub.set in the feedforward part 3, which, in an ideal scenario without disturbances d and deviations of the real behavior of the system 1 from an assumed behavior, already alone allows to solve the control objective, namely, to control the output y to the setpoint y.sub.set. However, the dynamics of a technical system 1 is oftentimes nonlinear and time variant. Thus, it is frequently difficult to determine the inverse of a nonlinear technical system model, making the implementation of the feedforward part 3 of a 2DoF controller 2 a challenging task.

[0032] To account for these considerations, in the case shown in FIG. 1, it is assumed that a disturbance d acts on the technical system 1, which must be compensated by the controller, and which may lead to a deviation of the real behavior of the technical system 1 from an assumed behavior, which assumed behavior may be reflected in a mathematical model of the technical system 1. In an ideal case without disturbances (d=0), in which the model parameters are identical to the real system parameters, u=u.sub.FF already by itself forms a control signal that leads to y=y.sub.set. In this ideal case (feedforward part 3 is the inverse of the real, acting system), the feedback part 4 has nothing to do (open-loop operation).

[0033] The feedback part 4, may solely consist of a proportional controller k.sub.p*e.sub.y, having no integral part, which allows to avoid issues that could potentially be connected the phenomenon of windup. As will be explained later, this aspect is particularly advantageous within the scope of present disclosure, as it avoids potential issues related to the well-known phenomenon of wind-up, see above. However, also other control architectures are conceivable for the feedback part 4, such as PI-controllers or PID-controllers or sliding-mode controllers etc.

[0034] FIG. 2 shows a measured 2DoF control signal, namely a sum-control-variable u, in the case of a mechanical system as technical system 1 to be controlled, specifically an axis of an industrial robot. In case of said robot, a set torque and a set position for the robot arm are calculated centrally on a central computing unit, which may be implemented in the form of a PLC, and are transferred via a field bus system, for example Ethernet or Ethercat or Ethernet Powerlink or Profibus etc., to specific control units on which control structures as the one shown in FIG. 1 are implemented to carry out the control for a specific arm. In the scope of the present disclosure, a control unit running a 2DoF controller 2 as shown in FIG. 1 may be implemented in the form of FPGA, a microcontroller etc.

[0035] As can be seen from FIG. 2, in case the controller parameters are selected appropriately, the first control output u.sub.FF generated by the feedforward part 3 and the sum-control-variable u eventually output by the controller 2 are highly alike. As mentioned previously, the deviation between the sum-control-variable u and the first control output u.sub.FF generated by the feedforward part 3 can be interpreted as a proportion of the control signal that the feedback part 4 still needs to apply in order to solve the control objective with sufficient accuracy. Ideally, the deviation is around zero and essentially consists of oscillations due to the control behavior or due to model deviations or due to disturbances d that cannot be taken into account when setting up a model. In an ideal scenario, without disturbances d and a perfect model of the technical system 1, thus a perfect inverse model used in the feedforward part 3, already the feedback part 3 alone would solve the control task. Hence, the deviation between the first control output u.sub.FF and the sum-control-variable u, which corresponds to the control error e.sub.y, can be interpreted as a measure of model inaccuracies and/or disturbances. Inaccuracies, disturbances and other effects that lead to deviations from an ideal model behavior are subsumed hereafter as anomalies. In case the inaccuracies become too large, the controller 2 may not function properly anymore. Thus, the present disclosure provides for a method that allows to react to said anomalies leading to modifications of the technical system 1 and hence to said inaccuracies.

[0036] To that end, within the scope of the present disclosure, it is provided to, at an anomaly detection time point t.sub.w during control of the output y of the technical system 1, derive an anomaly detection signal w from the control error e.sub.y present at said anomaly detection time point t.sub.w, to adapt at least one control parameter of the multiple of control parameters k.sub.p, k.sub.i, u.sub.max parametrizing the Two Degree of Freedom controller 2 depending on the anomaly detection signal w and to, from the anomaly detection time point t.sub.w onwards, control the output y of the technical system 1 is by means of the Two Degree of Freedom controller 2 parameterized by said at least one adapted control parameter k.sub.p, k.sub.i, u.sub.max. The present disclosure thus allows for a relatively simple way of detecting an anomaly and reacting to it accordingly (actively or proactively).

[0037] In course of the present disclosure, it was found that in case a two-degree-of-freedom controller is used, the control error is indicative of a deviation between the assumed model behavior and the behavior of the real technical system. In case of a (classical) controller that does not comprise a feed-forward path, such a conclusion cannot be drawn. In case of a classical controller, a control error is necessary, in order to allow for any control input to be generated. Hence, in such a case, a control error does not provide any insight regarding the presence of any kind of anomaly.

[0038] As it is typically the case in modern control engineering, the Two Degree of Freedom controller 2 may be implemented as a discrete-time Two Degree of Freedom controller 2, the control outputs u.sub.FB, u.sub.FF being ongoingly computed at equidistantly spaced discrete control time points t.sub.k, typically spaced by a constant, predefined sampling time Td of, for example, T.sub.d=10 s or T.sub.d=100 s or T.sub.d=1 ms or T.sub.d=10 ms etc., the anomaly detection signal w being ongoingly computed at equidistantly spaced discrete detection time points t.sub.w. Beneficially, the discrete control time points t.sub.k and the detection time points t.sub.w may coincide.

[0039] To allow for a more detailed discussion of the present disclosure, FIGS. 3a and 3b each show control error signal e.sub.y, an anomaly threshold W and an anomaly detection signal w obtainable by means of the present disclosure. An anomaly threshold W may in particular be provided to carry out the adaptation of the at least one control parameter of the multiple of control parameters k.sub.p, k.sub.i, u.sub.max only in case the absolute value of the control error e.sub.y surpasses a predefined anomaly-threshold W. As can be seen in FIG. 3a, the anomaly detection signal w is set equal to the control error e.sub.y. However, also other design options for deriving the anomaly detection signal w exist, which will be explained later. At the anomaly detection time point t.sub.w, the detection signal w exceeds said predefined anomaly-threshold W, which is additionally emphasized by the status signal s.

[0040] As can further be seen in FIG. 3a, a control parameter to be adapted, in FIG. 3a indicated by parameters k.sub.p and u.sub.max, is reduced as long as the anomaly-threshold W is exceeded. Reducing a control parameter is particularly useful in some embodiments. An important reason therefor is that an anomaly indicates some sort of deviation from an assumed model behavior. Hence, in order to reduce the risk of problems that might result from such deviations, such as a potential loss of stability, reducing control parameters allows to increase, for instance a stability margin, and thus make the operation safer again. Reducing the control parameters allows to refuse the controller as a whole.

[0041] Speaking again in general terms, to detect an anomaly, a non-linear function may be provided to arrive at the anomaly detection signal w:

[00001]$w=f\left(y_{\mathrm{set}},{\stackrel{}{y}}_{\mathrm{set}},{\stackrel{.Math.}{y}}_{\mathrm{set}},u_{\mathrm{FF}},{\stackrel{.}{u}}_{\mathrm{FF}},{\stackrel{.Math.}{u}}_{\mathrm{FF}},y,\stackrel{}{y},\stackrel{.Math.}{y},.Math.\right)$

[0042] In the simplest case, the deviation of the actual output y from the target setpoint yset can be used directly for this purpose:

[00002]$w:=e=y_{\mathrm{set}}-y$

[0043] If the signal quality allows it, higher derivatives of the variables mentioned can also be included:

[00003]$w=k_1\left(y_{\mathrm{set}}-y\right)+k_2\frac{d}{\mathrm{dt}}\left(y_{\mathrm{set}}-y\right)+k_3\frac{d^2}{{\mathrm{dt}}^2}\left(y_{\mathrm{set}}-y\right)+.Math.$

[0044] As indicated by the formula above, a control error e.sub.y=ysety and its derivations may of course also be scaled by appropriate scaling weights. Hence, the anomaly detection signal w may be derived from the control error e.sub.y as an, in some embodiments scaled, absolute value |e| of the control error e.sub.y or as a, in some embodiments scaled, squared value e.sup.2 of the control error ey or as an, in some embodiments scaled, L-norm L(e) of the control error ey, hence allowing for great flexibility, especially when it comes to fine tune the present disclosure to a specific practical use case.

[0045] With regards to the dependency between the anomaly value and the at least one control parameter, a linear or non-linear function is provided to describe a dependency between the anomaly detection signal w and the at least one control parameter of the multiple of control parameters kp, ki, umax. However, different approaches are conceivable in this regard. For instance, the sum-control-variable u may also be frozen to a pre-defined value in case said anomaly-threshold W is exceeded, hence effectively deactivating the 2DoF controller 2.

[0046] In the cases referred to above, the differentiating behavior can increase sensitivity and lead to a faster reaction in the status formation described below. The f( . . . ) function can also have an additional filtering effect in the case of noisy signals. The specified deviation can also be used to create a status signal s, as already indicated in FIG. 3a:

[00004]$s=\{\begin{array}{cc}1,& .Math.e.Math.>W\\ 0,& \mathrm{otherwise}\end{array}$

[0047] It is also conceivable to make the threshold W asymmetrical (using two values (W1, W2)) or variable in time (W1(t), W2(t)), in order to be able to adapt the detection even more individually.

[0048] As mentioned at the outset, a planar motor PM or a long stator linear motor LLM or an industrial robot, especially in the form of a delta robot, may be controlled as the technical system 1. The application may be used wherever electrically driven axes come into contact with other machine parts, for example in case of packaging machines, robotics, planar motors (PM), long stator linear motors (LLMs), etc. In case an industrial robot is controlled as the technical system 1, a detected anomaly may in particular be indicative of a collision of the robot with a mechanical object. By applying the teachings of the present disclosure, many types of anomalies can be detected, even unexpected unstable behavior, resulting from changes in the plant causing instability. Regardless of whether this is a disturbance from outside or a parameter variation of the controlled system (change in the mechanics). Specifically, collisions can be detected at an early stage. In the event of a collision, the control error ey and hence the anomaly detection signal w will increase abruptly. This change of state will be detected and the control action will be restricted in accordance with the parameterization.

[0049] Specifically, in the case of an industrial robot, by means of the present disclosure, a detection of the loss of an arm in a tripod can be detected, or an incorrect parameterization may be detected as an anomaly, or an incorrectly parameterized tool may be detected, or mechanical changes over time (friction, bearing damage, looseness, . . . ) may be detected, but also changes in a drivetrain (for example belt wear, blockage, slipping clutches, etc.) may be detected, indicating when model parameters should be adjusted (running-in behavior of the mechanics, temperature influence, process change, etc.).

[0050] Depending on the operating mode (centralized, decentralized), also other versions of a 2 DoF controller 2 are conceivable, particularly as shown in FIGS. 4 and 5. Specifically, as shown in FIG. 4, in the case of decentralized control (single-axis operation), the feedforward control u.sub.FF is determined from the setpoint y.sub.set, in the controller 2 itself. However, in case of centralized control (as it is typical when operating a group of axis, axis group), the feedforward control may also be calculated centrally, in other words, outside of the Two Degree of Freedom controller, hence a tuple [y.sub.set, u.sub.FF] being fed to the respective drive, as depicted in FIG. 5. Therefore, the first control output u.sub.FF may be computed outside of the Two Degree of Freedom controller 2, in some embodiments by means of an inverse model of the technical system 1, fed to the Two Degree of Freedom controller 2 as an input and processed forward by the feedforward path 3, or the feedforward path 3 may use an inverse model of the technical system 1 to compute said first control output u.sub.FF from the setpoint yset.

[0051] FIG. 6 further shows a possible implementation of the present disclosure based on an adapted limitation value. The limitation can be carried out in the feedforward branch, in the feedback branch, but can also act on entire manipulated variable, namely, the sum-control-variable u. However, also other options to implement a limitation are conceivable (symmetrical, asymmetrical, time-dependent, process-dependent, . . . ).

[0052] Moreover, as depicted in FIG. 6 as well, the anomaly detection signal w may, in some embodiments, be filtered by means of a predefined filter F, in some embodiments a low-pass filter or a band-pass filter or a high-pass filter or a notch filter, before being used to adapt the at least one control parameter of the multiple of control parameters k.sub.p, k.sub.i, u.sub.max.

[0053] As also shown in FIG. 6, after the filter F, the anomaly value is connected to a limitation element lim. To implement the present disclosure, a limitation element lim as the one shown in FIG. 6 parameterized by at least one limitation parameter lim1 may be provided in the Two Degree of Freedom controller 2 to limit the first control output u.sub.FF and/or to limit the second control output u.sub.FB and/or to limit the sum-control-variable u, such that the at least one limitation parameter u.sub.lim may be considered as one of the by a multiple of control parameters k.sub.p, k.sub.i, u.sub.max to parametrize the Two Degree of Freedom controller 2, and the at least one limitation parameter u.sub.lim is adapted depending on the anomaly detection signal w.

[0054] A second implementation of the present disclosure based on adapted limitation values is presented in FIG. 7. In both cases shown in FIGS. 6 and 7, a conceivable approach to implement the present disclosure is to provide for a first, regular value of said limitation values lim1, lim2, lim3, in case no anomaly is detected, and for a second set of values, whose absolute value is smaller, for the case that an anomaly is detected and that a reduction is required.

[0055] The disclosed systems and methods are not limited to the specific embodiments described herein. Rather, components of the systems or activities of the methods may be utilized independently and separately from other described components or activities.

[0056] This written description uses examples to disclose various embodiments, which include the best mode, to enable any person skilled in the art to practice those embodiments, including making and using any devices or systems and performing any incorporated methods. The patentable scope is defined by the claims and may include other examples that occur to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insubstantial differences form the literal language of the claims.