Method and apparatus for continuously determining a quality measure

Abstract

The invention relates to a method for determining a quality measure (G), in particular continuously, from input variables (u) and output variables (y), for a regulating mechanism of a technical device (3) or for a system model of a technical system; wherein time series of the input variables (u) and output variables (y) up to a time step are detected (S1), wherein a discrete ARMAX-model structure is adapted (S2) to the ascertained respective time series of the input variables and output variables in order to determine a first parameter set A for depicting the time series of the input variables (u) and a second parameter set B for depicting the time series of the output variables (y), wherein the quality measure (G) is determined from the first and second parameter sets A, B for the time step (t).

Description

Technical Field

[0001] The present invention relates to system models and regulating mechanisms for technical systems, and more particularly to methods for determining quality metrics of system models and regulators, especially for fault identification or for support in regulator design. Background Technology

[0002] In technical systems, regulating mechanisms are frequently used to manipulate a technical device by adjusting a certain amount to achieve a regulating objective. The regulating objective is defined by a pre-defined set of parameters, and its achievement is verified by detecting the actual parameters. The regulating mechanism is used to ensure that the actual parameters optimally track the pre-given set parameters.

[0003] Ideally, the adjustment mechanism would result in the actual parameter corresponding to the rated parameter at every moment. However, in reality, the actual parameter exhibits a delayed response due to dynamic changes in the rated parameter and the dynamic characteristics of the device.

[0004] The quality of such a regulator thus indicates a metric: how precisely and quickly the actual parameter follows the pre-given dynamic rated parameter. In misdesigned regulators, situations may arise where the actual parameter follows the rated parameter slowly, or where the actual parameter overshoots the rated parameter. Such misdesigns of the regulator can be quantified using quality metrics.

[0005] As an alternative, system models can be generated in other application areas, mimicking the physical characteristics of the technical system. These system models can be used, for example, in simulations or as observers of regulating mechanisms.

[0006] These system models often do not adequately depict the performance of the technical system across all operating ranges, and thus the corresponding quality metrics may vary. In particular, the complex frequency-dependent performance of the technical system can lead to significant discrepancies between the modeled performance and the actual performance of the technical system. Therefore, for such system models, it is possible to define a quality metric that indicates the accuracy of the modeled performance relative to the actual performance at a specific operating point of the technical system. Summary of the Invention

[0007] According to the present invention, a method for continuously determining a quality scale according to claim 1, an apparatus according to the parallel claims, and a fault identification system are provided, wherein the quality scale is used for an adjustment mechanism of a technical device or for a system model of a technical system.

[0008] Other design options are described in the dependent claims.

[0009] According to a first aspect, a method is described for determining a quality scale from input parameters and output parameters, the quality scale being used for the adjustment mechanism of a technical device or for a system model of a technical system; wherein time series of the input and output parameters up to a time step are detected, wherein a discrete ARMAX-model structure is adjusted for the identified corresponding time series of the input and output parameters to determine a first set of parameters for describing the time series of the input parameters and a second set of parameters for describing the time series of the output parameters, wherein the quality scale is determined based on the first and second sets of parameters for the time step.

[0010] For regulating mechanisms, quality metrics can indicate the quality of regulation, and for modeling, they can indicate the quality of the model, which should mimic system performance.

[0011] Typically, the regulation quality, used as a quality metric, cannot be determined during continuous operation or only through sampling over the entire operating range of the device to be regulated. Therefore, insufficient regulation characteristics can only be reliably identified within specific operating scenarios, and conclusions regarding other operating ranges and continuous variations in the quality metric are generally impossible. This is especially true for safety-critical systems or regulation mechanisms requiring minimum regulation quality within the range of internal combustion engines, which must adhere to pre-defined emission regulations at all possible operating points.

[0012] To model a technical system, a system model is typically used to simulate the system's performance. Such a system model can be optimized through quality assessments, either in simulations or when the technical device is used in the technical system.

[0013] For example, such a system model can be created using a PT2-stage model. For this PT2-stage model, model parameters—namely, amplification factor, time constant, and decay configuration data (bedaten)—must be provided. If only a limited amount of measurement data is available to determine these model parameters, it can lead to the following result: the identified model parameters are not representative of other scenarios, i.e., other operating ranges, resulting in poor modeling of the technical system. Adjustments to the system model using expert knowledge are typically difficult to make because this requires considering the system performance across many or all operating ranges.

[0014] According to the above method, the quality scale for the adjustment mechanism or system model used in the technical device should be determined by describing the system performance through the system model. Thus, the quality scale can be determined online, that is, during continuous operation, throughout the entire adjustment range or operating range, and for different operating scenarios.

[0015] The concept behind the above method is to determine the quality scale using a parameter set of an ARMAX model, which is matched with the time series of the input and output parameters. The time series of the input and output parameters represent the main dynamic characteristics of the control mechanism or system model.

[0016] The underlying model for ARMAX can be a virtually arbitrary discrete model, whose parameters should be adapted to the parameters to be observed. The number of parameters to be optimized is generally unlimited, as the system is overdefined depending on the length of the observed time series. For each time step, there exists a model equation, thus providing sufficient data for parameter adaptation in optimization methods (such as least squares).

[0017] To determine the quality metric used to regulate the mechanism, a PT2-stage transfer function can be selected, where the nominal parameter can be assumed as the input parameter and the actual parameter as the output parameter. From the discrete system, the corresponding identified parameter set for each time step, the model parameters of the time-continuous transfer function can be determined, particularly the amplification factor, time constant, and attenuation, from which the quality metric is determined.

[0018] The adjustment quality of the regulating mechanism can be measured by the corresponding speed (the actual parameter tracking the rated parameter at this speed) and the transient response when the rated parameter is reached. Specifically, the speed (the actual parameter tracking the rated parameter at this speed) can be expressed by the time constant T of the PT2-element, while the transient response when the rated parameter is reached is characterized by the attenuation D. If the actual value cannot consistently track the rated value, a residual adjustment deviation occurs. This residual adjustment deviation can be identified if the amplification factor K of the PT2-element is significantly different from K=1 for a prolonged period.

[0019] The quality metric can be determined using the known recursive ARMAX algorithm (ARMAX: Autoregressive Moving Average Module with Exogenous Input Model), which determines the model parameters based on the time series of the input and output parameters and recalculates the results taking into account previous time steps.

[0020] According to one embodiment, a quality metric of an adjustment mechanism can be determined by: pre-given rated parameters of the adjustment mechanism as input parameters and pre-given actual parameters of the adjustment as output parameters, wherein the adjustment mechanism has a transfer function, wherein model parameters of the transfer function, particularly amplification factor, time constant, and attenuation, are determined from the first and second parameter sets, wherein the quality metric is determined from the model parameters.

[0021] To optimize the transfer function in terms of its parameters, the time constant, decay, and amplification factor can be determined at each time step by inverse calculation in a time-continuous system, following an iterative optimization method. The ARMAX algorithm here functions as a least-squares optimization procedure, which uses a time-discrete PT2 = f(D, T, K) as a model and adjusts the time constant and decay to fit the time-discrete PT2 as well as possible to the measurement data. The quality metric can be derived from the model parameters, i.e., from the amplification factor, time constant, and decay.

[0022] The parameter set can be identified even with a noise share assumed to be 0.

[0023] Specifically, the quality scale can be determined based on the intervals between the attenuation values ​​that are 1 and the intervals between the time constant values ​​that are 1. Furthermore, the quality scale can be correlated with the intervals between the amplification factor values ​​that are 1.

[0024] Furthermore, the quality metric can be used to adapt the adjustment mechanism of the technical device.

[0025] It can be specified that a fault in the adjustment mechanism can be identified based on the value or the change in the quality scale, and that the fault can be reported with a signal when it is identified.

[0026] In order to evaluate the system model, the output parameters of the system model and the actual (measured) system parameters that describe the system performance can be assumed as input parameters in a corresponding manner.

[0027] The system model can be evaluated in a manner that provides a system parameter modeled by the system model as an input parameter based on a pre-given variation curve of one or more control parameters. The measured system performance is used as an output parameter, wherein the system parameter measured in the actual technical system is assumed to be the output parameter. The quality scale of the system model is generated according to the above processing method, by determining the model parameters of the system model from the first and second parameter sets. The advantage over data configuration from time-varying curves lies in the availability of a set of measurement data when determining the quality scale during continuous operation.

[0028] In particular, the system model can be adjusted according to the quality metric to better depict the actual performance of the technical system.

[0029] According to another aspect, there is provided an apparatus for determining a quality scale, particularly continuously, from input and output parameters, the quality scale being used for an adjustment mechanism of a technical device or for a system model of a technical system; wherein the apparatus is configured to detect time series of the input and output parameters up to a time step, wherein a discrete ARMAX-model structure is adjusted for the identified corresponding time series of the input and output parameters to determine a first set of parameters for describing the time series of the input parameters and a second set of parameters for describing the time series of the output parameters, and the quality scale is determined based on the first and second sets of parameters for the time step. Attached Figure Description

[0030] The embodiments are explained in detail below with reference to the accompanying drawings. Wherein: Figure 1 A schematic diagram of the regulated system is shown; Figure 2 A schematic diagram of transfer functions with different decay and time constants is shown; and Figure 3 A flowchart is shown, which is used to illustrate the process for... Figure 1 A method for online fault identification in regulated systems. Detailed Implementation

[0031] Figure 1A schematic diagram of an adjustable system 1 with an adjustment mechanism 2 and a technical device 3 is shown. The adjustment mechanism has parameterizable adjustment performance corresponding to the adjustment function 21. From the differential element 22, the adjustment mechanism obtains an adjustment deviation as an input parameter between a predetermined rated value of the rated parameter u and a measured actual value of the actual parameter y, which is derived based on the performance of the technical device 3.

[0032] In practice, a transfer function exists between the pre-given rated parameter u and the actual parameter y generated by the regulating mechanism. This regulating function 21 can be constructed in a variety of ways, particularly as a PID controller, etc.

[0033] The system performance of the regulating mechanism can be characterized using the PT2-transfer function. The PT2-transfer function is essentially determined by model parameters of the amplification factor K, the time constant T, and the attenuation D. The transfer function of the PT2-stage corresponds to: in Furthermore, ω0 corresponds to the natural angular frequency, and the symbol "s" in the formula represents the Laplace operator of the Laplace transform.

[0034] The time constant T here determines the following velocity: the actual parameter y uses this velocity to track the rated parameter u. The transient response, that is, overshoot or too slow approach to the corresponding pre-given rated value, is determined by the attenuation D.

[0035] exist Figure 2 The diagram illustrates the variation curves of the transfer function for different attenuations D and time constants T. Curve K1 shows the ideal, optimal possible variation curve of the transfer function of the regulating mechanism, while curve K2 has a variation curve with too high an attenuation, and curve K3 has a variation curve with too small a time constant and / or too small an attenuation.

[0036] A control unit 4 is provided, which monitors the regulating mechanism of the regulated system. Here, the control unit 4 is configured to determine a quality criterion G, adjust the regulating function of the technical system 1 according to the quality criterion G, or, for example, report a fault using a signal according to the quality criterion G when the regulating mechanism no longer meets a predetermined quality standard within a specific operating range of the technical system 1.

[0037] The following uses Figure 3The flowchart describes in detail the method for operating the regulated system 1. The method can be implemented in the control unit in hardware and / or software form.

[0038] In step S1, the nominal value u(t) of the nominal parameter and the actual value y(t) of the actual parameter are continuously detected in the regulated technical system 1 at predetermined times, especially at regular time steps t=1, 2, ... Now, at each time step, according to the ARMAX model (Autoregressive Moving Average Model with Exogenous Inputs), under the condition that the nominal value of the nominal parameter corresponds to the actual value of the actual parameter at each time step, the coefficients a and b can be determined from the time series of the nominal parameter (input parameter) u and the actual parameter (output parameter) y.

[0039] The ARMAX model structure corresponds to: or Where y(t) corresponds to the value of the output parameter at the currently observed time step t, n a Corresponding to the number of poles, n b The number corresponding to the root + 1, n c Corresponding to the number of coefficients used to describe the noise share, n k Corresponding to the number of input values ​​of the input parameter identified before the input of the regulating mechanism or system model affects the output parameter (rest time), y(t-1), ..., y(tn) a ) corresponding to the previous output parameter, wherein the current value of the output parameter depends on the previous output parameter, and u(tn) k ), ..., u(tn) k -n b +1) corresponding to the preceding and delayed values ​​of the input parameter, wherein the current output parameter depends on the input parameter, and e(t-1), ..., e(tn) c The curve corresponding to the change in the amount of interference over time.

[0040] The time-continuous transfer function of the PT2-stage In step S2, which determines the parameter sets a and b, a time-discrete model is transformed, for example, using a z-transform. For the PT2-transfer function, na = 2 (2 poles), nb = 2 (number of roots + 1), nc = 0 (number of coefficients for white noise), and nk = 1 (rest time). The time-discrete PT2-transfer function is determined by the current time step and two past time steps. This time-discrete PT2-transfer function is used in the ARMAX algorithm to allow the coefficients of the z-transformed PT2-transfer function to be determined. By performing an inverse transformation over the time range, the amplification constant (amplification factor) K, the attenuation D, and the time constant T can be determined.

[0041] Using the coefficients to be identified, the first parameter set A a1, a2, ..., a na The second parameter group B b1, b2, ... b nb And c1, c2, ..., c nc =0, which applies to the adjusted case, meaning that the nominal value u(t) of the nominal parameter u corresponds to the actual value y(t) of the actual parameter y for the observed current time step t. Therefore, as a standard, all coefficients a1, a2, ..., a... can be predetermined. na The sum equals all coefficients b1, b2, ... b nb The sum of the coefficients, b. Deviations between the sums of these coefficients can be identified as adjustment deviations and represent a quality measure. If this deviation persistently deviates from the value in the regulated state, then residual adjustment deviations can be identified.

[0042] The remaining adjustment deviation is thus derived from the difference between the determined sums, wherein the smaller the difference between the sums, the higher the adjustment quality.

[0043] The parameters a and b of the ARMAX model can be used to inversely deduce the model parameters of the conversion model, namely the amplification factor, attenuation D, and time constant T. Quality scales can then be assigned to the attenuation D and time constant T according to appropriate standards.

[0044] The calculation of the quality metric can be performed at predetermined intervals, either within each time step or after a predetermined number of time steps, so that the regulating mechanism can be continuously monitored during the operation of the regulated system 1. The quality metric can be determined, for example, as the distance from a predetermined value of the time constant and / or the distance from a decay value of 1.

[0045] In step S3, it is checked whether the currently identified quality metric conforms to a pre-defined quality standard. For example, the quality standard can be queried to determine whether the operating point determined by the current attenuation D and the current time constant T is within or outside a pre-defined range of attenuation D equal to 1 around the time constant T. If the attenuation D and the time constant T are within the pre-defined range (either option: yes), the method continues with step S1; otherwise (either option: no), a signal is used in step S4 to report a fault in the regulating mechanism.

[0046] The above method can be used alternatively to evaluate a system model that should depict the physical characteristics of a technical system. Here, the input values ​​of the input parameters correspond to the model values ​​of the system model, and the output parameters correspond to the measured parameters of the technical system. The system model is used to mimic the technical system, which can be determined based on quality metrics identified from the input and output parameters as described above. This allows the system model to be matched manually or automatically (iteratively) to the actual physical characteristics of the technical system, and, for example, to be used in conjunction with an observer for conditioning design or for system optimization.

Claims

1. A method for determining a quality scale (G) from input parameters (u) and output parameters (y), the quality scale being used for the adjustment mechanism of a technical device (3) or for a system model of a technical system; wherein the time series of the input parameter (u) and the output parameter (y) up to a time step are detected (S1), wherein a discrete ARMAX-model structure is adjusted (S2) for the identified corresponding time series of the input parameter and the output parameter, the ARMAX-model structure having a transfer function corresponding to a PT2-stage, in order to determine a first parameter set A for describing the time series of the input parameter (u) and a second parameter set B for describing the time series of the output parameter (y), wherein the model parameters of the transfer function are determined based on the first and second parameter sets A and B for time step (t), wherein, The model parameters include amplification factor (K), time constant (T), and attenuation (D), wherein the quality scale (G) is determined from the model parameters.

2. The method of claim 1, wherein the parameter groups A and B are determined under the assumption of a noise share of 0.

3. The method according to claim 1, wherein the quality metric (G) of the regulating mechanism is determined by: pre-given a pre-given rated parameter of the regulating mechanism as an input parameter (u) and pre-given an output parameter (y) of the regulating mechanism.

4. The method of claim 3, wherein the quality scale (G) is determined based on the interval of the attenuation (D) from a value of 1 and / or based on the interval of the time constant (T) from a predetermined nominal value.

5. The method of claim 1, wherein the quality scale is determined based on the difference between the sum of parameters in the first parameter group A and the sum of parameters in the second parameter group B, wherein the quality scale (G) depends on the spacing of the difference relative to a comparison parameter corresponding to the difference between the sum of parameters in the first parameter group A and the sum of parameters in the second parameter group B in an adjusted state.

6. The method according to claim 1, wherein the quality metric (G) is used to adapt the adjustment mechanism of the technical device (3).

7. The method according to claim 1, wherein the quality scale (G) is determined continuously in successive time steps, wherein a fault of the adjustment mechanism is identified (S3) based on the value of the quality scale (G) or the time-varying nature of the quality scale (G), and the fault is reported (S4) by a signal when the fault is identified.

8. The method of claim 1, wherein the quality metric (G) of the system model is determined by: pre-given the system parameters to be modeled as input parameters (u), and pre-given the system parameters to be measured as output parameters (y).

9. The method of claim 8, wherein the system model is adjusted according to the quality metric (G).

10. An apparatus for determining a quality scale (G) from input parameters (u) and output parameters (y), the quality scale being used for an adjustment mechanism of a technical device (3) or for a system model of a technical system (1), wherein the apparatus is configured to detect time series of the input parameters (u) and output parameters (y) up to a time step, wherein a discrete ARMAX-model structure is adjusted for the identified corresponding time series of the input and output parameters, the ARMAX-model structure having a transfer function corresponding to a PT2-stage, in order to determine a first set of parameters A for describing the time series of the input parameters (u) and a second set of parameters B for describing the time series of the output parameters (y), and in order to determine model parameters of the transfer function based on the first and second sets of parameters A and B for time step (t), wherein, The model parameters include amplification factor (K), time constant (T), and attenuation (D), wherein the quality scale (G) is determined from the model parameters.

11. A computer program product comprising instructions that, when executed by at least one data processing unit, cause the data processing unit to perform the steps of the method according to any one of claims 1 to 9.

12. A machine-readable storage medium comprising instructions that, when executed by at least one data processing mechanism, cause the data processing mechanism to perform the steps of the method according to any one of claims 1 to 9.

Images