Characteristic value estimation device and characteristic value estimation method

Abstract

To provide a characteristic value estimation device and a characteristic value estimation method capable of easily estimating a characteristic value of a control object.SOLUTION: A characteristic value estimation device 1 estimates a characteristic value of a control object W. The characteristic value estimation device 1 includes: a first characteristic value acquisition unit 11 which determines a first characteristic value of the control object W from a frequency characteristic of the control object W; a second characteristic value calculation unit 13 which calculates a second characteristic value of the control object W by using a state equation containing a variable which influences the first characteristic value, and changing the variable in a search region; an optimization processing unit 14 which obtains an optimum solution, in the search region, of an evaluation function being a relational expression of the first characteristic value and the second characteristic value by changing the second characteristic value calculated by the second characteristic value calculation unit 13 while changing the variable in the search region; and an estimation unit 15 which estimates the variable when the optimum solution is obtained by the optimization processing unit 14 as the characteristic value.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a characteristic value estimation device and a characteristic value estimation method for estimating a characteristic value of a controlled object. [Background technology]

[0002] In automotive testing equipment that applies a load from a motor (dynamo, dynamometer) to a test piece (power system such as a drive motor, engine, or transmission) to perform evaluation tests on the test piece, torsional vibrations occurring on each connecting shaft in a power transmission system in which the motor, the test piece that serves as the motor's load, and a torque detector are connected by shafts are a problem.One method of solving this problem is to apply a disturbance observer or the like to the automotive testing equipment.

[0003] Furthermore, as a method for seeking solutions to problems of a controlled object, a method for analyzing the controlled object using a simulation is known. In such a method, it is necessary to reflect the characteristic values ​​of the controlled object in the simulation.

[0004] In either of the above cases, it is necessary to estimate the characteristic value of the controlled object. As such, in the field of control, there is a demand for estimating the characteristic value of the controlled object. As a method for estimating such a characteristic value of the controlled object, for example, a parameter identification device disclosed in Patent Document 1 is known.

[0005] The parameter identification device in Patent Document 1 identifies model parameters of a two-inertia system model. Specifically, the parameter identification device generates a plurality of phase plane diagrams based on a torque command for a motor, actual measured values ​​of the angle and angular velocity of the motor, and actual measured values ​​of the angle and angular velocity of a load, and identifies model parameters indicating the inertia, viscosity, and friction of the motor, the inertia, viscosity, and friction of the load, the stiffness of a connecting member connecting the motor and the load, and the dead zone width of the connecting member based on the phase plane diagrams.

[0006] Another method for estimating the characteristic values ​​of a controlled object is to use a calculation program to find a mathematical formula or the like that fits time-series data to estimate the characteristic values. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Patent No. 6756653 Summary of the Invention [Problem to be solved by the invention]

[0008] However, the parameter identification device of Patent Document 1 targets a two-inertia system model and requires actual measurement of the angular velocities of both the motor and the load. Therefore, if it is attempted to use the parameter identification device of Patent Document 1 to estimate the characteristic values ​​of the controlled object even in the case of a multi-inertia system, it is necessary to significantly change the algorithm of the parameter identification device and to use a large number of sensors to actually measure the angular velocities.

[0009] Furthermore, the method of estimating characteristic values ​​by finding a mathematical formula or the like that fits time-series data requires time for calculation, and it is difficult to easily estimate characteristic values ​​that allow a mathematical formula or the like to be accurately fitted to time-series data.

[0010] An object of the present invention is to provide a characteristic value estimation device and a characteristic value estimation method that can easily estimate a characteristic value of a controlled object. [Means for solving the problem]

[0011] A characteristic value estimation device according to one embodiment of the present invention is a device for estimating a characteristic value of a controlled object. The characteristic value estimation device includes: a first characteristic value acquisition unit that calculates a first characteristic value of the controlled object at a predetermined frequency from the frequency characteristics of the controlled object; a second characteristic value calculation unit that calculates a second characteristic value of the controlled object when a variable affecting the first characteristic value is varied within a search region at the predetermined frequency using a state equation including the variable; an optimization processing unit that changes the second characteristic value calculated by the second characteristic value calculation unit by varying the variable within the search region to obtain an optimal solution within the search region of an evaluation function that is a relational expression between the first characteristic value and the second characteristic value; and an estimation unit that estimates the variable at which the optimal solution is obtained by the optimization processing unit as the characteristic value (first configuration).

[0012] This makes it possible to change the variables of the state equation for calculating the characteristic value of the controlled object within the search region to find an optimal solution to the evaluation function, and estimate the variables when the optimal solution is found as the characteristic value of the controlled object. Since the evaluation function is a relational expression between the first characteristic value calculated from the frequency characteristic of the controlled object and the second characteristic value calculated using the variables changed within the search region and the state equation, the characteristic value can be easily estimated by finding the optimal solution of the evaluation function.

[0013] In the case of commonly performed time response fitting and frequency response fitting, data must be fitted to each measurement value. Therefore, the accuracy of the fitting is affected by the amount of data. Therefore, in the case of time response fitting and frequency response fitting, in order to achieve accurate fitting, data must be fitted to many measurement points, which imposes a large computational load on the device.

[0014] In contrast, in the above-described configuration, the first feature value and the second feature value are used to perform optimization by the evaluation function, so that the amount of data calculation processing can be significantly reduced.

[0015] Therefore, the calculation processing load of the device can be reduced compared to the conventional case where characteristic values ​​are obtained by fitting to time series data or the like.

[0016] Therefore, the above-described configuration provides a characteristic value estimation device that can easily estimate the characteristic value of a controlled object.

[0017] In the first configuration, the evaluation function is a relational expression including the square of the difference between the first feature value and the second feature value, and the optimization processing unit determines, as the optimal solution, the value of the evaluation function when the square of the difference between the first feature value and the second feature value is minimum (second configuration).

[0018] This makes it possible to realize an evaluation function, which is a relational expression between the first feature value and the second feature value, in the first configuration. Moreover, by squaring the difference between the first feature value and the second feature value, the difference between the first feature value and the second feature value can be found regardless of whether the first feature value and the second feature value are large or small. The optimization processing unit determines the value of the evaluation function when the square of the difference between the first feature value and the second feature value is smallest as the optimal solution of the evaluation function, making it possible to easily find the optimal solution.

[0019] Therefore, the above-described configuration provides a characteristic value estimation device that can easily estimate the characteristic value of a controlled object.

[0020] In the first or second configuration, the first characteristic value and the second characteristic value each include at least a gain (third configuration).

[0021] This makes it possible to obtain characteristic values ​​such as the spring constant of the controlled object using the vibration frequency and gain.

[0022] In any one of the first to third configurations, the search region comprises a plurality of local regions, and the optimization processing unit includes: a local region optimization processing unit that changes the variables of the state equation in each of the local regions to change the second feature value calculated by the second feature value calculation unit to obtain an optimal solution of the evaluation function, and an optimal solution selection unit that selects an optimal solution of the evaluation function in the search region from the optimal solutions of the evaluation function obtained in each local region by the local region optimization processing unit (fourth configuration).

[0023] This allows the optimum solution of the evaluation function to be found in each of the local regions obtained by dividing the search region of the variables into multiple regions. Then, by finding the optimum solution for the search region from among the optimum solutions for each local region, it is possible to find an optimum solution for the entire search region, rather than a local optimum solution. Therefore, it is possible to find the optimum solution of the evaluation function in the search region with high accuracy. Therefore, it is possible to find the characteristic value of the controlled object with high accuracy.

[0024] Moreover, with the above-described configuration, the characteristic value can be estimated even if the search region is wide, so the characteristic value can be found with high accuracy without adjusting the range of the search region and without finding a local solution.

[0025] On the other hand, if the search region for variables is not divided into multiple regions, there is a high possibility of finding a local solution, and therefore the search region must be changed through trial and error. While it is possible to avoid the calculation of a local solution by increasing the number of optimization iterations or tightening the criteria for convergence evaluation of the evaluation function, this requires time for calculation. In contrast, by dividing the search region into multiple local regions as in the above configuration, it is possible to find an optimal solution without trial and error or changing the criteria for convergence evaluation. Therefore, by dividing the search region into multiple regions as described above and finding an optimal solution of the evaluation function in each local region, the calculation load of the characteristic value estimation device can be reduced.

[0026] A characteristic value estimation method according to one embodiment of the present invention is a method for estimating a characteristic value of a controlled object. This characteristic value estimation method includes: a first characteristic value acquisition step of determining a first characteristic value of the controlled object at a predetermined frequency from the frequency characteristics of the controlled object; an optimization step of calculating a second characteristic value of the controlled object when a variable affecting the first characteristic value is varied within a search region at the predetermined frequency using a state equation including a variable affecting the first characteristic value, thereby determining an optimal solution within the search region of an evaluation function that is a relational expression between the first characteristic value and the second characteristic value; and an estimation step of estimating the variable when the optimal solution is determined in the optimization step as the characteristic value (first method).

[0027] As a result, the variables of the state equation for calculating the characteristic value of the controlled object are changed within the search region to calculate an optimal solution of the evaluation function, and the variables at which the optimal solution is calculated can be estimated as the characteristic value of the controlled object. Since the evaluation function is a relational expression between the first characteristic value calculated from the frequency characteristic of the controlled object and the second characteristic value calculated using the variables changed within the search region and the state equation, the characteristic value can be easily estimated by calculating the optimal solution of the evaluation function.

[0028] Therefore, it is possible to realize a characteristic value estimation method that can easily estimate the characteristic value of the controlled object. [Effects of the Invention]

[0029] A characteristic value estimation device and a characteristic value estimation method according to an embodiment of the present invention calculate a first feature value of a controlled object at a predetermined frequency from the frequency characteristics of the controlled object, change a variable that affects the first feature value at the predetermined frequency within a search region to change a second feature value, and calculate an optimal solution within the search region of an evaluation function that is a relational expression between the first feature value and the second feature value.The characteristic value estimation device and the characteristic value estimation method then estimate the variable when the optimal solution is calculated as the characteristic value.

[0030] This provides a characteristic value estimation device and a characteristic value estimation method that can easily estimate the characteristic value of a controlled object. [Brief explanation of the drawings]

[0031] [Figure 1] FIG. 1 is a functional block diagram showing a schematic configuration of a characteristic value estimation device according to the first embodiment. [Figure 2] FIG. 2 is a diagram showing an example of a control target in the form of a model. [Figure 3] FIG. 3 is a Bode diagram showing an example of frequency characteristics of a controlled object expressed by a state equation. [Figure 4] FIG. 4 is a diagram showing the flow of the characteristic value estimation method. [Figure 5] FIG. 5 is a diagram showing an example of a search area. [Figure 6] FIG. 6 is a functional block diagram showing a schematic configuration of a characteristic value estimation device according to the second embodiment. [Figure 7] FIG. 7 is a flowchart illustrating a characteristic value estimation method according to the second embodiment. [Figure 8] FIG. 8 is a diagram showing an example of local regions obtained by dividing the search region. [Figure 9] FIG. 9 is a diagram showing an example of frequency characteristics obtained using variables when a local solution of the evaluation function is found. [Figure 10] FIG. 10 is a diagram showing an example of frequency characteristics when the optimal solution of the evaluation function is obtained by the method of the second embodiment. [Figure 11] FIG. 11 is a functional block diagram showing a schematic configuration of a characteristic value estimation device according to the third embodiment. [Figure 12] FIG. 12 is a diagram showing a schematic diagram of the relationship between the values ​​used in the evaluation function. [Figure 13] FIG. 13 shows an example of frequency characteristics when the moment of inertia, spring constant, and viscous damping coefficient differ between the actually measured values ​​and the calculated values. [Figure 14] FIG. 14 shows an example of frequency characteristics when the measured value and the calculated value of the viscous damping coefficient differ. [Figure 15]FIG. 15 shows an example of frequency characteristics when the measured spring constant differs from the calculated spring constant. DETAILED DESCRIPTION OF THE INVENTION

[0032] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to the accompanying drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals and the description thereof will not be repeated.

[0033] <Embodiment 1> 1 is a functional block diagram showing a schematic configuration of a characteristic value estimation device 1 according to a first embodiment of the present invention. The characteristic value estimation device 1 is a device for estimating a characteristic value of a control target W, and is configured by an arithmetic device such as a computer.

[0034] (Control object) The controlled object W is, for example, a multi-inertia system. FIG. 2 is a diagram showing an example of the controlled object W in the form of a dynamic characteristic model of a mechanical system. The controlled object W shown in FIG. 2 is a four-inertia system. The controlled object W includes a motor, a torque detector, a test object, etc. In the example shown in FIG. 2, j1 is the moment of inertia of the motor, and j2 to j4 are the moments of inertia of the torque detector, the test object, etc.

[0035] As shown in FIG. 2, structures having moments of inertia j1 ​​to j4 are connected by springs having spring constants k1 to k3 and dampers having viscous damping coefficients c1 to c3. A command torque u is input to a motor having a moment of inertia j1. The torsional torque generated between the structures in the controlled object W is expressed as T i =k i ∫(ω i -ω i+1 )dt. In the example shown in Figure 2, i=1, 2, 3.

[0036] In the control object W, when the torsional torque of the shaft between the structure having the moment of inertia j2 and the structure having the moment of inertia j3 can be detected by a sensor, an example of the state equation is as follows: Note that the state equation may be an equation other than the following one.

[0037]

number

[0038] 3 is a Bode diagram showing an example of the frequency characteristics of the controlled object W expressed by the above-mentioned state equation. As shown in FIG. 3, the controlled object W, which is a four-inertia system, generally has three resonances.

[0039] (characteristic value estimation device) As shown in FIG. 1, the characteristic value estimation device 1 includes a first characteristic value acquisition unit 11, a parameter setting unit 12, a second characteristic value calculation unit 13, an optimization processing unit 14, and an estimation unit 15.

[0040] The first feature value acquisition unit 11 obtains a resonant frequency and its gain from the measurement results of the frequency characteristics of the input and output of the controlled object W. Specifically, the first feature value acquisition unit 11 acquires three resonant frequencies and their gains in the Bode diagram shown in FIG. 3. The first feature value acquisition unit 11 may automatically acquire the resonant frequency and its gain of the controlled object W, or the resonant frequency and its gain of the controlled object W may be input to the first feature value acquisition unit 11. The first feature value acquisition unit 11 may have any configuration as long as it is capable of acquiring the resonant frequency and its gain of the controlled object W.

[0041] In this embodiment, the resonant frequency and its gain obtained from the frequency characteristics of the control object W by the first feature value acquisition unit 11 correspond to the first feature value. In this embodiment, the first feature value is an eigenvalue in the frequency characteristics of the control object W. The first feature value may include parameters other than the resonant frequency and its gain as long as they are feature values ​​of the control object W. In other words, the first feature value may be a value other than the eigenvalue. In this embodiment, the resonant frequency is the predetermined frequency in the present invention.

[0042] The parameter setting unit 12 acquires, for example, the moments of inertia j1 ​​to j4 and the spring constant k2 from design information, frequency characteristics, etc. of the controlled object W. The parameter setting unit 12 may automatically acquire the parameters of the controlled object W, or the parameters of the controlled object W may be input to the parameter setting unit 12. The parameter setting unit 12 may have any configuration as long as it is capable of acquiring the parameters of the controlled object W.

[0043] Furthermore, the parameter setting unit 12 sets a search region for variables that are the remaining parameters (e.g., spring constants k1 and k3, viscous damping coefficients c1 to c3) in the state equation of the controlled object W. More specifically, the parameter setting unit 12 sets a search region for the variables that are to be calculated by the optimization processing unit 14, which will be described later. The search region for the variables may be stored in advance in a storage unit (not shown) or the like, or may be input to the parameter setting unit 12.

[0044] The second feature value calculation unit 13 uses a state equation of the controlled object W to calculate the resonant frequency and its gain after the variables of the state equation are changed within the search region. The second feature value calculation unit 13 uses the state equation to calculate the resonant frequency and the gain when the variables are changed within the search region by an optimization processing unit 14, which will be described later. That is, the second feature value calculation unit 13 calculates the resonant frequency and the gain using the variables changed within the search region when the optimization processing unit 14 performs calculations for the optimization process.

[0045] In this embodiment, the resonant frequency and its gain calculated by the second feature value calculation unit 13 using the changed variables correspond to the second feature value. In this embodiment, the second feature value is an eigenvalue in the frequency characteristics of the control object W. The second feature value may include parameters other than the resonant frequency and its gain as long as they are feature values ​​of the control object W. In other words, the second feature value may include a value other than the eigenvalue. In this embodiment, the resonant frequency is the predetermined frequency in the present invention.

[0046] The optimization processing unit 14 changes the variables within the search area, and performs optimization processing to find an optimal solution of the evaluation function within the search area using the second feature value calculated by the second feature value calculation unit 13 using the changed variables, and the first feature value acquired by the first feature value acquisition unit 11. In detail, the optimization processing unit 14 uses the resonant frequency and its gain, which are the first feature value, and the resonant frequency and its gain, which are the second feature value, to find an optimal solution that minimizes the evaluation function within the search area, and also finds the variables at that time.

[0047] The evaluation function is a relational expression between the first feature value and the second feature value. That is, the evaluation function is a relational expression including the square of the difference between the first feature value and the second feature value. More specifically, the evaluation function is a relational expression including the square of the difference between the resonance frequency which is the first feature value and the resonance frequency which is the second feature value, and the square of the difference between the gain which is the first feature value and the gain which is the second feature value. In this embodiment, the evaluation function is a relational expression obtained by adding up feature values ​​of a plurality of modes (in this embodiment, three modes: first resonance, second resonance, and third resonance) of the controlled object W, taking into account a weighting coefficient for each mode. Specifically, the evaluation function is, for example, the following equation (1).

[0048]

number

[0049] Here, w1 to w6 are weighting coefficients, f1 to f3 are resonance frequencies which are first feature values, G(f1) to G(f3) are gains which are first feature values, fm1 to fm3 are resonance frequencies which are second feature values, and G(fm1) to G(fm3) are gains which are second feature values.

[0050] The optimization processing unit 14 can find a second feature value close to the first feature value by finding a value that minimizes the evaluation function when the variable is changed within the search range.

[0051] When the optimization processing unit 14 finds the minimum value of the evaluation function, the estimation unit 15 estimates the variable at that time as the characteristic value of the controlled object W.

[0052] (Characteristic value estimation method) Next, a method for estimating a characteristic value of the controlled object W using the characteristic value estimation device 1 having the above-described configuration will be described with reference to Fig. 4. Fig. 4 is a diagram showing the flow of the characteristic value estimation method.

[0053] 4 starts (START), first, in step SA1, the first feature value acquisition unit 11 acquires the frequency characteristics of the control object W. Then, in step SA2, the first feature value acquisition unit 11 calculates the resonant frequency and its gain from the frequency characteristics. That is, the first feature value acquisition unit 11 acquires the first feature value.

[0054] In the next step SA3, the parameter setting unit 12 acquires the moments of inertia j1 ​​to j4 and the spring constant k2 from the design information, frequency characteristics, etc. of the controlled object. That is, the parameter setting unit 12 acquires constants other than the variables. Thereafter, in step SA4, the parameter setting unit 12 determines a search region for the variables. FIG. 5 is a diagram showing an example of the search region. For the sake of explanation, FIG. 5 shows the search region when the spring constants k1 and k3 are changed as an example of the search region for the variables.

[0055] In step SA5, the optimization processing unit 14 changes the variables within the search region. For example, the optimization processing unit 14 sets the initial values ​​of the variables to the minimum values ​​within the search region. The second feature value calculation unit 13 calculates second feature values ​​using the changed variables and the state equation of the control object W. The optimization processing unit 14 performs optimization processing to find an optimal solution to the evaluation function using the first feature value acquired by the first feature value acquisition unit 11 and the second feature value acquired by the second feature value calculation unit 13. In detail, the optimization processing unit 14 uses the resonant frequency and its gain, which are the first feature values, and the resonant frequency and its gain, which are the second feature values, to find an optimal solution that minimizes the evaluation function and to find the variables at that time.

[0056] In step SA6, when the optimization processing unit 14 finds an optimal solution of the evaluation function, the estimation unit 15 estimates the variables at that time as the characteristic values ​​of the control object W. After that, this flow ends (END).

[0057] In the above flow, step SA2 corresponds to the first feature value acquisition step, step SA5 corresponds to the optimization processing step, and step SA6 corresponds to the estimation step.

[0058] As described above, the characteristic value estimation device 1 of this embodiment estimates a characteristic value of the controlled object W. The characteristic value estimation device 1 includes a first characteristic value acquisition unit 11 that calculates a first characteristic value of the controlled object W from the frequency characteristics of the controlled object W, a second characteristic value calculation unit 13 that calculates a second characteristic value of the controlled object W when a variable that affects the first characteristic value is changed within a search region using a state equation including the variable, an optimization processing unit 14 that changes the second characteristic value calculated by the second characteristic value calculation unit 13 by changing the variable within the search region, and calculates an optimal solution within the search region of an evaluation function that is a relational expression between the first characteristic value and the second characteristic value, and an estimation unit 15 that estimates the variable when the optimal solution is calculated by the optimization processing unit 14 as the characteristic value.

[0059] As a result, the variables of the state equation for finding the characteristic values ​​of the controlled object W can be changed within the search region to find an optimal solution to the evaluation function, and the variables when the optimal solution is found can be estimated as the characteristic values ​​of the controlled object W. Since the evaluation function is a relational expression between the first characteristic value found from the frequency characteristics of the controlled object W and the second characteristic value calculated using the variables changed within the search region and the state equation, the characteristic values ​​can be easily estimated by finding the optimal solution of the evaluation function.

[0060] Therefore, with the above-described configuration, the characteristic value estimation device 1 that can easily estimate the characteristic value of the controlled object W is obtained.

[0061] In this embodiment, the evaluation function is a relational expression including the square of the difference between the first feature value and the second feature value. The optimization processing unit 14 determines, as the optimal solution, the value of the evaluation function when the square of the difference between the first feature value and the second feature value is minimum.

[0062] By squaring the difference between the first feature value and the second feature value, the difference between the first feature value and the second feature value can be found regardless of whether the first feature value and the second feature value are large or small. The optimization processing unit 14 determines the value of the evaluation function when the square of the difference between the first feature value and the second feature value is smallest as the optimal solution of the evaluation function, and therefore can easily find the optimal solution.

[0063] Therefore, with the above-described configuration, the characteristic value estimation device 1 that can easily estimate the characteristic value of the controlled object W is obtained.

[0064] In this embodiment, the first characteristic value and the second characteristic value each include a vibration frequency and a gain, so that the characteristic value of the control object W, such as a spring constant, can be calculated using the vibration frequency and the gain.

[0065] <Embodiment 2> 6 is a functional block diagram showing a schematic configuration of a characteristic value estimation device 100 according to the second embodiment. The characteristic value estimation device 100 differs in configuration from the characteristic value estimation device 1 of the first embodiment in that the search region for variables is divided into multiple regions. In the following explanation, the same components as those in the first embodiment are denoted by the same reference numerals and explanations thereof will be omitted, and only the components different from those in the first embodiment will be explained.

[0066] (characteristic value estimation device) As shown in FIG. 6, the characteristic value estimation device 100 includes a first characteristic value acquisition unit 11, a parameter setting unit 120, a second characteristic value calculation unit 13, an optimization processing unit 140, and an estimation unit 15.

[0067] In this embodiment, the first feature value is a resonant frequency and its gain determined from the frequency characteristics of the controlled object W. The first feature value may be an eigenvalue or a value other than an eigenvalue. The second feature value is a resonant frequency and its gain determined from the frequency characteristics of the controlled object W. The second feature value may be an eigenvalue or a value other than an eigenvalue. The resonant frequencies in the first feature value and the second feature value are predetermined frequencies in the present invention.

[0068] The parameter setting unit 120 has a search area dividing unit 121. The search area dividing unit 121 obtains a plurality of local areas by dividing the search area of ​​a variable into a plurality of parts. The search area dividing unit 121 may divide the search area into a predetermined number of parts, or may divide the search area into a plurality of parts according to the range of the set variable. The search area dividing unit 121 may divide the search area into a plurality of local areas according to inputs such as the number of divisions and the division range.

[0069] The parameter setting unit 120, like the parameter setting unit 12 of the first embodiment, acquires, for example, the inertia moments j1 to j4 and the spring constant k2 from the design information and frequency characteristics of the controlled object W.

[0070] The optimization processing unit 140 changes variables in each local region obtained by the parameter setting unit 120, and performs optimization processing to find an optimal solution of the evaluation function using second feature values ​​calculated by the second feature value calculation unit 13 using the changed variables in each local region and first feature values ​​acquired by the first feature value acquisition unit 11. That is, the optimization processing unit 140 finds an optimal solution of the evaluation function in each local region. The optimization processing unit 140 finds an optimal solution in the search region from the optimal solutions of the evaluation function calculated in each local region.

[0071] Specifically, the optimization processing unit 140 includes a local area optimization processing unit 141 and an optimal solution selection unit 142 .

[0072] The local region optimization processing unit 141 uses the second feature value obtained by changing variables in each local region and the first feature value to find an optimal solution of the evaluation function in each local region. The local region optimization processing unit 141 may find optimal solutions of the evaluation function in multiple local regions sequentially, or may find optimal solutions of the evaluation function in at least some of the multiple local regions simultaneously. The optimal solution is the minimum value of the evaluation function in each local region.

[0073] The evaluation function is the same as the evaluation function in embodiment 1. The method for finding an optimal solution of the evaluation function in each local region is the same as the method for finding an optimal solution of the evaluation function in the search region in embodiment 1. Therefore, a detailed description of the method for finding an optimal solution in each local region will be omitted.

[0074] The optimal solution selection unit 142 selects the optimal solution in the search area from the optimal solutions of the evaluation function in each of the local areas obtained by the local area optimization processing unit 141. In this embodiment, the optimal solution selection unit 142 selects the smallest value from the optimal solutions of the evaluation function in each of the local areas as the optimal solution in the search area. The optimal solution selection unit 142 outputs variables when the optimal solution in the search area is obtained to the estimation unit 15. The estimation unit 15 estimates the variables as characteristic values ​​of the control object W.

[0075] (Characteristic value estimation method) Next, a method for estimating a characteristic value of a controlled object W using the characteristic value estimation device 100 having the above-described configuration will be described with reference to Fig. 7. Fig. 7 is a diagram showing a flow of a characteristic value estimation method according to the second embodiment.

[0076] 7 starts (START), first, in step SB1, the first characteristic value acquisition unit 11 acquires the frequency characteristics of the control object W. Then, in step SB2, the first characteristic value acquisition unit 11 calculates the resonant frequency and its gain from the frequency characteristics. That is, the first characteristic value acquisition unit 11 acquires the first characteristic value.

[0077] In the next step SB3, the parameter setting unit 120 acquires the moments of inertia j1 ​​to j4 and the spring constant k2 from the design information, frequency characteristics, etc. of the controlled object. That is, the parameter setting unit 120 acquires constants other than the variables. Thereafter, in step SB4, the parameter setting unit 120 determines the search region for the variables. Note that steps SB1 to SB4 are similar to steps SA1 to SA4 in the flow shown in FIG. 4 of the first embodiment.

[0078] In step SB5, the search area dividing unit 121 of the parameter setting unit 120 divides the search area into a plurality of local areas. Fig. 8 is a diagram showing an example of the local areas obtained by dividing the search area. For the sake of explanation, Fig. 8 shows an example of the search area when the spring constants k1 and k3 are changed.

[0079] In step SB6, the local region optimization processing unit 141 of the optimization processing unit 140 changes the variables in each local region. For example, the local region optimization processing unit 141 sets the initial value of the variable to the minimum value in each local region. The second feature value calculation unit 13 calculates the second feature value using the changed variable and the state equation of the control object W. The local region optimization processing unit 141 performs optimization processing to find an optimal solution of the evaluation function in each local region using the first feature value acquired by the first feature value acquisition unit 11 and the second feature value acquired by the second feature value calculation unit 13.

[0080] Specifically, the local region optimization processing unit 141 uses the resonance frequency and its gain, which are the first feature value, and the resonance frequency and its gain, which are the second feature value, to find an optimal solution that minimizes the evaluation function in each local region. Then, in the subsequent step SB7, the optimal solution selection unit 142 of the optimization processing unit 140 determines the smallest value among the optimal solutions in each local region as the optimal solution in the search region, and finds the variables at that time.

[0081] In step SB8, when the optimization processing unit 140 finds an optimal solution of the evaluation function in the search region, the estimation unit 15 estimates the variables at that time as the characteristic values ​​of the control object W. Then, this flow ends (END).

[0082] In the above-described flow, step SB2 corresponds to the first characteristic value acquisition step, steps SB6 and SB7 correspond to the optimization processing step, and step SB8 corresponds to the estimation step.

[0083] However, when an optimal solution of the evaluation function is obtained by changing variables over the entire search region as in the first embodiment, a local solution may be obtained instead of the optimal solution of the evaluation function. Fig. 9 is a diagram showing an example of frequency characteristics obtained using variables when a local solution of the evaluation function is obtained. As shown in Fig. 9, the frequency characteristics obtained using variables when a local solution of the evaluation function is obtained are significantly different from the measurement results of the actual frequency characteristics. In other words, the obtained variables are different from the actual values.

[0084] In contrast to this, by dividing the search region for variables into multiple local regions as in this embodiment, it is possible to obtain frequency characteristics close to the measurement results, as shown in Fig. 10. Fig. 10 is a diagram showing an example of frequency characteristics when the optimal solution of the evaluation function is obtained by the method of embodiment 2. Note that Fig. 10 shows an example of the case where the spring constants k1 and k3 are not divided, but the viscous damping coefficients c1 to c3 are each divided into five to obtain the optimal solution.

[0085] In this way, the configuration of this embodiment makes it possible to accurately find the optimal solution of the evaluation function in the search region, and therefore to accurately estimate the characteristic value of the controlled object W.

[0086] In this embodiment, the search region has a plurality of divided local regions. The optimization processing unit 140 changes the variables of the state equation in each of the plurality of local regions to change the second feature value calculated by the second feature value calculation unit 13, thereby obtaining an optimal solution of the evaluation function. The estimation unit 15 determines the optimal solution of the evaluation function in the search region to be the optimal solution of the evaluation function in the search region, among the optimal solutions of the evaluation function obtained in each local region by the optimization processing unit 140.

[0087] This allows the optimum solution of the evaluation function to be found in each of the local regions obtained by dividing the search region of the variables into multiple regions. Then, by finding the optimum solution for the search region from among the optimum solutions for each local region, it is possible to find an optimum solution for the entire search region, rather than a local optimum solution. Therefore, the optimum solution of the evaluation function in the search region can be found with high accuracy. Therefore, the characteristic value of the controlled object W can be found with high accuracy.

[0088] Moreover, with the above-described configuration, the characteristic value can be estimated even if the search region is wide, so the characteristic value can be found with high accuracy without adjusting the range of the search region and without finding a local solution.

[0089] On the other hand, if the search region for variables is not divided into multiple regions, there is a high possibility of finding a local solution, and therefore the search region must be changed through trial and error. While it is possible to avoid the calculation of a local solution by increasing the number of optimization iterations or tightening the criteria for convergence evaluation of the evaluation function, this requires time for calculation. In contrast, by dividing the search region into multiple local regions as in the above configuration, it is possible to find an optimal solution without trial and error or changing the criteria for convergence evaluation. Therefore, by dividing the search region into multiple regions as described above and finding an optimal solution of the evaluation function in each local region, the calculation load of the characteristic value estimation device 100 can be reduced.

[0090] <Embodiment 3> 11 is a functional block diagram showing a schematic configuration of a characteristic value estimation device 200 according to the third embodiment. The characteristic value estimation device 200 differs in configuration from the characteristic value estimation device 1 of the first embodiment in that it uses data of not only the resonant frequency but also other frequencies. In the following explanation, the same components as those of the first embodiment are denoted by the same reference numerals and explanations thereof will be omitted, and only the components different from those of the first embodiment will be explained.

[0091] The characteristic value estimation device 200 includes a first characteristic value acquisition unit 211 , a parameter setting unit 12 , a second characteristic value calculation unit 213 , an optimization processing unit 214 , and an estimation unit 15 .

[0092] The first characteristic value acquisition unit 211 acquires the resonant frequency and its gain of the controlled object W from the measurement results of the frequency characteristics of the input and output of the controlled object W, acquires the gain of the controlled object W at a desired frequency, and also acquires the gain of the controlled object W at a frequency near the resonant frequency. The first characteristic value acquisition unit 211 has a first acquisition unit 211a and a second acquisition unit 211b.

[0093] The first acquisition unit 211a acquires the resonant frequency and its gain of the controlled object W. The second acquisition unit 211b acquires the gain of the controlled object W at the desired frequency, and also acquires the gain of the controlled object W at frequencies near the resonant frequency. In this embodiment, the desired frequency and the near frequencies are predetermined frequencies in the present invention.

[0094] In this embodiment, the resonant frequency and gain acquired by the first acquisition unit 211a and the second acquisition unit 211b correspond to the first feature value.

[0095] The second characteristic value calculation unit 213 uses the state equation of the controlled object W to calculate the resonant frequency and its gain after changing the variables of the state equation within the search region, and also calculates the gain of the controlled object W at the desired frequency and the gain of the controlled object W at frequencies near the resonant frequency. The second characteristic value calculation unit 213 has a first calculation unit 213a and a second calculation unit 213b.

[0096] The first calculation unit 213a uses the state equation of the controlled object W to find the resonant frequency and its gain after changing the variables of the state equation within the search region.

[0097] The second calculation unit 213b calculates the gain of the controlled object W at the desired frequency and also calculates the gain of the controlled object W at frequencies near the resonant frequency, using the state equation of the controlled object W. In this embodiment, the desired frequency and the near frequencies are predetermined frequencies in the present invention.

[0098] In this embodiment, the resonant frequency and gain calculated by the first calculation unit 213a and the second calculation unit 213b correspond to the second characteristic value.

[0099] The optimization processing unit 214 changes the variables of the state equation of the control object W within the search region, and performs optimization processing to find an optimal solution of the evaluation function within the search region using the second feature value calculated by the second feature value calculation unit 213 using the changed variables, and the first feature value acquired by the first feature value acquisition unit 11. In detail, the optimization processing unit 214 uses the resonant frequency and gain, which are the first feature values, and the resonant frequency and gain, which are the second feature values, to find an optimal solution that minimizes the evaluation function within the search region, and also finds the variables at that time.

[0100] The evaluation function is a relational expression between the first feature value and the second feature value. That is, the evaluation function is a relational expression including the square of the difference between the first feature value and the second feature value. More specifically, the evaluation function is a relational expression including the square of the difference between the resonance frequency, which is the first feature value, and the resonance frequency, which is the second feature value, and the square of the difference between the gain, which is the first feature value, and the gain, which is the second feature value. Specifically, the evaluation function is, for example, the following equation (2):

[0101]

number

[0102] Here, w1 to w5 are weighting coefficients, fp is the resonance frequency which is the first feature value, fpm is the resonance frequency which is the second feature value, ε is the desired frequency, fpl and fph are frequencies near the resonance frequency, G(fp), G(ε), G(fpl), and G(fph) are gains which are the first feature value, and Gm(fpm), Gm(ε), Gm(fpl), and Gm(fph) are gains which are the second feature value. In this embodiment, fpl is a frequency lower than the resonance frequency, and fgh is a frequency higher than the resonance frequency. Figure 12 shows the relationship between the values ​​in the evaluation function. In the example shown in Figure 12, the desired frequency is a frequency lower than the resonance frequency.

[0103] In equation (2), w3(G(ε)-Gm(ε)) 2 performs fitting at the desired frequency, and w4(G(fpl)-Gm(fpl)) 2 +w5(G(fph)-Gm(fph)) 2 can improve the accuracy of the fit for the decay of the resonance.

[0104] The method of finding the optimum solution using the evaluation function in the optimization processing unit 214 is the same as in embodiments 1 and 2. Therefore, a detailed description of the calculation method for finding the optimum solution of the evaluation function will be omitted.

[0105] The configuration of this embodiment makes it possible to perform fitting at the desired frequency while improving the accuracy of fitting regarding resonance damping, thereby making it possible to estimate the characteristic value of the controlled object W easily and with higher accuracy.

[0106] The evaluation function is w3(G(ε)-Gm(ε)) 2 may not be included, and w4(G(fpl)-Gm(fpl)) 2 +w5(G(fph)-Gm(fph)) 2may not be included. Also, ε may be a frequency higher than the resonant frequency. The evaluation function may be an evaluation function that can accommodate a plurality of resonant frequencies, like the evaluation function in the first embodiment.

[0107] Figure 13 shows an example of frequency characteristics when the moment of inertia, spring constant, and viscous damping coefficient differ between the measured and calculated values. In Figure 13, the resonance frequency and its gain match between the measured and calculated values, but the gains of other frequencies differ between the measured and calculated values. In the case shown in Figure 13, the calculated values ​​can be fitted to the measured values ​​by using the above-mentioned equation (2).

[0108] In the case of FIG. 13, in equation (2), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 In addition to w3(G(ε)-Gm(ε)) 2 , w4(G(fpl)-Gm(fpl)) 2 , w5(G(fph)-Gm(fph)) 2 In the case of FIG. 13, an equation including at least one of the following may be used as the evaluation function. In addition, in the case of FIG. 13, in equation (2), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 is not included, and w3(G(ε)-Gm(ε)) 2 , w4(G(fpl)-Gm(fpl)) 2 , w5(G(fph)-Gm(fph)) 2 An expression including the above may be used as the evaluation function.

[0109] Figure 14 shows an example of frequency characteristics when the viscous damping coefficient differs between the measured value and the calculated value. In Figure 14, the measured value and the calculated value of the resonance frequency are the same, but the measured value and the calculated value of the gain at the resonance frequency are different. In the case shown in Figure 14, the calculated value can be fitted to the measured value by using the following equation (3).

[0110]

number

[0111] In the case of FIG. 14, in equation (3), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 In addition to w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 In the case of FIG. 14, an equation including either one of the following may be used as the evaluation function. In addition, in the case of FIG. 14, in equation (3), W1(G(fp)-Gm(fpm)) 2 In addition to w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 In the case of FIG. 14, an equation including either one of the following may be used as the evaluation function. In addition, in the case of FIG. 14, in equation (3), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 is not included, and w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 An expression including the above may be used as the evaluation function.

[0112] Fig. 15 shows an example of frequency characteristics when the measured and calculated spring constants are different. In Fig. 15, the measured and calculated resonant frequencies are different, but the measured and calculated gains at the resonant frequencies are the same. In the case shown in Fig. 15, the calculated values ​​can be fitted to the measured values ​​by using equation (3).

[0113] In the case of FIG. 15, in equation (3), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 In addition to w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 In the case of FIG. 15, an equation including either W2(fp-fpm) in equation (3) may be used as the evaluation function. 2is not included, W1(G(fp)-Gm(fpm)) 2 In addition to w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 In the case of FIG. 15, an equation including either one of the following may be used as the evaluation function. In addition, in the case of FIG. 15, in equation (3), W1(G(fp)-Gm(fpm)) 2 , W2(fp-fpm) 2 is not included, and w3(G(fpl)-Gm(fpl)) 2 , w4(G(fph)-Gm(fph)) 2 An expression including the above may be used as the evaluation function.

[0114] (Other embodiments) Although the embodiments of the present invention have been described above, the above-described embodiments are merely examples for carrying out the present invention. Therefore, the present invention is not limited to the above-described embodiments, and it is possible to appropriately modify the above-described embodiments within the scope of the spirit of the present invention.

[0115] In the second embodiment, the characteristic value estimation device 100 divides each of the viscous damping coefficients c1 to c3 into five without dividing the spring constants k1 and k3, and finds an optimal solution of the evaluation function. However, the characteristic value estimation device may divide all of the spring constants and viscous damping coefficients, or may divide at least one of the spring constants and viscous damping coefficients.

[0116] In each of the above embodiments, the controlled object W is a four-inertia system and includes a motor, a torque detector, a test object, etc. However, the controlled object W may be a multi-inertia system other than a four-inertia system, such as a three-inertia system or a five-inertia system, or may be a two-inertia system. Furthermore, the controlled object W may include a configuration other than those described above, or may include an axis system having a configuration other than those described above.

[0117] In each of the above embodiments, the first feature value is a resonant frequency and a gain. The second feature value is a resonant frequency and a gain. However, the first feature value may be only the resonant frequency. The second feature value may be only the resonant frequency. Furthermore, the first feature value may include a feature value other than the resonant frequency and the gain of the controlled object W (such as a current response of a motor), or may be only that feature value. The second feature value may include a feature value other than the resonant frequency and the gain of the controlled object W (such as a current response of a motor), or may be only that feature value.

[0118] In each of the above-described embodiments, the characteristic value estimation device 1, 100 estimates the spring constants k1, k3 of the controlled object W. However, the characteristic value estimation device may estimate other parameters.

[0119] In each of the above embodiments, the optimization processor 14, 140 determines the minimum value of the evaluation function in the search region of the variables as the optimal solution. However, the optimization processor may determine another value, such as the maximum value of the evaluation function in the search region, as the optimal solution. In this case, instead of the evaluation function of embodiment 1, an evaluation function that can determine the other value as the optimal solution may be used. The evaluation function used in the optimization processor may be an evaluation function different from the evaluation function used in each of the above embodiments.

[0120] In each of the above embodiments, the optimization processor 14, 140 sets the initial value of the variable to the minimum value in the search region or local region. However, the optimization processor may set the initial value of the variable to another value in the search region or local region.

[0121] In each of the above embodiments, the spring constants k1 and k3 are varied as an example of the search region, but the search region may be varied by varying other parameters that are variables of the state equation of the controlled object. [Industrial Applicability]

[0122] The present invention can be used in a characteristic value estimation device and a characteristic value estimation method for estimating a characteristic value of a controlled object. [Explanation of symbols]

[0123] 1, 100, 200 characteristic value estimation device 11, 211 First feature value acquisition unit 12, 120 Parameter setting section 13, 213 Second feature value calculation unit 14, 140, 214 Optimization processing section 15 Estimation part 121 Search area division part 141 Local area optimization processing unit 142 Optimal Solution Selection Unit 211a 1st acquisition part 211b 2nd acquisition part 213a 1st calculation section 213b 2nd calculation section W Control target

Claims

1. A characteristic value estimation device that estimates a characteristic value of a controlled object, a first characteristic value acquisition unit that obtains a first characteristic value of the control object at a predetermined frequency from the frequency characteristics of the control object; a second characteristic value calculation unit that calculates a second characteristic value of the controlled object when the variable is changed within a search region at the predetermined frequency using a state equation including a variable that affects the first characteristic value; an optimization processing unit that changes the variable within the search region to change the second feature value calculated by the second feature value calculation unit, and obtains an optimal solution within the search region of an evaluation function that is a relational expression between the first feature value and the second feature value; an estimation unit that estimates the variables when an optimal solution is obtained by the optimization processing unit as the characteristic values; and the evaluation function is a relational expression including a square of the difference between the first feature value and the second feature value, the optimization processing unit determines, as the optimal solution, a value of the evaluation function when a square of a difference between the first feature value and the second feature value is minimum. Characteristic value estimation device.

2. 2. The characteristic value estimation device according to claim 1, the first characteristic value and the second characteristic value each include at least a gain; Characteristic value estimation device.

3. 3. The characteristic value estimation device according to claim 1, the search region is made up of a plurality of local regions, The optimization processing unit a local region optimization processing unit that changes the variables of the state equation in each of the local regions to change the second feature value calculated by the second feature value calculation unit, thereby obtaining an optimal solution of the evaluation function; an optimal solution selection unit that selects an optimal solution from among the optimal solutions of the evaluation function obtained in each local region by the local region optimization processing unit as an optimal solution of the evaluation function in the search region; having Characteristic value estimation device.

4. A characteristic value estimation method for estimating a characteristic value of a controlled object, comprising: a first feature value obtaining step of obtaining a first feature value of the control object at a predetermined frequency from the frequency characteristics of the control object; an optimization processing step of calculating a second feature value of the controlled object when the variable is changed within a search region at the predetermined frequency using a state equation including a variable that affects the first feature value, and obtaining an optimal solution within the search region of an evaluation function that is a relational expression between the first feature value and the second feature value; an estimation step of estimating the variables when an optimal solution is obtained in the optimization processing step as the characteristic values; and the evaluation function is a relational expression including a square of the difference between the first feature value and the second feature value, the optimization processing step determines, as the optimal solution, the value of the evaluation function when a square of the difference between the first feature value and the second feature value is minimum. Property value estimation method.

Images