JP2014178242A - Noise reduction device of time-series measurement signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To appropriately reduce noise while suppressing time lag of an output signal even when a measurement signal by a load cell is rapidly changed.SOLUTION: An observation noise parameter calculation unit 13 calculates a standard deviation of a plurality of measurement signals obtained when calibrating a device having no test piece mounted thereon and stores it in a storage unit 14 as a parameter σrelating to the observation noise. During executing the measurement, a processing noise parameter calculation unit 12 acquires a gradient of the change based on a prescribed number of latest data values and assigns it to a Kalman filter 11 as a parameter σrelating to the processing noise. The Kalman filter 11 executes a correction based on an error between a state estimation using the parameters σand σand an actual measurement signal to smooth and output the measurement signal. The parameter σis dynamically changed according to a change in the measurement signal so as to appropriately reduce the noise while substantially following to the rapid change in the measurement signal.

Description

  The present invention relates to a noise reduction device that reduces noise superimposed on measurement signals obtained sequentially with time. The noise reduction device according to the present invention can be applied to a measurement signal obtained by, for example, a load cell attached to a material testing machine, an electronic balance, or the like.

  In a test using a material testing machine, generally, a load such as tension or compression is continuously applied to a test piece, and a test force acting on the test piece due to the load or a deformation such as elongation or shrinkage of the test piece is applied. Measurements are made every moment, and the characteristics of the materials are evaluated based on the test force and deformation measurement results obtained. In the material testing machine, the test force acting on the test piece is detected by a load cell, and the deformation of the test piece is detected by a strain gauge. At this time, for example, noise due to various factors is superimposed on the measurement signal from the load cell. Therefore, in order to accurately obtain the test force, it is necessary to remove the superimposed noise as much as possible.

  Conventionally, in order to remove noise superimposed on a load cell measurement signal, a smoothing process using a moving average is performed (see, for example, Patent Document 1). The moving average is an average of a plurality of signal values included in the moving average length. Although the smoothing effect increases as the moving average length increases, it becomes impossible to follow a sudden change in the measurement signal. Therefore, in order to follow a rapid change in the load cell measurement signal, an adaptive calculation process is usually performed in which the moving average length is changed according to the magnitude of the change in the obtained measurement signal.

  However, even if the moving average length is adaptively changed, an appropriate moving average length is not always set. Therefore, when the set moving average length is not appropriate, smoothing is not sufficiently performed, and as shown in FIG. 9A, noise superimposed on the load cell measurement signal may remain considerably. Further, if the moving average length is increased in order to enhance the noise removal effect, the smoothing effect is increased, but not only it becomes impossible to follow a sudden change in the measurement signal, but also an output as shown in FIG. 9B. A signal delay occurs in the signal and affects the measurement result. In a material testing machine, it is generally necessary to suppress this delay to 50 [msec] or less. However, when it is attempted to sufficiently reduce noise by a moving average, the signal delay may exceed the required value. For this reason, there is a need for a new method for more accurately reducing noise superimposed on the measurement signal by following a sudden change in the measurement signal.

  In addition to the measurement signal obtained by the load cell, the same noise is superimposed on the measurement signal obtained by a measurement device such as an electronic balance, for example, and there is a similar problem.

JP 2001-330468 A JP-A-11-183242 (paragraph [0008], FIG. 9 etc.) Japanese Patent Laid-Open No. 2002-6898

  Real-time processing is required for noise reduction processing of load cell measurement signals. A Kalman filter is known as a method for estimating a smoothed signal at a time point when a measurement signal is obtained in substantially real time. For example, Patent Document 2 and the like remove a natural vibration superimposed on a measurement signal using the Kalman filter. Is disclosed. The Kalman filter is a filter that calculates an estimated state quantity from time to time based on a measurement signal disturbed by noise so that the error is minimized, and is mainly process noise (also referred to as system noise) generated in the system itself. The degree of smoothing is adjusted by parameters relating to the observation noise superimposed on the measurement signal.

  However, even when the Kalman filter is used, if the process noise parameter and the observation noise parameter are set so that the noise superimposed on the measurement signal is sufficiently removed, the Kalman filter passes when the measurement signal changes suddenly. The problem that the output signal cannot follow the original measurement signal, that is, the time delay becomes unavoidable.

  The present invention has been made to solve the above-described problems, and the object of the present invention is to suppress noise superimposed on the measurement signal while suppressing a time delay in output even when the measurement signal changes suddenly. An object of the present invention is to provide a noise reduction device for time-series measurement signals that can be accurately removed.

  In various filters including the Kalman filter, so-called adaptive control is known in which filtering characteristics (specifically, filter coefficients and the like) are changed in accordance with the characteristics of an input signal. For example, in Patent Document 3, in an apparatus for reducing noise superimposed on an audio signal by a Kalman filter, information on the input audio signal is used for the purpose of preventing noise from being appropriately removed and preventing deterioration of the audio signal itself as much as possible. And the strength of the filter is changed. The inventor of the present application has come up with the idea of introducing adaptive control in order to satisfactorily remove the noise superimposed on the measurement signal and ensure that the output signal follows the sudden change in the measurement signal. Then, by repeating the simulation calculation based on the data obtained by actual measurement, a method suitable for application to a measurement signal or the like by the load cell has been found and the present invention has been achieved.

That is, the present invention made to solve the above problems is a noise reduction device for reducing noise superimposed on a discretized measurement signal input in time series order,
a) calculating a parameter value related to observation noise described later based on a plurality of measurement signals input under conditions where the measurement signal does not change or a plurality of measurement signals obtained before the input measurement signal changes; One parameter calculation unit;
b) a second parameter calculation unit that calculates a parameter value related to process noise, which will be described later, in accordance with the degree of change in a predetermined number of measurement signals in the past each time a measurement signal is input in chronological order;
c) Each time a measurement signal is input in chronological order, a parameter value related to observation noise obtained by the first parameter calculation unit and a parameter value related to process noise obtained by the second parameter calculation unit for the measurement signal A Kalman filter that executes a smoothing process based on
It is characterized by having.

  Here, the “discretized measurement signal input in chronological order” that is the object of noise reduction is typically a measurement signal initially such as a measurement signal from a load cell or a measurement signal from an electronic balance. The measurement signal is digitized such that the level changes abruptly or slowly in one direction (in the direction of increasing or decreasing) at a certain point from the state in which the level of is maintained substantially constant.

  In the present invention, the Kalman filter is used to smooth the measurement signal and remove the noise superimposed on the measurement signal. The Kalman filter estimates the current system state and corrects the estimated value based on the actual observation value. The parameter value related to at least the process noise used in the process is changed according to the degree of change in the predetermined number of measurement signals in the past each time a discretized measurement signal is input. As a result, the degree of smoothing by the Kalman filter dynamically changes according to the amount of change in the measurement signal. Specifically, the greater the amount of change in the measurement signal, that is, the more rapid the change, the lower the degree of smoothing by the Kalman filter. As a result, even when the measurement signal changes abruptly, the time delay of the output signal smoothed with respect to the measurement signal is reduced while maintaining the noise removal effect.

  In the noise reduction device according to the present invention, for example, the second parameter calculation unit has a configuration in which an average inclination of a measurement signal obtained by linear approximation with respect to a predetermined number of measurement signals in the past is used as a parameter related to process noise. Can do. Here, the “predetermined number” may be experimentally determined in advance.

  On the other hand, the first parameter calculation unit calculates a standard deviation of a plurality of measurement signals input under conditions where the measurement signal does not change or a plurality of measurement signals obtained before the input measurement signal changes. The obtained standard deviation can be used as a parameter value related to observation noise.

The first parameter calculation unit may be configured to calculate a fixed parameter value related to observation noise based on a plurality of measurement signals obtained at the time of device calibration performed prior to measurement of the target sample. Specifically, when the noise of the measurement signal obtained by the load cell attached to the test machine that performs mechanical measurement on the sample is reduced, the first parameter calculation unit is implemented without mounting the sample on the test machine. What is necessary is just to set it as the structure which calculates the fixed parameter value regarding observation noise based on the several measurement signal obtained at the time of the apparatus calibration performed.
According to this configuration, since it is not necessary to calculate the parameter value related to the observation noise every moment during the measurement, the processing load can be reduced. For example, relatively inexpensive and inexpensive hardware ( Processing is also possible with a computer or a digital signal processor.

  On the other hand, the first parameter calculation unit may be configured to calculate a parameter value related to observation noise based on a plurality of measurement signals obtained during a period before the measurement signals input in time series are changed. . In this configuration, unlike the above configuration, a fixed parameter value related to observation noise is not calculated before actual measurement execution, such as during device calibration. Therefore, the time required for apparatus calibration can be shortened.

In the noise reduction device according to the present invention,
The second parameter calculation unit uses, as a parameter relating to the first process noise, an average inclination of the measurement signal obtained by linear approximation to a predetermined number of measurement signals in the past.
A third parameter calculation unit that calculates a parameter value related to the second process noise according to the average change in inclination;
The Kalman filter has three types of system states based on a parameter value related to observation noise, a parameter value related to the first process noise, and a parameter value related to the second process noise every time measurement signals are input in time series. It is good also as a structure which performs the smoothing process which estimated this.

  According to this configuration, although the processing is complicated, the information corresponding to the differential value of the change amount of the measurement signal is reflected in the characteristics of the Kalman filter, so that the output signal particularly when the measurement signal changes abruptly in a curve Followability and noise reduction effect are further improved.

According to the noise reduction device for time-series measurement signals according to the present invention, for example, the degree of smoothing is dynamically adjusted according to the amount of change in the measurement signal such as the test force signal by the load cell. Even when it changes, it is possible to follow the change and sufficiently reduce the noise superimposed on the measurement signal over the entire measurement time without a large time delay.
In addition, it is not necessary for the user to set parameter values related to observation noise and process noise by trial and error each time measurement is performed, which reduces the burden on the user and allows the user to set inappropriate parameter values by mistake. It is also possible to prevent wasteful data collection due to the error.

The schematic block diagram of the load cell measurement signal noise reduction apparatus which is one Example of this invention. The conceptual diagram of the parameter setting used for a Kalman filter. The flowchart of the pre-processing in the noise reduction apparatus of a present Example. The flowchart of the noise reduction process performed in real time in the noise reduction apparatus of a present Example. The figure which shows an example of the measurement signal (test force) at the time of performing the noise reduction process by a present Example, and the output after a process. The figure which shows an example of the measurement signal (test force) at the time of performing the noise reduction process by a present Example, and the output after a process. The figure which shows an example of the measurement signal (test force) at the time of performing the noise reduction process by a present Example, and the output after a process. The figure which shows an example of the comparison with the output at the time of performing the noise reduction process by a present Example, and the output by a conventional method. The figure which shows an example of the measurement signal (test force) at the time of using the conventional moving average, and the output after a process.

Hereinafter, a load cell measurement signal noise reduction apparatus according to an embodiment of the present invention will be described with reference to the accompanying drawings.
An analog measurement signal (test force) that is an output of the load cell 2 equipped in a tensile tester (not shown) is input to an analog-digital converter (ADC) 3 and sampled at a predetermined time interval in the analog-digital converter 3. And converted into digital data. This digital data is input to the noise reduction device 1, and random noise superimposed on the original measurement signal is reduced and output. The noise reduction apparatus 1 includes a Kalman filter 11, a process noise parameter calculation unit 12, an observation noise parameter calculation unit 13, an observation noise parameter storage unit 14, and the like. The Kalman filter 11 includes a prediction calculation unit 111, a correction calculation unit 112, and the like.

  Prior to describing the operation of the load cell measurement signal noise reduction apparatus of the present embodiment, first, an outline of smoothing processing using a Kalman filter will be described.

The Kalman filter is an algorithm that calculates a least-square estimation value of a state vector of a linear system that receives Gaussian white noise sequentially, that is, every time a new observation value is obtained, based on an observation value disturbed by noise.
Now, it is assumed that the true system state x _k at a certain measurement time point k is expressed by the following state equation using the system state x _k-1 one time before the measurement time point (this is k−1). .
x _k = F _k x _k-1 + w _k (1)
Here, F _k is a linear model related to the time transition of the system, w _k is noise (process noise) related to the time transition, and an approximate value thereof is N (0, Q _k ). As will be described later, Q _k is a process noise covariance matrix.

At time k, the observed value z _k has the following relationship with the true system state x _k .
z _k = H _k x _k + v _k (2)
Here, H _k is a mapping model from the state space to the observation space, and v _k is an observation noise superimposed on the observation value. Equation (2) is generally called an observation equation.

In the Kalman filter, two kinds of estimated values of the system state, that is, <x _{k | k-1} >, which is a state estimated value (prior state estimated value) at the measurement time k at the measurement time _k−1 , and the measurement time <x _{k | k} > which is a state estimated value (post-state estimated value) at a measurement time point k obtained by correcting the prior state estimated value <x _{k | k-1} > based on the observed value z _k at _k . Handle. These two types of system state estimates can be related by the following equations.
<x _{k | k} > = <x _{k | k-1} > + K _k (z _k −H _k <x _{k | k−1} >) (3)
Here, K _k is the Kalman gain.

The Kalman filter estimates the system state quantity so as to minimize the error covariance (posterior error covariance) between the true system state x _k and the posterior state estimate <x _{k | k} > according to the following equation (4). It can be said that this is a technique for performing.
P _{k | k} = E [(x _k − <x _{k | k} >) (x _k − <x _{k | k} >) ^T ] (4)

Actually, the Kalman filter performs the following prediction and update calculations at each measurement time point, that is, whenever new measurement data input in time series is obtained, thereby obtaining a smoothing process result.
[Prediction calculation]
Based on the system estimation state at the previous measurement time from a certain measurement time point, the system state estimation value <x _{k | k} > and the system state error P _{k | k−1} are used as the system estimation state at the current measurement time point. presume.
<x _{k | k-1} > = F _k <x _{k-1 | k-1} > (5)
P _{k | k−1} = F _k P _{k−1 | k−1} F _k ^T + Q _k (6)
Here, Q _k is a process noise covariance matrix according to the following equation.
Q _k = [σ _a ² (Δt) ² ]

[Update calculation]
Using the observed value at the time of the current measurement, the value is updated by correcting the estimated value of the system state obtained as described above. That is, the optimal Kalman gain K _k is _{K k = (P k | k} -1 H k T) / (H k P k | k-1 H k T + R k) ... (7)
It is. Here, R _k is an observation noise covariance matrix according to the following equation.
R _k = [σ _z ² ]
The estimated value of the system state updated by applying the Kalman filter having the Kalman gain is expressed by the following equation (8).
<x _{k | k} > = <x _{k | k−1} > + K _k (z _k −H _k <x _{k | k−1} >) (8)
Further, the error of the updated system state is expressed by equation (9).
P _{k | k} = (I−K _k H _k ) P _{k | k−1} (9)

  By repeating the prediction calculation and update calculation for each measurement time point, the observed value is smoothed, and the noise superimposed on the observed value is reduced.

In order to apply the smoothing algorithm using the Kalman filter as described above to the smoothing of the load cell measurement signal (test force), the system state x _k having the test force x as an element as the state equation of the Kalman filter A kinematic equation that gives a velocity x ′ between one measurement time point k−1 and the next measurement time point k was set. That is, the equation of state is the same as equation (10) above (1).
x _k = F _k x _k-1 + w _k (10)
Here, the system state x _k is [x], x is the test force (N), the time transition transformation F _k of the system is [I], the process noise w _k is x′Δt, and x ′ is an approximation thereof. The time derivative (N / sec) of the test force whose value is N (0, σ _a ² ), Δt is the time interval (sec) at each measurement time point.

On the other hand, the observation equation is the same equation (11) as the above equation (2).
z _k = H _k x _k + v _k (11)
Here, z _k is the observed test force (N), the transformation H _k from the state space to the observation space is [I], and the observation noise v _k is approximately N (0, σ _z ² ).

That is, in order to perform the smoothing process in the Kalman filter, two parameters, _a parameter σ _a relating to process noise and a parameter σ _z relating to observation noise, are necessary. In order to smooth the load cell measurement signal in real time, the two parameters need to be set in advance (that is, before measurement) or during measurement execution. The observation noise is noise that is observed in a state where the measurement signal should be constant. On the other hand, the process noise is noise related to time transition, and thus becomes prominent when the measurement signal fluctuates. Therefore, here, the parameter σ _a relating to the process noise and the parameter σ _z relating to the observation noise are set as follows.

FIG. 2 is a conceptual diagram of parameter setting in the Kalman filter.
As shown in FIG. 2, the parameter σ _a relating to process noise is calculated based on a plurality of original test force signals (not corrected signals) obtained within a certain past period from the current measurement time point k. To do. Specifically, linear approximation using the last 100 measured data values is performed, an average signal slope (N / sec) at the time of the measurement is obtained, and this is set as _a parameter σ _a relating to process noise. did. Each time new measurement data is obtained during measurement execution, the latest 100 measurement data are replaced one by one, so that the parameter σ _a relating to process noise is updated each time.

On the other hand, the parameter σ _z related to the observation noise may be a standard deviation obtained from measurement data during a period in which the test force is substantially constant as shown in FIG. Since there is no great difference even if the results are used, it is determined in advance as part of the calibration of the device. The initial value <x _{0 | 0} > of the posterior state estimated value is [0], and the initial value P _{0 | 0} of the system state error is [1].

  FIG. 3 is a flowchart of prior processing in the noise reduction apparatus of the present embodiment using the algorithm as described above, and FIG. 4 is a flowchart of noise reduction processing executed in real time during measurement execution. The specific operation of the noise reduction apparatus of this embodiment will be described according to these flowcharts.

[Pre-setting parameters for observation noise]
Prior to the test on the test piece to be tested with the tensile tester, as a preparatory test, the test force is measured for a predetermined time without mounting the test piece on the tester, that is, in an unloaded state (step S1) . At this time, measurement data obtained by converting the measurement signal output from the load cell 2 by the analog-to-digital converter 3 is input to the observation noise parameter calculation unit 13 and temporarily held. At this time, since there is no load fluctuation, the measurement signal should ideally be constant, but the measurement signal fluctuates slightly due to the influence of various noises. The observation noise parameter calculation unit 13 calculates a standard deviation as a fluctuation amount of a predetermined number of measurement data (step S2), and stores this in the observation noise parameter storage unit 14 (step S3).

[Smoothing during measurement]
When the test piece is mounted on the tester and the user instructs the start of the test using a button or the like (not shown), the Kalman filter 11 and the process noise parameter calculation unit 12 respectively initialize the process noise parameter value and the initial value of the test force. (Step S11).

  Next, in the Kalman filter 11, the prediction calculation unit 111 estimates the system state and the test force at the current measurement time from the system estimated state at the previous measurement time, using the parameter value related to the process noise (step S12). However, in the estimation performed immediately after startup, initial values are used for the parameter values related to process noise, the system estimation state at the previous measurement time, the test force, and the like.

Next, the correction calculation unit 112 and the process noise parameter calculation unit 12 of the Kalman filter 11 read the measurement data (test force) output from the analog-digital converter 3 at the current measurement time (step S13). The process noise parameter calculation unit 12 obtains an average signal at the current measurement time from a plurality of (in this case, 100) measurement data obtained in the past fixed time from the current measurement time and accumulated in the buffer, as described above. The inclination is obtained and transferred to the Kalman filter 11 as _a parameter value σ _a relating to process noise (step S14). As is clear from the above description, measurement data obtained by actual measurement is not stored in the buffer until a certain time has elapsed from the start of the test. Therefore, the parameter value σ _a relating to process noise is output assuming that there is no signal inclination. Also good.

Next, in the Kalman filter 11, the correction calculation unit 112 sets the measurement data (test force signal) obtained at the time of the current measurement and the parameter σ _z related to the observation noise to which the value read from the observation noise parameter storage unit 14 is fixed. The system state and test force at the current measurement time estimated in step S12 are corrected (step S15). And the correction | amendment calculating part 112 outputs the test force corrected in step S15 as a Kalman filter application result (step S16).

When the corrected test force signal for one piece of measurement data is output, it is determined whether or not the test is finished (step S17), and if the test is not finished yet, the process returns to step S12. By repeating the processing of steps S12 to S16 for each measurement time point, that is, each time new measurement data is input, data indicating the test force signal obtained by smoothing the original test force signal is sequentially output. For example, as shown in FIG. 2, the parameter σ _a relating to the process noise is substantially constant during a period in which the test force signal is substantially constant except for noise, but when the test force signal is changed, the process noise is correspondingly changed. The parameter σ _a changes (becomes larger). That is, the parameter σ _a related to the process noise changes so as to follow the change in the original test force signal. Thereby, even when the original test force signal changes abruptly, the noise can be sufficiently reduced while suppressing the time delay of the output signal.

[Measurement results and evaluation]
An actual measurement result obtained by applying the load cell measurement signal noise reduction device of the present embodiment to an actual load cell measurement signal will be described.
FIG. 5 is a diagram illustrating an example of an original signal (test force) and an output after smoothing processing when a tensile test (test speed: 1 mm / min) is performed using metal as a test piece. FIG. 6 is a diagram illustrating an example of an original signal (test force) and an output after smoothing processing when a tensile test (test speed: 100 mm / min) is performed using rubber as a test piece. FIG. 7 is a diagram illustrating an example of an original signal (test force) and an output after smoothing processing when a tensile test (test speed: 500 mm / min) is performed using rubber as a test piece. From these results, it can be qualitatively understood that the noise superimposed on the measurement signal is sufficiently reduced, and that the output signal sufficiently follows the change of the measurement signal.

  FIG. 8 is a diagram showing a comparison of signal time delay between the above-described noise reduction method according to the present embodiment and the conventional noise reduction method based on moving average, in which (a) shows a test piece made of rubber, and (b) shows a test. It is an actual measurement result when a piece is a metal. FIG. 8 also shows a time delay range of 50 [msec] from the original signal (average value). In the conventional method, the time delay of the output signal exceeds 50 [msec] when the signal changes, whereas in the method according to this embodiment, the time delay of the output signal is sufficiently reduced to 50 [msec]. I understand that. Thereby, it can be said that the noise reduction method according to the present embodiment exhibits sufficiently high followability as compared with the conventional method.

  Further, when the standard deviation of the original test force signal and the standard deviation of the signal after smoothing are calculated and the ratio is obtained as an index indicating the noise reduction effect, the noise superimposed on the original signal is 1/25. It was confirmed that it was reduced to about 1/35. This is almost the same result as the conventional method, and the noise reduction effect and the high followability of the output signal according to the present invention can be sufficiently confirmed quantitatively.

  In the noise reduction device according to the above embodiment, the parameter related to the observation noise is obtained in advance and given to the Kalman filter as a fixed value. However, similarly to the parameter related to the process noise, the observation is performed based on the measurement signal obtained during measurement You may make it obtain | require the parameter regarding noise.

  In the noise reduction device according to the above embodiment, the Kalman filter estimates the two types of system states, the prior state estimation and the posterior state estimation value. The system state at the time of the previous measurement is estimated, and for this estimation, the amount of change in the slope of the signal indicating the degree of change in the measurement signal, that is, the differential value of the slope is used as a parameter for the second process noise. It is good also as a structure using. In this configuration, as a matter of course, although the processing is complicated as compared with the configuration of the above-described embodiment, it is possible to further improve the followability of the output signal especially in the vicinity where the measurement signal changes in a curved shape.

  Further, the above-described embodiments are merely examples of the present invention, and it is obvious that changes, modifications, and additions are appropriately included in the scope of the claims of the present application within the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 ... Noise reduction apparatus 11 ... Kalman filter 111 ... Prediction calculating part 112 ... Correction calculating part
DESCRIPTION OF SYMBOLS 12 ... Process noise parameter calculation part 13 ... Observation noise parameter calculation part 14 ... Observation noise parameter memory | storage part 2 ... Load cell 3 ... Analog-digital converter (ADC)

Claims (7)

A noise reduction device for reducing noise superimposed on a discretized measurement signal input in chronological order,
a) calculating a parameter value related to observation noise described later based on a plurality of measurement signals input under conditions where the measurement signal does not change or a plurality of measurement signals obtained before the input measurement signal changes; One parameter calculation unit;
b) a second parameter calculation unit that calculates a parameter value related to process noise, which will be described later, in accordance with the degree of change in a predetermined number of measurement signals in the past each time a measurement signal is input in chronological order;
c) Each time a measurement signal is input in chronological order, a parameter value related to observation noise obtained by the first parameter calculation unit and a parameter value related to process noise obtained by the second parameter calculation unit for the measurement signal A Kalman filter that executes a smoothing process based on
A noise reduction device for time-series measurement signals, comprising: It is the noise reduction device of the time series measurement signal according to claim 1,
The second parameter calculation unit is a noise reduction device for time-series measurement signals, wherein an average slope of the measurement signal obtained by linear approximation with respect to a predetermined number of measurement signals in the past is used as a parameter relating to process noise. A noise reduction device for a time-series measurement signal according to claim 1 or 2,
The first parameter calculation unit calculates a standard deviation of a plurality of measurement signals input under a condition in which the measurement signal does not change or a plurality of measurement signals obtained before the input measurement signal changes. A noise reduction device for time-series measurement signals, wherein the obtained standard deviation is used as a parameter value relating to observation noise. A noise reduction device for a time-series measurement signal according to any one of claims 1 to 3,
The first parameter calculation unit calculates a fixed parameter value related to observation noise based on a plurality of measurement signals obtained at the time of device calibration performed prior to measurement of a target sample. Noise reduction device. The apparatus for reducing noise of a time series measurement signal according to claim 4, wherein the apparatus reduces noise of a measurement signal obtained by a load cell attached to a test machine that performs mechanical measurement on a sample,
The first parameter calculation unit calculates a fixed parameter value related to observation noise based on a plurality of measurement signals obtained during apparatus calibration performed without mounting a sample on the test machine. Series measurement signal noise reduction device. A noise reduction device for a time-series measurement signal according to any one of claims 1 to 3,
The first parameter calculation unit calculates a parameter value relating to observation noise based on a plurality of measurement signals obtained during a period before a measurement signal input in time series changes. Series measurement signal noise reduction device. A noise reduction device for a time-series measurement signal according to any one of claims 1 to 6,
The second parameter calculation unit uses the average slope of the measurement signal obtained by linear approximation with respect to the predetermined number of measurement signals in the past as a parameter related to the first process noise,
A third parameter calculating unit that calculates a parameter value related to the second process noise according to the average amount of change in inclination;
The Kalman filter has three types of system states based on a parameter value related to the observation noise, a parameter value related to the first process noise, and a parameter value related to the second process noise every time measurement signals are input in chronological order. A noise reduction apparatus for a time-series measurement signal, characterized by executing a smoothing process for estimating the time series measurement signal.

Images