JP2013083559A - Signal processor and signal processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reliably remove influences by noise elements and to perform accurate weight estimation in a technology for estimating article weight on the basis of probability density function of amplitude of a measurement signal.SOLUTION: The signal processor has: a plurality of N filters CHto CHfor extracting from a measurement signal, signal components Sto Sin a plurality of N partial bands different within a frequency range except that of DC and its vicinity among bands to be processed of a measurement signal s(t); a whole band data processing part 25 for calculating amplitude probability density function PDFof an output signal Sof the whole band to be processed during a prescribed period when article is loaded on a load sensor; a plurality of N partial band data processing parts 30(1) to 30(N) for calculating probability density functions PDFto PDFduring the prescribed period of the amplitude of the signal components Sto Sin the plurality of N partial bands respectively; and a weight estimation part 40 for estimating weight of the article on the basis of the probability density functions PDFto PDFobtained.

Description

  The present invention relates to a signal processing apparatus and a signal processing method for accurately measuring the weight of an article sequentially loaded on a load sensor with a simple configuration and technique.

  2. Description of the Related Art Conventionally, as a weighing device for measuring the weight of an article, a weighing conveyor type that performs weight detection while the article is being conveyed by a conveyor has been widely used. This type of weighing device carries an article from a preceding conveyor to a weighing conveyor supported by a weighing machine as a load sensor, and outputs a weighing signal output from the weighing instrument while the loaded article is transported on the weighing conveyor. Based on this, the weight of the article is detected, and the article whose weight has been detected is carried out to the rear conveyor.

  Since such weighing devices are mainly installed in food production lines, floor vibrations caused by other production facilities and low-frequency vibration components due to the rollers and belts of weighing conveyors are measured as noise components. Superimposes on the output signal of the instrument, leading to deterioration of weighing accuracy.

  For this reason, there is a demand for a dynamic weight measurement method that can efficiently remove noise components from the output signal of the weighing instrument, perform weight measurement of articles closer to the true value, and achieve high accuracy and high speed. From a software perspective, the weight of an article can be determined from the measurement data that is measured by the weighing device and disturbed by the natural vibrations of the sensor measurement system that constitutes the weighing device and the vibration noise of the weighing and conveying system in a floor vibration environment. It can be understood as a research subject on how to estimate efficiently and accurately.

  Here, as a related research in the dynamic weight measurement field, for example, a study that modeled a hakari system with a linear state space representation and resulted in a mass state estimation problem, and removed the noise signal contained in the output signal of the load cell There are studies that apply linear system theory and system identification methods to signals, and research reports that combine acceleration sensors and Kalman filters.

  In addition, there is a study in which a compensation cell is added separately from the main body cell and the relative compensation principle is applied to floor vibration removal.

  On the other hand, as an example of research using prior knowledge of vibration noise generation, the output signal of a measuring instrument such as a load cell is A / D converted for weighing measurement with a multiple scale, and the filter processing is performed by a low-pass filter (FIR type LPF) There is also a report that applied to the removal of vibration noise suppression.

  In this way, research has been conducted to remove the vibration noise component from the output signal of the measuring instrument and extract the true mass component with high speed and high accuracy, but the calculation method is mainly based on linear calculation. there were. As signal processing using this linear operation, a low-pass filter LPF that is widely used generally can remove low-frequency noise components by setting the cutoff frequency width of the LPF low.

  However, in signal processing using such linear operations, the analysis and evaluation procedures are uniquely determined, but the filter response time based on the processing delay time and time frequency uncertainty is limited in principle. Accompany. Specifically, when the cut-off frequency of the LPF is set low, there is a problem that it is necessary to lengthen the sensing period, and the response speed is slowed accordingly. Further, there is a problem in that the LPF itself generates artificial noise by limiting the cutoff frequency of the LPF, and the artificial noise is superimposed on the output signal from the measuring instrument, and the measurement result lacks reliability. In addition, a noise component having a frequency lower than the cutoff frequency of the LPF set to the limit of the weighing capacity cannot be removed.

  For this reason, there is a means for effectively utilizing prior knowledge of physical phenomena encountered in one way of thinking to remove the above-mentioned restrictions. For example, a Kalman filter or the like is considered as an estimation method using prior knowledge of signal generation.

  However, in order to ensure that the mathematical modeling of signal generation matches the actual situation, there are uncertain elements such as setting of initial conditions and occurrence of sudden events, which is extremely difficult.

  As an effective technique for solving this problem, the inventor of the present application has developed a probability density function representing the frequency of occurrence of amplitude values obtained by sampling a weighing signal at a predetermined period within a certain period while an article is loaded. A system for calculating a PDF and estimating an article weight based on the probability density function PDF has been proposed (Non-Patent Document 1).

Naosuke Nori: “Examination of dynamic weight measurement using amplitude probability distribution (APD) measurement technology”, 27th Sensing Forum document, 27 (pp.198-202) (2010)

  The weight measurement method using the probability density function of the amplitude of the weighing signal described above calculates the amplitude probability density by taking in a signal of a certain period of the weighing signal after a certain period of time has elapsed since the load sensor was loaded with an article, For example, the expected value of the period is calculated and the weight of the article is estimated based on the expected value. The complicated process of mathematical modeling for signal generation is not required and simple and practically accurate. There is an advantage that weight estimation can be performed.

  However, even in this method, further improvement is desired in terms of eliminating the influence of various noise signals.

  The present invention improves this point, and in the technology for estimating the weight of an article based on the probability density function of the amplitude of the weighing signal, the signal processing that can more reliably remove the influence of the noise component and enable more accurate weight estimation. An object is to provide an apparatus and a signal processing method.

In order to achieve the above object, a signal processing device according to claim 1 of the present invention comprises:
In a signal processing device that receives a measurement signal output from a load sensor in a state where an article is loaded and performs processing for detecting a weight value of the loaded article,
Among the processing target bands of the measurement signal including the direct current corresponding to the load of the article and the noise component superimposed on the direct current, the signal components in a plurality of (N) subbands that are different within the frequency range excluding the direct current component and the vicinity thereof ( _{S 1,} S 2, ..., a filter _(CH 1 _{~CH N)} of the plurality (N) to be extracted from the weighing signal _{S N),}
An all-band data processing unit (25) that calculates a probability density function (PDF ₀ ) of the amplitude of the output signal (S ₀ ) of the entire processing target band during a predetermined period when an article is loaded on the load sensor;
A plurality (N) of partial band data processing units respectively receiving the plurality (N) of partial band signal components and calculating probability density functions (PDF _{1 to} PDF _N ) of the amplitudes of the respective signal components in the predetermined period. (30 (1) -30 (N)),
A probability density function obtained by the full-band data processing unit, and a weight estimation unit (40) for estimating the weight of the article based on the probability density function obtained by the partial-band data processing unit. It is characterized by being.

A signal processing device according to claim 2 of the present invention is the signal processing device according to claim 1,
The weight estimation unit calculates an expected value or an intermediate value between a lower limit value and an upper limit value of the probability density function obtained by the full band data processing unit and the partial band data processing unit, and based on the calculated value, The weight is estimated.

Further, the signal processing method of the present invention includes:
In a signal processing method for receiving a measurement signal output from a load sensor in a state where an article is loaded and performing processing for detecting a weight value of the loaded article,
Among the processing target bands of the measurement signal including the direct current corresponding to the load of the article and the noise component superimposed on the direct current, the signal components in a plurality of (N) subbands that are different within the frequency range excluding the direct current component and the vicinity thereof ( Extracting S ₁ , S ₂ ,..., S _N ) from the weighing signal;
Probability density function (PDF ₀ ) of the amplitude of the output signal (S ₀ ) of the entire processing target band during a predetermined period when an article is loaded on the load sensor, and the amplitude of the signal component of each partial band during the predetermined period Calculating respective probability density functions (PDF _{1 to} PDF _N ) of
The article based on the probability density function (PDF ₀ ) obtained for the calculated output signal of the entire processing target band and the probability density function (PDF _{1 to} PDF _N ) obtained for the signal components of each partial band. Estimating the weight of the.

A signal processing method according to claim 4 of the present invention is the signal processing method according to claim 3,
Estimating the weight comprises:
An expected value or an intermediate value between a lower limit value and an upper limit value of each calculated probability density function is calculated, and the weight of the article is estimated based on the calculated value.

  As described above, the present invention regards the amplitude variation of the weight measurement process as a stochastic process, and the entire processing target band set for the weighing signal output during a predetermined period while the load sensor is loaded with the article. And a probability density function for a plurality of different subband signals within the frequency range excluding the direct current and its vicinity in the processing target band, and the probability density calculated for these different bands. The weight value of the article is estimated based on the function.

  For this reason, noise components such as damped vibration components and floor vibration components superimposed on the DC component corresponding to the article weight can be associated with the probability density function of signals extracted from a plurality of partial bands, and the entire processing target band By removing the estimated value obtained from the probability density function for each sub-band signal from the weight estimated value obtained from the probability density function, a more accurate weight value is obtained by removing the effects of various noise components. Can do.

A diagram for explaining a probability density function PDF and a cumulative probability distribution APD The figure which shows the process sequence of the multi-channel PDF measurement method of this invention. The figure which shows the structure of the signal processing apparatus of embodiment of this invention. Diagram showing the configuration of the weighing unit Measured waveform diagram of weighing signal Show simulation results for the entire band CH ₀ It shows a simulation result of the partial band CH ₁ It shows a simulation result of the partial band CH ₂

  The present invention uses the amplitude of the weighing signal output from the load sensor loaded with the article as a random variable, and obtains the probability density function PDF and the cumulative probability distribution APD to estimate the article weight. The probability density function PDF and the cumulative probability distribution APD will be briefly described.

  A probability density function (also referred to as probability density distribution) PDF takes the time signal as shown in FIG. 1 as an example, and the occurrence frequency of amplitude values extracted at an appropriate sampling time within the measurement time T can be obtained with the required measurement accuracy. It is obtained by counting for each quantized amplitude level.

If this PDF value is integrated (accumulated) with a random variable, it becomes an APD value. However, if the definition of the APD value “time probability that the envelope signal of the signal amplitude exceeds a certain threshold level” is adopted, the signal amplitude r (t) the signal amplitude of the discrete values x _i of the random variable, the measurement time T, the signal amplitude threshold R occupancy at time W _{i (x k)} cumulatively adds, APD value of x _i divided by T is following for It is calculated | required like Formula (1).

APD (x _i ) = ΣW _i (x _k ) / T (1)
However, symbol Σ represents the sum total from i = 1 to N (x _k ).

  Conversely, if this APD value is subjected to a difference calculation with respect to a random variable, the PDF value can be obtained.

  Based on the above preparation, the signal processing method of the present invention will be described. In the present invention, the entire band set as the processing target for the measurement signal including the direct current corresponding to the weight value of the article and the harmful noise component superimposed thereon is set as the processing target band, and the signal of the entire processing target band This method estimates the weight of an article based on the probability density function for, and the probability density function for signals extracted from different bands in the frequency range excluding the direct current and its vicinity in the processing target band. Will be referred to as a “multi-channel PFD measurement technique”.

FIG. 2 is a diagram showing a processing procedure of the multi-channel PDF measurement technique of the present invention.
The weighing signal output from the load sensor until the load is released (carrying out) after the article is loaded (carrying in) on the load sensor (the weighing conveyor weighing device) used in the above-described weighing conveyor system is shown in FIG. (A) becomes a waveform in which a noise signal is superimposed on a trapezoidal wave.

This weighing signal is divided every time Tp (seconds) from measurement reference timing (appropriate article delivery timing) t = 0, and each time signal is divided on the frequency axis as shown in FIG. Input to the arranged filter group (CH _{0 to} CH _N ), and observe the output time signal of each filter.

Here, the filter CH ₀ is an LPF type filter that allows the entire processing target band to pass, and the other filters CH _{1 to} CH _N are BPF type, and within the frequency range excluding the direct current and the vicinity thereof in the processing target band. Different N subband signals are set to selectively pass therethrough. The filter CH ₀ may limit the band of the measurement signal before the other filter, and may also serve as a band limiting filter used before the A / D conversion process.

Then, as shown in FIG. 2C, the PDF value of the time signal output from each filter (CH _{0 to} CH _N ) is obtained, and the APD value, the expected value E [x], etc. are obtained from the PDF value. calculate.

The expected value calculated for the output of the filter CH ₁ (full-band signal) is not only the component value corresponding to the weight value (DC value) of the article to be acquired, but also the frequency of transient vibration, floor vibration, etc. Although the component value by the influence of a noise component is contained, these noise component values appear in the expected value with respect to the outputs of the other N filters CH _{1 to} CH _N.

Therefore, as shown in (d) of FIG. 2, each expected value E ₁ [x obtained for the outputs of the filters CH _{1 to} CH _N from the expected value E ₀ [x] in the entire processing band for the output of the filter CH _0. ] To E _N [x] are removed, and the weight of the article is estimated from the value after the removal, thereby accurately estimating the weight under the influence of various noise components. Note that it is not necessary to employ all of the expected values E ₁ [x] to E _N [x] obtained for the outputs of the filters CH _{1 to} CH _N , and they are compared with the threshold values set according to the measurement accuracy. It is also possible to ignore expected values below the threshold.

  As will be described later, not only the purpose of detecting the weight value of an article, but also the probability density function PDF, cumulative probability density APD, and expected value E [x] for all bands and partial bands for each type of article to be measured. , Etc., and the data such as the probability density function PDF, cumulative probability density APD, expected value E [x] newly obtained for the article to be measured, and the article were obtained by past measurement. By comparing the data such as probability density function PDF, cumulative probability density APD, expected value E [x], etc., the operation of the measuring unit (eg, motor rotation unevenness), the operating environment change (increase / decrease in floor vibration), etc. It is also possible to use for the purpose of state management to prevent a measurement result from containing a large error by being influenced by them.

  In the above method, each expected value calculated varies depending on which timing signal of the weighing signal is captured and at which length, but the sample product is repeatedly weighed while changing the acquisition timing to obtain the required accuracy. It is only necessary to start the actual weighing operation after obtaining a time period during which the weight can be estimated by.

  FIG. 3 shows the overall configuration of the signal processing apparatus 20 using the above principle. The signal processing device 20 uses the measurement signal from the measurement unit 10 shown in FIG. 4 as a processing target.

  The weighing unit 10 has a structure in which the article W is carried from the front conveyor 13 to the weighing conveyor 12 supported by the load sensor 11 and the article W conveyed by the weighing conveyor 12 is carried out to the rear conveyor 14. The carry-in timing of the article W to the weighing conveyor 12 is detected by the carry-in sensor 15. Further, here, for the sake of simplicity, the weighing signal s (t) output from the weighing conveyor 12 is treated as a value obtained by subtracting the conveyor weight. The article weight may be obtained by subtracting with.

  As shown in FIG. 3, the weighing signal s (t) from the weighing unit 10 is input to the A / D conversion unit 21, converted into a digital data string, and input to the filter block 22.

The filter block 22 extracts an LPF type filter CH ₀ that allows the entire processing target band to pass through, and a BPF that extracts signal components corresponding to a plurality of N different partial bands within the frequency range excluding the direct current and the vicinity thereof in the entire band. is constituted by the type of filter _CH 1 to CH _N, full-band signal _{S 0} output from the filter CH ₀ is input to the full-band data processing unit 25.

The all-band data processing unit 25 receives the signal from the carry-in sensor 15 of the weighing unit 10 and then receives the signal for the all-band signal S ₀ output from the filter CH ₀ during a predetermined period Tp after the preset time Ta has elapsed. A probability density function PDF or the like is calculated.

More specifically, the full bandwidth signal S ₀ is input to the normalization processing unit 26, when such negative data occurs performs a shift process to positive data as a whole, the normalized value Logarithmic conversion is performed by the logarithmic conversion unit 27, and the logarithmic value is quantized in units of accuracy to be measured. The logarithmic conversion unit 27 is for smooth processing of a large number of data, and this configuration can be omitted.

The logarithmic value is input to the PDF calculation unit 28. The PDF calculation unit 28 obtains each appearance frequency (histogram) of the input logarithmic value and divides this by the total number of data input during a certain time Tp, thereby calculating a normalized probability density function PDF ₀ .

Further, the APD calculation unit 29 cumulatively adds the calculated probability density function PDF ₀ to obtain a cumulative probability distribution APD ₀ . Note that the probability density function PDF and the cumulative probability distribution APD may be continuously obtained, and one or a plurality of locations as the measurement period and one or a plurality of times as the number of times can be arbitrarily selected. The measurement period is preferably one period (or more) of the lowest frequency component that is expected to affect the measurement among noise components.

Meanwhile, sub-band signals _S 1 to S _N to be outputted from the filter _CH 1 to CH _N are input to each partial band data processing section 30 (1) ~30 (N) .

These N partial band data processing units 30 (1) to 30 (N) have the same configuration as the full band data processing unit 25, and are output from the respective filters CH _{1 to} CH _N during the predetermined period Tp. Probability density functions PDF _{1 to} PDF _{N and the} like for the band signals S _{1 to} S _N are obtained.

That is, each of the partial band signals S _{1 to} S _N is input to the normalization processing unit 31, and when negative data or the like is generated, a shift process or the like is performed to make the data positive as a whole. Each logarithmic conversion unit 32 performs logarithmic conversion, and the logarithmic value is quantized in units of accuracy to be measured.

Then, the logarithmic values are respectively input to the PDF calculation unit 33, the appearance frequencies (histograms) of the input logarithmic values are obtained, and normalized by dividing this by the total number of data input at a certain time Tp. calculating a probability density functions _PDF 1 _{~PDF N} respectively, which was cumulatively added by APD calculation unit 34 obtains respectively the cumulative probability distribution _APD 1 _{~APD N.}

Each information obtained in this way is input to the weight estimation unit 40. The weight estimation unit 40 includes the probability density function PDF ₀ obtained by the entire band data processing unit 25 and the probability density functions PDF _{1 to} PDF _N obtained by the partial band data processing units 30 (1) to 30 (N). Based on the above, the weight of the article W is estimated.

Specifically, as described above, the expected values E ₀ [x] and E ₁ [x] to E _N [x] obtained from each probability density function are used, for example, corresponding to the weight of the article by the following calculation. The expected value E _W [x] is obtained. Here, the expected value is a value obtained by dividing the total of the product of the value and the frequency by the total number of appearances for all random variables (in this case, amplitude values) appearing at a certain time Tp. This corresponds to a simple average of signal amplitudes output at a certain time Tp.

E _W [x] = E ₀ [x] −ΣE _i [x]
The symbol Σ indicates the sum of i = 1 to N.

The weight value of the article W can be obtained by converting the expected value E _W [x] by weight.

  On the other hand, the data management unit 50 stores the probability density function PDF, cumulative probability density APD, expected value E [x], etc. for all bands and partial bands when measured in the past, together with the item type information, etc. Next, data such as the probability density function PDF, cumulative probability density APD, expected value E [x], etc. newly obtained for the article to be measured, probability density function PDF, cumulative probability obtained in the past measurement for the article Data such as density APD and expected value E [x] are compared.

  Here, for example, when there is an operation modulation of the measuring unit 10 (for example, uneven rotation of the motor) or a change in the operating environment (increase / decrease in floor vibration), the expected value of the probability density function PDF for all bands and partial bands, Since a change appears in the pattern of the intermediate value and the cumulative probability density APD, the weight estimation unit 40 is instructed to permit or reject the weight estimation process according to the degree of the change, and the state change of the weighing unit 10 is caused. It is possible to prevent a large measurement error from being affected.

  Hereinafter, the result of simulating the operation of the signal processing device 20 having the above-described configuration using a three-channel filter (N = 2) will be described.

  The larger the number of samplings necessary for calculating the APD probability distribution, the better for grasping the phenomenon. However, the calculation time also increases accordingly, and here, about 10 times the maximum frequency of the sensor signal band. Data sampled at the time is used.

Further, as simulation conditions, the filter CH ₀ for the entire band is a 1 kHz signal band rectangular filter, the filter CH ₁ for the partial band is a BPF having a cutoff frequency of 20 Hz (low band) and 400 Hz (high band), and a filter CH _2. Is a BPF with a linear phase characteristic of 400 Hz to 600 Hz. As the measurement signal, the measurement data of the measurement signal shown in FIG. 5 (the true weight value is estimated to be 90.3409 dB) is used.

  The measurement target time Tp is 50 milliseconds required to include one period of the natural vibration frequency 30 Hz included in the measurement signal, and a period of 700 to 750 milliseconds is measured from the load start timing of the article.

The measurement results of PDF / APD for the outputs of the filters CH ₀ , CH ₁ , and CH ₂ are shown in FIGS. In each figure, the horizontal axis indicates the logarithmic amplitude level value (dB), and the vertical axis indicates the probability. The PDF value is displayed 10 times larger. The outputs of the filters CH ₁ and CH ₂ are alternating current components such as vibration noise, but are added and displayed with a constant α (65535/2 = 90.3088 dB display) for easy comparison.

From the results of the filters CH ₀ and CH ₁ in FIGS. 6 and 7, the frequency behavior of the vibration component within the measurement period can be observed microscopically / macroscopically. That is, in this example, the PDF characteristic of the entire band in FIG. 6 is similar (has high correlation) to the PDF characteristic of CH ₁ in FIG. , CH ₁ components are dominant, and it can be clearly and easily understood that the vibration noise components of CH ₂ are so small that they cannot be observed.

Component of CH ₀ in FIG. 6, because it contains components of vibration components _CH 1, CH _2, as the weight estimate, when assuming linearity, as described above, the expected value of the CH _{0 E} 0 [ It can be seen that a highly accurate estimated value can be obtained by removing the expected values E ₁ [x] and E ₂ [x] of CH ₁ and CH ₂ from x].

The calculation results are shown below. However, the constant α is added to the calculation results of CH ₁ and CH ₂ .
Expected value E ₀ [x] of CH ₀ = 90.3064 dB
Expected value E ₁ [x] + α = 90.2475 dB of CH ₁
Expected value of CH ₂ E ₂ [x] + α = 90.3103 dB
Article weight estimated value E _W [x] = 90.3659 dB

From this result, the estimated product weight E _W [x] (= 90.3659 dB) obtained by removing the expected values of the plurality of partial bands is the expected value E ₀ [x] ( It was confirmed that it was closer to the true value R (= 90.3409 dB) than = 90.3064 dB).

In the above embodiment, the expected value is obtained from the probability density function PDF obtained for the entire band and the partial band, and weight estimation is performed, but the distribution of the probability density function PDF is around the true weight amplitude value. When the generated uniform noise distribution is considered to have symmetry in the distribution shape, the lower limit value min (x) of the random variable in which the probability density function PDF exists is used instead of each expected value. Intermediate value of the upper limit value max (x),
C = {max (x) −min (x)} / 2
The weight may be estimated based on this.

  As described above, the signal processing device 20 of the present invention includes the probability density function PDF for the signal of the entire processing target band set for the metric signal, and the frequency range within the processing target band excluding the direct current and the vicinity thereof. Since the probability density function PDF is obtained for signals extracted from a plurality of different sub-bands, and the weight of the article is estimated based on the probability density function PDF, it is not necessary to prepare detailed modeling, and the effects of various noise components Can be measured with high accuracy.

  Here, the processing for the weighing signal from the weighing conveyor type weighing unit has been described, but the present invention can be similarly applied to any weighing device that repeatedly loads and releases an article with respect to a load sensor. . For example, a plurality of weighing hoppers supported by a weighing instrument are provided, articles are put into these weighing hoppers, the weights thereof are obtained, and combinations in which the combined weight falls within the target range are determined from the combinations of articles for which the weights are obtained. The present invention can also be applied to a combination weighing device that selects and discharges them collectively.

  DESCRIPTION OF SYMBOLS 10 ... Weighing part, 11 ... Measuring device (load sensor), 12 ... Weighing conveyor, 15 ... Carry-in sensor, 20 ... Signal processing device, 21 ... A / D conversion part, 22 ... Filter block, 25... All band data processing unit, 30 (1) to 30 (N)... Partial band data processing unit, 26, 31... Normalization processing unit, 27, 32. PDF calculation unit, 29, 34... APD calculation unit, 40... Weight estimation unit, 50.

Claims (4)

In a signal processing device that receives a measurement signal output from a load sensor in a state where an article is loaded and performs processing for detecting a weight value of the loaded article,
Among the processing target bands of the measurement signal including the direct current corresponding to the load of the article and the noise component superimposed on the direct current, the signal components in a plurality of (N) subbands that are different within the frequency range excluding the direct current component and the vicinity thereof ( _{S 1,} S 2, ..., a filter _(CH 1 _{~CH N)} of the plurality (N) to be extracted from the weighing signal _{S N),}
An all-band data processing unit (25) that calculates a probability density function (PDF ₀ ) of the amplitude of the output signal (S ₀ ) of the entire processing target band during a predetermined period when an article is loaded on the load sensor;
A plurality (N) of partial band data processing units respectively receiving the plurality (N) of partial band signal components and calculating probability density functions (PDF _{1 to} PDF _N ) of the amplitudes of the respective signal components in the predetermined period. (30 (1) -30 (N)),
A probability density function obtained by the full-band data processing unit, and a weight estimation unit (40) for estimating the weight of the article based on the probability density function obtained by the partial-band data processing unit. A signal processing device.   The weight estimation unit calculates an expected value or an intermediate value between a lower limit value and an upper limit value of the probability density function obtained by the full band data processing unit and the partial band data processing unit, and based on the calculated value, The signal processing apparatus according to claim 1, wherein the weight is estimated. In a signal processing method for receiving a measurement signal output from a load sensor in a state where an article is loaded and performing processing for detecting a weight value of the loaded article,
Among the processing target bands of the measurement signal including the direct current corresponding to the load of the article and the noise component superimposed on the direct current, the signal components in a plurality of (N) subbands that are different within the frequency range excluding the direct current component and the vicinity thereof ( Extracting S ₁ , S ₂ ,..., S _N ) from the weighing signal;
Probability density function (PDF ₀ ) of the amplitude of the output signal (S ₀ ) of the entire processing target band during a predetermined period when an article is loaded on the load sensor, and the amplitude of the signal component of each partial band during the predetermined period Calculating respective probability density functions (PDF _{1 to} PDF _N ) of
The article based on the probability density function (PDF ₀ ) obtained for the calculated output signal of the entire processing target band and the probability density function (PDF _{1 to} PDF _N ) obtained for the signal components of each partial band. Estimating the weight of the signal processing method. Estimating the weight comprises:
4. The signal processing method according to claim 3, wherein the expected value of each calculated probability density function or an intermediate value between a lower limit value and an upper limit value is calculated, and the weight of the article is estimated based on the calculated value. .

Images