WO2004072858A1

WO2004072858A1 - Condition identifying metod and condition identifying system

Info

Publication number: WO2004072858A1
Application number: PCT/JP2004/000344
Authority: WO
Inventors: Ho Jinyama
Original assignee: Ho Jinyama
Priority date: 2003-01-17
Filing date: 2004-01-16
Publication date: 2004-08-26
Also published as: CN1757022A; CN100375076C; JPWO2004072858A1

Abstract

A simple-to-use, high-in-diagnosis-accuracy, high-in-processing-speed, portable diagnosis system used for on-side condition diagnosing needs to be provided. A condition identifying method and a condition identifying system which execute the learning step of diagnosis functions by means of a computer having a high calculating power and a large memory capacity, transfer to a pocket-size portable diagnosis device elements necessary for diagnosis functions constituted by learning at the computer, and quickly perform a condition diagnosis using the portable diagnosis device. In addition, since feature waveform data obtained by removing noises from waveform data measured for diagnosing is converted into parameter waveform data in order to efficiently process information for condition diagnosis to allow parameter waveform data to play a role of compressing a measured signal on a time scale, a quick condition diagnosis is possible using parameter waveform data.

Description

Description State identification method and state identification system

-. Technical field

The present invention relates to a state identification method and a state identification system for performing half-state I] definition signal identification of an object in the fields of equipment diagnosis, infrastructure facility diagnosis, medical diagnosis, voice recognition, and pattern recognition. . 2. Background technology

Conventional state diagnosis is performed as follows.

(1) Perform state identification processing on a computer (for example, reference (1)). The “computer” described in this specification is an information processing device that is practically difficult to realize with a pocket size.

(2) All state identification processing is performed by a pocket-sized portable identification device (for example, reference (2)).

(3) State discrimination is performed by using the time-series signal measured for state discrimination as it is for fast Fourier transform processing 線 envelope processing and feature parameters ゃ information theory (for example, reference (3)) . 3. Problems to be solved by the invention

However, the conventional methods have the following problems.

In the first method, complicated signal processing, learning processing, and identification processing can be easily performed, so that an intelligent Eich state identification system can be easily constructed. However, it should be realized as a portable (pocket size) identification apparatus. Can be difficult.

In the second method, the pocket-sized portable identification device not only takes a very long time due to complicated signal processing and learning processing due to limitations in calculation capacity and memory capacity, but also requires high accuracy. In many cases, it is not possible to deal with the state identification problem.

Method (3) is a method generally used in the field of state identification. When measuring force waveform data, the sampling frequency is high, and if the measurement time is long, an enormous amount of waveform data must be processed. Therefore, processing by a pocket-sized portable identification device is difficult. 4. Means to solve the problem

In order to solve the problems described above, in the present invention, the process of learning and constructing a state identification function that requires a long processing time with a complicated processing algorithm has a high calculation capability and a large memory capacity. It can be performed on a large computer, and the elements required for the state identification function constructed by learning on the computer can be transferred to the portable identification device, and the state can be quickly identified using the portable identification device. In addition, it is possible to transfer the waveform data obtained by the portable identification device and the identification result to a computer, and perform more advanced processing such as cause analysis and state trend management by the computer.

Furthermore, in order to efficiently perform information processing for state identification, characteristic waveform data obtained by removing noise from waveform data measured for state identification is converted into parameter waveform data, and parameter data is obtained. Since the data waveform data also serves to compress the characteristic waveform data on the time scale, the state can be quickly identified using the parameter waveform data. 5. Effects of the Invention

In the present invention, in order to efficiently identify the state of an object in the field, a learning process in which the processing algorithm is complicated and requires a long processing time is performed by a computer having a high computational capacity and a large memory capacity. The necessary elements for the state identification function constructed by the learning in the above are transferred to the portable identification device, and the signal measurement and the state identification can be performed quickly using the portable identification device. In addition, it is possible to transfer the waveform data at the time of state identification and the identification result obtained by the portable identification device to a computer, and to perform more advanced processing such as cause analysis and state tendency management by the computer.

Further, in order to efficiently perform information processing for state identification, the characteristic waveform data acquired for state identification is converted into parameter waveform data, and the role of the parameter waveform data is to compress the characteristic waveform data on a time scale. Therefore, the state can be quickly identified using the parameter waveform data. 6. Best mode for carrying out the invention

FIG. 1 shows a processing flow of the present invention. FIG. 2 is a configuration diagram of hardware for realizing the state identification system shown in FIG. FIG. 3 is an example of a circuit diagram of the portable identification device. In Figure 3, 1 is a sensor, 2 is an amplifier, 3 is a filter, 4 is a processing unit, 5 is a result display, 6 is data RAM, 7 is an AD converter, 8 is a DC port, and 9 is SCI Reference numeral 10 denotes one chip, 11 denotes a flash ROM, and 12 denotes an external computer.

The flow of signal measurement and state identification processing shown in FIG. 1 will be described below.

In the field of equipment diagnosis and medical diagnosis, “state identification” is also referred to as “state diagnosis”, but in this specification, it will be referred to as “state identification”.

1 · Construction of a state identification function by learning on a computer

1-1. Setting the state to be identified

If waveform data that reflects the characteristics of various states to be identified can be measured, the state identification function is constructed using these waveform data. In the case of equipment diagnosis and medical diagnosis, the standard condition is generally normal, but other conditions are abnormal.

1-2 Signal measurement

Waveform data is measured and acquired according to the characteristic frequency band of each state to be identified. For example, in the case of equipment diagnosis, when various abnormal conditions occur, as shown in Fig. 4, signals indicating the characteristics of the abnormalities appear in low, medium, and high frequency regions, respectively. In order to reduce the effects of noise, it is necessary to measure the waveform data separately for these low, medium, and high frequency regions depending on the abnormal state to be identified at the time of signal measurement.

1-3 Extraction of waveform data by noise removal

For early detection of abnormal conditions, it is necessary to further remove noise from the measured waveform data and extract characteristic waveform data. Many methods for removing noise have been reported so far (for example, references (4) and (5)).

1-4 Calculation of parameter waveform If the sampling frequency is high and the measurement time is long when measuring and obtaining waveform data, the amount of waveform data becomes enormous, it takes time for the state identification processing, and the efficiency of the state identification deteriorates. Therefore, after extracting the extracted characteristic waveform data into parameter waveform data, the state is identified using the parameter waveform data. The feature parameters used for the conversion include dimensional feature parameters and non-dimensional feature parameters (for example, reference (6)).

For example, Fig. 5 (a) shows raw waveform data measured in the outer ring damage state of a bearing, Fig. 5 (b) shows characteristic waveform data with noise removed, and Fig. 5 (c) shows dimensional special parameters. This is the parameter waveform data calculated by (effective value). Since the number of raw waveform data is 8192, but the number of RMS parameter waveform data is only 128, the effect of data compression can be confirmed. The number of feature waveform data used when calculating parameter waveform data is calculated by the following formula.

M = 5 to (1) 2f _x

Here, f _x is the characteristic frequency to be analyzed, f _m is the sampling frequency of the time-series waveform data.

For example, in the case of bearing state identification, the number of characteristic waveform data used when calculating parameter waveform data is calculated by the following formula.

M = 5 ~~ (2) 2f ₀

Where f. It is characterized (path) frequency during the outer ring wound state, f _m is the sampling frequency of the time-series waveform data.

The spectrum of the parameter waveform data of the effective value obtained by FFT to identify the bearing state is shown in Fig. 5 (d). The frequency of the first peak in the spectrum in Fig. 5 (d) is 110Hz, which matches the characteristic (path) frequency of the bearing when the outer ring is in a damaged state, so it can be determined that the bearing is in the "outer ring damaged state". .

Similarly, Fig. 5 (e) shows the parameter waveform data calculated using the dimensionless feature parameters (effective value ratio, that is, the ratio of the section effective value of the waveform data to the total effective value). It is also possible to judge that the outer ring is in the injured state from the star in Fig. 5 (e) (Fig. 5 (f)).

1-5 Establishing knowledge to identify various states

(1) In the case of a neural network or a multi-valued neural network In order to identify a state using a neural network or a multi-valued neural network, a finite number of indices indicating the characteristics of the characteristic waveform data (or parameter waveform data) are used. Need to calculate. Such an index is called a “feature parameter.” Many feature parameters have been defined in the past (for example, reference (7)).

To train a neural network or a multi-valued neural network, the following input data and teacher data are required. Input data (3)

Teacher data (4)

Here, _Pij is the ith feature parameter, which is a value obtained from the feature waveform data (or レ is parameter waveform data) extracted at the jth time. n is the number of types of feature parameters, and m is the number of measurements of waveform data. _du is the occurrence rate of the i-th state corresponding to the j-th row of the input data. An example of how to obtain input data and teacher data is shown in (for example, Reference (8)).

(2) G A special case

The characteristic parameters obtained from the characteristic waveform data (or parameter waveform data) in state a and state b are respectively

Then, the input data is obtained as follows.

Input data at state a:

p ^ib) U p)), η Input data in state b Ρ y (6)

Ρ

Here, η is the number of types of feature parameters, and m is the number of times of measurement of waveform data. Good feature parameters that can distinguish between state a and state b are determined by genetic algorithms. The good feature parameters obtained by the genetic algorithm are called GA feature parameters. An example of a specific method is shown in (for example, Reference (9)).

(3) In the case of fuzzy identification mechanism

In the case of the fuzzy identification mechanism, the antecedent (input) and consequent (conclusion) of fuzzy inference are characterized by It is determined using waveform data (or parameter waveform data). The specific way of finding is

(For example, reference (10)).

(4) Other methods

There are other methods besides the three types of state identification methods described above.However, in order to construct a state identification function even in the case of a deviation, each state is identified using characteristic waveform data or nomometer waveform data. Establish knowledge to identify.

1-6 Building a state identification function by learning

As described above, if knowledge for identifying each state is established, a state identification function can be constructed by learning for a portable identification device. Examples of constructing the state identification function include the case of a neural network (for example, reference (8)), the case of a GA feature parameter (for example, reference (9)), and the case of fuzzy identification (for example, reference (8)). 10))).

1- 7 Transfer necessary elements for status identification function to portable identification device

The elements required for the state identification function to be transferred from the computer to the portable identification device are weighting factors in the case of a neural network or a multi-valued dual network, and are calculated by a genetic algorithm in the case of GA feature parameters. This is a good GA feature parameter and state judgment criterion for state identification, and a membership function for identification inference in the case of a fuzzy identification mechanism.

2. Preparation and execution of state identification by band type identification device

2-1 Preparation for identification

After receiving the elements required for the state identification function transferred from the computer, the portable identification device constructs the state identification function so that it can identify the state independently. For example, in the case of a neural network, a trained neural network obtained by a computer is prepared so that it can be executed by a portable identification device, and measurement conditions of waveform data and determination criteria for state identification are set. Put. 2 -2 Perform status identification

After the portable identification device is equipped with the state identification function, the signal measurement, noise elimination and parameter waveform data calculation for the target object are actually performed by the above-mentioned computer during learning (1-1 to: 1-4). Content). The signal identification and the state identification are performed by executing the state identification function obtained by the computer learning on the portable identification device.

2 3 Display of status identification result and transfer of identification result to computer

The identification result obtained by the portable identification device is displayed on the display unit of the portable identification device, and the state identification result is shown. If necessary, the waveform data measured at the time of state identification and the state identification result are stored in a portable identification device and transferred to a computer, after which further cause analysis and state trend management are performed by the computer. 7. Examples of implementation

1. Multi-valued neural network example

According to the flow shown in Fig. 1, an example of construction of a state identification system using a multi-valued neural network is shown.

Figure 6 shows the target bearing and a microphone for signal measurement. There are four states to be distinguished: normal, rolling body wound, inner ring wound, and outer ring wound. The waveform data used when learning the state identification function is shown in the figure. 6 is waveform data of an acoustic signal measured at a distance of lm from the target bearing shown in FIG. After removing noise from the sound signal measured by a bandpass filter (5 kHz 40 kHz), the signal was normalized by the following equation.

_Xi = ^ (7)

¹ S

Here, X '; is discrete waveform data of measured signals, mu and S each _X'; the mean value and the standard deviation of.

In this example, since the sampling frequency is 40 kHz and the number of pieces of waveform data is 4096, parameter waveform data as shown in FIG. 5 is not obtained. When the number of pieces of waveform data is large, learning and identification may be performed by obtaining the parameter waveform data shown in FIG. 5 and then obtaining the following characteristic parameters.

The following 11 state identification feature parameters were calculated from the feature waveform data.

here,

N

I is the absolute mean, and Ν is the total number of data.

Is the standard deviation.

Ν

L- ³

i = l (11)

P,

² (^ 1) σ ³

Ν

L ι "= 1«-μ) ⁴

(-1) σ ⁴ (12)

Here, _ρ is the average value of the maximum value (peak value) of the waveform.

Where μ is the average of the 10 maximum values of the waveform c

-6- Here, σ _ρ is the standard deviation value of the maximum value. 7

Where and a _L are the mean and the standard deviation ^ j of the local minimum (valley value), respectively.

2ST i = X

(18)

Kb ²

Ριο— (19)

Figure Ί shows examples of characteristic parameter (Pi to p) values obtained from characteristic waveform data (30 each) in each state.

The feature parameter value is converted into an integer by the following equation.

P i

i max {p}-min { _P -w _{P i} ) -to.5] (2 l) where int [x] is a function that rounds down the decimal point of X and calculates an integer. N _pi indicates the number of divisions from max { _Pij } force to minfci. In this example, m = 120 and i = 1-11.

The relationship between the combination of characteristic parameter values and the occurrence rate (degree of possibility) of state k is calculated by the following equation.

Fiber ₍ ( _n 22) where the number of occurrences of ^k sets of parameter-like value states in tok 1 is 赚. For example, the combination of values of ~ Pi i is {2, 5, 12, 1, 12, 4, 9, 16, 17, 3, 5}, if state k occurs three times and states other than state k occur seven times, the probability of state k at that time (occurrence rate ) Is 0.3, and the probability (occurrence rate) of the state other than state k is 0.7.Figure 8 shows an example of the learning data for the obtained state identification function (multi-valued neural network). Redundant parts of the input data were removed using the rough set described in (8).

FIG. 9 shows an example of a multi-valued neural network for discriminating the state of a bearing. Using the learning data shown in Fig. 8, the computer learned the multi-valued neural network shown in Fig. 9. Then, the weighting factor of the learned multi-valued neural network is transferred to the portable identification device.

After receiving the weighting factors of the multi-valued neural network, the portable identification device is prepared to execute the learned multi-valued neural network shown in Fig. 9. At the time of state identification, if the multi-valued neural network shown in FIG. 9 is executed according to the signal measurement and state identification execution procedure shown in FIG. 1, the identification result shown in FIG. 10 is obtained. In FIG. 10, for example, a combination of feature parameter values {3, 2, 1, 16, 14, 17, 16, 3, 4} obtained from waveform data measured in a normal state is a learned multi-valued neural network. , The likelihood (occurrence rate) of each state output from the multi-valued neural network is normal: 0.79, rolling body wound: 0.34, inner ring wound: 0.46, outer ring wound: 0.34. Status ". Similarly, the identification results of other states are shown in FIG.

2. Examples of GA feature parameters (Reference (9))

According to the flow shown in Fig. 1, an example of construction of a state identification system using GA feature parameters will be shown.

Fig. 11 shows characteristic waveform data and parameter waveform data extracted from acceleration signals measured in the four states (normal state, outer ring injury, inner ring injury, and rolling body injury) to be identified in the rotating machine in Fig. 6. At a sampling frequency (f _m) is 25600Hz features waveform data path frequency of the outer ring wound because it is 54 Hz, the number of feature waveform data to be used when calculating the parameter waveform data according to the equation (2) is 241 did. Since the number of points in the characteristic waveform data is 32768 and the number of points in the parameter waveform data is only 136, the efficiency of the state identification process is improved.

To identify these conditions, calculates the PP _2, the value of _{_{_{P 3, P 4, P 5}}} . P 6 of the formula (8) to Formula (15) using the parameter waveform data.

To identify each state efficiently, the sequential state identification shown in FIG. 13 is performed. In this case, a special feature parameter for identifying each state is required. Therefore, using a genetic algorithm (GA) or genetic programming (GP), we search and find good GA special parameters to identify each state. For example, the GΑ feature parameters obtained to identify the four states shown in Fig. 13 are as follows.

GA feature parameters for normal state identification:

P _N = (Pi + Ρ ₅ ) ^/ β ^/ Pl ^{+ 2} P4 ^/ P3) ^X {P4 ^/ Pi '(P4P6)} (23) GA feature parameters for discriminating outer ring injury

G Α feature parameters for discriminating inner ring injury status

S = (AP _{4 6} ³ ) x ₂ Ζρ ₅ ) ^{33 5} (25) G 識別 feature parameter for identifying rolling element wound state

5) X 2 3) ° · 33 + Ρ 5 0 · 75 (26) In addition, these GA characteristic parameters of variation and statistical theory and potential by regulating base ambiguity creates a criterion for state identification. For example, 'G Α feature parameter for normal state identification And P _N is approximately follows a normal distribution, if you the average value and the standard deviation / ½ and sigma _New for P _N in the normal state, the value of the _New [rho during the identification actual state satisfies the following condition Then, it is judged as “normal state” with about 99.9% confidence. Otherwise, it is judged as “not normal” with about 99.9% confidence.

_{_{β Ν - 3 σ Ν <Ρ}} Ν <μ Ν + 3 σ Ν ^ ') Thus, forward each GA characteristic parameters and criteria determined by the computer to the portable identification device.

After the portable identification device receives each G Α feature parameter and judgment criterion, at the time of state identification, it performs state identification using each GA feature parameter and judgment criterion according to the signal measurement and state identification execution procedure shown in Fig. 1. For example, an identification result is obtained.

3. Example of fuzzy identification mechanism

According to the flow shown in Fig. 1, an example of construction of a state identification system using a fuzzy identification mechanism is shown (references (10)).

Figure 13 shows an example of acceleration waveform data and spectra of four states (normal, eccentric, worn, and local flaws) measured to identify the state of a gear device of a rotating machine. As a result of examining the knowledge to identify these states, the effective frequency domain feature parameters for identifying the four states of this rotating machine are P ^ P ^ Ps ^.

Here

Here, f is the frequency, f ^ is 1/2 of the sampling frequency, and F〗 (f) is the spectrum.

In this case, state identification is performed according to the flow shown in FIG. The characteristic waveform data is a spectrum of 5 kHz or less for identifying a normal state, and a spectrum of 8 kHz or less for identifying other states. The feature parameters used to identify each state are as shown in Fig.14.

Furthermore, the variation and ambiguity of these feature parameters are examined by statistical theory and possibility, and a membership function (criterion) for state identification is created. For example, the characteristic parameter _Pl for normal state identification determined by potential theory, the membership functions of p ₂ p (x) is shown in FIG. 1 5 and FIG 6. For example, the probability distribution function obtained from the characteristic waveform data obtained during actual state identification (the “membership function at the time of identification” in FIGS. 15 and 16) is called the “membership function at the normal state”. the results were matched with "membership function of the state is not the normal state", identification result of the P _l:

Probability of normal condition = 0.95

Possibility of abnormal state = 0.8

p ₂ by the identification result:

Possibility of normal condition = 0.4

Probability of abnormal state = 0.99

Finally, according to the rules of fuzzy inference,

Possibility of normal state = minimum {0. 95, 0.4} = 0.4

Normal state, possibility of state = minimum {0. 8, 0. 99} = 0.8

Accordingly, since the possibility of the abnormal state is greater than the possibility of the normal state, the state is determined to be “non-normal state”.

The other states can be similarly identified according to the flow of FIG.

As described above, the computer selects characteristic parameters for identifying each state by checking the performance, creates a membership function (judgment criterion) for state identification, and then transfers it to the portable identification device.

After the portable identification device receives each feature parameter and membership function (judgment criterion), at the time of state identification, each portable device identifies each feature parameter and membership function (judgment criterion) according to the state identification execution procedure shown in FIG. If it is used to identify the state, an identification result can be obtained. 8. Brief description of drawings

FIG. 1 is a flowchart showing the processing flow of the present invention.

FIG. 2 is a graph showing the configuration of the hardware of the present invention, and the symbols in the figure are as follows. 1 sensor, 2 signal measuring device, 3 computer, 4 portable identification device, 5 sensor Fig. 3 is a graph showing the construction of the portable identification device, and the symbols in the figure are as follows.

1 sensor, 2 amplifiers, 3 filters, 4 processing section, 5 result display, 6 data RAM, 7 AD touch, 8 DC port, 9 SCI, 101 chip CPU, 1 1 Flash R〇M, 1 2 External calculator.

FIG. 4 is a graph showing an example of characteristic waveform data in each frequency domain.

FIG. 5 is a graph showing an example of parameter waveform data and an example of state identification.

Figure 6 is a graph showing a rotating machine.

FIG. 7 is a table showing an example of characteristic parameter values.

Figure 8 is a table showing an example of learning data for a multi-valued neural network (state identification function).

FIG. 9 is a diagram illustrating an example of a multilevel neural network for discriminating the state of a bearing.

FIG. 10 is a table showing an example of the identification result by the multi-valued neural network.

FIG. 11 is a graph showing an example of characteristic waveform data and parameter waveform data in each state of the bearing.

FIG. 12 is a flowchart showing the flow of sequential state identification for identifying each state of the bearing.

FIG. 13 is a graph showing an example of vibration acceleration waveform data and a spectrum in each state of the gear.

FIG. 14 is a flowchart showing a flow of sequential state identification for identifying each state of the gear.

Fig. 15 is a graph showing an example of the membership function of the characteristic parameter p i for identifying the normal state of the gear.

Figure 1 6 is a graph showing an example of the membership function of the characteristic parameter p ₂ for identifying the normal state of the gear. References

(1) Patent publication Hei 9—3 3 2 9 8

(2) Patent publication 2 0 0 0—1 7 1 2 9 1

(3) Patent publication 7--3 1 8 4 5 7

(4) Toshio Toyoda, Peng Chen, Taketo Mizoda: Extraction of fault signal by statistical test of spectrum, Journal of the Japan Society of Precision Engineering, Vol. 58, No. 6, pp. 1041-1046, 1992.

(5) Peng CHEN, Toshio TOYOTA: Extraction Method of Failure Signal by Genetic Algorithm and the Application to Inspection and Diagnosis Robot, IEICE TRANSACTIONS on Fundamenta 丄 s of ectronics, Communications and Computer Science, VOL.E78—A, No. 12 , pp. 1622-1626, 1995.

(6) Patent publication 2 0 0 1-1 8 0 7 0 2

(7) Peng Chen, Toshio Toyoda: Genetic programming for self-repetition of characteristic parameters in the frequency domain, J .; «Memoir of Papers (C), Vol. 65 No. 633, ρ · 1946-1953, 1998.

(8) Tomo Chen, Toshio Toyoda: A method for acquiring diagnostic knowledge using rough sets and a fault diagnosis method using linear interpolation type dual networks, Journal of the Society of Equipment Management, Vol. 9, No. 3, p. 383-388, 1997.

(9) Peng CHEN, Toshio TOYOTA: Sequential Self-reorganization Method of Symptom Parameters and Identification Method of Membership Function for Fuzzy Diagnosis, Proceedings of FUZZ-IEEE '97, Vol. 2, pp. 433-440, 1997.

(10) Chen Peng, Feng Fang, Toyoda Toshio: Extraction of characteristic frequency bands and sequential diagnosis of equipment abnormalities based on possibility theory, Journal of Japan Society for Reliability Studies, Vol. 24, No. 4, pp. 311-321, 2002.

Claims

The scope of the claims

1. In order to realize the state identification of an object on site with a pocket-sized portable identification device, the computer must learn the state identification function that the portable identification device should have, using the measured waveform data. A second step of transferring elements required for the state identification function constructed by the computer to a portable identification device, and a state identification function identical to the state identification function using the elements. A third step of constructing the object in the portable identification device, a fourth step of identifying the behavior of the object using the portable identification device, waveform data acquired by the portable identification device, and obtained data. Was! ! A state identification method and state identification system, comprising: a fifth step of transferring another result to a computer; and a sixth step of performing a necessary cause analysis and state prediction in the computer.

2. In the first step and the second step described in the above item 1, a first step of determining various states to be identified in advance with respect to the object, and a plurality of waveform data reflecting the characteristics of the various states. A second step of measuring and acquiring the waveform data in the same frequency band, a third step of obtaining characteristic waveform data from which noise has been removed from the waveform data, and identifying the various states using the characteristic waveform data. A fourth step of establishing knowledge for establishing the state, a fifth step of establishing a state identification function for identifying the various states by learning the knowledge, and a portable identification of elements required for the state identification function. A method for learning a state identification function, comprising: a sixth step of transferring to a device.

3. In the third step and the fourth step described in the above item 1, a first step of receiving an element necessary for the state identification function, a second step of constructing a state identification function using the element, A ^third step of setting measurement conditions and identification conditions for signal measurement and state identification of an object, a fourth step of measuring and acquiring waveform data of an object in a plurality of frequency bands, and A fifth step of obtaining characteristic waveform data from which noise has been removed, a sixth step of performing state identification by the state identification function using the characteristic waveform data, and a seventh step of displaying the result of the identification. A state identification method comprising:

4. The method according to item 1, wherein the state identification function is constructed using a neural network, a multi-valued neural network, a GA feature parameter, or a fuzzy identification mechanism. Method and construction method of state identification system.

5. In the method described in the above item 1, the element required for the state identification function is a weighting factor in the case of a dual network or a multi-valued dual network, and a factor in the case of a GA feature parameter. A method for constructing a state identification system that is characterized by good GA feature parameters and state determination criteria for identification obtained by a genetic algorithm, and a membership function in the case of a fuzzy identification mechanism.

6. In the fourth step described in the second item and the sixth step described in the third item, the characteristic waveform data is further converted into parameter waveform data in accordance with a measurement condition of the waveform data, A state identification method and a state identification system construction method, wherein the state identification function is constructed using parameter waveform data.

7. The method according to item 6, wherein the parameter waveform data includes parameter waveform data calculated by a dimensional feature parameter and parameter waveform data calculated by a non-dimensional feature parameter. A method for calculating parameter waveform data, which is parameter waveform data.

8. The method according to the above item 6, wherein the number of characteristic waveform data for determining the parameter waveform data is determined.

9. A state identification system for measuring a signal and identifying a state of an object, comprising a sensor, a signal measuring device, a computer, and a portable identification device for acquiring waveform data of the object. The method according to the first aspect is performed.

10. Portable identification device equipped with a sensor, amplifier, filter, processing unit, data storage memory, and display output device for acquiring waveform data of the object, which performs signal measurement and state identification of the object Wherein the apparatus performs the method of item ³ above.