WO2010116222A1 - 信号識別方法および信号識別装置 - Google Patents
信号識別方法および信号識別装置 Download PDFInfo
- Publication number
- WO2010116222A1 WO2010116222A1 PCT/IB2010/000579 IB2010000579W WO2010116222A1 WO 2010116222 A1 WO2010116222 A1 WO 2010116222A1 IB 2010000579 W IB2010000579 W IB 2010000579W WO 2010116222 A1 WO2010116222 A1 WO 2010116222A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- learning
- inspection
- time
- data
- subject
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 34
- 230000002159 abnormal effect Effects 0.000 claims abstract description 24
- 238000007689 inspection Methods 0.000 claims description 91
- 241001455214 Acinonyx jubatus Species 0.000 claims description 11
- 230000000295 complement effect Effects 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 9
- 230000001174 ascending effect Effects 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 4
- 102100030386 Granzyme A Human genes 0.000 claims 1
- 101001009599 Homo sapiens Granzyme A Proteins 0.000 claims 1
- 230000009466 transformation Effects 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract description 2
- 230000004044 response Effects 0.000 abstract 1
- 238000012545 processing Methods 0.000 description 7
- 238000006243 chemical reaction Methods 0.000 description 5
- 230000011664 signaling Effects 0.000 description 5
- 238000000926 separation method Methods 0.000 description 4
- 230000033001 locomotion Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 241000209202 Bromus secalinus Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010561 standard procedure Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/78—Detection of presence or absence of voice signals
Definitions
- the present invention relates to a signal-specific device for identifying the state of an inspection object.
- the other device that is coming is learning
- Another conventional device that created a rastering map at the time of learning inputs inspection data into the clustering map based on the state of the inspection / inspection object, and calculates the crid separation for the inspection on the cluster link. Therefore, the conventional separate device has an output that minimizes crid separation.
- the inspection data is classified into categories, and the state of the inspection target is identified according to the category into which the inspection data is classified. If you use a different device, you can learn automatically without any specialized knowledge.
- the conventional separate device can automatically learn without specialized knowledge.
- a signal-separating device comprising a plurality of each for identifying the state of the object to be examined, learning to set a constant standard for the object to be examined, Based on the standard method for inspection, it is a row signal method that identifies the state of the inspection target, and at the time of learning, a number of numbers including normal and abnormal numbers are input.
- the features are extracted from each learning issue using a specified method, and is set as learning, and the weight is set or changed for each of the above-mentioned titers.
- the standard of the range of each learning data is matched, and whether the learning number that is each learning is normal
- the sum of only the above-mentioned determined data is output as an erroneous determination rate for each of the above-mentioned criteria, and the determination rate is minimized for each of the above.
- the standard is applied to create a complement and only the above is set or changed. The complement with the smallest decision rate is selected as the complement from the complements created for each increment, and is calculated using the decision rate for each increment.
- an inspection number indicating the state of the subject of inspection is input, and the characteristics are extracted from the inspection number using the method, and is used as the inspection cheetah.
- the examination data is checked against the above-mentioned standard to determine whether or not the inspection target is in a normal state, and the inspection target is determined to be in a normal state.
- a signaling method characterized in that if the sum is determined after the subject to be examined is in an abnormal state, the subject is finally judged to be in a normal state.
- the above-mentioned examination may be acceptable.
- the range of the learning data may be set in ascending or descending order, and the arranged values may be set in order.
- 1 means the constant () of the inspection object with respect to time.
- the extraction method may be a short-time Fourier transform.
- continuous weblet replacement may be used.
- each of the two states is provided with a plurality of for identifying the state of the subject of the examination, and learning to set the standard of the subject, and based on the state of the subject of the subject.
- This is a device that separates the signal that contains the normal number and the abnormal number at the time of learning, and the inspection number that indicates the state of the inspection object is input to the inspection.
- a feature stage for extracting a feature from the number using a predetermined method;
- the feature extracted from each learning issue is set as a weight, a weight updating method for setting or changing the weight for each data, By setting the range of each learning cheetah to, and comparing the above with each other, the criteria for the range of each learning teeter is matched while changing the judgment criteria.
- a stage that outputs the sum of only the above as an erroneous determination rate, and a supplementary standard are created by applying the standard when the determination rate that is output in the previous stage is the minimum.
- the complement having the smallest determination rate is selected as the complement from the complements created for each step by the step, and the determination rate is used for each of the steps.
- a weight stage instructing the further means to increase the titer determined by, and extracted from the examination number in the stage
- the determined feature is used as the inspection data to determine whether or not the inspection target is in a normal state by collating with the standard for the inspection, and the inspection target is in a normal state
- An inspection stage that finally determines that the inspection object is in a normal state when the sum of the determination is that the inspection object is in an abnormal state. Signal Separate device is provided.
- the first state when selecting the smallest judgment rate, only the learning data judged by the previous selection is increased, and only the electric signals normally determined by the previous selection are increased. Change to decrease • By selecting each time only the setting or change is made, the optimum setting can be automatically made during learning without any specialized knowledge. In addition, according to the request, it is only necessary to integrate the results in the inspection, so that the interval between inspections can be shortened as compared with the case of using a neural network.
- the final judgment can be made as to whether or not the inspection object is in a normal state, taking into consideration the degree of calculation separately.
- a short-time free change will show signs of It is necessary to have multiple fills because it is configured with a criterion that uses only simple elements. As a result, it is possible to reduce the required memo amount compared to other conventional devices.
- the enclosure is limited to a short time frame, the learning time can be shortened compared to other conventional devices, and the program size can be reduced.
- the continuous wavelet conversion is configured with a criterion that uses only the necessary elements, and has a criterion.
- the required amount of memos can be reduced compared to other conventional devices.
- the learning time can be shortened and the program size can be reduced as compared with other conventional devices.
- Another apparatus in the embodiment uses the method of O to identify the inspection object (whether the inspection object is in a normal state or not).
- Another device of the implementation is used during learning.
- the other devices are: signal, feature () 2, weighting unit 3, learning (training), () 5, (, (7, () Equipped with 8 and.
- the inspection object (not) is mainly equipment and equipment including rotating equipment, but is not limited to the above.
- the sensor converts the detected motion into a analog signal. This is a symbol representing the movement of the test object with respect to electricity and time converted by the sensor.
- Micro 1 converts the detected sound into a signal. Electricity converted by a microphone is a symbol representing the change in the sound of a test object with respect to time.
- a normal sample and an abnormal sample that are are manually trained as learning numbers. For each learned learning, the noise is removed from the learning signal.
- the feature is a short-time Fourier transform of the learning signal using, and the feature F () (bf in the combination of time (number) b frequency f is featured.
- the sample data is the sum of F f of time b and wave number f given by the feature. For example, sample of time 3 and wave number f 2 ((bf is expressed as F bf of sumb b and wave number f 2 represents Rd b) of Rd b is a us window centered at time b.
- R tb R tb
- an analog signal is manually input from the vibration sensor or microphone as the inspection signal based on the state of the inspection object (not).
- “2”, which is the input of the survey code is a short-time Fourier transform of the test signal and is characterized by the combination of time b and wave number f (bf).
- bf wave number f
- (bf) with wavenumber f is the inspection cheetah.
- the time b and the wave number f 5 are expressed as bf 5 () Weight to instruct to change () 3 are provided.
- W represents only sample data. Only the sample is represented by W and the sample
- F () (bf) of b wavenumber f was taken out from all cheetahs for each combination of time b and wavenumber f, and the extracted elements F () (bf were arranged in.
- Element c) is F in ascending order.
- the sum of W (the W of the learning data mis-determined at the time of learning is Judgment refers to judging whether an abnormal sample is included even though it is a sample that contains element F c), or a normal sample is judged although the above is a sample. If the misjudgment rate you have issued is greater than 5, set (1 judgment rate) to the new misjudgment rate.
- () D C b f) is created for each combination of time b and wavenumber f by applying the constant criterion when the error determination rate E given in 5 is minimized.
- the standard is that if the misjudgment rate given in 5 is 5 or lower and the element is, include element F
- D C (b f) selects D C (b f) with the smallest decision rate from the number of D C f created for each combination of time b and wave number f as an identification (investigation) D C). Is a number for identifying D C).
- D C) represents the selected for the eye. 7 is calculated by substituting the decision rate of D C) into Equation 3 for the selected D C, and substituting the confidence () for.
- Each time W is set or changed, 7 can select (and select a total number of identifiers D C ().
- the numbers D C) are stored in memory 7 alongside the selection.
- the element F b f) extracted from the inspection code in step 2 is compared 6 times to determine whether the inspection target is in a normal state. If DC () determines that the subject to be examined is in the normal state, Inspection 8 sets the value obtained by adding the previous DC) to) to the new S. If D C) determines that the subject to be examined is abnormal, Inspection 8 takes the value obtained by subtracting () of D C) from the previous S as the new S. After the above determination is made for all DCs, the final judgment is that the subject of inspection is normal if the trust S is above. If S of is, the test will finally determine that the test object is abnormal. The final judgment of 8 is output to the output.
- a sample including normal samples and abnormal samples is input to feature 2 (S).
- Samples are initially set (S).
- the features and learning numbers are converted into short-term frying elements (elements) (b
- the learning that is the combination is extracted (3).
- step 7 it is determined whether (b) from step S9 to step S has been performed for all b (S). If not all b are processed, b is set (S7, step S9 to step S are repeated.
- weight 3 changes ⁇ () for each learning data (S 8). 3, the training data corrected by the latest DEC) is instructed to multiply ()) and reduce the weight W (). Instructs weight 3 to weight 3 to increase the weight VI by dividing by ().
- step 7 it is determined whether or not D C) in () has been selected (S). If all D C) are not selected, they are set (S2) and step S to step S are repeated. If all D C () are selected, learning is completed.
- step S 3 the determination rate VI is calculated using ⁇ ⁇ including the determined element () (S). Part 6 determines whether the determination rate is greater than 5 (S5). If the decision rate is greater than 5, 6 will reverse the decision criteria and (1 decision rate e will be the new false decision rate (). Change the criteria so that a normal sample of data containing () is judged. Or, if the misjudgment rate is 5 or below, 6 is the smallest judgment rate so far.
- step S27 If the decision rate of the number of times is the smallest, DC (update the constant standard of bf (S 8), and 6 is processed for all (step S 3 From step S to step S28 is determined (S2) If all processing is not performed, c C is set (S3), and step S to step S28 are repeated. If processing is performed for all, 6 is the DC of time b and wavenumber f (determine bf (S). The final criteria have been renewed.
- an inspection signal is input from the signal to the feature (S).
- the trust S is set to the initial value (S3, is set to the initial value (S4).
- the test 8 is set to DC), and the test data is set (S5).
- inspection 8 is (S of the previous DEC)) (S7)
- the inspection target is in an abnormal state. If it is judged as DEC) (S), the inspection 8 shall be ()) of () S) up to (S). In 8, it is determined whether or not the inspection data has been set in all DC () (S). If all DCs are not tested, (S5), and steps S5 to S48 are repeated. If inspection data is set for all DCs (), inspection 8 determines whether or not the trust S is up (if S in S5 is up)
- S 8 is the final decision that the test object is in a normal state (S5). If S of is •, the final determination is made that the object to be inspected is in an abnormal state (the final determination of S53 is output to the output).
- the learning W determined by the previous DC is increased, and the previously selected DC ()
- the optimum setting during learning is automatically set. In particular, it is possible to make a final decision as to whether or not the inspection target is in a normal state, taking into consideration the degree of calculation calculated separately.
- the time can be shortened as compared with the case of using a global network.
- the necessary element F (DC with a criterion using only bf)) is formed from that.
- the required amount of memo can be reduced as compared with other conventional devices, and according to the present embodiment, the enclosure is limited to the free space for a short time. Compared to devices, the learning time can be shortened and the program size can be reduced.
- the signal-separating device according to 2 is different from the signal-separating device according to the implementation in that continuous wavelet conversion is performed instead of short-time free-wave conversion of the electric signal.
- Equation 5 each time a normal sample and an abnormal sample are included, noise is removed from the learning signal, and then the learning signal is continuously wavelet-converted using Equation 5 to match the time parameter a.
- Element Y is characterized by the combination of ba. 5 indicates Y ba of sample, b, and parameter a, and 6 indicates the wavelet V () of 5.
- x) is an electric signal.
- the feature 2 is to convert the inspection signal into a continuous wavelet and extract the inspection data that is the result of time b and parameter a Y (ba). .
- the individual subjects including normal samples and abnormal samples are human-powered by features (S. Set to initial sample (S62).
- the features are the continuous wavelet transform of the learning numbers, element Y) ba
- the learning that is the result of) is extracted (). (Evaluate whether or not learning has been extracted from the No. (). If no learning cheetah has been issued from all No, it is set (S 5), and Step S 3 is repeated.)
- the weight 3 sets the W of each learning data to the initial S). , Is set to initial (S7), time b is set to initial (S6), parameter a is set to initial (S6)
- step S7 it is determined whether or not steps (S7) to (S73) have been performed for all parameters a at time b (S7). If the process 3 is not performed for parameter a in b, a is set (S75), and steps S7 to S73 are repeated. If the process 3 is performed for the parameter a, it is determined whether or not (from step S to step S7) is performed for all b (S7). If the processing of 4 is not performed for b, bb is set (S7) Steps S6 to S7 are repeated.
- the weight changes each learning VI (S78). 3. Instruct the weight 3 to reduce the weight by applying () () to the training data corrected by the newly updated DEC). On the other hand, for the learning data determined, the weight weight 3 is instructed to increase the weight VI (by dividing by.
- step 7 it is determined whether or not (DEC) in () has been selected (S7). If all DC) are not selected, they are set (S), and steps S8 to S78 are repeated. If all D C (are selected, complete the learning.
- step S7 the method of DC b a) of time b and parameter a in step S7 will be described.
- every cheetah Y) (a) is arranged into (S2) for each combination of time b and parameter a.
- the obtained element Y) becomes Y) Y ( ⁇ ⁇ Y in order from the beginning.
- set to initial Y) S.
- create DC (ba) at Criteria for comparing Y c) to DC (ba) in 6 and judging whether the cheetah containing full Y c) is a normal sample and judging the data containing Y) above as abnormal samples (S 3)
- the ⁇ is calculated in step S 3 by using the W of the data including the determined element Y (), and the determination rate (is calculated (S2).
- step S3 to step S8 determines whether or not processing (step S3 to step S8) has been performed for all items (S2; set to C if all processing has not been performed). (S), and repeats step S to step S. If processing is performed for all, is: • DC ba at the time of parameter a is determined (S3. The specified DC (ba) With the determined criteria.
- step 2 the inspection code is converted into continuous wavelets, and the inspection data that is the result of element Y b) is extracted (S).
- the subsequent operation is the same as in the implementation (S to S53).
- a continuous weblet conversion has a sign of, and DC (with a criterion using only the necessary element Y (ba)) is formed, so multiple fills are included. It is necessary to prepare. As a result, the required memo amount can be reduced as compared with other conventional devices.
- the learning time can be shortened and the program size can be reduced as compared with another conventional apparatus.
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
Claims
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010800120253A CN102362282A (zh) | 2009-03-26 | 2010-03-18 | 信号识别方法及信号识别装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2009-077688 | 2009-03-26 | ||
JP2009077688A JP2010231455A (ja) | 2009-03-26 | 2009-03-26 | 信号識別方法および信号識別装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010116222A1 true WO2010116222A1 (ja) | 2010-10-14 |
Family
ID=42935684
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/IB2010/000579 WO2010116222A1 (ja) | 2009-03-26 | 2010-03-18 | 信号識別方法および信号識別装置 |
Country Status (4)
Country | Link |
---|---|
JP (1) | JP2010231455A (ja) |
KR (1) | KR20110122748A (ja) |
CN (1) | CN102362282A (ja) |
WO (1) | WO2010116222A1 (ja) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101804051B1 (ko) * | 2016-05-17 | 2017-12-01 | 유광룡 | 대상체 검사를 위한 센터링장치 |
JP6970694B2 (ja) * | 2017-01-27 | 2021-11-24 | 株式会社 エニイワイヤ | 状態判定システム |
JP6924413B2 (ja) * | 2017-12-25 | 2021-08-25 | オムロン株式会社 | データ生成装置、データ生成方法及びデータ生成プログラム |
JP6844564B2 (ja) * | 2018-03-14 | 2021-03-17 | オムロン株式会社 | 検査システム、識別システム、及び学習データ生成装置 |
CN108836574A (zh) * | 2018-06-20 | 2018-11-20 | 广州智能装备研究院有限公司 | 一种利用颈部振动的人工智能发声系统及其发声方法 |
JP7175216B2 (ja) * | 2019-02-15 | 2022-11-18 | ルネサスエレクトロニクス株式会社 | 異常検知装置、異常検知システム、異常検知方法 |
JP2020201143A (ja) * | 2019-06-11 | 2020-12-17 | 株式会社日立製作所 | 自動点検システム |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006154961A (ja) * | 2004-11-25 | 2006-06-15 | Sumitomo Electric Ind Ltd | 交通音識別装置、コンピュータを交通音識別装置として機能させるための交通音判定プログラム、記録媒体および交通音判定方法 |
JP2008065544A (ja) * | 2006-09-06 | 2008-03-21 | Toshiba Corp | 識別器及びその方法 |
JP2008217589A (ja) * | 2007-03-06 | 2008-09-18 | Toshiba Corp | 学習装置及びパターン認識装置 |
JP2008250908A (ja) * | 2007-03-30 | 2008-10-16 | Toshiba Corp | 映像判別方法および映像判別装置 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3340357B2 (ja) * | 1997-08-25 | 2002-11-05 | 三菱電機株式会社 | 情報処理装置管理システム及び情報処理装置管理方法 |
JP2004234175A (ja) * | 2003-01-29 | 2004-08-19 | Matsushita Electric Ind Co Ltd | コンテンツ検索装置およびそのプログラム |
JP2006031387A (ja) * | 2004-07-15 | 2006-02-02 | Yamaha Motor Co Ltd | 画像認識装置、画像認識方法、画像認識プログラムおよび画像認識プログラムを記録した記録媒体 |
-
2009
- 2009-03-26 JP JP2009077688A patent/JP2010231455A/ja active Pending
-
2010
- 2010-03-18 KR KR1020117022207A patent/KR20110122748A/ko not_active Application Discontinuation
- 2010-03-18 CN CN2010800120253A patent/CN102362282A/zh active Pending
- 2010-03-18 WO PCT/IB2010/000579 patent/WO2010116222A1/ja active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2006154961A (ja) * | 2004-11-25 | 2006-06-15 | Sumitomo Electric Ind Ltd | 交通音識別装置、コンピュータを交通音識別装置として機能させるための交通音判定プログラム、記録媒体および交通音判定方法 |
JP2008065544A (ja) * | 2006-09-06 | 2008-03-21 | Toshiba Corp | 識別器及びその方法 |
JP2008217589A (ja) * | 2007-03-06 | 2008-09-18 | Toshiba Corp | 学習装置及びパターン認識装置 |
JP2008250908A (ja) * | 2007-03-30 | 2008-10-16 | Toshiba Corp | 映像判別方法および映像判別装置 |
Also Published As
Publication number | Publication date |
---|---|
KR20110122748A (ko) | 2011-11-10 |
CN102362282A (zh) | 2012-02-22 |
JP2010231455A (ja) | 2010-10-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2010116222A1 (ja) | 信号識別方法および信号識別装置 | |
KR101493552B1 (ko) | 신호판정방법, 신호판정장치, 프로그램, 신호판정시스템 | |
KR101075824B1 (ko) | 신호 식별 장치용 학습 데이터 세트 최적화 방법 및 학습 데이터 세트를 최적화할 수 있는 신호 식별 장치 | |
CN111914883A (zh) | 一种基于深度融合网络的主轴轴承状态评估方法及装置 | |
CN114155879B (zh) | 一种利用时频融合补偿异常感知及稳定性的异音检测方法 | |
KR20200092143A (ko) | 딥러닝 신경망을 이용한 디스플레이 패널 불량 진단 시스템 및 방법 | |
US5860062A (en) | Speech recognition apparatus and speech recognition method | |
CN111178548A (zh) | 集成学习预测方法与系统 | |
CN110992985A (zh) | 识别跑步机异音的识别模型确定方法、识别方法、系统 | |
CN111277459A (zh) | 一种设备异常检测方法、装置和机器可读存储介质 | |
CN109240276B (zh) | 基于故障敏感主元选择的多块pca故障监测方法 | |
CN108470194B (zh) | 一种特征筛选方法及装置 | |
CN108829878A (zh) | 一种工业实验数据异常点检测方法及装置 | |
Renouard et al. | Toward false event detection and quarry blast versus earthquake discrimination in an operational setting using semiautomated machine learning | |
TW202111352A (zh) | 地震觀測裝置、地震觀測方法以及記錄地震觀測程式的記錄媒體 | |
JP3778177B2 (ja) | 物体検査装置 | |
CN110910900B (zh) | 音质异常数据检测方法、装置、电子设备及存储介质 | |
CN109829479A (zh) | 一种用于传感器的分类器模型信息自动更新系统及传感器更换方法 | |
US20220171984A1 (en) | Determination difference display apparatus, determination difference display method, and computer readable medium storing program | |
KR101983603B1 (ko) | 증강현실을 활용한 원자력 발전설비의 기계학습식 예측진단장치 | |
EP3913383B1 (en) | Method and system for detecting anomalies in a spectrogram, spectrum or signal | |
JP2020030138A (ja) | 信号解析装置及び信号解析方法並びに信号解析プログラム | |
JP5134605B2 (ja) | 信号識別装置 | |
CN112782365A (zh) | 一种基于多级分类器的混合气体识别方法 | |
CN113240156A (zh) | 一种架空输电线路舞动状态在线预测方法及系统 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201080012025.3 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10761242 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20117022207 Country of ref document: KR Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 10761242 Country of ref document: EP Kind code of ref document: A1 |