CN106096634B - Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm - Google Patents

Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm Download PDF

Info

Publication number
CN106096634B
CN106096634B CN201610392752.6A CN201610392752A CN106096634B CN 106096634 B CN106096634 B CN 106096634B CN 201610392752 A CN201610392752 A CN 201610392752A CN 106096634 B CN106096634 B CN 106096634B
Authority
CN
China
Prior art keywords
primitive
data
window
length
sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201610392752.6A
Other languages
Chinese (zh)
Other versions
CN106096634A (en
Inventor
顾晓丹
邓方
刘畅
孙健
陈杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN201610392752.6A priority Critical patent/CN106096634B/en
Publication of CN106096634A publication Critical patent/CN106096634A/en
Application granted granted Critical
Publication of CN106096634B publication Critical patent/CN106096634B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6201Matching; Proximity measures
    • G06K9/6215Proximity measures, i.e. similarity or distance measures

Abstract

The invention discloses the fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm, this method is using interval halving method extraction identification trend, and constantly change the initial point and end point of section window as the case may be in extraction process, adaptively changing section size, to obtain higher extraction accuracy, then the characteristic trend of real-time tendency and the various typical faults in rule-based knowledge base is matched by smudge tendency matching algorithm, the real-time diagnosis system failure;The present invention can improve the accuracy and real-time of sensor fault identification.

Description

Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm
Technical field
The invention belongs to the technical field of Intelligent Information Processing, and in particular to one kind based on Adaptive windowing mental arithmetic method with The fault detection method of interval halving algorithm.
Background technology
Sensor (English name:Transducer/sensor) it is a kind of detection means, measured letter can be experienced Breath, and the information that will can be experienced, electric signal or the information output of other required forms are for conversion into according to certain rules, to meet The requirement such as transmission, processing, storage, display, record and control of information, is widely used in various control systems.And as The window of systematic procedure state is solved, the accuracy of its measurement result directly affects the operation of system.While most of sensors Working environment is all relatively more severe, therefore they inevitably often break down because of various reasons in use. Mended once sensor fault is had by system detectio, it is necessary to carry out certain online or offline failure for different faults type Repay, therefore, sensor fault is recognized and is just particularly important.
Fault identification is carried out to sensor and belongs to pattern recognition problem, wherein mainly including feature extraction and pattern classification two Individual part.The selection of feature then directly affects follow-up with the basis that extraction is pattern classification, the extraction efficiency of characteristic information Practise the accuracy rate with recognition result.Because nonlinear system has the characteristics that the limitation of complexity and modeling method, thus it is right Very limited in the result of study of fault of nonlinear system diagnosis, existing certain methods are mainly by means of intelligent control method and line Property method.These methods are largely all based on known mathematical modeling.And modern control system generally has higher-dimension, non-thread The features such as property, close coupling, random noise and input delay, it is difficult to establish accurate mathematical modeling, or even it may not use Analytic equation describes.
At present, the common method independent of mathematical modeling mainly has several classes representational:Side based on wavelet transformation Method, the method based on neutral net, the method based on fuzzy logic, the method based on Statistic analysis models, based on expert system Method, the method based on fault tree, method based on qualitative model etc..As the fault detect side independent of mathematical modeling One of method, qualitiative trends analysis is exactly a kind of analysis method based on data-driven, has and has only required that process data can be real The advantages of now monitoring to process.This feature has very important meaning in actual applications, because in some industrial mistakes Cheng Zhong, unique available information is exactly process data.In addition, the method independent of process mathematical model also easily utilizes The information such as operating experience, process knowledge, director's failure logging, and correct utilize of these information often plays what is got twice the result with half the effort Effect.But traditional qualitiative trends analysis method still has many problems, the segment width of such as extraction is difficult to adaptive, calculation Method is complicated, relies on artificial setting threshold value etc..
The content of the invention
In view of this, the invention provides a kind of failure inspection based on Adaptive windowing mental arithmetic method with interval halving algorithm Survey method, it is possible to increase the accuracy and real-time of sensor fault identification.
Realize that technical scheme is as follows:
A kind of fault detection method based on Adaptive windowing mental arithmetic method and interval halving algorithm, comprises the following steps:
Step 1: the sensing data to be detected to one section starts with interval halving algorithm extraction original from data starting point Language, and judge whether its primitive is " A " in primitive, if it is not, then judging since primitive is not the data segment original position of " A " Point occurs for failure, point occurs up to the end point of data to be tested section is fault data section from failure;If data to be tested section Primitive is " A ", then carries out the detection of next section of data to be tested;
Step 2: since occurring point for failure, it is right with the 1/10 of data to be tested length sliding window size by default Data in first sliding window section, which carry out offline interval halving fitting, makes the data be converted to polynomial fitting sequence, Polynomial sequence is converted to by Sequence of Primitive Elements according to polynomial derived indice;
Step 3: determine the starting point and length of window of next sliding window:Judge that a upper sliding window is last Whether one primitive is linear basis element, if then judging whether its length is more than default linear primitive length critical value, if greatly In the end point that the starting point of then next window is a upper window, otherwise the data corresponding to last primitive are included into Next window;If last primitive is non-linear primitive, judge whether its length is more than default most short primitive length Value, if the next window more than if starting point be a upper window end point, otherwise by corresponding to last primitive Data are included into next window;If the data corresponding to last primitive of a upper window are included into next window, window Length=last primitive length+acquiescence sliding window size;Otherwise, length of window=acquiescence sliding window size;
Step 4: the data in fixed sliding window are converted into Sequence of Primitive Elements;
Step 5: repeat step three and step 4 are until the data of fault data section are all converted to Sequence of Primitive Elements, by institute State Sequence of Primitive Elements and carry out fuzzy logic matching with multiple fault signature Sequence of Primitive Elements in rule-based knowledge base, draw similarity respectively SI;
Step 6: find out the highest similarity SI of gained in step 5max, and itself and default decision-making critical value are carried out Compare, if SImaxMore than decision-making critical value, then the fault type of data to be tested is SImaxCorresponding fault type;If SImax Less than or equal to decision-making critical value, then the fault type for assert data to be tested is new failure.
Further, linear primitive length critical value is more than most short primitive length value.
Beneficial effect:
1st, the invention provides a kind of online fault detect being combined based on sliding window algorithm with interval halving algorithm Method, sensor normal mode can be distinguished with fault mode.
2nd, compared with traditional offline interval halving algorithm, the efficiency and accuracy of trend abstraction all obtain the inventive method Arrived lifting, and the size of sliding window can be flexibly selected according to size, flexibility is higher, generalization compared with It is good.
Brief description of the drawings
Fig. 1 is inventive sensor fault diagnosis block flow diagram.
Fig. 2 is primitive schematic diagram.
Fig. 3 is the fitting result chart that sliding window size is 75.
Fig. 4 is the fitting result chart that sliding window size is 180.
Fig. 5 is Adaptive windowing mouth algorithm flow chart of the present invention.
Fig. 6 is fault diagnosis framework of the present invention based on qualitiative trends analysis.
Embodiment
The present invention will now be described in detail with reference to the accompanying drawings and examples.
, should the invention provides a kind of fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm Method constantly changes the first of section window as the case may be using interval halving method extraction identification trend in extraction process Initial point and end point, adaptively changing section size, to obtain higher extraction accuracy, then matched and calculated by smudge tendency Method matches the characteristic trend of real-time tendency and the various typical faults in rule-based knowledge base, the real-time diagnosis system failure.This In using the partial data in Tennessee-Yi Siman chemical processes, Tennessee-Yi Siman processes (TEP) be U.S. Tian Na Created according to real chemical industry production process one of two experts of the western one entitled Yi Siman in state chemical company Chemical industry process emulation platform based on multivariate technique.TEP simulation system datas include a kind of normal status data and 21 kinds of fault state datas, wherein 15 are known fault, 16-21 is unknown failure, and the data under every kind of state include training Collection and test set part, wherein training set have 480 groups of data, and test set has 960 groups of data, and whole TEP systems include 52 Variable, because what we studied is the trend abstraction of single sensing data, therefore which part is taken to be known to occur here The variable of failure carries out data analysis.
Choose each 480 groups of sensor output data under 5 kinds of different faults states, i.e. Xij(i=1,2 ..., 5;J=1, 2 ..., 480), to make the signal characteristic of extraction not influenceed by amplitude, signal is standardized first:
Wherein:XijRepresent the sensor output signal of different mode, E (Xij) it is XijAverage,For XijStandard deviation.
As shown in figure 1, the present invention comprises the following steps:
Step 1: the sensing data to be detected to one section starts with interval halving algorithm extraction original from data starting point Language, and judge whether its primitive is " A ", if it is not, being then determined as that failure is sent out since primitive is not the data segment original position of " A " Raw point, point occurs up to the end point of data to be tested section is fault data section from failure;If data to be tested section primitive is A, Then carry out the detection of next section of data to be tested;
Primitive basic conception introduction:
In qualitiative trends description language, the seven kinds of primitives represented with first derivative and quadratic derivative symbols are defined (i.e. Primitive), as shown in Figure 2.First derivative and second dervative are represented respectively in figure bracket.
Represent for any one group trend data carry out interval halving (for interval halving algorithm specific steps this In just no longer do excessive introduction) after, the trend fragment that an exponent number is up to second order can be obtained.(1) if the fragment is single order, It is carried out being easy to obtain its corresponding primitive according to Fig. 2 after seeking first derivative;(2) if the fragment is second order, it is first asked First derivative at two-end-point, if the first derivative jack per line at two-end-point, by it compared with zero after be easy to according to Fig. 2 Obtain corresponding primitive;(3) if first derivative contrary sign at two-end-point, illustrate that the section is not unimodal section, it is necessary to again Split, find the point that the fragment first derivative is zero and be segmented in the section again, take its front half section to repeat in (2) Step obtains primitive.Remaining Data duplication said process.The primitive being made up of multiple primitives is finally obtained, or is called primitive Sequence.
Step 2: since occurring point for failure, with the 1/10 of data to be tested segment length sliding window size by default, Data in first sliding window section are carried out with offline interval halving fitting makes it be converted to polynomial fitting sequence, will Fault data section is divided into several unimodal sections, is fitted unimodal section with quadratic polynomial, then fault data section is converted into plan Close polynomial sequence:
Y (t)={ y1,y2,...yN}→y(t)≈{Q1(t),Q2(t),...QN(t)}
Wherein, y (t)={ y1,y2,...yNIt is the cycle sampled data points such as a series of, t is the sampling time, y1,y2, ...yNData respectively corresponding to several unimodal sections,For k-th of unimodal area Between quadratic fit multinomial.K=1,2 ..., N;Q can be obtained according to least square methodk(t) regression coefficient tki,tkfThe starting point and end point in respectively k-th unimodal section;
Polynomial sequence is converted to by Sequence of Primitive Elements according to polynomial derived indice;I.e.:
y(t)≈{Q1(t),Q2(t),...QN(t) } → tr={ P1,P2,...PN}
Wherein, P1,P2,...PNPrimitive corresponding to respectively each multinomial.
Fig. 3 and Fig. 4 illustrates the partial fitting effect of the different sliding window sizes for same one piece of data, it is clear that from The length of it can be seen from the figure that sliding window is influential for fitting precision, and length of window is 75 in Fig. 3, fits what is come Primitive is ' C ', and length of window is 180 in Fig. 4, and it is ' A ' to fit the primitive come, it is clear that essence of the precision of the latter than the former Degree is high, therefore it is concluded that:Sliding window size can not be too small, otherwise can influence to be fitted the degree of accuracy, in guaranteed efficiency In premise, it should set tune up window as far as possible.Drawn after multiple test, acquiescence sliding window size is in number to be detected It is more suitable according to 1/10 or so of segment length, sliding window size is given tacit consent to here elects 100 as.
Step 3: determine the starting point and length of window of next sliding window:Judge that a upper sliding window is last Whether one primitive is linear basis element, if then judging whether its length is more than default linear basis element length critical value, if greatly In the end point that the starting point of then next window is a upper window, otherwise the data corresponding to last primitive are included into The starting point of next window, i.e. next window is the starting point of last primitive of a upper window;If last base Member is non-linear primitive, then judges whether its length is more than default most short primitive length value, if the next window more than if Starting point be a upper window end point, the data corresponding to last primitive are otherwise included into next window, i.e., under The starting point of one window is the starting point of last primitive of a upper window;If last primitive of upper window institute is right The data answered are included into next window, then length of window=last primitive length+acquiescence sliding window size;Otherwise, window Mouth length=acquiescence sliding window size;
When it is determined that after starting point and length of window, sliding window determines;
Linear basis element length critical value is more than most short primitive length value.
We are by analyzing last primitive PNInformation, including primitive length, the type of primitive and rising for primitive Initial point, to determine the starting point of next sliding window.Obviously, if primitive PNLength is too small, less than most short base as defined in us First length value (minNLP_len), then may be too short due to data length, data message deficiency, causes true primitive to be divided, So that there is larger error, therefore the starting point using the starting point of the primitive as next sliding window, in next slip It is fitted again with new data again in window;And if last primitive is linear basis element (A, C or F) and length Less than linear basis element length critical value (maxCFlen), because linear basis element is more sensitive for length, a longer primitive Part may all be synthesized to linear basis element, it is also possible to bring larger error, therefore also it be grouped into next window Carry out again new fitting;Otherwise the starting point of next window is then normally counted since the end point of a upper window Calculate, then constantly promote window to repeat said process until whole fault sample data are converted into primitive sequence.Wherein pass through Initial parameter value is set, adjusting parameter value can change efficiency and the degree of accuracy of the algorithm.Concrete principle is shown in the flow of accompanying drawing 5 Figure.
Step 4: the data in fixed sliding window are converted into Sequence of Primitive Elements;
Step 5: repeat step three and step 4 are until the data of fault data section are all converted to Sequence of Primitive Elements, by institute State Sequence of Primitive Elements and carry out fuzzy logic matching with multiple fault signature Sequence of Primitive Elements in rule-based knowledge base, draw similarity respectively SI;
Sensing data is converted into after Sequence of Primitive Elements using above-mentioned trend abstraction, recognizer, it is necessary to by sensor Real-time tendency matches with the characteristic trend under system normal condition in knowledge base, and whether detecting system breaks down;If detection Go out system jam, then real-time tendency and the characteristic trend under various typical faults in knowledge base match, real-time diagnosis Failure.This Fuzzy Logic Reasoning Algorithm is a Multivariable Inferential algorithm, generally comprises following three steps:Primitive matches;Trend Match somebody with somebody;Process status matches.Here due to being single sensor, therefore the first two matching step is only discussed.
Primitive matches:Trend is made up of Sequence of Primitive Elements, first has to consider the similarity between primitive.Its similarity is seen below Table:
Trend matches:Each group of sensing data is all convertible into by M base after interval halving extracts identification trend The trend of member composition, trend matching compared with knowledge base trend, obtain the similarity between them exactly by real-time tendency. The real-time tendency is made to be:
Wherein, Tr is the Sequence of Primitive Elements of whole section of data to be tested,For primitive PiSampling time;ti-1,tiRespectively Pi The starting point and end point of corresponding data interval;
Knowledge base trend is:
Wherein, Tr*For the Sequence of Primitive Elements of failure in knowledge base,For primitiveSampling time;RespectivelyIt is right The starting point and end point for the data interval answered;
Generally, M ≠ S.Need to be Tr and Tr*It is placed on same timeline tuOn, tiTime shaft tuIt is divided into R Individual section, i-th of section are [tui-1,tui].I-th of siding-to-siding block length is Δ tui=tui-tui-1, trend can be estimated in terms of two Similarity:The order of primitive;The duration of primitive.By smudge tendency matching process, Tr and Tr is calculated*It is similar Spend SI.
Step 6: such as Fig. 6, the highest similarity SI of gained in step 5 is found outmax, and it is critical with default decision-making Value is compared, if SImaxMore than decision-making critical value, then the fault type of fault data section is SImaxCorresponding fault type; Then by it in rule-based knowledge base corresponding fault type return-to-operator, and provide response fault type corresponding to behavior police Accuse and prompt;If SImaxLess than or equal to decision-making critical value, then the fault type for assert fault data section is new failure;Renewal rule Then knowledge base, and add its corresponding DTC in rule-based knowledge base after being diagnosed by handbook.
In an experiment, we are by have selected 5 groups of more obvious failure samples in the emulation data in TEP chemical processes Notebook data, an and small-sized rule-based knowledge base is established with this, below we by by the real-time tendency of each sample with Feature primitive in rule-based knowledge base is compared with this, to verify the fault diagnosis rate of the algorithm.
Fault diagnosis result is as shown in the table:
As can be seen from the table, for each above-mentioned failure, can be obtained in corresponding similarity mode higher SI, and the similarity mode result between different faults is significantly smaller.
In summary, presently preferred embodiments of the present invention is these are only, is not intended to limit the scope of the present invention. Within the spirit and principles of the invention, any modification, equivalent substitution and improvements made etc., it should be included in the present invention's Within protection domain.

Claims (2)

1. the fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm, it is characterised in that including following Step:
It is extracted as Step 1: the sensing data to be detected to one section starts with interval halving algorithm from data starting point Primitive sequence, the primitive sequence are the set of primitive, and what the primitive represented is the trend of data segment, by a data segment The value obtained after one, second order derivation can obtain most seven kinds of different trend, be designated as A~G respectively, and judge its primitive Sequence whether the set for being one or more primitives " A ", if it is not, then from primitive sequence be " A " data segment original position Start to be determined as that point occurs for failure, point occurs up to the end point of data to be tested section is fault data section from failure;If so, then Carry out the detection of next section of data to be tested;
Step 2: since occurring point for failure, with the 1/10 of data to be tested length sliding window size by default, to first Data in individual sliding window section, which carry out offline interval halving fitting, makes the data be converted to polynomial fitting sequence, according to Polynomial sequence is converted to Sequence of Primitive Elements by polynomial derived indice;
Step 3: determine the starting point and length of window of next sliding window:Judge a upper sliding window last Whether primitive is linear basis element, if then judging whether its length is more than default linear primitive length critical value, if more than if The starting point of next window is the end point of a upper window, is otherwise included into the data corresponding to last primitive next Individual window;If last primitive is non-linear primitive, judge whether its length is more than default most short primitive length value, if Starting point more than then next window is the end point of a upper window, otherwise returns the data corresponding to last primitive Enter next window;If the data corresponding to last primitive of a upper window are included into next window, length of window= Last primitive length+acquiescence sliding window size;Otherwise, length of window=acquiescence sliding window size;
Step 4: the data in fixed sliding window are converted into Sequence of Primitive Elements;
Step 5: repeat step three and step 4 are until the data of fault data section are all converted to Sequence of Primitive Elements, by the base Metasequence carries out fuzzy logic matching with multiple fault signature Sequence of Primitive Elements in rule-based knowledge base, draws respectively and each failure Multiple similarity SI of feature primitive sequences match;
Step 6: find out the highest similarity SI of gained in step 5max, and by it compared with default decision-making critical value, If SImaxMore than decision-making critical value, then the fault type of data to be tested is SImaxCorresponding fault type;If SImaxIt is less than Equal to decision-making critical value, then the fault type for assert data to be tested is new failure.
2. the fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm as claimed in claim 1, its It is characterised by, linear primitive length critical value is more than most short primitive length value.
CN201610392752.6A 2016-06-06 2016-06-06 Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm Active CN106096634B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610392752.6A CN106096634B (en) 2016-06-06 2016-06-06 Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610392752.6A CN106096634B (en) 2016-06-06 2016-06-06 Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm

Publications (2)

Publication Number Publication Date
CN106096634A CN106096634A (en) 2016-11-09
CN106096634B true CN106096634B (en) 2018-04-06

Family

ID=57447303

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610392752.6A Active CN106096634B (en) 2016-06-06 2016-06-06 Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm

Country Status (1)

Country Link
CN (1) CN106096634B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107451004B (en) * 2017-07-01 2020-07-31 南京理工大学 Turnout fault diagnosis method based on qualitative trend analysis
CN112113595B (en) * 2020-09-25 2021-03-30 西门子交通技术(北京)有限公司 Sensor fault detection method, device and computer readable medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105631596A (en) * 2015-12-29 2016-06-01 山东鲁能软件技术有限公司 Equipment fault diagnosis method based on multidimensional segmentation fitting

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7580812B2 (en) * 2004-01-28 2009-08-25 Honeywell International Inc. Trending system and method using window filtering

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105631596A (en) * 2015-12-29 2016-06-01 山东鲁能软件技术有限公司 Equipment fault diagnosis method based on multidimensional segmentation fitting

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Constructive analysis of periodic solutions with interval halving;A Rontó等;《Boundary Value Problems》;20131231;第1卷(第57期);第1-34页 *
基于数据特征及模糊逻辑的光电编码器故障诊断;邓方 等;《中南大学学报(自然科学版)》;20110805;第42卷;第876-882页 *

Also Published As

Publication number Publication date
CN106096634A (en) 2016-11-09

Similar Documents

Publication Publication Date Title
CN100489870C (en) Method and multidimensional system for statistical process control
EP2930578B1 (en) Failure cause classification apparatus
CN104712542B (en) A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults
CN104915568A (en) Satellite telemetry data abnormity detection method based on DTW
CN106096634B (en) Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm
CN106404441B (en) A kind of failure modes diagnostic method based on non-linear similarity index
CN108875772B (en) Fault classification model and method based on stacked sparse Gaussian Bernoulli limited Boltzmann machine and reinforcement learning
CN108875771B (en) Fault classification model and method based on sparse Gaussian Bernoulli limited Boltzmann machine and recurrent neural network
CN104471542A (en) Discriminative hidden kalman filters for classification of streaming sensor data in condition monitoring
CN105678332A (en) Converter steel-making endpoint determination method and system based on flame image CNN recognizing and modeling process
Zhu et al. Two-dimensional contribution map for fault identification [focus on education]
CN110867196A (en) Machine equipment state monitoring system based on deep learning and voice recognition
CN104596780A (en) Diagnosis method for sensor faults of motor train unit braking system
CN109298633A (en) Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization
CN109983482A (en) Learning model generation method, learning model generating means, signal data method of discrimination, signal data discriminating gear and signal data discriminating program
CN106250937B (en) A kind of failure modes diagnostic method based on non-index of similarity
US20200012257A1 (en) Diagnostic Device and Method for Monitoring a Technical Plan
CN104635146B (en) Analog circuit fault diagnosis method based on random sinusoidal signal test and HMM (Hidden Markov Model)
CN107067034B (en) Method and system for rapidly identifying infrared spectrum data classification
CN103411913A (en) Infrared spectroscopy PLS (Partial Least Square) modeling method based on genetic algorithm self-adaptive selection segment point
CN110320802B (en) Complex system signal time sequence identification method based on data visualization
CN107491058A (en) A kind of industrial control system sequence attack detection method and equipment
CN112087443B (en) Sensing data anomaly detection method under physical attack of industrial sensing network information
EP3686812A1 (en) System and method for context-based training of a machine learning model
US20200143292A1 (en) Signature enhancement for deviation measurement-based classification of a detected anomaly in an industrial asset

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
COR Change of bibliographic data
CB03 Change of inventor or designer information

Inventor after: Gu Xiaodan

Inventor after: Deng Fang

Inventor after: Liu Chang

Inventor after: Sun Jian

Inventor after: Chen Jie

Inventor before: Deng Fang

Inventor before: Liu Chang

Inventor before: Gu Xiaodan

Inventor before: Sun Jian

Inventor before: Chen Jie

GR01 Patent grant
GR01 Patent grant