CN104781741B - Operation monitors diagnostic device and operation monitoring diagnostic program - Google Patents

Operation monitors diagnostic device and operation monitoring diagnostic program Download PDF

Info

Publication number
CN104781741B
CN104781741B CN201380058447.8A CN201380058447A CN104781741B CN 104781741 B CN104781741 B CN 104781741B CN 201380058447 A CN201380058447 A CN 201380058447A CN 104781741 B CN104781741 B CN 104781741B
Authority
CN
China
Prior art keywords
mentioned
data
definition
contribution amount
daily mode
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
CN201380058447.8A
Other languages
Chinese (zh)
Other versions
CN104781741A (en
Inventor
山中理
平冈由纪夫
吉泽直人
杉野寿治
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.)
Toshiba Corp
Original Assignee
Toshiba Corp
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 Toshiba Corp filed Critical Toshiba Corp
Publication of CN104781741A publication Critical patent/CN104781741A/en
Application granted granted Critical
Publication of CN104781741B publication Critical patent/CN104781741B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
    • G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
    • G05B23/0227—Qualitative history assessment, whereby the type of data acted upon, e.g. waveforms, images or patterns, is not relevant, e.g. rule based assessment; if-then decisions
    • G05B23/0235—Qualitative history assessment, whereby the type of data acted upon, e.g. waveforms, images or patterns, is not relevant, e.g. rule based assessment; if-then decisions based on a comparison with predetermined threshold or range, e.g. "classical methods", carried out during normal operation; threshold adaptation or choice; when or how to compare with the threshold
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
    • G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

Operation monitoring diagnostic device possesses：Division；Model construction portion, with identification part and abnormality detection data definition portion；And monitoring diagnostics division, with extraction unit and exception level judging part.Division is classified based on attribute information to operation data or time series data.The time series data that identification part is based on being classified makes daily mode data and non-daily mode data.Abnormality detection data definition portion is based on non-daily mode data and makes abnormity diagnosis model.Extraction unit extracts deviation degree of the operation data relative to daily mode data being classified.Whether deviation degree is applied to abnormity diagnosis model to calculate abnormality detection data and threshold value by exception level judging part, and exceedes threshold decision exception level with data based on abnormality detection.

Description

Operation monitors diagnostic device and operation monitoring diagnostic program
Technical field
Embodiments of the present invention are related to operation to monitor diagnostic device and the work used in operation monitoring diagnostic device Sequence monitor diagnostic program, the operation monitor diagnostic device to lower water process operation, drainage sunk well operation, sludge digestion operation, The exception produced in the working procedure systemses such as water purification operation, water supply and distribution operation, chemical process and iron steel operation is diagnosed.
Background technology
The water process such as lower water process operation, sludge digestion operation, water purification operation and water supply and distribution operation/water transport operation, In the factories of working procedure systemses such as petrochemistry operation, iron steel operation and semiconductor fabrication sequence, it is provided with to multiple working procedure states The multiple on-line sensors being measured.Operation monitoring arrangement (SCADA:Supervisory Control And Data Acquisition) will generally pass through obtained from the measurement of set sensor group in working procedure systemses, such as flow, temperature, The operation data such as water quality and/or operational ton, are converted to such as time series data such as trend chart.Plant manager (keeper) and/ Or operating personnel (operating personnel) is monitored by time series data, the state of operation is thus grasped, carry out the operating of operation Change, control.
The time series data of each operation data is generally positioned the upper lower limit value for being referred to as managing the limit etc..Operation is monitored Device time series data value exceed the management limit in the case of or less than the management limit in the case of, alarm of transmitting messages. Plant manager and/or operating personnel carry out the confirmation of factory's utilization and examine closely again based on the alarm.Based on this alarm hair The running management of report is the basic of factory's utilization.
More advanced plant operation management in, do not require nothing more than simple operation it is unstable when reply, also require reaching The cost-effective utilization of energy-conservation is realized on the basis of to the regulation target capabilities of operation.Herein, it is stipulated that target refers to, for example, If lower water process operation, then corresponding to discharge water quality limitation in accordance with etc..Additionally, if clean water treatment operation, then only Concentration of residual chlorine in water for the situation below set upper limit and/or, in the absence of the micro- life of pathogenicity with Cryptosporidium as representative Situation of thing etc. corresponds to regulation target.Additionally, if chemical process and iron steel operation, then by product (petroleum refining product and Iron steel) quality (such as purity and intensity etc.) be maintained prescribed limit situation correspond to regulation target.When wanting to use work When the running management of factory is to cause to realize that energy-conservation is cost-effective on the basis of the regulation target capabilities for reaching operation, emphasis exists In the state of pair operation related to target capabilities is monitored to avoid sinking into not reaching the state of regulation target, rapid inspection The state change and/or abnormality for hindering to reach regulation target are measured, and is taken some countermeasures in advance.And, it is preferred that emphasis is, will be with Target capabilities and the cost-effective related working procedure states of energy-conservation remain kilter, and detect to get married and start a new life rapidly Good state deviates such working procedure states change.
However, as the state change to operation and the abnormal method for being diagnosed, it is known that be referred to as multivariate statistics work Sequence monitors (MSPC：Multi-Variate Statistical Process Control) method, the method use in stone " the multivariate statistics analytic method " utilized always in the field of oily chemical process and Tie Gang factories.As the most Chang Li in MSPC Method, it is known that principal component analysis (PCA：Principal Component Analysis) and latent variable sciagraphy/partially most Small square law (PLS：Projection to Latent Structure/Partial Least Square).
In MSPC, using multivariate analysis such as PCA and/or PLS, mainly reach and the abnormal sign of factory is detected The first purpose and the second purpose for being speculated to the process variable as abnormal factorses.For the first purpose, by profit With the relevant information of multiple process variables, the slight abnormal sign to that cannot be detected by a variable is detected.It is right In the second purpose, the abnormality detection synthesized by multiple process variables in basis data are (for example, Q statistical magnitude and Hotelling T2Statistic) detect exception after, using the contribution for representing contribution degree of each operation data for the abnormality detection data Measure to speculate the process variable of the candidate as abnormal factorses.If in this way, using MSPC, with conventional to independent operation The monitoring that the variable uses simple management limit is carried out (in the monitoring of production line etc., is contrasted and referred to as SPC with MSPC sometimes： Statistical Process Control (statistical process control)) compare, can carry out for plant manager and/or fortune Turn more useful advanced monitoring diagnosis for personnel.
As described above, if using MSPC, can reach the abnormal sign of factory is detected the first purpose, with And the second purpose speculated to the process variable as abnormal factorses, but it is used as factory in MSPC is introduced into SCADA Monitoring system in the case of realizing, hears more following live reflection：Only by when first and second purpose is reached The information for obtaining, enough information can not be provided for user is plant manager and/or operating personnel.Its reason For, plant manager and/or operating personnel want to make it is final should take what action is such to judge in exception, but only Speculated by the detection of abnormal sign and its factor, it is difficult to taken an immediate action in exception.
Prior art literature
Patent document
Patent document 1：Japanese Unexamined Patent Publication 8-241121 publications
Patent document 2：Japanese Unexamined Patent Publication 2004-303007 publications
Patent document 3：Japanese Unexamined Patent Publication 2007-65883 publications
Non-patent literature
Non-patent literature 1：http://tech.chase-dream.com/spc.html
Non-patent literature 2：C.Rosen,“Monitoring Wastewater Treatment Systems”, Lic.Thesis,Dept.of Industrial Electrical Engineering and Automation,Lund University,Lund,Sweden,1998.
Non-patent literature 3：Mia Hubert,Peter J.Rousseeuw,Karlien V,“ROBPCA:a New Approach to Robust Principal Component Analysis(2005)”,Technometrics.
Non-patent literature 4：C Croux,A Ruiz-Gazen,“High breakdown estimators for principal components:the projection-pursuit approach revisited”,Journal of Multivariate Analysis.
Non-patent literature 5：K.-R.Muller,S.Mika,G.Ratsch,K.Tsuda,and B.Scholkopf,“An introduction to kernel-based learning algorithms”,IEEE Trans.Pattern Anal.Machine Intell.,12(2):181-201,March,2001.
Non-patent literature 6：B.Scholkopf,A.J.Smola,and K.-R.Muller,“Nonlinear component analysis as a kernel eigenvalue problem”,Neural Computation,10(5): 1299-1319,1998.
The content of the invention
The invention problem to be solved
As described above, in the conventional device for being introduced into MSPC, can carry out the abnormal sign of factory detection and its Factor speculates, but only by these information, plant manager and/or operating personnel are difficult to judge what should take in exception Action.
Therefore, it is an object of the present invention to provide operation monitoring diagnostic device and the operation for using in the apparatus are monitored Diagnostic program, operation monitoring diagnostic device can be provided plant manager and/or operating personnel and should taken in exception The judgement policy of what action.
Means for solving the problems
According to implementation method, a kind of operation monitors diagnostic device, obtain by the sensor measurement set in object operation to Operation data, possess：Division；Model construction portion, with identification part and abnormality detection data definition portion；And monitoring is examined Disconnected portion, with extraction unit and exception level judging part.Above-mentioned division keeps the environment according to residing for above-mentioned object operation and divides The attribute information of class, based on above-mentioned attribute information to above-mentioned operation data and by past during multiple set in advance Multiple time series datas that operation data are constituted are classified.Above-mentioned identification part is classified as multiple classifications by robustly speculating The above-mentioned respective typical value of time series data makes the daily mode data of each above-mentioned classification, and based on each above-mentioned classification Daily mode data makes non-daily mode data.Above-mentioned abnormality detection is based on the non-daily pattern of above-mentioned multiple with data definition portion Each of data, makes the first definition for calculating abnormality detection data and is used for calculating above-mentioned abnormality detection Second definition of the threshold value of data.The above-mentioned operation data being classified are extracted relative to above-mentioned multiple day norms in said extracted portion The deviation degree of formula data.Above-mentioned exception level judging part is respectively applied to above-mentioned first definition by by above-mentioned multiple deviation degrees To calculate abnormality detection data according to each above-mentioned classification, determined by the way that above-mentioned multiple deviation degrees are respectively applied into above-mentioned second Adopted formula to calculate threshold value according to each above-mentioned classification, and whether above-mentioned abnormality detection data are judged by according to each above-mentioned classification Judge exception level more than above-mentioned threshold value.
Brief description of the drawings
Fig. 1 is the functional structure for representing the monitoring system that the operation monitoring diagnostic device of first embodiment possesses Figure.
Fig. 2 is to represent that the division shown in Fig. 1 carries out the figure of sorted time series data.
Fig. 3 is the flow chart when monitoring diagnostics division for representing shown in Fig. 1 is judged the exception level of operation data Figure.
Fig. 4 is the figure of the judged result of the exception level for representing the exception level judging part shown in Fig. 1.
Fig. 5 is to represent the contribution amount calculated based on element paritng portion as shown in Figure 1 and the contribution amount curve for making Figure.
Fig. 6 is to represent the contribution amount calculated based on element paritng portion as shown in Figure 1 and the contribution amount curve for making Figure.
Fig. 7 is the block diagram of the functional structure of the operation monitoring diagnostic device for representing second embodiment.
Fig. 8 is the figure for representing the matrix that the normal data register shown in Fig. 7 is registered.
Specific embodiment
Hereinafter, it is explained with reference to implementation method.
(first embodiment)
Fig. 1 is the functional structure for representing the monitoring system that the operation monitoring diagnostic device 30 of first embodiment possesses The structure chart of example.In monitoring system shown in Fig. 1, in order that implementation idea is clearer and more definite, by for the purpose of except denitrification and phosphorus Lower water height treatment process is used as supervision object.However, the supervision object of monitoring system is not limited to lower water height treatment process.
Monitoring system shown in Fig. 1 possesses lower water height treatment process 10, Data Collection storage unit 20, operation monitoring and examines Disconnected device 30 and user interface 40.
Lower water height treatment process 10 possesses initial sedimentation basin 101, anaerobism groove 102, anaerobic groove 103, aerobic groove 104 and most Whole sedimentation basin 105.
Additionally, lower water height treatment process 10 possesses draw-off pump 1111, air blower 1121, circulating pump as actuator 1131st, loopback sludge pump 1141 and draw-off pump 1151.Draw-off pump 1111 extracts the excess sludge of initial sedimentation basin 101.Air blower 1121 supply oxygen to aerobic groove 104.Circulating pump 1131 makes the water from the output of aerobic groove 104 be circulated to anaerobic groove 103.Loopback sludge Pump 1141 is by the excess sludge of final sedimentation tank 105 to the loopback of anaerobism groove 102.Draw-off pump 1151 extracts final sedimentation tank 105 Excess sludge.Various actuators 1111~1151 are acted with the cycle for specifying.
Additionally, actuator 1111~1151 possesses operation quantity sensor respectively.That is, draw-off pump 1111 possesses extraction flow and passes Sensor 1112, air blower 1121 possesses supply air flow sensor 1122, and circulating pump 1131 possesses circular flow sensor 1132, loopback sludge pump 1141 possesses loop back traffic sensor 1142, and draw-off pump 1151 possesses extraction flow sensor 1152.
The flow for extracting 1112 pairs of sludge extracted by draw-off pump 1111 of flow sensor is measured.Supply air mass flow The flow of 1122 pairs of oxygen supplied by air blower 1121 of sensor is measured.Circular flow sensor 1132 pairs is by circulating pump The flow of the water of 1131 circulations is measured.The flow of the 1142 pairs of sludge by loopback sludge pump loopback of loop back traffic sensor enters Row measurement.The flow for extracting 1152 pairs of sludge extracted by draw-off pump 1151 of flow sensor is measured.Various operational tons are passed Sensor 1112~1152 is measured with the cycle for specifying to operation data.
Additionally, in lower water height treatment process 10, possessing as operation sensor：Rain sensor 121；To under inflow The lower water that water is measured flows into quantity sensor 122；The inflow TN measured to flowing into the total nitrogen content contained by lower water is passed Sensor 123；To flowing into the inflow TP sensors 124 that the full phosphorus amount contained by lower water is measured；To flowing into contained by lower water Inflow UV sensors 1251 or flow into COD sensors 1252 that organic matter amount is measured；ORP to anaerobism groove 102 is carried out The anaerobism groove ORP sensors 126 of measurement；The anaerobism groove pH sensors 127 measured to the pH of anaerobism groove 102；To anaerobic groove The anaerobic groove ORP sensors 128 that 103 ORP is measured；The anaerobic groove pH sensors measured to the pH of anaerobic groove 103 129；The phosphoric acid sensor 1210 measured to the phosphoric acid concentration of aerobic groove 104；Oxyty to aerobic groove 104 is counted The DO sensors 1211 of survey；The ammoniacal sensor 1212 measured to the ammonia density of aerobic groove 104；In anaerobism groove 102, anaerobic groove 103 and aerobic groove 104 at least one groove in the MLSS sensors 1213 that are measured to active mud content；Anaerobism groove 102, The cooling-water temperature sensor 1214 measured to water temperature at least one groove of anaerobic groove 103 and aerobic groove 104；To being precipitated from final The excess sludge SS sensors 1215 that the solid concentration of the sludge quantity that pond 105 is extracted is measured；To from final sedimentation tank 105 The discharge SS sensors 1216 that the SS concentration of the discharge water of discharge is measured；To the sludge interface level of final sedimentation tank 105 The sludge interface sensor 1217 for being measured；Quantity sensor 1218 is discharged to the lower water that the lower water of discharge is measured；To row Put down the discharge TN sensors 1219 that the total nitrogen content contained by water is measured；Full phosphorus amount contained by the lower water of discharge is counted The discharge TP sensors 1220 of survey；And to discharging the discharge UV sensors that the organic matter amount contained by lower water is measured 12211 or discharge COD sensors 12212.Various operation sensors 121~12212 are entered with the cycle for specifying to operation data Row measurement.Additionally, figure 1 illustrates following situation：Be provided with inflow UV sensors 1251 and flow into COD sensors 1252 it In inflow UV sensors 1251, and be provided with the discharge among discharge UV sensors 12211 and discharge COD sensors 12212 UV sensors 12211.
Data Collection storage unit 20 is collected and obtained with the cycles for specifying from various operation quantity sensors 1112~1152 Operation data and the operation data obtained with the cycle for specifying from various operation sensors 121~12212.Data Collection The operation data that storage unit 20 will be collected into are exported to operation monitoring diagnostic device 30.Additionally, Data Collection storage unit 20 will The operation data being collected into are converted to time series data and are preserved according to form set in advance.Data Collection storage unit 20 According to the request that diagnostic device 30 is monitored from operation, the time series data that will be preserved is exported to operation monitoring diagnostic device 30.
Operation monitoring diagnostic device 30 for example includes CPU (Central Processing Unit) and ROM (Read Only Memory) and RAM (Random Access Memory) etc. store the storage of program or data that treatment is performed for CPU Deposit region etc..Operation monitors diagnostic device 30 by making CPU configuration processors, be achieved in division 31, model construction portion 32 and Monitor the function of diagnostics division 33.
Division 31 possesses attribute supply unit 311 and determined property portion 312.
Month information, season (spring (March~May), summer (June~8 when attribute supply unit 311 keeps operation data to be collected Month), the autumn (September~November), the winter (December~2 month)) information, climatic information, temperature or water temperature information and operation mode letter The attribute informations such as breath.
Determined property portion 312 to data Collection and conservation portion 20 ask preset during past time series data. The attribute information that determined property portion 312 is kept using attribute supply unit 311, to when Data Collection storage unit 20 is supplied Ordinal number evidence is classified.Attribute information is the relevant letter of the attribute being easily classified with the external condition according to residing for operation Breath, attribute for example has month, season, weather, temperature, water temperature and operation mode etc..For example, determined property portion 312 is based on category Property information, by time series data be categorized as each weather such as data monthly, the data per the four seasons, fine, cloudy, rain or snow data, By temperature or water temperature etc. be divided into multiple grades each temperature or water temperature data and operation mode it is dispar every Data of individual operating etc..Determined property portion 312 would be classified as time series data of all categories and be exported to model construction portion 32.
Additionally, when by monitoring that diagnostics division 33 carries out the diagnosis of the operation data at current time (online), determined property The attribute information that portion 312 is kept using attribute supply unit 311, it is online by what is be collected into by Data Collection storage unit 20 Operation data are classified as described above.Determined property portion 312 would be classified as operation data of all categories and be diagnosed to monitoring Portion 33 exports.
Model construction portion 32 possesses identification part 321, abnormality detection data definition portion 322 and contribution amount definition portion 323.Mould Type structure portion 32 makes multiple daily mode datas and respectively originally making based on these daily mode datas by identification part 321 Make non-daily mode data.Additionally, model construction portion 32 is by abnormality detection data definition portion 322, based on non-daily pattern Whether data are to as producing abnormality detection data of abnormal index etc. to be defined.Additionally, model construction portion 32 passes through Contribution amount definition portion 323, is defined to contribution amount, and the contribution amount is represented carries out table to operation quantity sensor and operation sensor The process variable shown makes much contributions to produced exception.
Identification part 321 is based on being categorized as time series data of all categories by division 31, makes daily mode data and base In the non-daily mode data of the daily mode data.Daily mode data corresponds to the object for wanting confirmation trend, according to The relevant definition of calendar and made multiple, the phase of such as time series data in the same time, the daily mode data of this day unit, and/ Or, identical what day (is for example all Monday, or is all Saturday etc.) of time series data and mutually in the same time, i.e. what day list Daily mode data of position etc..
For example, in the case where the daily mode data of day unit is made, identification part 321 is based on climatic information and week On the basis of the information that several information etc. in advance become process actions in the case of exception is removed, based on the time series data for receiving Phase measured value in the same time, the typical value that calculating is robustly deduced at each moment.Here, the typical value for robustly deducing It is the average value of phase measured value in the same time, prunes average value, intermediate value and mode value etc..May be included in time series data different In the case of constant value, not using average value as typical value, and using intermediate value etc., the value relatively low relative to exceptional value sensitivity as generation Tabular value.Additionally, in the case where the daily mode data of what day unit is made, identification part 321 is identical for time series data Moment and identical what day measured value carries out treatment similar to the above.Identification part 321 can continue to carry out calculating generation Tabular value and make the treatment of daily mode data.
Additionally, the method for robustly speculating typical value, it is also known that other are several.If for example, using HL (Hodges- Lehmann) the method such as supposition, MCD (Minimum Covariance Determinant), bootstrap or subsample method, i.e., Make in the case where more complexly abnormal data has been mixed into, it is also possible to deduce typical typical value.
Identification part 321 calculate time series data among making daily mode data when used measured value, relative to system Which kind of degree is value in the daily mode data made deviate from, using result of calculation as non-daily mode data.
The quantity of the daily mode data for defining is set to m by identification part 321, in the classification classified by division 31 In the case that quantity is l, m × l daily mode data and m × l non-daily mode data are made by above-mentioned treatment.
Additionally, the making of the daily mode data carried out based on time series data and non-daily mode data, such as also can Implemented using the digital filter such as low pass filter or discrete wavelet conversion.If however, hypothesis be mixed into abnormal data, Robustly speculated using known various methods in Robust Statistics if typical value, reliability day norm higher can be made Formula.
Abnormality detection data definition portion 322 receives the non-daily mode data produced by identification part 321.Now, it is non- Daily mode data is individual with the l classification and the definition number m of daily mode data classified by division 31 accordingly as l × m data collection and be provided.
Additionally, in the following description, by it is contained by the data set of non-daily mode data, for during presetting Past time series data away from value, be recited as ＜ Xk ＞, wherein, k=1,2 ..., l × m.＜ Xk ＞ are in the row direction Represent it is relative with the non-day regular data of the operation data by operating quantity sensor and operation sensor to measure away from value, arranging The matrix of sequential is represented on direction.That is, ＜ Xk ＞ are for r rows, key element ＜ Xk ＞ (r, s) of s row, store and basis The relative matrix away from value of non-day regular data that r-th sensor measured value of moment s is calculated.If quantity sensor will be operated And the quantity of operation sensor is set to n, the data number of sequential is set into Q, then ＜ Xk ＞ include Q × n data.Additionally, with Down, it is necessary to ＜ Xk ＞, k=1,2 ..., the abnormality detection of l × m is defined with data.But, only difference is that, institute is defeated The non-daily mode data for entering is from different classes of offer.The definition method of abnormality detection data is one, as long as therefore The worry not confused, then be simply recited as ＜ X ＞ by ＜ Xk ＞, omits subscript k.
Abnormality detection data definition portion 322 is by various sides such as ＜ X ＞ applications multivariate analysis or rote learnings Method generates abnormality detection data.It is the method being commonly used as operation diagnostic techniques to the method for ＜ X ＞ applications, Referred to as MSPC (process management of multivariate statistics).In MSPC, projected using usual principal component analysis (PCA) or latent variable Method (PLS) is referred to as the T of Q statistical magnitude and Hotelling to generate2The abnormality detection data of statistic.
Hereinafter, the specific calculating formula generated using PCA in the case of abnormality detection data is recorded.If using PCA, ＜ X ＞ are decomposed as described below.Additionally, in (1) formula, ＜ X ＞ are recorded with the X of thick word.
[number 1]
Here, ＜ Ta＞ (the thick word T in (1) formulaa)∈Rq×n, it is the quilt being made up of q sample and n number of principal components The referred to as matrix of score matrix.＜ Pa＞ (the thick word P in (1) formulaa)∈Rn×n, be represent n constitute variable and n it is main into / relation be referred to as the matrix of loading matrix.＜ T ＞ (the thick word T in (1) formula) ∈ Rq×p, it is by p ＜ ＜ n Individual principal component is come the ＜ T that interceptaThe part matrix of ＞, commonly known as score matrix.Equally, ＜ P ＞ (the thick texts in (1) formula Word P) ∈ Rn×p, it is to represent the principal component by p ＜ ＜ n interceptions relative to the n ＜ P of the relation of variableaThe part matrix of ＞ (also referred to as submatrix), commonly known as loading matrix.Additionally, ＜ E ＞ (the thick word E in (1) formula) ∈ Rq×n, it is by q sample The error matrix that this and n variable are constituted, expression has intercepted the error in the case of principal component by p ＜ ＜ n.
Hereinafter, by ＜ Ta＞ and ＜ T ＞, ＜ Pa＞ and ＜ P ＞ are clearly distinguished, by ＜ Ta＞ and ＜ Pa＞ is referred to as Score matrix and loading matrix, main score matrix and Main Load matrix are referred to as by ＜ T ＞ and ＜ P ＞.
Using these matrixes, following Q statistical magnitude and T is defined as abnormality detection data2Statistic.
[number 2]
Q (x (t))=xT(t)(I-PPT)x(T) (2)
[number 3]
T2(x (t))=xT(t)PT∧-1Px(t) (3)
Here, ＜ Λ ＞ (the thick word Λ in (3) formula) are the squares as diagonal key element with the dispersion of principal component Battle array, is meant that carry out normalization to dispersion.Additionally, ＜ I ＞ (the thick word I in (2) formula) are the unit matrixs of appropriate size. Additionally, x (t) is t-th key element of matrix L T.LT.LT X ＞.Additionally, when monitoring that diagnostics division 33 carries out anomaly monitoring diagnosis, to the x T () substitutes into operation data being calculated away from value relative to daily mode data of online measurement.
Abnormality detection data definition portion 322 is set as the Q statistical magnitude and T defined by (2), (3) formula2Statistic To recognize the threshold value of abnormal and normal judgment standard.Threshold value is significantly related to the detection of state change and/or abnormal sign, Therefore its establishing method is more important.However, due to unrelated with the technological thought of present embodiment, therefore only record typical setting Method.
Assuming that in the case of there is no any prior information relative to the past time series data during presetting, as The setting method of acquiescence, can use statistics confidence limit angle value and the T with Hotelling related to Q statistical magnitude2Statistic Related statistics confidence limit angle value (with reference to non-patent literature 2).Statistics confidence limit angle value, the Yi Jiyu related to Q statistical magnitude The T of Hotelling2The related statistics confidence limit angle value of statistic, can write as described below.
[number 4]
h0：=1-2 Θ1Θ3/3Θ2 2
Θi：=λp+1 i+λp+2 i+.........+λn i
Here, p is the quantity of the variable of residual in model.caIt is that the limit of confidential interval is the standard in the case of 1-a Skew (the example of the standard deviation of regular distribution：In the case of a=0.01, ca=2.53, in the case of a=0.05, ca= 1.96).Additionally, λiIt is diagonal key element (that is, the Θ of ＜ Λ ＞iBe each composition contained by error term i powers and.).
[number 5]
T2Limit=p (q-1)/(q-p) F (p, q-p, a) (5)
Here, p is the quantity of the variable of residual in model, q is the quantity of entire variable.(p, q-p are a) that the free degree is to F (p, q-p), the F being set to cofidence limit in the case of a are distributed.Additionally, the situation for being set to a=0.01 or 0.05 is more.
Abnormality detection defines (1)~(5) with data definition portion 322 according to each of l × m non-daily mode datas Formula.
Additionally, in the case where being assumed in data containing a large amount of outliers etc., abnormality detection data definition portion 322 Non-patent literature 3 and the various robusts for considering the robustness for outlier of the grade of non-patent literature 4 can also for example be used PCA algorithms.Or, it is also possible to it is extended and is used as robust PLS.
Also, in that case of being assumed to that there is stronger non-linear correlation between data, abnormality detection data Definition portion 322 for example can also consider non-thread using core PCA described in non-patent literature 5 and the grade of non-patent literature 6 etc. The PCA of shape.Or, it is also possible to it is extended and is used as core PLS.
Also, in the case where there are the both sides of non-linear and outlier, abnormality detection data definition portion 322 can also use the method for being combined with robust PCA and core PCA.
Additionally, as the technology similar with MSPC, it is also possible to using field mouthful method used in the field of quality engineering etc., make Abnormality detection data are generated with mahalanobis distance.
Contribution amount definition portion 323 sets the process variable, relative being indicated to operation quantity sensor and operation sensor In the definition of the contribution amount of the abnormality detection data defined by (2) formula and (3) formula.The definition method of contribution amount there is also many It is individual, but can for example be defined as described below.
[number 6]
[number 7]
T2Cont (n, t)=xT(t)PT∧-1P(：, n) x (t, n) (7)
Here, n is meant that n-th process variable, t is to represent variable sometime.(6) shown in formula and (7) formula Definition, defines according to l × m non-daily mode datas each.By using (6) formula and (7) formula, contribution amount definition portion 323 can to monitoring diagnostics division 33 provide can calculation process variable which kind of degree is made to the value of abnormality detection data respectively The definition of contribution.
If as long as additionally, contribution amount definition portion 323 is based on the process number that is supplied to being transfused to by monitoring diagnostics division 33 The possibility of the factor that then which process variable can turn into exception to according to the abnormality detection data for calculating is high to be ranked up And the framework for exporting, can be arbitrary framework.
As described above, when model construction portion 32 is on the classification and daily mode data that make the moon or season etc. Definition, (1)~(7) formula is defined respectively, thus build for the deviation degree to online operation data and daily mode data The multiple abnormity diagnosis models evaluated.
Monitoring diagnostics division 33 possesses extraction unit 331, exception level judging part 332, element paritng portion 333 and anomalous identification portion 334.Monitoring diagnostics division 33 is extracted by the non-daily data pattern data in 331 pairs of online operation data of extraction unit. Additionally, monitoring diagnostics division 33 uses what is calculated based on the non-daily data pattern extracted by exception level judging part 332 Abnormality detection data, judge produced abnormal exception level.Additionally, monitoring diagnostics division 33 passes through element paritng Portion 333, calculates produced abnormal contribution amount each operation variable.Additionally, monitoring diagnostics division 33 is known by abnormal Other portion 334, recognized based on the contribution amount for calculating produced by exception be by the different of operation quantity sensor and operation sensor It is often causing or being caused by the exception of operation.
Extraction unit 331 receives to be categorized as online operation data of all categories by division 31.Extraction unit 331 is to model structure Build portion 32 and ask the daily mode data relevant with the classification of the operation data for being received.Extraction unit 331 is obtained according to request The difference of the daily mode data provided from identification part 321 and the operation data from the supply of division 31, thus abstraction process number Non- daily mode data in.
The non-daily mode data that exception level judging part 332 will be extracted by extraction unit 331 is substituted into be used by abnormality detection L × m (2) formula and (3) formula of the definition of data definition portion 322, calculate Q statistical magnitude and T2Statistic.
Exception level judging part 332 judges Q statistical magnitude and the T calculated for multiple abnormity diagnosis models2Statistic is It is no more than the threshold value calculated based on (4) formula and (5) formula in each abnormity diagnosis model.Multiple abnormity diagnosis models are adapted for The abnormity diagnosis model evaluated the exception level of online operation data, is preset.For example, these abnormity diagnosis Model belongs to the adjacent classification among the classification of identical type, is made based on the daily mode data of identical definition.Additionally, The acquirement situation of online operation data, is to belong to mutually similar with some the abnormity diagnosis model in multiple abnormity diagnosis models Other situation.
Exception level judging part 332 is based on Q statistical magnitude and T2Statistic determines to represent online with the comparative result of threshold value Operation data in whether generate abnormal exception level.For example, exception level judging part 332 first determines whether to be directed to and operation Q statistical magnitude and T that adaptability highest abnormity diagnosis model between data is calculated2Whether statistic exceedes is directed to the exception The threshold value that diagnostic model is calculated.Grade of fit between abnormity diagnosis model is predefined, adaptability highest abnormity diagnosis model It for example refer to the abnormity diagnosis model for belonging to identical category with the acquirement situation of operation data.In Q statistical magnitude and T2Statistic is small In the case of threshold value, exception level judging part 332 is judged as that operation data are normal, and exception level is 0%.Additionally, here, The value of exception level is evaluated according to percentage, but is not limited to be evaluated according to percentage.
In Q statistical magnitude and T2In the case that statistic exceedes threshold value, exception level judging part 332 is judged as that operation data are It is abnormal, to operation data setting V%.Additionally, V% is the arbitrary numerical value that can be evaluated exception level.Next, different Normal grade judging part 332 judges that the Q calculated for the abnormity diagnosis model high with adaptability second between operation data unites Metering and T2Whether statistic exceedes the threshold value calculated for the abnormity diagnosis model.The abnormity diagnosis mould high of adaptability second Type, for example, refer to the abnormity diagnosis model of the classification adjacent with the classification belonging to operation data.In Q statistical magnitude and T2Statistic is small In the case of threshold value, set V% is judged as exception level by exception level judging part 332.In Q statistical magnitude and T2Statistics In the case that amount exceedes threshold value, exception level judging part 332 is judged as operation data exception, set V% is increased and is set It is V ' %.
Exception level judging part 332 is in Q statistical magnitude and T2In the case that statistic exceedes threshold value, as Q statistical magnitude and T2 Statistic is switched over less than untill threshold value to abnormity diagnosis model, and Q statistical magnitude and T is repeated2The ratio of statistic and threshold value Compared with.Additionally, exception level judging part 332 all implements Q statistical magnitude and T for multiple abnormity diagnosis models set in advance2 The comparing of statistic and threshold value, in Q statistical magnitude and T2In the case that statistic exceedes threshold value, it is judged as that operation data are 100% It is abnormal.
Element paritng portion 333 in the case where the exception level judged by exception level judging part 332 is not as 0%, i.e. work Ordinal number is counted according to for the Q in the case of exception, calculated using (6) formula and (7) formula to being calculated by exception level judging part 332 Amount and T2The contribution amount of statistic.Now, element paritng portion 333 is used for the adaptability between online operation data most The definition of abnormity diagnosis model high.Element paritng portion 333 makes contribution amount curve based on the contribution amount for calculating, and will The contribution amount curve produced is shown in user interface 40.Thus, the user of user interface 40 can speculate abnormal as producing Factor process variable.
Anomalous identification portion 334 recognizes exception based on the process variable that abnormal factor is speculated as by element paritng portion 333 Species caused by the exception of operation quantity sensor and operation sensor or caused by the exception of operation.Anomalous identification Recognition result is shown in user interface 40 by portion 334.
For example, in contribution amount curve, in the case where the data of only some process variable show prominent exceptional value, For the abnormal possibility of sensor is higher.Particularly, have under related environment to multiple operation measurement informations, at one It is that the possibility of sensor abnormality is high in the case that the data of process variable show the exceptional value for detecting.Therefore, it is abnormal to know Other portion 334 is in the case where the data of only some process variable show prominent exceptional value, thus it is speculated that be sensor abnormality.
Conversely, in the case where the contribution amount of multiple process variables shows exceptional value, be operation abnormal possibility compared with It is high.Therefore, anomalous identification portion 334 is in the case where the contribution amount of multiple process variables shows exceptional value, thus it is speculated that be operation exception. But, even if being the abnormal state of operation, there is also the state of certain sensor failure.Even if in this case, sending out The contribution amount of the process variable of raw failure shows that the situation of relatively extreme exceptional value is more, therefore is in most cases capable of identify that Go out sensor abnormality and operation exception.
As this knowledge method for distinguishing is quantitatively carried out, it can be considered that method as described below.
The data being made up of the contribution amount of each operation variable are considered as a data set by anomalous identification portion 334, carry out for The abnormity diagnosis of the data.When by contribution amount curve to speculate abnormal factorses, due to having been detected by exception, it is therefore necessary to To be wherein mixed into premised on the situation of certain abnormal data.Due to can not be by common average for contribution amount data set Or disperse (standard deviation) to calculate exception, therefore anomalous identification portion 334 utilizes and replaces average and scattered robust to speculate method. For example, can replace average, and use intermediate value and prune average.Furthermore it is possible to instead of standard deviation, and utilize intermediate value definitely inclined Difference (MAD) and the standard deviation pruned.Anomalous identification portion 334 is using the situation of intermediate value and median absolute deviation Under, calculated using the contribution amount X of certain process variable
K=(X- intermediate values)/median absolute deviation,
Value to the K sets the threshold k max of such as 5~10 degree.The process variable for exceeding the Kmax in the value of K is carried In the case of getting only one, anomalous identification portion 334 is judged as sensor abnormality, is not also extracted at one or extracts many In the case of individual, it is judged as operation exception.
Additionally, as other methods, method as following can also be used.Between contribution amount and statistic, have The summation of the contribution amount of each operation variable this property consistent with statistic.That is, if (6) formula is obtained into total for each operation variable Then turn into (2) formula, or, (3) formula is turned into if (7) formula is obtained into summation for each operation variable.Using the property, can Adopt with the following method：Maximum abnormal factorses candidate contribution amount relative to statistic ((2) formula for detecting the abnormal moment Or (3) formula) ratio exceed regulation setting value in the case of, be judged as sensor abnormality.For example, being set as setting value Fixed 0.75 (75%), in abnormality detection, contribution amount ((6) formula or (7) formula) ÷ of the maximum abnormal factorses candidate of t is examined extremely In the case that the value of the statistic ((2) formula or (3) formula) of t exceedes setting value 0.75 during survey, anomalous identification portion 334 is judged as passing Sensor exception.
In the case that the measurement information of the sensor of identical type contains multiple in the process variable, the identification can be made Precision is further improved.For example, in the case where the DO sensors 1211 of multiple respectively different principles are provided with, it is assumed that in place The different position in some different, flow directions is provided with situation of multiple DO sensors 1211 etc..Here, in only certain DO sensings The contribution amount of the measured value of device is significantly high, DO sensors beyond it contribution amount it is not high in the case of, strong doubt contribution amount DO sensor failures high.In this case, can interpolate that to be sensor abnormality.
Additionally, be once judged as sensor abnormality, then be judged as abnormal sensor be maintained and recover it is normal it Before, lasting detection is abnormal all the time.Therefore, monitoring system is difficult to proceed abnormity diagnosis after sensor recovers normally.For To should situation, anomalous identification portion 334 is for by being judged as the operation data that abnormal sensor is measured, notifying the biography Before the reparation of sensor terminates, the object moment of the subject sensor being input into the daily mode data defined by identification part 321 Data.Or, the nearest past phase in 334 pairs, anomalous identification portion in the same time, identical what day the several samples of Data Mining, Intermediate value by the sample of acquirement etc. is to by being judged as the operation data input that abnormal sensor is measured.Thus, anomalous identification Even if portion 334 also can continue to carry out abnormity diagnosis after sensor fault is diagnosed as.Even if additionally, in multiple sensors In the case of breaking down, if not phase break down in the same time, anomalous identification portion 334 be just capable of identify that sensor fault and Operation exception.
Next, the action of the operation monitoring diagnostic device 30 to constituting as described above is specifically described.
First, the situation that model construction portion 32 builds abnormity diagnosis model is illustrated.In the following description, with point The time series data supplied from Data Collection storage unit 20 is categorized as the feelings of the data in June in the determined property portion 312 in class portion 31 Illustrated as a example by condition.Fig. 2 is the figure of the example for representing the time series data that the data in June are classified.It is sorted according to Fig. 2 Time series data includes 1 hour data of unit across multiple date-time ground.For example, the moment 0：00 data are by X0=[5 46 635 111 7] this data set is constituted.
Identification part 321 makes the daily mode data of day unit based on the time series data shown in Fig. 2.For example, calculating Moment 0：In the case of 00 typical value, according to X0=[5 46635 111 7], the moment 0 is predicted as：00 typical value into It is of about the value near 5,6.If the most straightforward procedure that calculating typical value is taken to X0 is i.e. average, the average value of X0 turns into 18.375.The value is significantly different from the typical value predicted.Its reason may be considered, and the 8th data are exceptional value.Cause This, if not this simple averaging operation but calculate intermediate value, then as 5.5, close to predicted typical value.Additionally, If for example, calculating the mode value of X0, as 6, the value is also close to the typical value predicted.If additionally, repairing cuts flat (pruning Remove maximum and minimum value in averagely) and be averaged, then as 5.5, the value is also close to the typical value predicted.Identification part The 321 daily mode datas that June is made by the way that typical value is computed as described above at each moment.
Identification part 321 is by calculating the measured value for using and the daily pattern count produced when daily mode data is made Non- daily mode data is made according to away from which kind of degree.For example, being carried out to the typical value of daily mode data with intermediate value In the case of calculating, the 1st data in X0=[5 46635 111 7] and daily mode data are away from being 5-5.5 =-0.5, the 7th data and daily mode data are away from being 111-5.5=105.5.
Abnormality detection data definition portion 322 makes (2) with reference to the non-daily mode data produced by identification part 321 T shown in the definition of the Q statistical magnitude shown in formula and (3) formula2The definition of statistic.Additionally, abnormality detection is fixed with data Adopted portion 322 calculates the statistics confidence limit angle value relevant with Q statistical magnitude based on (4) formula, is calculated based on (5) formula and T2Statistic Relevant statistics confidence limit angle value.
Contribution amount definition portion 323 is set by operation quantity sensor and operation sensor as shown in (6) formula and (7) formula The process variable of measurement is to the Q statistical magnitude and T by (2) formula and the definition of (3) formula2The definition of the contribution amount of statistic.
Daily mode data of the model construction portion 32 based on June is this concludes the description of to build the abnormity diagnosis model in June Situation, but model construction portion 32 evaluated for the exception level to online operation data, and assume to January to December The situation that the abnormity diagnosis model of every month is built.That is, model construction portion 32 makes the daily of the every month in January to December Mode data, and make the definition of (2)~(5) formula based on these daily mode datas.
Then, illustrate to monitor the situation that the online operation data of diagnostics division 33 pairs are diagnosed.Fig. 3 is to represent monitoring diagnosis The figure of flow chart when portion 33 is judged the exception level of operation data.Additionally, in the following description, being received with from data Integrate the moment 0 that the operation data of the supply of storage unit 20 were as June xx days：Data collected by 00, by determined property portion 312 Illustrated in case of being categorized as June.
Extraction unit 331 is by calculating day of the operation data classified by division 31 from the June produced by identification part 321 Which kind of degree normal mode data deviates from, and thus makes non-daily mode data (step S31).For example, collected process number According to value be 8, in the case where identification part 321 has made daily mode data using intermediate value, the value quilt of non-daily mode data It is calculated as 8-5.5=2.5.
The non-daily mode data that exception level judging part 332 will be made by step S31, substitutes into the exception for June (2) formula and (3) formula of diagnostic model definition, calculate Q statistical magnitude and T2Statistic.Exception level judging part 332 will be by step The non-daily mode data that S31 makes, substitutes into (4) formula and (5) formula of the abnormity diagnosis model for June, calculates Q statistics The threshold value and T of amount2The threshold value (step S32) of statistic.The Q statistics that exception level judging part 332 will be calculated by step S32 Amount and T2The threshold value and T of statistic and the Q statistical magnitude calculated by step S322The threshold value of statistic is compared (step S33), Q statistical magnitude and T are judged2Whether statistic is less than threshold value (step S34).In Q statistical magnitude and T2Statistic is less than threshold value In situation (step S34's be), exception level judging part 332 is judged as exception level and (is walked for normal for 0%, i.e. operation data Rapid S35), terminate treatment.
In Q statistical magnitude and T2Statistic exceedes in the situation (step S34's is no) of threshold value, and exception level judging part 332 is sentenced Whether the disconnected abnormity diagnosis model as comparison other is that last exception in multiple abnormity diagnosis models set in advance is examined Disconnected model (step S36).Further, since the classification belonging to operation data is June, therefore last abnormity diagnosis model herein It is the abnormity diagnosis model in December.The abnormity diagnosis model as comparison other is the abnormity diagnosis model in June in step S34 (step S36's is no), thus exception level judging part 332 by exception level it is tentative be set to 10% (step S37).
Then, monitoring diagnostics division 33 be based on classification May adjacent with the abnormity diagnosis model in June abnormity diagnosis model, And the abnormity diagnosis model in July, carry out the diagnosis of operation data.That is, extraction unit 331 will turn into the abnormity diagnosis of comparison other Model is from June increment to May and July (step S38).Extraction unit 331 calculates the operation data classified by division 31 from logical The daily mode data in the May and July produced identification part 321 is crossed away from which kind of degree, the non-day in May and July is thus made Normal mode data (step S39).
May that exception level judging part 332 will be produced by step S39 and the non-daily mode data in July generation respectively Enter (2) formula and (3) formula of the abnormity diagnosis model for May and July, calculate the Q statistical magnitude and T in May and July2Statistics Amount (step S310).Q statistical magnitude and T that exception level judging part 332 will be calculated by step S3102Statistic and in May And the threshold value calculated by abnormality detection data definition portion 322 in the abnormity diagnosis model in July is compared (step S311), Treatment is set to be shifted to step S34.In the Q statistical magnitude and T relevant with the abnormity diagnosis model in May or July2Statistic is less than threshold In the situation (step S34's be) of value, exception level judging part 332 is judged as that operation data are suitable for the exception in May or July Diagnostic model, and exception level is judged as YES tentative 10% (the step S35) for setting, terminate treatment.With May and 7 Month the relevant Q statistical magnitude of abnormity diagnosis model both sides and T2Statistic exceedes in the situation (step S34's is no) of threshold value, abnormal Grade judging part 332 judge as comparison other abnormity diagnosis model whether be December abnormity diagnosis model (step S36). The abnormity diagnosis model as comparison other is the abnormity diagnosis model (step S36's is no) in May and July in step S34, because This exception level judging part 332 by exception level it is tentative be set as 30% (step S37).
Then, monitoring diagnostics division 33 be based on classification April adjacent with the abnormity diagnosis model in May abnormity diagnosis model, And the abnormity diagnosis model of the classification August adjacent with the abnormity diagnosis model in July, carry out the diagnosis of operation data.Monitoring is examined Portion 33 break after step S38~step S311 is implemented, in step S34, in the abnormity diagnosis model with April or August Relevant Q statistical magnitude and T2In statistic is less than the situation (step S34's be) of threshold value, exception level judging part 332 is judged as Operation data are suitable for the abnormity diagnosis model of April or August, and by exception level be judged as YES it is tentative set 30% (step S35), terminates treatment.In the Q statistical magnitude and T relevant with the abnormity diagnosis model both sides in April and August2Statistic surpasses Cross in the situation (step S34's is no) of threshold value, exception level judging part 332 judge be as the abnormity diagnosis model of comparison other No is abnormity diagnosis model (step S36) in December.The abnormity diagnosis model as comparison other is April and 8 in step S34 Month abnormity diagnosis model (step S36's is no), therefore exception level judging part 332 by exception level it is tentative be set as 50% (step S37).
Exception level judging part 332 is repeated the treatment of step S34~step S311.With December in step S34 The relevant Q statistical magnitude of abnormity diagnosis model and T2Statistic exceedes the situation (the no and step S36 of step S34 is) of threshold value Under, exception level is set as 100% (step S312) by exception level judging part 332, treatment is shifted to step S35.
Fig. 4 is the figure for representing exception level judging part 332 to the judged result of exception level.The transverse axis of Fig. 4 is meant that The classification of the operation data as diagnosis object.In the present note, by the moment 0 of June xx days：Operation data collected by 00 As diagnosis object, therefore it is conceived to the data of the row represented by oblique line in Fig. 4.The longitudinal axis of Fig. 4 is meant that the exception of each moon Diagnostic model.Numerical value described in each table represents the exception level in the case of the model for being not suitable for being represented by the longitudinal axis.Judge Result is exported to user interface 40.
Additionally, in Fig. 3 and Fig. 4, show as an example be not suitable for January in operation data~December is all of different Monitor that the exception level of operation data is judged as diagnostics division 33 100% situation in the case of normal diagnostic model, but if not It is suitable for several abnormity diagnosis models, then also exception level can be set as 100% certainly.
Then, illustrate that the species for monitoring 33 pairs of exceptions of diagnostics division is caused by sensor abnormality or drawn extremely by operation The situation about being identified for rising.
Element paritng portion 333 is being supplied with the moment 0 of June xx days：In the case of operation data collected by 00, make The contribution amount to the abnormity diagnosis model in June is calculated with (6) formula and (7) formula.Fig. 5 and Fig. 6 are based on by element paritng portion 333 The contribution amount for calculating is come the contribution amount curve that makes.The transverse axis of Fig. 5 and Fig. 6 is the process variable being input into MSPC, and the longitudinal axis is To Q statistical magnitude or T2The contribution amount of statistic.
Anomalous identification portion 334 as shown in Figure 5 only the contribution amount of some process variable exceed threshold k max feelings Under condition, thus it is speculated that be sensor abnormality.Additionally, anomalous identification portion 334 surpasses in the contribution amount of multiple process variables as shown in Figure 6 In the case of crossing threshold k max, thus it is speculated that be operation exception.
As described above, the identification part 321 of present embodiment uses what is robustly speculated according to past time series data Typical value, makes the daily mode data of multiple species.Identification part 321 is based on the daily mode data of the multiple species produced To make the non-daily mode data of multiple species.Abnormality detection data definition portion 322 is based on the multiple species produced Non- daily mode data makes multiple abnormity diagnosis models.Extraction unit 331 calculates the operation data phase from the supply of division 31 For the deviation degree of daily mode data, the non-daily mode data of diagnosis is thus made.Exception level judging part 332 will be examined Disconnected non-daily mode data is applied to the multiple abnormity diagnosis models built by model construction portion 32, thus makes abnormal Detection data.Exception level judging part 332 is used according to the abnormality detection that each judgement of multiple abnormity diagnosis models is produced Whether data show exceptional value, thus determine the exception level of operation data.Thus, operation monitoring diagnostic device 30 can not only Carry out the judgement of abnormal or normal this 2 value, additionally it is possible to provide abnormal degree.In the conventional monitoring carried out based on MSPC In system, the detection of abnormal sign and the separation of variable factors can be only carried out, on the other hand, according to the operation of present embodiment Monitoring diagnostic device 30, engineering (engineering) can be nearly free from and mechanically provide factory abnormality it is detailed The information of thin grade.
Additionally, phase of the identification part 321 of present embodiment based on received time series data measured value in the same time, each Moment robustly calculates typical value using average value, intermediate value, mode value etc. is pruned.Then, identification part 321 using typical value come Make daily mode data, and non-daily mode data is made based on the daily mode data produced.In this way, identification part 321 implement fairly simple statistical disposition by time series data, can automatically recognize daily pattern and non-daily pattern.Cause This, the user of monitoring system in abnormity diagnosis, can readily recognize be deviation from daily mode data exceptional value, also It is paroxysmal exceptional value.
Additionally, the identification part 321 of present embodiment makes daily mode data by being continued for calculating typical value Treatment, can automatically learn the slowly varying of daily mode data.Thereby, it is possible to easily carry out operation monitoring diagnosis dress The maintenance put.
Additionally, the identification part 321 of present embodiment by being calculated in typical value in using pruning average, intermediate value or more The each method HL suppositions of the Robust Statistics of height, MCD, bootstrap or subsample method etc., even if thus containing a large amount of exceptions In the case of value, it is also possible to robustly synthesize daily pattern.
Additionally, multiple operations that the multiple abnormity diagnosis models in 323 pairs, the contribution amount definition portion of present embodiment contain become Each of amount is defined to abnormal contribution amount.Element paritng portion 333 is by the non-daily mode data of diagnosis to contribution amount Definition input, calculate multiple process variables each to abnormal contribution amount.Thus, operation monitoring diagnostic device 30 can There is provided for the information to producing abnormal factor to be speculated.
Additionally, the anomalous identification portion 334 of present embodiment in the contribution amount calculated by element paritng portion 333 exist it is many In the case of the individual contribution amount more than threshold value, thus it is speculated that be operation exception, in the case where the contribution amount more than threshold value is one, push away It is sensor abnormality to survey.Thus, operation monitoring diagnostic device 30 can be provided for being operation abnormal still to abnormal factor The information that sensor abnormality is speculated.Particularly, the feelings of the redundant system of the sensor of identical type are provided with the factory Under condition, can further accurately identification sensor be abnormal and operation exception.Therefore, anomalous identification portion 334 can clearly know Individual sensor failure and operation exception, in the case of sensor fault, urge the maintenance of sensor and/or the replacing of sensor Deng reply, in the case of operation exception, the abnormality of operation can be quickly accounted for.
Therefore, the operation monitoring diagnostic device 30 according to present embodiment, can be to plant manager and/or operating personnel Offer should take the judgement policy of what action in exception.Thus, the row when user of monitoring system easily takes exception It is dynamic.
Additionally, the abnormality diagnostic a series of action of present embodiment assumes the operation monitoring diagnosis dress in monitoring system Realized in putting 30, but not limited to this.For example, it is also possible to be function, the i.e. division 31, mould for making operation monitor diagnostic device 30 Type structure portion 32 and monitoring diagnostics division 33 function is realized on Cloud Server, by exception level judge, abnormal factorses supposition and The information remote ground of sensor operation anomalous identification is provided to the user for needing.
For example, monitoring diagnostics division 33 is to providing the operation quantity sensor 1112~1152 that is measured in the factory and operation The office of the sensor manufacturer of sensor 121~12212 or the mobile terminal of the attendant of sensor manufacturer, The information of exception level judgement, abnormal factorses supposition and sensor operation anomalous identification is provided.Then, sensor manufacturer base In the information of exception level judgement, abnormal factorses supposition and sensor operation anomalous identification, judge whether sensor needs dimension Shield, the raising thus, it is possible to realize maintenance efficiency.
If additionally, on Cloud Server realize monitoring diagnostics division 33 by exception level judge, abnormal factorses speculate and The information of sensor operation anomalous identification sends to the mobile terminal in long-range plant operation personnel, then confirming It is not sensor fault but in the case of certain exception of operation, plant operation personnel just can quickly be located to abnormal Reason.In this way, by using modes such as clouds, information remotely can be provided to multiple users.
(second embodiment)
Fig. 7 is the block diagram of the example of the functional structure of the operation monitoring diagnostic device 50 for representing second embodiment.Fig. 7 institutes The operation monitoring diagnostic device 50 for showing possesses division 31, anomaly monitoring diagnostics division 33, model construction portion 51 and normal data and steps on Note portion 52.Additionally, in the figure 7, identical symbol is assigned for the part common with Fig. 1.
Normal data register 52 receives to be categorized as time series data of all categories by division 31.Normal data register 52 Time series data to being received is registered, and according to the request from model construction portion 51, the time series data that will be registered is to mould Type structure portion 51 exports.
Additionally, normal data register 52 is judged as that exception level is in the judged result of exception level judging part 332 In the case of 0%, it is judged as the data set of diagnosis object as in the case of normal, it is believed that in the data set not Containing abnormal data, therefore the data set is registered.
As the register method of normal data, it is possible to use arbitrary method.For example, normal data register 52 prepares There is process variable, the in a column direction matrix (table) with the time (interval between diagnosis) on line direction.Then, it is being judged as diagnosis In the case of the data set of object is normal, normal data register 52 writes data set to the matrix for being prepared.Additionally, In the case of being judged as that exception level is not 0%, normal data register 52 is not by data set for example by -99999 to square Battle array write-in.Fig. 8 represents the example of the matrix that normal data register 52 is recorded.Data beyond -99999 are registered Normal data.Then, the normal data registered is endowed in starting in zero day zero month etc. according to preassigned specified period In the case of the triggering of condition set in advance, or in the case where user has transferred, exported to model construction portion 51.
Model construction portion 51 possesses identification part 511, abnormality detection data definition portion 512 and contribution amount definition portion 513.
Normal data is read in identification part 511 from normal data register 52, and is defined with data to abnormality detection Additional the read in normal datas of matrix L T.LT.LT X ＞.Or, the information that the old times in matrix L T.LT.LT X ＞ carve is given up in identification part 511 Abandon, replace and be taken into normal data.
Abnormality detection data definition portion 512 is updated using matrix L T.LT.LT X ＞ to (1) formula~(5) formula, to abnormality detection Redefined with data.
Contribution amount definition portion 513 is updated using matrix L T.LT.LT X ＞ to (6) formula~(7) formula, and it is fixed contribution amount to be carried out again Justice.
Additionally, identification part 511 can also be updated using normal data to daily mode data as needed.But It is the various methods for using robust to speculate originally due to pattern recognition device generation mode from past time series data, therefore The normal data can not also be used and proceed to update with the cycle for specifying.
As described above, the normal data register 52 of second embodiment is judged as by exception level judging part 332 In the case of operation data are normal, normal data are registered.Then, model construction portion 51 passes through registered number According to being updated to abnormity diagnosis model.Thus, operation monitoring diagnostic device 50 can be while from abnormality diagnostic result only Normal data is extracted while being automatically updated to abnormity diagnosis model.
Also be present the known method such as ecad MSPC in the renewal of abnormity diagnosis model, but in ecad MSPC, do not have Have and abnormity diagnosis model is automatically updated and normal, the abnormal judgement of data is carried out with the cycle for specifying.Therefore, exist In the case of containing abnormal data, abnormity diagnosis model can be adapted to abnormal data sometimes, and diagnostic accuracy can deteriorate.It is real second Apply in the operation monitoring diagnostic device 50 of mode, only extract and be judged as that normal data are come to different by initial abnormity diagnosis model Normal diagnostic model is updated, and therefore, it is possible to automatically improve abnormality diagnostic performance, and can constantly be suitable for lentamente The characteristic of the factory of change.Therefore, there is no need to the maintenance of the abnormity diagnosis model as bottleneck in the realization of diagnostic system.
Additionally, the abnormality diagnostic a series of action of present embodiment is assumed to monitor diagnosis in the operation of monitoring system Realized in device 50, but not limited to this.For example, it is also possible to be make operation monitor diagnostic device 50 function, i.e. division 31, The function in model construction portion 51, monitoring diagnostics division 33 and normal data register 52 is realized on Cloud Server, by exception level Judge, abnormal factorses speculate and the information remote ground of sensor operation anomalous identification is provided to the user for needing.
Additionally, being operation monitoring diagnostic program by the program that operation monitoring diagnostic device 30,50 is performed, it is also possible to be recorded in The recording medium of embodied on computer readable.
Several embodiments of the invention is illustrated, but these implementation methods are intended only as example and point out, It is not intended to limit the scope of invention.These new implementation methods can be implemented in other various modes, without departing from invention Various omissions, displacement, change can be carried out in the range of purport.These implementation methods and its deformation be contained in invention scope and Purport, and be contained in invention and its equivalent scope of claims record.
The explanation of symbol
10 ... lower water height treatment process, 101 ... initial sedimentation basins, 102 ... anaerobism grooves, 103 ... anaerobic grooves, 104 ... need Oxygen groove, 105 ... final sedimentation tanks, 1111 ... draw-off pumps, 1112 ... extract flow sensor, 1121 ... air blowers, 1122 ... confessions To air flow sensor, 1131 ... circulating pumps, 1132 ... circular flow sensors, 1141 ... loopback sludge pumps, 1142 ... return Flow sensor, 1151 ... draw-off pumps, 1152 ... extraction flow sensors, 121 ... rain sensors, 122 ... lower water is sent to flow into Quantity sensor, 123 ... flow into TN sensors, and 124 ... flow into TP sensors, and 1251 ... flow into UV sensors, and 1252 ... flow into COD Sensor, 126 ... anaerobism groove ORP sensors, 127 ... anaerobism groove pH sensors, 128 ... anaerobic groove ORP sensors, 129 ... nothings Oxygen groove pH sensors, 1210 ... phosphoric acid sensors, 1211 ... DO sensors, 1212 ... ammoniacal sensors, 1213 ... MLSS sensors, 1214 ... cooling-water temperature sensors, 1215 ... excess sludge SS sensors, 1216 ... discharge SS sensors, 1217 ... sludge interfaces sensing Device, 1218 ... lower water discharge quantity sensors, 1219 ... discharge TN sensors, 1220 ... discharge TP sensors, 12211 ... discharge UV Sensor, 12212 ... discharge COD sensors, 20 ... Data Collection storage units, 30,50 ... operations monitoring diagnostic device, 31 ... Division, 311 ... attribute supply units, 312 ... determined property portions, 32,51 ... model construction portions, 321,511 ... identification parts, 322, 512 ... abnormality detections data definition portion, 323,513 ... contribution amount definition portions, 33 ... monitoring diagnostics divisions, 331 ... extraction units, 332 ... exception level judging parts, 333 ... element paritng portions, 334 ... anomalous identification portions, 40 ... user interfaces, 52 ... normal datas Register.

Claims (14)

1. a kind of operation monitoring diagnostic device, obtains the operation data measured by the sensor set in object operation, wherein,
Possess：Division；Model construction portion, with identification part and abnormality detection data definition portion；And monitoring diagnostics division, tool There are extraction unit and exception level judging part,
The attribute information that above-mentioned division keeps the environment according to residing for above-mentioned object operation and classifies, and based on above-mentioned attribute letter Cease to above-mentioned operation data and entered by multiple time series datas that the past operation data during multiple set in advance are constituted Row classification,
Above-mentioned identification part is classified as the respective typical value of above-mentioned time series data of multiple classifications by robustly speculating to make The daily mode data of each above-mentioned classification, and the daily mode data based on each above-mentioned classification makes non-daily pattern count According to,
Above-mentioned abnormality detection is based on each of the non-daily mode data of above-mentioned multiple with data definition portion, makes for calculating exception First definition of detection data and the second definition for calculating the threshold value of above-mentioned abnormality detection data,
Said extracted portion extracts deviation degree of the above-mentioned operation data being classified relative to above-mentioned multiple daily mode datas,
Above-mentioned exception level judging part is respectively applied to above-mentioned first definition come according to each by by above-mentioned multiple deviation degrees Above-mentioned classification calculates abnormality detection data, by by above-mentioned multiple deviation degrees be respectively applied to above-mentioned second definition come according to Each above-mentioned classification calculates threshold value, judges whether above-mentioned abnormality detection data exceed above-mentioned threshold by according to each above-mentioned classification Value judges exception level.
2. such as the operation monitoring diagnostic device that claim 1 is recorded, wherein,
Above-mentioned model construction portion also has contribution amount definition portion, and the contribution amount definition portion is based on the non-daily mode data of above-mentioned multiple Each, make for calculate each process variable for above-mentioned abnormality detection data contribution amount the 3rd definition,
Above-mentioned monitoring diagnostics division also have element paritng portion, the element paritng portion by the above-mentioned operation data being classified relative to The deviation degree of the daily mode data of adaptability highest between the above-mentioned operation data being classified, is applied to be directed to and is divided with above-mentioned Above-mentioned 3rd definition that the non-daily mode data of adaptability highest is produced between the operation data of class, thus calculates each The contribution amount of above-mentioned operation variable.
3. such as the operation monitoring diagnostic device that claim 2 is recorded, wherein,
Above-mentioned monitoring diagnostics division is also equipped with anomalous identification portion, operation of the anomalous identification portion in the contribution amount more than contribution amount threshold value In the case that variable is 1, thus it is speculated that for sensor produces exception, in the process variable of the contribution amount more than above-mentioned contribution amount threshold value In the case of for more than 2, thus it is speculated that for above-mentioned object operation produces exception.
4. such as the operation monitoring diagnostic device that claim 1 is recorded, wherein,
Above-mentioned identification part makes above-mentioned daily mode data according to the definition of multiple species relevant with calendar, and making is based on The non-daily mode data of each daily mode data defined above,
Above-mentioned abnormality detection data definition portion is non-for each produced according to each above-mentioned classification and each above-mentioned definition Daily mode data, makes above-mentioned first and second definition,
Said extracted portion is directed to each the daily mode data produced according to each above-mentioned classification and each above-mentioned definition, extracts Above-mentioned deviation degree,
Above-mentioned exception level judging part is to each calculating according to multiple deviation degrees that above-mentioned definition is identical but above-mentioned classification is different The abnormality detection data and threshold value for going out are compared, and exception level is judged based on comparative result.
5. such as the operation monitoring diagnostic device that claim 3 is recorded, wherein,
Above-mentioned anomalous identification portion the contribution amount more than contribution amount threshold value process variable in the case of 1, instead of generating The measured value of above-mentioned abnormal sensor, will be based on generating above-mentioned abnormal sensor it is normal when the value of data that measures, Supplied to above-mentioned extraction unit.
6. such as the operation monitoring diagnostic device that claim 1 is recorded, wherein,
Normal data register is also equipped with above-mentioned exception level judging part, when not producing abnormal in being judged as operation data, The normal data register will not produce above-mentioned abnormal operation data to be registered as normal data,
Above-mentioned abnormality detection data definition portion is updated using above-mentioned normal data to above-mentioned first and second definition.
7. such as the operation monitoring diagnostic device that claim 6 is recorded, wherein,
Above-mentioned normal data is incorporated into the above-mentioned time series data being classified by above-mentioned identification part, and robustly speculate be incorporated into it is above-mentioned just The typical value of the time series data after regular data, is thus updated to above-mentioned daily mode data.
8. such as the operation monitoring diagnostic device that claim 2 is recorded, wherein,
Normal data register is also equipped with above-mentioned exception level judging part, when not producing abnormal in being judged as operation data, The normal data register will not produce above-mentioned abnormal operation data to be registered as normal data,
Above-mentioned abnormality detection data definition portion is updated using above-mentioned normal data to above-mentioned first and second definition,
Above-mentioned contribution amount definition portion is updated using above-mentioned normal data to above-mentioned 3rd definition.
9. such as the operation monitoring diagnostic device that claim 8 is recorded, wherein,
Above-mentioned normal data is incorporated into the above-mentioned time series data being classified by above-mentioned identification part, and robustly speculate be incorporated into it is above-mentioned just The typical value of the time series data after regular data, is thus updated to above-mentioned daily mode data.
10. such as the operation monitoring diagnostic device that claim 1 is recorded, wherein,
Above-mentioned division, above-mentioned model construction portion and above-mentioned monitoring diagnostics division are realized on Cloud Server.
A kind of 11. operation monitoring diagnostic devices, obtain the operation data measured by the sensor set in object operation, its In,
Possess：Division；Model construction portion, with identification part, abnormality detection data definition portion and contribution amount definition portion；And Monitoring diagnostics division, with extraction unit and element paritng portion,
The attribute information that above-mentioned division keeps the environment according to residing for above-mentioned object operation and classifies, and based on above-mentioned attribute letter Breath is classified to above-mentioned operation data and by the time series data that the past operation data during presetting are constituted,
Above-mentioned identification part makes daily mode data by robustly speculating the typical value of the above-mentioned time series data being classified, and Non- daily mode data is made based on above-mentioned daily mode data,
Above-mentioned abnormality detection is based on above-mentioned non-daily mode data with data definition portion, makes for calculating abnormality detection data The first definition,
Above-mentioned contribution amount definition portion is based on above-mentioned non-daily mode data, makes for calculating each process variable for above-mentioned different Second definition of the contribution amount of normal detection data,
Said extracted portion extracts deviation degree of the above-mentioned operation data being classified relative to above-mentioned daily mode data,
Above-mentioned factor separation unit calculates each above-mentioned work by the way that above-mentioned deviation degree is applied into above-mentioned first and second definition The contribution amount of sequence variable.
The 12. operation monitoring diagnostic devices that such as claim 11 is recorded, wherein,
Above-mentioned monitoring diagnostics division also has anomalous identification portion, operation of the anomalous identification portion in the contribution amount more than contribution amount threshold value In the case that variable is 1, thus it is speculated that for sensor produces exception, in the process variable of the contribution amount more than above-mentioned contribution amount threshold value In the case of for more than 2, thus it is speculated that for above-mentioned object operation produces exception.
The 13. operation monitoring diagnostic devices that such as claim 12 is recorded, wherein,
Above-mentioned anomalous identification portion the contribution amount more than contribution amount threshold value process variable in the case of 1, instead of generating The measured value of above-mentioned abnormal sensor, will be based on generating above-mentioned abnormal sensor it is normal when the value of data that measures, Supplied to above-mentioned extraction unit.
The 14. operation monitoring diagnostic devices that such as claim 11 is recorded, wherein,
Above-mentioned division, above-mentioned model construction portion and above-mentioned monitoring diagnostics division are realized on Cloud Server.
CN201380058447.8A 2012-11-09 2013-09-05 Operation monitors diagnostic device and operation monitoring diagnostic program Active CN104781741B (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
JP2012247560A JP5996384B2 (en) 2012-11-09 2012-11-09 Process monitoring diagnostic device, process monitoring diagnostic program
JP2012-247560 2012-11-09
PCT/JP2013/073998 WO2014073261A1 (en) 2012-11-09 2013-09-05 Process monitoring/diagnosis device and process monitoring/diagnosis program

Publications (2)

Publication Number Publication Date
CN104781741A CN104781741A (en) 2015-07-15
CN104781741B true CN104781741B (en) 2017-07-04

Family

ID=50684378

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201380058447.8A Active CN104781741B (en) 2012-11-09 2013-09-05 Operation monitors diagnostic device and operation monitoring diagnostic program

Country Status (3)

Country Link
JP (1) JP5996384B2 (en)
CN (1) CN104781741B (en)
WO (1) WO2014073261A1 (en)

Families Citing this family (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9767671B2 (en) * 2014-11-05 2017-09-19 Intel Corporation System for determining sensor condition
JP6514494B2 (en) * 2014-11-28 2019-05-15 ローム株式会社 Sensor network system and its operation method
WO2016088362A1 (en) * 2014-12-05 2016-06-09 日本電気株式会社 System analyzing device, system analyzing method and storage medium
CN104731083B (en) * 2015-02-03 2017-06-20 浙江大学 A kind of industrial method for diagnosing faults and application based on self-adaptive feature extraction
JP6595245B2 (en) * 2015-07-27 2019-10-23 株式会社東芝 Monitoring system, information providing method, and computer program
JP6148316B2 (en) * 2015-07-31 2017-06-14 ファナック株式会社 Machine learning method and machine learning device for learning failure conditions, and failure prediction device and failure prediction system provided with the machine learning device
DE102016008987B4 (en) 2015-07-31 2021-09-16 Fanuc Corporation Machine learning method and machine learning apparatus for learning failure conditions, and failure prediction apparatus and failure prediction system including the machine learning apparatus
JP6504089B2 (en) * 2016-03-10 2019-04-24 横河電機株式会社 Process monitoring apparatus, process monitoring system, process monitoring method, process monitoring program and recording medium
JP6140331B1 (en) * 2016-04-08 2017-05-31 ファナック株式会社 Machine learning device and machine learning method for learning failure prediction of main shaft or motor driving main shaft, and failure prediction device and failure prediction system provided with machine learning device
JP6607821B2 (en) * 2016-04-12 2019-11-20 株式会社神戸製鋼所 Blast furnace sensor failure detection method and abnormal situation prediction method
WO2018009733A1 (en) * 2016-07-07 2018-01-11 Aspen Technology, Inc. Computer system and method for monitoring key performance indicators (kpis) online using time series pattern model
CN106371939B (en) * 2016-09-12 2019-03-22 山东大学 A kind of time series data method for detecting abnormality and its system
JP6888291B2 (en) * 2016-12-16 2021-06-16 富士電機株式会社 Process monitoring device, process monitoring system and program
EP3575908B1 (en) * 2017-01-25 2021-08-04 NTN Corporation State monitoring method and state monitoring apparatus
EP3367261A1 (en) * 2017-02-28 2018-08-29 Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. Method for classifying information and classification processor
US11467570B2 (en) * 2017-09-06 2022-10-11 Nippon Telegraph And Telephone Corporation Anomalous sound detection apparatus, anomaly model learning apparatus, anomaly detection apparatus, anomalous sound detection method, anomalous sound generation apparatus, anomalous data generation apparatus, anomalous sound generation method and program
WO2019073512A1 (en) * 2017-10-10 2019-04-18 日本電気株式会社 System analysis method, system analysis device, and program
JP7127304B2 (en) * 2018-03-12 2022-08-30 株式会社リコー Diagnostic device, diagnostic method and program
JP6763910B2 (en) * 2018-05-30 2020-09-30 横河電機株式会社 Anomaly detection device, anomaly detection method, anomaly detection program and recording medium
WO2020044533A1 (en) * 2018-08-31 2020-03-05 東芝三菱電機産業システム株式会社 Manufacturing process monitoring device
JP6615963B1 (en) * 2018-08-31 2019-12-04 株式会社日立パワーソリューションズ Abnormal sign diagnostic apparatus and abnormal sign diagnostic method
JP7163117B2 (en) * 2018-09-14 2022-10-31 株式会社東芝 Process monitoring support device, process monitoring support system, process monitoring support method, and process monitoring support program
JP7408911B2 (en) * 2018-10-26 2024-01-09 富士電機株式会社 Anomaly detection device and anomaly detection method
JP7126256B2 (en) * 2018-10-30 2022-08-26 国立研究開発法人宇宙航空研究開発機構 Abnormality diagnosis device, abnormality diagnosis method, and program
DE112019006255B4 (en) 2019-01-21 2024-07-04 Mitsubishi Electric Corporation ANOMALY CAUSE ESTIMATION DEVICE, ANOMALY CAUSE ESTIMATION METHOD AND MEDIUM
JP7030072B2 (en) * 2019-03-14 2022-03-04 株式会社日立製作所 Time-series data monitoring system and time-series data monitoring method
US12007758B2 (en) 2019-03-26 2024-06-11 Jfe Steel Corporation Process state monitoring device and process state monitoring method
CN110347561B (en) * 2019-06-11 2022-12-30 平安科技（深圳）有限公司 Monitoring alarm method and terminal equipment
JP7357494B2 (en) * 2019-09-13 2023-10-06 住友重機械工業株式会社 diagnostic system
JP7398270B2 (en) * 2019-12-23 2023-12-14 株式会社トヨタプロダクションエンジニアリング Transmitting device, receiving device, transmitting method, receiving method, transmitting program and receiving program
JP7222939B2 (en) * 2020-02-03 2023-02-15 株式会社日立製作所 Explanatory information generation device for time-series patterns
EP4128087A1 (en) * 2020-03-31 2023-02-08 Abb Schweiz Ag Data processing for industrial machine learning
WO2021255783A1 (en) * 2020-06-15 2021-12-23 株式会社日立製作所 Automatic inspection system and wireless slave machine
CN113377078B (en) * 2021-04-20 2023-04-07 山东师范大学 Industrial process fault detection method and system based on kernel principal component regression
JP7052914B1 (en) 2021-08-05 2022-04-12 富士電機株式会社 Abnormality diagnosis system, abnormality diagnosis device, abnormality diagnosis method, and program
CN114265390B (en) * 2021-12-22 2024-02-20 苏州华星光电技术有限公司 Equipment data acquisition diagnosis method, device, server and storage medium
WO2024201718A1 (en) * 2023-03-28 2024-10-03 樫山工業株式会社 Vacuum pump device

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004303007A (en) * 2003-03-31 2004-10-28 Mitsubishi Chemicals Corp Process monitoring method
JP2007065883A (en) * 2005-08-30 2007-03-15 Toshiba Corp Process monitoring device and its method
CN102282516A (en) * 2009-02-17 2011-12-14 株式会社日立制作所 Abnormality detecting method and abnormality detecting system
CN102460529A (en) * 2009-06-11 2012-05-16 株式会社日立制作所 Device abnormality monitoring method and system
JP2012141712A (en) * 2010-12-28 2012-07-26 Toshiba Corp Process monitoring diagnostic system

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3495129B2 (en) * 1995-03-03 2004-02-09 株式会社東芝 Plant abnormality detection device
JP4762088B2 (en) * 2006-08-31 2011-08-31 株式会社東芝 Process abnormality diagnosis device

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004303007A (en) * 2003-03-31 2004-10-28 Mitsubishi Chemicals Corp Process monitoring method
JP2007065883A (en) * 2005-08-30 2007-03-15 Toshiba Corp Process monitoring device and its method
CN102282516A (en) * 2009-02-17 2011-12-14 株式会社日立制作所 Abnormality detecting method and abnormality detecting system
CN102460529A (en) * 2009-06-11 2012-05-16 株式会社日立制作所 Device abnormality monitoring method and system
JP2012141712A (en) * 2010-12-28 2012-07-26 Toshiba Corp Process monitoring diagnostic system

Also Published As

Publication number Publication date
CN104781741A (en) 2015-07-15
JP5996384B2 (en) 2016-09-21
JP2014096050A (en) 2014-05-22
WO2014073261A1 (en) 2014-05-15

Similar Documents

Publication Publication Date Title
CN104781741B (en) Operation monitors diagnostic device and operation monitoring diagnostic program
JP5284433B2 (en) Process monitoring / diagnosis / support equipment
JP4762088B2 (en) Process abnormality diagnosis device
CN112863134B (en) Intelligent diagnosis system and method for rural sewage treatment facility abnormal operation
CN105027014B (en) Process monitors diagnostic device
KR20190023152A (en) The water quality monitoring system for fish farm
Corona et al. Monitoring nitrate concentrations in the denitrifying post-filtration unit of a municipal wastewater treatment plant
Ba-Alawi et al. Missing data imputation and sensor self-validation towards a sustainable operation of wastewater treatment plants via deep variational residual autoencoders
Villez et al. Combining multiway principal component analysis (MPCA) and clustering for efficient data mining of historical data sets of SBR processes
JP2012155361A (en) Process monitoring device
Wei Modeling and optimization of wastewater treatment process with a data-driven approach
KR20120001116A (en) A system and method for diagnosis algorithm development at wastewater treatment plant operation
CN116293454A (en) Management system, method and medium based on steam pipe network dynamic balance
CN110426996B (en) Environmental pollution monitoring method based on big data and artificial intelligence
Riss et al. Detection of water quality failure events at treatment works using a hybrid two-stage method with CUSUM and random forest algorithms
CN113221442A (en) Construction method and device of health assessment model of power plant equipment
Liu et al. A method of detecting contamination events using multiple conventional water quality sensors
CN117278591A (en) Park abnormity alarm system based on cloud platform
Lee et al. Autoencoder-based detector for distinguishing process anomaly and sensor failure
JP7163117B2 (en) Process monitoring support device, process monitoring support system, process monitoring support method, and process monitoring support program
Kim et al. Optimizing coagulant dosage using deep learning models with large-scale data
Dürrenmatt Data mining and data-driven modeling approaches to support wastewater treatment plant operation
JP7566525B2 (en) Process monitoring support device, process monitoring support method, process monitoring support program, and terminal device
CN117287640B (en) Early warning method, device, equipment and storage medium for water supply risk
Heikkinen et al. SOM-based modelling for an activated sludge treatment process

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
EXSB Decision made by sipo to initiate substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant