WO2014142152A1 - プロセス監視診断装置 - Google Patents
プロセス監視診断装置 Download PDFInfo
- Publication number
- WO2014142152A1 WO2014142152A1 PCT/JP2014/056439 JP2014056439W WO2014142152A1 WO 2014142152 A1 WO2014142152 A1 WO 2014142152A1 JP 2014056439 W JP2014056439 W JP 2014056439W WO 2014142152 A1 WO2014142152 A1 WO 2014142152A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- principal component
- output
- variable
- monitoring
- variables
- Prior art date
Links
Images
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
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
Definitions
- Embodiments of the present invention relate to a process monitoring diagnostic apparatus.
- Examples of specified target performance include compliance with effluent quality standards in the sewage treatment process, stable generation of generated energy (such as methane and hydrogen) in the sludge digestion process, and supply and distribution water quality standards through disinfection and sterilization in the water purification process. Compliance, the achievement of a predetermined target value for the yield of refined products such as petroleum in the petrochemical process, and the achievement of a predetermined target value for the yield of semiconductor products in the semiconductor manufacturing process.
- Examples of energy-saving and cost-saving operations include reduction of blower and pump drive power and chemical injection in the sewage treatment process, maximization of generated energy efficiency in the sludge digestion process, and minimization of chemical injection in the water purification process. In the petrochemical process and semiconductor manufacturing process, the yield can be maximized.
- the process status related to the target performance is monitored so as not to fall into a state where the predetermined target cannot be achieved, and a state change or abnormal state that impedes the achievement of the predetermined target is quickly detected and countermeasures are taken in advance. Taking it is an important point in operation management.
- the process state related to the target performance and energy savings and cost savings is always kept in a good state and likely to deviate from the good state. Need to quickly detect process state changes.
- Multivariate statistical process monitoring using the “multivariate statistical analysis method” developed mainly in the petrochemical process field as a method for diagnosing such process changes and abnormalities
- a method called “Control” is known.
- Principal component analysis PCA: Principal Component Analysis
- PLS Latent Variable Projection Method / Partial Least Square Method
- Common point Generate time series data of several summary statistics using correlation information between process data from a large number of measurement data, detect changes in process status with a small number of generated statistics data, and then The variables that cause this are estimated.
- the method of generating summary statistics is the same for both PCA and PLS.
- a highly correlated data set (data subspace) is generated, and a statistic called T2 statistic that is a distribution index of data in this subspace.
- a statistic called a Q statistic which is a deviation index from the partial space, is used.
- PCA treats all monitoring items (variables) for constructing a monitoring diagnostic model in the same line, and does not distinguish between explanatory variables (inputs) and result variables (outputs). However, PLS does not distinguish between explanatory variables (inputs) and results. Differentiate variables (outputs).
- the diagnostic method using PCA and the diagnostic method using PLS are similar to each other, but they are selectively used depending on whether input / output is distinguished.
- the final product quality (result variable / output) quality ( ⁇ normal / abnormal) we want to diagnose the final product quality (result variable / output) quality ( ⁇ normal / abnormal), but it is difficult to measure this product quality online.
- diagnosis is often performed using PLS or PCR (principal component regression) related thereto.
- MLR multiple regression analysis
- PCA is mainly used for detecting and diagnosing various abnormal states that may occur in a plant for general purposes.
- Nonlinear component analysis as a kernel eigenvalue problem Neural Computation, 10 (5): 1299-1319, 1998. '' C.Rosen “Monitoring Wastewater Treatment Systems”, Lic.Thesis, Dept. of Industrial Electrical Engineering and Automation, Lund University, Lund, Sweden (1998)
- Problem 1 Although it is desired to make a diagnosis by distinguishing between a result variable and an explanatory variable, it is difficult to select a result variable and an explanatory variable in advance, and it is desired to define a result variable and an explanatory variable while developing a diagnostic system.
- Problem 2 I want to switch the result variable that should be monitored according to the situation during plant operation.
- Problem 3 You do not necessarily want to identify and diagnose a result variable, but you do not want to make an unnecessary diagnosis. For example, it is not necessary to issue an alarm when the sensor is under maintenance and in an abnormal state. In other words, we do not want to make an overly generalized abnormality diagnosis, but only an abnormality diagnosis related to some result variables.
- the present invention has been made in view of the above circumstances, and when a result variable is appropriately designated while based on a PCA process abnormality diagnosis system that does not distinguish between a result variable and an explanatory variable, the result variable It is an object of the present invention to provide a process monitoring and diagnosis apparatus capable of extracting a factor variable strongly related to the process.
- the process monitoring and diagnosing apparatus is a process monitoring and diagnosing apparatus that monitors an arbitrary process having at least two process sensors capable of measuring a state quantity and an operation quantity of a target process at a predetermined cycle.
- output-related principal component extraction means for extracting principal components having an influence on output variables, and principal components extracted by the output-related principal component extraction means
- Summary statistic monitoring means for calculating and monitoring the defined summary statistics for abnormality detection
- summary statistic abnormality determining means for determining whether the statistics monitored by the summary statistic monitoring means are abnormal or not .
- FIG. 2 is a diagram for explaining an example of principal component extraction means that is primarily related to an output variable after primary extraction of main principal components in the process monitoring diagnostic apparatus shown in FIG. 1.
- FIG. 2 is a diagram for explaining an example of removing a process diagnosis configuration variable having a small correlation with an output variable from output related principal components in the process monitoring diagnosis apparatus shown in FIG.
- FIG. 1 shows a configuration example of a process monitoring diagnostic apparatus and a monitoring target process according to the present embodiment.
- the process monitoring diagnostic apparatus according to the present embodiment is a monitoring apparatus that monitors the advanced sewage treatment process 1 for the purpose of removing nitrogen and phosphorus.
- the sewage advanced treatment process 1 to be monitored includes an initial sedimentation tank 101, an anaerobic tank 102, an oxygen-free tank 103, an aerobic tank 104, a final sedimentation tank 105, an actuator, an operation amount sensor, and a process sensor. And.
- the actuator and its operation amount sensor are the first settling tank excess sludge pulling pump 111P and its extraction flow rate sensor 111S, the blower 112B for supplying oxygen to the aerobic tank 104 and its supply air flow rate sensor 112S, the circulation pump 113P and its A circulation flow rate sensor 113S, a return sludge pump 114P and its return flow rate sensor 114S, a final sedimentation basin surplus sludge removal pump 115P and its extraction flow rate sensor 115S are included.
- the process sensor measures a rain sensor 121, a sewage inflow sensor 122 for measuring the inflow sewage, an inflow TN sensor 123 for measuring the total nitrogen contained in the inflow sewage, and a total phosphorus amount contained in the inflow sewage.
- Inflow TP sensor 124, inflow UV sensor (or inflow COD sensor) 125 for measuring the amount of organic matter contained in inflow sewage, anaerobic tank ORP sensor 126 for measuring ORP of anaerobic tank 102, and pH of anaerobic tank 102 are measured.
- the various actuators 111P, 112B, 113P, 114P, and 115P operate at a predetermined cycle.
- the operation amount sensors 111S to 115S and the various process sensors 121 to 1221 measure at a predetermined cycle.
- the process monitoring diagnosis apparatus includes a process measurement data collection / storage unit 2, a past data (offline data) extraction unit 3, a process monitoring model construction / supply unit 4, and a current data (online data) extraction unit. 5, a process monitoring / diagnostic unit 6, and a user interface unit 7 capable of notifying a plant manager and an operator of information on state changes and abnormal signs detected by the process monitoring / diagnostic unit 6 and the cause variable candidates; It is equipped with.
- the process measurement data collection / storage unit 2 collects and holds process data obtained at predetermined intervals from the operation amount sensors 111S to 115S and the process sensors 121 to 1221.
- the past data (offline data) extraction unit 3 extracts past offline data from various time series data stored in the process measurement data collection / storage unit 2.
- the process monitoring model construction / supply unit 4 uses the offline data extracted by the past data (offline data) extraction unit 3 to construct a process monitoring / diagnosis model offline in advance.
- the process monitoring model construction / supply unit 4 includes a monitoring model configuration variable definition unit 41, a principal component calculation unit 42, an output variable designation unit 43, an output related principal component extraction unit 44, and a statistic and contribution definition unit 45. And a statistic threshold value setting unit 46.
- the monitoring model configuration variable definition unit 41 is necessary for constructing a process monitoring model from information of past time series data of measurement variables extracted from the process measurement data collection / storage unit 2 through the past data (offline data) extraction unit 3. Monitor items are selected, and variables such as management indices are newly synthesized from the monitor items as necessary to define input variables for n monitoring models.
- the principal component calculation unit 42 performs n pre-processing using PCA, robust PCA, kernel PCA, etc. after performing appropriate preprocessing such as outlier removal, filtering, and normalization (scaling) in the data. To do.
- the output variable designation unit 43 designates an output variable that the user wants to monitor from among the monitoring model configuration variables defined by the monitoring model configuration variable definition unit 41.
- the output related principal component extraction unit 44 extracts only the principal components having a strong relationship with the output variable designated by the output variable designation unit 43.
- the statistic and contribution amount definition unit 45 uses the principal components extracted by the output-related principal component extraction unit 44, the Q statistic that is a summary statistic for abnormality detection, the Hotelling T2 statistic, and the monitoring model configuration variable. The contribution amount to the Q statistic and Hotelling's T2 statistic defined by the definition unit 41 is defined.
- the statistic threshold value setting unit 46 sets threshold values for the two statistic values defined by the statistic and contribution amount defining unit 45.
- the current data (online data) extraction unit 5 extracts the current online data from various time series data stored in the process measurement data collection / storage unit 2.
- the process monitoring / diagnosis unit 6 monitors the state of the process using the online data extracted by the current data (online data) extraction unit 5 and the process monitoring model constructed by the process monitoring model construction / supply unit 4. Detect state changes and abnormal signs.
- the process monitoring / diagnosis unit 6 includes a variable selection unit 61, a statistic monitoring unit 62, a statistic abnormality determination unit 63, and a factor variable estimation unit 64.
- variable selection unit 61 defines the variables defined by the monitoring model configuration variable definition unit 41 for the current time series data of the measurement variables extracted from the process measurement data collection / storage unit 2 through the current data (online data) extraction unit 5. Is taken out and calculated.
- the statistic monitoring unit 62 performs appropriate preprocessing such as removal and normalization of missing values and outliers on the current data of the variable selected by the variable selection unit 61, These statistics are calculated according to the Q statistic determined by the contribution amount definition unit 45 and the Hotelling T2 statistic calculation formula.
- the statistic abnormality determining unit 63 detects a process state change or an abnormal sign when the statistic monitored by the statistic monitoring unit 62 exceeds the threshold defined by the statistic threshold setting unit 46. .
- the factor variable estimator 64 when the statistic abnormality determination unit 63 detects a change in the Q statistic or the Hotelling T2 statistic, determines the contribution amount of the constituent variable that becomes the change factor, and a statistic and contribution amount definition unit. The calculation is performed according to the formula defined in 45, and the variable as a factor is estimated.
- the user interface unit 7 can notify the plant manager and the operator of information on the state change and the abnormal sign detected by the process monitoring / diagnostic unit 6 and the cause variable candidate.
- the user interface unit 7 includes a display unit for notifying the operator of information and input means such as a mouse and a keyboard for the operator to input commands.
- process information is measured at predetermined intervals by the operation amount sensors 111S to 115S and the process sensors 121 to 1221. These pieces of measurement information are stored as time series data by the process measurement data collection / storage unit 2 according to a predetermined format.
- past process data stored in the process measurement data collection / storage unit 2 over a predetermined period of time is extracted as past data (offline data) extraction unit 3.
- Extract with The process monitoring model construction / supply unit 4 constructs a process monitoring model using past process data for a predetermined period extracted by the past data (offline data) extraction unit 3.
- the monitoring model configuration variable definition unit 41 in FIG. 1 defines and extracts measurement variables necessary for constructing the process monitoring model.
- the equipment such as the pumps 111P, 113P, 114P, and 115P.
- measurement variables such as measurement variables or integrated values accumulated over time.
- these variables are excluded and appropriate variables are defined as configuration variables of the monitoring model.
- blower or pump flow data that rarely starts up is input as it is, it is 0 in most time zones, so the monitoring model must be constructed correctly. Since these are not possible, these are also excluded.
- an MLSS sensor 1213 that measures the amount of activated sludge in at least one tank among the reaction tanks 102 to 104, an excess sludge SS sensor 1215 that measures the solid matter concentration of the amount of sludge pulled away from the final sedimentation basin 105, and a final SRT (sludge retention time), A-SRT (aerobic tank sludge retention time), etc. are determined from the information of the sedimentation tank surplus sludge drain pump 115P and its extraction flow rate sensor 115S and the volumes of the reaction tanks 102 to 104. It can also be one of the variables.
- a sewage discharge sensor 1218 for measuring the amount of discharged sewage a discharge TN sensor 1219 for measuring the total amount of nitrogen contained in the discharged sewage, a discharge TP sensor 1220 for measuring the total amount of phosphorus contained in the discharged sewage, or
- the discharge load amount is calculated by the product of the water concentration measured by the discharge UV sensor (or discharge COD sensor) 1221 for measuring the amount of organic matter contained in the discharged sewage, and this may be made a part of the configuration variable. it can.
- the principal component calculation unit 42 appropriately performs removal of missing data and outliers, normalization of a plurality of monitoring items (monitoring model configuration variables) having different physical dimensions, and the like. Is calculated.
- PCA which is well known as multivariate analysis is usually used.
- Non-Patent Document 1 Non-Patent Document 2
- Various robust PCA algorithms that take into account robustness against outliers may be used.
- the principal component calculation unit 42 in the present embodiment can calculate n principal components having the same number as the configuration variables without distinguishing between output variables and input variables. In the following description, explanation will be given using a normal PCA.
- a matrix X ⁇ Rm ⁇ n consisting of n configuration variables and m samples (time series data) is decomposed by PCA as follows. (However, it is assumed that X is normalized.)
- Ta ⁇ Rm ⁇ n is a matrix called a score matrix composed of m samples (or time series data) and n principal components
- Pa ⁇ Rn ⁇ n is a relationship between n constituent variables and n principal components. Is a matrix called a loading matrix.
- T ⁇ Rm ⁇ p is a Ta partial matrix truncated by p ⁇ n principal components, and is usually called a score matrix.
- P ⁇ Rn ⁇ p is a submatrix of Pa that represents the relationship with the principal component truncated by p ⁇ n for n variables, and this P is normally called a loading matrix.
- E ⁇ Rm ⁇ n is an error matrix composed of m samples (or time series data) and n variables, and represents an error when the principal component is cut off with p ⁇ n.
- Ta and T and Pa and P are clearly distinguished, Ta and Pa are called a score matrix and a loading matrix, respectively, and T and P are called a main score matrix and a main loading matrix. I will decide.
- the output variable designating unit 43 selects an output variable that is particularly noted by a user such as a plant administrator from among the variables defined by the monitoring model configuration variable defining unit 41. At this time, the output variable that the plant manager pays attention to must be defined as a configuration variable in the monitoring model configuration variable definition unit 41.
- water quality considered to be the main performance index of the plant discharge phosphorus concentration, discharge nitrogen concentration, discharge COD concentration, discharge phosphorus load amount, discharge nitrogen load amount, discharge COD load considering these as load amounts
- the blower 112B for supplying oxygen to the aerobic tank 104 and its supply air flow rate sensor 112S, the circulation pump 113P and its circulation flow rate sensor 113S, the return sludge pump 114P and its return flow rate From the flow rate of the sensor 114S, the final sedimentation basin surplus sludge removal pump 115P and its extraction flow rate sensor 115S, and the ratings and operating states (such as the number of starts and the number of revolutions) of the pumps 111P, 113P, 114P, 115P and the blower 112B,
- the amount of power in the pumps 111P, 113P, 114P, 115P and the blower 112B may be calculated, and the amount of power or the amount of power per unit processing amount (energy intensity) may be used as an output variable.
- the output variable designating unit 43 selects and defines the water quality and cost indicators related to the plant performance.
- FIG. 2 is a diagram for explaining an example of the principal component extracting means strongly related to the output variable in the process monitoring diagnostic apparatus of the present embodiment.
- the output-related principal component extraction unit 44 selects a certain output variable and extracts a loading matrix of the principal component that has a strong influence on the output variable.
- the loading matrix Pa of the principal component is represented by a 15 ⁇ 15 matrix. Note that an element of i rows and j columns of this matrix is described as Pa (i, j).
- the tenth output variable for example, electric energy
- This variable is, for example, a variable that the plant operator wants to actually monitor, such as the amount of electric power and the concentration of discharged phosphorus.
- looking at the vertical axis direction (the direction in which the principal component items are arranged) of the tenth variable of interest the magnitude of the contribution of this output variable to each principal component can be seen.
- Pa (i, j) has a property that the loading matrix Pa becomes an orthonormal matrix, and its element takes a value in the range from ⁇ 1 to 1, and the i-th principal component (hereinafter, referred to as “i” principal component of i row of Pa).
- > PMa (i,:) / 15 may be set as an output variable related principal component.
- the output related principal component extraction unit 44 constructs a matrix composed of principal components that are strongly related to the output variables.
- FIG. 3 is a diagram for explaining an example of principal component extraction means when a plurality of output variables are selected in the process monitoring diagnostic apparatus.
- the output variable designating unit 43 defines a plurality of output variables less than the number of the plurality of measurement variables, and the output related principal component extracting means 44 extracts the principal components having an influence on each output variable and then The principal component defined by the logical sum is used as the output-related principal component.
- principal components related to two or more outputs can be extracted.
- the principal component strongly related to the output variable can also be extracted by this method.
- Such an extraction method is particularly effective when it is desired to make a comprehensive judgment on a plurality of monitoring items such as the quality of discharged water.
- FIG. 4 shows a diagram for explaining an example of principal component extraction means that is strongly related to an output variable after primary extraction of main principal components in the process monitoring diagnostic apparatus.
- the output related principal component extraction unit 44 extracts one or a plurality of main principal components based on the cumulative contribution ratio of the principal components or the value of the eigenvalue in each principal component direction, and then extracts the extracted main principal components.
- the principal components having an influence on the output variable designated by the output variable designation unit 43 are extracted from the inside.
- principal components are selected from the first principal component by, for example, a method in which major principal components having a cumulative contribution rate of about 80% to 90% are sequentially extracted from above. Further, after extracting main principal components based on criteria such as the cumulative contribution rate, output-related principal components that are strongly related to the output variables are extracted from the main principal components based on the criteria shown in FIG. 3 or FIG. It is also possible.
- the principal component obtained by principal component analysis has the largest variation in the first principal component direction, and the variation of the n-th (in this case, fifteenth) principal component is considered to be minimal. Therefore, for the principal components other than the main principal components extracted by the index such as the conventional cumulative contribution rate, even if the influence of the selected output variable is strong, the fluctuation in the direction itself is small, so it is rather ignored. Sometimes it is better to do this.
- the principal components related to the output variable may be extracted after first extracting the main principal components based on the cumulative contribution rate.
- the output-related principal component extraction unit 44 includes principal component display means (not shown) for displaying a loading matrix of the same number of principal components as the plurality of measurement variables defined by the principal component computation unit 42, and a principal component Principal component selection means for selecting an output-related principal component according to a user instruction based on the loading matrix displayed by the display means.
- FIGS. 2 to 4 are presented on the monitoring screen, and a check box BX as shown on the right side of each diagram is provided so that a human can determine which principal component is selected and respond. It is also possible to have a mechanism for instructing selection to the principal component selecting means by checking the check box to support selection of the principal component.
- a loading matrix consisting of output related principal components is defined as follows.
- principal component analysis normally, after calculating a variance covariance matrix of X composed of normalized (standardized) variables, a loading matrix and a score matrix are obtained by using singular value decomposition.
- the loading matrix Poutput composed of the eighth, ninth, twelfth, fourteenth, and fifteenth principal components is constructed as follows.
- Poutput [Pa (8,1), Pa (8,2), ..., Pa (8,15), Pa (9, 1), Pa (9, 2), ..., Pa (9, 15), Pa (12, 1), Pa (12, 2), ..., Pa (12, 15), Pa (14, 1), Pa (14, 2), ..., Pa (14, 15), Pa (15, 1), Pa (15, 2), ..., Pa (15, 15)]
- the output-related principal component extraction unit 44 can be modified as follows. When the output-related principal component and the loading matrix Poutput are extracted, there are many monitoring variables that hardly contribute to the principal component.
- the first, third, fifth, seventh, eighth, eleventh, and twelfth variables surrounded by the dotted line in FIG. Has little effect. This means that there is almost no correlation between the selected tenth output variable and these variables. Therefore, the abnormality of the selected output variable and the abnormality of these variables are almost unrelated.
- a measurement variable having a small influence of the absolute value of loading (factor load) representing the magnitude of the influence of each measurement variable on each principal component of the principal component extracted by the output-related principal component extraction unit 44 A correction may be made so that the value of is replaced with zero and the corresponding measurement variable is excluded from the input items.
- the variance covariance 1 (correlation matrix) C X ′ * X used for the calculation of the principal component analysis is output before the principal component analysis.
- a correlation coefficient between the variable and another variable is obtained, and the principal component analysis may be performed using only the C submatrix from which the variable having a very weak correlation with the output variable (the tenth in FIG. 2) is removed.
- the output-related principal component extraction unit 44 outputs the output variable designation unit 43 from the correlation matrix or the variance covariance matrix that is calculated immediately before the principal component calculation is performed before the calculation of the principal component calculation unit 42 is performed.
- a highly correlated variable extracting means (not shown) for extracting a correlation matrix or a partial matrix of a variance-covariance matrix that extracts a correlation with the output specified by the above.
- the row is first, third, fifth, seventh, eighth, eleventh, twelfth
- the tenth output variable first, third, fifth, seventh
- the principal component analysis is performed using the C submatrix Cr from which the first, third, fifth, seventh, eighth, eleventh, and twelfth rows and columns of C are removed.
- the output related component extraction unit 44 extracts the loading matrix Poutput having a strong correlation with the output variable including the correcting means as described above.
- the statistic and contribution definition unit 45 defines a calculation formula for calculating the Q statistic and the Hotelling T2 statistic using the loading matrix Poutput having a strong correlation with the output as follows.
- Equation (3) and (4) are statistics defined by the statistics and contribution amount definition unit 45.
- the definition formula of the contribution amount of the input variable of each diagnosis model to the statistic defined by formula (3) and formula (4) is set.
- the contribution amount can be defined as follows.
- Qcont (n, t) x T (t, n) F (:, n) T F (:, n) x (t, n) (5)
- F (I ⁇ PoutputPoutput T )
- Contribution amount of Hotelling's T2 statistic: T 2 cont (n, t) x T (t) Poutput T ⁇ 1 Poutput (:, n) x (t, n) (6 )
- n means the nth variable
- t is a variable representing a certain time.
- the conventional method is obtained by replacing Poutput in the equations (3) to (6) with P.
- P is extracted from Pa in accordance with the magnitude of data dispersion in each principal component direction.
- statistics and contributions are defined, in this embodiment, Poutput is extracted from principal components that are strongly related to a specific output variable from Pa, and statistics and contributions are defined. Thereby, it is possible to improve the detection of an abnormality related to a specific output and the factor estimation accuracy.
- the loading matrix Poutput can be implemented by the same method as the conventional method. Therefore, if the output to be specified is changed, there is no change other than simply changing the principal component to be extracted. Therefore, it is possible to construct a process monitoring and diagnosing apparatus with almost the same effort as in the past, that is, avoiding complicated work such as repeatedly constructing a diagnostic model according to output variables.
- the statistic threshold value setting unit 46 sets the threshold values of the expressions (3) and (4). Since this threshold setting value is greatly related to the detection of state changes and abnormal signs, its setting method is important. In this embodiment, a typical setting method will be described.
- the statistical confidence limit value of the Q statistic and the statistical confidence limit value of the Hotelling T2 statistic can be used as the default setting method.
- Qlimit theoretical calculation formula: Qlimit ⁇ 1 [(c ⁇ (2 ⁇ 2 h 0 2 ) 1/2 ) / ⁇ 1 +1+ ( ⁇ 2 h 0 (h 0 ⁇ 1)) / ( ⁇ 1 2 )] 1 / h 0 (7)
- p is the number of variables left in the model.
- ⁇ i is a diagonal element of ⁇ (that is, ⁇ i is the sum of i powers of components included in the error term).
- T2limit p (m-1) / (mp) F (p, mp, ⁇ ) (8)
- m is the number of all variables.
- F (p, mp, ⁇ ) is an F distribution when the degree of freedom is (p, mp) and the confidence limit is ⁇ (often 0.01 or 0.05). It is.
- the threshold value of the statistic can be set based on the equations (7) and (8).
- the process monitoring / diagnosis unit 6 supplies the process monitoring model constructed by the process monitoring model construction / supply unit 4. The process is monitored using this process monitoring model.
- the current data (online data) extraction unit 5 extracts the online data at the point of time (hereinafter referred to as the present or present) at which diagnosis is collected in the process measurement data collection / storage unit 2.
- the process monitoring / diagnostic unit 6 monitors the process state, and if there is a change in the state or an indication of an abnormality is observed. Detect it.
- the variable selection unit 61 takes out the current data corresponding to the variables defined by the monitoring model configuration variable definition unit 41 and appropriately normalizes them using the average or variance of each variable. In addition, the outlier is removed as necessary.
- the statistic monitoring unit 62 substitutes the current Q statistic and the T2 statistic by substituting the X statistic and the T2 statistic defined by the equations (3) and (4) into X (t). Monitor. Since this statistic changes from time to time, it may be monitored in the form of a time series graph (trend graph).
- FIG. 6 shows an example of a display in which the output variable and the abnormal point of the Q statistic and / or T2 statistic are overlapped for monitoring.
- the main purpose is to estimate an abnormal factor for a specific output variable. Therefore, as shown in FIG. 6, abnormal points due to Q statistics and / or T2 statistics on the time-series data of output variables. Are preferably displayed in an overlapping manner. In FIG. 6, the abnormal points of the Q statistic and / or the T2 statistic are plotted overlapping the output variable graph.
- the abnormality related to the output variable (A in FIG. 6) Although it is qualitatively determined whether it is other abnormality (B in FIG. 6), it can be visually determined.
- FIG. 7 shows an example of a switching display in which the output variable and the abnormal point of the Q statistic and / or T2 statistic are overlapped for monitoring.
- diagnosis ON (ON) / OFF (OFF) operation screen as shown in FIG. 7
- the link between the trend graph monitoring screen for the output variable and the process diagnosis screen of the present embodiment is switched based on the judgment of the user. Can do.
- Clicking the “Process diagnosis ON” button switches the display so that abnormal points are plotted and displayed on the output variable graph.
- Clicking the “Process diagnosis OFF” button displays only the output variable graph. The display switches as shown.
- FIG. 8 is a diagram for explaining an example of a concept of cooperation between a plurality of process diagnosis menu lists and a process monitoring monitoring screen. As shown in FIG. 8, a diagnostic menu for a plurality of output variables can be prepared in advance.
- process diagnosis is performed using time-series data of Q statistics and / or T2 statistics.
- process diagnosis is performed using time-series data of Q statistics and / or T2 statistics.
- the time series data of the Q statistics and / or T2 statistics in the upper left of FIG. Switch to the function to display the overlapping points of the quantity and / or T2 statistics in an overlapping manner.
- Process monitoring by statistics can be performed by the method as described above.
- the current Q statistic or T2 statistic is defined by the threshold value set by the statistic threshold value setting unit 46, for example, the equations (7) and (8) If the threshold is exceeded, it is determined that a state change has occurred in the process.
- This is the minimum function of the statistic abnormality determination unit 63, but this function can be modified as follows.
- a Q statistic and / or a T2 statistic based on a diagnosis model that uses a loading matrix P of several higher-order principal components from the conventional first principal component in advance and a main correlation strongly related to an output variable are prepared, and the following abnormality determination can be performed.
- Judgment A P Q and T2 statistics normal, loading matrix Poutput Q and T2 statistics normal, and within output variable control value ⁇ Process is normal
- Judgment B1 P or T2 statistics abnormal in P, Q and T2 statistics normal in loading matrix Poutput, and within output variable control value ⁇
- Judgment B2 P or T2 statistics abnormality of P, Q or T2 statistics abnormality of loading matrix Poutput, and within output variable control value ⁇
- Process abnormality not related to output variable Determination C Q and T2 statistics of P are normal, Q and T2 statistics of loading matrix Poutput are normal, and other than output variable control value ⁇
- Judgment D P Q and T2 statistics normal, loading matrix Poutput Q or T2 statistics abnormal, and within output variable control value ⁇
- Judgment E P Q and T2 statistics normal, loading matrix Poutput Q or T2 statistics abnormal and other than output variable control value ⁇
- the factor variable estimation unit 64 estimates a process diagnosis configuration variable that causes the abnormality.
- the variable that becomes the factor may be estimated only when it is determined that the abnormality is related to the designated output variable.
- the contribution amount of each variable of the process diagnosis configuration variable is calculated based on the equations (5) and (6).
- A The variable with the largest amount of contribution is used as the state change factor variable.
- B Three variables in order from the largest contribution amount are used as state change factor variables.
- C The value of the contribution amount exceeds the average contribution amount ⁇ k * standard deviation of the contribution amount, k: parameter, and is defined as a state change factor variable.
- the user interface unit 7 presents not only the abnormality detection result and the factor variable estimation result as described above, but also a time series graph (trend graph) of statistical data such as Q statistics and T2 statistics as described above. It may be possible to always monitor.
- FIGS it is preferable to have a monitoring screen as shown in FIGS. It is also preferable to have display means for extracting the output-related main components as shown in FIG. Furthermore, it is preferable to provide input means such as a mouse and a keyboard used when inputting a command when the user clicks a check box on the screen.
- the present embodiment it is possible to easily estimate an abnormal sign and a factor related to a specific process variable that the plant operation manager wants to monitor. That is, in the plant monitoring and diagnosis apparatus according to the present embodiment, only one monitoring diagnosis model using PCA is used, and means for specifying an output variable to be monitored and selecting a corresponding main component are added. It is very easy to estimate abnormal signs and factors of specific process monitoring items that the plant operation manager wants to monitor.
- the factor variable strongly related to the result variable is determined.
- a process monitoring and diagnosis device that can be extracted, and is particularly flexible with respect to changes in the result variables. The number of steps and costs required for development and maintenance of the diagnosis system It is possible to provide a process monitoring diagnostic apparatus that can be realized at a low cost.
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
Description
共通点:多数の計測データからプロセスデータ間の相関情報を利用して数個の要約統計量の時系列データを生成し、生成された少数の統計量データによってプロセス状態の変化を検出し、その後にその要因となる変数を推定する。PCAでもPLSでも要約統計量の生成方法の考え方は共通であり、相関の強いデータ集合(データの部分空間)を生成し、この部分空間内のデータの分布指標であるT2統計量と呼ばれる統計量と、この部分空間からの乖離指標であるQ統計量と呼ばれる統計量を用いることが多い。
課題2:プラント運用時に状況に応じて監視すべき結果変数を適宜切り替えたい。
課題3:必ずしも結果変数を特定して診断したいわけではないが、不必要な診断を行いたくない。たとえば、センサがメンテナンス中で異常状態になっている状況でのアラーム発報は不要である。すなわち、過度に汎用的な異常診断ではなく、いくつかの結果変数に関連する異常診断のみを行いたい。
(1)処理場を管理する管理者にとって優先順位の高い要求は何か。
(2)優先順位の高い要求の中で最も効果が高いと見込める具体的ターゲットは何か。
(3)絞ったターゲットに対し、どの変数を結果変数として監視すればよいか。
(4)結果変数に影響を与えそうな説明変数の候補は何か。
このような状況で、結果変数と説明変数とを区別するPLS(およびPCRやMLR)を用いて診断モデルを構築すると、診断モデルの構築に非常に開発工数がかかり、これに伴い開発コストが増加する。加えて、プラントの運転状況やプラント管理者の好みに応じて性能指標としての結果変数の変更を伴う場合には、複数の結果変数に対する診断モデルを繰り返し構築する必要があり、診断モデルのメンテナンスに多大の労力とコストを必要とする。
一方、PCAに基づく方法では、結果変数と説明変数とを区別しないため、複数のモデルを構築する必要はなく、一旦一つの診断システムを開発できれば、PLSに基づく方法と比較して診断モデルの維持にかかるコストは小さい。しかし、PCAに基づく方法は、ある特定の結果変数に着目した診断方法としてはPLSと比較すると適していない。
X = Σi=1n ti*piT
= TaPaT
= Σi=1p ti*piT+Σi=p+1n ti*piT
= TPT + E (1)
TaεRm×nはm個のサンプル(あるいは時系列データ)とn個の主成分数とからなるスコア行列と呼ばれる行列であり、PaεRn×nはn個の構成変数とn個の主成分との関係を示すローディング行列と呼ばれる行列である。
Poutput=[Pa(8,1),Pa(8,2),…,Pa(8,15),
Pa(9,1),Pa(9,2),…,Pa(9,15),
Pa(12,1),Pa(12,2),…,Pa(12,15),
Pa(14,1),Pa(14,2),…,Pa(14,15),
Pa(15,1),Pa(15,2),…,Pa(15,15)]
なお、出力関連主成分抽出部44に対して、以下の様な修正行うこともできる。出力関連主成分、ローディング行列Poutputを抽出すると、その主成分に対して、ほとんど寄与しない監視変数がある場合も多い。
F=(I-PoutputPoutputT) HotellingのT2統計量の寄与量:T2cont(n,t)=xT(t)PoutputTΛ-1Poutput(:,n)x(t,n) (6)
ここで,nはn番目変数という意味であり,tはある時刻を表す変数である。
Qlimitの理論計算式:Qlimit=Θ1[(cα(2Θ2h0 2)1/2)/Θ1+1+(Θ2h0(h0-1))/(Θ1 2)]1/h 0 (7)
h0:=1-2Θ1Θ3/3Θ2 2 Θi:=λp+1 i+λp+2 i+………+λn i
ここで、pはモデルの中に残された変数の数である。cαは,信頼区間の限界が1-αである場合の標準正規分布の標準偏差のずれ(例:α=0.01の場合,2.53,α=0.05の場合,1.96)である。また、λiはΛの対角要素である(つまり,Θiは,誤差項に含まれる各成分のi乗和である。)。
T2limit=p(m-1)/(m-p)F(p,m-p,α)(8)
ここで、pは選択した(=モデルの中に残された)変数の数であり,mは全変数の数である。F(p,m-p,α)は,自由度が(p,m-p)であり,信頼限界をα(=0.01あるいは0.05とすることが多い)とした場合のF分布である。
判定B1:PのQまたはT2統計量異常、かつ、ローディング行列PoutputのQおよびT2統計量正常、かつ、出力変数管理値以内 ⇒ 出力変数には関係しないプロセス異常
判定B2:PのQまたはT2統計量異常、かつ、ローディング行列PoutputのQまたはT2統計量異常、かつ、出力変数管理値以内 ⇒ 出力変数には関係しないプロセス異常
判定C:PのQおよびT2統計量正常、かつ、ローディング行列PoutputのQおよびT2統計量正常、かつ、出力変数管理値以外 ⇒ 出力変数の単独異常
判定D:PのQおよびT2統計量正常、かつ、ローディング行列PoutputのQまたはT2統計量異常、かつ、出力変数管理値以内 ⇒ 出力変数が異常になる兆候あり
判定E:PのQおよびT2統計量正常、かつ、ローディング行列PoutputのQまたはT2統計量異常かつ出力変数管理値以外 ⇒ 出力変数異常
判定F:PのQまたはT2統計量異常、かつ、ローディング行列PoutputのQまたはT2統計量異常、かつ、出力変数管理値以外 ⇒ 恐らく出力変数に関連する異常
判定不能:PのQまたはT2統計量異常、かつ、ローディング行列PoutputのQまたはT2統計量異常、かつ、出力変数管理値以外 ⇒ 不明 上記の様な統計量を用いた異常判定を行い、その旨をユーザインタフェース部7を通してオペレータあるいはプロセス管理者に通知する。この場合、例えば、しきい値を超える回数が連続してr回(rは正の整数)続いた場合にオペレータに通知するなどのルールを入れてアラームの頻発を避けるようにしておいてもよい。
(a)寄与量の最も大きいものを状態変化要因変数とする
(b)寄与量の大きいものから順に3個を状態変化要因変数とする
(c)寄与量の値が、寄与量の平均±k*寄与量の標準偏差、k:パラメータ、を超えたものを状態変化要因変数とする
などのルール(a)~(c)を予め決めておくことによって、状態変化の要因と考えられる計測変数あるいは指標を推定し、ユーザインタフェース部7を通して推定した指標をオペレータあるいはプロセス管理者に通知する。
Claims (11)
- 対象プロセスの状態量や操作量を所定の周期で計測可能な少なくとも2つのプロセスセンサを有する任意のプロセスを監視するプロセス監視診断装置であって、
前記プロセスセンサによって計測される複数の計測変数の時系列データを収集し、保持するデータ収集・保存手段と、
入力と出力とを区別しない多変量解析により、前記計測変数と同じ数の主成分を定義可能な主成分演算手段と、
前記複数の計測変数の中から1又は複数の出力変数を指定する出力変数指定手段と、
前記主成分演算手段の中から、前記出力変数指定手段によって指定された出力変数に影響を持つ主成分を抽出する出力関連主成分抽出手段と、
前記出力関連主成分抽出手段によって抽出した主成分を用いて、予め定義した異常検出用の要約統計量を計算し監視する要約統計量監視手段と、
前記要約統計量監視手段によって監視されている統計量に対する異常か否かの判断を行う要約統計量異常判定手段と、を備えることを特徴とするプロセス監視診断装置。 - 前記多変量解析は、主成分分析、ロバスト主成分分析、カーネル主成分分析、のいずれかであることを特徴とする請求項1記載のプロセス監視診断装置。
- 前記異常判定手段によって異常と判定された場合に、前記要約統計量に対する各計測変数の寄与量を計算し要因となる変数を推定する要因変数推定手段をさらに備えることを特徴とする請求項1または請求項2記載のプロセス監視診断装置。
- 前記出力変数指定手段は、前記複数の計測変数の数未満の複数個の出力変数を定義し、
前記出力関連主成分抽出手段は、各出力変数に影響を持つ主成分を抽出した後にその論理和で定義される主成分を出力関連主成分とすることを特徴とする、請求項1乃至請求項3のいずれか1項記載のプロセス監視診断装置。 - 前記出力関連主成分抽出手段は、主成分の累積寄与率あるいは各主成分方向の固有値の値に基づいて1又は複数の主な主成分を抽出した後、前記主な主成分の中から前記出力変数指定手段によって指定された出力変数に影響を持つ主成分を抽出することを特徴とする、請求項1乃至請求項4のいずれか1項記載のプロセス監視診断装置。
- 出力関連主成分抽出手段は、前記主成分演算手段で定義された前記複数の計測変数と同じ数の主成分のローディング行列を表示する主成分表示手段と、前記主成分表示手段で表示されたローディング行列に基づいて出力関連の主成分をユーザの指示に従って選定する主成分選定装置と、を備えることを特徴とする、請求項1乃至請求項5のいずれか1項記載のプロセス監視診断装置。
- 前記出力関連主成分抽出手段によって抽出した主成分の、各計測変数の各主成分への影響の大きさを表すローディングの絶対値が、所定値以下の計測変数の値をゼロに置換し、対応する計測変数を入力項目から除外するように修正を加えることを特徴とする請求項1乃至請求項6のいずれか1項記載のプロセス監視診断装置。
- 前記主成分演算手段の演算を実行する前に、主成分計算実行直前に演算される相関行列あるいは分散共分散行列の中から、前記出力変数指定手段によって指定された出力との相関が所定値以上のものを抽出し、前記相関行列あるいは前記分散共分散行列の部分行列を抽出する高相関変数抽出手段をさらに有することを特徴とする、請求項1乃至請求項6のいずれか1項記載のプロセス監視診断装置。
- 前記出力変数指定手段で指定した出力変数を時系列データとして監視する表示装置を備え、
前記表示装置は前記要約統計量異常判定手段によって異常と判定された箇所を前記出力変数のグラフに重複して表示可能であることを特徴とする請求項1乃至請求項8のいずれか1項記載のプロセス監視診断装置。 - 前記表示装置は、前記出力変数のグラフを表示する異常診断オフ表示と、前記要約統計量異常判定手段によって異常と判定された箇所を前記出力変数のグラフに重複して表示する異常診断オン表示とを切替え可能であることを特徴とする、請求項9記載のプロセス監視診断装置。
- 複数の出力変数から所定の出力変数を選択可能なプロセス診断メニューを表示する診断メニュー表示および選択手段を備え、
前記メニュー表示および選択手段は、前記プロセス診断メニュー上で前記所定の出力変数を指定すると、指定された前記出力変数に関するプロセス診断モデルを構築し、それに対応する診断結果を表示することを特徴とする、請求項1乃至請求項10のいずれか1項記載のプロセス監視診断装置。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201480010812.2A CN105027014B (zh) | 2013-03-14 | 2014-03-12 | 工序监视诊断装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013-051930 | 2013-03-14 | ||
JP2013051930A JP6214889B2 (ja) | 2013-03-14 | 2013-03-14 | プロセス監視診断装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2014142152A1 true WO2014142152A1 (ja) | 2014-09-18 |
Family
ID=51536803
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2014/056439 WO2014142152A1 (ja) | 2013-03-14 | 2014-03-12 | プロセス監視診断装置 |
Country Status (3)
Country | Link |
---|---|
JP (1) | JP6214889B2 (ja) |
CN (1) | CN105027014B (ja) |
WO (1) | WO2014142152A1 (ja) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104808626A (zh) * | 2015-04-20 | 2015-07-29 | 中国石油化工股份有限公司 | 化工装置开车时工艺参数的报警方法 |
CN104914821A (zh) * | 2015-04-20 | 2015-09-16 | 中国石油化工股份有限公司 | 化工装置精馏系统开车时的报警方法 |
US20170299565A1 (en) * | 2016-04-15 | 2017-10-19 | Beijing University Of Technology | Computing system for detecting total phosphorus in effluent using data driven A2/O process |
WO2018052015A1 (ja) * | 2016-09-14 | 2018-03-22 | 日本電気株式会社 | システムの分析支援装置、システムの分析支援方法及びプログラム |
CN107949814A (zh) * | 2016-04-18 | 2018-04-20 | 胜科工业有限公司 | 用于废水处理工艺控制的系统和方法 |
CN110333474A (zh) * | 2019-06-28 | 2019-10-15 | 中国电力科学研究院有限公司 | 一种在线检测互感器计量异常状态的方法和系统 |
CN110687895A (zh) * | 2019-10-24 | 2020-01-14 | 上海工程技术大学 | 一种基于自适应核主成分分析的化工过程故障检测方法 |
CN113741224A (zh) * | 2021-11-08 | 2021-12-03 | 南京铖联激光科技有限公司 | 一种基于上位机的监控系统和方法 |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105809923A (zh) * | 2016-03-08 | 2016-07-27 | 胡秀娟 | 净水系统数据采集方法及系统 |
WO2017152359A1 (zh) * | 2016-03-08 | 2017-09-14 | 胡秀娟 | 净水系统数据采集方法及系统 |
JP6504089B2 (ja) * | 2016-03-10 | 2019-04-24 | 横河電機株式会社 | 工程監視装置、工程監視システム、工程監視方法、工程監視プログラム及び記録媒体 |
JP7059526B2 (ja) * | 2017-06-30 | 2022-04-26 | 横河電機株式会社 | 水処理施設における運転監視装置 |
WO2019202700A1 (ja) * | 2018-04-19 | 2019-10-24 | 三菱電機株式会社 | 異常設備を特定する装置、方法、およびコンピュータプログラム |
US11460364B1 (en) | 2019-02-07 | 2022-10-04 | Starrycom Sensing Technologies Inc. | Multiplexed inductive tactile sensor array |
CN111083662B (zh) * | 2019-12-18 | 2020-10-27 | 深圳市快鱼环保技术有限公司 | 基于云计算的水质监测物联网系统 |
CN113048402A (zh) * | 2019-12-27 | 2021-06-29 | 柯思科技股份有限公司 | 中低压管线监控系统及其方法 |
WO2021168308A1 (en) * | 2020-02-21 | 2021-08-26 | Nanotronics Imaging, Inc. | Systems, methods, and media for manufacturing processes |
WO2024180746A1 (ja) * | 2023-03-01 | 2024-09-06 | 富士通株式会社 | 因果探索プログラム、因果探索方法および情報処理装置 |
JP2024129372A (ja) * | 2023-03-13 | 2024-09-27 | オムロン株式会社 | 情報処理装置、情報処理方法、プログラムおよび生産システム |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH08241121A (ja) * | 1995-03-03 | 1996-09-17 | Toshiba Corp | プラント異常検知装置 |
JP2004303007A (ja) * | 2003-03-31 | 2004-10-28 | Mitsubishi Chemicals Corp | プロセス監視方法 |
JP2007065883A (ja) * | 2005-08-30 | 2007-03-15 | Toshiba Corp | プロセス監視装置及びその方法 |
JP2012141712A (ja) * | 2010-12-28 | 2012-07-26 | Toshiba Corp | プロセス監視診断装置 |
JP2012155361A (ja) * | 2011-01-21 | 2012-08-16 | Toshiba Corp | プロセス監視装置 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101251754B (zh) * | 2008-03-13 | 2010-06-02 | 西安交通大学 | 一种多工序加工过程误差流的处理及控制方法 |
-
2013
- 2013-03-14 JP JP2013051930A patent/JP6214889B2/ja active Active
-
2014
- 2014-03-12 CN CN201480010812.2A patent/CN105027014B/zh active Active
- 2014-03-12 WO PCT/JP2014/056439 patent/WO2014142152A1/ja active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH08241121A (ja) * | 1995-03-03 | 1996-09-17 | Toshiba Corp | プラント異常検知装置 |
JP2004303007A (ja) * | 2003-03-31 | 2004-10-28 | Mitsubishi Chemicals Corp | プロセス監視方法 |
JP2007065883A (ja) * | 2005-08-30 | 2007-03-15 | Toshiba Corp | プロセス監視装置及びその方法 |
JP2012141712A (ja) * | 2010-12-28 | 2012-07-26 | Toshiba Corp | プロセス監視診断装置 |
JP2012155361A (ja) * | 2011-01-21 | 2012-08-16 | Toshiba Corp | プロセス監視装置 |
Non-Patent Citations (1)
Title |
---|
OSAMU YAMANAKA ET AL.: "Application of Multivariate Statistical Process Control Technique for Efficient and Stable Plant Operation of Wastewater Treatment Process", THE JOURNAL OF THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS, vol. 94, no. 4, 1 April 2011 (2011-04-01), pages 299 - 304 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104808626A (zh) * | 2015-04-20 | 2015-07-29 | 中国石油化工股份有限公司 | 化工装置开车时工艺参数的报警方法 |
CN104914821A (zh) * | 2015-04-20 | 2015-09-16 | 中国石油化工股份有限公司 | 化工装置精馏系统开车时的报警方法 |
CN104914821B (zh) * | 2015-04-20 | 2017-12-12 | 中国石油化工股份有限公司青岛安全工程研究院 | 化工装置精馏系统开车时的报警方法 |
CN104808626B (zh) * | 2015-04-20 | 2018-01-02 | 中国石油化工股份有限公司青岛安全工程研究院 | 化工装置开车时工艺参数的报警方法 |
US20170299565A1 (en) * | 2016-04-15 | 2017-10-19 | Beijing University Of Technology | Computing system for detecting total phosphorus in effluent using data driven A2/O process |
US10788473B2 (en) * | 2016-04-15 | 2020-09-29 | Beijing University Of Technology | Computing system for detecting total phosphorus in effluent using data driven A2/O process |
EP3353613A4 (en) * | 2016-04-18 | 2019-08-21 | SembCorp Industries Ltd | SYSTEM AND METHOD FOR CONTROLLING A PROCESS FOR WASTEWATER TREATMENT |
CN107949814A (zh) * | 2016-04-18 | 2018-04-20 | 胜科工业有限公司 | 用于废水处理工艺控制的系统和方法 |
JPWO2018052015A1 (ja) * | 2016-09-14 | 2019-06-24 | 日本電気株式会社 | システムの分析支援装置、システムの分析支援方法及びプログラム |
WO2018052015A1 (ja) * | 2016-09-14 | 2018-03-22 | 日本電気株式会社 | システムの分析支援装置、システムの分析支援方法及びプログラム |
US11455203B2 (en) | 2016-09-14 | 2022-09-27 | Nec Corporation | Abnormality detection support device, abnormality detection support method, and program |
JP7183790B2 (ja) | 2016-09-14 | 2022-12-06 | 日本電気株式会社 | システムの分析支援装置、システムの分析支援方法及びプログラム |
CN110333474A (zh) * | 2019-06-28 | 2019-10-15 | 中国电力科学研究院有限公司 | 一种在线检测互感器计量异常状态的方法和系统 |
CN110687895A (zh) * | 2019-10-24 | 2020-01-14 | 上海工程技术大学 | 一种基于自适应核主成分分析的化工过程故障检测方法 |
CN110687895B (zh) * | 2019-10-24 | 2022-11-18 | 上海工程技术大学 | 一种基于自适应核主成分分析的化工过程故障检测方法 |
CN113741224A (zh) * | 2021-11-08 | 2021-12-03 | 南京铖联激光科技有限公司 | 一种基于上位机的监控系统和方法 |
CN113741224B (zh) * | 2021-11-08 | 2022-02-08 | 南京铖联激光科技有限公司 | 一种基于上位机的监控系统和方法 |
Also Published As
Publication number | Publication date |
---|---|
CN105027014B (zh) | 2017-09-05 |
JP2014178844A (ja) | 2014-09-25 |
CN105027014A (zh) | 2015-11-04 |
JP6214889B2 (ja) | 2017-10-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6214889B2 (ja) | プロセス監視診断装置 | |
JP5284433B2 (ja) | プロセス監視・診断・支援装置 | |
JP5996384B2 (ja) | プロセス監視診断装置、プロセス監視診断プログラム | |
JP4762088B2 (ja) | プロセス異常診断装置 | |
Aguado et al. | Multivariate SPC of a sequencing batch reactor for wastewater treatment | |
JP5813317B2 (ja) | プロセス状態監視装置 | |
JP4468269B2 (ja) | プロセス監視装置及びその方法 | |
Wang | Robust data-driven modeling approach for real-time final product quality prediction in batch process operation | |
Garcia-Alvarez et al. | Fault detection and isolation in transient states using principal component analysis | |
JP6933899B2 (ja) | プラント運転支援装置、プラント運転支援方法、及びプラント運転支援プログラム | |
JP4604987B2 (ja) | 水処理プラントの運転管理方法および装置 | |
García-Muñoz et al. | Model predictive monitoring for batch processes | |
JP6606003B2 (ja) | プロセス診断装置、プロセス診断方法及びコンピュータプログラム | |
JP2018120343A (ja) | プロセス診断装置、プロセス診断方法及びプロセス診断システム | |
Marais et al. | Comparing statistical process control charts for fault detection in wastewater treatment | |
JP2024001315A (ja) | プロセス監視装置、プロセス監視方法及びプログラム | |
JP5722371B2 (ja) | ノウハウ可視化装置及びノウハウ可視化方法 | |
WO2020054850A1 (ja) | プロセス監視支援装置、プロセス監視支援システム、プロセス監視支援方法、プロセス監視支援プログラム及び端末装置 | |
JP2014106841A (ja) | エネルギー管理支援装置および方法、ならびにプログラム | |
Rusinov et al. | Real time diagnostics of technological processes and field equipment | |
Kaneko et al. | Data density-based fault detection and diagnosis with nonlinearities between variables and multimodal data distributions | |
JP6749219B2 (ja) | プラント運転データ解析システム | |
Ivan | Fault Detection in Wastewater Treatment: Process Supervision to Improve Wastewater Reuse | |
JP7551425B2 (ja) | プラント監視支援装置 | |
JP7566525B2 (ja) | プロセス監視支援装置、プロセス監視支援方法、プロセス監視支援プログラム及び端末装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 201480010812.2 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 14765435 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: IDP00201506444 Country of ref document: ID |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 14765435 Country of ref document: EP Kind code of ref document: A1 |