JP2002526852A - Method and system for monitoring and controlling a manufacturing system - Google Patents

Abstract

(57) [Summary] A neural network model interfaces with a distributed control system associated with a manufacturing facility to operate the manufacturing facility. The neural network model receives measured variables of the manufacturing process, predicts process performance data, and provides the performance data to the communication server in real time. The graphical user interface communicates over a network, such as the Internet or an intranet of a company, to receive and present real-time performance data, including performance measurements, such as a key performance real-time analyzer, so that managers can make decisions about the manufacturing process. To help. The communication server also interfaces with the offline model engine to transfer neural network models and real-time performance data for analysis on the offline engine. Object-oriented box transformation enhances the publishing and subscribing of performance data from neural network models.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001] The present invention relates generally to the field of control systems, and more particularly, to improved methods and systems for monitoring and controlling performance indicators of a manufacturing process.

BACKGROUND OF THE INVENTION Complex manufacturing facilities perform a wide variety of resource-manufacturing processes to produce the desired end product. For example, a cogeneration plant of a petrochemical plant produces steam for operating the plant and electricity for use by the plant or for sale in the utility grid. Resources used in such cogeneration facilities include tangible resources such as petrochemical raw materials and fuels for steam generation, environmental regulations, facility restrictions,
And intangible resources such as restrictions on the configuration of the electrical distribution network.

[0003] The equipment associated with a typical cogeneration facility has a wide range of operating conditions,
Interacting and interconnected unit operations are distributed throughout the manufacturing facility. These unit operations produce a tremendous amount of performance information that can be used to monitor the manufacturing process, including process measured variables such as pressure, temperature, flow, and production rate performance measurements. Is included. Performance information is generally measured by unit operation, converted to electrical signals,
It is sent to a distributed control system (DCS) or stored in a process data storage device (storage device). Similar performance measures are commonly measured in other types of manufacturing processes, including pulp and paper processes, refining processes, food processes, and power production.

[0004] Process performance measurements provide manufacturing facility management with the essential information to monitor and optimize the manufacturing process. One of the difficulties associated with the large amount of performance data available for equipment management is that there is too much information from a management perspective to integrate decisions that can optimize equipment production. Using the example of a cogeneration plant, managers can change the distribution of electricity and petrochemical production based on fluctuations in electricity prices and fluctuations in steam demand, but these can be varied within the plant. Is a function of the process production speed of the unit operation. Managerial decisions on production may also face regulatory restrictions, such as environmental restrictions, facility restrictions, and restrictions on the configuration of electrical distribution networks. Managers trying to make production decisions with production conditions fluctuating face the daunting task of overflowing information.

Many companies use a set of standard performance matrices to monitor and analyze the economic results of manufacturing facilities and processes so that decision makers can focus on the most relevant information possible. For example, petroleum refiners typically use the Solomon Ut
Refining performance is evaluated using an index provided by Solomon Associates Inc., such as an ilization index. The utilization index calculates the utilization of the purification processing unit compared to the theoretical production capacity. The use of a standard set of performance matrices is beneficial from a job monitoring perspective as it limits the need for an administrator to manage detailed manufacturing performance data. However, since the performance matrix is generally calculated from accounting data generated after the end of the accounting period, it is less useful for managers' decision making over multiple accounting periods.

In addition to the vast amount of information available to monitor and control the manufacturing process, managers also face significant decisions about the quality of information they must rely on. For example, historical data about a manufacturing process provides insight into the production variability that has occurred so far and how such variability is likely to affect future production goals. However, the use of historical data tends to rely on subjective decisions, limiting its usefulness in situations where production varies according to unpredictable parameters.

One technique that can be used to help managers make decisions about production is to generate a model of the manufacturing process and use that model to examine changing production parameters. Pavilion Technolog
ies Inc.) uses neural network technology to train manufacturing process models and develop a number of tools. For example, Pavilion's SoftSensor ™ technology can create a neural network using historical data to model a given manufacturing process, with quality measurements, production measurements,
Alternatively, it is possible to predict the behavior of a process such as other relevant process information. Once a high fidelity model of the manufacturing process is created, the model can use the measured process variables to predict process behavior and optimize and control unit operation to achieve desired results. it can. For example, the measured variables are distributed by the runtime application engine (RAE)
To manipulate and control such variables to optimize production in terms of economy, production capacity, and regulations regarding the operation of each unit. Pavilion's product, Process Insights, Boost, and Process Perfecter, enhances the runtime application engine's ability to optimize and control unit operations.

[0008] Neural network models, such as those made possible by technological advances in the pavilion, significantly increase the efficiency and optimization of the manufacturing process by interacting directly with the distributed control system. For example, a neural network model can predict the outcome of a process using real-time measured variables of a process controlled by the distributed control system, instruct the distributed control system to change certain parameters, Process results can be obtained. In general, using a neural network model that communicates with the control system will make more efficient use of the resources of the process.
Limited utilities to support real-time decision-making by administrators. One of the difficulties is establishing an interface between the neural network model and the decision makers who manage it. Neural network models generally perform critical control functions, which can disrupt the manufacturing process if interrupted by information requests from other systems. Furthermore, since a neural network model is generated using a large amount of data, it is difficult to select and provide the most relevant information. Another difficulty associated with using neural network models is the format in which the models use to store performance data. In general, a neural network model is
It has a flat table-driven database from which it is inconvenient and difficult to extract the desired data points.

SUMMARY OF THE INVENTION Accordingly, a need has arisen for a method and system that allows for the display and control of real-time and predictable performance information of a manufacturing process to enhance and optimize process control. .

[0010] There is a further need for methods and systems that allow access to neural networks or other (first principles) model information without interrupting the manufacturing process.

There is a further need for methods and systems that allow individual users to customize the presentation of key performance measurements on an individual basis and provide presentations of information relevant to the interests and responsibilities of individual users. I do.

There is a further need for methods and systems that provide facilities for publishing and subscribing to manufacturing process performance data from a neural network model associated with the manufacturing process.

[0013] There is a further need for methods and systems that increase the accessibility of certain manufacturing process performance data used to manage, optimize, and control a manufacturing process. The method and unit for presenting performance data enhances the ability of decision makers to access and utilize the process performance data predicted by the neural network model. The uniform flexible architecture creates a distributed virtual plant to enable real-time display and manipulation of performance data, and enables modeling, simulation, model predictive control, and optimization of manufacturing processes.

More specifically, the present invention provides a method for presenting performance data of a manufacturing process performed in a manufacturing facility. A neural network model of the manufacturing process is created using the measured variables of the process. Neural network models can predict process behavior such as quality measurements, manufacturing measurements, or other relevant process information. Neural network models provide real-time and future predictions of measured variables of a process and predicted performance variables of the process. For example, a key performance index (KPI), typically calculated from historical data and accounting data, is instead calculated as a key performance real-time analyzer (KPRA) using a real-time neural network model of performance measurement. The neural network model is interfaced with the manufacturing facility, which provides real-time performance data, such as measured variables of the process, to the neural network model. The neural network model predicts process performance data and transfers the predicted performance data to a network for display on a graphical user interface.

[0015] In one embodiment, the neural network model is interfaced to a distributed control system and provides predicted performance data to improve control of the manufacturing process by the distributed control system. The neural network model is also interfaced to a communication server. The manufacturing process can use multiple neural network models, and one or more neural networks or other models interface with the distributed, interconnected unit operations associated with the manufacturing process. Each such neural network model can be interfaced with a communication server via the network, and the communication server can predict the process variables from the neural network, the process variables predicted by the neural network, the neural network including KPRA. Request and receive real-time performance data, including performance variables.

The communication server interfaces with a graphical user interface through a network such as a corporate intranet or the Internet. The graphical user interface presents performance data of the manufacturing process, including real-time and predicted performance data, in a user-friendly format to reinforce management decisions regarding the manufacturing process. The graphical user interface allows the selection of predetermined performance data to be displayed, so that the administrator can make a decision based on the selected relevant information. For example, a graphical user interface can display KPRA for a standard set of administrator performance measurements based on real-time data received online from a communication server.

In addition, the graphical user interface accepts inputs that enable “what-if” and analysis of setpoints. For example, a graphical user interface allows a change in value for one or more predetermined performance data points and forwards the change over a network to a communication server. The change may be, for example, a change in a setting value for determining a KPRA associated with a new setting value, or a change in a KPRA that determines a setting value for optimally implementing the KPRA. The communication server
The neural network model can be requested to perform what-if or setpoint analysis based on input changes to the performance data, or the offline model engine can perform the analysis in real time so that the offline model engine can perform the analysis. Models can be transferred. The communication server is what-i
Provide the results of f or set point analysis in a graphical user interface. Thus, for example, a decision maker can enter a hypothetical situation for a manufacturing facility and view estimated KPRA values for the hypothetical situation based on real-time facility parameters. Then, the graphical user interface transfers the manufacturing setting values from the user to the communication server via the network, and allows the setting to be performed by the control model and the optimization model related to the manufacturing process.

The present invention offers a number of important technical advantages. One important technical advantage is that
Real-time KP with user-friendly graphical user interface
Presenting real-time process performance data such as RA values.
By making available predetermined and selected real-time performance data from a neural network via a communication server, the present invention allows an administrator to study manufacturing processes in real time at a remote location, and via a graphical user interface. A decision on manufacturing can be made.

Another important technical advantage of the present invention is that performance data is made available in real time while minimizing the risk of disruption of the control process. For example, the communication server mediates requests for performance information to minimize the risk of overloading the neural network model for process control. The communication server makes the performance data available in real time to the offline model engine for what-if or setpoint analysis, thereby minimizing any impact on the computing platform supporting the neural network model. Supports future state diagrams of the manufacturing process.

Another important technical advantage of the present invention is the ability of a neural network model to accurately predict future process state conditions and present them to a decision maker in a graphical user interface. The performance data provided in the graphical user interface can be used to monitor and analyze real-time measured variables of the process, real-time measured variables of the predicted process, and economic results.
A performance indicator of the process, such as RA. Predicting in real time and calculating future performance indicators facilitates quantitative and qualitative economic decisions faced by manufacturing process managers.

Another important technical advantage of the present invention is that the present invention
A uniform and flexible architecture for manufacturing process simulation, manufacturing process model predictive control, and manufacturing process optimization. By combining these four functions into a single system and making performance data available to a remote graphical user interface, the present invention simplifies and enhances control of the manufacturing process. Simulate variables for manufacturing conditions and predict optimal settings based on expected future manufacturing conditions to make manager-level manufacturing decisions with reference to actual real-time manufacturing conditions. it can. This knowledge, condensed into KPRA user-friendly presentations, allows the administrator to make optimal decisions and provide settings for these decisions directly to the control system, boiling down to user-friendly KPRA presentations.

Another important technical advantage of the present invention is the “pick-up and copy” of neural or other network models and real-time performance data by a communication server for use with an offline model engine. This allows the current state of the neural network model and the configuration of the manufacturing process to be separated and frozen in real time for what-if and / or setpoint analysis. The present invention provides a hierarchical tree using active transform building blocks, which allows for variable naming relationships for the transform blocks and allows the desired data to be easily published and subscribed to a graphical user interface.

A more complete understanding of the present invention and its advantages may be obtained by reference to the following description taken in conjunction with the accompanying drawings, in which like reference numerals designate like features.

DETAILED DESCRIPTION OF THE INVENTION [0024] Advantageous embodiments of the present invention are illustrated in the drawings, wherein like numerals are used to refer to like and corresponding parts of the various drawings.

Neural network technology has greatly enhanced manufacturing equipment operations by providing a model that simplifies and optimizes process control decisions at the equipment level. However, the great power of neural networks is that
Providing high-level information for equipment management with an overall view of the equipment towards optimizing the manufacturing process economy is not yet available. This is due in part to the large amount of information made available by the neural network and in part to the complex data structures utilized by the neural network. The present invention provides a link between performance information available via a neural network and performance measurements used by managers to make economical decisions on manufacturing across the facility.

FIG. 1 shows a simplified block diagram of a production facility for generating electricity. The boiler 12 produces steam that rotates the turbine 14 so that the turbine can generate electricity. Boiler 12 provides measured variables such as steam temperature and pressure to distributed control system 16. The decentralized control system 16 controls the measurement variables provided by the boiler 12 by operating the equipment operation. For example, the decentralized control system 16 can maintain a predetermined temperature or pressure set point for the boiler 12 by manipulating the electromechanical control to flow fuel to the boiler 12. Similarly, the decentralized control system 18 includes a turbine speed (RPM),
It interfaces with the turbine 14 to receive power output and measured variables of turbine operation such as turbine steam pressure and temperature. The boiler 12, turbine 14 and its associated distributed control system form an interacting and interconnected unit operation for generating electricity. Thus, for example, the distributed control system 16 and the turbine distributed control system 18 can operate the electromechanical control to maintain a predetermined set point to generate a predetermined amount of electricity. The history database 20 tracks and stores variables set for analysis and accounting purposes.

The block diagram of the electric production facility 10 illustrated by FIG. 1 is the simplest as compared to a typical cogeneration facility. For example, a typical cogeneration facility has multiple boilers and several turbines. In addition, steam from the boiler also supports petrochemical production. Therefore,
Small changes by a single decentralized control system to achieve existing or new setpoints can cause derivative problems along manufacturing processes that are difficult to predict. In addition, some process variables are difficult to measure. Neural networks are
Accurately model measured variables and predict unmeasured variables for use in estimating the impact of setpoint changes on the manufacturing process to support the manufacturing process in a manufacturing facility. For example, the neural network model 22 for the boiler 12 can predict changes in temperature and pressure associated with changes in electrical manufacturing settings. Similarly, the neural network model engine 24 for the turbine 14 uses the turbine R associated with the change in steam pressure setting.
A change in PM can be predicted. Through communication within the manufacturing facility 10 in co-operation with the distributed control system, the neural network model 22 and the neural network model 24 optimize their settings over time to achieve desired manufacturing results. Can be. The neural network model is
Unmeasured variables, such as fuel consumption, can also be predicted to provide an estimate of manufacturing costs.

First, to establish a neural network model to control the manufacturing process
Is the construction of a model. The "Soft Sensor" (trade name) sold by Pavilion Technology enables the development of neural network models of manufacturing processes with multiple technologies. For example, a neural network model can be trained using historical data gathered from the manufacturing process. Alternatively, the neural network model is
By training the neural network from the data generated by the first principle model. Once the neural network model is trained,
This is saved as a file and transferred to the computing platform associated with the manufacturing facility.

The neural network model is based on the Pavilion Data Interface (PDI)
Interface to the distributed control system via an application programming interface 26 such as The pavilion data interface couples the neural network model to a distributed control system to provide the neural network with measured variables of the manufacturing process. The Pavilion Uptime Application Engine (RAE) accepts the measured variables of the process online from the distributed control system and processes the data to provide predicted values and desired settings to the distributed control system.

The uptime application engine can utilize the pavilion's dataflow architecture 28 to provide the forecast values. Pavilion's data flow architecture 28 is described in "Method and A" by Godbale et al.
Further details are described in US Pat. No. 5,768,475 entitled "Pparatus for Automatically Constructing a Data Flow Architecture". In summary,
The uptime application engine utilizes the neural network model to accept the measured variables of the process in the form of a temporary flat table and to provide the desired predictions. For example, the uptime application engine:
Calculate the desired value to be used from the row of the flat table and re-enter the calculated value in a new row of the table. The uptime application engine can interface with another facility uptime application engine to generate forecasts and populate tables. Thus, for example, the uptime application engine of the neural network model 22 for the boiler 12 accepts variables measured from the boiler decentralized control system 16 and the neural network model 24 for the turbine 14 to predict electrical production levels. The uptime application engine returns data, such as setpoint data, to the distributed control system to achieve the desired process results. This process is repeated at predetermined time intervals, allowing the neural network model to be controlled via a distributed control system.

Once a high fidelity network model of a manufacturing process operation unit has been developed, another pavilion tool is available to use the model to optimize and control unit operation. For example, the pavilion “Process Insight, BOOST,
Process Perfecter products leverage economical process and constraint information,
Enables optimization and control of unit operation. In this way, the neural network model provides a powerful tool for enhancing control at the facility level.

The data available from the neural network model also provides important process information for determining the management level for the equipment, assuming the relevant data is available at the required management level in a timely manner. The present invention provides a two-tier system for obtaining neural network data from a location remote from the facility,
The system is used to allow decision makers to avoid cumbersome operations on the equipment. Communication server 30 interfaces to neural network models 22 and 24 via a network 32, such as an intranet or the Internet. The communication server 30 can interface to another equipment data source 33, such as information generated in Excel, SAP and ODBC formats, if needed. The communication server 30 mediates requests for information from the associated neural network model, and depends on the supply capacity to supply one or more predetermined neural network model data, from one or more of the equipment 10. Perform a "pick and copy" operation that transfers the current state of a given neural network model. The one or more neural network models received by the communication server 30 are effectively disconnected from the facility 10 and frozen at the current time to maintain its configuration in its actual real-time state.

The communication server 30 can interface with the offline model engine 34 and forward the one or more neural network models received from the facility 10 to the offline model engine 34. Once the communication server 30 receives the neural network model, the communication server 30 provides a unified virtual plant tool for analyzing the manufacturing process of the facility 10. Communication server 3
0 acts as a central base for modeling, simulation, model predictive control and optimization. The communication server 30 can mediate requests for these functions to the equipment neural network model as soon as it becomes available,
Alternatively, the off-line model engine 34 can be utilized to perform functions with real-time data received from the facility 10.

Once the model for predicting the equipment manufacturing process is completed and transferred to the customer server architecture supported by the communication server 30, the KPRA generated by the model can be made available to the manager. , A plant manager or operator can have a real-time indicator of how the equipment is operating. For example, real-time KPRA can provide key information regarding manufacturing speed, quality and restrictive constraints. Furthermore, once this information is displayable, it can be used to perform equipment operations in a more informative manner. Pavilion multi-unit optimization techniques can utilize real-time pricing information, contract information, operating objectives and constraints, and environmental costs or constraints to optimize overall equipment operation. Multi-unit optimization is available online via the communication server 30 to optimize the operation of each unit and the combined units. The multi-unit optimizer calculates the most useful setpoints and sends the setpoints to the open / closed loop controller / optimizer of each individual unit to perform the setpoint change.

The off-line model engine 34 supports optimization analysis by receiving changes in operation settings and destinations to operate “what-if”, as well as various constraints, priorities and unit structures and states. Supports set point analysis for changes in Based on this analysis, the user can modify the settings provided by the multi-unit optimization and execute them at the facility. As an example, an analysis of the total fuel cost can include not only the price of the fuel, but also fuel efficiency, fuel efficiency based on the rate of heat consumption for the unit burning the fuel, emissions from burning the fuel, and burner sludge deposition and catalyst degradation. As well as costs associated with equipment degradation caused by fuel combustion. Off-line engines support modeling that includes such fuel cost factors to aid in fuel purchase and sale decisions. The communication server can receive input about these values from the user or utilize tools such as an Internet web browser to automatically update the required information in real time from outside the manufacturing facility .

The communication server 30 is connected to the graphical user interface 3 from the manufacturing facility 10 via a network 36 such as an intranet or the Internet.
8, the performance data is transferred. Graphical user interface 3
8 provides a point-and-click environment that allows real-time performance data for the manufacturing facility 10 to be viewed from a distance. For example, by using TCP / IP and HTML pages transferred over the Internet, administrators at headquarters facilities can view real-time manufacturing facility performance data for facilities located around the world.

By selecting the KPRA option in the graphical user interface 38, a display such as the display shown by FIG. 2 provides KPRA performance data for viewing. The KPRA column 40 typically includes a number of performance measurements associated with a cogeneration facility, but calculated as KPRA using real-time data received from the communication server 30. For example, as described in description line 42, the graphical user interface (GUI) 38 illustrated by FIG. 2 provides information such as total power generation, which includes measured variables and utilization factors of the process. , Can be a Solomon index, usually calculated using historical accounting data. KP shown in column 44
RA values can result from directly downloading variables measured in real time of a process, variables of a process calculated by a neural network model, or KPRA calculated by a neural network. The performance data shown in the column 44 indicates that the communication server 3
0 to required by GUI 38. The communication server 30 obtains real-time data by mediating with the neural network model of the facility 10. Communication server 30
Can obtain the real-time data and provide the data directly to GUI 38 and offline model engine 34. Subsequent requests for the same data by the GUI 38 will result in updates being made directly from the neural network model of the facility 10 or the predicted values being generated by the offline model engine 34.

A given manufacturing facility may have multiple unit operations, each with performance data including KPRA. However, the decision maker may be interested in only a limited number of performance data. To allow the user to display only desired information, an index addition / deletion selection 46 is available. Selection of the index add / delete option 46 provides the user with the display shown in FIG. The user selects KPRA from the selection window 48, and then
Choose add or delete as appropriate. Once a user has identified the KPRAs of interest, they can save these KPRAs so that this information can be recalled repeatedly. Thus, many users can identify the KPRA on an individual basis and store the individual configuration in the communication server 30 or a computing platform associated with the GUI 38 for later reference. The user can also customize the KPRA by selecting the create new KPRA option 50 or by selecting the new data source option 52. New KPRA calculations are facilitated by the Pavilion's data flow architecture and transformation calculation engine.

The graphical user interface 38 also cooperates with the communication server 30 to
Provide detailed historical data. By selecting a specific KPRA such as the total power generation and then selecting the chart option 54, the chart display of the history data of the selected KPRA becomes available. For example, referring now to FIG. 4, a chart display of the total power generation for the last 24 hours is shown. Further historical data is available by scrolling the window to the left. The status window 56 provides information about unit operation related to power generation at the plant site (location) XYZ.
By selecting the "about" option 58, the user obtains updated information about the unit operation shown in the status window 56, as shown in FIG. Alternatively, the user can receive a top view of the unit operation, as shown in FIG.

The client server architecture supported by communication server 30 allows real-time performance data to be presented on graphical user interface 38. However, the flat table data format typically used in neural network models is generally not well suited for publishing and subscribing purposes. For large projects with multiple unit operations, such a flat table format is difficult to configure. Further, the flat table format can in some cases introduce defects in the neural network output.

Referring now to FIG. 7, a hierarchical neural network model configuration is illustrated that enhances the issuance and subscription of data from the neural network model to other models or communication servers. The system neural network 60 includes a unit A neural network model 62. Unit A model 62 includes a series of transformations A, B, C, and E connected together by wires. Data flows from the distributed control system 64 to the conversion A for processing, and processed data flows from the conversion A to the conversion B for processing. The feedback loop converts the conversion B to the conversion E, the conversion C, and the conversion A.
This is indicated by data flowing through. Transform C also receives input from dependent variable 66. The transformations included in the unit A model 62 together convert the data into the system model 6
It operates as a composite transform provided to a transform F of 0. The conversion F also provides the processed data to the distributed control system 64.

The object-oriented box approach shown in FIG. 7 offers significant advantages for data publishing and subscription as compared to the flat-table approach of the runtime application engine. For example, the box approach provides a hierarchical structure with an identifiable subset of data that enhances the ability to browse models for specific data. One example is the data name 68 system. unit. A. B. x
(System.UnitA.Bx). The data name 68 is a unit A
Identify data X that is the result from conversion B in 62. Contrary to table location, the availability of data names greatly simplifies the publication and subscription of data.

Referring now to FIG. 8, there is shown a neural network model 70 in communication with a distributed control system historian 72 via a pavilion data interface 74. Interface 74 provides distributed control system tag 76 for model 70. Tag 76 is processed by transformations C, A, and B. Conversion C outputs data Z, conversion A outputs data Y, and conversion B outputs data X. Thus, model 70 involves the exchange of pre-configured variables for a specified distributed control system. Data monitoring client 78 retrieves data X, Y, and Z by interactively subscribing to variables in model 70 or directly querying variables in model engine 70. The object-oriented box architecture enhances interactive browsing by placing variables in a browsable, hierarchical namespace.

Referring now to FIG. 9, an airflow filter object-oriented wrapper 80 provides one example of how a transformation models a processing element. Airflow filter transform 80 models the filter airflow as an exponential moving average. The input flow 1, flow 2, flow 3, and CF to the transform 80 can include other neural network objects, empirical models, first principles models, signal processing, sensor validation, and pavilion data interfaces, OPC , Or from multiple sources, including direct data access, such as ODBC. Transform 80 processes the input and generates a filtered flo
w) 1, filtered flow 2, and filtered flow
flow) 3 output is provided. Transform 80 provides a unified model for the airflow filter that can be reused in a number of different neural network models for a given installation using a number of different airflow filters. Similarly, a generic object-oriented wrapper can model other elements in the facility. An object-oriented box approach that uses a generic object-oriented wrapper for element processing
Provides a flexible architecture that allows fast modeling of related unit operations.

Next, with reference to FIG. 10, the object-oriented box transform 82 includes an algebraic transform 8
4 and the exponential transformation 86. Transforms 84 and 86 illustrate the flexibility of the box architecture for online and offline solutions. In the online neural network model, since the data for each of the object-oriented wrappers in the transforms is already available for processing, these transforms do not need to refer to tabulated historical data; Perform a satisfactory calculation. These transformations can be easily transferred to the offline engine along with the current state data for further analysis without having to recall the data from the tabular format.

Referring now to FIG. 11, to provide setpoint and what if analysis, transform 8
8 is shown. Setpoint analysis is achieved by selecting a target and / or range limit, solving the transformation, and determining optimal setpoints to achieve the target. What-if analysis is accomplished by setting a set point and determining the result provided by the set point at the desired reference. What-if analysis is a mixed-in optimization where the on / off decision is combined with continuous parameter optimization.
teger optimization).

Having described the invention in detail, various changes, substitutions, and alterations may be made to the invention without departing from the spirit and scope of the invention as defined by the appended claims. I want to be understood.

[Brief description of the drawings]

FIG. 1 is a block diagram of a manufacturing process neural network control system that interfaces with a graphical user interface.

FIG. 2 is a diagram showing a graphical user interface display of a key performance real-time analyzer.

FIG. 3 shows a graphical user interface for selecting a key performance real-time analyzer.

FIG. 4 is a chart diagram of a graphical user interface showing a temporary view of the key performance real-time analyzer.

FIG. 5 is a diagram showing a graphical user interface display of selected manufacturing equipment information.

FIG. 6 is a plan view of a graphical user interface of a manufacturing facility.

FIG. 7 is a block diagram of a neural network model.

FIG. 8 is a block diagram showing a neural network building block and a data storage table.

FIG. 9 is a generic object oriented wrapper for modeling processing elements.

FIG. 10 is a block diagram of neural network processing of algebraic and iterative simulations.

FIG. 11 is a block diagram of a neural network model for set value and what-if analysis.

[Procedure amendment]

[Submission date] June 23, 2000 (2000.6.23)

[Procedure amendment 1]

[Document name to be amended] Drawing

[Correction target item name] All figures

[Correction method] Change

[Correction contents]

FIG.

FIG. 2

FIG. 3

FIG. 4

FIG. 5

FIG. 6

FIG. 7

FIG. 8

FIG. 9

FIG. 10

FIG. 11

──────────────────────────────────────────────────続 き Continuation of front page (81) Designated country EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, IT, LU, MC, NL, PT, SE ), OA (BF, BJ, CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, KE, LS, MW, SD, SL, SZ, TZ, UG, ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, CA, CH, CN, CR, CU, CZ, DE, DK, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID, IL , IN, IS, JP, KE, KG, KP, KR, KZ, LC, LK, LR, LS, LT, LU, LV, MD, MG, MK, MN, MW, MX, NO, NZ, PL, PT, RO, RU, SD, SE, SG, SI, SK, SL, TJ, TM, TR, TT, UA, UG, UZ, VN, YU, ZA, ZW (71) Applicant Suite 700, 11100 Metric Boulevard, Austin, TX 78758, U.S. Pat. S. A. (72) Inventor Plummer, Edward S. United States, Texas, Georgetown, River Road 120 (72) Inventor Ellinger, Joshua Brennan United States, Texas, Austin, Waterston Avenue 1622 F-term ) 3C100 AA57 AA59 BB02 CC08 5H223 AA01 AA05 AA11 BB01 BB09 CC08 DD03 DD05 DD09 EE06 EE30 FF05 FF06

Claims (15)

[Claims] 1. A method for presenting performance data of a manufacturing process performed in a manufacturing facility, comprising: generating at least one model of the manufacturing process; and interfacing the model with the manufacturing facility. Supplying process performance data to the model; predicting process performance data using the model; transferring the predicted performance data from the model via a computer network and displaying on a graphical user interface A method that includes and. 2. The method of claim 1, wherein said model comprises a neural network. 3. The method of claim 1, wherein the model transfers real-time predicted performance data to the graphical user interface. 4. The method of claim 1, wherein the model transfers future predicted performance data to the graphical user interface. 5. The method of claim 1, wherein the model transfers real-time performance measurement data to the graphical user interface. 6. The method of claim 5, wherein the performance measurement data comprises at least one key performance real-time analyzer. 7. The method of claim 6, wherein the key performance real-time analyzer includes a Solomon utilization index. 8. Transferring the model and the performance data to an offline engine; accepting a change in performance data from a graphical user interface; transferring the change in performance data to the offline engine; The method of claim 1, further comprising: determining predicted performance data; transferring the predicted performance data and displaying the data on a graphical user interface. 9. The method of claim 8, wherein the change in the performance data includes a change in a process setting, and the predicted performance data includes a predicted performance measurement associated with the change in the process setting. 10. The method of claim 8, wherein the change in the performance data comprises a change in the performance measure and the predicted performance data comprises a predictive process setting to achieve the change in the performance measure. Method. 11. The method of claim 10, further comprising transferring the predicted process settings to a neural network model and performing the steps in a manufacturing process. 12. The method of claim 1, further comprising the step of selectively adding or removing KPRAs and displaying them on a graphical user interface for each individual user.
The described method. 13. A system for presenting performance data of a production facility unit operation, the plurality of control systems being associated with at least one production facility unit operation and maintaining a measured variable setting of the unit operation. A plurality of models each interfaced with at least one control system to receive measured variables of the process from the control system; a communication server interfaced with the models to receive real-time performance data from the models; A graphical user interface interfaced with the communication server via a network to display real-time performance data to a user. 14. The system of claim 13, wherein the at least one model includes a plurality of object-oriented box transforms. 15. An offline model engine interfaced with the communication server and receiving a model and associated performance data from the communication server, the offline model engine analyzing a response of the model to changes in the performance data. The system according to claim 14.