Abstract: An on-line training neural network for process control system and method trains by retrieving training sets from the stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network.When multiple presentations are needed to effectively train, a buffer of training sets is filled-and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed.

Abstract: An on-line training neural network for controlling a process for producing a product having at least one product property that trains by retrieving training sets from a stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network. When multiple presentations are needed to effectively train, a buffer of training sets is filled and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed.

Abstract: An on-line training neural network for process control system and method trains by retrieving training sets from the stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network.When multiple presentations are needed to effectively train, a buffer of training sets is filled and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed.

Abstract: A computer neural network process measurement and control system and method uses real-time output data from a neural network to replace a sensor or laboratory input to a controller. The neural network can use readily available, inexpensive and reliable measurements from sensors as inputs, and produce predicted values of product properties as output data for input to the controller. The system and method overcome process deadtime, measurement deadtime, infrequent measurements, and measurement variability in laboratory data, thus providing improved control. An historical database can be used to provide a history of sensor and laboratory measurements to the neural network. The neural network can detect the appearance of new laboratory measurements in the history and automatically initiate retraining, on-line and in real-time. The system and method can use either a regulatory controller or a supervisory control architecture.

Abstract: An on-line process control neural network using data pointers allows the neural network to be easily configured to use data in a process control environment. The inputs, outputs, training inputs and errors can be retrieved and/or stored from any available data source without programming. The user of the neural network specifies data pointers indicating the particular computer system in which the data resides or will be stored; the type of data to be retrieved and/or stored; and the specific data value or storage location to be used. The data pointers include maximum, minimum, and maximum change limits, which can also serve as scaling limits for the neural network. Data pointers indicating time-dependent data, such as time averages, also include time boundary specifiers. The data pointers are entered by the user of the neural network using pop-up menus and by completing fields in a template. An historical database provides both a source of input data and a storage function for output and error data.

Abstract: An on-line training neural network for process control system and method trains by retrieving training sets from the stream of process data. The neural network detects the availability of new training data, and constructs a training set by retrieving the corresponding input data. The neural network is trained using the training set. Over time, many training sets are presented to the neural network. When multiple presentations are needed to effectively train, a buffer of training sets is filled and updated as new training data becomes available. The size of the buffer is selected in accordance with the training needs of the neural network. Once the buffer is full, a new training set bumps the oldest training set off the top of the buffer stack. The training sets in the buffer stack can be presented one or more times each time a new training set is constructed.

Abstract: A computer neural network regulatory process control system and method allows for the elimination of a human operator from real time control of the process. The present invention operates in three modes: training, operation (prediction), and retraining. In the training mode, training input data is produced by the control adjustment made to the process by the human operator. The neural network of the present invention is trained by producing output data using input data for prediction. The output data is compared with the training input data to produce error data, which is used to adjust the weight(s) of the neural network. When the error data is less than a preselected criterion, training has been completed. In the operation mode, the neutral network of the present invention provides output data based upon predictions using the input data. The output data is used to control a state of the process via an actuator.

Abstract: An on-line process control neural network using data pointers allows the neural network to be easily configured to use data in a process control environment. The inputs, outputs, training inputs and errors can be retrieved and/or stored from any available data source without programming. The user of the neural network specifies data pointers indicating the particular computer system in which the data resides or will be stored; the type of data to be retrieved and/or stored; and the specific data value or storage location to be used. The data pointers include maximum, minimum, and maximum change limits, which can also serve as scaling limits for the neural network. Data pointers indicating time-dependent data, such as time averages, also include time boundary specifiers. The data pointers are entered by the user of the neural network using pop-up menus and by completing fields in a template. An historical database provides both a source of input data and a storage function for output and error data.

Abstract: A neural network for adjusting a setpoint in process control replaces a human operator. The neural network operates in three modes: training, operation, and retraining. In operation, the neural network is trained using training input data along with input data. The input data is from the sensor(s) monitoring the process. The input data is used by the neural network to develop output data. The training input data are the setpoint adjustments made by a human operator. The output data is compared with the training input data to produce error data, which is used to adjust the weights of the neural network so as to train it. After training has been completed, the neural network enters the operation mode. In this mode, the present invention uses the input data to predict output data used to adjust the setpoint supplied to the regulatory controller. Thus, the operator is effectively replaced.

Abstract: A neural network/expert system process control system and method combines the decision-making capabilities of expert systems with the predictive capabilities of neural networks for improved process control. Neural networks provide predictions of measurements which are difficult to make, or supervisory or regulatory control changes which are difficult to implement using classical control techniques. Expert systems make decisions automatically based on knowledge which is well-known and can be expressed in rules or other knowledge representation forms. Sensor and laboratory data is effictively used. In one approach, the output data from the neural network can be used by the controller in controlling the process, and the expert system can make a decision using sensor or lab data to control the controller(s). In another approach, the output data of the neural network can be used by the expert system in making its decision, and control of the process carried out using lab or sensor data.

Abstract: Batch process control is improved by defining a step endpoint condition in an expert system knowledge base; using the expert system to monitor for the occurrence of the endpoint in the batch process; and triggering a change in a batch process condition when the endpoint is found. Preferably, the expert system and the batch process condition change are implemented as modules which execute under control of timing and sequencing functions in a supervisory control system, and the change affects a setpoint (or other control objective) in a continuous control system. Multiple instances of modular expert systems allow parallel process units to be easily controlled.

Abstract: An integrated system for process control in which a process supervisor procedure (which is preferably the top-level procedure) is configured as a modular software structure, with modules which can be revised by a user at any time, without significantly interrupting the operation of the process supervisor. The modular software structure can define control parameters for many process control procedures, and can retrieve data from many sources (preferably including a historical database of process data, which can provide time-stamped data). The supervisor can also call on various expert subprocedures. Preferably the expert subprocedures can also be modified by an authorized user at any time, by calling up and editing a set of natural-language rule templates which correspond to the rules being executed by the expert subprocedure.

Abstract: An integrated system for process control in which a process supervisor procedure (which is preferably the top-level procedure) is configured as a modular software structure with modules which can be revised by a user at any time, without significantly interrupting the operation of the process supervisor. Users can define or redefine modules by editing highly constrained templates. These templates use a standardized data interface (as seen by the user), which facilitates communications with an extremely wide variety of systems. The template set preferably contains highly constrained portions (which are optimized for the most common functions), and also contains pointers to user-customized functions. Thus, rapid set-up and modification are possible, but sophisticated users still have full flexibility to do customization.

Abstract: An expert system for process control, which permits the inference rules to be revised at any time without requiring the specialized skills of a "knowledge engineer". The inference rules are intitially defined by a domain expert who fills in blank fields in a set of highly constrained substantially natural-language templates. The rule set thus specified is automatically translated to define an operational expert system. Updating can be performed by a domain expert at any time: the set of templates with the data fields as originally entered is redisplayed, so that the domain expert can edit the accessible fields and then command the modified rule set to be automatically translated, to define a modified operational expert system.

Abstract: An integrated system for process control in which a process supervisor procedure (which is preferably the top-level procedure) is configured as a modular software structure, with modules which can be revised by a user at any time without significantly interrupting the operation of the process supervisor. A user can define or redefine modules by editing highly constrained templates, which preferably include module timing and sequencing options including: block becomes active if another specified block has become active; block becomes active if a new value has been entered for a specified data source; block becomes active if a specified time of inactivity has elapsed; and combinations of these.

Abstract: An integrated system for process control in which a process supervisor procedure defines parameters for one or more controller systems (or control procedures). The supervisor procedure changes control parameters only in discrete changes, and the decision to act is sufficiently constrained that every change must be a significant change. Every change is logged (or otherwise reported out to human experts). Since every change is significant, the history of changes will provide a meaningful record which can be reviewed by human experts.

Abstract: An expert system wherein the rules are of three classes: (1) retrieval rules, which each associate one of several attributes to an object in accordance with the values of inputs; (2) analysis rules, which selectively associate an attribute with an object, and which are somewhat analogous to the natural-language inference rules which would be used in communications between domain experts; and (3) action rules, which selectively carry out the output and control actuation options, based on the attributes associated with objects by the other rules.Preferably only the action rules can enable execution of an external command procedure. Preferably each of each of the action rules requires no logical operations other than a test for association between an attribute and an object. Preferably none of the action rules can associate an attribute with an object. Preferably only the retrieval rules include numeric operations.