New Neural Network System Designed for Process Intensive Industries
A new neural-network system designed for process intensive industries such as pulp and paper, petroleum refining, and pharmaceutical manufacturing has been introduced by Gensym Corp. Known as NeurOn-Line Studio (NOL Studio), the system guides users step-by-step through the process of data preprocessing, model configuration, training, validation, and deployment.
Designed for maximum ease-of-use, NOL Studio's powerful visualization tools allow users to analyze large, complex data sets with more than 100,000 records and 100 variables. As a result, it enables users to build models needed for improved control and optimization of process efficiency and product quality.
Applications of NeurOn-Line include inferential measurements of product quality, model-based control, and process fault detection. NOL Studio can be used off-line, for model analysis and optimization, or online for real-time model-based diagnostics and control. Off-line, it is a desktop tool in the Windows (NT/98/95) environment for analysis of process behavior, modeling, and design optimization where typically the source of data is a data historian or spreadsheet data arrays.
Designed to be easy for process engineers to use, many technical decisions such as selection of relevant inputs, time delays, and network architecture are automated. Once a model has been built, NOL Studio allows users to discover more profitable ways to run their process through simulation and optimization.
Based on an objective function that expresses profitability in terms of predicted and measured process variables, NOL Studio applies the neural network models to determine optimal operating conditions, within stated process constraints.
Gensym began shipping NOL Studio in December. Pricing starts at $25,000/ user, with initial sales focused on minimum 10-user site packages and OEM partners.
Online applications (back to top)
For online use, there are two options. The predictive models and optimization capability can be deployed as COM objects in embedded applications. Alternatively, the models can be loaded into Gensym's G2 software, for execution in a G2-based environment, bringing the added capabilities of G2 for intelligent operations management. Typical online uses of the learned-models include inferential measurement, model-based control, and prediction of future conditions.
"For online use, the new NOL embedded deployment module can deploy neural network models as ActiveX controls in a distributed architecture, independent of Gensym's G2 software environment," according to Mark Kramer, chief technology officer of Gensym. "As a result we are bringing the power of neural network technology to a broader audience that could benefit from model-based diagnostics and control. Our goal with NOL Studio is to make it easy and economical for process engineers to deploy," he added.
NOL models can also be integrated into G2 applications that add knowledge-based components such as expert systems to create a more robust and powerful application. This deployment strategy leverages G2's extensive connectivity libraries, object orientation, ability to represent expert rules in structured natural language, and open-ended character for application enhancements.
"With NOL Studio we have extended our highly successful NeurOn-Line product family to a broader audience, who can benefit from neural network technology that is easy-to-use and deploy," said Bob Moore, Gensym president. "The new, enhanced development environment, optimization capabilities, and embedded deployment options increase NeurOn-Line's flexibility, modeling power, efficiency, and economy to help our customers improve their operating margins and enhance their product quality."
NeurOn-line features (back to top)
NeurOn-line features powerful blocks that are easily accessed through pull-down menus. These blocks are arranged in several groups:
Vector Blocks provide the means to consolidate, condition, and manipulate time series and other real-time data patterns.
Data Set Blocks provide tools for collecting, filtering, inspecting, summarizing, and archiving the incoming data.
Training Blocks provide full control over real-time network training and validation procedures.
Neural Net Blocks compute a vector of output values for each vector of input values.
More about G2 (back to top)
Gensym's G2, the company's core product, is a comprehensive, object-oriented environment for building and deploying mission-critical, intelligent applications designed to dramatically improve complex business operations. G2 enables organizations to apply knowledge to operational data in order to reach conclusions, provide advice, and execute decisions—all in real time.
G2 can follow numerous lines of reasoning based on this knowledge and analyze large amounts of data and trends concurrently. G2 maintains an understanding of the behavior of processes over time and the currency of information, both of which are important for real-time decision support and control.
Applications and products based on G2 augment G2's broad capabilities with additional functionality specific to such problems as fault diagnosis, dynamic scheduling, process design, process simulation, and network management. Such products are offered both by Gensym and Gensym's Solution Partners.
Founded in 1986, Gensym Corp. is a leading supplier of software and services for intelligent operations management. Common applications include quality management, process optimization, dynamic scheduling, network management, energy and environmental management, and process modeling and simulation.
Gensym has sold more than 10,000 product licenses to organizations in manufacturing, communications, aerospace, transportation, government, and other industries. The company currently has more than 300 commercial installations. With headquarters in Cambridge, MA, Gensym has offices in North America, Europe, Africa and the Pacific Rim.
For more information: Gensym Corp. Tel: 617-588-3327.