STATISTICA's Exploratory Data Analysis tools helped us identify that environmental conditions at one plant were contributing to process variation. An overall reduction in variation is the end result.
Robert Breyer, R&D Group, Georgia-Pacific Resins, Inc
STATISTICA Process Optimization is a specialized add-on to STATISTICA Data Miner and it offers a powerful software solution designed to analyze, monitor, and optimize complex processes based on all available data. The products combine the most powerful tools for monitoring quality and other characteristics of a process with cutting-edge data mining technology and process optimization capabilities.
STATISTICA Process Optimization is applicable to any situation where (often a large amount of) data are available describing a complex organizational, service-delivery, manufacturing, or continuous or batch manufacturing environment.
- Typical Areas of Application
- A Complete Solution
- Integration with the STATISTICA System
- Go Beyond Simple Process Monitoring and Single-Factor Optimization
Typical Areas of Application
Using years' worth of past data collected for a manufacturing process, STATISTICA Process Optimization can find trends that reoccur over time. These trends are then used to predict means, minimums, maximums, and ranges for samples not yet created. Understanding the sample trends and forecasting future samples proves invaluable to the manufacturing process.
In a process with hundreds or thousands of process inputs, all potentially affecting the final product, STATISTICA Process Optimization will determine a subset of those predictors with the most influence. This allows us to focus on a hand full of important parameters of the complex process while at the same time, giving greater ability to influence the final product.
A Complete Solution
The tools available in STATISTICA Process Optimization, and the supporting services available through StatSoft's specialized consulting practice world-wide, will allow StatSoft clients to apply the cutting edge technologies required in today's challenging business environment, in order to boost competitiveness and achieve superior return on investment for sustained success. This specialized add-on to STATISTICA Data Miner and other tools is offered either (a) as a complete "turn-key" (deployed) solution, custom tailored by StatSoft consultants and engineers to your specific needs, and/or (b) as a comprehensive set of tools that enables you to easily build new or customize existing solutions.
STATISTICA Process Optimization provides these advanced features of the STATISTICA system:
- Methods and algorithms for modeling complex processes (e.g., for building predictive models of quality, process outcomes, key performance indicators; model-based process monitoring)
- Advanced methods for root cause analysis (e.g., to identify important process parameters from among thousands of parameters available for process monitoring)
- Optimization (e.g., optimize arbitrary cost functions based on one or more models for key performance/process outcomes)
- Simulation (e.g., simulate non-normal multivariate processes and models to identify expected performance, reliability, etc.)
STATISTICA Process Optimization integrates:
- The most complete comprehensive selection of data mining methods offered in STATISTICA Data Miner, tools for root cause analysis, STATISTICA software for quality control and improvement, Multivariate Quality Control charting, and software for uncovering trends/drift, explaining known patterns, and predicting/forecasting.
- Specialized, unique tools for multivariate process simulation.
- Specialized, unique tools for process optimization of arbitrary goal functions (and data mining models for one or more outcomes and/or cost functions).
- The complete set of superior analytic graphics and exploratory methods for drill down and problem "understanding".
Specifically, STATISTICA Process Optimization provides capabilities for:
Statistical Process Control (SPC), Multivariate SPC, Advanced Process Monitoring
- All quality control charts, multivariate quality control charts, process capability analyses, experimental design procedures, or Six Sigma methods and charts are integrated with a comprehensive library of cutting-edge techniques for exploratory and predictive data mining.
- Capabilities for connecting directly to external databases, to monitor live data streams
- Methods for standard control charting, trending, as well as multivariate charting and trending (e.g., MCUSUM, MEWMA, T²)
- All tools and methods for data exploration and review, to identify excursions, patterns, etc.
- All tools and methods for data preparation (e.g., recoding of outliers and excursions, identification and imputation of missing data, etc.)
- All methods, algorithms, and techniques available in STATISTICA Data Miner, which contains the most comprehensive collection of data mining algorithms and methods in a single package (e.g., including Automated Neural Networks, various Tree/Recursive-Partitioning algorithms, Boosted Trees and Forests, Support Vector Machine, Multivariate Adaptive Regression Splines, etc.), providing the most powerful toolset available for modeling/predicting complex process outcomes
- Optimization of multiple data mining prediction models (e.g., application of Simplex optimization, Genetic Algorithm optimization, etc., to minimize/maximize the prediction from one or more predictive data mining models)
- Generic optimization of user-defined loss/cost functions (e.g., application of Simplex optimization, Genetic Algorithm optimization, Grid search methods to optimize arbitrary goal functions defined by the user via the convenient STATISTICA Visual Basic (SVB) scripting language; thus; multiple models for multiple outcomes (complex cost functions) can be optimized simultaneously; note that STATISTICA can be integrated with the Enterprise environment, so that optimization can be offloaded to powerful 64-bit multi-processor servers to tackle problems of significant complexity)
- Distribution fitting to multivariate datasets; automatic selection of normal, non-normal, or mixtures of distributions
- Estimation of variable covariance matrices
- Advanced simulation of multivariate non-normal distributions and covariance structures, to identify expected distributions of reliability, yield, quality, success/response-rate to marketing campaign strategies, investment and portfolio strategies, etc.
Integration with the STATISTICA System
STATISTICA Process Optimization is part of the STATISTICA family of software products, and is fully integrated with all tools for enterprise deployment.
- For example, data mining prediction models can be automatically connected to STATISTICA Enterprise for model-based predictive quality control, virtual sensors, etc. (see also references and white papers listed below) deployed against live data streams.
- Data Mining projects and modeling can be off-loaded to more powerful servers running STATISTICA Enterprise Server; for example, optimization can be performed simultaneously over multiple server processors.
- Model based process monitoring models (e.g., for control of batch maturation processes, continuous processes such as power generation, chemical manufacturing, etc.) can be deployed to STATISTICA Monitoring and Alerting Server, to enable enterprise-wide comprehensive multivariate process monitoring.
Go Beyond Simple Process Monitoring and Single-Factor Optimization
StatSoft has been at the forefront of providing practical and effective solutions for advanced process monitoring and optimization worldwide for nearly two decades! For recent publications describing some of the applications and use cases of typical deployment scenarios and success stories, refer to the White Paper section, or see these recent papers:
Hill, T., Eames, R., Lahoti, S. (2008). Finding direction in chaos: Data mining methods make sense out of millions of seemingly random data points. Quality Digest, December, 20-23.
Hill, T., EPRI/StatSoft Project 44771: Statistical Use of Existing DCS Data for Process Optimization, EPRI, Palo Alto, CA, 2008). (Note: This paper is available to EPRI members. It can also be purchased. Search for paper title on http://epri.com)
Grichnik, T., Hill, T., & Seskin, M. (2006). Predicting quality outcomes through data mining. Quality Digest, September, 42-47.
Lewicki, P; Hill, T; Qazaz, C. (2007). Multivariate Quality Control. Quality Magazine, April, 38-45.