Inferential Measurements and Prediction
Modeling
Inferentials are a supplement for infrequently measured
qualities or critical sensors and help support environmental
compliance. Aspen IQ enables modeling and implementation of
inferred product qualities, and makes it possible to implement
linear or non-linear inferential sensors online. Inferential
sensors are fundamental elements of many advanced process control
systems. Key parameters (such as naphtha 95% point and polymer melt
index) are often inferred rather than directly measured.
Through Aspen IQ, you can model and implement inferential
sensors across your site.
Example Applications:
- Refining: Distillation,
viscosity, cloud point
- Emissions Monitoring: NO2,
CO2, particulate
- Polymers: Melt Index,
viscosity, coatability
- Food & Beverage: Food
taste, wine grade
- Steel: 30-day hardness
- Pulp and Paper: Brightness /
Kappa Index
- Semiconductors: Plasma etch
selectivity, variance, and rate
Inferential model types include FIR, PLS, fuzzy PLS, BDN, hybrid
neural net, monotonic neural net, linearized rigorous model-based,
and custom equations. Along with empirical inferentials, the Aspen
Control Platform provides the ability to incorporate Aspen's
industry-leading engineering simulation models for building
rigorous inferential applications. Aspen HYSYS®, Aspen HYSYS®
Petroleum Refining (formerly Aspen RefSYS®), Aspen Polymers
(formerly Aspen Polymers Plus), and Aspen Customer Modeler® are
supported.
Online Features
- Built-in steady-state detector
- Provides both analyzer and lab model
updating
- Wide range of DCS and information system
interfaces available
- Enabled for remote monitoring
- No code generation or programming required
Offline Features
- A broad suite of tools for developing
inferential sensors: PLS, fuzzy PLS, and neural-network and
linear-ized rigorous model-based approaches
- Graphical analysis tools for model
evaluation
- Model building tools such as variable
selection, dead-time detection, and dynamics analysis
- Prediction libraries allow for future expansion
(e.g., with process-specific models)
Data Pre-processing Features
- Handles multiple files
- Allows both graphical and numeric cutting of
bad data
- Allows interpolation for replacing bad or
missing data
- Allows averaging of training data to support
cumulative lab samples
Supported Model Types
- Linear PLS
- Fuzzy PLS
- Hybrid Neural Net (HNN)
- Linear-ized Rigorous Models
- Algebraic and FIR Models