Soft Sensor Recalibration

Soft Sensor Recalibration

A Review of Current Research and Industry Practice

 

Researchers:

Mr. Alain Bartels
(alain.bartels@uwa.edu.au)

Prof. Melinda Hodkiewicz

Dr. Ed Cripps

Soft Sensor..?

Soft Sensors, also known as inferential sensors and virtual sensors,  are the names given to models which have been designed to infer parameters from real-time measurable measurable variables.

Traditionally Soft Sensors are used for:

Process Monitoring:

  • Compliment on-line instrumentation
  • Monitoring process trends
  • Fault detection

Process Control

  • Develop advanced control strategies
  • Plan and schedule process operations

Off-line operation assistance

  • Diagnosis of process operations
  • Knowledge based engineering design
  • Development of plant operations

Recently, the potential of virtual sensors has been greatly expanded by developments in the statistics and computing world. Our focus is on Soft Sensing within a Bayesian framework as shown in Figure 1.

BayesianModel

Figure 1: Bayesian Model

Models currently being tested for soft sensing include:

  • Generalized Linear Models (GLMs)
  • Random Forests
  • Neuro-Fuzzy Models (NFMs)
  • General Bayesian Learning Techniques
  • Neural Networks

Concept

“If we knew what it was we were doing, it would not be called research, would it?”-Albert Einstein

Superficially the concept is simple. Apply the idea of Soft Sensors to the Sensors themselves. Allow your network of sensors to determine when recalibration of an individual member is required.

We have entered the age of BIG DATA. Companies are competing on analytics and decisions are based on evidence as required in in rapidly changing competitive environments.  Data is now a commodity and often referred to as the new oil.

Where does this data come from?
For the resource industry it’s physical sensors of all forms.

Sensors are imperfect devices. Drift and bias are ever present and varying over time . To account for the changes of long term drift and local area effects a calibration schedule needs to be determined. Currently, not taking into account complete sensor failure, this calibration schedule is determined  empirically. The DATA SCIENCE revolution has very old fashioned underpinnings.

Motivation

Figure 2: Calibration Costs

Figure 2: Calibration Costs

An improved calibration schedule will lead to lower costs and improved decision making on all organizational levels.

 

Participation

We are seeking industry involvement in this project to:

  • To identify sensors which would most benefit from a new calibration approach
  • Highlight problematic sensor applications
  • Provide data sets for us to work with
  • Allow for Distributed Control System (DCS) integration
  • Address regulatory compliance requirements

With data intensive computing being named the fourth paradigm of scientific discovery it is essential to ensure confidence in the underlying source of our data.

Acknowledgements:

This work is supported by the Australian ARC Industrial Transformation Research Hub for Offshore Floating Facilities (IH140100012).

Further Links:

Soft Sensor: Summary

Soft Sensor: Challenges

Soft Sensor: Key Figures

Soft Sensor: Research Gaps

Submit a Comment