Researcher
Research institution
Champion
Focus team
Topic
Project status
Year ended
2012
Project ID
201103
Abstract

Dr. Ricardo Dunia of the University of Texas picked up on the attempt to develop a method to identify event signatures in process data before alarms would alert the operator to a problem. Parallel coordinate analysis was applied to principal components identified in the precursor project. Data on 24 compressor surge events and eight column flooding events were modeled. High false alarm rates still plagued the results, but visualization of the upset was far superior to PCA alone.

Objective

Evaluate methods of early detection of off-normal operations

Driving questions
  • Why do equipment and process conditions leading up to an unplanned event get missed by operators?
  • Are there generalized techniques that can be used to identify precursors to abnormal situations and system behavior?
  • Which of the conditions can be detected?
  • How does the operator know what the design constraints and operating boundaries are for different operating points?
  • How can these constraints and boundaries be incorporated into displays so that conditions leading up to unplanned events are either avoided or detected by the operator prior to failure and alarm actuation?
  • How can decision aids be incorporated?
Background

To start the motivation of this work compressor surge was defined as dynamic unstable operations due to low volumetric flow given the amount of power provided to the compressor. It is interesting to highlight that optimal operations are reached closed to surge conditions.
As general physical characteristics of this phenomena, we have that flow reverse inside the compressor, the temperature of the equipment and the amount of vibrations increases considerably , and the output flow becomes pulsing.
Regarding anti-surge systems – they consist of mainly recycling the compressor outlet flow after cooling it. During anti-surge activation the compressor efficiency drops considerably for a long period of time.

Deliverables

(1) Identify techniques for showing what is normal, what is not normal.
(2) Provide identifying precursors that leads up to an unplanned event.
(3) Suggest guidelines for developing displays and navigational aids.
(4) To show how decision aids can be incorporated to support operators.