Detecting cooling coil fouling automatically. 2. Results using a multilayer perceptron.

Author(s) : VERONICA D. A.

Type of article: Article

Summary

A new concept to automatically detect cooling coil fouling is introduced in a companion paper (Veronica 2010). The concept maintains a data-driven dynamic model of the cooling coil on-line with coil instruments, so all steady and transient coil behaviour is assimilated. The fouling surveillance is to periodically apply a specific transient sequence (the query) to this model off-line as a surrogate for applying such a test to the real coil. A query sum, calculated from the transient response of the model to a query, is used to detect changes in the real coil, to discern fouling from other faults such as instrument drift, and to further discern water-side from air-side fouling. A crucial element to viability of the concept is that the data-driven coil model respond to queries dynamically and recursively, with sufficient stability and accuracy so that fouling can be discerned. Yet, the model should be of reasonably simple architecture, permitting a plausible embedded implementation. Based on trials using simulator data, a relatively simple, three-node, multilayered perceptron demonstrates sufficient accuracy and stability so that the concept can be regarded as viable and as warranting further development. Also presented are the factors found to be critical for dynamically employing a data-driven model of any architecture to detect process faults.

Details

  • Original title: Detecting cooling coil fouling automatically. 2. Results using a multilayer perceptron.
  • Record ID : 30002577
  • Languages: English
  • Source: HVAC&R Research - vol. 16 - n. 5
  • Publication date: 2010/09
  • DOI: http://dx.doi.org/10.1080/10789669.2010.10390923

Links


See other articles in this issue (3)
See the source