Implementation Costs and Output Reliability on Virtual On-line Analysers

2018 
Abstract Virtual Analysers are used in industrial processes due to its capacity to replace essential equipment by mathematical models. In order to guarantee the precision of the model output, one has to assure the model input will be available and reliable, implying greater efforts to sustain a high maintainability of the measuring devices used to generate model input. Thus, there is a significant trade-off between model costs and output reliability. This work introduces a new index, called Ana-Braga-Barbosa-Adelmo (ABBA), which embodies such economic trait when selecting an empirical model for virtual analysers. Associated costs of different Partial-Least-Square (PLS) models used to estimate the content of contaminants in a trickle bed reactor are simulated through Monte Carlo assuming a Poisson distribution for instrument failure. The performance of different PLS models are analysed in light of the new proposed index, which indicates the most adequate one. Such is different from the PLS model developed based on selection of variables through Variable Importance in Projection (VIP) as the latter relies on a greater number of input variables for similar performance.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    3
    References
    0
    Citations
    NaN
    KQI
    []