文摘
An efficient formulation of the data reconciliation problemapplied to batch reactor experimentsis proposed and discussed in comparison with other works reported onthis subject in theliterature. Under the assumption that all components in thereaction are known experimentallyand that their chemical compositions are defined, atomic balances areused as constraints inorder to correct raw data values. The formulation is based on ana posteriori simultaneoustreatment of all measurements over the whole time of experimentation.The main point ofinterest in this formulation is the preservation of the measurementidentity in each sample.Even if component holdups are systematically involved in massbalances by pairs to estimate adifference in time, here we correct each component measurement and eachsample independently.On this basis, it is shown how classical statistical tests can beused to discriminate gross errors.A case study built from simulated true and noisy data gives anoverview of the performances ofthe method. The results state that the overall time formulation ismore powerful than theclassical one, which is essentially based on a sequential treatment byperiods of time. Evaluationof the performance for three adapted statistical tests leads us toconclude that the generalizedlikelihood ratio (GLR) is not only the most appropriate but also themost efficient test.