文摘
A nonlinear multivariate analysis, artificial autoassociative neuralnetwork (AANN),was applied to bioprocess fault detection. In an optimalproduction process of arecombinant yeast with a temperature controllable expression system,faults in testcases with faulty temperature sensors and plasmid instability ofrecombinant cellscould be detected by the AANN. Since the raw data of measuredvariables includedhigh-frequency noise, a wavelet filter bank (WFB) was applied to noiseeliminationbefore training of the AANN. The filtering performance of the WFBwas comparedwith those of some classical first-order digital filters. Thefiltered signals at severalresolution scales by the WFB were employed as the training data of theAANN. Thecomputing time and summation of square of errors in training werecompared, andthe appropriate degree of the noise filtering and the density of thetraining data ofthe AANN were discussed. The performance of the feature capturingby the AANNwas compared with that by a linear multivariate analysis, principalcomponentanalysis. A J index defined in this paper, using inputsand outputs of the AANN,was used for fault detection successfully. The output of the firstunit of the trainedAANN functioned effectively for the discrimination of the data in theabnormal casesfrom the data in the normal cases.