The use of armax strategy and self organizing maps for feature extraction and clasification for fault diagnosis

Nowadays model-based fault diagnosis is restricted to a-priory knowledge of the plant model where in the ease of a mnodel-free strategy it is neeessary to have enough informnation in termns of the frequeney Tesponse of the observed plant. This approaeh presents the advantage of nsing several strategies for featnre extraction and classification to achieve pattern Tecognition based upon linear o nonlinear strategies. For instanee Prineipal Component Analysis wavelets time frequeney distributions and partial model build patamneters (like ARMAX) are teehniques feasible to extraet key ehara eteri sties fromn data either in termns of time series o elusters. However these may not be suitable for every data analysis in termns of unkn own seen arios; therefore it is neeessary to combine some of them to achieve a valid classification. In that respect, the nse of non snpervised nenral networks like ARTMAP or SOM as powerfnl classifiers to organize the data in accnrate terms as post-processing techniqnes becomes snitable in specific cases, where, the most common eharacteristies to be found in data are hard nonlinearities and a great var ation of frequeneies. Based upon these two iss two strategies are followed; ARMAX (for p pToeessing the data) and SOM (for pos-proeessing the data) both have been cho sen in terms of partial linear model representation and the related classification, where, some mnportant Testr are the Telated to inherent online eharacteristies and time var This novel strategy is validated by uising a heuristie proposal of error mneasure which is studied and implemented in order to determnine the most suitable paramneters for this sort of combination fromn both algorithmns. An imnportant iss to be taken into account is sampling to avoid quantization at fault diagnosis algorithmn. A benchmnark example with two typical faults is Teviewed and implemented in order to highlight the benefits of this novel strategy. Resnlts of this evalnation are presented in terms of several s experiments considering fault and fault-free scenarios.