The paper proposes an approach to the problem of characterizing the automotive systems/components defectiveness by means of forecasting methodologies based on soft computing techniques. A specific code based on ANN already used from an automotive industry is studied. The acquisition phase, the treatment and the best collection of data are analysed. Moreover, a sensitivity analysis is developed in order to find the parameters more influent on ANN response. Then an optimisation algorithm of the already existing model is carried out to find the best compromise among parameters as ANN architecture, number of neurons for input and hidden layer, learning rate, momentum etc. Then the optimized ANN is integrated with a Fuzzy logic algorithm to reduce the error of the output results. The complete model, implemented in Matlab environment, is tested and validated, with fundamental support of an automotive industry, on different categories of parts as well as electronic and mechanical components.
RELIABILITY FORECAST OF AUTOMOTIVE SYSTEMS BASED ON SOFT COMPUTING TECHNIQUES / P. CITTI; DELOGU M; FONTANA V; AMMATURO M. - STAMPA. - (2005), pp. 373-379. (Intervento presentato al convegno European Safety & Reliability Conference - ESREL 2005 tenutosi a Tri City (PL) nel 27-30 giugno).
RELIABILITY FORECAST OF AUTOMOTIVE SYSTEMS BASED ON SOFT COMPUTING TECHNIQUES
CITTI, PAOLO;DELOGU, MASSIMO;
2005
Abstract
The paper proposes an approach to the problem of characterizing the automotive systems/components defectiveness by means of forecasting methodologies based on soft computing techniques. A specific code based on ANN already used from an automotive industry is studied. The acquisition phase, the treatment and the best collection of data are analysed. Moreover, a sensitivity analysis is developed in order to find the parameters more influent on ANN response. Then an optimisation algorithm of the already existing model is carried out to find the best compromise among parameters as ANN architecture, number of neurons for input and hidden layer, learning rate, momentum etc. Then the optimized ANN is integrated with a Fuzzy logic algorithm to reduce the error of the output results. The complete model, implemented in Matlab environment, is tested and validated, with fundamental support of an automotive industry, on different categories of parts as well as electronic and mechanical components.I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.