Comparative study between ANN and regression model in prognostics

Research in prohnostics

Abstract:

Data driven prognostics employs many types of

algorithms some are statics and other are dynamics. Dynamic

complex engineering systems such as automobiles, aircraft, and

spacecraft require dynamic data modeling which is very efficient

to represent time series data. Dynamic models are complex and

increase computational demands. In previous work performed by

the author, linear regression model is provided to estimate the

remaining useful life left of an aircraft turbofan engines and

overcome the complexity of using dynamic models. It was simple

and efficient but it had some drawbacks and limitations. The

same task is resolved again here using multilayer perceptron

neural network (MLP NN). Results show that MLP NN as a static

network is extensively superior to linear regression model and

does not involve the complexity of dynamic models. Phm08

challenge data are used for algorithms training and testing. The

final score obtained from MLP NN can be placed in the fifteenth

position of the top 20 scores as published on the official site of the

Phm08