Effective approaches for prognostics

Abstract

Today’s manufacturing industries aims at reducing time and cost for maintenance of the products. As far as data driven approaches are considered for prognostics of life of the tool component, different prognostics models have been proposed so far. Although these models have aided the advancement of the discipline, they have made only a limited contribution to developing an effective machinery health prognostic system. One major cost incurred in production industry is the tool cost. In the past and even currently, continuous efforts have been made in this field. Prognostic which refers to determining the remaining useful life of the component play a significant role in bringing down the tool cost. Two most widely used approaches in this direction are the analytical physics based model approach and data based approach. Former lacks accuracy due to the number of assumptions and latter approach varies widely when switching from one model to another or when there is a certain variation in dataset which is always a case in a real world problem. Recent advances in condition monitoring technologies have given rise to many prognostic models for forecasting machinery health based on condition data. So there lies a need for more robust and reliable model for prognostics of machine tool component. Most of the prognostic models are not intuitive as far as their computational strategy is considered. The set of logics and rules followed by these models are intuitive only to a lower dimensionally cases like 2-dimentional cases. As we switch to higher dimensional problem the representation of the problem becomes difficult hence it gets less intuitive. One cannot wonder the logics behind the classification or regression results proposed by a machine learning models based on a given dataset but it is believed they often provide accurate results. There always lies little insecurity about the results proposed by given prognostic model. To deal with uncertainty in the results proposed by the model we have defined a novel approach so that we can get more intuitive results in the prognostics for tool failure life. This work presents a novel approach in addressing the above-mentioned challenges. In our present novel approach, we have tried many machine learning models on our dataset to predict the life of the tool. Also, we have tried to devise a novel approach which involves the combination of different machine learning approaches into the single model of machine learning.The new novel approach proposed tries to add more confidence to the prediction compared to the traditional approaches. .