Oil condition monitoring, an AI application study using the Classification Learner Technics

Abstract

Abstract The present study uses an estimation and prediction method and characteristic techniques to predict the life of the lubricating oil based on data collected directly from the mechanical or hydraulic system. The collected data is part of a complex data set with 19 lubricating oil status parameters resulting from online measurements on an experiment stand built and operated under conditions similar to those in a mechanical machining company. The data set was collected during six months, continuously validating the data in several 258646 instances for 19 operating parameters. To predict the values of the next steps of a sequence, the Classification Learner Technics has been approached by support vector machines (SVM) models. The answers obtained characterize and equate the training sequences with values changed by a step of the time, and this means that at each stage of the input sequence, the data structure learns to predict the output value at the next time step. To prevent divergence of the forecast, it was necessary to standardize the training data so that it will achieve a zero mean and unit variance. Also, the test data set has been normalized in the same way as training data.