A microsensor array for diesel engine lubricant monitoring using deep learning with stochastic global optimization

Abstract

Measurement of concentrations of the critical properties in a diesel powered machine’s lubricant oil is an important task in preventing machine failure and excessive damage. While there are several existing sensors for detecting these properties individually, they suffer from cross sensitivity issues and tedious calibrations for varying operating temperatures. We developed an interdigital sensor array based on an artificial neural network (ANN) automatically tuned with a stochastic global optimization (SGO) method for measuring water, base, soot, and diesel fuel contaminant concentrations in a lubricant oil. The temperature effect was compensated with the neural network. The neural network architecture was automatically selected through a unique simulated annealing process which resulted in an increased prediction accuracy when compared to the ANN with traditionally selected architecture. Dropout and data augmentation techniques were used during training to prevent overfitting. Experiment results demonstrated the ANN’s ability in accurately determining the oil properties from the overlapped sensor responses as well as removing the need to calibrate for a variety of operating temperatures. The sensor array is able to provide comprehensive information about a diesel powered machine’s health status.