Abstract

Data-driven smart sensors are being widely used in industrial applications to estimate and evaluate the quality of critical variables. By using physical devices, most of the critical variables are being measured with difficulties. When process data is discarded, The quality variables sampling rate of majority of the smart sensors have been developed on labeled and labeled number of samples, small and large. The prediction accuracy enhancement quality will limit the large loss of information being generated from a measuring device. However, utilizing all available process data contained information, is a major and common issue of data-driven smart sensor. In this article, a new Multi-Layer Learning Machine (MLLM) approach is recommended for the applications of smart sensors Using the Extreme Learning Machine as a foundation (ELM). Initially, a semi-supervised auto-encoders deep network structure is being used as an extraction for unsupervised feature with the reference to all process samples. At that point, extreme learning machine is used for regression with the quality variable added. In the meantime, the manifold regularization technique is presented for MLLM. The new strategy can deeply separate, and information is extracted from the data contains, and get more data from the unlabeled samples. The proposed MLLM procedure is implemented in a debutanizer column application to assess the C4 contents. Finally, the simulation result verify that our approach enhances both of the expectation and prediction accuracy in comparison with the available methods.

Statistics