Semi-Supervised Learning of Soft Sensor Model for Multi-Grade Process Quality Prediction Using Multi-Source Data Transfer | AIChE

Semi-Supervised Learning of Soft Sensor Model for Multi-Grade Process Quality Prediction Using Multi-Source Data Transfer

Authors 

Lik Teck Chan, L. - Presenter, Chung-Yuan Christian University
Chen, J., Chung-Yuan Christian University
Starting up and sustaining production levels consistently from an ethylene plant does not always match expectations. Maintaining and managing furnace operations, the Cracked Gas compressors, Transfer Line Exchangers, chiller health and the distillation operations in the downstream cold section, to get the best Overall Asset Effectiveness (OAE), is a constant challenge. OAE is a combination of Throughput, Availability and Yield. Any one of these three parameters could impact performance of the plant. Big Data Analytics can provide insights and prescriptions that help balance these three parameters to sustainably maximize the OAE.

Data availability is not a constraint, but utilization of the available data is. Data capable of providing insights is available from diverse sources like, the DCS, operator logs, laboratory, maintenance and engineering documents. Traditionally about 90% of the collected data is archived and not even looked at. More complete utilization of this data, utilizing Big Data Analytics techniques modified to suit the process manufacturing application, yields valuable insights to help improve the overall asset effectiveness as well as individual equipment effectiveness.

While several Ethylene plants have used subsets of these big data analytical techniques quite successfully to enhance OAE, the full application of the big data techniques tailored to process manufacturing allows improved OAE levels beyond the current paradigms.

Successful implementations have shown that Big Data Analytics in Process Manufacturing is not just an off the shelf tool that can be picked up and applied. Success of the analytics initiative for Chemical Process Manufacturing entails several aspects, to ensure effective implementation and sustenance. Effective data handling, a synergistic use of Machine Learning and fundamental models for Descriptive, Predictive, Prescriptive and Diagnostic applications, an effective business intelligence interface to provide readily actionable visual information based on a solid data and analysis foundation and a work process are all important aspects to be considered.

Data quality issues and skills availability are two of the top challenges organizations typically face when trying to implement analytics systems. The paper shows how these challenges were successfully addressed using combined models based analysis and the Internet of Things (IoT).

This paper showcases ethylene facilities where a Big Data Analytics initiative has been successfully implemented at the process manufacturing level to improve OAE. The paper also covers the key ingredients and success factors based on lessons learnt from these applications.