466909 Improving Industrial Polyethylene Production Via Data Analysis

Wednesday, November 16, 2016: 8:30 AM
Monterey II (Hotel Nikko San Francisco)
Mohsen Nikkhoo, Process Technology, SABIC, Sugar Land, TX, Job D. Guzman, Process Technology, SABIC, Geleen, Netherlands and Francesco Bertola, Technology PE, SABIC, Geleen, Netherlands

Plant historians accumulate extensive data about plant operation and performance. Hidden in these data are possible solutions to chronical problems such as, polymer fouling, production bottlenecks, and product quality variations. Process engineers routinely analyze historian data in search of solutions, but the sheer amount of data and the complexity of industrial plants conspire to keep these solutions hidden from engineers using conventional data-analysis tools. Small data sets miss recurrent root causes. Plant disturbances obscure steady-state trends. Instrument errors propagate through data manipulations.

Here we give examples where data mining and rigorous statically analyses have uncovered simple solutions to real plant problems. We take advantage of specialized software to handle large amounts of data, covering a comprehensive set of plant variables and several years of plant operation. In addition, we place special emphasis on data reconciliation: we use noise reduction techniques and outlier analysis to refine data points before mining them for root causes. This approach has been successful in identifying simple solutions that could have been identified by some other means, but simply weren’t.


Extended Abstract: File Not Uploaded