(35b) Combining Big Data Analytics and the Industrial Iot with Reliability and Predictive Models to Improve Furnace Reliability and Performance

Furnace reliability and run length play critical roles in overall ethylene plant operation strategy and consistently maintaining high production rates. This paper explains how application of big data analytics through multi‐model techniques and systematic analysis of furnace operation identifies opportunities for furnace performance, reliability and run length improvement.

Ethylene plants have common furnace issues which limit performance, of which lower than expected run length is the most prevalent. Run length issues limit plant throughput, impact product yield, and thereby reduce plant profitability. The majority of furnace issues can be addressed effectively if the abnormal deviations and anomalies are identified quickly and proficiently. This requires close monitoring of the furnace and its downstream parameters. A continuous multi-dimensional data analytics process can serve as an early warning mechanism. The complete analytics process must include appropriate data collection, information development through analytics (the core), proper visualization, prescriptive action identification, and a tracking and feedback mechanism to convert the learning to wisdom.

Data è Information è Knowledge è Wisdom

Process, equipment and compliance analytics require multiple types of model. Application of the appropriate models and technology greatly help maximize profit/production and extend furnace run lengths. Most ethylene plants try a single model approach. While this allows rudimentary control; a multiple model approach for more detailed management of furnace operation is the most effective approach to increase run length and drive production without affecting furnace reliability.

A furnace reliability matrix based on big data analytics can provide a visual, faster identification of furnace anomalies and deviations based on variations from model prediction.

Key features that can be included in a furnace reliability matrix:

• Deviations from normal in key furnace parameters can be flagged
• Key furnace operating parameters like COT, CIP, CIT, etc. can be predicted using statistical-based models/equations utilizing historical data and cross-verified by fundamental models
• As furnace feed type and run length vary the matrix parameter limits for deviations must vary
• Parameters with high level of significance and criticality can be highlighted

The information provided by this matrix formed by a combination of statistical and fundamental models identify and drill down to the cause of deviation; helping to take quicker measures to improve furnace reliability and performance. While certainly challenging to operational paradigms, such a matrix has proven to be a reliable tool to flag early warnings.

Furnace Yield models provide insights on impact of feed, cracking severity and other parameters on coking rate and production. Fundamental yield models provide insight on key operating parameters such as coil outlet temperature, coil pressure and dilution steam to hydrocarbon ratio, and allows appropriate adjustments to be made to optimize production and inhibit differential coking.

Parallel data driven machine learning models allow early warning indications and quick primary adjustments, which can be cross verified by kinetic models. Statistical techniques like Principle Component Analysis (PCA), Partial Least Squares (PLS) etc. can be applied successfully on furnaces. The PCA approach saves the effort of looking at multiple trends for multiple variables and judging faults or abnormal operation based on “mental” models. PCA/PLS analysis accurately detects abnormal deviation in furnace operating parameters and aids in deciding subsequent actions to reduce differential coking and to improve furnace reliability.

A process monitoring approach applying statistical techniques and coke prediction models in conjunction with a predictive reliability matrix identifies abnormalities and deviations in key furnace operating parameters precisely. Corrective action can then be devised by appropriate use of model results, and its impact on coking rate and other parameters can be analyzed before implementing in the field. Combining the above approach with a convection section model determines the parameter settings for maximum safe production. Combination of radiant and convection section models, aids to monitor and adjust heat flux to maximize run length and production while ensuring that tube metal temperatures are within allowable limits.

Analytics coupled with bird’s eye view dashboards, provide a daily view with context for identification of problems and enabling of action, reducing the cycle time from problem identification to solution and action. Effective multi-modeling is the key to go from data collection to transforming it into information and knowledge. Through using the models, visualization and dashboards, effective data interpretation and early problem detection can be realized and an implementation and tracking process to ensure benefit realization developed.