(169e) Improving Ethylene Furnace Performance with Big Data Analytics

Parameters around furnace performance include throughput, yield, runlength, coil life, and energy efficiency. Increasing the factors that allow maximization of any of one of these furnace performance criteria will reduce the level of performance of the other criteria. Big Data analytical techniques have been repeatedly proven to effectively find the optimal settings to meet the furnace operational objectives.

A key to successful implementation of Big Data to Ethylene furnaces is, however, to combine it with first principles reaction kinetic models. Pure empirical models work well with predicting equipment failure and thereby addressing availability issues. Availability is a variable for Ethylene furnaces to be managed. While one can argue that with sufficient data and time, empirical models can be made to replicate furnace performance over a range of feedstocks and conditions; however first principles models can also replicate performance closely with an enhanced ability to view relationships. The logical approach is to utilize the modeling approach (i.e. first principles versus empirical) that can best illustrate actual performance within the constraints of difficulty and time in building and maintaining the model. This approach can be very effectively utilized to increase production and runlength.

Adjustment variables for the furnaces include firing, draft, flow, feed balancing, and dilution steam and understanding the ultimate effect of these on coil inlet pressure, coil outlet pressure, tube metal temperature, pressure drop in TLE, firing duty, furnace efficiency, residence time, and coil outlet temperature is key to better operation of the furnace. By combining first principles modeling with predictive modeling using Big Data techniques results the ability to completely predict the performance of the furnace over the life of its run. The impact of adjustments is reflected in the performance and production of the furnace over the likewise variable runlength of the furnace. With this information furnaces can be set to optimize the production goals and targets that themselves change with feedstock variability and planned decoke scheduling.

In this presentation, models utilized will be described and their output provided. Use of this output to identify furnace anomalies at various stages in its life and the ability to take corrective actions will be shown. While dependent on feedstock availability and downstream limitations, achieving anywhere from 5% to 20% increase in production has occurred in numerous facilities.