Concluding Remarks

Data analytics has become a standard tool for reliability improvement. With sufficient data and failure events, failure signatures can be recognized much earlier than might otherwise be possible. This approach can be improved upon by principle component analysis. Multivariate analysis allows recognition of a change in variable relationships. Triggering warnings when variable relationships are not following historical convention allows an even earlier indication of failure potential versus following a single variable approach.

However, an indication that relationships between variables has changed must be understood before action can take place. The goal is to go from just having a possible indication that something has changed to being able to have predictive and prescriptive advice. One successful approach to doing this is to bring expected performance from equipment specific sources such as pump or compressor curves and domain expertise to the problem and build a model of the expected performance depending on the status of each variable. An example will be shown where an incorrect installation of the bearings on the cracked gas compressor was resulting in high bearing temperatures and variable constraints and options to keep the bearing temperature down until the next shut down were modeled to allow continued operation.

Incrementally applying data analytics to Ethylene furnace to optimize performance can similarly result in very effective tools. Simple analytics looking at feed conditions and historical best can provide insightful guidance. An example of this will be demonstrated.

Taking this concept to the next level is possible by establishing variable relationships where machine learning can be applied on both historical data and automated first principle model runs to accurately predict runlengths (i.e. coking/ CPRs / TMTs) based on feed and cracking conditions. Getting value from this requires being able to understand the implications of adjustments and being able to push furnaces to match optimized performance while staying within a decoke schedule. A demonstration of how this modeling results in a “digital twin” enabling predictive and prescriptive advice will be shown.

Data analytics greatly increases the power of fundamental models and domain expertise by enabling real time feedback to operators and engineers. When applied to Ethylene furnaces, significantly more production is possible without risking excessive coil life diminishment.