Oil & Gas

AI enabled Prescriptive Analytics: Recommendation of the Optimal Coil-Out-Temperature

Business Context

Optimization of even a single critical KOP (Key Operating Parameter) has significant impact for manufacturers, as will be shown in this case study – it highlights the bottom-line impact of optimizing a single petroleum refining KOP widely known as the coil-out-temperature. There are thousands of such parameters & KPIs/KOPs in any industrial operation where AI/ML-driven optimization can translate to substantial economic outcomes.

The use-case : Optimal Coil-Out-Temperature

One of the first steps in the petroleum refining process is the separation of crude oil into various fractions or "cuts" by the process of distillation in a Crude oil Distillation Unit (CDU).  

A crude mix, after getting desalted, is passed into the pre-heat furnace to be heated at an optimum temperature before being transferred to the CDU (Crude Distillation Unit) for atmospheric distillation. The output temperature of the furnace is called the coil out temperature (COT).

The Needs:

Yield Optimization with minimal processing costs :

The optimal value of COT depends on various factors, such as the characteristics of the crude oil, the desired product yields/cut (e.g., kerosene, diesel, fuel oil, etc.), and associated strategies such as "maximize middle distillates (e.g., diesel)," and so on.

Both underheating and overheating have costs and risks associated with them –  such as:

  1. underheating and leaving too much diesel at the bottom for secondary recovery in the VDU (Vacuum Distillation Unit) and consequent extra cost, or,  
  1. overheating leading to the mixing of some heavy distillates into the diesel leading to quality issues and consequent costs, revenue loss, etc.  

Therefore, the refiners want to reach the optimal temperature as fast as possible without spoiling the broth. This minimizes secondary recovery (at the VDU – Vacuum Distillation Unit) and thus costs.

This refinery processes a new crude mix every 48 hours, and the COT needs to be re-setup for each one. To avoid undesirable thermal cracking (overheating), the operators cannot set this COT in one go. It takes at least 8 – 12 hours to arrive at the optimal COT. It is a manual iterative process that involves sample testing and adjustments in small increments – a process known as trimming. Trimming involves monitoring various parameters such as Diesel Draw Temperature, Quality, Atmospheric Residue, Cumulative Yield, Over-flash ratio, and Kerosene Draw while slowly adjusting the COT. The process depends heavily on the expertise of the operator and the process engineer and necessitates frequent sample testing.

The objective of the customer was to ensure that the operator should be able to arrive at the optimal COT within 4 hours.

Other concerns regarding data-driven solutions:

This refiner, like most manufacturers, had valid concerns about IP rights. They could not just move all data to a cloud or cloud SaaS, especially since the use case involved various process-related data points, proprietary data such as “crude-assay” etc.

The IP concerns of the refiner could be defined in two segments:  

Data IP: How much data should we allow beyond the company's firewall (including private cloud)?

Process IP: How do I protect my process IP (domain expertise and specialized algorithms)?

The Solution – The UNLSH Platform & COT Advisor

The UNLSH Platform Architecture – the differentiator

Concerns regarding data ownership & intellectual property rights are major roadblocks to the large-scale adoption of analytics. UNLSH uniquely addresses the issue through its DataOps architecture – it is the key differentiator that allowed the refiner to address both their mission-critical need as well as the concerns regarding the IP issues.

UNLSH was already powering several other use cases for the refiner. The platform enables the ingestion of all IT & OT data, continuous "change-captures, harmonization, homogenization, and modelling to provide a common & contextualized data feed for all analytics use-cases. This eliminates the possibility of "disparate analytics" being built on top of "disparate data."  

Moreover, the platform provides granular & absolute control over the movement and hosting of the data and its analytical derivatives.  

This enabled two things:

  1. Only the necessary data had to be fed to the cloud for advanced analytics.
  1. The domain expert & data scientist roles (which would lay the foundation for any custom analytics unique to the customer) could remain within the customer organization. At the same time, the data and ML engineering could be handled by the vendor (dDriven in this case).  

The UNLSH COT Advisor Application

The UNLSH COT advisor takes into account all of the relevant parameters such as Crude Properties (Assay), Crude Mix (Recipe), Historical Crude Runs, Reference Parameters (Quality, Diesel Draw Temperature, etc.), laboratory test results, and time-series data (process tags). It churns the data and prescribes the optimal setup parameters (COT) under given constraints & objectives.

It helps process engineers & operators to set up faster and reduces the risk of the wrong setup. Whenever an upcoming crude change/run is sensed, it recommends an optimal COT for the crude mix. Also, it offers all supporting information, including the optimal value of the reference parameters.

It helps to reduce the setup time by 50% - 70%%, saving a lot of energy and other operations costs.  


Reduction in COT discovery time to 3-4 Hours (an 80% Reduction).
Estimated Direct Annual savings of USD 300 – 500K from reduced secondary processing costs (energy) from just this one Key Operating Parameter.

Stay updated
about our blogs, case studies and events

Connect with dDriven