[V11N1] – Online Versus Lab Analysis: Understanding Why, Where, and When


In the last article, we walked with a lab sample from collection to analysis and discussed its journey, hurdles, hardship, and happy ending in the lab. This article will take a step further and discuss and compare the online analysis versus lab analysis in a specific context to fuel blending.

Three sampling points require an analysis of qualities to accomplish an efficient and profitable blending system. We all know that a refinery makes 80-90% of products as blended products, and any inefficiencies in their manufacturing can cost money, in the range of 25-30 Million dollars, for a refinery of 100 KB/day crude capacity.

Let us look at the options available to analyze the qualities of three sources and destinations in an end-to-end fuel blending system.

1718832962552?e=1724284800&v=beta&t=1 s2GVBN Fc09VGr g ZzbvKzm7jwdryPfDJGA4zb9c

Now, let us discuss all the options to analyze or estimate qualities at three sources required for the fuel blending system.


The qualities of blend component tanks can be analyzed in the following methods and have pros and cons.

a. Lab Analysis – This method is standard for analyzing component tank quality. However, it has low investment and use of existing lab infrastructure pros. It has the disadvantages of infrequent sampling frequencies lab intensive to take samples. Lab analyses are not available in real-time and are not current with qualities, creating issues predicting blend product qualities.

b. Online Analysis – This method has two configuration options of using a single set of multiplexed analyzers to analyze the qualities of stock and product tanks in slow and faster scan frequencies. The pros of this method are that the availability of component qualities in teal-time during blending facilitates better blend quality prediction. The second method uses two sets of analyzers for stocks and product tanks and facilitates a faster scan of component tanks every two minutes per tank-like a product tank. Hence, the qualities are synced faster than the single analyzer set option. Both configurations require investment, maintenance, and updating the analyzer with changing process conditions.

c. Blend Model-Based Prediction – This software-based method uses the first principle blend models. The technology uses lab analysis of process streams feeding to the components tanks or installing stand-alone online analyzers or multiplexed with blend header as discussed above by changing the sampling point from tank outlet to tank inlet. The technology facilitates the real-time computing qualities of all component and product tanks and makes the qualities available 24/7 for planning and optimization. The technology has been installed for Singapore Refinery, and a published white paper can be accessed here for details.

d. AI / Machine Learning Prediction – Component tank qualities vary due to crude switch process changes and mainly depend on lab analysis by taking samples at scheduled frequencies. It would require a large set of historical data for sample timestamp, actual analyzed qualities and qualities and flows of incoming process streams to the component tanks. These AI/machine learning models are difficult, if not impossible, to develop but would require a concentrated effort from the enterprise. The pros and cons of installing multiplexed multi-analysis analyzers such as NIR vs complex AI models must be weighed carefully.


The blend qualities at the header are very crucial and an important factor in accurately predicting the final blend qualities of the product tank. Qualities of the blend header can be determined again by four methods but offer more flexibility than determining the qualities of the component tanks.

a. Lab Analysis – The lab analysis of the qualities at the blend header point is not done very often if the analyzers are installed. It may be required to check the online analyzers against the lab values only to determine bias or error terms. Again, the lab does not sample and analyze blend header qualities for every blend run.

b. Online Analysis – The blend header qualities may be analyzed by installing a set of analyzers like NIR, RVP, and Sulfur as NIR can not determine all qualities accurately for RVP and sulfur. The following table compares quality measurement accuracies by NIR online analyzer and lab analysis.

Online analyzers pose many challenges for the refinery. They are: updating the NIR model, availability of in-house NIR modeling experts, maintenance of NIR, etc. Although analyzer manufacturers are adopting AI / machine-based self-tuning and updating analyzer models, some human attention is still mandatory for the proper functioning of the blending system.

1718832963612?e=1724284800&v=beta&t=Kh51MGVhmBLhy K5ZYSSOWOgBeQL7o9aRYeGMraIrRY

c. Blend Model-Based Prediction – Blend header qualities are predicted using traditional first principle blend models. This has been the preferred installation in the refineries for the last decades. This method of prediction is again a backup of lab analysis but again requires customization and updating of blend models for each quality. Most of such requirements for the blend models are automated but require a dedicated blending engineer and LP expert for the proper function of the blending system.

d. AI / Machine Learning Prediction – This method of predicting blend header qualities is up-and-coming for an analyzer-less blending system. OMS eLearning Academy is pioneering AI / Machine learning prediction models for the blend header qualities using the historically certified lab analyses of the product tanks by back-casting to simulate the blend header qualities.

The following diagrams show how AI/machine learning can be implemented for an analyzer-less system.



a. Lab Analysis – All qualities of the final product tank are analyzed and certified before selling to the customer. The entire blending system flow of qualities is the only point where blend qualities are accurate and can be used to back-cast, predict, and update blend models. Let us consider this gold standard of blend qualities.

b. Online Analysis – This method of estimating final blend product tank qualities is not applicable.

c. Blend Model-based Prediction – In a traditional blending system using first principle blend models, the predicted or analyzed blend qualities are integrated online, and blend product tank qualities are updated every minute. This means that product tank quality is updated with 150 to 200 barrels of blend in the final product tank. This represents only .50 to .60% of the heel volume. Integrating a new blended amount can be linear without loss of accuracy. However, this may create round-off errors when the entire batch of 50 to 60 thousand barrels is blended. This round-off error is considered in bais calculated by comparing the final lab analysis and predicted quality of the product tank and used in the next blend run to offset the error.

d. AI / Machine Learning Prediction – AI/machine learning algorithm can play an important for an analyser-less blending system. The blend header qualities are back-casted from the final tank qualities in this method. Then, we can create an AI model using linear or non-linear AI predictions as in the hybrid model. We will leave the discussion of this aspect for a future article with a case study of 750 blends data from an actual refinery.

The diagram below shows an analyzer-less system’s AI / Machine learning architecture.

1718832962599?e=1724284800&v=beta&t=E61YbgNgbB0UC HFw5vWrrLDNOgc08snhwL9LgjXTlw


We have discussed different sources in a refinery blending system where the qualities are essential for success and minimizing the quality giveaways.

OMS eLearning Academy is an online platform to teach refinery professionals new or renewed skills to design, develop, implement, and maintain a profitable and efficient system. We do so by offering 200+ courseswith high technical content in audi-video-subtitled-interactive format.

You can try accessing the entire academy free of cost, no strings attached, for 3-6 months as an individual student or a refinery professional with corporate credentials.

Disclaimer: OMS eLearning Academy and ChatGPT collaborated as Humans and AI to generate this article for you.

Stay tuned for more groundbreaking publications and enrich your expertise with OMS Academy. 🌐✨

Sign up todayΒ to start your learning journey.

➑️ Sign Up Now πŸš€

Thank you for being a part of the OMS eLearning Academy community.

If you have any questions or feedback, please contact us.

“At OMS eLearning Academy, excellence in education isn’t just our goal; it’s our promise – a promise to fuel minds and forge futures in Downstream Refining.”

– Online Versus Lab Analysis: Understanding Why, Where, and When" sku=""]
Explore Our Courses