(152b) Effective Use of Hybrid Modeling (ML + Digital-Twin) to Predict and Maintain Uniform Conversion across Furnaces in VCM Plant | AIChE

(152b) Effective Use of Hybrid Modeling (ML + Digital-Twin) to Predict and Maintain Uniform Conversion across Furnaces in VCM Plant


Doshi, S., Ingenero
Hybrid modeling is a combination of fundamental first principle models and AI/ML algorithms to predict important process specific KPI’s like Conversion, Selectivity. Generating effective AI/ML models need large amount of “good data”, this data can be generated through fundamental models using multi-year historical operating data. AI/ML models can be trained on this data and deployed to predict these KPIs on real-time plant data. These work as a soft sensor providing real-time insight to the predicted parameters/KPIs.

VCM Furnaces are one of the most critical equipment’s in VCM manufacturing process. Generally multiple furnaces are operating in parallel and their combined effluent is processed in downstream distillation columns. Based on the total VCM product flow from VCM column the overall plant conversion can be determined (Even this is a challenge for some plants where interconnections with other plants exists). However, it is generally difficult to determine the EDC conversion in individual VCM furnaces. In absence of real-time individual furnace conversion values, one furnace may be operated at higher conversion and thus subjected to higher firing rates while other furnaces may be operating at lower conversion and thus subjected to lower firing rates. Due to this, coking rates and hence run-length will differ across furnaces, leading to earlier decoke-shutdowns and production loss. Therefore, severity balancing in addition to load balancing across furnaces is crucial for VCM operations.

For a VCM unit in USA, a hybrid modelling approach was used for calculating the EDC conversion in VCM furnaces. Using a huge volume of “clean” historical data, a first principle-based model (digital-twin) was developed to estimate the EDC conversion in each furnace, by tuning it to match with furnace downstream operating conditions. This generated data was fed to an ML model to predict conversion on real-time plant data. The concerned unit had two furnaces operating in parallel. The model provides individual and combined conversion of the furnaces as output.

Application Utilization:

  1. Control and achieve desired EDC conversion in individual furnaces.
  2. Perform severity tuning across furnaces to ensure uniform conversion.

Application Benefits:

  • The model has been made online in the discussed unit few months ago, and is performing as expected.
  • Quantitively, potential benefits of implementing this model in the unit are estimated to be ~ $0.8 million/annum.
  • Qualitatively, augmenting the intelligence of field/panel engineers with such a tool will help in pro-active and continuous improvement of furnace operation leading to better asset utilization without requiring any major CAPEX.