Omer Miranda

Senior Manager, Technology

At

Digital Centre of Excellence

Biography:

BE in electronics and telecoms, MBA in product leadership,18 years of experience developing and delivering technology projects across HVAC, oil and gas and industrial automation domains. Worked across the technology stack from embedded hardware to cloud computing. Has significant exposure to working in multiple geographical locations such as the USA, Germany, China and India. Currently associated with JSW Steel as technology head of the Digital Centre of Excellence, leading the IoT, automation and data science initiatives across the JSW steel manufacturing plants.

Abstracts:

Dynamic Ferro Alloy recovery prediction and addition using ML and optimization by Omer Miranda, senior manager, technology, Digital Centre of Excellence, JSW; and Keivaly Pujara, data scientist, Digital Centre of Excellence, JSW.

After the steel is tapped at the EAF, the ladle is brought to the LRF area for the secondary steelmaking process, during which ferroalloys and fluxes are added. The recovery of elements from the ferroalloys being added is not known accurately at the initial stages. Currently, an historical average-based model is used to calculate the amount of ferroalloys to be added. It is static in nature and does not take into consideration the current process parameters. Hence, the addition of ferroalloys in the LRF process is still primarily driven by operator knowledge and experience. Also, multiple test samples need to be taken during the process so that the final output matches the desired chemistry. While this achieves the end results, the time taken to achieve the output is high.

The dynamic ferro alloy recovery prediction and addition model uses machine learning to predict the recovery rate of the alloys and optimization techniques to communicate the optimized amounts of ferro alloys to be added. The objective is to develop a dynamic predictive model that will take as input the process parameters and the input steel chemistry and then provide as output the amount of ferroalloys that need to be added to achieve the target chemistry level for the different elements. The model is self-learning whose accuracy will increase with time as the size of the underlying database increases. This model will hence reduce the overall cost of ferro alloys through the right choice and the right amount of additions across all grades of steel to be produced. This model will also reduce the overall time of the process by 10% thus increasing the throughput of the LRF process.