PTQ Q2 2025 Issue

Model recommendation in units’ operation for extending catalyst life

Daily avg. op. data (T’put/pressure/ WABT/HbyHC)

Kinetic model

Model Output i) Product properties S, N, density, T95, Cl ii) Catalyst RLA

Refinery B

Throughput

Life of the catalyst, months

Simulated data

Catalyst activity

Base operation with

Base

limitation to increase WABT until design EOR Hybrid model suggested solution, considering future feed streams processing

AI/ML model

90% Base

Lab data - Feed S, N, Cl distillation, density

One sample set Historical data

Table 2

Figure 4 Architecture of DHDT hybrid model

Case studies The model has been used for catalyst health monitoring throughout its life cycle and unit optimisation and has acted as an advisory tool in making decisions. The applications of the model at Refineries ‘A’ and ‘B’ is briefed in the following case studies: Refinery A Catalyst of the DHDT unit (three reactors in series) in Refinery A was loaded in 2021 during the revamp to meet BS-VI standards. As the unit approached the end of the cat- alyst guarantee period, Refinery A started catalyst procure- ment in the first reactor as per licensor recommendations to extend guarantees by another term in all the reactors. However, the hybrid digital twin model, installed at the refinery in 2023 and trained on the unit’s historical per- formance data, estimated that the current catalyst could run for the next three to four years. The model forecasted the catalyst life, taking into consideration future variations in feed properties due to crude changes. This helped the refinery defer catalyst tendering and utilisation of the cur- rent catalyst to the maximum extent saving catalyst cost to the tune of INR 10 Cr. Also, the model predictions were validated for product properties, including density, T95, CI, and hydrogen con- sumption, continuously for almost 1.5 years and found

within +1.84 Kg/m3, 2.57ºC, 0.62 units, and 0.04 wt%, respectively of plant data (see Figure 5 ). The effect of crude mix on H 2 consumption is also predicted accurately, as seen in the fourth quadrant of Figure 5. Hence, the model is per - forming as a predictive tool for optimum make-up hydro- gen required whenever there is a change in crude mix. Refinery B In Refinery B, the DHDT unit catalyst (two reactors in series) was expected to run for 12 months with a 0.51ºC/month deactivation rate based on current operating conditions and design throughput. Due to feed/effluent exchanger train problems, the second reactor’s WABT could not be increased until the design EOR, resulting in decrease in cat - alyst life by six months. However, the refinery aims to extend the catalyst life until the next shutdown schedule. The model estimated the effect of feed quality variations on deactivation rates and pass on to Kinetic model to generate case scenarios (see Figure 4). The model results showed that catalyst life can be extended with 10% reduction in throughput, considering most stringent quality of feed properties in future. In this way, the hybrid model facilitated an increase in catalyst life from six to ten months (see Table 2 ) and in accordance with scheduled catalyst procurement.

20 30 25 10 5 15

RMSE = 2.52 ˚C MAE = 2.57 ˚C

RMSE = 2.52 Kg/m MAE = 1.84 Kg/m

Model predictions Plant data

+ Base

40 30 20 10 60 50

0.6 0.4 0.2 1.0 0.8

CrudeMix-1

CrudeMix-2

RMSE = 0.77 MAE = 0.62

RMSE = 0.05 wt% MAE = 0.04 wt%

Feed density

+ Base

Days on stream (DOS)

Figure 5 Validation of DHDT hybrid model for diesel product properties and H2 consumption

PTQ Q2 2025