Factors that impact BPCL KR DHDT unit’s product diesel sulphur content
Causes
1
2
3
4
5
6
7
8
9
10
11
High and immediate
Low and delayed
Feed
Cracked AGO D95 LCGO D95 LCO D95 VD D95 Combined Combined Reactor
GOR
System pressure
sulphur
feed, %
feed density feed D95 WABT
Table 1
driven manufacturing and service environments, fluctuations in product sulphur of product diesel are not expected. Investigations have found that traditional quality control methods, often reliant on post-production inspection and manual analysis, can be time-consuming and reactive. To overcome these limitations, artificial intelligence (AI)-based inferential prediction has emerged as a transformative solution. Inferential prediction involves estimating product quality attributes that are difficult, expensive, or time-consuming to measure directly by using easily accessible process and operational data. By leveraging AI techniques – such as machine learning, deep learning, and data-driven modelling – organisations can develop predictive models that infer quality metrics in real time, enabling proactive decision-making and rapid response to deviations. This AI-driven approach not only enhances accuracy and efficiency but also supports continuous improvement through adaptive learning from historical and real-time data. As a result, we can achieve greater consistency, reduce off-grade generation and improve customer satisfaction. Based on the above understanding of AI-based inferential prediction, it was decided to apply this technology to develop an inferential model for the DHDT feed stream distillation, focusing primarily on the D95% recovery temperature. Various inferential AI models have been
removed or reduced from the DHDT feed. Feed density is to be maintained within the design value. v Changes in H₂ consumption can be observed during tank changeover to assess changes in feed quality. w AGO, VD, LCGO, and LCO draw temperatures to be provided in DHDT DCS for immediate identification of the problematic stream. x Reactor weighted average bed temperatures (WABTs) to be optimised as per product sulphur. But to increase by 0.5-1°C as an immediate action to reject feed quality change disturbances. WABTs are to be normalised back again after the identification and rectification of the disturbances. Discussion Although all actions were taken as per the above suggestions, they focused mainly on the DHDT units. No action was able to prevent feed quality fluctuations, which are the primary cause of product sulphur control issues. However, whenever a product sulphur peak was observed, it became standard practice to send different feed stream samples for distillation analysis. Necessary actions were then taken on the problematic stream to reduce the D95%/FBP temperature to within the design value. Following these actions, some improvement in feed quality was observed. Figure 4 shows the trend of
the CDU AGO D95% recovery temperature after increased awareness among personnel and prompt actions to control it. Introduction of AI- based inferential prediction In today’s highly competitive, quality-
AGO D95 % recovery temperature
300 310 320 340 330 360 370 380 350 390
14.4.23
3.6.23
23.7.23
11.9.23
31.10.23
20.12.23
Figure 4 AGO D95% recovery temperature
Refining India
55
Powered by FlippingBook