PTQ Q2 2022 Issue

events is rapidly established, and feedback is obtained quickly to determine the next course of action. There is no need for reams of historical data for reinforcement learning to crunch through. That makes DRL perfectly suitable for optimising hydrocarbon processing tasks using established metrics in the form of inputs, actions, and rewards. Being a powerful tool, which requires no historical data for reinforcement learning, DRL has one significant weakness, which makes it more difficult to implement for hydrocarbon processing with wide ranges of opera- tion. DRL is likely to improve performance only where the pre-trained parameters are already close to yielding the correct process stream qualities. The observed gains may be due to effects unrelated to the training signal, but rather from changes in the shape of the distribution curve. Therefore, there is a need for real-time monitoring of process yields, rather than only their prediction using the pre-trained models. This can be reached using on-line analysers installed in the process to determine the chem- ical composition or physical properties of substances involved in hydrocarbon processing. The Beacon 3000 is an inline, multi-channel Process NIR Analyser. It enables non-contact, real-time mon- itoring and closed-loop control of physical properties and chemical composition in industrial process appli- cations. Based on novel algorithms, the Beacon 3000 measures the absorption spectrum in the near infrared (NIR) quickly and accurately without labour and mate- rial waste. Up to eight field units, which use no elec - tricity and contain no moving parts, can be connected to one main analyser. When used as a part of the DRL implementation, to allow a high fidelity unit model, the Beacon 3000 enables tighter process optimisation and identifies process excursions before they affect yield. The incorporated analyser Freetune software compares its measurements with laboratory test results on a constant basis. In case of continuous deviations, Freetune will automatically correct the analyser’s predicted results to comply with laboratory test results. The optimised dynamic model obtains real-time data from process analysers, which is verified and validated against laboratory results and predicted product quali- ties. The simulated process’s digital twin is continuously updated by process analysers to allow the highest possi- ble efficiency of the process at lowest cost. This method enables global process optimisation through integration of the network input and target KPIs, using DRL based optimisation to maximise overall profit and reduce envi - ronmental impact. Advanced ML methodologies are capable of leverag- ing historical, contextual, and often high-dimensional data in order to produce accurate prediction models. At the same time, even if one is equipped with an excellent prediction model, the optimisation model(s) underlying a particular analytics application may be difficult to solve due to complicated objective functions and/or constraint structures. Therefore, it important to continuously mon- itor the accuracy of prediction, optimisation, and control – which are presented to the operator on HMI. Artificial planning and scheduling represents a branch Koch-Glitsch focused on the most convenient way of removing water prior to the affected columns, lead - ing to the recom endation to revamp the HP separa - tor immediately upstream. Once the problem was fully understood, and the focus was ut on the HP sep - arator, the Koch-Glitsch process engineering group evaluated the performance shortfalls of the existi g arrangement and pre ared a solution to remedy the problem. By incorporating experience and proprietary separations software, Koch-Glitsch provided a solu - tion that would meet Essar’s capacity goal. By under - standing and targeting t e source of the problem, the solution was implemented in the HP separator vessel which debottlenecked and improved the performance of the GCU. sation of diolefins in the presence of water at high tem - perature on the reboiler process side (shell side). The duty of the debutaniser reboiler was limited due to this fouling which affected the operation of the debutan - iser. The deposits on the reboiler were cleaned in the shutdown. Results Post-installation of the Ky-Flex media in the HP sep - arator, the water carry-over was reduced significantly to the debutaniser overhead receiver. As a result, the control valve in the debutaniser overhead receiver boot opened less frequently with minimal and intermit - tent water withdrawal (see Figure 7 ). The correspond - ing water flow rate was reduced from approximately 450 l/h to 108 l/h. The bottom temperature of the strip - per could be reduced with a resulting reduction in reboiler duty (see Table 1 ). The operation became stable and the capacity of the gas concentration unit could be increased. The results of the pre-revamp and post-revamp operation are shown in Table 1 . Conclusion Component trapping by physical carry-over of water can cause capacity bottlenecks that are not easily pre - dicted by performing process simulations and hydrau - lic rating of internals. Koch-Glitsch worked closely with Essar Oil and utilised their experience to troubleshoot and thoroughly review the equipment in the unit, espe - cially those performing physical separations, to iden - tify potential problems. Based on the symptoms and unsteady operation of the stripper and debutaniser under increased loading, it was suspected that water accumulation within the system was the cause. Parameter Feed from FCC main column receiver to HP separator, t/h 147 0.26 125 Pressure drop across stripper, kg/cm 2 Temperature at stripper bottom, °C Water from debutaniser overhead receiver to debutaniser feed, % 0.132 Tempe ature at debutaniser bottom, °C 164 Debutaniser reboiler duty, GCal 9.2 Table 1

of the Modcon-AI package that concerns the realisa- tion of strategies and planning in refinery activities. This ranges from determining which crude oil types to acquire to which products should be produced and sold in the market. The simulated process’s digital twin is continuously updated and enables planning and sched- uling of feedstock unloading and blending, production planning and process scheduling, and product blending and recipe optimisation. Investment in the AI/ML technologies enables the sustainable development of the modern hydrocarbon industry by adopting business strategies and developing artificial solutions that meet the needs of the enterprise today while protecting, sustaining, and enhancing the human and natural resources that will be needed in the future. predicted H data using the X factor method with an average k ideal value which is assumed constant from one flow rate to another, and the k ideal values at suction and discharge conditions have been derived from a simula - tion program on the basis of the new molecular weight (115% of base case). Again, for a new operating case where the new P d is not yet known, iteration of P d is likely needed to calculate the k ideal at the discharge con - dition. This discharge k ideal is needed to calculate the average k ideal (or k 2 in Equation 7 ) for the new operating case. The k 2 value can be used for estimating the P d profile when P d assumed to calculate the discharge k ideal value agrees with P d calculated from Equation 7 . For this example, Case B, the green dots in Figure 9 show the predicted discharge pressures using the X factor method. The predicted or calculated P d at the rated flow rate is less than 1% different from the P d from the vendor. At flow rates close to the end of the curve, the P d difference is about 2% maximum. For this example, the X method gives reasonable agreement with the vendor data which are typically derived from ‘wheel to wheel’ calculations and likely based on in-house k correlation data: (7) Gregory Shahnovsky is President and CEO of Modcon Systems with extensive business development and professional skills in analytical chemistry, industrial process control, machine learning and artificial intelligence. He holds a MSc in chemical engineering, a PhD in industrial automation, and MBA in high technology enterprises business administration. Ariel Kigel is R&D Manager with Modcon Systems. He leads market- oriented development and day-to-day operations of on-line process analytical systems for conventional industry, and leads the definition and development of new applications into new industries, overseeing multinational teams. He holds a PhD in physical chemistry and a MBA. Gadi Briskman is Product Manager with Modcon Systems. He has long, practical experience with computer vision, robotics, process optimisation, deep learning and artificial intelligence. He holds a ME in systems engineering, a BSc in electrical engineering and a LLB law degree. Fi ld signature method – area monitoring In critical locations, getting a deeper insight into cor- rosion in a large area can provide valuable insight into understanding asset health. Field signature method (FSM) measures the metal loss trend within a pre - defined area and quantifies general and localised cor - rosion. This allows pipeline operators and refineries to optimise their crude feedstock, chemical inhibitor pro - gramme, and their integrity assessment programmes. The measure ent principle, which is based on electrical potential drop, measures voltages at multiple sections on the pipe or vessel, generating a topography image of metal loss phenomenon. RMadhavan is General Manager Process with Koch Chemical Technology Group India Private Limited in the Koch-Glitsch division. He has over 20 years of experience in mass transfer technolo y and holds an M. Tech. degree in chemical engineering from Indian Institute of Technology. Sandeep Yadav is a Senior Manager Process Engineering with Koch Chemical Technology Group India Private Limited in the Koch-Glitsch division. He has over 10 years of experience in mass transfer technology and holds a bachelor’s degree in chemical engineering from Shivaji University, India. 𝑋𝑋𝑋𝑋 " = 𝑋𝑋𝑋𝑋 % 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 " 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 % 𝑘𝑘𝑘𝑘 % 𝑘𝑘𝑘𝑘 " (𝑘𝑘𝑘𝑘 " − 1) (𝑘𝑘𝑘𝑘 % − 1) 𝑋𝑋𝑋𝑋 = $ 𝑃𝑃𝑃𝑃 & 𝑃𝑃𝑃𝑃 ' ( (*+* ,) − 1 Where commonly referred to as grid scan. When g mma scanning a tower, I o remains fixed and µ is essen - tially a constant. For a grid scan, the multiple paths of t e next plant turnaround, typically up to 10 year , s no maintenan e is req ired between turnarounds. This simplicity of inst llation makes ultrasonic sen- s rs ideal for use in remote locations which are only accessibl uring turnarounds. To protect the ultra - so ic electronics from heat, the sensor uses stain- less steel waveguides to keep the electronics safely away from hot m tal surfac s up to 600˚C (1100˚F). Adva ced proc ssing software makes use of previ us recorded ultrasonic waveforms o improve the resil- ience of the measurement when the internal metal sur- face morphology is very rough, which is a situation where norm l ltrasonic wall thickness m asurements break down. 182 0.25 Feed rate increased by 124% 116 Reduction in stripper bo tom temperature reduced reboiler duty 0.05 170 Fouled reboiler tubes limited t mperatu e 13.6 Reboiler duty debottlenecked Reference 1 Kister H Z, Component trapping in distillation towers: causes, symptoms and cures, CEP, Aug 2004. J Rajesh is Vice-President Process with Essar Oil Limited India. He has over 20 years of experience in refinery operations and process and holds a bachelor’s degree in chemical engineering fromAnnamalai University, India. Pawan Gupta is Deputy General Manager with Essar Oil Limited India. He has over 12 years of experience in process design and simulation and holds a bach lor’s d gree in chemical engineering from MJP Rohilkhand University, India.

Mecha Compr lers ( I ) perform result i ( Q ). Fo and P d the co increas have m these o not on also ev dischar It wou lapse of ity to pr and fine of reserv the form greater techniqu sand ma The a during w free rate rate (MA ing you Operato ing max the rese reservoir ment pri when it algorithm compre increas For a ble spe motor increas effectiv predict to be m correla ered, a perform the cos ing resu Another equipme pipes. Non-in multiple scan cou Since all the lengt uniform fo r scan ty ical e lines ma Howev ation me seem to would be Conclu A corr gal com to pred operati The pr with th vendor sensor s Hydrost permeat rosion. T scan dat cated mo cated les example Figure 2b This is open to in the results dischar design variatio tion w a chan cussed for scr stages TekSutik Engineer the proce engineer Email: te provide operatio to anoth system t toring un Advance An dv dat fro istently c nclusi Sand aco Pulling t structura A Berth Corrosio (for inst probes) of the eq occurs t pipe soc Local co coupon radiation remains tion is so On the far high gies are corrosion pipe soc fixed ou without devices induced method sensor p an electr possible rosion in In gen from tow ysis does tity f li conc usio very amb of any l q The be c mod stag

Comparison of pre- and post-revamp results

Pre-revamp Post-revamp Results/comments

Figure 2 (A) Gamma scan results showing good liquid distribution due to unifor distribution since there was no uniform response among the scanlines

PTQQ 2 2022 71

www.digitalrefining.com

28 Revamps 2018

www.eptq.com

PTQ Q4 2017 49

16 PTQ Q4 2018 70 PTQ Q2 2018

Powered by