A. JAWANDHIYA and D. PANCHAL, Honeywell UOP, Bracknell, UK; and M. K. Gellaboina, Honeywell UOP, Kuala Lumpur, Malaysia
Fluid catalytic cracking units (FCCUs), equipped with cyclones as catalyst retention devices in reactors and regenerators, often face unplanned shutdowns primarily due to catalyst loss issues. A significant contributor to catalyst loss is the erosion or deterioration of the abrasion-resistant lining (ARL) within the cyclones. Monitoring this erosion rate is challenging due to the difficulty in accessing it for direct measurements, often leading to its oversight until erosion is discovered during a turnaround (TAR). This unexpected discovery typically extends the TAR period due to the unplanned repairs required for the ARL. Presently, the industry lacks accurate methods to estimate a cyclone’s reliability based on the estimated remaining ARL’s thickness, leading to difficulties in predicting the extent of cyclone repairs needed during TARs. This article focuses on developing an approach for estimating ARL thickness loss over time in the cyclone inlet, and then predicting the remaining useful life (RUL) by using piecewise linear projections.
The FCCU. The FCC process converts heavy crude oil fractions into lighter, more valuable hydrocarbon products at high temperatures and moderate pressures in the presence of a silica/alumina-based catalyst. A typical FCCU schematic diagram is shown in FIG. 1. In the course of cracking large hydrocarbon molecules into smaller molecules, a nonvolatile carbonaceous material—commonly referred to as coke—is deposited on the catalyst. The coke laid down on the catalyst acts to deactivate the catalytic cracking activity of the catalyst by blocking access to the active catalytic sites. To regenerate the catalytic activity of the catalyst, the coke deposited on the catalyst is burned off with air in a regenerator vessel (FIG. 1).
One of the important advantages of FCC is the ability of the catalyst to flow easily between the reactor and the regenerator when fluidized with an appropriate vapor phase. The vapor phase on the reactor side contains vaporized hydrocarbon and steam, while, on the regenerator side, the fluidization media comprises air and combustion gases. In this way, fluidization permits hot regenerated catalyst to contact fresh feed; the hot catalyst vaporizes the liquid feed and catalytically cracks the vaporized feed to form lighter hydrocarbon products. After the gaseous hydrocarbons are separated from the spent catalyst in riser termination devices and/or cyclones, the hydrocarbon vapor is cooled and then fractionated into the desired product streams. The separated spent catalyst flows via steam fluidization from the reactor to the regenerator vessel where the coke is burned off the catalyst to restore its activity. Two stages of cyclones are typically used in regenerator vessels to separate combustion gases from the regenerated catalyst. In the course of burning the coke, a large amount of heat is liberated. Most of this combustion heat is absorbed by the regenerated catalyst and is carried back to the reactor by the fluidized regenerated catalyst to supply the heat required to drive the reaction side of the process. The ability to continuously circulate fluidized catalyst between the reactor and the regenerator allows the FCCU to operate efficiently as a continuous process.
Nearly all FCCUs have experienced a catalyst loss that is mainly due to cyclone efficiency loss and/or catalyst attrition. This is also evident from 2017 benchmark survey results conducted on 45 of the authors’ company’s FCCUs, as shown in FIG. 2. The survey identified different causes for unplanned TARs such as catalyst losses, main air blower failures, main column fouling and refractory failures. Out of all failures, catalyst losses contribute to over 35% of unplanned TARs.
Cyclones use centrifugal force to separate catalyst particles from gas. These particles are forced to the walls and fall down into the dipleg, as shown in FIG. 3. The gas accelerates to the outlet tube at the top of the cyclone.
Utilizing refractory materials to establish an ARL within the cyclone interior serves as a protective measure against the erosive impact of gas flow contents. This standardized thickness lining (e.g., 19 mm, 25 mm) acts primarily to safeguard cyclone walls from erosion over time. Erosion of the ARL can ultimately lead to exposing metal walls, followed by the formation of holes—ultimately causing disruptions of pressure balances. Such exposure may draw gases into the cyclone, causing particle concentrations in the center and undesirable catalyst loss with the gas flow.
Various factors—including higher cyclone inlet velocities, increased catalyst loadings and elevated outlet tube velocities—influence the rate of erosion. Assessing erosion during the operation can be challenging, necessitating reliance on empirical data and predictions.
Velocity is a key operating parameter for cyclone performance. Collection efficiency first increases with velocity up to a certain value and then drops off due to catalyst re-entrainment and attrition inside the cyclones. Catalyst attrition to micro-fines does occur within cyclones and increases with velocity. The overall cyclone collection efficiency depends on numerous factors, including the number of spirals within the barrel and cone, the inlet velocity, particle density and size, and catalyst loading.
The regenerator superficial velocity is defined as flow velocity calculated as if the flue gas was the only one flowing in a given cross-sectional area based on the regenerator’s diameter. Other phases, particles and internals present in the regenerator are disregarded. Superficial velocity is particularly important for bubbling-bed regenerators, where the entrainment from bed into cyclones is a function of superficial velocity.
Current industry-standard method. Cyclone ARL thickness loss is a function of regenerator superficial velocity in a bubbling-bed regenerator. The higher the superficial velocity, the more catalyst will be entrained into the cyclones, and the faster the catalyst is going to accelerate into the cyclones and out of the cyclones, thereby increasing the erosion potential. This simple model sums up daily superficial velocity averages to some exponent. The typical range for this exponent is tuned for each unit based on operating experience. The summation throughout the run is then used as a proxy for the condition of lining. This method does not take into consideration the catalyst properties or the actual geometry of cyclones.
The next section will explain the authors’ method and compare results with the current industry method.
Proposed method. This method focuses on predicting ARL thickness loss in the target area or inlet scroll of the cyclone.
ARL loss computation. Erosion of cyclone refractory in FCC systems correlates with the inlet velocity and catalyst loading. A recommended inlet feed velocity range of 55 ft/sec to 75 ft/sec aims to balance efficiency with erosive effects. Most FCCUs operate within this range, while a small percentage exceeds 75 ft/sec. Very high inlet velocities can result in rapid ARL wear.
A suggested correlation exists between the rate of inlet velocity increase and accelerated ARL wear, possibly exhibiting an exponential relationship. FIG. 4 illustrates the exponential increase in the probability of catalyst loss with higher inlet velocities, impacting cyclone efficiency and indicating ARL erosion. Catalyst loss not only affects cyclone efficiency but may also necessitate unplanned TARs for inspection and repair.
A cyclone’s ARL life, along with superficial velocity, is also influenced by parameters such as inlet velocity, catalyst loading, catalyst hardness, refractory quality and cyclone design. Refractory quality, catalyst hardness and cyclone design can be treated as constants, which could vary from unit to unit. Therefore, the erosion rate is dependent on the particulates load into the cyclones, particle density, particle size and velocity. Using the combination of these variables, an operator should be able to visualize the loss of ARL thickness with days on stream, as shown in FIG. 5.
Piecewise linear segmentation. Erosion monitoring of FCCU cyclones is crucial for ensuring the efficiency and reliability of the process. This article focuses on observing the ARL loss over time as a key indicator. However, due to the dynamic nature of operating conditions spanning 3 yr–7 yr, ARL loss is not linear but rather exhibits a piecewise linear behavior. To address this, the authors propose a digital algorithm that can detect these piecewise linear segments in real time using streaming data. This approach estimates the time it takes for ARL loss to reach a specified threshold (x mm).
The digital algorithm, deployed on several of the authors’ company’s FCCUs across the globe, receives daily average time series data and computes ARL thickness. It follows a systematic process to filter outliers in ARL thickness, employing both process variables and the Hampel filter, which can detect outliers based on a sliding-window method. Once outliers are removed, the data is smoothed using moving average and median filters. ARL is then plotted on the Y-axis, with days on stream plotted on the X-axis, as shown in FIG. 6. The algorithm assesses if there are enough points to fit a line and then proceeds to fit the line.
This approach offers a novel alternative to regression by leveraging eigenvalues to detect linear regions in real-time streaming data. The eigen-based method is employed to compute eigenvalues for the given ARL thickness and time (T) data. In a perfectly linear scenario, the first eigenvalue is high, and the second eigenvalue is zero. As the data deviates from linearity, the second eigenvalue increases. By controlling the threshold on the second eigenvalue, one can effectively fit piecewise linear lines to the streaming data, as shown in FIG. 6. This methodology allows for flexibility in adjusting eigen values (providing options for both relaxed and tight fits), and specifically considers lines with negative slopes to capture the decreasing trend in ARL thickness over time. One additional advantage of this algorithm is that it runs automatically without manual intervention, unlike some algorithms where break points need to be provided for a piecewise linear algorithm, with the need to continuously change the number of break points. The authors’ algorithm preserves the previous linear segment time stamps, so that it will be easy for operators to understand each segment’s behavior and to determine which variables are impacting cyclones in those segments.
Estimating the remaining cyclone length. Once the piecewise linear model is fitted to the ARL thickness for the current day, the user can take the slope of the last line and project it further into the future, as shown by the dark blue line in FIG. 7. The intersection point of the ARL projection and the minimum threshold line (shown by the yellow horizontal line in FIG. 7) provide the remaining cyclone ARL thickness. Apart from the current ARL thickness, ±10% of primary cyclone inlet velocity is used to obtain respective projections to calculate the cyclone’s remaining ARL life if the unit operations varied the cyclone inlet velocities by +/- 10% from current operation. The authors also provide a projection using maximum primary cyclone inlet velocity to test the impact on the remaining cyclone’s ARL thickness if the unit runs at the maximum primary cyclone inlet velocity in the future. Projections for the four cases are shown in FIG. 7.
The algorithm operates autonomously without the need for manual intervention. Once the configuration engineer updates the start of the run, uploads the configuration file and triggers the algorithm, it runs continuously every day to update the ARL thickness and respective projections. Operators can make decisions by visualizing the projections on screen. They can select historical dates for the same run and view historical projections. Operators can also adjust cyclone velocities and catalyst circulation rates to understand their impact on projections, which will be computed by the algorithm and displayed on screen for the operator.
Takeaway. The authors’ company has developed a novel approach with a digital algorithm to effectively estimate the current cyclone erosion and predict future erosion based on operating conditions. With this approach, refiners get an automatic way to better predict abrasion lining erosion for both reactors and regenerator cyclones daily from digital analytics, thus helping refiners quantify their monitoring. This provides them with advantageous insights for the maintenance requirements during a TAR, thus avoiding extended TARs that could cause a higher loss of revenue and higher repair costs. Unplanned shutdowns are prone to causing substantial impacts on production revenues. Using this digital approach, refiners can avoid conservative unit operations and utilize assets to their maximum potential. HP
REFERENCES
Amit Jawandhiya is a Global Offering Manager for Refining Digital Solutions at Honeywell UOP. He leads development initiatives for connected services offerings for refining technologies. He has worked more than 17 yr at Honeywell UOP in roles spanning process design, technology technical sales, strategic marketing and business management. He earned a chemical engineering degree from LIT, India, and an MBA from Warwick University, UK.
Dharmesh Panchal is a Service Fellow and a member of Honeywell UOP’s FCC/Alkylation/Treating Technology Services group based in the UK. Currently, Panchal has responsibility for providing troubleshooting, unit optimization, projects, engineering and start-up support for new and revamped FCCUs, and turnaround support for the EMEA/FSU/Asia regions. He has 30 yr of experience in the industry focused on commissioning, startup, operations, optimization, revamp and troubleshooting of FCCUs, including 5 yr as a Senior Staff Consultant working on a range of design projects, simulations, unit optimization and troubleshooting. Panchal is a Chartered Chemical Engineer and holds several patents with Honeywell UOP.
Mahesh Kumar Gellaboina is a Principal Data Scientist at Honeywell International, based in Malaysia. He earned an MTech degree in electronics and communications engineering from the Indian Institute of Technology Madras (IITM) in Chennai, India. With 17 yr of experience, Gellaboina specializes in machine learning and artificial intelligence, and has extensive experience in developing data analytics-based solutions for the petrochemical and automation industries.