- Posted by Okechukwu Anosike on November 29, 2011
- [Permalink]
Although modelling and simulation is traditionally the reserve of process design – used in the feasibility and FEED stages of an EPC project – it has been progressively adopted in operations activities. Most operational plants have a working model of their process, whether it is a refinery, chemical plant, an oil and gas platform, and so on. These models are being utilised in more sophisticated ways to increase productivity, profitability, efficiency, safety, operational flexibility and many other such reasons.
Historically, steady state simulations have been used to optimise various production processes. The objective functions of such optimisation are themselves functions of important operational inputs like feed composition and temperature, or of financial indicators like raw material costs or projected revenue. A good example of this application is in the crude refinery process where the ratio of a mix of crude types is determined as the ratio that generates the highest profit from product sales via optimisation. Such optimisations require accurate models before substantial value can be realised.
A refined implementation is the Real Time Optimisation (RTO). In addition to an accurate model, a high accuracy solver such as an equation-oriented solver is also prerequisite, along with a Distributed Control System (DCS) and a plant historian. The model is reconciled with the real plant to match the plant operations to within 1 – 2% offset. The offsets between plant variables and the model variables are determined, along with variable constraints, while plant parameters are also estimated. The plant historian provides all process conditions and inputs, and the model is optimised using the high accuracy solver. The result of the optimisation provides new operational points in terms of process conditions and inputs for the plant, which the DCS utilises to provide new set points for critical controllers in the process, while the historian logs measurement data. This process is repeated to determine new set points, with the frequency governed by how quickly the plant operations stabilises. This practice is proven to improve plant output and profitability.
Steady-state models are also used in endeavours such as energy management, where the objective is to reduce energy demand and consumption in plants, while also reducing the supply costs of the energy used. The least expensive ways to generate energy is determined, taking into account operational and system constraints, process unit interactions, electricity contracts and so on. Accurate and rigorous utilities demand for the process units are also modelled. With the supply side and demand side energy usage available on a single dashboard, operations are initiated to maximize use of most efficient process units, choose the best fuels and equipment drives, better adhere to contract terms and reduce penalties, reduce venting of steam, better cost accounting and so on.
For dynamic models, initial applications involved the investigation of transient operations issues like start-up and shut-down, however this was limited due to the complex mathematic operations required to solve time derivatives and other complex differentials. With the arrival of more powerful computers and solvers, dynamic simulation has become an integral part of FEED and has slowly moved into the operations.
A popular application is in Operator Training Simulators, where a dynamic model of the plant is used to train new plant operators. The trainee is set in front of a life control panel identical to that of the plant, and is forced to intervene in the plant operation by handling pre-programmed operational challenges. Instead of the plant bearing the brunt of the training exercises, the actions the trainee takes to stabilise operations are inputs on a dynamic model, which is under the hood of the training simulator representing the real plant. In this way, new operators are able to get up to speed in their duties within a relatively short period, with minimal risk to plant operations due to the experience they gain on the simulators.
Another application of dynamic models is in Advanced Process Control, specifically Model Predictive Control (MPC), using a dynamic model, a DCS and plant historian. The model, which runs at speeds up to a hundred times the real time due to powerful processors, must be robust enough to handle every conceivable operational point of the plant. The process conditions and inputs are read from the historian into the model, which then predicts the plant behaviour before such behaviour is observed in real time. The plant controllers via the DCS respond to correct for and handle process disturbances that have been predicted in the model before they occur in the plant to maintain stable plant operations. This control system is popular in unmanned oil and gas platforms and other remote or automated operations.
Modelling and simulation has come a long way over the past decades. These are only a few examples of the innovative applications in plant operations to improve the performance of process plants. As process industries continue to automate their plant operations, one thing that we can expect is that more novel applications for modelling and simulation would be introduced to tackle operational challenges.
Tags:
Category: Chemical Engineering, General Engineering, Plant Scale Up's, Process Engineering
Actions:
E-mail | Comments (46)
Share With: Twitter | FaceBook
- Posted by Okechukwu Anosike on January 12, 2011
- [Permalink]
The implications of a well-designed crystallization process, including how Tim Bell of DuPont Engineering wrote: “Crystallization is notoriously difficult to scale-up…”, we will now address why crystallization is such a complex process.
The efficiency and profitability of a crystallization process is directly tied to the crystal size distribution produced in the crystallizer vessel. Problems with solids bulk density, flowability, crystal size and shape can often be directly related to the operations of the crystallization step. Missing the crystal size distribution specifications may result in costly rework. The production of excessive fines in the crystallizer vessel can dramatically cut production yields and can reduce throughput due to serious bottlenecks in downstream processes such as filtration and drying. Excessive milling of crystal product can result in further yield losses and result in potential dust hazards. Therefore, an engineered crystal size distribution that meets particle size specifications repeatedly, and avoiding excessive downstream modification through milling and sieving, can dramatically improve overall production efficiency and profitability.
Crystallization, however, is extremely difficult to directly transfer from the laboratory to pilot and production scale. Scale-up difficulties are compounded by the importance of both thermodynamic and kinetic properties in determining the final crystal size distribution.
Supersaturation, the thermodynamic driving force of crystallization, is a critical parameter in determining the final crystal population. In a laboratory vessel that is relatively well-mixed, supersaturation may be effectively constant throughout the vessel. At a larger scale, there are undoubtedly gradients of supersaturation throughout the crystallizer – due to the manner in which the supersaturation is created (most often by cooling or anti-solvent addition), and due to the mixing configuration (including parameters such as the vessel dimensions, baffles, impeller type, and agitation speed) which determines how effectively the supersaturation is dispersed throughout the vessel. The introduction of the supersaturation gradient plays a very significant role in the difference between laboratory-scale and full-scale crystallization.
Crystallization is further complicated by the fact that it is a multiphase system. The solid crystal product is very often a different density than the liquid phase (mother liquor). Crystals that are denser than the mother liquor have a desire to settle to the bottom of the crystallizer, and the mixing required to keep a liquid based system well-mixed is no longer sufficient to keep the solids suspended. Although a first instinct response might be to increase the agitation speed or to add baffles, one must also be aware of the possibility of crystal breakage and attrition due to the increased energy input. Attrition can be a significant cause of fine crystals or a source of secondary nucleation which will dramatically alter the final crystal size distribution and causes a number of scale-up headaches.
The kinetics of crystallization adds additional complexity to scale-up. Crystallization kinetics is commonly simplified to two parameters: nucleation (birth of new crystals) and growth (rate of increase of crystal dimensions). (More complex crystallization models may also include rates of dissolution, breakage, attrition, and agglomeration to more fully predict the population balance in the crystallizer – but these are usually situations we will work to avoid in a practical crystallization process.)
The nucleation and growth rates are primarily functions of supersaturation. At relatively low levels of supersaturation, growth tends to dominate. Nucleation rates, however, have a higher order relationship with supersaturation – so that if supersaturation reaches a high enough level, nucleation will dominate the crystallization process.
In addition, the presence of crystals – and the related crystal size distribution – is another critical factor in determining how the supersaturation is consumed and therefore the combination of the the final crystal product. This is because the quantity of crystals present – and specifically the viable crystal surface area available for growth – determines the rate at which supersaturation can be consumed by the existing crystal population.
If avoiding nucleation is desirable, the crystallizer can only generate supersaturation at a rate that the existing crystal surface area can handle at the corresponding growth rate. If the amount of surface area is insufficient to handle the generated supersaturation, then the overall supersaturation level will rise, eventually to the point where nucleation becomes a significant factor. If the amount of surface area is more than the current growth rate will sustain, the supersaturation will drop (implying that the crystallizer is running at less than full capacity.)
If that is not complicated enough, the fact that you have potential gradients of supersaturation and gradients of crystal size distribution throughout the full-scale vessel, also means that you have potential gradients of nucleation and growth rates making the final crystal product extremely difficult to predict from simple kinetic models based on laboratory data.
There have been two traditional ways of dealing with this complexity of crystallization – either crash the solids out of solution and deal with the headaches of the downstream processing bottlenecks, or simply tune down the crystallizer so far that it runs at a low enough level of supersaturation that problems of nucleation are generally avoided. Clearly, neither of these situations is optimized for maximum production yield and throughput, and this in part has fueled the recent push towards real-time monitoring of crystallization using Process Analytical Technology (PAT) that can directly measure critical parameters such as the crystal size and shape distribution, the crystal form, and even the level of supersaturation.
Tags:
Category: Chemical Engineering, Plant Scale Up's, Process Engineering
Actions:
E-mail | Comments (13)
Share With: Twitter | FaceBook