1,720,993 research outputs found
Learning to navigate a crystallization model with Deep Reinforcement Learning
No full textIn this work, a combination of a Convolutional Neural Network (CNN) based measurement sensor and a reinforcement learning (RL) framework that speeds up the control loop is presented. The objective of the controller is to reach a target mean size and to reduce the variability of the crystal sizes. The CNN based sensor improves the quality of crystal size measurement and reduces the time to process images while the RL framework learns to navigate the crystallization model optimally even in the face of disturbances. The proposed data driven strategy is validated against an unseeded crystallization of sodium chloride in water using ethanol as antisolvent in an experimental bench-scale semi-batch crystallizer. We find that the RL-based controller can be trained offline to optimize multiple target conditions while the CNN provides accurate feedback for the controller to recompute the optimal actions in the face of disturbances and guide the system towards the target
Optimal strategies to control particle size and variance in antisolvent crystallization operations using deep RL
Full text linkSolution crystallization operations have complex dynamics that are typically lumped into two competing processes namely nucleation and growth. Mathematical models can be used to describe these two processes and their effect on the crystal population when subject to variables like temperature, addition of anti-solvent, etc. To ensure that the crystals meet specific performance objectives, the models need to be solved and the control variables need to be optimized. This has largely been done until now using algorithms from dynamic programming or optimal control theory. Recently, however, it has been shown that learning frameworks like Reinforcement Learning can solve large optimization problems efficiently while offering distinct advantages. In this work, we explore the possibility of computing the optimal profiles of a semi-batch crystallizer to control the mean size and variance using four different deep RL algorithms. The performance on one of the tasks is evaluated experimentally on the anti-solvent crystallization of NaCl in a water-ethanol system
On the closed-loop stochastic dynamics of two-state nonlinear exothermic CSTRs with PI temperature control
Full text linkFokker-Planck (FP) partial differential equation (PDE) theory is applied to characterize the stochastic dynamics of a class of open-loop (OL) 2-state nonlinear exothermic continuous reactors with: (i) zero and time-varying mean noise disturbances, and (ii) linear proportional-integral (PI) temperature control. The characterization includes: (i) the stochastic on deterministic dynamics dependency, (ii) gain condition for robust probability density function (PDF) stability over deterministic-diffusion time biscale with stationary monomodality at prescribed most probable (MP) state, (iii) evolutions of along nearly deterministic time scale of MP state and control and their variabilities, (iv) attainment of random motion in-probability (IP) stability over deterministic-diffusion time biscale, and (v) identification of the compromise between MP state regulation speed, robustness, and control effort. The methodological developments and findings are illustrated with three indicative examples with OL complex (bimodal and vulcanoid) stationary state PDFs, including analytic assessment as well as state PDF and random motion numerical simulation
Characterization with Fokker–Planck theory of the nonlinear stochastic dynamics of a class of two-state continuous bioreactors
No full textThe nonlinear stochastic dynamics of a class of two-state bioreactors with isotonic or nonisotonic kinetics is analytically characterized with Fokker–Planck (FP) theory, with emphasis on: (i) the spatiotemporal geometry of the two-state probability density function (PDF) motion, (ii) conditions for metastability-based bio-extinction/revival (inexistent in deterministic systems), and (iii) state PDF behavior of the optimal (maximum yield) operation. It is found that, depending on the kind of kinetics: (i) the stationary state PDF is mono or bimodal, (ii) the state PDF motions can be either non-metastable along deterministic-diffusion time scale, or metastable towards probabilistic extinction/revival along deterministic-diffusion-escape time scale, and (iii) the optimal operation can have robust (or fragile) stationary PDF, depending on the particular kinetics and operation condition. The developments and results are illustrated with representative examples with Monod and Haldane kinetics, and put in perspective with the ones drawn before with Monte Carlo (MC) and FP methods
A geometric observer design for a semi-batch free-radical polymerization system
No full textIn this work, a geometric observer (GO) is formulated and tested using experimental data from the Automatic Continuous Online Monitoring of Polymerization reactions (ACOMP) system for a semi-batch free-radical polymerization reactor that synthetizes polyacrylamide. The available measurements include the weight-average molecular weight, the concentration of monomer, and the volume of internal contents. Different combinations between innovated states and measurements offer a number of possible structures of the GO. The computation of the minimum singular values and condition numbers permit to select the most promising structures. To overcome inadequacies observed in full-order architectures, low order GOs with passive structures are explored. The best observer is selected, compared with a standard estimation strategy and tested under various experimental operating conditions. The observer performance is evaluated qualitatively and quantitatively as a tradeoff between dynamic property estimation and signal processing
Effect of Boundary Conditions and Turbulence Treatment on the Simulated Performance of a Ribbed Heat Exchanger
No full textRibbed surfaces are widely employed in heat exchangers to enhance the convective heat transfer and hence the overall thermal efficiency. This study aims to investigate the effect of two important assumptions made in computational fluid dynamics simulations, i.e. the thermal boundary conditions and the turbulence modeling, using a popular test case for the heat transfer over a continuous ribbed plate was taken as a reference. Numerical simulations were performed both neglecting and considering the conduction within the solid, to verify the effect of different thermal boundary conditions on the fluid domain, and with several turbulence treatments, ranging from common Reynolds-averaged Navier-Stokes approaches to higher fidelity but more computationally intensive Large Eddy Simulations. The results demonstrate that both aspects are important for an accurate prediction of the thermal performance of ribbed channels
On the dynamics and robustness of the chemostat with multiplicative noise
Full text linkThe stochastic dynamics of a two-state bioreactor model with random feed flow fluctuations and non-monotonic specific growth rate is analyzed. Using the Fokker-Planck equation approach for describing the probability density function (PDF) evolution the lack of stochastic robustness due to deterministic bifurcation phenomena for the open-loop reactor operating under optimal (maximum production) operation condition is established, and the associated stochastic stabilization problem is addressed. Inherent differences between the presence of multiplicative noise, due to the feed flow fluctuations, and additive background noise are analytically established. Numerical simulation results illustrate these inherent differences, the stochastic fragility of the open-loop operation yielding a stochastic extinction phenomenon, as well as the stochastic PDF stabilization with a proportional feedback control
Solvent recovery system for a CO2-MEA reactive absorption-stripping plant
Full text linkThe solvent recovery section from the exhaust gas represents an important auxiliary part for an industrial CO2 post-combustion capture plant by the reactive absorption-stripping process. In this work, a partial condenser and a water-wash section configuration were designed to reach 1 ppm of solvent in the exhaust gas, and compared using the Total Annual Cost (TAC) as economic index. Both the configurations ensured the required recovery performance. The results highlighted that the partial condenser alternative is more convenient in terms of capital annualized costs and water make-up, but at the same time it is strongly penalized by the high operating costs for the cooling water. Therefore, the configuration in which the absorber is equipped with the water-wash section resulted the option with the minimum TAC
- …
