Batch-to-batch optimal control of a batch polymerisation process based on stacked neural network models

A neural network based batch-to-batch optimal control strategy is proposed in this paper. In order to overcome the difficulty in developing mechanistic models for batch processes, stacked neural network models are developed from process operational data. Stacked neural networks have enhanced model generalisation capability and can also provide model prediction confidence bounds. However, the optimal control policy calculated based on a neural network model may not be optimal when applied to the true process due to model plant mismatches and the presence of unknown disturbances. Due to the repetitive nature of batch processes, it is possible to improve the operation of the next batch using the information of the current and previous batch runs. A batch-to-batch optimal control strategy based on the linearisation of stacked neural network model is proposed in this paper. Applications to a simulated batch polymerisation reactor demonstrate that the proposed method can improve process performance from batch to batch in the presence of model plant mismatches and unknown disturbances.     

Error correction in hydrostatic spindles by optimal bearing tuning

In this paper a high precision grinding wheel is considered as a rigid rotor mounted on two hydrostatic bearings. The equations for small perturbations of the wheel on the bearings are derived in the form of a multi-input, multi-output transfer function matrix, enabling the frequency response function of the wheel to be determined. Thereafter an optimisation algorithm is proposed which considers speed, load and dimensions of the spindle, and computes optimal stiffness and damping of the bearings. The dynamic characteristics of the bearings, tuned for minimum radial displacement of the spindle, is achieved maximising thereby the accuracy of the grinding process. Simulation results show that by stiffness coarse adjustment, and fine adjustment of the damping in the bearings, a spindle with 35 μm manufacturing error, can produce components with 3 μm accuracy.     

Contents

The attempts of researchers to obtain accurate and high-quality surfaces have led to the invention of new methods of finishing. Magnetic abrasive finishing (MAF) is a relatively new type in which magnetic field is used to control the abrasive tools. Surfaces of moulds, for instance, are among those which require very high-surface smoothness. Usually, this type of part has freeform surfaces. In this study, the effect of magnetic abrasive process parameters on finishing freeform surfaces of aluminium parts has been examined. This method was achieved through a combination of the magnetic abrasive process and computer numerical control. The use of a simple hemisphere to be joined on the flat area of the magnet as well as spark machining for forming a sphere at the end of a magnet were performed during experimentation. Gap, rotational speed of the machining head, amount of abrasive powder and feed rate were among the parameters that were tested in experiments. The design of experiments is based on the response surface methodology. Significant parameters and the regression equations governing the process were also determined. The impact of intensity of the magnetic field was obtained using MAXWELL finite element software. In the MAF process, magnetic abrasives play the role of cutting tools. However, the magnetic abrasives are not easily available as these are produced by special techniques such as sintering method, adhesive based, plasma based or gel based. This study presents the basic polishing characteristics of the magnetic abrasives produced by the mechanical alloying process. After the mechanical alloying process fine magnetic abrasives are obtained, in which the abrasive particles adhere to the base metal matrix without any bonding material. In this study, investigation was performed only on the convex area of workpiece. Optimum parameters are gap size of 0.5?mm, feed rate of 10?mm?min^?1 rotational speed of 2100?rev?min^?1 and powder amount of 1.75?g. To help understand the effectiveness of the MAF process, scanning electron microscopy and atomic force microscopy of the machined surfaces have been carried out.     

Finite element simulation of butt welded 2·25Cr–1·6W steel pipe incorporating bainite phase transformation

Abstract

This article presents a newly developed global optimisation method for the finite element simulation of welding process considering bainite transformation. In this method, the pattern search algorithm was applied to determine kinetic parameters in Johnson–Mehl–Avrami–Kolmogorov (JMAK) equation during a continuous cooling process. Meanwhile, the JMAK equation was modified into an explicit form as a function of welding temperature field to improve calculation efficiency in the optimisation process. This methodology improves the accuracy as calculating the temperature dependent volume fraction of bainite transformation in finite element simulation. The calculated welding residual stresses considering phase transformation effects exhibited better agreement with the measured results than those calculated without phase transformation. The influences of variable cooling rates on welding residual stresses were also investigated.     

Response surface methodology for optimisation of protein concentratepreparation from cowpea [Vigna unguiculata (L.) Walp]

The optimum conditions for the preparation of protein concentrate from cowpea were determined using response surface methodology. A central composite rotatable design, consisting of eight experimental points and five replications at the centre point, was used to investigate the effects of two independent variables, namely pH and NaCl concentration on four responses: protein content (%,), protein yield (%), reactive lysine (g/16 g N) and zinc content (mg/100 g). A second-order polynomial model was used for predicting the responses. Regression analysis indicated that more than 80% of the variation was explained by the fitted models. Experimental results showed that under optimum conditions (pH and NaCl concentration of 9.91 and 0.15 M, respectively) the protein content was 84%, protein yield 87%, reactive lysine 1.175 g/16 g N and zinc content 7.75 mg/100 g. These results were in agreement with those predicted, hence indicating the suitability of the model used.     

An impact assessment of electricity and emission allowances pricing in optimised expansion planning of power sector portfolios

The present work concerns a systematic investigation of power sector portfolios through discrete scenarios of electricity and CO₂ allowance prices. The analysis is performed for different prices, from regulated to completely deregulated markets, thus representing different electricity market policies. The modelling approach is based on a stochastic programming algorithm without recourse, used for the optimisation of power sector economics under multiple uncertainties. A sequential quadratic programming routine is applied for the entire investigation period whilst the time-dependent objective function is subject to various social and production constraints, usually confronted in power sectors. The analysis indicated the optimal capacity additions that should be annually ordered from each competitive technology in order to substantially improve both the economy and the sustainability of the system. It is confirmed that higher electricity prices lead to higher financial yields of power production, irrespective of the CO₂ allowance price level. Moreover, by following the proposed licensing planning, a medium-term reduction of CO₂ emissions per MW h by 30% might be possible. Interestingly, the combination of electricity prices subsidisation with high CO₂ allowance prices may provide favourable conditions for investors willing to engage on renewable energy markets.     

Experimental Nonlinear Model Based Predictive Control for a Class of Semi-Batch Chemical Reactors

This paper investigates experimentally the application of a nonlinear model based predictive control (NLMBPC) to a class of semi-batch chemical reactors equipped with a mono-fluid heating/cooling system. We present the experimental results dealing with a strongly exothermic reaction carried out in a small pilot plant according to a procedure commonly used in industry. The application of the NLMBPC is based on a constrained optimisation problem solved repeatedly on-line. The control objective is to keep the reactor temperature within safe operating specifications by manipulating the heating power. Experimental results demonstrate that this control strategy works well in the presence of hard constraints and load dis-turbances.     

DESIGN AND TUNING OF A FUZZY CONTROLLER FOR INDOOR AIR QUALITY AND THERMAL COMFORT MANAGEMENT

This paper presents the development and testing of a multi objective fuzzy controller. It is dedicated to the control of thermal comfort and indoor air quality while minimising energy consumption and reducing instabilities are other constraints for this controller. After a brief introduction on the difficult task of indoor climate control, the test cell and its corresponding model is described in the second section. The model has been developed within TRNSYS. Section 3 describes the architecture of the fuzzy controller applied to these control objectives in the test cell. A hierarchical architecture has finally been selected since it is a very flexible architecture and it results in reducing the total number of fuzzy rules. The rule optimisation technique, based on genetic algorithms is also presented in the third section. Simulation results are finally presented. They prove that, after optimisation, the fuzzy controller greatly outperforms a typical ON-OFF controller on all control criteria. Energy consumption is in particular reduced by more than 10%. The fourth section of this chapter presents field experiments among 20 subjects of both ON-OFF control and fuzzy control. If major differences are not observed on the thermal comfort and indoor air quality point of view, these experiments however illustrate the importance of on-line tuning of specific fuzzy rules.     

MODELISATION D'UN SECHOIR SOLAIRE A POISSONS FONCTIONNANT EN CONVECTION NATURELLE

A mathematical model of the behaviour of a fish solar dryer has been described. This model has allowed an optimization of the design of this dryer (solar-chimney area and height, insolator area, etc...). The natural convection had to pulse drying air inside apparatus, whose capacity should be 30 kg of fresh fish.

Then the dryer has been built in National Fishing School of Anjouan (Comores Island). The first results mean a good fit between theoretical and experimental results.

The air-insolator, used for air heating, is a porous type, made with blacked coco fibers, crossed by the drying air. This local material is confirmed as a good thermic interface.

The most sensitive parameter, on the drying time, is the thickness of the fish slices. The choice of the fish quality is also very important: each different fish is characterized by one or several specific value of the water diffusivity constant, D_m. The model uses experimental values for the calculation.

Then the Fick's diffusion equation may be solved by decomposition of the moisture content inside fish slice (Fourier analysis).

Un modeéle matheématique du comportement d'un seéchoir solaire, destineé aà seécher du poisson, a eété eécrit. II a permis d'optimiser au mieux la geéomeétric de ce seéchoir qui a pu eêtre construit sur le site de I'eécole Nationale de peêche d'Anjouan aux Comores. II s'agit d'un seéchoir fonctionnant grace aà la convection naturelle, et dont la capacite est d'environ 30 kg.

L'air de seéchage est chauffeé par un capteur solaire de type absorbeur poreux, reéaliseé en bourre de noix de coco noircie. Le choix de ce mateériau local est bien suÛr volontaire; il s'est par ailleurs aveéreé comme un excellent milieu d'eéchange thermique.

Le parameètre le plus sensible sur la dureée de sechage est l'eépaisseur des filets de poisson. La qualiteé du poisson est aussi un choix important, chacun posseédant une constante de diffusiviteé de l'eau, D_m, bien speécifique. Le modeèle utilise les valeurs expeérimentales de cette constante. L'eéquation de Fick, qui reégit la migration de l'eau dans un mateériau poreux, peut alors eâtre reésolue en utiiisant une deécomposition en seérie de Fourier, de la reépartition de l'eau dans les tranches de poisson.