José María Nougués

On-line fault diagnosis system support for reactive scheduling in multipurpose batch chemical plants

Abstract In this work, a simple strategy for the development and implementation of a fault diagnosis system (FDS) that interacts with a schedule optimiser in batch chemical plants is presented. The proposed FDS consists of an artificial neural network (ANN) structure supplemented with a knowledge-based expert system (KBES) in a block-oriented configuration. The system combines the adaptive learning diagnostic procedure of the ANN and the transparent deep knowledge representation of the KBES. The information needed to implement the FDS includes a historical database of past batches, a Hazard and Operability (HAZOP) analysis and a model of the plant. Two motivating case studies are presented to show the results of the proposed methodology. The first corresponds to a fed-batch reactor. In this example, the FDS performance is demonstrated through the simulation of different process faults. The second case study corresponds to a multipurpose batch plant. In this case, the results of reactive scheduling are shown by simulating different abnormal situations. A performance comparison is made against the traditional scheduling approach without the support of the proposed FDS.

Parameter estimation with genetic algorithm in control of fed-batch reactors

The temperature control strategy of a fed-batch reactor has been presented. The fed-batch operation is very useful to increase the productivity in some type of batch reactors but may incur in reactant overdose, which in case of exothermic reaction would result in possible runaway condition. The main problem is to get the best feeding profile (temporal trajectory) to minimize the reaction time with a temperature constraint. The genetic algorithm (GA) has been used for this task and the profile obtained has been used to drive a pilot scale experimental reactor. This work also demonstrates the usefulness of the GA in parameter estimation of first principle models.

Fault diagnosis support system for complex chemical plants

Abstract A process fault detection and diagnosis system (PFD&D) is proposed for complex chemical plants. The system combines an artificial neural network (ANN) based supplement of a fuzzy system in a block-oriented configuration. A methodology for designing the system is described. As a motivating example, a chemical plant with a recycle stream is considered. Faults in the supply of raw materials and in controllers are simulated. The performance of the system in handling simultaneous faults is also analysed. A comparison of the proposed approach is made with a classification method (ANNs) and inference methods (knowledge-based system). Results of system implementation in a fluidised bed coal gasifier at pilot plant scale are also shown.

Batch and semibatch reactor performance for an exothermic reaction

Abstract In this work, a study of a batch and semibatch reactor has been carried out based on a very exothermic reaction between thiosulfate and hydrogen peroxide. The experiments were carried out in a glass-jacketed reactor of 5 l, provided with different sensors and a data acquisition system. Thermal and kinetic studies were carried out previously using an adiabatic batch reactor. Then, these results have been used for experiments in semibatch mode of operation with heat transfer. Several experiments have been carried out at different operating conditions (addition flow, initial temperature, initial concentration of reagents…). In batch and adiabatic mode of operation, experimental measures of the reaction mass temperature provided concentration profiles of reagents and products which have been compared to those determined by simulation. In semibatch mode of operation, temperature profiles have also been simulated and validated with experimental results. An algorithm describing the mathematical model has been developed and implemented in a software module written in FORTRAN 77 language. With this mathematical model it has been possible to obtain concentration and heat profiles for the semibatch mode of operation.

On-line process fault detection and diagnosis in plants with recycle

Abstract A process fault detection and diagnosis system is performed for the complex case of plant-wide control in processes with recycle in which the control system is the inventory control. It is considered an artificial neural network based supplement of a fuzzy system in a block oriented configuration. A methodology to design the system is described. As a case study, a chemical plant with a recycle stream is considered. Faults in supply of raw materials and in controllers are simulated. Performance of the system to handle simultaneous faults is also analysed. A comparison is made against a classification method (artificial neural networks) and an inference method (knowledge — based system).

Integrated on line production and financial scheduling with intelligent autonomous agent based information system

This paper proposes a methodology to cope the development of a system achieving the integration of financial and production planning with on-line predictive/reactive transaction-oriented scheduling in batch process industries. The best solution of delivery dates is found taking into account the trade off between customer satisfaction (meeting due dates) and company liquidity (minimum net cash-flow constraint). The resulting schedules are improved by means of an objective function to be minimized under the constraint of minimum cash by a simulated annealing (SA) algorithm. Production schedules are updated with process data. Predictive models upgrading control set points and product recipes close the loop of the managerial production cycle. A planning layer with autonomous web-based agents in an order-entry information system is used to obtain a quick feedback and response of the whole system to order requests. This system architecture can be used in an inner-network information system within a manufacturing Networked Enterprise (NE). In addition to the knowledge in real time of the electronic market, the integrated system performs the status/usage profile of the resources including the net cash-flow. These new capabilities in scheduling assure better financial decisions and a quick response to clients, a response which is backed up with a system that outputs an updated optimal price-time trade-off of the order requested. The methodology is illustrated with a case study.

Design of sensor networks to optimize PCA monitoring performance

In this work, a design method for sensor placement for fault detection based on PCA is presented. Cost is minimized, including in it the investment and the cost of not detecting a fault. A case study is used to illustrate the performance of the proposed technique. Genetic algorithms are used.

Modelling Superstructure for Conceptual Design of Syngas Generation and Treatment

In this chapter a description of how the process synthesis problem can be casted as a superstructure optimisation problem is done. The first section draws on how the superstructure can be built, while the second section depicts the different techniques that can be used to reduce the computational time required to run a superstructure optimisation. Section 3 describes the different integration and control considerations embedded in a possible superstructure for analysing syngas generation and treatment. This last section also shows the results of different scenarios of this superstructure model.

Control application study based on PROCEL

Abstract This is the compare study of control strategy in different operating conditions. To achieve this, a configurable process scenario at pilot plant scale (PROCEL) has been built at the Universidad Politecnica de Catalunya (UPC). In this work, the following study has been carried out in sequence: o 1. Walk through the general steps for process control development 2. In Sattline DCS, different configurations have been programmed for continuous mode case study. 3. Using the commercial software MATLAB, there have been reviewed the performance of various tuning techniques optimisation techniques Genetic Algorithm Results of the above study have been tested.