Catherine Cadet

Trends extraction and analysis for complex system monitoring and decision support

This paper presents an effective trend extraction procedure, based on a simple, yet powerful, representation. Its usefulness for complex system monitoring and decision support is illustrated by three examples. The method extracts semi-qualitative temporal episodes on-line, from any univariate time series. Three primitives are used to describe the episodes: {Increasing, Decreasing, Steady}. The method uses a segmentation algorithm, a classification of the segments into seven temporal shapes and a temporal aggregation of episodes. It acts on noisy data, without prefiltering. The first illustration is devoted to decision support in intensive care units. The signals contain information and noise at very different frequencies, and smoothing must not mask some interesting high-frequency data features. The second illustration is dedicated to a food industry process. On-line trends of key variables represent a very useful monitoring tool to control the end product quality despite high variations of raw materials at the input and a long delay. The last example concerns operator support and predictive maintenance. The results issued from a diagnostic module are complemented by the extrapolation of the key variable trends, which gives an idea of the time left to repair or reconfigure the process.

Towards the modelling and control of NOx emission in a fluidized bed sludge combustor

Sludge incineration is a widely used technology because of its large volume reduction and complete organic destruction. However, incineration does encounter a number of drawbacks, among which are notably carbon monoxide (CO) and nitrogen oxides (NOx) emissions. Carbon monoxide emissions are efficiently avoided by oxygen regulation in the furnace. Nitrogen oxide (NOx) formation, however, is very complex and not well known. This paper proposes a dynamic model of sludge combustion in an industrial fluidized bed combustor mainly focused on NOx formation. The model is tuned with industrial data. A control strategy is proposed to improve on the current industrial regulation, simultaneously ensuring CO and NOx control, yet without decreasing the combustion efficiency. The results are sufficiently acceptable for this process to be carried out in a real incinerator plant.

Diagnosis of PEM fuel cell stack based on magnetic fields measurements

Abstract Non-uniform current density distribution in PEM fuel cell results in local over-heating, accelerated ageing, and lower power output. This paper proposes a diagnosis approach for PEM fuel cell system based on monitoring the current density distribution inside the stack. A magnetic sensor network has been used to provide an image of the current density distribution. The diagnosis method has three steps: residuals generation, residual analysis to obtain symptoms and decision by classifying these symptoms. The proposed approach has been applied on virtual measurements and validated on experimental measurements under performance degradation due to long term functioning of the PEM fuel cell stack and under low air stoichiometric ratio due to an actuator fault. Results show that the proposed method is a remarkable tool for diagnosis and taking compensatory actions.

Activated sludge processes state estimation: a nonlinear moving horizon observer approach

This paper deals with activated sludge processes state estimation using moving horizon approach. In order to decrease the high computing cost of searching the global minimum of the error criterion, a descent-like approach is used. An associated model is designed so as to respect a tradeoff between model complexity and accuracy. The results point out the interest of this approach for this process and are sufficiently advanced to think to implement it further on a semi-industrial pilot plant

Dynamic modeling of clarifier-thickeners for the control of wastewater treatment plants: a critical analysis

Sedimentation is central activated sludge process, and its performance has a major impact on that of the whole wastewater treatment process. Nevertheless, there is still no satisfying model for secondary settling tanks. This paper explores the reasons why the existing one dimensional models are not relevant, from the lack of knowledge on the physical phenomena to the difficulties to solve the partial differential equations. Finally, the most important modeling challenges are presented, highlighting scientific advances that have to be done.

Discrete-time recurrent high order neural observer for activated sludge wastewater treatment

This paper presents a recurrent neural observer to estimate substrate and biomass concentrations in an activated sludge wastewater treatment. The observer is based on a discrete-time high order neural network (RHONN) trained on-line with an extended Kalman filter (EKF)-based algorithm. This observer is then associated with a hybrid intelligent system to control the substrate/biomass concentration ratio. The neural observer performance is illustrated via simulations.

Modeling and Nonlinear Predictive Control of Evaporators in Sugar Industry

Abstract This paper deals with a modeling and a control study of an industrial process. An evaporator station in sugar industry is modelled by a knowledge model which includes some aspects of representation type. This model is validated from data collected on industrial plants. The model is used in order to design a nonlinear control system of the industrial plant. A predictive control law with the Internal Model Control structure is simulated. The originality of this control structure is the use of feedback linearization in order to get a linear model implemented in the predictive control law.

Model-based fault detection using analytical redundancy for automotive proton exchange membrane fuel cell

In this paper, we present a model-based approach for fault detection and isolation of faulty operating conditions of proton exchange membrane fuel cells, and we analyse experimental results of fault detection obtained on a 20 cells fuel cell stack. The system is modelled using MEPHYSTO-FC, a 2D + 0D multi-physics fuel cell model based on lumped and bond-graph approach. Parameters of the model are identified on the 20 cells stack. For the experiments, the fuel cell is operated in nominal condi- tions and in seven different faulty conditions. The model computes the estimated fuel cell voltage and the real part of the high frequency impedance. This model-based estimation is compared to the measured data to generate two residual signals used for the fault detection. The detection algorithm is finally veri- fied during the time evolution of operating conditions, creating faults in the fuel cell and observing the residuals.

Dynamic Modeling of an Activated Sludge Process: Case Study on Paper Mill Effluents

AbstractThe aim of this paper is to propose a step by step method to establish and validate a relevant, dynamical model of the activated sludge process for the paper mill industry. The model is established by considering the specificities of the effluent to be treated. It uses the biochemical oxygen demand (BOD) measurement, which allows a direct correlation between the state variables of the model and the measurements. This model is validated on a pilot plant of semi-industrial size fed with industrial paper mill effluents. The parameter estimation is based on an analysis of the sensitivities. The results are compared with a model based on chemical oxygen demand (COD) partitioning which is issued from the simplification of the activated sludge model number 1 (ASM1) model. Finally, the model dealing with BOD measurements not only shows the best performances, but is also the easiest to implement, overcoming one main obstacle to efficient monitoring and control in paper mill wastewater treatment plants.

Management of a hybrid fuel cell system using a predictive control approach

The objective of this work is to improve the power management subsystem of a hybrid fuel cell / supercapacitor power generation system. The predictive approach is a relevant available control strategy that can explicitly handle constraints including soft ones and that can also deal with multiple control inputs. Some improvements are presented to shorten the computation time which is very important in practical cases. The first results show the interest of the proposed approach and the possible improvements.