Ph. D. Project
Title:
Dynamic diagnosis for nonlinear process. Application to a liquid-liquid extraction recycling unit.
Dates:
2021/10/01 - 2024/09/30
Other supervisor(s):
Dr Duterme Amandine (Amandine.DUTERME@cea.fr)
Description:
Knowledge of the state of an industrial chemical process, including operating parameters, state indicators and other available monitoring measures, is a key-point for process control. Indeed, if hazards or operating changes occur, it is necessary to be able to detect their presence, their nature and possibly their amplitudes in order to maintain the process as close as possible to the target state. However, the uncertainties and errors affecting the collected measurements often make it difficult to estimate the process state. Specific treatment of the measured data, for example by data reconciliation techniques, is necessary to determine a consistent data set (i.e. closest to the measured data set and in accordance with the knowledge-based models).
Recently, A. Duterme's PhD thesis [1] has led to the definition and the implementation of a tool for the state estimation and the operating diagnosis of a liquid-liquid extraction recycling unit in static regime, using the CEA simulator and data reconciliation techniques.

The proposed thesis consists in methodological developments aiming to extend existing results in order to take into account the dynamic regime of the process and its non-linear operation for its diagnosis. Two complementary aspects are to be developed, the first related to the data validation [1, 6] and the second dedicated to the detection and analysis of operating changes of a system [2, 3].

For both points, a preliminary work is required to complete the design of a simulator of the complex chemical process studied, as well as the study of the structure of the associated model, through sensitivity studies of the parameters and the design of a signature matrix reflecting the influence of variations in inputs and specific parameters on system outputs.

Then, the first objective of the thesis will be to reconcile the dataset, by taking into account the measurement uncertainties and the model previously developed. Although data reconciliation of nonlinear dynamic systems is not a new topic [4,5], little work has been published and methodological developments are expected. The extension of well-established methods from the static to the dynamic case could, for example, be done by discretization of the continuous model and a time sliding window approach. This first step should make it possible, from the data collected on the process, to generate malfunction indicators, i.e. indicators detecting biased sensors, if there are any, and indicators detecting a modification of the process behavior (for instance : leaving a steady state towards a dynamic regime of the process).

Finally, the second objective concerns the analysis of these indicators in order to determine the operating variations at the origin of the observed evolution. The use of the dynamic simulator (taking into account, in particular, changes in process rate, changes in reagent batches or even disturbances) should allow to simulate the responses of the state indicators according to the operating variations. The next step is to develop an algorithm, based on an inverse model to trace symptoms back to their causes, i.e. to determine the operating variation(s) that caused the observed behavior change.

The final tool would be a « digital twin », innovative in the field, capable of following the process at all times, whether it is at steady-state equilibrium or in expected or accidental transient dynamic phases.

This thesis is of twofold interest, the first concerning the definition, realization and implementation of a "digital twin" type tool for a complex chemical process, an innovative field that has been expanding rapidly in recent years. The candidate will thus acquire skills in data science, in addition to knowledge in process engineering. These skills will be highly appreciated by industrial companies involved in monitoring and advanced control of complex processes. The second interest is the methodological development and implementation of operating change detection. Associated with diagnostic tools, these methods are in full expansion because they provide synthetic and consistent information with the knowledge of the process materialized by the simulator, which is very useful for complex systems operation. In addition, the person recruited will benefit from multiple supervision, both by the industrialists providing their knowledge of the process, by the CEA research team for their expertise in the process modeling and supervision and by the research team in the Nancy Automatic Research Center (CRAN, a laboratory affiliated to the University of Lorraine and the CNRS) for their expertise in model or data-based diagnosis.

References
[1] Duterme A. Estimation de l'état d'un procédé basée sur la réconciliation de données couplée à un simulateur. Thèse de doctorat en Génie des Procédés et de l'Environnement, INPT, Toulouse, 2019.
[2] Marx B., Ichalal D., Maquin D., Ragot J. Operating mode recognition. Application to a grinding mill process. Symposium on Control, Optimization and Automation in Mining, Mineral and Metal Processing, Vienna, Austria, 2016.
[3] Nagy Kiss, A.M. Analyse et synthèse de multimodèles pour le diagnostic : application à une station d'épuration. Thèse de doctorat en Automatique, traitement du signal et des images, génie informatique, Université de Lorraine, 2010.
[4] Kallas M., Mourot G., Maquin D., Ragot J. Data driven approach for fault detection and isolation in nonlinear system. International Journal of Adaptive Control and Signal Processing, 32 (11), 1569-1590, 2018.
[5] Bai S., Mclean D., Thibault J. Simultaneous Measurement Bias Correction and Dynamic Data Reconciliation. The Canadian Journal of Chemical Engineering, 85, 2008.
[6] D'Emilia G., Gaspari A. Data Validation Techniques for Measurements Systems Operating in a Industry 4.0 Scenario a Condition Monitoring Application. Workshop on Metrology for Industry 4.0 and IoT, Brescia, 112-116, 2018.
Keywords:
diagnosis , nonlineair sysytem , digital twin , industrial application
Conditions:
Duration: 3 years
Location: CRAN Nancy (France) and CEA Marcoule (France)
Salary: the net salary is around 1500¬/month


To apply to this PhD, please provide the following elements:
- A detailed CV including your university curriculum and a description of your professional and internship experiences
- A Master degree or equivalent degree certificate (this equivalence should be validated by the graduate school in the case of foreign testamurs)
- Copies of testamurs and diploma supplements, as well as grades and rankings from the candidate
- Dissertations and/or internship reports and/or publications from the candidate
- Letters of recommendation from the scientist(s) who supervised the Master thesis or internship
- A scan of your passport (for non-EU) or identity card
- Any item which demonstrates the candidate's value and ability to prepare a PhD in automatic control

Application contacts:
Amandine Duterme (CEA) : Amandine.DUTERME@cea.fr
Gilles Mourot (CRAN, UL) : gilles.mourot@univ-lorraine.fr
Benoît Marx (CRAN, UL) : benoit.marx@univ-lorraine.fr
José Ragot (CRAN, UL) : jose.ragot@univ-lorraine.fr
Department(s): 
Control Identification Diagnosis
Funds:
The PhD student will be hired by the CEA