Analysis of PDE and mathematical models from Multiphase Fluid Dynamics. I work mainly on models for compressible fluids, and their thermodynamical closure [27,30], for di-phasic/bi-component systems using either Eulerian-Eulerian [34,32,29,18] or Eulerian-Lagrangian [5,6,8] frameworks.

Recently I am interested also in the analysis of mathematical models for biological systems for medical applications.

I am interested in efficient modeling and simulation of multiphase/multiphysics complex flows in industrial processes for process design and process optimisation/intensification.

Computational Fluid Dynamics models can be very useful to understand, analyse and improve existing processes. In my research I collaborate with different teams from group OCP and UM6P to build such CFD models for key operations in the Phosphates industry. As Examples :

• Industrial mixing in agitated tanks [13,17].

• Phosphates Flottation [13,14].

• Sulfur combustion [26].

• Preneutralisation of phosphoric acid with nitrogen [16,36,].

• Granulation of fertilizers (CFD/DEM) [21,38].

• Pipe flow for phosphates slurry transport [37].

Statistics, Estimation theory, Parameter Estimability, Surrogate models... 

I am interested in the developement and the application of a wide range of mathematical tools to the field of mathematical modeling of physical and chemico-physical industrial processes. Rigorous statistical methodologies are of utmost importance to incorporate data-driven information into models, and are a powerfull tool when used correctly to extract the maximum information. 

• Example of my research work : Sensitivity analysis and parameter identification for thermodynamic Pitzer models for complex electrolytic solutions. 

My co-authors and I, worked on the application of the Sobol global sensitivity indices to estimability analysis and the calibration of Pitzer models to a number of electrolytic solutions of interest for the phosphates industry (Phosphoric acid and Sulfuric acid based solutions).

In the paper Global sensitivity for identification of Pitzer models [23], we present the general method consisting on improving the estimability analysis algorithm proposed by Wu et al [Wu], by using statistical and global sensitivity indices (i.e. Sobol indices). Furthermore in this paper we present the application to a Sulfuric acid solution.

The slides here describe briefly this work : We use the global Sobol indices to assess the importance of input variables and deduce their estimability from avalaible data. The general principle is that the less important variables are not estimable from the model outputs.

The Github code here gives an introductory example for the calculation of Sobol indices. These indices are generally used to assess the importance of input variables for a model or a system, for example in order to simplify it by removing the 'inimportant' variables. The obtained model is called then a surrogate model.

The same methodology has been applied in our most recent paper [39] to much complex electrolytic solutions, containing Sulfuric acid, Phosphoric Acid and many other metalic ions (naturally present in the phosphates rock) qualified as 'impurities'. We can quantify the complexity of the solution by the number of chemical elements and the number of chemical species (compounds) that are considered/present and that for which we need to predict chemical activities at different operating conditions. In the paper [39], we consider a solution with 11 chemical elements and more than 30 species.

The calibrated thermodynamic model reproduces very well experimental measures for many quantities of interest. This model is now a cornerstone submodel for futur models in phosphates chemical transformation process such as rock digestion/attack, cristalization of phosphogypsum, preneutralisation, etc.

We already exploited this model in many applications aimed to improve the industrial wet process for phosphoric acid production at OCP plants.

• In the papers [19,28,35], we use calibrated thermodynamic Pitzer models to model and them to optimise industrial plant operations. In [35], two decision variables are considered, namely the distribution ratios of sulfuric acid over the sections of the Multi-tank reactor, and the operating temperature (controled through cooling). A multi criterea (Pareto front) approach is necessary to optimize two conflicting objective functions :  phosphate losses (to minimize) and gypsum production (to maximise). 

• In the paper [20], we consider the formation of species that tend to precipitate or cristalize and cause fooling issues in the process. Here 7 elements are considered and 25 species including the solid minerals for which we predict the precipitation conditions in order to prevent it.

[Wu] Wu S, Mc Lean KA, Harris TJ, Mc Auley KB. Selection of optimal parameter set using estimability analysis and MSE-based model-selection criterion. Int J Adv Mecha tr Syst 2011;3(3):188–97.

I collaborate with other reseachers at @UM6P and elsewhere, to use and apply mathematical modeling frameworks or analyis tools, in different fields like agriculture  and biology.  Examples are :

• System dynamics for analysis of national agriculture systems and macro-economics [9,  11]  

• Statistical analysis for biology/agriculture/medical sciences [24]. 

• Scientific consulting for compagnies and organisations : e.g. System dynamics modeling for OCP value chain, System thinking for OCP group strategy.

My main activities in the field of Artificial inteligence and Data science are on the discimination and the promotion of the training and research in the field, particularly at the local level of Morocco, and the regions of MENA and Africa. I aim to participate into breaking two important barriers that burden the developement of this field in Morocco, and MENA region :

• Quality training on state of the art topics. With collaborators from or originating from  MENA region, we created the NASSMA initiative; which role is to promote education and research in Machine Learning and AI, in the region. We organised our first international summer school at @um6p in June 2019. A second summer school was planned in 2020 in Istanbul but had to be canceled due to COVID. In 2022, we organised a one week training workshop on #AI4science in Rabat in collaboration with Mascir, Deepmind and UM6P.

• In Morocco, the lack of openly accessible datasets and dababases for local applications can be a barrier to the developement of research in data science or to the creation of entrepreneurial solutions based on AI. With a team in our lab @MSDA we created an initiative for collecting/preparing/labeling and openly sharing various type of datasets [22,25] that may be used either by researchers or by startups or compagnies for AI applications and research. Our main undergoing effort today is to build an NLP dataset for Moroccan Darija and we are doing this as part of the global initiave Commonvoice from the Mozilla Foundation. Our labeled Moroccan car plates OCR dataset has already been used by at least two compagnies in Morocco to build car plates recognition solutions (up to our current knowledge).