ARR - Annual Report
This project aims in developing a system engineering approach for understanding, optimizing and scaling industrial dry grinding processes, with a special focus on the manipulation of the material properties and, thus, the grinding and classification efficiency by the use of grinding aids. In terms of process units, the project deals with a pilot scale dry grinding circuit consisting out of a ball mill and air classifier. During milling operations, grinding aids affect the milling process mainly in: tendency of fine particle agglomeration; amount of material coated on equipment surfaces; powder flowability; total mass of product inside the mill and residence time; product fineness after grinding.
The project work for the initial three years was divided in four main work packages:
- Identify which aspects of the ball milling process are affected by different GA
- Identify which aspects of the air classification process are affected by different GA
- Modify process models from the literature to account for the presence of GA, implement a flowsheet simulation tool and validate the flowsheet with mill-classifier circuit data
In the first year of the project (2020) batch grinding tests and powder flowability measurements of the product were conducted in order to assess grinding aid contribution to the breakage aspect of milling, without powder transport. The second project year (2021) focused on the air classification step of the circuit. Trials in two air classifiers, in laboratory and industrial scales, were conducted. It was compared which aspects of this process are influenced by grinding aids and which are determined by machine design.
The third project year (2022) dealt with two aspects. First, continuous milling trials in passage mode to study the effect of grinding aids on powder transport, mill holdup and process dynamics and stabilization. Second, modification of process models from the literature to consider the effect of GA. The main conclusion of the open-circuit milling trials can be summarized as:
- Powder flowability should be kept within an intermediary range (easy-flowing). Both excessive and too low flowability should be avoided in order to improve throughout and process stability.
- High flowability can be detrimental to the amount of stressed material and energy transfer from ball to product particles.
- Beyond flowability GAs should be selected in order to reduce caking on equipment surfaces and reducing ball coating. Once an amount of powder is stressed between two balls, it should be readily fall off to allow another sample of powder to be stressed.
The mill model proposed is formulated as mechanistic population balance for media mills capable of predicting grinding of particles sizes from the lower millimeter size range down to the sub-micrometer scale. The mechanistic approach to media mill models is a very flexible tool that allows full separation of material and process aspects. The proposed mill model assumes stead-state operation, requires input from Discrete Element Method (DEM) simulations and accounts for impact of powder flowability on powder stressing.
Wet granulation is a key process used to make formulated particulate products across a wide range of industries. Granular products typically have at least one desired function, and in many cases there are several key performance characteristics which are required. Recent work has shown great improvement in the ability to model granulation process to predict granular properties such as size, however the ability to predict granular function is lacking, as is the ability to design processes to give desired granular function.
The primary aim of this work is to develop linked process and product performance models for wet granulation, and to initiate the inverse problem solving process, i.e. to investigate the ability to predict required process parameters to give desired performance characteristics. This is being performed for a case study of a high shear wet granulation process, coupled with a new model which describes granule disintegration.
Due to the relative immaturity of granular product performance models, much of the focus of this work has been on the development of a model to describe granule disintegration. Of particular importance is the suitability of this model for coupling with existing population balance models to enable model linking.
In this report, an improved model for granule disintegration is presented, which has been simplified to reduce the number of parameters required. A local sensitivity analysis is shown, which shows that decreasing granule porosity and constituent particle size contribute to smaller granule populations over time, due to an increased number of breakage events. Increasing the maximum absorption ratio of disintegrants in the model acts to decrease particle size. The effect of starting granule size is somewhat more complex, but indicates a potential threshold in normalized granule size behavior, above which the normalized size distribution becomes independent of the starting granule size. This however requires further research to confirm.
Initial experimental validation has been presented using a bespoke flow cell, optical microscopy and Optical Coherence Tomography (OCT), alongside a new image analysis app to provide data required for model parameterisation and validation. Preliminary parameterization has been performed, and a good fit to experimental data is demonstrated, however further work is required to verify, validate and parameterise the model.
A summary of the mechanistic high shear wet granulation model is presented, which is well developed and implemented in gPROMS FormulatedProducts. Tasks for the remainder of this project will focus on experimental validation of the new disintegration model, global sensitivity analysis, linking of the process and product performance models, and inverse problem solving. Additional resources at the University of Sheffield and the University of Strathclyde are being used to assist in the experimental validation, global sensitivity analysis and inverse problem solving.
In the first year of this project, we had derived a mechanics-based model for the feed rate in a single-screw feeder, making several simplifying assumptions. The model makes the prediction that the feed rate is maximum at a particular value of the ratio of pitch to diameter p/2R of the screw. This prediction matches exactly with the results obtained by DEM simulations under the same conditions. When the simplifying assumptions are relaxed in the DEM simulations, the dependence of the feed rate on p/2R was found to be qualitatively similar, suggesting that the simple model captures the essential physics of the problem.
In the second year, our work was extended in several directions:
- experimental measurement of the feed rate for different p/2R, the stress at the barrel surface using sensitive force sensors, and the velocity profile adjacent to the (transparent) barrel surface by flow imaging;
- application of a newly developed non-local constitutive model to the screw feeder problem and solving the governing equations to obtain the velocity and stress fields;
- DEM simulations to study the effect of particle cohesion on the feed rate and obtain the detailed spatial variation of the stress and velocity in the particulate medium.
Overall, we found good agreement between results of the DEM simulations, model predictions, and experimental data. The important conclusion was that combination of the three components of our investigation, namely theoretical analysis, DEM simulations and experiments, led to substantial insight.
In the third year (for which this report is written) we have further extended our experimental studies in some directions: we first completed the determination of the feed rate for a larger range of p/2R for glass beads and confirmed the existence of a maxima in the feed rate at an optimum value of p/2R. We then measured the feed rate for two cohesive powders – though measurements for sufficient large p/2R are yet to be made, the data strongly suggest the presence of the maximum in the feed rate. The experiments also throw light on the feed rate fluctuations, which are quite different for non-cohesive and cohesive powders. Our earlier DEM simulations restricted the feeder length to one pitch and assumed periodic inlet and outlet conditions. We have now conducted simulations for the full feeder, from the inlet hopper to the feeder exit. The results show a gradual fall in the fill level with axial distance from the inlet, as observed in the experiments.
Our ongoing work is to obtain solutions of the non-local model for the more general cases of finite screw friction in the presence of gravity. We are measuring the feed rate of cohesive powders for a large range of p/2R to confirm the presence of the maxima. We will soon conduct DEM simulations for cohesive powders for the full length of the feeder.
This report summarizes the main achievements during the year 2022 of the project with the aim of developing process systems engineering approaches for improved crystal size and purity control during crystallization processes. The successful crystallization process and system design requires an interdisciplinary effort, which ranges from population balance model (PBM) development of the system concept, through efficient implementation of model equations to soft-sensor development, which is required for the model predictive control (MPC) design as well. This report gives a deeper insight into these interdisciplinary development efforts, which also highlights the achievable improvements enabled by the combination of process modeling, high performance process simulation and optimization.
This year was focused on designing a novel, integrated crystallization system capable of establishing increased control of the properties of crystalline materials. The attainable region of crystal size distribution (CSD) is widened by the application of a recirculation stream and multiple MSMPR units as well as the integration of a downstream wet mill and classification units with recirculation stream(s) for continuous operation. This approach was then applied in finding an attainable region for a commercial active pharmaceutical ingredient (API) to observe real life application of the system. Also, a Quality-by-Control (QbC) guided framework is developed for crystallization processes to meet the CQAs of studied processes. QbC allows for control over techno-economics of the system through productivity and yield in addition to properties such as crystal size and narrow distribution.
The last part of this work focuses on production of Atorvastatin calcium with higher yield and lower cost. To address the limitations that the conventional batch manufacturing possesses, operating end-to-end in a continuous mode to shorten and strengthen product supply chains and add agility and flexibility to manufacturing is proposed. This leads to an integrated cascade of crystallizers that control polymorphism and agglomeration to be the connection between reaction and continuous filtration and drying in carousel (CFC) device.
- Development of an integrated continuous crystallization system, wet mill, classifiers with recycle
- Development of attainable regions for model and commercial compounds
- Robustness studies via kinetic parameter uncertainties and inlet seed distribution
- Validation experimental results for continuous crystallization with wet mill, classifier and recycle
Achieved Deliverable
In this report, we provide an update on the progress of the second part of the IFPRI Robin. In our Part 1 report, we quantitatively assessed the effectiveness of discrete element method (DEM) calibration methods utilised by 8 industrial DEM practitioners for a number of differing experimental geometries, particulate media, and combinations thereof. The accuracy of the methods was assessed by comparing the outputs of simulations performed following the procedures of 8 industrial participants with detailed experimental data produced using Positron Emission Particle Tracking (PEPT), a technique which allows the dynamics of particulate systems to be imaged, in three dimensions, with sub-millimetre spatial resolution and sub-millisecond temporal resolution. Strikingly, of all the participants surveyed, no two institutions adopted the same practices, highlighting the need for a more standardised approach and best practice. Our results showed that while most contemporary calibration methods were able to successfully capture the dynamics of simple, free-flowing, spherical particles under low-shear conditions, and a reasonable percentage of participants could correctly predict the dynamics of angular particles, the majority of procedures tested were unable to correctly reproduce the behaviours of smaller, more cohesive particles, or higher-shear environments. For the latter case, though qualitative agreement and visual similarity between simulated and experimental systems could be observed, deeper and more quantitative analysis using PEPT revealed significant disparities. Of the calibration methods examined, the most effective – indeed the only one to consistently reproduce the experimentally-measured dynamics of the cohesive systems tested – involved the combination of both static and dynamic powder characterisation tests, suggesting this to be the best practice for multi-parameter DEM calibration.
In the second part of the project, we will assess the ability of DEM, and the practitioners thereof, to handle a series of still more complex particles, including binary systems whose components possess strongly differing PSDs; fine particles (both free-flowing and cohesive); and highly elongated particles. We will also explore additional industryrelevant test systems (a Resodyn acoustic mixer and a Pascall mixer), and create additional digital twins of characterisation tools used by our industrial project partners (a Granutools GranuPack, Granutools GranuFlow. aerated Freeman FT4, and Anton Paar powder rheometer). We will also use detailed sensitivity analysis to assess the suitability and efficacy of all characterisation tools explored for the determination of different DEM parameters. This brief report highlights progress made so far toward these aims, and showcases a selection of the new tools developed and data obtained. All tools and data are available, free and open source, to IFPRI members upon request.
.
Our goals within the IFPRI project are threefold
- To explore how, moving away from model systems containing spherical colloids with near hard interactions, we can widen the range of rheological responses by changing the properties of the building blocks of the suspensions, so that even in simple formulations a wide range of behaviors can be ”built in”, i.e. obtaining formulation guidelines to do “more with less” or simplifying formulations from within. The properties aimed for, after discussing with IFPRI members, are the control of shear thinning/thickening and the control of the thixotropic response.
2. To further develop a limited number of rheological and structural tools to interrogate the rheological response of the such dispersions, focusing on
- Advanced rheological methods which allow for stress deconvolution such as high frequency rheometry and superposition rheometry, which help identify the nature of the stress during flow (elastic or viscous), which helps to identify which aspect of the particles or formulation controls the rheology
- High resolution confocal microscopy to probe structural development in situ during flow (4D imaging)
- Local scale tribological measurements using AFM.
3. Apply these methods to simplified industrial dispersion by industrial partners and compare with the formulation guidelines obtained from (1).
The present report discusses the progress made in the last 12 months.
The focus of the report is on development of a contact model for usage in CFD-DEM simulations. Great effort is placed in this step since it provides a basis for all future results upon which the continuum model will be built.
We present a contact model able to capture the response of interacting adhesive elastic-perfectly plastic particles under a variety of loadings. The model is built upon the Method of Dimensionality Reduction which allows the problem of a 3D axisymmetric contact to be mapped to a semi-equivalent 1D problem of a rigid indenter penetrating a bed of independent Hookean springs. Plasticity is accounted for by continuously varying the 1D indenter profile subject to a constraint on the contact pressure. Unloading falls out naturally, and simply requires lifting the plane indenter out of the springs and tracking the force. By considering the incompressible nature of this plastic deformation, the contact model is also able to account for the nonlocal effects of neighboring contacts, including formation of secondary contacts from outward displacement of the free surface. JKR type adhesion is recovered easily by simply allowing the springs to ‘stick’ to the 1D indenter’s surface. Additionally, we account for the rapid stiffening in the force-displacement curve under high confinement (e.g. during powder compaction) by triggering a superimposed bulk elastic response based on a simple criterion related to contact area.
Given that the model arises from rigorous contact mechanics formulations and simple geometric arguments only a few physical inputs are needed to run the model. Namely, the average radius of the particles Ro, Young’s modulus E, Poisson ratio ν, yield stress Y, and effective surface energy Δγ. An outline of the numerical implementation of the model is included. Every aspect of the contact model is validated, either through comparison to finite element simulations or in the case of adhesion directly to the JKR theory of adhesion. These comparisons show that the proposed contact model is able to accurately capture plastic displacement at the contact, average contact stress, contact area, and force as a function of displacement under a variety of complex loadings. This gives us confidence in the predictive capability of the contact model and its ability to reflect reality when used in the upcoming CFD-DEM simulations.
This annual report presents key advances made during year 2 of which, a major component is a concise treatise on our key advances in model-guided dry coating-based enhancements of poor flow and packing of fine cohesive powders, included in Appendix A. The report also includes a review of the available particle contact models for both smooth and rough particles and presents a database of industry relevant materials and their key properties. In terms of IFPRI Member interactions, we have held regular update meetings and worked with a member company on their powder characterization device. We plan to prepare a manuscript on that and include that in the report next year.
Major Accomplishments
Major accomplishments include a review of the existing van der Waals force-based particle contact models to elucidate the main mechanism of flow enhancement through silica dry coating. Our multi-asperity model explains the effect of the amount of silica, insufficient flowability enhancements through conventional blending, and the predominant effect of particle surface roughness on cohesion reduction. Models are presented for the determination of the amount and type of guest particles, and estimation of the granular Bond number, used for cohesion nondimensionalization, based on particle size, particle density, asperity size, surface area coverage, and dispersive surface energy.
Processing Conditions
Selection of the processing conditions for LabRAM, a benchmarking device, is presented followed by key examples of enhancements of flow, packing, agglomeration, and dissolution through the dry coating. Powder agglomeration is shown as a screening indicator of powder flowability (Appendix B). The mixing synergy is identified as a cause for enhanced blend flowability with a minor dry coated constituent at silica <0.01%. The analysis and outcomes presented in this paper are intended to demonstrate the importance of dry coating as an essential tool for industry practitioners.
Objectives:
- To establish a predictive criteria
- To identify the key factors
- To establish a test methodology
The original aim and objectives of the project remained unchanged: to design a diagnostic tool to determine if a powder formulation will stick to the punch-face during tablet production.
Timeline
Work on the project started effectively in August 2021 with the arrival of PhD student Ahmad Ramahi. In February 2022 PhD student Vishal Shinde joined the project. The first two objectives have been completed and work on the third objective is ongoing aiming to complete by the AGM in June 2023.
Approximately 20 characterisation techniques were employed or explored at different levels of detail as described in this report. Measurement of the % area of the punch covered by the sticking powder was selected as the main method to quantify sticking. Following regular monthly project meeting with the IFPRI advisory group, given the complexity involved in the sticking phenomena (summarised in Section 1), the focus was maintained on empirical studies to identify the relevant mechanisms for the materials of interest and on creating a database of approximately 20 powder materials (including sticking and non-sticking APIs, sticking and non-sticking excipients, powder formulations and lubricated formulations), and delve into the science of each mechanism in the follow-on proposal.
The database contains material properties including chemical information (formula, structure, molecular weight), physical characteristics (particle size distribution, density, shape, morphology of the particles, bulk density of powders), mechanical properties of particles (Young’s modulus, Poisson’s ration, yield strength), interaction properties between particles (friction coefficient between particles, surface energy), thermal properties (conductivity, heat capacity, thermal expansion coefficient) and humidity related properties (water adsorption isotherms, water activity).
The sticking behaviour of powders during single compression events is characterised considering 4 processing parameters: compaction pressure, temperature, humidity, and compaction rate. Work is ongoing, only 4 materials were characterised so far. The compression tooling used was provided by iHolland (B tooling). Long term sticking (multiple compaction events) are planned.
The deliverable the end of the 3 years is a predictive toolkit comprising of the database analysed using Principal Component Analysis to extract functional relationships between material properties, process parameters (compaction pressure and rate) and environmental conditions (temperature and RH) and finally assign a risk for sticking.
A proposal is being developed for a follow-on project to extend the database for new materials to further validate the predictive capability, establish the science base to understand the underlying mechanisms, link molecular level information to sticking behaviour and develop mitigating strategies for sticking at early stage product development. A collaboration with Professor Jerry Heng at Imperial College is proposed to cover the Chemistry/Chemical Engineering aspects.
In the current funding period we have made advances on three fronts:
- use of kinetic Monte Carlo (kMC) simulations to predict the morphology of organic crystals grown from solution for cases where the solution is pure and also when it contains growth inhibitor molecules,
- completed incorporating three new all-atom force fields into ADDICT in order to test the sensitivity of our morphology predictions with respect to different estimates of intermolecular interaction energy, and
- developed a new crystal growth model for asymmetric organic molecules with two molecules in the unit cell. This is a precursor to developing a more general model with many molecules in the unit cell.
We report progress on all three topics.
We have continued to develop our kinetic Monte Carlo (kMC) modeling codes to predict the morphology of organic crystals grown from solution both with and without the inclusion of impurity molecules on the crystal surfaces. We have used these codes to make morphology predictions for naphthalene grown from ethanol solvent at increasing supersaturations in impurity-mediated solutions. The results were quite satisfactory and are reported in more detail later in the report.
In addition to the Generalized Amber Force Field (GAFF) which is already included in ADDICT, we have added three new all-atom force fields to ADDICT. They are the Coulomb-London-Pauli (CLP) force field, the Consistent Force Field (CFF), and the Universal Force Field (UFF). Each has its own specific advantages and limitations.
- CLP is the most general with over 90 atom types (similar to GAFF).
- CFF is specific to organic molecules containing intermolecular hydrogen bonds, especially carboxylic acids and amines.
- UFF has one atom type for every element in the periodic table - it is very general but with only one atom type for each element it is not very accurate.
Later in the report we describe these force fields in more detail and show new results for predicting the morphology of five organic crystals using the CLP force field. We continue to develop a new crystal growth model for asymmetric organic molecules under the restriction of that the unit cell contains exactly two molecules. This allows us to make a big leap from essentially one (symmetric) molecule per unit cell to two asymmetric molecules. Once this theory is fully tested and validated on real molecular systems it will lead to further extensions to 4 molecules in the unit cell, and then many.