Improved particle characterisation from in-line PAT : comparison of deep learning and white-box methods

Boyle, Christopher and Brown, Cameron and Sefcik, Jan and Cardona, Javier (2022) Improved particle characterisation from in-line PAT : comparison of deep learning and white-box methods. In: AIChE Annual Meeting 2022, 2022-11-13 - 2022-11-18, Phoenix Convention Center.

[thumbnail of Boyle-etal-AIChE2022-Improved-particle-characterisation-from-in-line-PAT]
Preview
 Text. Filename: Boyle_etal_AIChE2022_Improved_particle_characterisation_from_in_line_PAT.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (1MB)| Preview

Abstract

In-line Process Analytical Technologies (PAT) are useful for measurement of particle characteristics (e.g. particle size distribution, PSD) in a non-destructive manner and with high time resolution inaccessible with off-line techniques. These in-situ, time-varying measurements can be essential for process monitoring and accurate population balance modelling. This work is concerned with assessing the performance of using in-line image and chord length distribution (CLD) analysis for PSD measurement. Imaging is obtained through in-line microscopy (e.g. BlazeMetrics’ Blaze 400, or Mettler Toledo’s Particle Vision and Measurement, PVM), while CLD can be obtained via light-scattering methods (e.g. Mettler Toledo’s Focused Beam Reflectance Measurement, FBRM). Sensor data is analysed through different methods to yield PSDs (amongst other key particle characteristics). Traditional white-box (i.e. not machine learning) analyses can involve length “tuning” steps wherein the analysis is adapted to work with the system (Cardona et al. 2018), or are derived mathematically and are restricted to a small domain (Agimelen et al. 2015). Despite these drawbacks, these techniques are demonstrably effective, once tuned or when applied to the appropriate domain. Deep learning (black box) models have potentially increased performance and flexibility over white box techniques. This is in part due to their large number of parameters and due to modern breakthroughs in model design (He et al. 2017). Deep learning models are widely used in image analysis, with further applicability to other domains such as CLD transformation. Different PAT sensors perform differently under different conditions. For image-based measurement: small particles may not be measurable with great accuracy due to resolution issues, concentrated systems will impact measurement of the particles due to particle overlaps, and large particles are more likely to interact with the edge of the image field of view and therefore are less likely to be sized resulting in left-skewed PSDs. For FBRM-based measurements: translucent particles cause a phenomenon called “chord splitting” wherein a single chord is mis-registered as multiple smaller chords resulting in modes appearing at lower bins than expected on the PSD. In addition, low concentration systems (small number of particles) can result in few counts in the CLD and present as noise and thus impacting the transformation from CLD to PSD.

ORCID iDs

Boyle, Christopher ORCID logoORCID: https://orcid.org/0000-0001-8926-6590, Brown, Cameron ORCID logoORCID: https://orcid.org/0000-0001-7091-1721, Sefcik, Jan ORCID logoORCID: https://orcid.org/0000-0002-7181-5122 and Cardona, Javier ORCID logoORCID: https://orcid.org/0000-0002-9284-1899;
CORE (COnnecting REpositories)