AI-Based Particle Analysis & Particle Measurement

In many industries, product quality stands and falls with particles – in pharmaceuticals, food technology, or special materials such as paints or pigments. Capturing high-resolution images using light and electron microscopy is only the first step. The real challenge lies in the evaluation: How can we quickly, accurately, and reproducibly convert a particle image into valid data on particle size and shape?

The status quo: Manual evaluation. An experienced specialist often needs 30 to 60 minutes to accurately measure just 100 particles. Regulatory requirements exacerbate the problem, as a minimum of over 1,000 particle detections are required for statistical significance in safety assessments.

Initial automation: Classic image processing methods. Rule-based algorithms often fail in practice: as soon as particles significantly overlap or contrast fluctuates, these methods see only light and dark pixel values – not physical objects.

The solution of the future: AI-based object recognition. A neural network learns from examples and recognises particles even at extremely high densities or with high noise. The effort for initial training pays off quickly – the output is fully automated analysis in real time.

Direct images instead of indirect signals. Laser-based systems measure particle sizes via scattering patterns and thus provide indirect approximations. Image-based sensors deliver real images of the particles directly in the process – allowing not only size, but also shape, structure, and agglomeration to be detected.

Shape and structure are crucial. In many chemical and pharmaceutical processes, the shape of particles is just as important as their size. Image-based systems provide clear differentiation and quantifiable data on all relevant particle types. Foreign particles or unwanted inclusions can also be detected, significantly improving quality assurance.

Resistance to harsh production conditions. Image-based inline systems continuously detect particles directly in the process – allowing aggregation, dispersion, or breakage to be detected immediately and process adjustments made in real time.

AI-supported analysis. Image-based systems can use artificial intelligence to automatically detect and classify particles, immediately quantifying parameters such as size, shape, aggregate state, or foreign particles.

Optical measuring devices all work on the same principle: a light beam is sent into the fluid and altered there. A detector records the change.

Challenges in implementation: The optical access to the apparatus must be kept clear at all times. The sensitive electronics must be shielded against high temperatures and humidity. The alignment of the optical axis must remain stable even under vibrations.

Particularly tricky: choosing the right light source. The lighting must be strong and homogeneous, while exposure times must be kept short to avoid motion blur when fluids are moving.

Conclusion: Image-based optical inline measuring devices are most worthwhile in processes with high savings potential, in sensitive process steps, and in safety-relevant plant components.

Particles can be so much more than solid chunks in a stream! Whenever a second phase is created due to a different state of aggregation, we refer to it as a particle flow or multiphase flow – this can be a solid particle in a flowing liquid, a bubble in liquid, a droplet in a gas stream, or a droplet in an immiscible liquid.

All the magic of chemistry and process engineering takes place at the interface between medium and particle: heat transfer, reaction, and changes in concentration. Without particle flows, many reactions would not work.

With our images, we provide the key data: How many particles are there? How big are they, what shape do they have, and how do they change over time?

During crystallisation, crystals are formed from a liquid solution through cooling or evaporation. Which crystal shape will emerge is difficult to predict: different shapes can form depending on the solvent, stirring speed, and cooling rate. Crystal breakage, agglomerates – all these phenomena overlap in the crystalliser.

Crystal shape is decisive for all subsequent steps: filtration, drying, and formulation (tablet, powder, fertiliser pellet) only work well if the crystal shape is correct. Uniform, large crystals filter better; round particles flow well, while needle-shaped crystals tend to clump.

If parameters are not correct, crystals must be discarded or re-dissolved – an expensive, time-consuming process. That is why we want to measure these parameters directly during crystallisation: less waste, fewer process disruptions, higher product quality.

In separation apparatus (distillation, extraction), phases must be separated again at the outlet. For perfect separation, the apparatus would have to be infinitely large. The compromise: a small part of the foreign phase is carried along – this is entrainment.

Entrainment causes a loss of valuable material, increases energy consumption, and can damage downstream equipment such as pumps or compressors. Even installed separators eventually reach their limits and "break through".

Conclusion: Entrainment cannot be prevented and is factored into plant design. However, if it is greater than calculated, performance losses are to be expected. Real-time monitoring of entrainment in critical process steps is therefore strongly recommended!

A general distinction is made between 4 types of measurement data acquisition: OFFLINE (sample to the laboratory), ATLINE (analysis close to the plant), ONLINE (sample loop), and INLINE (measuring device directly in the plant – the supreme discipline).

Inline means: real real-time data without distortion at the point of action. Measurement variables that are easy to detect inline include temperature and mass flow. Particle size, however, is currently still often measured offline!

The result: no real-time data, no possibility to counteract incorrect particle size in the process, no data for non-samplable particles (droplets, bubbles) – and thus: unwanted plant downtime, disposal of failed products, trial and error in troubleshooting.

Therefore: Measure particle sizes inline – save money and valuable time!