Discuss your analysis
APPLICATION
Foreign particle investigation
An unexpected particle count becomes a chemical population fingerprint: identify dominant material classes, compare suitable references, and assess a probable source to focus CAPA and root-cause work.
Small sample sets are typically reported within 1–3 business days, including final review.
Discuss this application
CUSTOMER QUESTION
What changed, which materials dominate, and what should we investigate next?
Start with the investigation
A routine count increase, unexpected particles or a changed batch signature calls for more than another count.
Ask whether the population resembles tubing, packaging, stopper, silicone, formulation material or process residue.
The answer you receive
A defined population with counts, size classes and particle images.
Raman material classifications linked to individual measurements, including unresolved outcomes.
Comparison with suitable customer and material references.
Reviewed interpretation and focused next steps for root-cause/CAPA work.
Result: the investigated population was strongly PET-dominated, with PET accounting for 80.8% of Raman-analyzed particles.
Example A · Supplier A · n = 156 Raman-analyzed particles
PET · 126 · 80.8%
Polystyrene · 17 · 10.9%
Epoxy · 7 · 4.5%
Cellulose · 4 · 2.6%
Polyethylene · 1 · 0.6%
Unknown ester · 1 · 0.6%
This chemical population example and the cellulose reference study below are separate studies demonstrating two capabilities.
Identifying a particle as cellulose, polypropylene or PET may only be the first answer. SizeID.bio can compare unknown particles with customer-supplied process, packaging and device references to determine which reference fingerprint best explains the finding.
Reference-study result: six closely related cellulose source classes were distinguished with 96.4% accuracy on the held-out test dataset.
Unknown cellulose-type particle
Raman fingerprint and image-derived context guide reference comparison.
Reference library
Cellulose
Cotton fiber
Swab
Filter fiber
Lab-coat fiber
Paper fiber
Raman data → reference-trained classification
Probable source class
96.4%
Test-set classification accuracy
Balanced accuracy 96.7%
Cohen’s κ 0.957
Reference study using automated Raman to discriminate closely related cellulose source classes. Reported test metrics: CNN-LDA-PCA Raman model. Combined Raman and image-derived analysis is discussed separately as a future approach in the poster.
Source attribution strength depends on the available reference library, sampling design and process context.
Traditional database search
“What material is it?”
Fast first-pass chemical identification and a useful troubleshooting overview. Structurally similar cellulose sources can be difficult to distinguish reliably.
Reference-trained ML fingerprinting
“Which known source class does it most resemble?”
Higher discrimination between closely related reference materials, with additional training, planning and maintenance. Particularly useful for controlled in-house contamination libraries.
Reference study: Markus Lankers and Lia Ivanov, “Pinpointing Contamination Sources: Evaluating AI supported Spectroscopic Data Analysis Methods for Cellulose Fiber Contamination”, mibiC GmbH & Co. KG.
FOREIGN PARTICLE INVESTIGATION
From particle field to chemical identification
Particles are located in the population, targeted individually and linked directly to their Raman spectrum for material classification.
Analytical workflow
1 · Define the question
Clarify what happened, where the particle was observed and which suspected source materials may be available.
2 · Prepare and count
Prepare particle population for imaging and Raman targeting.
3 · Automatically target Raman
Apply the agreed targeting logic, classify Raman results and compare suitable reference materials.
4 · Compare the population
Summarize particle classes and investigation-relevant patterns.
5 · Review and report
Technical and independent quality review connect the evidence to practical follow-up options.
Deliverables
Defined particle counts, size-class results and a complete Raman class summary.
Linked particle images, calibrated morphology and measured Raman spectra.
Reference comparisons and representative particle evidence.
Specialist interpretation with supporting raw data as agreed in the study scope.
Technically reviewed and independently quality-reviewed reporting.
Rapid screening, with deeper investigation when needed
A rapid screen may target, for example, the 50 largest particles using a defined Raman wavelength. For suitable samples, a single wavelength can identify the large majority of targeted particles. Fluorescence, metals, weak Raman response, spectral quality and matrix interference can limit identification.
Unknown particles can be compared with customer-supplied materials and relevant spectral references. A sufficiently discriminating reference set and sampling design can support identification of a probable source and focused root-cause/CAPA work.
Automated particle Raman analysis