All materials, natural or human-made, contain unique traces of their origin and history as represented by thousands of variables. Materialytics® detects these traces to characterize unclassified samples of material. The process is automatic and objective, without reliance on opinions or doubtful documentation. Accuracy is high because the system uses tens of thousands of variables, rather than dozens.
Materialytics analysis compares an unclassified sample of material with reference collections of well-classified and well-documented materials. Further analysis determines which class of materials the new sample best matches.
Collection and database building start with gathering a statistically significant number of reference samples (bare minimum of 30) from each class of interest. They might be from a mine, a manufacturer, a supplier, a coffee farm, a particular production line, or an archaeological site. Each set is a class of materials determined by human judgment to have something in common.
Each reference sample is subjected to physical analysis, typically producing millions of data points per sample; tens of millions of data points per class.
From this big data the system determines the signature of the material in each class. Observers have commented that the signature is as distinctive as an individual’s DNA.
The signature of each unclassified sample is compared with the signatures of all classes in the reference database. The best match of the unknown to any class is reported as a probability. For example:
- The system is 99.3 percent confident that the sample matches Class X, or
- The system cannot match this test sample with a high degree of confidence to any particular class in the reference database, so it has probably not encountered this material before.
The system learns empirically, not guided by theory. The more robust the reference collection/database, the better. The accuracy of the system increases as the size of the database increases.
After creation of the reference database, comparison results with unclassified samples can be available rapidly.