An Algorithm to Identify Material Mixtures from Raman Spectra

This challenge is externally hosted.

You can view the challenge details here: https://www.innocentive.com/ar/challengeSeeker/details/9933552

Description

Background Raman spectroscopy is commonly utilized in the field for the identification of bulk powders due to its reliability and high specificity; however, this specificity poses a problem when the unknown bulk powder is a mixture. Each mixture produces unique spectra at different concentrations (non-linear) and current algorithms or systems are not proficient at identifying mixtures based on the pure components recorded in its library. In the field of explosive detection, this poses a problem when dealing with homemade explosives (HMEs) especially oxidizer/fuel mixtures. Vendors have attempted to address this issue with their Raman systems; however, they have not yet found a robust solution.

The Challenge From this Challenge we ideally seek to receive an algorithm that can be translated into a software update for any commercial Raman system in the field. This solution will allow the system to better identify mixtures and preferably be able to calculate the ratio of the components in a 2 component mixture. Solvers are challenged to come up with an algorithm that can identify the referenced 2 component mixtures by utilizing the spectra for the individual pure components in the library and not a library of mixture spectra. The algorithm should still identify pure components correctly.  In addition, it would be nice to have their relative concentrations from the Raman spectra, but it is not absolutely required as some assumptions about ratios can be made.

Typically for oxidizer and fuel mixtures the Raman signal is dominated by the oxidizer signal. The identification of a mixture with both fuel and oxidizer requires identification of the fuel component.

Solvers are not obligated to use this approach, but a potential solution could include saturating the oxidizer signal and running an algorithm to identify the smaller return from the fuel component, while ignoring the saturated peaks.

Rules

The submitted proposal should include the following:

1. A detailed description of the proposed algorithm/system and how it was derived.

2. Rationale as to why the Solver believes that the proposed algorithm/system will work. Goodness of fit alone is not guaranteed an award. The Solver needs to explain their algorithm and the reasoning behind it based on sound principles. The Solver should address each of the Solution Requirements described in the Detailed Description.

3. Demonstration – The Solver should show the goodness of fit using the given data and demonstrate the algorithm works.  The algorithm should correctly identify the spectra as the mixture component rather than the pure components at least 95% of the time.

4. The Algorithm itself should be delivered in a file format readable by the Seeker. This could be done by delivering the source code or an executable file. If necessary, instructions should be provided on how to run the solution.  Permitted programming languages are C, C++, C#, Java, MatLab, and Python.  The Seeker is flexible in the regard, and other languages may be acceptable with prior acceptance by the Seeker. If the Seeker cannot read the format, the submission will not be evaluated or awarded.

The proposal should not include any personal identifying information (name, username, company, address, phone, email, personal website, resume, etc.) This Challenge will be awarded based on deliverables. For an algorithm that can meet the requirements the award amount is $25,000. The award is contingent upon theoretical evaluation and experimental validation of the submitted Solution by the Seeker. The algorithms will be tested on the Seeker’s data of known mixtures. To receive an award, the Solvers will have to transfer to the Seeker their exclusive Intellectual Property (IP) rights to the solution. The Seeker is willing, in return, to grant a license to the Solver so they can continue to use the solution.  However, the Seeker will be willing to consider a licensing agreement for a partial award, if exclusive IP cannot be granted by the Solver.