Learning the Treatment Effects on FTIR Signals Subject to Multiple Sources of Uncertainties

TL;DR

This paper introduces a two-step optimization method to quantify treatment effects on FTIR signals under uncertainties, validated through simulations and a real case study involving plasma exposure effects.

Contribution

It proposes a novel two-step optimization approach to decompose and interpret treatment effects on FTIR signals considering multiple uncertainties.

Findings

Method accurately estimates treatment effects in simulations.

Real case study shows plasma height influences FTIR signals as expected.

Pattern of modification aligns with engineering knowledge.

Abstract

Fourier-transform infrared spectroscopy (FTIR) is a versatile technique for characterizing the chemical composition of the various uncertainties, including baseline shift and multiplicative error. This study aims at analyzing the effect of certain treatment on the FTIR responses subject to these uncertainties. A two-step method is proposed to quantify the treatment effect on the FTIR signals. First, an optimization problem is solved to calculate the template signal by aligning the pre-treatment FTIR signals. Second, the effect of treatment is decomposed as the pattern of modification $\mathbf{g}$ that describes the overall treatment effect on the spectra and a vector of effect $\boldsymbol{\delta}$ that describes the degree of modification. $\mathbf g$ and $\boldsymbol{\delta}$ are solved by another optimization problem. They have explicit engineering interpretations and provide useful information on how the treatment effect change the surface chemical components. The effectiveness of the proposed method is first validated in a simulation. In a real case study, it's used to investigate how the plasma exposure applied at various heights affects the FTIR signal which indicates the change of the chemical composition on the composite material. The vector of effects indicates the range of effective plasma height, and the pattern of modification matches existing engineering knowledge well.