Multi-sensor data fusion and parallel factor analysis reveals kinetics of wood weathering

Jakub Sandak,Anna Sandak,Marina Cocchi

doi:10.1016/j.talanta.2020.122024

Jakub Sandak, Anna Sandak + Show 1 more

Open Access

https://doi.org/10.1016/j.talanta.2020.122024

Copy DOI

Abstract

Understanding mechanisms of materials deterioration during service life is fundamental for their confident use in the building sector. This work presents analysis of time series of data related to wood weathering acquired at three scales (molecular, microscopic, macroscopic) with different sensors. By using several complementary techniques, the material description is precise and complete; however, the data provided by multiple equipment are often not directly comparable due to different resolution, sensitivity and/or data format. This paper presents an alternative approach for multi-sensor data fusion and modelling of the deterioration processes by means of PARAFAC model. Time series data generated within this research were arranged in a data cube of dimensions samples × sensors × measuring time. The original protocol for data fusion as well as novel meta parameters, such as cumulative nested biplot, was proposed and tested. It was possible to successfully differentiate weathering trends of diverse materials on the basis of the NIR spectra and selected surface appearance indicators. A unique advantage for such visualization of the PARAFAC model output is the possibility of straightforward comparison of the degradation kinetics and deterioration trends simultaneously for all tested materials.

Highlights

Weathering is a natural process occurring to all materials exposed to environmental conditions
This paper presents an alternative approach for multi-sensor data fusion and modelling of the deterioration processes by means of PARAllel FACtor analysis (PARAFAC) model
This paper presents the results regarding in service performance of modified wood during natural weathering test

Summary

Introduction

Weathering is a natural process occurring to all materials exposed to environmental conditions. It is expected that combining data from complementary analyses will improve the performance of statistical models Such multi-sensor monitoring introduces new issues and challenges in the data analysis phase, where the strategy for merging different sources of information is fundamental [5,6,7]. Data acquired by diverse sensors usually rely on the measurement of different physical phenomena that, in the majority of cases, are not directly correlated to each other or respond with different kinetics to material alterations. Both data integration as well as interpretation of obtained results are challenging.

Methods

Results

Conclusion