(148c) Energy Dispersive X-Ray Hyperspectral Image Analysis and Chemometrics for Catalyst Characterization | AIChE

(148c) Energy Dispersive X-Ray Hyperspectral Image Analysis and Chemometrics for Catalyst Characterization


Gonzalez Martinez, J. M. - Presenter, Shell Global Solutions Int. BV
Prats Montalban, J. M., Universidad Politecnica de Valencia
Haswell, R., Shell Global Solutions International B.V.
Ferrer, A., Universidad Politécnica de Valencia
Multi-dimensional images with chemical information embedded at pixel-level (so-called Hyperspectral Imaging) is the next generation of Image Analysis and Machine Vision. This advanced technology has attracted the attention of different industry sectors in the last two decades, ranging from metalurgic, mining and pharma, to high-value-added manufacturing industries such as biotechnology and biomedical imaging. Recently, the combination of Scanning Transmission Electron Microscopy (STEM) and Energy Dispersive X-ray spectroscopy (EDX) has been developed to characterize the surface area of catalysts in the chemical industry. The measurements are done by scanning the electron beam over the sample and at the same time measuring the X-ray spectrum – so-called spectral imaging [1]. The spectral dimension of the hyperspectral images consists of X-ray counts acquired at different energy channels. Chemical data is corrupted with noise, which comes from the time-dependent arrival of discrete particles on the sensor. This noise typically follows a Poisson distribution, which represents the probability of occurrence or events (X-ray counts) during a given period of time.

The application of Multivariate Image Analysis (MIA) techniques and Multivariate Curve Resolution (MCR) models becomes essential for the analysis of EDX hyperspectral images. This contribution proposes a modeling framework that permits segregating hyperspectral X-Ray images into simpler images (so-called Distribution Maps, DM’s), which can be directly related to each of the chemical compounds present in the mixture [2]. From these DM’s, chemical-textural score images (SI’s) are further obtained. Finally, from all these DM’s and SI’s, different types of features, such as quantitative, morphological or textural can be extracted and combined into new data structures [3]. This new source of information is used to build multivariate statistical models for process understanding and prediction purposes at a MIA image-based level. This approach allows us to study similarities and differences between and within types of catalysts -i.e. different samples of the same catalyst across batches, and across locations, and the potential effects on quality properties of interest. To illustrate the chemometric framework for catalyst characterization, STEM-EDX images of real industrial catalyst will be used.

[1] R. Haswell, D. W. McComb and W. Smith, Preparation of site‐specific cross‐sections of heterogeneous catalysts prepared by focused ion beam milling, Journal of Microscopy (2003) Vol. 211, Pt 2, pp. 161-166
[2] J.M. Prats-Montalbán, A. de Juan, A. Ferrer, Multivariate image analysis: a review with application, Chemometrics and Intelligent Laboratory Systems, 107: 1-23, 2011.
[3] C Duchesne, JJ Liu, JF MacGregor, Multivariate image analysis in the process industries: A review, Chemometrics and Intelligent Laboratory Systems 117, 116-128, 2012.