(507a) Surface Characterization and Quality Monitoring of Polymer Nanofibre Membranes Using Multivariate Image Analysis

In the manufacturing of several high-technology products quality monitoring is often carried out manually. Image analysis has been receiving a wider and wider interest either to build off-line systems for laboratory metrology, or to implement online automatic control schemes. In particular, multivariate image analysis (MIA) can be used to compact the most important features of the product in terms of few significant latent variables and few statistical indices, giving a full and meaningful characterization of the quality of the manufactured product [1]. In this way, real or virtual measurement of the relevant quality properties can be identified and related to the process parameters. This is important to limit the lab assays that may represent a considerable time and economic burden in a manufacturing process. Furthermore, several important qualitative features of the product, which are not directly measurable, can be highlighted by using a multivariate-multiresolution approach combining MIA with wavelet decomposition, in such a way as to assess the important qualitative attributes on different resolution scales [2].

A very promising application field is concerned with the production of high value added nanostructured materials, where equipment for the morphological, chemical and physical characterization is seldom available, and (when it is) is usually complex and expensive. In fact, microscopic analysis is often the only tool to describe and classify nanostructured materials, in which the most minute details play an essential role in the usability of the product [3]. Such issues are common to a variety of industries (e.g., advanced materials, pharmaceuticals, food, electronics, national security, etc?).

This work presents the application of multivariate and multiresolution techniques to the quality monitoring of polymer nanofibre membranes produced by electrospinning [4] through the analysis of scanning electron microscope images. In particular, the research work focuses on i) metrology issues, such as the measurement of the fibres diameter, the inter-fibre pore size, the probability distribution of the fibre diameters and of the pore sizes, and the orientation of fibres and pores; and ii) estimation issues, i.e. the monitoring of the most significant morphological and physical features through textural analysis of the image (e.g.: the estimation of the fibres diameter, of the inter-fibre pore sizes, and of membrane permeability).

The metrology system is developed following a complex algorithm, which comprises the following steps:

? filtering (e.g.: multiresolution denoising, multivariate image elaboration to enhance spatial information, etc?);

? projection of the image onto a latent space of scores via PCA (principal component analysis);

? segmentation of the latent image in different layers of fibres to consider the effects of the third dimension;

? measurement of the fibres diameter and of the pore size with a novel strategy.

The results are very satisfactory in terms of accuracy: the size of the pores, the diameter of the fibres and their probability distribution exhibit a measurement error comparable to the limit of resolution (linear dimension of 1 pixel).

A computationally faster estimation approach via multiresolution texture analysis [5] of the product surface is also devised. This approach is particularly effective if used simultaneously to the information from the analysis of the gray level co-occurrence matrix method, i.e.: homogeneity, contrast, etc? In this case, the output of the algorithm is quickly available (more then 1 image inspected/s) and the accuracy of the estimation is almost the same as the one of the measurements. The relative estimation error is lower than the dimension of one pixel in 90% of the cases, and lower then the dimension of 2 pixels for all samples. Moreover, the estimation system based on the multivariate and multiresolution textural analysis of the product can be used for the estimation of the permeability of the membranes. It is shown that this strategy can be applied to images from optical microscopes, too, and therefore it can be used in an automatic fashion for online analysis of images of lower resolutions. Finally, it should be highlighted that the same modeling structure can provide easily consultable monitoring charts to discriminate between membranes of different characteristics and uses. In a broader and more general extent, this work shows how automatic quality monitoring strategies based on image analysis can be developed and extended to new industrial fields and to the most critical inspection scales.

References

[1] Geladi, P. e H. Grahn (1996). Multivariate Image Analysis, John Wiley & Sons, Inc., New York (U.S.A.).

[2] Liu, J.J. e J.F. MacGregor (2006). On the Extraction of Spectral and Spatial Information from Images. Chemom. Intell. Lab. Sys., 85, 119-130.

[3] Facco, P., R. Mukherjee, F. Bezzo, M. Barolo e J.A. Romagnoli (2009). Monitoring Roughness and Edge Shape on Semiconductors through Multiresolution and Multivariate Image Analysis. AIChE J., 55, 1147-1160.

[4] Roso, M., S. Sundarrajan, D. Pliszka, S. Ramakrishna e M. Modesti (2008). Multifunctional Membranes Based on Spinning Technologies: the Sinergy of Nanofibers and Nanoparticles. Nanotech. , 19, 1-6.

[5] Bharati, M.H. e J.F. MacGregor (2004). Image Texture Analysis: Methods and Comparisons. Chemom. Intell. Lab. Sys., 72, 57-71.