The development and application of chemometrics to process analysis in an industrial environment

This thesis describes two main sections of work, an examination of a commercial product, Intrasite Gel, and the development of an algorithm for variable selection using projected latent structures.Following on from the successful development of a variable selection procedure for multivariate linear regression this work looks at transferring this idea for use with projected latent structures. The first part of this thesis will show how the variable selection algorithm was developed and used with three different data sets. The algorithm will be shown to be superior to standard projected latent structures, for linear multi-component data. Although the final algorithm developed requires considerable computing resources to carry out this is compensated for by significantly improved model predictions and robustness. The final algorithm developed is written to run using MATLAB on any computer platform that supports this application, though the principles of operation could be transferred to another method of execution, for example custom code written in C or Pascal. The approach used in the development of this method is that the ability of the model to predict unknownsamples is of far greater importance than the internal performance of the model. All the assessments of the procedures developed are based on the ability of the model to predict accurately and precisely samples that were not presented to the model during the training stage.The second section of this thesis is concerned with the study of Intrasite Gel, produced by Smith & Nephew Ltd. Hull. The material in question is a medical device intended to assist in the treatment and healing of wounds that are necrotic, sloughy or granulating. The product is characterised by its ability to maintain moisture equilibrium in a wound environment and to provide a suitable medium to encourage the growth of new cell tissue. Medical devices require registration, and as part of that registration a number of tests are made on samples to ensure that the material meets the required specifications. There was some concern at Smith & Nephew that the tests they were required to carry out as part of the device registration were not providing appropriate information about the product. Of particular interest was the fluid absorption property as it was suspected that the test has a large amount of random error associated with it and an investigation was required to examine this test and to provide an alternative procedure should the fluid absorption test prove inadequate. Also of interest to Smith & Nephew was the issue of sampling frequency, as it was felt that this should also be examined to determine whether the correct rate of sampling to ensure product quality was being carried out. The work reported here shows that the fluid absorption test as it stands is insufficient to the task of monitoring this property of Intrasite gel and that an alternative test should be considered. This work also showed that current sampling rate was too high and that the high sampling rate may in fact cause misleading assumptions as to the stability and quality of the product