Rishi Relan

The real-time fault monitoring and control of the Vacuum Assisted Resin Transfer Moulding (VARTM) production process requires a knowledge of the position of the epoxy flow-front inside the mould. Therefore, a fast and accurate flow-front tracking system capable of combining the underlying physics of the flow-front dynamics with the measured data is highly prized. Stochastic differential equations (SDEs) based grey-box models deliver a good trade-off between high fidelity models and data-driven black-box models for designing such a flow-front position tracking system. In this paper, we propose a simple yet novel coupled SDE based spatiotemporal grey-box model of the flow-front dynamics in case of missing sensor information. The proposed method uses the finite difference approximation of the spatial domain of the flow-front for estimating spatial flow pattern of the epoxy. Furthermore, to accommodate for the missing sensor data, we utilise a modified version of the continuous-discrete extended Kalman filter (CD-EKF) based estimation framework for SDEs that takes into consideration the effective dimension of the measurement space during the identification process. The performance of the method is evaluated for various common sensor faults scenarios at different levels of measurement noise and sampling rates.

For control and monitoring purposes, knowledge of the current state of the flow-front in a vacuum assisted resin transfer moulding (VARTM) process is essential. The permeability of the medium and viscosity of the epoxy can change during the infusion process. Especially for online monitoring of the infusion process there is a need for a fast and fairly accurate, possibly virtual sensor system which can handle such parameter variations. Stochastic-differential equations (SDEs) based estimation of the flow-front dynamics can offer a good trade-off between physics and data-driven estimators. In this paper, we analyze the effect of parameter variations on an SDE based spatio-temporal estimator of the flow-front dynamics in a VARTM process.

Atmospheric new-particle formation (NPF) is an important source of climatically relevant atmospheric aerosol particles. NPF can be directly observed by monitoring the time-evolution of ambient aerosol particle size distributions. From the measured distribution data, it is relatively straightforward to determine whether NPF took place or not on a given day. Due to the noisiness of the real-world ambient data, currently the most reliable way to classify measurement days into NPF event/non-event days is a manual visualization method. However, manual labor, with long multi-year time series, is extremely time-consuming and human subjectivity poses challenges for comparing the results of different data sets. These complications call for an automated classification process. This article presents a Bayesian neural network (BNN) classifier to classify event/non-event days of NPF using a manually generated database at the SMEAR II station in Hyytiälä, Finland. For the classification, a set of informative features are extracted exploiting the properties of multi-modal log normal distribution fitted to the aerosol particle concentration database and the properties of the time series representation of the data at different scales. The proposed method has a classification accuracy of 84.2 % for determining event/non-event days. In particular, the BNN method successfully predicts all event days when the growth and formation rate can be determined with a good confidence level (often labeled as class Ia days). Most misclassified days (with an accuracy of 75 %) are the event days of class II, where the determination of growth and formation rate are much more uncertain. Nevertheless, the results reported in this article using the new machine learning-based approach points towards the potential of these methods and suggest further exploration in this direction.

Digital images often suffer from contrast variability and non-uniform illumination, which seriously affect the evaluation of biomarkers such as the arteriolar to venular ratio. This biomarker provides valuable information about many pathological conditions such as diabetes, hypertension etc. Hence, in order to efficiently estimate the biomarkers, correct classification of retinal vessels extracted from digital images, into arterioles and venules is an important research problem. This paper presents an unsupervised retinal vessel classification approach which utilises the multiscale self-quotient filtering, to pre-process the input image before extracting the discriminating features. Thereafter the squared-loss mutual information clustering method is used for the unsupervised classification of retinal vessels. The proposed vessel classification method was evaluated on the publicly available DRIVE and INSPIRE-AVR databases. The proposed unclassified framework resulted in 93.2 and 88.9% classification rate in zone B for the DRIVE and the INSPIRE-AVR dataset respectively. The proposed method outperformed other tested methods available in the literature. Retinal vessel classification, in an unsupervised setting is a challenging task. The present framework provided high classification rate and therefore holds a great potential to aid computer aided diagnosis and biomarker research.