Intelligent Feature Extraction and Selection Framework for Fault Diagnosis and Prognosis

The objectives of the project are: Literature review (includes, but not limited to, feature selection techniques, machine learning methods, data mining, decision making strategies) Identify, based on literature and experts experience, industrial use cases worth considering in the project from an operational perspective Develop a novel framework  of feature selection by fusing different feature selection methods (domain knowledge, model-based, and data driven) Validate the proposed framework using selected industrial use cases The student will be based at Cranfield University in the Integrated Vehicle Health Management (IVHM) Centre, which is part of the Manufacturing Department. The student will have the opportunity to work with experts in the diagnostics, prognostics and condition monitoring field, as well as being part of our strong and dynamic research centre at Cranfield. As a part of working in this industrially sponsored project, the student will work in close collaboration with Boeing Company. He/she will need to present their research findings regularly to the project team.



Feature extraction and selection play a key role in fault diagnosis and prognosis research for identifying informative features and reducing feature dimensionality. Feature extraction, or the automated selection of attributes that are of information value in a data set, is a diverse area of data analysis in many disciplines. Numerous model-based and data-driven approaches exist to perform feature extraction. Model based techniques use simplified mathematical representations of more complicated physical process to predict and extract meaningful information from the available data. In contrast, data-driven techniques use statistical or algorithmic approaches, devoid of any physical process representation, for the same purpose. The need for advanced data-driven feature selection techniques has grown immensely due to the popularity of...