CME8124 : Multivariate Methods
- Offered for Year: 2017/18
- Module Leader(s): Dr Chris O'Malley
- Owning School: Engineering
- Teaching Location: Newcastle City Campus
|Semester 1 Credit Value:||10|
To introduce students to multivariate statistical methodologies that allow them to analyse industrial data thereby realising an enhanced understanding of the process. The course covers three aspects in terms of focus – data pre-screening, feature extraction and process modelling.
This module addresses a real challenge that is of extreme relevance to the process industries. There are many cases where it is not possible to undertake experimental design and utilise the resulting data to attain enhanced process understanding, or the development of models for process monitoring or optimisation. The only data available may be that collected directly on the plant. Consequently it is not necessarily of an appropriate form, in the first instance, to execute the required task, be it finger printing or modelling. Therefore to extract the maximum amount of information, necessitates in the first instance the cleaning of the data (removing outliers, filtering the data, in-filling for missing data, time shifting variable, for example) prior to the application of appropriate statistical methodologies, such as principal component analysis if the objective is feature extraction. Multivariate statistical control of both continuous and batch industrial processes will be explored through the extension to principal component analyses. Additionally for process control and optimisation, it is essential to develop a robust model of the process of interest. The course introduces a number of linear modelling techniques to help realise the aforementioned objectives.
Outline Of Syllabus
Multivariate Data Analysis: Introduction: What problems can be addressed using these techniques; Preliminary Data Analysis – Handling of Inhomogeneous Data (Missing Data; Outliers; Noisy Data; Time Alignment); Graphical Procedures. Dimensionality Reduction (Principal Component Analysis); Modelling techniques: Multiple linear regression, Principal component regression; Projection to Latent Structures. Multivariate Statistical Performance Monitoring – Continuous and Batch Processes.
|Guided Independent Study||Assessment preparation and completion||1||4:30||4:30||Problem Solving Exercise 2 and subsequent writing up in report format|
|Guided Independent Study||Assessment preparation and completion||1||4:30||4:30||Problem Solving Exercise 1 and subsequent writing up in report format|
|Scheduled Learning And Teaching Activities||Lecture||16||1:00||16:00||N/A|
|Scheduled Learning And Teaching Activities||Small group teaching||16||2:00||32:00||Numerical practice sessions|
|Guided Independent Study||Independent study||1||43:00||43:00||Review lecture material and prepare for small group teaching|
Teaching Rationale And Relationship
Lectures convey the statistical concepts and theory and their application in process engineering. Numerical practice sessions support the learning introduced in lectures through the students having the opportunity to apply the concepts to a number of problems varying in terms of complexity. The numerical practice sessions allow the completion of assignment work.
The format of resits will be determined by the Board of Examiners
|Prob solv exercises||1||M||50||Assessed report - Data Prescreening, Principal Component Analysis & process monitoring (maximum 25 pages)|
|Prob solv exercises||1||M||50||Assessed report - Process Data Modelling (set week 11), maximum 20 pages|
Assessment Rationale And Relationship
Assignments allow engineering problems to be set and solved using computer software. They also provide the opportunity for the key skills listed above to be assessed and implemented.