|
" Quantitative risk analysis using real-time data and change-point analysis for data-informed risk prediction "
Jordan, G.
| Document Type
|
:
|
Latin Dissertation
|
| Record Number
|
:
|
833077
|
| Doc. No
|
:
|
TLets795083
|
| Call number
|
:
|
http://doi.org/10.15123/uel.87789
|
| Main Entry
|
:
|
Jordan, G.
|
| Title & Author
|
:
|
Quantitative risk analysis using real-time data and change-point analysis for data-informed risk prediction\ Jordan, G.
|
| College
|
:
|
University of East London
|
| Date
|
:
|
2019
|
| student score
|
:
|
2019
|
| Degree
|
:
|
Thesis (D.Prof.)
|
| Abstract
|
:
|
Incidents in highly hazardous process industries (HHPI) are a major concern for various stakeholders due to the impact on human lives, environment, and potentially huge financial losses. Because process activities, location and products are unique, risk analysis techniques applied in the HHPI has evolved over the years. Unfortunately, some limitations of the various quantitative risk analysis (QRA) method currently employed means alternative or more improved methods are required. This research has obtained one such method called Big Data QRA Method. This method relies entirely on big data techniques and real-time process data to identify the point at which process risk is imminent and provide the extent of contribution of other components interacting up to the time index of the risk. Unlike the existing QRA methods which are static and based on unvalidated assumptions and data from single case studies, the big data method is dynamic and can be applied to most process systems. This alternative method is my original contribution to science and the practice of risk analysis The detailed procedure which has been provided in Chapter 9 of this thesis applies multiple change-point analysis and other big data techniques like, (a) time series analysis, (b) data exploration and compression techniques, (c) decision tree modelling, (d) linear regression modelling. Since the distributional properties of process data can change over time, the big data approach was found to be more appropriate. Considering the unique conditions, activities and the process systems use within the HHPI, the dust fire and explosion incidents at the Imperial Sugar Factory and the New England Wood Pellet LLC both of which occurred in the USA were found to be suitable case histories to use as a guide for evaluation of data in this research. Data analysis was performed using open source software packages in R Studio. Based on the investigation, multiple-change-point analysis packages strucchange and changepoint were found to be successful at detecting early signs of deteriorating conditions of component in process equipment and the main process risk. One such process component is a bearing which was suspected as the source of ignition which led to the dust fire and explosion at the Imperial Sugar Factory. As a result, this this research applies the big data QRA method procedure to bearing vibration data to predict early deterioration of bearings and final period when the bearing's performance begins the final phase of deterioration to failure. Model-based identification of these periods provides an indication of whether the conditions of a mechanical part in process equipment at a particular moment represent an unacceptable risk. The procedure starts with selection of process operation data based on the findings of an incident investigation report on the case history of a known process incident. As the defining components of risk, both the frequency and consequences associated with the risk were obtained from the incident investigation reports. Acceptance criteria for the risk can be applied to the periods between the risks detected by the two change-point packages. The method was validated with two case study datasets to demonstrate its applicability as procedure for QRA. The procedure was then tested with two other case study datasets as examples of its application as a QRA method. The insight obtained from the validation and the applied examples led to the conclusion that big data techniques can be applied to real-time process data for risk assessment in the HHPI.
|
| Added Entry
|
:
|
University of East London
|
| |