Forensic engineering to avoid future infrastructure failures

Lead Research Organisation: University of Edinburgh
Department Name: Sch of Engineering


The engineering infrastructure sector is seeing unprecedented investment in the transport, energy and water markets globally. A failure in this infrastructure can be catastrophic in terms of economic impact and potential for loss of life.

Forensic engineering is the process of collecting and analysing data related to failure in order to determine causes and learn from failure. Together with project partner Costain, this project seeks to investigate, through this Forensic Engineering methodology, data on real infrastructure failures to understand the reasons for that failure and allow learning to influence future industry decision making. Data will be obtained from publicly available data sources and also through direct investigation of live infrastructure projects.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/P510488/1 01/10/2016 30/09/2021
1812058 Studentship EP/P510488/1 01/10/2016 31/10/2020 Henrietta Rose Baker
Description Findings from initial interview analysis show that that the concept of failure in relation to a construction project signifies different things to different individuals. As well as evocative catastrophic failures, other definitions include: project failure, such as going over budget or over time; engineering failures, like non-compliance or having to undertake rework; and safety failures which encompass incidents, near-misses and injuries. Thematic analysis also revealed that attitudes towards different failure modes vary, especially when considering blame and ownership of the failure itself. However, there is also a general appreciation that exploiting the learning opportunities presented by failures is important to facilitate progress, increase safety and improve efficiencies.
The main body of the research concerns extracting key attributes about failures from descriptions of incidents and quality failures in order to facilitate systematic learning. This also anonymise the data for industry wide sharing. Supervised classification methods using machine learning and Natural Language Processing (NLP) were applied. Proof of concept was proven using abstract data from scientific articles, with key words as attribute classes. For the real failure data, the attribute classes were developed from existing literature and expert labelling of data. Results from a collaboration in the USA show that deep learning methods do not necessarily generate a significant increase in accuracy. Real UK failure data is now being investigated.
Exploitation Route These findings have impacts on how best to progress this research in relation to learning from failure data. They can also have impact on current learning practices in the construction industry.
Sectors Construction

Description John Moyes Lessell's Scholarship (travel sponsorship)
Amount £3,200 (GBP)
Organisation Royal Society of Edinburgh (RSE) 
Sector Charity/Non Profit
Country United Kingdom
Start 11/2018 
End 10/2019