Health monitoring, damage prognosis and possible futures: applying machine learning to the appraisal and analysis of existing structures

The repair and maintenance of existing structures represents a significant cost, both now and in the future. This has been estimated as being in the region of £15 billion per year in the United Kingdom alone. Moreover, existing structures contain significant quantities of embodied carbon and there is thus an environmental and economic imperative to facilitating their continued effective service when possible. Many structures exist that that have exceeded, or will soon exceed their intended design life. For some years it has been recognised that effective ways of tracking the condition and health of these structures would be valuable.

The future of structural appraisal will rely on effective decision-making using large data sets. The data will be collected from instruments embedded within structures (at the time of construction or retrofitted) and from state of the art surveying techniques. It has been recognised that there is significant potential for applying machine learning techniques to these large data sets derived from structures to assist engineers in pattern-recognition, detection and diagnosis of defects, and in understanding potential future scenarios.

The implementation of structural health monitoring using machine learning to make effective judgements regarding the prognosis of existing structures has been called a "grand challenge problem for engineers in the twenty-first century". It has been suggested that implementation of such an approach would add significantly to the value of built assets. The development of a functional framework for assisting decision-making in real world scenarios is therefore a worthwhile goal.

The project will focus on the development of integrated instrumentation, measurement and data processing methodologies for evaluating health and integrity for service of existing structures, in additon to future prognosis. This will complement traditional engineering approaches and offer assistance to engineers assessing structures based on limited information. Opportunities for applying machine learning techniques to support and enable this process will be investigated. Collaboration with relevant UK and international industrial partners will be pursued. It is hoped that the project will be pivotal to establishing a collaborative international laboratory for invesitgation of such problems, based at the University of Edinburgh.