Machine-Learning Framework for Efficient Multi-Asset Rehabilitation Planning
MetadataShow full item record
While smart cities are viewed as the way of the future, the infrastructure assets expected to support the different smart services are currently managed using frameworks that are outdated, subjective, and inefficient. Such inefficiencies have led to huge maintenance and rehabilitation backlogs that are far beyond the financial capabilities of cities, municipalities and large asset owners like school boards. For example, the cost to bring Ontario schools facilities to an acceptable level of service is estimated to be as high as $16 billion. Currently, most “smart asset initiatives” are geared towards building new assets and using sensors to get periodic info about their condition, with little thought given regarding the condition of existing assets. As such, there is a need to introduce a “smart rehabilitation” framework that answers the question “how to bring the current infrastructure assets up to speed to satisfy the needs of current and future generations?”. To contribute to the overall vision of smart cities (data-driven interconnected services), the introduced framework uses machine learning and smart analytics to tackle three main functions of smart asset rehabilitation frameworks: (1) it automates the inspection and condition assessment processes by using convolutional neural networks (CNNs) to develop a machine learning system where defects can be automatically detected, classified, and quantified from images; (2) it uses data mining and clustering techniques to classify the assets according to their condition and need for repairs, and then uses optimization to select which assets are most worthy of immediate repairs subject to the existing funding constraints, thus enhancing the fund allocation phase by reducing its subjectivity; and (3) it uses novel computations, visualizations, and algorithms to facilitate cost-effective and fast-tracked delivery of the required rehabilitation works by considering them as units of a large repetitive project. To verify the strengths and versatility of the model, the proposed framework is applied to built-up roofs of educational buildings such as schools and university campuses. First, images were collected from the University of Waterloo campus buildings to develop the image-based analysis module; a two-step CNN framework that can detect damages and classify them according to their type. Information from the image-based analysis were then combined with textual information related to building age and description and unsupervised learning was applied to develop the prioritization and fund allocation module. Results from this module are used as the inputs to an optimization procedure where the overall performance of the entire asset portfolio is maximized by selecting which buildings should undergo immediate repairs, given strict budgetary constraints. Finally, the selected rehabilitation works were scheduled as units in a large repetitive project for delivery planning. Accordingly, novel computations and algorithms were developed to create compact schedules with minimal gaps that comply with deadline constraints, and novel visualizations were introduced to showcase the crews movements and the timing of all tasks required in each unit. The proposed framework offers powerful decision support features for a proposed smart rehabilitation layer to be included into the overall smart city vision. This framework deals with existing assets and provides objective assessments, cost-effective prioritization, and time-effective delivery plans. While this study used the case of built-up roofs as an example application, the framework is scalable towards other asset components as well as other assets in general. For example, components such as parking lots and concrete elements would rely heavily on the image-based inspection module, while other components such as HVAC systems would place more emphasis on the data analytics component, including more parameters related to different performance metrics as part of the analysis. Overall, this framework has the potential to revolutionize the multi-billion-dollar business of infrastructure renewal and provide cost effective decisions that save taxpayers’ money on the long run.
Cite this version of the work
Kareem Tarek Mostafa (2021). Machine-Learning Framework for Efficient Multi-Asset Rehabilitation Planning. UWSpace. http://hdl.handle.net/10012/17664