Clients with multiple assets require intimate knowledge of the condition of each of their operational assets to enable them to effectively manage their portfolio and improve business performance. This is being driven by the increasing adverse effects of climate change, demanding legal and regulatory compliance requirements for sustainability, safety and wellbeing, and increasing competitiveness. As such, clients are looking for fast and effective means to quickly and frequently survey and communicate the condition of their buildings so that essential maintenance and repairs can be done in a proactive and timely manner before it becomes too dangerous and expensive.
Traditional methods for this type of work commonly necessitate engaging building surveyors to undertake a condition assessment which involves a lengthy site inspection resulting in a systematic recording of the physical conditions of the building elements with the use of photographs, note taking, drawings and information provided by the client. The data collected is then analysed to produce a report that includes a summary of the condition of the building and its elements. This is also used to produce estimates of immediate and projected long-term costs of renewal, repair and maintenance of the building. This enables facility managers to address current operational requirements, while also improving their real estate portfolio renewal forecasting and addressing funding for capital projects. Current asset condition assessment procedures are extensively time consuming, laborious, and expensive and pose health and safety threats to surveyors, particularly when inspections are carried out at height and roof levels which are difficult to access.
Image analysis techniques for detecting defects have been proposed as an alternative to the manual on-site inspection methods. Whilst the latter is time-consuming and not suitable for quantitative analysis, image analysis-based detection techniques, on the other hand, can be quite challenging and fully dependent on the quality of images taken under different real-world situations (e.g. light, shadow, noise, etc.). In recent years, researchers have developed a number of computer vision and machine learning-based detection techniques as an attempt to increase the level of automation of asset condition inspection. Current research efforts have been mainly geared towards the automated detection of defects in infrastructural assets such as cracks in road surfaces, bridges, water and sewerage pipelines, earthquake damaged structures.
However, there has not been a similar level of research effort directed at automated condition assessment of buildings despite buildings representing a significant financial asset class. Buildings are generally considered durable because of their ability to last hundreds of years. However, those that survive long have been subject to care through continuous repair and maintenance throughout their lives. All building components deteriorate at varying rates and degrees depending on the design, materials and methods of construction, quality of workmanship, environmental conditions and the uses of the building. Our intention is to develop advanced techniques for automated condition assessment of buildings with a focus on the automated detection and localisation of a range of defects in real assets from images, preferably in a single application.
There has been significant research geared towards the automated detection of defects in infrastructural assets such as cracks in road surfaces, bridges, water and sewerage pipelines. However, there has not been a similar level of research effort directed at automated condition assessment of buildings despite buildings representing a significant financial asset class.
Most of the research work reported in the literature to date on automated detection of defects focus on one type of defect. The challenge we set ourselves was to explore the use of emerging deep machine learning techniques to investigate the potential of developing a single application that can detect multiple types of non-structural and structural defects on real assets. This has the advantage that in practice a surveyor will only have to acquire and run a single application which detects multiple defects on an asset as it would be impractical for surveyors to acquire multiple applications; each one running individually to detect each type of defect, not unless it cannot be helped. We are therefore developing a generalised surveying technique for detecting different types of structural and non-structural defects instead of focusing on one type only which would be both an expensive and impractical proposition for surveyors.
To the best of our knowledge, there is no available deep learning-based application that is currently used for real-time detection of multiple types of building defects. The lack of such application is, in fact, the motivation behind conducting this research. We believe that our practical and affordable application will contribute towards the efforts for the automation of condition assessment and health monitoring of built assets. Although the immediate focus is on building defects, we intend to extend this to other real assets as the work progresses.
Computer vision and deep machine learning techniques require the acquisition and use of large image datasets to train an algorithm to be able to recognise the aspect of interest in an image. In our case will need to assemble large datasets for each type of defect on interest. A database is being established of images solicited from many sources, including photos taken by mobile phone, a hand-held camera, drones/UAVs, and copyright-free images obtained from the internet. The data currently contains four types of defects which include mould, stain, paint deterioration (which includes peeling, blistering, flacking, and crazing), cracks, and rust on metal roofs. We are therefore seeking to establish strategic partnerships with organisations that can provide us with large quantities of images of different types of defects or assist us is collecting such information from their assets or projects.
Perez, H, Tah, J.H.M, Mosavi, A. (2019). Deep Learning for Detecting Building Defects Using Convolutional Neural Networks, Sensors 2019, 19(16), 3556;. https://doi.org/10.3390/s19163556.
Perez, H, Tah, J.H.M. Improving the Accuracy of Convolutional Neural Networks by Identifying and Removing Outlier Images in Datasets Using t-SNE. Mathematics 2020, 8, 662. https://doi.org/10.3390/math8050662.