Recent Trends in Artificial Intelligence Research for Facility Maintenance and Management
Recent Trends in Artificial Intelligence Research for Facility Maintenance and Management
▲ Senior Researcher Won Ji-sun, Department of Future & Smart Construction Research, KICT
Prologue
The global artificial intelligence (AI) market in the construction sector is predicted to grow at an average annual rate of 35%, reaching KRW 2.33 trillion by 2023 (Market Research Future, 2020). At the same time, the Korean AI market is anticipated to reach KRW 1.9 trillion by 2025, with an average annual growth rate of 15.1% for five years from 2021 (IDC Korea, 2022). In recent times, the construction industry has adopted or has been actively considering adopting various AI technologies across the design, construction, and maintenance phases. The adoption of AI technologies has shown positive effects in the construction industry, including shortened construction durations, cost savings, improved safety, and enhanced quality (Lee, 2020).
The adoption and utilization of AI technology are recognized as essential strategies, not optional choices, for enhancing corporate competitiveness. At the national level, there is a need for research strategies and direction-setting in the field of AI technology that align with the role of the public sector in securing the competitiveness of AI technology in construction. This article aims to introduce a portion of the research conducted (Won et al., 2022) in the field of future AI technology for facility maintenance, focusing on numerical data and case studies, to help establish research directions and preparations.
AI Research Trends Seen Through Numbers
Over the past five years (2016-2021), I've analyzed a total of 33 documents, including research papers and reports, related to AI technology development in the field of facilities maintenance and management. During the document collection, basic search keywords, such as "artificial intelligence," "maintenance," "machine learning," "deep learning," and "convolutional neural network," as well as various model names commonly employed in research, were used. I've analyzed research trends based on the 41 collected AI application cases, considering four perspectives: 1. Purpose of utilizing AI, 2. Targeted facilities, 3. Collected raw data, and 4. Types of learning data.
Through my analysis of documents from the perspective of utilizing AI, two main types of AI technology usage were identified: direct utilization of AI technology for maintenance works, and utilization of AI in the intermediate stage for data collection and processing for learning purposes. The research areas that directly employ AI in maintenance work were further categorized into Inspection and assessment, Continuous measurement, Repair and reinforcement, and Aging prediction. The research areas in which research is active are as follows, and are listed in order of prominence: Inspection and assessment (62%), Building learning data for AI (17%), Continuous measurement (7%), Repair and reinforcement (7%), and Aging prediction (7%). To summarize, the current status of research regarding the five purposes of utilizing AI is as follows:
1. [Inspection and assessment] AI applications for inspection and assessment primarily focus on damage detection, such as crack detection using facility inspection photos. Local governments and construction corporations are increasingly adopting automated inspection technologies utilizing unmanned vehicles, such as drones and robots, for inspections in hard-to-reach areas. The development of classification models, primarily around concrete crack detection, is the most prevalent trend. Additionally, the technology is being expanded to include other types of damage beyond cracks, as well as techniques for quantifying damage location, size, and area.
2. [Repair and reinforcement] Regarding the application of AI for repair and reinforcement, research is being conducted with the purpose of training AI with repair and reinforcement data to predict maintenance methods and costs, as well as to predict repair timings exceeding the criteria by using time-series accumulated images of visual inspection grids.
3. [Continuous measurement] Regarding the application of AI for continuous measurement, research has primarily been focused on predicting changes in the condition of facilities and detecting real-time defects for immediate response. This research utilizes accelerometer sensor data and IoT sensor data to detect damaged locations or measurement anomalies, with the purpose of managing performance changes and risks.
4. [Aging prediction] In the application of AI for aging prediction, research is being conducted primarily on creating concrete degradation models or estimating remaining service life based on accelerometer data and structural health data, and utilizing this information for preventive maintenance.
5. [Building learning data for AI] In building data for AI learning, research is being conducted during the data collection and preprocessing stages with the purpose of creating learning data that is currently lacking, such as accelerometer data and crack images, or enhancing low-resolution images to high-resolution ones.
The target facilities in AI application research, ranked in descending order, have been bridges (58%), concrete structures (22%), road facilities with a focus on road pavement/surfaces (15%), and buildings (5%). In terms of the types of learning data for AI, images (56%) outnumbered text (44%). Among the detailed types of text data derived from collected raw data, it was observed that measurement data obtained from equipment, database data acquired from system databases, and documents such as inspection reports were utilized, in that order. Of the 33 collected documents that specified data collection methods, it was found that 5 (15%) used retained data, 9 (27%) used publicly available data, and 19 (58%) collected data directly through measurements and crawling, among other methods.
AI Research Trends Seen Through Cases
Through the analysis of 34 previous research cases related to inspection and assessment, repair and reinforcement, continuous measurement, and aging prediction, this study presents the main research status for each specific maintenance and management work type, utilization purposes, data utilized, and representative research cases.
Epilogue
We have examined trends in AI research in the field of facility maintenance through numbers and cases. In terms of works where AI is applied, the area of inspection and assessment stands out as the most active and technologically mature field within the maintenance domain. It is expected that the adoption of AI will accelerate, particularly for facilities that are difficult to inspect visually and are dangerous to access. Furthermore, as maintenance technology in Korea transitions towards proactive and preventive maintenance systems in response to aging infrastructure and facilities, there is a growing demand for AI research in aging prediction. In terms of data, image-based research is currently the most active, with text-based research acquired through measuring equipment also being quite prevalent. With recent advancements in natural language processing technology, the expansion of text-based research utilizing construction documents such as inspection reports in the future is anticipated. Many documents highlight the limitations of insufficient AI learning data in their research. Given that this significantly impacts the efforts to secure AI performance, it is expected that the establishment of specialized AI learning datasets for the field of maintenance and research on data quality will become increasingly important.
