Stochastic analysis and machine learning techniques for maintenance of heritage buildings with multi-source data

The maintenance of buildings, enabled by a fusion of diverse techniques, becomes integral to heritage conservation efforts. According to the principle of minimal intervention, applied methodology employs constrained survey and monitoring, resulting in an information deficit that may challenge comprehensive structural assessments. To address the uncertainty introduced in assessing structural risk assessment and performance prediction, stochastic analysis based on reasonable assumptions is required. Machine learning algorithms offer approaches to obtain valuable insights from limited datasets. This study presents a maintenance case where a monitoring system was deployed at a heritage site, underscoring the adaptability of these methods. The probability density evolution method (PDEM) and a credibility index were employed to quantify the possible failure and evolution of the City Wall stability, providing prior knowledge based on experience. Potential failure modes and unstable trends consistent with the observed results were revealed. The integration of machine learning techniques offers robust predictive models that address data scarcity and identify environmental influences. Vision and sound-based techniques also demonstrated significant promise in risk identification and forming multi-source archive for heritage maintenance.