Machine Learning to Model the Risk of Alteration of Historical Buildings

Detaild, realistic knowledge of the weather effects on building materials is essential to efficiently maintain and manage historical buildings. The main aim of this study is to model the long-term weathering behavior of stone materials under different climate change scenarios. In order to achieve this, data have been collected from an adapted IoT architecture on an emblematic monument in the city of Reims, the “Basilique St-Remi”. An Adhoc wireless sensor network has been deployed during two years. These sensors have been located in various locations of the building walls. A large amount of data has been analyzed together with general weather data of Météo France. Some features about the data variations have been extracted and split into different clusters. The behavior of stones regarding humidity and temperature has been modeled. The last step was the prediction of the behavior of the whole building in the future (in the next 50 or 100 years) according to the weather expectations in various scenarios given by climate changes. This work is a first attempt to assess precise behavior of historical buildings and gives enough information to decision makers to choose relevant measures to consider for building preservation. This study has used around 1400 mega bytes of inputs and has considered three possible scenarios for climate change. The prediction process has been computed on a personal computer during few minutes. The obtained results are precise enough compared to general predictions studied in the past.