Improving Cultural Heritage conservation: LSTM neural networks to effectively processing end-user’s maintenance requests

Abstract

Preventive conservation of cultural heritage can avoid or minimize future damage, deterioration, loss and consequently, any invasive intervention. Recently, Machine Learning methods were proposed to support preventive conservation and maintenance plans, based on their ability to predict the future state of the built heritage by collected data. Several data sources were used, such as structural data and images depicting the evolution of the deterioration state, but till now textual information, exchanged by people living or working in historical buildings to require maintenance interventions, was not used to support conservation programmes. This work proposes a method to support preventive conservation programs based on the analysis of data collected into CMMS (computer maintenance management software). In a Cultural Heritage building in Italy, hosting a University Campus, data about end-user’s maintenance requests collected for 34 months were analysed, and LSTM neural networks were trained to predict the category of each request. Results show a prediction accuracy of 96.6%, thus demonstrating the potentialities of this approach in dynamically adapting the maintenance program to emerging issues.