Work Order Prioritization Using Neural Networks to Improve Building Operation

TR Number



Journal Title

Journal ISSN

Volume Title


Virginia Tech


Facility management involves a variety of processes with a large amount of data for managing and maintaining facilities. Processing and prioritizing work orders constitute a big part of facility management, given the large number of work orders submitted daily. Current practices for prioritizing work orders are mainly user-driven and lack consistency in collecting, processing, and managing a large amount of data. Decision-making methods have been used to address challenges such as inconsistency. However, they have challenges, including variations between comparisons during the actual prioritization task as opposed to those outside of the maintenance context. Data-driven methods can help bridge the gap by extracting meaningful and valuable information and patterns to support future decision-makings. Through a review of the literature, interviews, and survey questionnaires, this research explored different industry practices in various facilities and identified challenges and gaps with existing practices. Challenges include inconsistency in data collection and prioritizing work orders, lack of data requirements, and coping strategies and biases. The collected data showed the list of criteria and their rankings for different facilities and demonstrated the possible impact of facility type, size, and years of experience on criteria selection and ranking. Based on the results, this research proposed a methodology to automate the process of prioritizing work orders using Neural Networks. The research analyzed the work order data obtained from an educational facility, explained data cleaning and preprocessing approaches, and provided insights. The data exploration and preprocessing revealed challenges such as submission of multiple work orders as one, missing data for certain criteria, long durations for work orders' execution, and lack of correlation between collected criteria and the schedule. Through hyperparameter tuning, the optimum neural network configuration was identified. The developed neural network predicts the schedule of new work orders based on the existing data. The outcome of this research can be used to develop requirements and guidelines for collecting and processing work order data, improve the accuracy of work order scheduling, and increase the efficiency of existing practices using data-driven approaches.



Work Order, Prioritization, Asset Management, Maintenance, Neural Networks