In recent years, there has been a growing focus on implementing strategies for the maintenance and control of commercial buildings. One such strategy is predictive maintenance (PdM) 4.0, which utilizes machine learning (ML) models and artificial intelligence (AI) to detect and predict equipment/system faults. This approach has shown promising results in improving various aspects of building management, including maintenance costs, equipment downtime, and overall profitability. However, there are still challenges to overcome in implementing PdM 4.0 effectively. This blog post presents a comprehensive framework for implementing PdM 4.0 in commercial buildings, addressing these challenges and providing practical solutions.
The Setup Part
The first stage of the framework involves understanding the building's components and layout. By analyzing the as-built drawing of the building, the number and location of each component, such as chillers, cooling towers, pumps, and terminal units, can be determined. This information provides a clear picture of the building's infrastructure.
The second stage focuses on the placement of reading tools for monitoring operational parameters of the building systems. These tools, such as sensors and meters, should be strategically located to ensure accurate readings.
The third stage involves data collection. The recommended operational parameters for monitoring include water temperatures, pressures, and space temperatures. By following the proposed data collection plan, datasets can be created for each component, which are crucial for building the prediction model in the next part of the framework.
The ML Part
In this stage, ML algorithms, specifically the decision tree (DT) algorithm, are used to build a prediction model for detecting and predicting faults in the building systems. The C4.5 and CART algorithms are recommended for this purpose. The DT algorithm is chosen due to its high accuracy and ability to capture nonlinear relationships.
The prediction model is trained using the collected datasets, and different training parameters are optimized to improve the model's accuracy. The model is implemented in Python, and the prediction accuracy for each component is evaluated. The results show a high accuracy rate for all components, indicating the effectiveness of the DT algorithm in predicting faults.
The Quality Control Part
The final stage of the framework focuses on monitoring and responding to the predicted faults. A control plan is proposed, which includes monitoring the system using the prediction model and responding immediately to any detected faults. The response actions involve inspecting and rectifying the faults based on the solutions provided in the industry survey study.
The implementation of the framework at a university in Riyadh, Saudi Arabia, demonstrates its practicality and effectiveness. The DT model shows a significant improvement in predicting faults compared to the existing building management system. The most common faults detected during the empirical period align with the findings of the industry survey study.
Conclusion
The proposed framework provides a comprehensive approach to implementing PdM 4.0 in commercial buildings. By addressing the challenges related to fault detection and prediction, the framework offers practical solutions for improving building maintenance and operation. The successful implementation of the framework at a university validates its effectiveness. Future research can explore further insights into integrating ML models with building automation and management systems and expanding the framework to other utility systems in commercial buildings.
Reference
Almobarek M.; Mendibil K.; Alrashdan A. (2023). Predictive Maintenance 4.0 for Chilled Water System at Commercial Buildings: A Methodological Framework. Buildings, 13(2), -, DOI: 10.3390/buildings13020497.