| Author: | Chen, Zhijie |
| Title: | Real-time benchmarking and predictive maintenance of air conditioners in public transportation based on physics-guided machine learning |
| Advisors: | Wang, Shengwei (BEEE) Fu, Xiao (BEEE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Buses -- Air conditioning Buses -- Air conditioning -- Maintenance and repair Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Building Environment and Energy Engineering |
| Pages: | xxi, 173 pages : color illustrations |
| Language: | English |
| Abstract: | People spend a lot of time in public transportation for daily commutes in densely populated cities. Air conditioners (ACs) consume large amounts of energy to create a comfort and healthy cabin environment for passengers. Due to the continuous operation of vehicle ACs under highly dynamic and poor environment, they are more prone to performance degradation and even complete failure compared with residential ACs. Real-time benchmarking and predictive maintenance of a large number of ACs in public transportation are critical to the development of smart cities in the future. Today’s Internet of Things (IoT) technology enables collection and transmission of massive measurements from public transportation and the cloud technology provides platforms for deployment of big data analysis and machine learning algorithms. Real-time benchmarking and predictive maintenance of a large number of public transportation ACs in the context of IoT face new challenges in terms of high dynamics, data quality and quantity and cost-effectiveness. Three key issues need to be addressed to tackle the challenges: 1) transient modeling of ACs in highly dynamic operation conditions, 2) analysis of big data from a large number of ACs, and 3) synergy of domain knowledge and data-driven methods to accomplish different tasks. This PhD study aims to develop real-time benchmarking and predictive maintenance methodologies for a large number of electric bus ACs by adopting a novel hybrid approach based on both domain knowledge and data-driven methods. Dynamic modelling of AC is important for both real-time benchmarking and predictive maintenance methodologies. In this study, a novel physics-guided deep learning methodology is proposed for dynamic modeling of electric bus ACs. To maximize the practical values of the model in control and diagnosis of ACs, this research aims at developing an integrated model consisting of individual models of major AC components rather than a black-box model of the entire system. The rich domain knowledge in air conditioning, including heat and mass transfer, thermodynamics, fluid dynamics, etc., guides the determination of model inputs and outputs, design of the model structure, and understanding of temporal relationship in developing dynamic heat exchanger models. A newly developed NARX-LSTM-MLP neural network is proposed for heat exchange process modelling. Component models are integrated by referring to the physical system structure. Field AC operation data from a bus AC collected by IoT sensors is used in this study for validation. Validation results indicate good accordance between measurements and simulation results. Compared with other dynamic modelling methodologies, the model developed in this research is less expensive, more convenient, and more feasible. Apart from dynamic modelling, identifications of bus ACs with abnormal performance from a large number of peers is another important yet challenging task. This study develops a data-driven benchmarking methodology for detecting anomalous operations from a large number of bus ACs by comparing a newly defined key performance indicator (the power consumption ratio, PCR). The indicator is calculated based on comparable peer samples, i.e., similar multivariate time series representing similar dynamic operation conditions of bus ACs. A novel LSTM-autoencoder based similarity measurement method for measuring the distance of multivariate time series operation data is proposed to identify comparable peer samples. Data-driven reference models are developed based on the comparable peer samples to benchmark energy performance of a large number of bus ACs in daily operation. Statistical analysis-based trend and change detection algorithms are designed to identify anomalous operation. Experimental data obtained from field operating bus ACs are used to validate the developed benchmarking methodology. The validation results show encouraging potentials of the proposed benchmarking methodology. The anomalies detected by benchmarking methodology should be further diagnosed. In this study, a novel fault diagnosis methodology combining physical knowledge and machine learning, more specifically time series classification, is proposed for diagnosing bus ACs. The dynamic modelling methodology developed in previous stage is used to build the reference model, predicting AC performance under normal dynamic operations. Rules obtained from physical knowledge are used to label part of the operation data by comparing predicted and measured system variables. The time series analysis-based labeling method is developed to classify and diagnose the remaining unrecognized data. Dynamic AC operation data under both normal and faulty conditions are obtained from experiments conducted on a full-scale bus in a large environmental chamber and used to test the proposed fault diagnosis methodology. The integration of the dynamic modelling, benchmarking and fault diagnosis methodologies enable the predictive maintenance of a large number of bus ACs. Validations are conducted to evaluate the accuracy and feasibility of the proposed methodologies. This study represents an earlier effort in integrating physical domain knowledge with machine learning knowledge in solving real-world engineering problems. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12488