PRIN 2022 MUR - UTMOST FDD: an aUToMated, Open, Scalable and Transparent Fault Detection and Diagnosis process for air-handling units based on a hybrid expert and artificial intelligence approach

Involved academic partners

DESCRIPTION OF THE PROJECT

Heating Ventilation and Air-Conditioning (HVAC) systems equipped with air-handling units (AHUs) are frequently operated in faulty conditions due to lack of proper maintenance, failure of components or incorrect installation. Faulty operation in AHUs leads to uncomfortable indoor environment, poor indoor air quality and significant wastes of energy and money. To this purpose, Fault Detection and Diagnosis (FDD) processes make it possible to automatically recognize fault occurrence and identify the causes and the location of fault, contributing to enhance both energy efficiency and indoor environmental quality. This project mainly aims to develop an automated, open, scalable and transparent FDD process for AHUs based on a hybrid expert and artificial intelligence-based approach. The project will start by creating a reference dataset based on experimental campaigns characterized by high resolution measurements of both normal and faulty operation of a typical existing monitored AHU (serving a 4x4x3.6 m³ test room) under different modes and weather/load conditions. The experimental dataset will represent a fundamental source of knowledge for assessing the real impact of several typical faults in terms of operating cost, energy consumption, GHG emissions and indoor comfort/air quality. Moreover, the dataset will be exploited to validate a digital twin capable to mimic the operation of the AHU in both faulty and normal conditions; the simulation model will make it possible to conduct robust fault impact scenarios and to extend the operating ranges of training measured data. Both the experimental and simulation datasets will be made publicly available on a data repository well-recognised by researchers, thus opening the opportunity for the scientific community to perform replicability and benchmark studies on FDD processes for AHUs. Novel hybrid FDD strategies including both data-driven and knowledge based models will be then developed based on the obtained datasets. The hybrid FDD framework will make it possible to exploit the potentialities of physics-based models for the description and interpretation of faults occurrence and artificial intelligence techniques to extract non-trivial knowledge from experimental and simulated data. Finally, the transferability and scalability of the conceived FDD strategy, exploiting ontology schema and applying a transfer learning framework with reference to a target AHU different from the one used for the development of the FDD strategy itself will be tested. The project will represent a cutting-edge experience thanks to the proposed holistic approach aiming to the resolution of the main challenging issues in the field of FDD for AHUs. The flow of activities can be replicated also for other AHUs with the aim of supporting an easier penetration of advanced automatic FDD tools in the automation industry as a key and low-cost solution to enhance energy management in buildings.

The UTMOST FDD Project is organised into the following Work Packages:

WP1 - EXPERIMENTAL ANALYSIS FOR THE CHARACTERIZATION OF FAULTY AND NORMAL OPERATION OF A TYPICAL AHU

The WP1 aims at achieving the following main objectives: 1) categorize scientific studies on FDD methods for AHUs based on experimental datasets, 2) create a reliable experimental dataset of a typical AHU covering a wide range of fault free and faulty
scenarios. This WP1 will help to a) better understand the current research gaps, b) address the lack of datasets for FDD with verified ground truth information, c) support the development of, and benchmark the performance accuracy of FDD methods, d) recognize fault patterns of AHUs and gain insights on AHUs performance.

Figure: Graphical abstract WP1

WP2 - KPIs IDENTIFICATION FOR ASSESSING FAULT IMPACT IN AHU OPERATION

The aim of the WP2 is the definition of KPIs to evaluate fault impacts in terms of energy consumption, energy cost, thermal comfort and indoor air quality. The assessment of fault impact based on these KPIs will be performed starting from the experimental/simulation data collected in the WP1 and WP3.

Figure: Graphical abstract WP2

WP3 - AHU DIGITAL TWIN AND SIMULATION-BASED FAULT IMPACT SCENARIO ANALYSIS

The WP3 aims at achieving the following main objectives: 1) analyse and categorize the scientific studies on digital twins of AHUs for FDD purposes, 2) publicly provide a detailed digital twin of a typical AHU validated against measured data, 3) publicly provide labelled simulation data accurately representing the performance of a typical AHU under a wide range of fault free and faulty conditions, 4) perform an impact scenario analysis associated to faults’ occurrence based on simulation results. This WP3 will help in a) better understanding the current research gaps, b) address the lack of simulation tools of AHUs for impact scenario analyses, c) support the development of, and benchmark the performance accuracy of FDD methods, d) recognize fault patterns of AHUs and rank the faults with most adverse impacts, e) assess the gap between AHU faulty operation and design expectations.

Figure: Graphical abstract WP3

WP4 - DEFINITION OF HYBRID FDD STRATEGIES FOR AHUs

The aim of the WP4 is the definition of FDD strategies that can be exploited for detecting and diagnosing anomalies in AHUs during their operation. The definition of FDD logics will be based on a hybrid approach, considering both data-driven and knowledge-based strategies.

Figure: Graphical abstract WP4

WP5 - TRANSFERABILITY ANALYSIS OF THE FDD STRATEGIES

The WP5 aims at assessing the transferability of the hybrid FDD strategies conceived in the WP4 for enhancing its penetration and adoption in buildings. A variety of transfer learning methods, which differ in their transfer methods (i.e., instance-based, feature-based, parameter-based, and relational knowledge-based) and application natures (i.e., transductive, inductive and unsupervised), have been reviewed in the literature. The main aim of this WP is to transfer the FDD strategies, pre-trained on the existing labeled datasets derived from the WP1 and WP3, exploiting them with reference to an external target AHU for accomplishing the same task.

Figure: Graphical abstract WP5

POSSIBLE APPLICATION POTENTIALITIES AND SCIENTIFIC, TECHNOLOGICAL, SOCIAL AND ECONOMIC IMPACT

FDD is one of the most active fields of research and commercial product development in the building and automation sector and this project will provide a general procedure for a wider development, application and transferability of FDD tools. The major project output is a novel hybrid approach for the definition of FDD strategies capable to couple both data-driven and knowledge-based methods aiming at more efficiently operate AHUs in building HVAC systems. The developed hybrid FDD strategies could be applied to typical AHUs (representing a significant portion of the overall energy demand and GHG emissions), with a high degree of transferability among similar configured systems.
The project will publicly deliver an experimental dataset including high resolution measurements of a reference AHU monitored during both normal and faulty conditions. The public dataset, with ground truth data on the presence of faults, will span a wide range of boundary conditions and contain information on fault type and severity. Similarly, the digital twin of the AHU, the extensive simulated dataset and the ontology-based schema of the metering infrastructure of the experimental facility will be made publicly available for enabling the highest usability of the project outcomes, thus representing a valuable knowledge source for the scientific and professional community.
A novel methodology will be also introduced to demonstrate the transferability of the developed hybrid FDD strategies and define a reference approach to be pursued while designing generalizable FDD tools for building energy systems. In this framework, the main application potentialities of the project will be focused on the maximization of the interoperability and transferability of FDD strategies during their deployment in building energy systems, while maintaining high performances in terms of fault detection and diagnosis rates. This is a key aspect considering that the deployment of advanced FDD tools in existing AHUs can be a cost-prohibitive task requiring significant manual efforts and specific domain expertise to analyze, organize and exploit collected data. Detailed guidelines will be delivered to support technicians and practitioners involved in continuous commissioning activities of buildings.
The implementation of the developed FDD strategies will prove the applicability of novel approaches to the energy management systems and the benefits related to the hybrid integration of AI-based and expert-based solutions in building automation systems. In this project, a challenge-based approach will bring together resources and expertise across different disciplines (mechanical, energy, environmental engineering, economics, and data science) and technologies.
The results of the project will be also used for:

• dissemination to industry, designers, energy managers, local authorities, and publicutilities;
• supporting pilot and demonstration projects;
• supporting procurement of novel AI-based energy management systems in the building sector;
• establishing links with the activities of EU and Italian institutions;
• promoting a local green economy and a conscious use of energy in buildings.

The project has a cross-cutting impact between the fields of AI, smart monitoring, energy management and building physics, consistent with the research trajectories defined in the Strategia Nazionale di Specializzazione Intelligente - SNSI (Agenda Digitale, Smart Communities, Sistemi di mobilità intelligente) and in the Programma Nazionale Ricerca - PNR to enable the penetration of an open approach to innovation (Articolazione 5. Better data and models for optimizing the building performance - GRANDE AMBITO DI RICERCA E INNOVAZIONE: CLIMA, ENERGIA, MOBILITÀ SOSTENIBILE). 

Figure: Impact of the UTMOST-FDD project

Acknowledgements

The project is funded by the European Union in the framework of the initiatives "Next Generation EU"