Performance Comparison of Machine Learning Fault Detection and Diagnosis Algorithms in Organic Rankine Cycle Systems

Aug 4, 2026

Upload by Ophelia

The study investigates fault detection and diagnosis (FDD) methods to improve the reliability and operational efficiency of organic Rankine cycle (ORC) systems for waste-heat recovery. It reduces maintenance costs and unplanned downtime through better fault management by using machine learning models, specifically support vector machines (SVM) and random forest (RF), to identify key faults. Simulation-based experiments address data scarcity and sensor noise common in industry, demonstrating robust detection and diagnostic accuracy and highlighting trade-offs between computational efficiency and diagnostic performance for online monitoring and predictive maintenance adoption.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Fault Detection and Diagnosis (FDD) Background

Machine Learning and Hybrid Approaches

Motivation for ORC-Specific Research

Methodology

Classifiers and Fault Scenarios

Paper ID: 9282, Page 1 PERFORMANCE COMPARISON OF MACHINE LEARNING FAULT DETECTION AND DIAGNOSIS ALGORITHMS IN ORGANIC RANKINE CYCLE SYSTEMS Andres Hernandez 1∗ , Aitor Cendoya 1 , Basile Chaudoir 1 , Vincent Lemort 1 1Thermodynamics Laboratory / Universit de Lige, Lige, Belgium *Corresponding Author: [jahernandez@uliege.be](jahernandez@uliege.be) ABSTRACT This study investigates fault detection and diagnosis (FDD) techniques aimed at enhancing the reliability and operational efficiency of organic Rankine cycle (ORC) systems used for waste heat recovery. By improving fault management, this approach can significantly reduce maintenance costs and unplanned downtime, contributing to the broader adoption of ORC technology in sustainable energy applications. The proposed methodology leverages advanced machine learning algorithms, specifically support vector machines (SVM) and Random Forest (RF), to identify and classify critical system faults. These faults include evaporator fouling, the presence of non-condensable gases, and mechanical issues in the expander and pump components. To evaluate the performance of the FDD framework, simulation-based experiments are conducted to address practical challenges such as data scarcity and noise, which are common in industrial applications. Results demonstrate the robustness of the SVM and RF models in accurately detecting and diagnosing faults, highlighting their potential to maintain high system performance in real-world scenarios. Additionally, the analysis provides insights into selecting appropriate strategies based on the quantity and quality of data available. Furthermore, the study explores the trade-offs between computational efficiency and diagnostic accuracy, offering insights into the applicability of these techniques for online monitoring systems. The findings underscore the critical role of machine learning in predictive maintenance strategies, paving the way for smarter and more resilient energy recovery solutions. 1 INTRODUCTION The increasing global demand for energy efficiency and sustainability has intensified interest in Organic Rankine Cycle (ORC) systems for waste-heat recovery and renewable energy generation (Wieland et al., 2023) . ORC technology offers a viable solution to improve energy utilization in industrial applications by converting low-grade heat into electricity. However, the fluctuating nature of waste heat sources often forces ORC systems to operate under off-design conditions (Dickes, 2019) . This variability poses significant challenges in distinguishing between normal operational deviations and actual system faults based solely on available sensor data. Consequently, the development of effective Fault Detection and Diagnosis (FDD) techniques is crucial for maintaining optimal performance and ensuring the reliability of these complex energy systems (Wang et al., 2021) . FDD methodologies have evolved significantly, encompassing model-based, data-driven, and hybrid approaches. Model-based techniques utilize the physical laws governing system behavior to detect anomalies by comparing measured and predicted operational parameters (Dragan, 2011) . Such methods have been successfully implemented in heat exchangers and other industrial equipment, enabling early fault identification and predictive maintenance (Choi and Krumdieck, 2016) . However, they often require extensive calibration and may struggle with highly nonlinear or uncertain system dynamics. Machine learning-based FDD approaches have gained traction due to their ability to learn complex patterns from historical data, offering adaptability and high detection accuracy. These methods have been extensively applied to Heating, Ventilation, and Air Conditioning (HVAC) and refrigeration systems (Elmouatamid et al., 2023; van de Sand, 2021) . For instance, predictive frameworks leveraging unsupervised learning have been developed to assess system degradation states (van de Sand, 2021) . Ad

FAQs

Which machine learning algorithms are evaluated for ORC fault diagnosis?

The study evaluates support vector machines (SVM) and random forest (RF) classifiers for detecting and diagnosing critical ORC faults.

What types of faults are considered in the organic Rankine cycle system?

The faults include evaporator fouling, the presence of non-condensable gases, and mechanical issues in expander and pump components.

How does the research handle practical challenges like limited data and noise?

Simulation-based experiments are used to model data scarcity and sensor noise, and the study analyzes how these factors affect model performance and feature selection.

Aug 4, 2026

Upload by Ophelia