Novelty Detection in Rotating Machinery - Assessment of Unsupervised Machine Learning Models for Medium-Sized Industrial Bearings

Aug 4, 2026

Early anomaly detection in industrial rotating equipment is essential for improving predictive maintenance and avoiding unexpected failures. In real industrial environments, structured datasets with fault labels are often unavailable, making supervised classification impractical. This study evaluates unsupervised novelty detection for industrial bearing condition monitoring, comparing machine learning methods with fixed threshold strategies based on standards. Experiments use a dedicated dataset for spherical bearings with localized damage under variable speed and load, showing fixed thresholds produce many false positives. Isolation Forest yields the highest recall, while LOF achieves superior precision, accuracy, F1-score and precision-recall AUC.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Abstract

Research focus

Dataset and experimental setting

Key findings

Introduction

Predictive maintenance benefits

Limits of labeled-data fault diagnosis

Alternatives using anomaly/novelty detection

POLITECNICO DI TORINO Repository ISTITUZIONALE Novelty Detection in Rotating Machinery: Assessment of Unsupervised Machine Learning Models for Medium-Sized Industrial Bearings Original Novelty Detection in Rotating Machinery: Assessment of Unsupervised Machine Learning Models for Medium-Sized Industrial Bearings / Di Maggio, Luigi Gianpio; Brusa, Eugenio; Delprete, Cristiana. - (2025), pp. 1-7. ( 2025 International Conference on Control, Automation and Diagnosis, ICCAD 2025 Barcelona (ESP) 01-03 July 2025)[10 . 1109/iccad64771 .2025. 11099203] . Availability: This version is available at: 11583/3002858 since: 2025-09-08T09:10:12Z Publisher: Institute of Electrical and Electronics Engineers Published DOI:10.1109/iccad64771.2025.11099203 Terms of use: This article is made available under terms and conditions as specified in the corresponding bibliographic description in the repository Publisher copyright IEEE postprint/Author's Accepted Manuscript ©2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collecting works, for resale or lists, or reuse of any copyrighted component of this work in other works. (Article begins on next page) 21 February 2026 Novelty Detection in Rotating Machinery: Assessment of Unsupervised Machine Learning Models for Medium-Sized Industrial Bearings Luigi Gianpio Di Maggio Dept. of Mechanical and Aerospace Engineering Politecnico di Torino Torino, Italy luigi.dimaggio@polito.it Eugenio Brusa Dept. of Mechanical and Aerospace Engineering Politecnico di Torino Torino, Italy eugenio.brusa@polito.it Cristiana Delprete Dept. of Mechanical and Aerospace Engineering Politecnico di Torino Torino, Italy cristiana.delprete@polito.it Abstract—Early anomaly detection in industrial rotating equipment is crucial for enhancing predictive maintenance methodologies and preventing unexpected failures. However, real-world industrial settings often lack structured fault-labeled datasets, making supervised classification approaches impractical. In such contexts, unsupervised novelty detection methodologies emerge as the most viable alternative, as they do not require labeled fault data and operate under the assumption that only normal conditions are available during training. This research investigates the application of novelty detection techniques for industrial bearing condition monitoring, comparing machine learning-based methods with fixed threshold-based strategies derived from standards. The analysis is conducted on a dedicated experimental dataset, representing the first contribution of its kind for spherical bearings with localized damage, tested under variable speed and load conditions. The findings reveal that fixed-threshold strategies result inan excessive number of false positives, limiting their practical applicability in industrial monitoring systems. Among the assessed machine learning algorithms, Isolation Forest achieved the highest recall, detecting the largest number of anomalies, while Local Outlier Factor (LOF) demonstrated superior precision, accuracy, F1-Score and precision-recall AUC. Index Terms—novelty detection, rotating machinery, bearings, condition monitoring, vibration analysis, unsupervised learning I. INTRODUCTION Industrial rotating machinery can greatly benefit from the adoption of predictive maintenance technologies, which optimize resource utilization and generate significant economic returns by reducing unplanned downtime and preventing catastrophic failures. These advantages justify the investment in acquisition, processing, and analysis of vibration signals for condition monitoring and early fault detection [1]–[3] . Over the past decade, there has been an exponential increase in research focusing on the use of machine learning (

FAQs

Why is supervised fault classification often unsuitable for industrial rotating equipment?

Industrial settings frequently lack structured, fault-labeled datasets across damage conditions and operating states, which prevents supervised methods from being applied effectively.

What problem does this research address in bearing monitoring?

It assesses unsupervised novelty detection techniques for detecting early bearing degradation signs when training data includes only normal conditions and labeled faults are unavailable.

How do fixed-threshold strategies compare with machine learning methods?

Fixed-threshold approaches produce excessive false positives, reducing practical usability in industrial monitoring systems.

Aug 4, 2026

Upload by Valentina