[Paper Review] Predictive Maintenance for Industrial IoT of Vehicle Fleets using Hierarchical Modified Fuzzy Support Vector Machine

Arindam Chaudhuri|arXiv (Cornell University)|Jun 24, 2018

Machine Fault Diagnosis Techniques19 references16 citations

TL;DR

This paper proposes a hierarchical modified fuzzy support vector machine (HMFSVM) for predictive maintenance in industrial IoT-enabled vehicle fleets, integrating vehicle attributes like mileage, age, and type to improve fault prediction. The method outperforms logistic regression, random forests, and standard SVMs in experimental evaluations, demonstrating superior accuracy and robustness in fault detection and maintenance scheduling.

ABSTRACT

Connected vehicle fleets are deployed worldwide in several industrial IoT scenarios. With the gradual increase of machines being controlled and managed through networked smart devices, the predictive maintenance potential grows rapidly. Predictive maintenance has the potential of optimizing uptime as well as performance such that time and labor associated with inspections and preventive maintenance are reduced. In order to understand the trends of vehicle faults with respect to important vehicle attributes viz mileage, age, vehicle type etc this problem is addressed through hierarchical modified fuzzy support vector machine (HMFSVM). The proposed method is compared with other commonly used approaches like logistic regression, random forests and support vector machines. This helps better implementation of telematics data to ensure preventative management as part of the desired solution. The superiority of the proposed method is highlighted through several experimental results.

Motivation & Objective

• To address the growing challenge of maintaining large-scale industrial vehicle fleets in IoT environments.
• To improve fault prediction accuracy by incorporating key vehicle attributes such as mileage, age, and type.
• To reduce unplanned downtime and maintenance costs through data-driven, proactive maintenance strategies.
• To develop a robust machine learning model that outperforms traditional methods like logistic regression and random forests.
• To enable effective telematics data utilization for preventative maintenance in real-world industrial IoT deployments.

Proposed method

• The proposed HMFSVM uses a hierarchical structure to classify faults based on vehicle attributes such as mileage, age, and type.
• Fuzzy logic is integrated into the support vector machine framework to handle uncertainty and imprecision in sensor and telematics data.
• The method applies modified fuzzy membership functions to weight training samples, reducing the impact of noisy or outlier data.
• A multi-level classification strategy is employed, where faults are first grouped into high-level categories before fine-grained classification.
• The model is trained on real-world telematics data from vehicle fleets, with performance evaluated using standard classification metrics.
• Cross-validation and comparative analysis are used to benchmark HMFSVM against logistic regression, random forests, and standard SVMs.

Experimental results

Research questions

• RQ1 How can vehicle fleet maintenance be optimized using predictive analytics in industrial IoT environments?
• RQ2 To what extent does integrating fuzzy logic improve fault prediction accuracy in heterogeneous vehicle data?
• RQ3 Can a hierarchical classification approach better handle complex, multi-attribute vehicle fault patterns than flat models?
• RQ4 How does HMFSVM compare in performance to conventional machine learning models like logistic regression and random forests?
• RQ5 What is the impact of incorporating vehicle-specific attributes (mileage, age, type) on predictive maintenance outcomes?

Key findings

• The HMFSVM model achieved higher prediction accuracy than logistic regression, random forests, and standard SVMs in classifying vehicle faults.
• The integration of fuzzy logic significantly reduced the influence of noisy or ambiguous data points on model performance.
• The hierarchical structure enabled better generalization across diverse vehicle types and fault patterns.
• The method demonstrated robustness in handling imbalanced datasets common in real-world telematics data.
• Experimental results confirmed that the proposed model reduces false positives and improves early fault detection.
• The model's performance was consistently superior across multiple evaluation metrics, including F1-score and AUC-ROC.