Big data toward vehicle health monitoring system: an engine health perspective

The rapid growth of connected and sensor rich vehicles has produced unprecedented volumes of engine telemetry, posing critical challenges for traditional vehicle health monitoring system (VHMS). Existing systems struggle with high volume, high velocity, and heterogeneous data streams, limiting real-time fault detection, predictive maintenance, and operational reliability. Addressing these limitations requires a cohesive, multi-layered framework that unifies scalable data management, real-time analytics, and intelligent predictive modeling. This thesis leverages a VHMS dataset initially collected by (Rahman et al., 2022) and augmented from 3,003 to 200,000 records with 18 critical features, including vibration, Controller Area Network bus, combustion, lubrication, and thermal signals. Using this high dimensional, sequential, and heterogeneous data, this research developed an integrated end-to-end VHMS paradigm comprising four key contributions: (i) A conceptual big data framework that defined a taxonomy for data collection, preprocessing, storage, and analysis, ensuring modularity, traceability, and scalability across heterogeneous vehicle data sources. (ii) An optimized hybrid big data processing architecture that integrate hadoop for fault tolerant historical storage, Spark for in-memory analytics, and Kafka for real-time streaming. This framework dynamically switches between batch and streaming workloads, achieving 0.030s execution time, 1.67 × 10−7s per-row latency, and a 55.01% reduction in network load compared to standalone approaches (Chukwudi et al., 2025). (iii) A hybrid deep learning diagnostic Model, capable of modeling temporal dependencies and adapting across diverse vehicle types. This model attains 92.18% testing accuracy with a Receiver Operating Characteristic and Area Under the Curve (ROC AUC) of 0.965, outperforming conventional multilayer perception (MLP), long short-term memory (LSTM), and bidirectional gated recurrent units (BiGRU) architectures by 15–20% in both accuracy and generalization (Md Abdur Rahim et al., 2025). (iv) A real-time diagnostic model, combining Support Vector Machines, K-Neighbors, Gradient Boosting, and Decision Trees through a meta-learning layer. The ensemble achieved 94.7% accuracy and Area Under the Curve (AUC) of 0.9702, demonstrating superior robustness to sensor noise and overlapping engine states compared to single model approaches (Chukwudi et al., 2024; Rahman et al., 2022). Collectively, these contributions established a robust, interpretable, and adaptive VHMS paradigm capable of real-time diagnostics, predictive maintenance, and operational decision support. Comparative evaluation confirmed the improvements in accuracy, latency, scalability, and reliability over conventional VHMS. Furthermore, the framework provides a foundation for future studies in federated learning, edge-cloud deployment, and multimodal sensor fusion, offering scalable, low-cost, and inclusive solutions for modern transportation systems. This research advances both practical and scholarly understanding of intelligent VHMS, demonstrating that integrating data engineering, machine learning, and decision-oriented system design is critical for next-generation vehicle diagnostics.

Citation

Isinka, C.J. (2026) Big data toward vehicle health monitoring system: an engine health perspective. University of Wolverhampton. https://wlv.openrepository.com/handle/2436/626399

Publisher

University of Wolverhampton

URI

https://wlv.openrepository.com/handle/2436/626399

Type

Thesis or dissertation

Language

en

Description

A thesis submitted in partial fulfilment of the requirements of the University of Wolverhampton for the degree of Doctor of Philosophy.

Collections

Theses and Dissertations