Predictive Maintenance by the Detection of Gradual Faults in an IoT-Enabled Public Bus

The widespread incorporation of Internet of Things (IoT) devices in various systems such as mobile phones, vehicles, and security systems is used to collect data. These large volumes of data can be used to train models to make predictions about various things. One such application that uses data from IoT devices is called predictive maintenance. Predictive maintenance involves collecting data from machines and using algorithms to analyze the machine's condition or determine if the machine requires maintenance or repairs. Prior to such a predictive approach, organizations would have to conduct scheduled check-ups for their vehicles to find out if the vehicles needed any repairs or maintenance; however, the scheduled check-ups resulted in downtime and shut-downs and so this proved to be costly and time-inefficient, as many of the vehicles would not require any kind of maintenance or repairs. The shift to predictive maintenance addresses these challenges by leveraging the continuous data stream from IoT devices to predict and prevent failures. This thesis presents a clustering-based algorithm to perform predictive maintenance by detecting potential faults and gradual deterioration for IoT-based buses. The algorithm is tested for the cooling system and the engine torque system of a bus from the Société de Transport de l'Outaouais (STO) transit service. In order to overcome the lack of availability of real-world data, this work also generates two synthetic time series datasets that contain sensor readings belonging to the cooling system and the engine torque system to simulate normal buses, and buses with underlying potential faults and gradual deterioration that the standard maintenance systems would not detect. The synthetic data generation process leverages deep learning frameworks such as Generative Adversarial Networks and Long Short-Term Memory networks along with other statistical methods to make the data appear as realistic as possible. The results from the experiments conducted using the predictive models validate the reliability and accuracy of the proposed algorithm. The results showed that using the predictive maintenance algorithm resulted in all the faults being clustered accurately and appropriately based on the type of fault. This thesis demonstrates that predictive maintenance not only enhances cost and time efficiency but also significantly improves user safety by enabling preemptive maintenance actions. While the predictive models implemented in this thesis focus on the cooling and engine torque systems, the methodology proposed is flexible and can be extended to other subsystems. Overall, this work contributes to the field of predictive maintenance by presenting an efficient and practical solution that ensures the reliability and safety of transportation systems.