Multi-tiered architecture for automatic sensor failure detection in traffic systems: edge-fog-cloud collaboration for enhanced accuracy
| Author(s) | Chang, Ethan | |
| Date Accessioned | 2025-04-14T16:53:49Z | |
| Date Available | 2025-04-14T16:53:49Z | |
| Publication Date | 2025 | |
| SWORD Update | 2025-04-14T04:01:35Z | |
| Abstract | The increasing reliance on traffic sensor networks in intelligent transportation systems requires robust methods for ensuring the accuracy and reliability of sensor data. Traditional fault detection techniques often struggle with the high volume of data generated by these sensors, the dynamic variability in traffic patterns, and the lack of sufficient failure data for training detection models. The problem becomes larger when we take into consideration the need to distribute the computational workload to minimize the stress and latency of the main traffic control server. ☐ A multi-tiered structure combined with machine learning models is an ideal framework for realizing these goals in near real-time. In the proposed system, the edge tier employs lightweight statistical models to flag anomalous sensor readings that deviate from expected thresholds. The fog tier clusters spatiotemporally-dependent sensors in the network. The cloud tier takes a sensor’s filtered data and compares it to other relevant sensor data within the same cluster and classifies whether the sensor is operating normally or anomalously. This multi-tiered approach allows for efficient processing by balancing localized data analysis at the edge with more sophisticated, large-scale processing in the cloud. ☐ Innovations are needed to overcome barriers in addressing these challenges of fault detection as well as the issue of fault detection in a spatiotemporal network system. This MS thesis introduces three main innovations: (1) a novel framework for fault detection in traffic sensor systems in non-dilemma zones; (2) a new use of a Density-Based Spatial Clustering of Applications with Noise model modified with dynamic time warping to detect sensor failures in traffic systems, which has not been explored in previous research; and (3) the systematic tuning of parameters of the machine learning models used for the specific context of traffic sensor failure detection. | |
| Advisor | Nejad, Mark | |
| Degree | M.S. | |
| Department | University of Delaware, Department of Computer and Information Sciences | |
| Unique Identifier | 1515021463 | |
| URL | https://udspace.udel.edu/handle/19716/36033 | |
| Language | en | |
| Publisher | University of Delaware | |
| URI | https://www.proquest.com/pqdtlocal1006271/dissertations-theses/multi-tiered-architecture-automatic-sensor/docview/3189990862/sem-2?accountid=10457 | |
| Keywords | Data fusion | |
| Keywords | Edge-fog-cloud computing | |
| Keywords | Graph neural networks | |
| Keywords | Intelligent transportation systems | |
| Keywords | Sensor anomaly detection | |
| Title | Multi-tiered architecture for automatic sensor failure detection in traffic systems: edge-fog-cloud collaboration for enhanced accuracy | |
| Type | Thesis |