AN AI-ENABLED HIERARCHICAL LSTM AUTOENCODER FRAMEWORK FOR PREDICTIVE MAINTENANCE IN INTELLIGENT TRANSPORTATION SYSTEMS USING MULTIMODAL WIRELESS SENSOR DATA
L. Arokia Jesu Prabhu, R. T. Thivya Lakshmi, Mokana Muthu Kumarasamy, P. V. V. S. Srinivas, Selvamani Indrajith, S. Navaneethan, Aseel Smerat, K. Prakash
Abstract: For smart transportation systems (STS), predictive maintenance is essential to improve reliability and efficiency, but traditional models struggle to detect complex, time-dependent fault patterns. This work proposes a Hierarchical-LSTM-autoencoder (H-LSTM-AE) based predictive maintenance model (PMM) within a 4-layer hierarchical model to process multi-modal sensor data and enable real-time proactive maintenance. The proposed H-LSTM-AE model includes layers for data measurement, transmission, analysis, and prognosis. The H-LSTM-AE is designed to capture temporal dependencies in transportation data and perform anomaly detection (AD) across different operational conditions. The proposed H-LSTM-AE achieved a detection accuracy of 96.82%, a false positive rate of 2.34%, and an average response time of 32.5 seconds under steady-state operating conditions. The model achieved a detection delay of 37.4 seconds for the whole evaluation. Compared with baseline models such as Isolation Forest (iForest), autoregressive integrated moving average (ARIMA), convolutional neural network (CNN) + LSTM, and variational autoencoder (VAE), the proposed H-LSTM-AE outperformed them in precision (94.37%), recall (92.83%), and F1 (93.59%). The prognosis layer also allows for prioritized maintenance scheduling based on anomaly severity and operational criticality. These results validate the proposed H-LSTM-AE for reliable, proactive maintenance in intelligent transportation systems. This study employs a prediction layer to schedule maintenance tasks by anomaly type and functional importance. The above results validate the H-LSTM-AE for reliable, proactive maintenance in STS.
Keywords: Smart transportation systems; Long short-term memory; Autoencoder; Anomaly detection; Sensor networks; Artificial Intelligence; Machine learning; Internet of Things
DOI: 10.24874/PES08.03.027
Recieved: 01.06.2026 Revised: 16.07.2026 Accepted: 11.08.2026
UDC:
Reads: 11 
