A Novel Approach to Elevator Safety: Semi-Supervised Data Labeling for Function Classification and Uneven Rope Tension Detection Using FBG Sensors

Automatic recognition of elevator functions and monitoring the tension of suspension ropes are essential for ensuring safe operations and facilitating predictive maintenance strategies for elevators. Current monitoring systems often depend on indirect measurements, which lack the granularity and real-time capabilities necessary for effective predictive maintenance. Additionally, traditional signal processing methods struggle to extract valuable insights from the high volume of noisy and unlabeled data generated by various sensors. Deploying an artificial intelligence (AI)-driven solution for this purpose faces several challenges, including unlabeled data, variability in strain values due to environmental factors and elevator configurations, and the absence of standardized datasets. This paper introduces a novel semi-supervised data labeling approach that utilizes six fiber Bragg grating (FBG) sensors to address these challenges. The FBG sensors provide direct monitoring of strain on individual ropes, allowing for real-time detection of uneven tension. To tackle the issue of unlabeled data, a unique semi-supervised data labeling strategy is implemented. This includes principal component analysis for dimensionality reduction, K-means clustering for pseudo-labeling, and transition-based temporal labeling method. A two-layer long short-term memory network is then trained on this processed data to validate the labeling approach, achieving high accuracy in classifying elevator functions, as assessed by metrics such as accuracy, precision, recall, and F1-score. This method illustrates the feasibility and effectiveness of AI-driven FBG sensor technology in enhancing elevator safety and enabling predictive maintenance.