Elevator Online Monitoring & Intelligent Fault Diagnosis

Elevators are among the most frequently used and safety-critical assets in high-rise buildings, yet fault samples such as shaking and emergency stops are scarce, and signal patterns vary widely across elevator models and service ages — thresholds and single-model approaches struggle to keep up. The team links car-mounted triaxial accelerometers, intelligent visual monitors, and a centralized monitoring platform into a live system, with an Attention-based Multi-Task Learning Bidirectional Long Short-Term Memory network (AT-MTL-BiLSTM) at its diagnostic core that automatically zeroes in on the critical moments around fault onset. Validated on operational data from real in-service elevators, the system achieves accuracy, recall, and F1 scores above 0.94 for all three fault types: front-back shaking, left-right shaking, and emergency stop.

1. 1

   Turn Elevator Shaking into Measurable Signals

   An integrated sensor module mounted on the car captures X, Y, and Z acceleration along with angular velocity, angle, speed, and noise channels; an intelligent visual monitor simultaneously records door operations and cabin footage. Professional maintenance personnel labeled operational data from multiple in-service elevators segment by segment, producing several real-world datasets.

2. 2

   Let the Model Find the Seconds That Matter

   Elevators of different service ages exhibit faults at different time scales, requiring adaptive window lengths. A Bidirectional LSTM (BiLSTM) reads each window in both directions; a temporal attention layer then scores every time step, concentrating weight on the critical moments around fault onset while suppressing noise. The attention weights themselves are interpretable, revealing exactly which segment drove each diagnosis.

3. 3

   Three Tasks, One Model — Accurate with Few Samples

   The primary task classifies fault type, while auxiliary tasks detect anomaly status and estimate severity; all three tasks share one representation and are trained jointly. The auxiliary tasks force the shared layers to learn more generalizable features under small-sample conditions, compensating for the bias introduced by scarce, imbalanced fault samples.

4. 4

   From Algorithm to a Deployable Platform

   Diagnostic outputs feed back into an industrial IoT platform: a monitoring center aggregates data from multiple elevators, the server processes it centrally and pushes results to phones, tablets, large-screen dashboards, and desktops for real-time condition monitoring, health assessment, and automatic fault alerts, with extensions toward predictive maintenance scheduling and remote fault handling. An elevator simulation testbed in the lab has completed end-to-end integration of sensing, vision, and the platform.