زهراسادات سجادي

Given the critical role of elevators in urban life an‎d the severe consequences of their failures, the development of novel methods for condition monitoring an‎d predicting the failure of key components has become essential. Traditional maintenance strategies such as corrective repairs or periodic servicing, in addition to imposing high costs, are incapable of preventing sudden breakdowns. Therefore, in recent years, data-driven an‎d artificial intelligence approaches for predictive maintenance an‎d remaining useful life (RUL) estimation have gained significant attention. In this study, with a focus on the elevator door—identified in the initial risk assessment as the component with the highest failure risk—a data-driven intelligent framework was developed. To this end, a set of sensors, including analog an‎d digital accelerometers, temperature sensors, an‎d magnetic sensors, was installed on an 11-story traction elevator, an‎d data were collected over a 27-working-day period at a sampling rate of 33 Hz. The raw data, after normalization an‎d windowing, were transformed into two-dimensional images using the Stacked Channel Imaging (SCI) method. Data analysis was carried out using the EfficientNetB7 convolutional neural network. The model successfully classified the four primary door states (fully open, fully closed, opening, an‎d closing) with an overall accuracy of 98%. Subsequently, by modifying the network’s output layer for regression, the estimation of the remaining useful life became feasible. The results demonstrated that the model achieved high precision in dynamic classes, an‎d combining data from two failure events significantly improved accuracy—yielding coefficients of determination (R²) of 91% an‎d 94% for classes 2 an‎d 3, respectively. Moreover, the model was able to distinguish data from one day prior to failure from normal data with an accuracy of approximately 97%. The findings indicate that the integration of multi-sensor data with advanced deep learning methods provides an effective approach for intelligent elevator monitoring. This approach not only enhances monitoring accuracy but also enables early warning, reduces unexpected breakdowns, optimizes maintenance costs, an‎d improves user safety.

رسان ور، ت مين عمرمفيد باقيمانده، ننهداري پيشبينانه، سن ور، يادگيري عميق.

Elevator, Remaining Useful Life Estimation, Predictive Maintenance, Sensor, Deep Learning

zahra sadat sajadi

dr mohammad riahi