AI Driven Approach in Smart Manufacturing in Bangladesh

Abstract

Predictive Maintenance (PdM) has become essential in smart manufacturing for reducing Downtime, improving efficiency, and cutting operational costs. The primary aim is to develop an Artificial Intelligence (AI) driven PdM framework for induction motors, leveraging IoT-based condition monitoring and time-series forecasting to estimate Remaining Useful Life (RUL) and enable intelligent maintenance scheduling. For this purpose, real-time Vibration and temperature data were collected from 2022 to 2024 using MPU-6050 sensors, followed by preprocessing, feature extraction, and fault trend analysis. The Prophet algorithm, known for handling seasonality and holiday effects, was employed for forecasting failure patterns and RUL estimation. Experimental analysis revealed distinct fault stages: unbalance, mechanical looseness, and bearing degradation; captured through Fast Fourier Transform (FFT) and time-domain features. Model validation across three axes showed strong performance with Coefficient of Determination (R²) up to 0.958, Root Mean Square Error (RMSE) as low as 0.110, and Mean Absolute Error (MAE) of 0.088, enabling accurate prediction of failure windows and proactive scheduling. However, limitations include a narrow dataset, reliance on two sensor modalities, and the exclusive use of Prophet, which struggles with highly non-linear dynamics. Future work would address these by incorporating hybrid AI models and multi-sensor fusion for improved prediction accuracy and scalability in large-scale deployments.