Vol. 1, Issue 1, Part A (2024)

Industrial systems today are increasingly reliant on automation and sensor-enabled equipment to ensure continuous production and minimal downtime. The ability to predict equipment failures before they occur is a core element of predictive maintenance strategies, which aim to reduce unscheduled interruptions and maintenance costs. This study proposes a data-driven methodology for failure prediction using time-series sensor data. By leveraging the Python-based Pandas library for efficient data manipulation and pre-processing, and Long Short-Term Memory (LSTM) models implemented through the Keras framework, the study demonstrates how raw sensor data can be transformed into predictive insights. The Sliding Window Method is used to segment continuous sensor data into structured sequences suitable for LSTM training. Experimental results on real-world manufacturing datasets show that the proposed framework can successfully detect potential failures ahead of time, enabling timely intervention and enhancing equipment reliability.