Hybrid Deep Learning-Based Fault Prediction for Industrial Machinery Using CNN-LSTM and CNN-GRU: CLKKSM Facility Application and Data-Driven Predictive Maintenance Action Plan
Year: 2026
Type: article
Abstract: Unplanned machine downtime in manufacturing facilities leads to significant production losses, rising maintenance costs, and occupational safety risks.This study compares two hybrid deep learning architectures-Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) and Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) -for monthly fault frequency prediction using an eight-year operational dataset from the CLKKSM facility.One-dimensional convolutional layers extract local temporal features from the raw time series, which are subsequently encoded by recurrent cells to capture long-range dependencies.Both models were trained on historical data and validated on a held-out period, then evaluated against actual fault events in early 2026.CNN-LSTM consistently outperformed CNN-GRU across all error metrics on the out-of-sample test set.The performance gap is attributed to the four-gate memory mechanism of LSTM, which models medium-and long-range temporal dependencies more effectively than the two-gate GRU structure.Building on CNN-LSTM forecasts, a three-phase predictive maintenance action plan was developed for the April-December 2026 period, incorporating machine-level priority classification, risk-stratified schedules, and trackable Key Performance Indicator (KPI) targets.The proposed methodology bridges the gap between model outputs and field implementation, offering a transferable framework for data-driven predictive maintenance in industrial settings.
Authors: Ermiş, Salih, Taşdemir, Oğuz, Başgül, Çağatay
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
Funder: Strong
OpenAlex ID: W7208769612