A Hybrid Expert System Using Symbolic Reasoning and Neural Networks for Predictive Maintenance in Mechatronic Systems

Predictive maintenance (PdM) in mechatronic systems demands high-precision failure prediction and interpretability for real-time operational decisions. This study presents a hybrid expert system integrating symbolic reasoning and Deep Neural Networks (DNNs) to enhance predictive accuracy and semantic traceability. The symbolic layer consists of 42 fuzzy inference rules, enabling domain expert interpretability, while the neural network layer comprises a 4-layer feedforward architecture with 128-64-32-1 units using ReLU and sigmoid activations. Experiments were conducted on a real-world dataset, and the hybrid model achieved an accuracy of 96.8%, a precision of 94.22%, and a recall of 97.31%, outperforming conventional Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) models, and rule-based systems by margins of 3.2–7.8%. The proposed method reduced false positives by 21.4% and improved time-to-failure prediction by 18.7% compared to standalone models. Maintenance scheduling optimized using the proposed model yielded a 14.5% reduction in unplanned downtime. The hybrid inference strategy not only improved prediction granularity but also supported rule-based diagnostics. This framework significantly advances predictive intelligence in safety-critical mechatronic domains.