Using ChatGPT for the intelligent diagnostics of complex technical systems

Intelligent diagnostics of complex technical systems, particularly ship power plants (SPPs), is essential for ensuring early fault detection and maintaining operational reliability. This study presents a methodological approach for integrating the ChatGPT language model into automated SPP diagnostics. This study aimed to develop a methodological approach for using the ChatGPT language model in the automated diagnostics of complex technical systems (CTSs), particularly SPPs. The proposed methodology consists of several stages: data collection, preprocessing, model training, anomaly detection, and the generation of diagnostic recommendations. The system integrates ChatGPT with real-time data streaming (Apache Kafka) and neural network-based anomaly detection using autoencoders and Long Short-Term Memory (LSTM) models. Experimental validation was carried out using real operational datasets from ship power plant systems. The proposed approach demonstrated a significant improvement in fault detection accuracy, increasing it by 15% compared with conventional threshold-based methods. The diagnostic time was reduced by a factor of nine, which enabled near real-time identification of deviations. The model achieved an accuracy rate of 92.8% when classifying abnormal states and correctly identifying early-stage faults in key parameters such as pressure, temperature, and rotation speed. The analysis of reconstruction error distributions confirmed the ability of the system to distinguish between normal and anomalous system behaviour. Detected anomalies were crossvalidated with expert assessments, confirming the practical applicability of the model in real-world diagnostics. Furthermore, the implementation of the proposed approach enables predictive maintenance planning, which contributes to reducing operational risks and lowering maintenance costs. The integration of ChatGPT into ship power plant diagnostic systems enhances the automated processing of technical documentation and operational logs, increasing the transparency and accuracy of fault explanations. The results of this study may be applied in ship engineering, industrial automation, and technical maintenance, contributing to the improved safety and reliability of complex technical systems