DEEP LEARNING ARCHITECTURES FOR ENGINEERING DATA ANALYSIS AND PROCESS AUTOMATION

Abstract

This study investigates the application of deep learning architectures for engineering data analysis and process automation using a real-world manufacturing dataset. The dataset, consisting of temperature, pressure, vibration, flow rate, tool wear, energy consumption, and machine failure indicators, was analyzed to explore predictive modeling approaches. Descriptive statistics and correlation analysis revealed key relationships, particularly the strong influence of tool wear, vibration, and pressure on machine failures. Several deep learning models, including Deep Neural Networks, Convolutional Neural Networks, Long Short-Term Memory networks, Autoencoders, and hybrid architectures, were evaluated against a logistic regression baseline. Performance was assessed using accuracy, precision, recall, F1-score, and ROC analysis, demonstrating that deep learning approaches achieve superior predictive reliability while capturing nonlinear dependencies. Interpretability techniques, such as feature importance analysis, further enhanced understanding of model behavior. The results highlight deep learning’s potential to enable predictive maintenance, minimize downtime, and support intelligent process automation in industrial systems.