Adaptive Engineering Systems: Redundancy-Driven Optimization And System Reliability

Abstract

Deep learning (DL) models are widely used in applications such as image classification, speech recognition, and autonomous systems, but their deployment is limited by high computational cost and sensitivity to hardware faults that reduce system dependability. Traditional fault tolerance methods rely on hardware redundancy, increasing system complexity and resource usage, making scalability difficult. This research proposes an adaptive engineering framework based on redundancy optimization to address these limitations. Data is preprocessed using Z-score normalization for standardization and autoencoder-based feature extraction to obtain compact representations. The proposed FAT-DRNN (Fault-Aware Training-tuned Dynamic Recurrent Neural Network) enhances robustness under simulated hardware faults. It integrates Fault-Aware Training to improve reliability through redundancy-based learning, while DRNN enables temporal modeling and adaptive system dynamics. Fault injection is applied within convolutional layers to make the network resilient without relying on external redundancy. The Python implementation optimizes learning-based redundancy instead of hardware redundancy using adaptive techniques. Results show improved fault tolerance and worst-case accuracy compared to conventional networks, with reduced redundancy. The model achieves 9.6 ms execution time, 97.1% accuracy, 98.2% precision, 97.8% recall, and 98.0% F1-score. Overall, the proposed FAT-DRNN framework demonstrates that integrating fault-aware training with recurrent architectures can significantly improve robustness and efficiency in deep learning systems deployed in unreliable hardware environments while minimizing dependence on external redundancy The approach provides a scalable solution for resilient intelligent systems balancing computational efficiency accuracy and fault tolerance across critical applications requiring dependable performance under real-world operational conditions in practical industrial and autonomous system deployments scenarios use.