A Systematic Literature Review on Machine Learning Techniques for Enhancing Embedded Hardware Reliability

Embedded systems (ES) have played a vital role in industrial automation and critical infrastructure, but their reliability has often been compromised by hardware faults, leading to downtime and safety concerns. Traditional threshold-based fault detection methods have frequently failed to adapt to dynamic environments and have struggled to identify early-stage failures. This study reviewed the effectiveness of artificial intelligence (AI), specifically machine learning (ML) models, for fault detection in ES. A systematic review methodology was employed to analyze the diagnostic performance of several deep learning (DL) architectures, including hybrid convolutional neural network-long short-term memory (CNN-LSTM) models, when implemented on resource-constrained edge devices. The results showed that optimized AI models achieved higher diagnostic accuracy and earlier fault identification compared to conventional approaches. Furthermore, these models enabled real-time, energy-efficient operation on platforms such as Raspberry Pi and ESP32 microcontrollers. It was concluded that AI-driven solutions significantly enhanced predictive maintenance and operational reliability in embedded system applications, demonstrating their transformative potential for future industrial systems.