Predictive maintenance (PdM) is critical to ensure optimal operating efficiency and minimize costly failures of industrial machinery. The PdM leverages a machine learning (ML) method to predict remaining useful life (RUL) for implementing minimal-cost and reliable maintenance. RUL prediction involves multiple steps, such as data collection, data pre-processing, and RUL estimation, each of which incorporates various methods. This study conducted a critical review of RUL estimation and data pre-processing specifically for turbofan engines and bearings, categorizing existing models to offer a high-level perspective on RUL prediction. This review demonstrates that the indirect mapping method exhibits outstanding prediction accuracy compared to the direct mapping method. Moreover, it highlights that autoencoder techniques and their variants demonstrate commendable performance in extracting features from turbofan engines and bearing datasets. Furthermore, the paper proposes potential areas for future research to improve RUL prediction in this domain.