Abstract : Uncertainty is inherent to real-world engineering systems, and reliability analysis aims at quantitatively measuring the probability that engineering systems successfully perform the intended functionalities under various sources of uncertainties. In this dissertation, heterogeneous uncertainties including input variation, data uncertainty, simulation model uncertainty, and time-dependent uncertainty have been taken into account in reliability analysis and reliability-based design optimization (RBDO). The input variation inherently exists in practical engineering system and can be characterized by statistical modeling methods. Data uncertainty occurs when surrogate models are constructed to replace the simulations or experiments based on a set of training data, while simulation model uncertainty is introduced when high-fidelity simulation models are built through idealizations and simplifications of real physical processes or systems. Time-dependent uncertainty is involved when considering system or component aging and deterioration. Ensuring a high level of system reliability is one of the critical targets for engineering design, and this dissertation studies effective reliability analysis and reliability-based design optimization (RBDO) techniques to address the challenges of heterogeneous uncertainties. First of all, a novel reliability analysis method is proposed to deal with input randomness and time-dependent uncertainty. An ensemble learning framework is designed by integrating the Long short-term memory (LSTM) and feedforward neural network. Time-series data is utilized to construct a surrogate model for capturing the time-dependent responses with respect to input variables and stochastic processes. Moreover, a RBDO framework with Kriging technique is presented to address the time-dependent uncertainty in design optimization. Limit state functions are transformed into time-independent domain by converting the stochastic processes and time parameter to random variables, and Kriging surrogate models are then built and enhanced by a design-driven adaptive sampling scheme to accurately identify potential instantaneous failure events. Secondly, an equivalent reliability index (ERI) method is proposed for handling both input variations and surrogate model uncertainty in RBDO. To account for the surrogate model uncertainty, a Gaussian mixture model is constructed based on Gaussian process model predictions. To propagate both input variations and surrogate model uncertainty into reliability analysis, the statistical moments of the GMM is utilized for calculating an equivalent reliability index. The sensitivity of ERI with respect to design variables is analytically derived to facilitate the surrogate model-based product design process, lead to reliable optimum solutions. Thirdly, different effective methods are developed to handle the simulation model uncertainty as well as the surrogate model uncertainty. An active resource allocation framework is proposed for accurate reliability analysis using both simulation and experimental data, where a two-phase updating strategy is developed for reducing the computational costs. The framework is further extended for RBDO problems, where multi-fidelity design algorithm is presented to ensure accurate optimum designs while minimizing the computational costs. To account for both the bias terms and unknown parameters in the simulation model, Bayesian inference method is adopted for building a validated surrogate model, and a Bayesian-based mixture modeling method is developed to ensure reliable system designs with the consideration of heterogeneous uncertainties.