DescriptionThermal systems play significant roles in the engineering practices and our lives. To improve those thermal systems, it is necessary to model and optimize the operating conditions. More importantly, the design uncertainties should be considered because the failures of the thermal systems may be very dangerous and produce large loss. This study focuses on the parametric modeling and the optimization of the thermal systems with the considerations of the design uncertainties. As an example, the material processing thermal system, the Chemical Vapor Deposition (CVD), is simulated with different inlet velocities and susceptor temperatures. Several responses are considered to represent the performance of the thin-film deposition, including the percentage of the working area, the mean of the deposition rate, the root mean square of the deposition, and the surface kurtosis. Those responses are then parametrically modeled by one of the Response Surface Method (RSM), the Radial Basis Function (RBF), and utilized to formulate the optimization problems to enhance the system performances. However, it is not until the design uncertainties are considered that the thermal system designs have high risk of the violations of the performance constraints. One of the Reliability-Based Design Optimization (RBDO) algorithms, the Reliability Index Approach (RIA), is used to solve the optimization problems with the design uncertainties However, the algorithm suffers from a convergence problem when the design point falls into the infeasible domain during the optimization process. A Modified Reliability Index Approach (MRIA) is proposed with a modified definition of the reliability index, and utilized to solve the RBDO problems of the CVD process. The MRIA converts the design space to the standard normal space and finds the Most Probable Points (MPPs) to evaluate the failure probabilities of the performance constraints. The probabilistic optimization problem is then solved using the approximate probabilistic constraints generated in terms of the MPPs. The MRIA has been used to solve several different optimization formulations with both normally and lognormally distributed random variables. The Monte Carlo Simulation (MCS) results verify that the optimal solutions have acceptable failure probabilities. As a result, the proposed strategy of parametrically modeling and optimizing with design uncertainties can be applied to the experiments or the simulations of other thermal systems to improve their productivity, maintain the quality control, and reduce the probability of system failure.