In recent years, swift changes in market demands toward achieving carbon neutrality have driven significant developments within the automotive industry. Consequently, employing computer simulations in the early stages of vehicle development has become imperative for a comprehensive understanding of performance characteristics. Of particular importance is the cooling performance of vehicles, which plays a vital role in ensuring safety and overall performance. It is crucial to predict optimal cooling performance, particularly about the heat generated by the powertrain during the initial phases of vehicle development. However, the utilization of thermal analysis models for assessing vehicle cooling performance demands substantial computational resources, rendering them less practical for evaluating performance associated with design changes in the planning phase. This paper introduces a method for constructing a low-dimensional model capable of predicting the time series response of cooling performance using a surrogate model based on thermal analysis. The thermal analysis model in this study is evaluated using design variables obtained through experiments, and a training data matrix is constructed from the corresponding time series responses. Unsupervised learning is employed to extract key features from these responses. For the regression model, a Gaussian process is applied to each latent variable derived from the unsupervised learning process. This transformation allows for a reduction in computational costs, shifting from high-dimensional calculations to a low-dimensional latent space for prediction. The proposed method is then applied to analyze the time series response of engine coolant temperature, obtained from the engine thermal analysis model, effectively demonstrating its utility.
