Abstract: In response to the current issue of low annual production of marine diesel engines, resulting in insufficient quality data and an inability to accurately assess assembly quality in a timely manner, a quality assessment method for high-dimensional small samples has been proposed. In consideration of data imbalance, a data generation approach based on VAE-GAN has been introduced, wherein the coding process is enhanced using VAE networks, and the original dataset is effectively expanded. Furthermore, a feature selection network has been constructed to eliminate redundant features and extract key processes, thereby improving the effectiveness of training. Finally, the assembly process has been modeled as time series data by CNN-LSTM networks, enhancing the accuracy of assembly quality evaluation. Experimental verification has been conducted using quality data from a marine diesel engine, providing theoretical guidance for evaluating high-dimensional small sample datasets.