Deep Learning Approaches for the Classification of Diesel Injector Spray Patterns

Abstract

Accurate classification of diesel injector spray patterns is critical for evaluating fuel atomization quality and optimizing combustion efficiency in internal combustion engines. Traditional image analysis methods often struggle to capture the complex and dynamic nature of spray structures. In this study, we explore deep learning-based approaches particularly Convolutional Neural Networks (CNNs) for the automated classification of diesel spray images. A comprehensive dataset of high-speed diesel spray images was used, capturing a range of injector conditions and operating parameters. Several CNN architectures, including VGG16, ResNet50, and custom shallow networks, were trained and evaluated for their performance in classifying spray patterns into predefined categories based on shape, dispersion, and uniformity. Data augmentation techniques were employed to improve generalization and robustness. The results demonstrate that deep learning models significantly outperform traditional machine learning approaches in terms of classification accuracy and resilience to noise and image distortion. The best-performing model achieved over 95% accuracy in distinguishing between normal and defective spray patterns. These findings highlight the potential of deep learning techniques to assist in real-time diagnostics and quality control of fuel injection systems, ultimately contributing to cleaner and more efficient diesel engine performance.