The advancement of the Industrial Internet of Things and digital-intelligent machinery presents considerable demands in detecting impurities in industrial equipment lubricants or oils. Herein, a Tubular oil–solid triboelectric nanogenerator (TOS-TENG) is designed as a self-powered sensor for impurity detection in insulating oil flow. Specifically, an oleophobic triboelectric sensing layer with superior electron trapping capability is created based on Nano silica and 1H,1H,2H,2H-perfluorodecyltriethoxysilane. Additionally, a periodic start-stop oil flow control method that facilitates the transition between flowing and stationary insulating oil is proposed, thereby generating periodic output signals from the TOS-TENG. Furthermore, a correlation between the output characteristics of the TOS-TENG and impurity properties (type, content) is established through a deep learning algorithm based on convolutional neural networks, enabling the identification of common impurities such as moisture, metal, and fiber particles in oil. As a proof of concept, an industrial-grade TOS-TENG-based device for in situ online monitoring of impurities in oil-immersed transformers is designed. This work offers novel insights into the compositional analysis of continuous fluids and provides a novel strategy for oil condition sensing in smart industrial equipment.