The paper aims to realize the fault real-time fault detection of the temperature sensor in the heat sealing process of the automatic packaging machine. The generalized regression neural network (GRNN) was used to construct a state automatic detection network of heat-sealed temperature sensor, and then the Chaos Accelerated Fruit Fly Optimization Algorithm (CAFOA) was used to study the generalized regression neural network. Factor optimization was selected to solve the optimal learning factor. By establishing a CAFOA-GRNN automatic detection model, combined with the statistical confidence interval method, the faults were classified and diagnosed. In the sensor failure experiment, the ideal fault function was superimposed with the historical operation data to generate the fault data set, to verify the established model. Good detection effect was obtained, and the accuracy was high. The method realizes the real-time detection of sensor failure, and can be used to improve the reliability of production, and has certain engineering practical significance.