Fault diagnosis of radio frequency circuit using heterogeneous image fusion

Abstract

Radio frequency (RF) technology covers all fields of life, but with the increase in the size and complexity for circuit devices, traditional contact detection methods have difficulty solving the fault diagnosis problem of electronic circuits. The infrared thermal image fault diagnosis technology can realize perfectly real-time monitoring of the circuit due to it being non-contact and non-destructive and having fast recognition. At present, infrared thermal diagnosis technology can locate the fault but not know what causes the fault. Therefore, we propose a heterogeneous fusion image fault diagnosis method for RF circuits based on image perspective transformation and convolutional neural network (CNN). First, a variety of faults are set on the circuit board, and the faults are numbered; then the optical image, infrared thermal image, and oscilloscope image of circuit board under each fault are obtained by the corresponding instrument and fused heterogeneously. The fused image set is divided into the training set and validation set after enhancement and finally is fed into the CNN for training and classification. The accuracy of the final validation set reached 97.3%, which was 3.4% higher than that of only using infrared thermal image and optical image fusion, 2.7% higher than that of the LeNet5 model, and 1.5% higher than that of Vgg16 model, and the time was shortened by 15 min.