Proceedings Article | 10 April 2023
KEYWORDS: Neural networks, Education and training, Semiconductor lasers, Matrices, Image segmentation, Design and modelling, Titanium dioxide, Resistance, Diagnostics, Convolutional neural networks
With the rapid development and breakthrough of semiconductor laser technology, the quality of semiconductor laser products, wavelength ranges and output power are rapidly improving, and the product range is becoming increasingly rich. Semiconductor lasers have the advantages of small size, light weight, high electro-optical conversion efficiency, stable performance, high reliability and long life, etc., and have revealed their dominant position in the field of lasers At present in the semiconductor laser research, semiconductor lasers are prone to defects under the influence of various environmental factors. In the traditional way, manual miscopy is a commonly used method of detecting surface defects. However, because of its low sampling rate, poor accuracy, low efficiency and large labor costs, manual microscopy cannot meet the needs of quality inspection. In order to solve the problem of high quality control and production costs in the microscopy process, we designed a trouble shooting method for convolutional neural networks. This paper attempts to implement a single-conductor laser radiation fault diagnosis system based on memristor convolutional neural network, and studies the integrated diagnosis system of integrated circuit electrical faults and radiation faults. Firstly, we should classify the problems that often arise in semiconductor lasers. Then the data in the fault state is collected through modeling simulation to form a data set. Finally, the dataset is classified and diagnosed by writing program. After training, the neural network fully realizes the classification diagnosis of radiated fault data and normal working data. The accuracy of the program is up to 95% or more.