Research On The Application Of Deep Learning In The Detection Of Strain Clamps In Power Equipment

In the power industry,the quality of the strain clamp in the overhead line is directly related to the safe and stable operation of the transmission line.At present,the defect detection of strain clamps is mainly realized manually.In the process of manual detection,X-ray non-destructive testing technology is usually used to collect the digital images of the strain clamps,and then the inspector can identify the defects of the professionally processed image.The disadvantage of this detection method is that the detection efficiency is low,and the corresponding detection results may be artificially interfered.In order to meet the need of establishing the digital image AI application platform for power equipment,this thesis applies deep learning to the power industry and uses convolutional neural network models to classify the structural defects of strain clamps,thereby solving the problems and deficiencies of traditional manual detection methods.The main tasks include:(1)According to the requirements of actual engineering projects,we sorted and collected the digital images of the strain clamps and prepared the data,combining with the composition structure of the strain clamps of compression type and the defect assessment details;(2)We delved into the structure and principle of several basic models of convolutional neural network,and used these models to carry out defect classification detection on the data set of strain clamps.Then we compared the performance of these models,and selected the DenseNet with the best comprehensive performance as the benchmark model;(3)The densely connected structure of DenseNet can reuse features,and the SE module of SENet can calibrate the importance of feature channels.Based on this,we propose a defect detection model that integrates SE modules and dense connections,design and implement the SEPreDenseNet network structure,compare the effects before and after model improvement,and evaluate the performance of the improved model.In this thesis,the convolutional neural network is first applied to the task of classification and detection of strain clamps.At the same time,the structure of the benchmark model was optimized and improved.It contains work on applied innovation and theoretical innovation,which has provided a theoretical basis for the application and development of deep learning in the power industry and certain reference significance for the follow-up research.