The Research Of Laser Welded Seam Real-time Detection Device Based On FPGA

In the system of automatic welding,the accuracy and real-time performance of weld detection are important factors affecting the welding quality.In order to complete the high-quality welding,the weld seam detection device should have high processing accuracy and fast processing speed.However,the research on detection devices at this stage is mostly based on serial processors,and the complexity of the algorithms used is high,which leads to a long time for the detection process.Therefore,in this paper,a parallel programmable logic device(FPGA)is used as a processor to design a set of real-time detection devices for welds.Algorithm research and experimental verification are implemented in the realization of the device.Firstly,the shortcomings of high cost,low integration and low real-time detection of the weld seam detection device using industrial cameras,image acquisition cards and serial processors are analyzed.A weld seam detection device with an FPGA as the processor was constructed under the active vision mode.FPGA is used to drive the CMOS image sensor,and the image transcoding and parameter configuration are completed,which solves the problem of high processor occupancy in the image acquisition process.Secondly,the traditional weld seam extraction algorithm based on Gabor transform,image thinning and Hough transform is simulated.Because of the complexity and time-consuming of the algorithm,it is difficult to meet the requirements of welding on the detection speed.Therefore,in order to improve the detection speed,this paper designs a parallel pipelined image processing architecture and improves the algorithm for the detection function.A compensation filter algorithm proposed can effectively enhance the characteristics of laser images and reduce the difficulty of image processing.According to different types of welds,a variety of weld feature extraction algorithms based on data stream operation are designed to reduce the delay of the algorithm to 10?s.The effect of the detection device was verified by experiments,and the detection accuracy was 0.1mm.Finally,for a variety of weld types,the detection device cannot adaptively select the corresponding feature extraction algorithm.This paper uses CNN to carry out the study of automatic weld recognition algorithm.The Inception framework was used to build the network model framework,and the parallel acceleration scheme based on FPGA was designed.The weld type image library was established.In the process of model training,the network structure and parameters were adjusted to increase the recognition accuracy to 97.15%.