On-line Prediction Technology Of Arc Welding Quality Based On Spectrum And Process Cooperative Perception

With the rapid development of the manufacturing industry,traditional prediction methods can no longer meet the industrial requirements.As an important part of intelligent manufacturing,the research on the online arc welding quality prediction technique is imminent.The arc spectrum and electrical signals reflect various changes in arc composition during welding,and contain rich information related to arc welding quality.This paper focuses on the online arc welding quality prediction technology based on cooperative awareness of spectrum and processes.The main contributions of this paper are listed as follows:(1)Qualitative prediction Technology of arc components: FPGA is used to acquire the electrical signal in real-time,and trigger the spectrometer at the peak current for arc spectrum.Then,qualitative prediction is realized by signal preprocessing,dimensionality reduction and classification.Aiming at information preprocessing,a spectral manifold structure extraction algorithm and a new electrical signal preprocessing method are proposed to realize information noise reduction and feature extraction.Aiming at dimensionality reduction,an improved local retention projection algorithm is proposed to achieve efficient dimensionality reduction,which mainly includes adding intra-class and regular constraints and improving the weight matrix construction method.Aiming at classification,an Ada Boost-ELM algorithm based on rule tree is proposed to achieve accurate classification,which establishes a rule tree with known process parameters as decision features,and then uses Ada Boost-ELM algorithm to predict.Finally,the effectiveness of the proposed method is verified by artificial controllable experiments of doping components,shielding gas components and flow rate.(2)Quantitative prediction Technology of arc components: Based on qualitative prediction,the quantitative prediction is realized by signal preprocessing and neural network regression.In order to meet the demand of quantitative prediction,the signal preprocessing of qualitative prediction is modified.Aiming at neural network regression,Res Net regression is first introduced into the field of online arc welding quality prediction based on cooperative awareness of spectrum and processes,and modified to better meet the requirement of arc spectrum and electrical signal regression.Finally,the effectiveness of the proposed method is verified by artificially controllable experiments of shielding gas flow rate and wire composition.