激光深熔焊接过程实时监测系统设计

Design of Real-time Monitoring System for Laser Penetration Welding

基于LabWindows/CVI开发环境,采用红外光、蓝紫光和麦克风3种传感器构成了焊缝熔透状态的实时监测系统。在多传感器的数据融合中,首先对3路测量信号进行时频分析,提取能够反映焊缝状态的特征,然后利用人工神经网络对特征值进行融合计算,得到焊缝是否熔透的判断结果。在对网络权值的训练中采用了随机变异粒子群算法,大大地提高了训练速度和精度。实验结果表明该系统应用于焊接过程的实时质量监测是可行的。

A monitoring system was constructed to classify the welding states as adequate or inadequate penetration by using infrared, ultraviolet and sound sensors. Within a procedure of multi-sensor data fusion, three kinds of signals were analyzed in time and frequency domain, then the welding states related features were extracted, the welding states were determined by an artificial neural network with the feature values as inputs. The weights of the neural network were trained by an improved algorithm of particle swarm optimizer with stochastic mutation （SMPSO） ,which improved the speed and accuracy of the training. The experimental results show that it is feasible to use the system for real time monitoring on laser welding process.