基于红外热像采集的BPSO-WD-ESN电路板故障预测

摘要

Some faults of circuit board are difficult to predict accurately due to high integration of the board, disorderly fluctuations of timing module failure,and pseudo-periodicity. A wavelet echo state network( WD-ESN)mode based on bootstrap sampling technique and particle swarm optimization( BPSO)was proposed using thermal infrared imaging to dynamically analyze the function of the board and obtain the changes in thermal image of relative module. The variation range of different components characterized by WD analy-sis,corresponding reserve pool network prediction model was established based on the heat variation,and combination weight matrix of WD-ESN network was optimized using Bootstrap PSO algorithms. Three dif-ferent algorithms and WD-ESN model optimized by BPSO were used respectively to analyze the features cir-cuit of a certain avionics product. Simulative results show that WD-ESN model optimized by PSO can more accurately predict the change of temperature( error Low to 15%)and meet the prediction requirements such as high-speed,real-time,and accurate.%针对电路板集成度高、故障规律波动无序、伪周期性等难以准确预测的问题，借助红外热成像仪应用于电路板故障诊断中，提出了一种基于Bootstrap采样技术的粒子群优化（ BPSO）小波回声状态网络（ WD-ESN）模型，运用红外热像故障分析模式，对功能电路板进行动态分析，获取相关模块的热成像变化数据。利用WD解析出表征不同元器件变化区间，根据不同热量变化规律，建立相应的储备池网络预测模型，采用Bootstrap粒子群算法对WD-ESN网络的组合权值矩阵进行优化。分别采用三种不同算法和经过BPSO优化后的WD-ESN模型对某型航电产品功能电路温度图谱进行分析，仿真结果表明，经过PSO优化后的WD-ESN模型能够更加准确预测温度变化趋势（误差低于15％），满足了高速、实时、准确的红外温度预测的要求。