智能结构损伤自诊断的PSO-LSWSVM方法

PSO-LSWSVM used to self-diagnose damages for smart structures

损伤检测方法是关联智能结构实现损伤自诊断功能的一个重要研究内容。针对支持向量机研究的关键与难点——核函数的构造及核参数的优化问题,基于小波核函数的近似正交性,研究了基于小波核的最小二乘支持向量机(LSWSVM)方法,并采用粒子群优化算法(PSO)对LSWSVM参数进行优化,从而构造PSO-LSWSVM方法。基于压电元件的正逆压电效应,搭建损伤自诊断压电智能结构实验系统,对各压电传感器的响应信号进行功率谱密度最大值(PSM)特征提取。基于各压电传感器的响应信号特征,将该PSO-LSWSVM方法应用于智能结构损伤自诊断,并对该方法进行性能评价。结果表明,在同等条件下,LSWSVM有比基于高斯核函数的最小二乘支持向量机(LSSVM)更高的损伤检测精度及更强的推广能力,相比于传统遗传算法(TGA),该方法中粒子群优化算法(PSO)具有较好的寻优能力和收敛速度。

The method of damage detection is an important issue related to the self-detecting damage function for smart structures. Aimed to the key and difficult problem on construction of kernel functions and optimization of kernel parameters for SVM,based on the approximately orthonormal property of wavelet kernel function,this paper proposed least square wavelet support vector machine( LSWSVM),and optimized its tuning parameters by particle swarm optimization( PSO),and thus constructed the method of PSO-LSWSVM. On basis of direct and converse piezoelectric effects of PZT materials,it set up a piezoelectric smart structures testing system with the self-diagnosing damage function,and obtained its piezoelectric responsive signals' features through the method of power spectrum density maximum( PSM). Then,based on the features of piezoelectric responsive signals,it applied the method of PSO-LSWSVM to self-diagnose damages for smart structures,and evaluated its performance by some tests. The results show that LSWSVM possesses the higher detecting accuracy,bitter dissemination ability than least square support vector machine( LSSVM) with Gaussian kernel function under the same conditions. Compared to traditional genetic algorithm( TGA),PSO in the method possesses the better searching ability and the faster convergence speed.