模块化神经网络容差模拟电路故障检测

Modular neural network tolerance analog circuit fault detection

摘要:

容差模拟电路故障检测对于电子设备的稳定运行而言至关重要,针对传统检测算法计算代价大、训练时间长及检测误差率高的不足,提出基于模块化神经网络的容差模拟电路故障检测算法研究。对神经网络检测模型的功能模块进行划分,并基于功能模块提取容差模拟电路的故障信号特征；基于样本中心到故障特征点的欧式距离,对比故障样本的特征向量,依据模块化神经网络决策分类函数,实现对容差模拟电路故障的准确定位和检测。仿真数据表明,在不同样本容量条件下提出检测算法均具有优势,最低误差值为0.382%.

Abstract:

The fault detection of tolerance analog circuit is very important for the stable operation of electronic equipment. In view of the shortage of large calculation cost, long training time and high error rate, the fault detection algorithm of tolerance analog circuit based on modular neural network is proposed. The function modules of the neural network detection model are divided, and the fault signal features of the tolerance analog circuit are extracted based on the functional modules, and the Euclidean distance based on the sample center to the fault feature point is compared with the feature vectors of the fault samples, and the fault tolerance analog circuit fault is realized by the modular neural network decision classification function. Accurate location and detection. The simulation data show that the detection algorithm has advantages under the condition of different sample capacity, and the minimum error is 0.382%.