针对瓦斯传感器常见的偏置型、冲击型、漂移型和周期型4种突发型故障,以小波分析和RBF神经网络为基础,提出了由小波包分解提取特征能量谱与扩展Kalman滤波算法(EKF)优化的RBF神经网络进行模式分类辨识的瓦斯传感器故障诊断方法.对瓦斯传感器的输出信号进行小波包分解,运用基于代价函数的局域判别基(LDB)算法进行裁剪,获取最优的特征能量谱,经处理后作为特征向量训练EKF-RBF神经网络,采用参数增广和统计动力学方法,通过带有整定因子的EKF参数估计,用来辨识瓦斯传感器的故障类型.实验结果表明:该方法的辨识正确率在95%以上,误报率和漏报率都明显优于其他算法,能够有效用于瓦斯传感器的故障在线诊断.%For four types of common abrupt faults of gas sensor, namely offset, impact, drift and periodic types, on the basis of wavelet analysis and RBF neural network, a method of the gas sensor fault diagnosis was proposed based on the pattern classification of characteristic energy spectrum extracted by the decomposition of wavelet packet and RBF neural network optimized by Extended Kalman Filter (EKF). The optimal characteristic energy spectrum was obtained through the decomposition of wavelet packet of output signal of gas sensor and optimally cut by Local Discriminant Base (LDB) based on the cost function. After processed, as the characteristic vector for training EKF-RBF neural network, adopted augnented parameters and method of statistical mechanics, and through the EKF parameter estimation with tuning factor, it was used to identify the fault type of sensor. The experimental results show that the identification accuracy is above 95 ％ ,its rate of false alarm and fail alarm is superior to other algorithms, and the method can be effectively applied to the online fault diagnosis of gas sensor.
展开▼
