基于广义形态学与多场信息融合的复杂供输弹系统早期故障预示方法研究

In order to overcome the limitations of the methods that were applied in the ammunition feeding system fault diagnosis, a new method for complex ammunition feeding system incipient-fault prediction based on generalized morphology and multi-field information fusion will be proposed. This method can make up the deficiencies of the existing incipient-fault prediction and solve the key issues such as weak feature extraction, multi-field information fusion, dynamic pattern recognition and evolution tendency prediction. The research approach of this method is as follows. Firstly, the idea of using the generalized morphological method to implement weak feature extraction, multi-field information fusion, dynamic pattern recognition and evolution tendency prediction will be proposed. A new theory of incipient-fault prediction will be established. Secondly, a new approach utilizing the new method of multi-field information fusion will be proposed to implement weak-signal fault features extraction, incipient-fault prediction and decision making. Thirdly, a new dynamic pattern recognition method based on the multi-space morphological embedded intelligent decision making mechanism will be proposed to implement dynamic pattern recognition of the incipient-fault. Fourthly, a new trend prediction method based on the multi-space morphological fusion evolution prediction mechanism will be proposed to solve the problems of the incipient-fault trend prediction from multi-space perspective. Finally, the ammunition feeding system incipient fault prediction system prototype based on generalized morphology will be developed. This method has important significance to insure the steady and reliable operation of large calibre ammunition feeding systems and it is of great significance for the development of the incipient-fault prediction theory.

提出基于广义形态学与多场信息融合的复杂供输弹系统早期故障预示新方法，目的是克服典型故障诊断方法的用于供输弹系统的局限性，弥补现有方法在故障预示上的不足，在广义形态学框架内解决供输弹系统早期故障微弱特征提取、多场信息融合、动态模式识别及演变趋势预测等关键问题。该方法研究思路为：提出集微弱特征提取、多场信息融合、动态模式识别及演变趋势预测等功能于一体的广义形态学新思想，形成早期故障预示的新理论和新方法；提出运用多场信息融合新方法，实现多场微弱故障特征提取与预示决策；提出多形态空间嵌入智能决策机制的动态模式识别新方法，实现早期故障动态模式识别；提出多形态空间融合演变预测机理的趋势预测新方法，解决早期故障跨时间尺度与多维空间的趋势预测问题；最后研发基于广义形态学的供输弹系统早期故障预示系统原型。该方法对保证大口径供输系统的稳定可靠工作，丰富和发展复杂系统早期故障预示理论具有重要意义。

本课题针对中大口径火炮供输弹系统的早期故障预示问题，采用理论与实验研究相结合的策略，解决实际应用中连续冲击响应信号相互叠加与干扰，强噪声恶劣环境中早期微弱故障信息难于有效提取，故障类别难于区分等问题。为保证中大口径火炮的战术技术性能的充分发挥，针对供输弹系统，对其机电各部件进行在线测试，进行故障自动诊断。为保证检测和诊断的快速性和精确性，必须发展真正高速、高精度、多参数、多功能的在线监测与早期故障诊断系统。.课题提出基于现代信号处理与多场信息融合的复杂供输弹系统早期故障预示新方法，克服典型故障诊断方法的用于供输弹系统的局限性，弥补现有方法在故障预示上的不足，在广义形态学框架内解决供输弹系统早期故障信号降噪、微弱特征提取、多场信息融合、模式识别等问题，并在此基础上进行了供输弹系统早期故障快速检测系统研发。.研究内容包括：降噪方法上主要研究了双正交样条小波的自适应分块阈值降噪、EEMD与相关系数结合的降噪、自适应广义形态滤波去噪、改进的可调品质因子小波变换降噪。特征提取方面研究了Kullback-Leibler散度（KLD）特征、小波时频图、熵特征、能量特征、同步压缩变换时频图像、纹理特征、运用堆叠降噪自动编码器智能提取特征、基于MRSVD理论的奇异差分谱。模式识别方面研究了引入了脱离率的深度置信网络、基于核的模糊C均值聚类、SOFTMAX分类器、Elman神经网络识别。多源信息融合方面研究了KPCA的特征级融合、基于灰色理论的D-S证据理论、经邓熵和Jensen-Shannon散度优化的D-S证据理论。开发了基于频响特征的供输弹系统早期故障快速检测软件。