基于虚实融合数据的悬臂式掘进机截割部故障预警技术研究

Research on fault warning technology for cutting part of cantilever roadheader based on virtual and real fusion data

目前悬臂式掘进机截割部故障预警技术依赖于传统的数据采集方法,在掘进机截割部实际运行过程中存在信号获取困难、噪声较多等问题,导致掘进机截割部故障预测预警能力受到限制。针对上述问题,提出一种基于虚实融合数据的悬臂式掘进机截割部故障预警方法。对悬臂式掘进机截割部进行三维实体建模,利用机械系统动力学自动分析软件(ADAMS)获取截割部机械系统虚拟数据,构建其动力学仿真模型以获取虚拟数据,并采用余弦相似度函数表征其与真实数据的相似度,验证虚拟数据的可信度。将虚拟、真实数据分别采用贝叶斯估计与自适应互补加权融合方法进行相似关联与互补关联融合,获得虚实融合数据。针对传统自组织映射(SOM)神经网络学习效率易受学习速率的影响问题,建立了基于改进SOM神经网络的故障预警模型,引入关于时间的单调递减函数对SOM神经网络进行训练,在保证学习速率的同时,兼顾模型的稳定性。将融合数据输入基于SOM神经网络的故障预警模型以确定获胜神经元并进行权值调整,计算真实数据与获胜神经元间的距离并进行权值调整,进而实现故障预警。实验结果表明,改进SOM神经网络的平均运行效率可提高35.84%;基于虚实融合数据的悬臂式掘进机截割部故障预警方法可成功实现单一故障和复合故障的类型预测,其预测准确率达83.33%。

Currently,the fault warning technology for the cutting part of cantilever roadheader relies on traditional data collection methods.In the operation process of the cutting part of the roadheader,problems such as difficulty in obtaining signals and high noise limit the capability to predict and warn faults in the cutting part of the roadheader.In order to solve the above problems,a fault warning method for the cutting part of cantilever roadheader based on virtual and real fusion data is proposed.The method performs three-dimensional solid modeling of the cutting section of a cantilever roadheader.It uses the automatic dynamic analysis of mechanical systems(ADAMS)to obtain virtual data of the cutting section's mechanical system,constructs its dynamic simulation model to obtain virtual data.The method uses the cosine similarity function to characterize its similarity with real data to verify the credibility of the virtual data.The method uses Bayesian estimation and adaptive complementary weighted fusion methods to perform similarity association and complementary association fusion on virtual and real data,respectively,to obtain virtual and real fusion data.In response to the problem that the learning efficiency of traditional self-organizing mapping(SOM)neural networks is easily affected by the learning rate,a fault warning model based on an improved SOM neural network is established.A monotonic decreasing function about time is introduced to train the SOM neural network,ensuring both the learning rate and the stability of the model.The method inputs the fused data into the fault warning model based on SOM neural network to determine the winning neuron and adjust its weight.The method calculates the distance between the real data and the winning neuron and adjusts its weight to achieve fault warning.The experimental results show that the average operating efficiency of the improved SOM neural network can be improved by 35.84%.The fault warning method for the cutting part of a cantilever roadheader based on virtual and real fusion data can successfully predict the types of single and composite faults,with a prediction accuracy of 83.33%.