矿井主风机将地面新鲜空气送入井下，降低井下有害气体浓度，为现场作业人员安全生产提供了重要保障。而风机一旦出现故障，致使停风，将危及井下工作人员的生命安全。因此，对矿用主风机进行故障诊断具有重要意义。

本文以煤矿主风机为研究对象，分析了风机转子不平衡等六种常见故障的振动机理，以振动信号为切入点，提出了一种基于振动数据驱动的主风机故障诊断方法，该方法提取振动信号相关特征后输入至分类器中对风机进行运行状态识别。对振动信号进行特征提取时，针对单域特征对信号特征挖掘不充分的问题，提出了一种主风机振动信号多域特征提取及降维融合的方法：提取振动信号时域特征、频域特征和IMF能量特征，得到相对全面的高维特征集，再通过基于类内、类间标准差的特征评价机制对特征进行筛选，剔除对分类无效及效果不明显的特征，筛选出俏度、重心频率等高效特征，最后采用核主成分分析将7维高效特征集降为3维融合特征集，消除了特征间的冗余。将3维融合特征分别输入至BP神经网络和深度置信网络（DBN）中建立主风机故障诊断模型。其中，针对于传统BPNN陷入局部最优解、收敛速度慢导致诊断结果准确率低，诊断时间长的问题，采用分数阶粒子算法（FPSO）优化BPNN的权值和阈值，在解决了BPNN陷入局部最优解问题的同时提高了故障诊断准确率，减少了诊断时间。在此基础上，将FPSO-BPNN和DBN故障诊断模型的输出结果采用D-S证据理论进行融合，得到最终诊断结果。仿真结果表明，最终诊断准确率高达98.63%，相比于FPSO-BPNN和DBN模型的诊断结果，准确率提高了5%左右。

在仿真模型的基础上设计并测试了矿用主风机故障诊断系统，可有效实现矿用主风机故障类型的识别，对促进智能煤矿安全生产具有良好的理论研究价值和工程实用价值。

The mine main fan sends the ground fresh air underground to reduce the concentration of harmful gas underground, which provides an important guarantee for the safe production of on-site operators. Once the fan fails, the air will stop, which will endanger the life safety of underground workers. Therefore, the fault diagnosis of mine main fan is of great significance.

Taking the main fan of coal mine as the research object, this paper analyzes the vibration mechanism of six common faults such as fan rotor imbalance. Taking the vibration signal as the starting point, a fault diagnosis method of main fan based on vibration data drive is proposed. This method extracts the relevant characteristics of vibration signal and inputs it into the classifier to identify the running state of fan. When extracting the features of vibration signals, aiming at the problem of insufficient mining of signal features by single domain features, a method of multi domain feature extraction and dimension reduction fusion of vibration signals of main fan is proposed: extract the time domain features, frequency domain features and IMF energy features of vibration signals, obtain the corresponding comprehensive high viterbilt collection, and then screen the features based on the feature evaluation mechanism of standard deviation within and between classes, eliminate the features that are invalid to the classification and have no obvious effect, and screen out the high-efficiency features such as kurtosis and center of gravity frequency. Finally, the kernel principal component analysis is used to reduce the 7-dimensional high-efficiency feature set to the 3-dimensional fusion feature set, so as to eliminate the redundancy between features. The three-dimensional fusion features are input into BPNN and deep confidence network (DBN) respectively to establish the fault diagnosis model of main fan. Among them, aiming at the problem that the traditional BPNN falls into the local optimal solution and the slow convergence speed leads to the low accuracy of diagnosis results and long diagnosis time, the fractional particle algorithm (FPSO) is used to optimize the weight and threshold of BPNN, which not only solves the problem of BPNN falling into the local optimal solution, but also improves the accuracy of fault diagnosis and reduces the diagnosis time. On this basis, the output results of FPSO-BPNN and DBN fault diagnosis models are fused by D-S evidence theory to obtain the final diagnosis results. The simulation results show that the final diagnosis accuracy is as high as 98.63%, which is about 5% higher than that of FPSO-BPNN and DBN models.

Based on the simulation model, the fault diagnosis system of mine main fan is designed and tested, which can effectively identify the fault types of mine main fan, and has good theoretical research value and engineering practical value for promoting the safety production of intelligent coal mine.