采煤机作为与煤岩体直接接触的机械，其发生故障的概率是综采设备中最高的，故障的发生极大的影响了煤矿的开采效率，甚至对开采人员的生命安全造成了极大的威胁，因此如何保证采煤机处于长时间稳定安全的运行状态，成为了现如今急需解决的问题。本文以采煤机为研究对象，构建概率神经网络（PNN）模型对采煤机进行故障诊断，并且使用人工水母搜索算法（JS）与粗糙集理论（RS）对概率神经网络进行优化，提高模型的诊断率，并以此为基础构建采煤机故障诊断系统，主要研究内容如下：

    （1）分析采煤机常见故障机理，提出故障诊断总体方案，使用两级式故障检测系统进行设计。一级子系统用于识别故障发生的大体部位，如截割部、液压系统等；二级子系统用于识别具体故障类型，并提出故障原因及解决方法。

    （2）一级诊断子系统研究。使用人工水母搜索算法（JS）对概率神经网络模型的平滑因子进行优化，并与粒子群优化概率神经网络的模型（PSO-PNN）进行对比实验，并以此模型为基础构建一级诊断子系统。

    （3）二级诊断子系统研究。每个部位需要设置对应的二级子系统，该子系统分为主系统和副系统，主系统使用JS-PNN网络对采煤机的详细部位进行诊断，副系统将粗糙集与神经网络结合建立RS-JS-PNN网络模型，以采煤机截割部轴承为例，使用粗糙集理论对该轴承的故障数据进行处理，达到去除不完整、模糊数据的目的，并且将粗糙集与ROSETTA软件相结合，使数据的处理更加方便快捷，最后将处理后的数据作为训练集带入人工水母搜索算法优化后的概率神经网络，以完成各部位的故障诊断神经网络模型的建立，并以轴承故障数据为例进行模拟实验。

    （4）采煤机故障诊断多级诊断系统构建。采用模块化设计思想，对软件的结构框架做出分析，完成系统的功能模块划分，并展示所设计系统的部分内容，同时对采煤机故障诊断系统的各项功能进行测试。

    本文设计了两级式采煤机故障诊断系统，通过分级诊断来对采煤机的各项故障进行诊断。将所采集到的数据首先经过一级诊断子系统，判断出故障的大体部位，如截割部、牵引部等。再将对应部位的二级诊断子系统调出，将采集信号输入到二级诊断子系统中，若数据完整则二级子系统中的主系统即可诊断结果，若数据不完整、模糊则使用副系统来进行诊断，最终得出详细的故障类型。

    Shearer, as a machine in direct contact with coal and rock, has the highest probability of failure in fully mechanized mining equipment. The occurrence of failure has greatly affected the mining efficiency of coal mine, and even caused a great threat to the life safety of mining personnel. Therefore, how to ensure that shearer is in a long-term stable and safe operation state has become an urgent problem to be solved. In this paper, shearer as the research object, the construction of probabilistic neural network (PNN) model for shearer fault diagnosis, and artificial jellyfish search algorithm (JS) and rough set theory (RS) to optimize the probabilistic neural network, improve the model diagnosis rate, and on this basis to build shearer fault diagnosis system, the main research content is as follows:

    (1)Analyze the common fault mechanisms of coal mining machines, propose an overall fault diagnosis plan, and use a two-stage fault detection system for design. The first level subsystem is used to identify the general location of the fault, such as the cutting part, hydraulic system, etc; The secondary subsystem is used to identify specific fault types and propose fault causes and solutions.

    (2)Research on the primary diagnostic subsystem. The artificial jellyfish search algorithm (JS) was used to optimize the smoothing factor of the probabilistic neural network model, and compared with the particle swarm optimization probabilistic neural network model (PSO-PNN), and based on this model, a primary diagnosis subsystem was constructed.

     (3)Research on secondary diagnostic subsystem. The main system uses JS-PNN network to diagnose the detailed parts of the shearer. The second system combines the rough set with the neural network to establish the RS-JS-PNN network model. Take the shearer cutting part bearing as an example. The rough set theory is used to process the fault data of the bearing, so as to remove incomplete and fuzzy data. Moreover, the rough set is combined with ROSETTA software to make the data processing more convenient and fast. Finally, the processed data is brought into the probabilistic neural network optimized by artificial jellyfish search algorithm as a training set. To complete the establishment of the fault diagnosis neural network model of each part, and bearing fault data as an example to simulate the experiment.

    (4)Construction of a multi-level diagnostic system for coal mining machine fault diagnosis. Adopting a modular design concept, analyze the software structure framework, complete the functional module division of the system, and display some parts of the designed system. At the same time, test the various functions of the coal mining machine fault diagnosis system.

    In this paper, a two-stage shearer fault diagnosis system is designed to diagnose various faults of shearer by hierarchical diagnosis. The collected data is first passed through the first-level diagnosis subsystem to determine the general parts of the fault, such as the cutting part, the traction part, etc. Then, the secondary diagnosis subsystem of the corresponding part is called out, and the collected signals are input into the secondary diagnosis subsystem. If the data is complete, the primary system in the secondary subsystem can diagnose the results; if the data is incomplete and fuzzy, the secondary system is used for diagnosis, and the detailed fault types are finally obtained.