基于模糊滑模的卫星姿态控制系统故障诊断

基于模糊控制的航天器姿态控制系统设计航天器姿态控制是航天器的关键技术之一，是指对航天器的姿态进行控制，使其保持或者改变特定的方向、角度或者位置。

目前，航天器姿态控制系统越来越受到人们的关注，因为它是实现卫星定位、导航和控制的关键技术之一，并对整个航天事业发挥着重要的作用。

现今，基于模糊控制的航天器姿态控制系统在航天器领域被广泛应用，成为研究热点之一。

本文将详细介绍基于模糊控制的航天器姿态控制系统设计。

一、概述基于模糊控制的航天器姿态控制系统设计在处理非线性、复杂、不精确、不确定以及模糊的问题上相对于传统控制方法优越。

该设计方案主要采用模糊控制算法构建的控制器，通过传感器获得航天器姿态信息并输入控制器，控制器运用模糊算法来产生合适的控制量，从而实现航天器姿态控制。

二、控制系统设计本设计方案的航天器姿态控制系统由三个部分组成，分别为传感器部分、控制器部分和实际控制对象部分。

1.传感器部分该部分主要用于采集航天器的姿态信息，包括姿态角度、角速度等参数。

传感器一般分为惯性传感器和光学传感器，惯性传感器适合用于短期姿态控制，而光学传感器则适合用于长期姿态监测。

2.控制器部分该部分是本系统的核心部分，主要负责产生控制量来控制航天器姿态。

本设计方案的控制器主要采用模糊控制算法，该算法具有处理不确定因素和非线性因素的特点，能够适应实际控制对象的动态特性。

为了保证该控制器的精度和稳定性，需要对其进行模糊规则库的建立和调整。

首先，需要根据航天器的姿态参数构建模糊规则库，然后对模糊规则库进行优化和调整，以满足控制器的精度和稳定性要求。

3.实际控制对象部分该部分主要是实际控制对象，也就是要控制的航天器。

本设计方案的控制对象为六自由度的刚体，可以采用非线性动力学模型来描述。

基于上述三个部分，本设计方案的姿态控制系统可以实现航天器的姿态控制，具有精度高、响应速度快、适应性强等优点。

三、实验结果基于模糊控制的航天器姿态控制系统设计的实验结果表明，该系统具有很好的姿态控制效果。

测控技术2019年第38卷第5期・7・试验与测试基于模糊贝叶斯风险和T-S模糊模型的故障诊断路涛J梁智超2,索明亮3，(1.空军装备部外场保障局，北京100843；2.复杂航空系统仿真实验室，北京100076；3.北京航空航天大学可靠性与系统工程学院，北京100191；4.可靠性与环境工程技术重点实验室，北京100191)摘要:为解决复杂装备故障诊断中的知识获取和决策制定问题，提出一种数据驱动的故障诊断方法。

利用模糊贝叶斯风险模型以风险最小化原则挖掘数据中有价值知识,最，其中的概率分布用于T-S(Takagi-Sugeno)模糊规则提取，以分段线性化思想逼近复杂的数据知识。

在数值实验中，以C-MAPSS(Commercial Modular Aero-Propulsion System Simulation)发动机数据为研究对象，验证方法的有，结果表明本文方法适用于复杂装备的故障诊断。

知识获取方法明，本文方法的诊断准确率。

关键词：故障诊断；数据驱动；模糊贝叶斯风险;T-S模糊；航空发动机中图分类号:TP277文献标识码:A文章编号：1000-8829(2019)05-0007-06doi：10.19708/j.ckjs.2019.05.002Fault Diagnosis Based on Fuzzy Bayes Risk and T-S Fuzzy ModelLU Tao1,LIANG Zhi-chao2,SUO Ming-liang3,4(1.Field Support Bureau of Air Force Equipment Department,Beijing100843,China;2.Science and Technology on Complex Aviation Systems Simulation Laboratory,Beijing100076,China;3.School of Reliability and Systems Engineering,Beihang University,Beijing100191,China;4.Science and Technology on Reliability and Environmental Engineering Laboratory,Beijing100191,China)Abstract：A kind of data-driven fault diagnosis approach is proposed to solve the problems of knowledge acqui­sition and decision making in the diagnosis of complex ing the fuzzy Bayes risk model to mine the valuable knowledge under the raw data with the principle of risk minimization to obtain the relative optimal at­tribute subset.The generated probability distribution is used for T-S fuzzy rule extraction,and the complex data knowledge is approximated according to the idea of piecewise linearization.In the numerical experiments,two diagnostic cases were carried out to illustrate the efficiency of the proposed method by utilizing the data of C-MAPSS,and the results show that the proposed method is suitable for fault diagnosis of complex equipment.Compared with other knowledge acquisition methods,this method can achieve higher diagnostic accuracy.Key words：fault diagnosis;data-driven;fuzzy Bayes risk;T-S fuzzy;aeroengine复杂装备的诞生是工业技术快速发展的必然结果，随之而来的是对复杂装备的健康管理问题。

《电液位置伺服控制系统的模糊滑模控制方法研究》一、引言随着工业自动化技术的快速发展，电液位置伺服控制系统在各种高精度、高动态性能的机械设备中得到了广泛应用。

然而，由于系统中的非线性和不确定性因素，传统的控制方法往往难以达到理想的控制效果。

因此，研究新型的控制方法，提高电液位置伺服控制系统的性能，具有重要的理论意义和实际应用价值。

本文重点研究了模糊滑模控制在电液位置伺服控制系统中的应用，为解决该类问题提供了新的思路。

二、电液位置伺服控制系统概述电液位置伺服控制系统是一种以液压传动为基础，通过电机驱动液压泵，进而控制执行机构位置的系统。

其核心目标是实现对执行机构位置的精确控制。

由于系统中存在非线性和不确定性因素，如液压缸的摩擦力、外部负载扰动等，使得系统控制变得复杂。

传统的控制方法如PID控制、模糊控制等，虽然在一定程度上可以实现对系统的控制，但往往难以达到理想的控制效果。

三、模糊滑模控制方法研究针对电液位置伺服控制系统的特点，本文提出了一种模糊滑模控制方法。

该方法将模糊控制和滑模控制相结合，通过模糊控制器对系统的不确定性进行估计和补偿，同时利用滑模控制的快速性和鲁棒性，实现对系统的高精度控制。

1. 模糊控制器设计模糊控制器是本方法的核心部分。

通过对系统的不确定性因素进行观察和学习，模糊控制器可以自动调整其参数，以适应系统状态的变化。

在电液位置伺服控制系统中，模糊控制器通过接收系统的位置、速度等信息，利用模糊推理机制对系统的不确定性进行估计和补偿。

2. 滑模控制器设计滑模控制是一种变结构控制方法，其核心思想是根据系统状态的变化，实时调整系统的控制策略。

在电液位置伺服控制系统中，滑模控制器通过设计适当的滑模面和滑模控制律，使系统在受到外部扰动时，能够快速地回到预设的滑模面上，从而实现高精度的位置控制。

四、实验验证与分析为了验证本文提出的模糊滑模控制方法的有效性，我们进行了大量的实验。

实验结果表明，与传统的控制方法相比，模糊滑模控制方法在电液位置伺服控制系统中具有更好的控制性能。

故障诊断论文：基于聚类分析的卫星姿态控制系统故障诊断方法研究【中文摘要】应用于卫星系统的故障诊断技术是确保卫星系统正常工作必不可少的一项关键技术,它不仅让卫星系统具有智能化修复功能,而且其诊断经验可以优化卫星系统观测点的分布。

随着卫星系统越来越复杂与智能化,卫星故障诊断的能力在未来卫星技术发展中变得突出重要,并不断向更高水平发展。

现代卫星系统的复杂化,使得基于模的故障诊断方法——建模难度大、灵活性差,导致基于模型的方法很难取得较好的故障诊断结果。

由于基于聚类分析的故障诊断方法利用历史数据进行建模,不需要实际的物理结构模型,它能够克服基于模型的诊断方法的缺点。

本论文采用基于聚类分析的故障诊断方法,开发和利用卫星监控系统所采集的历史数据。

论文的研究内容主要包括以下几个方面:首先,研究基于聚类分析的数据分布特征提取方法的特点,学习聚类分析的两种算法,利用聚类算法分析拟合数据与异常数据,设计拟合数据与异常数据的差值模型,初步建立基于聚类分析的故障诊断模型。

其次,对标准的飞机控制系统模型进行数值仿真,获取该模型的健康数据和故障数据,采用基于聚类分析的诊断方法进行故障诊断,依据诊断结果来完善该故障诊断模型的建模与诊断推理步骤。

再次,结合卫星姿态控制系统的结构,对主要部件的数学模型做了详细分析,并研究飞轮摩擦力矩对姿态角控制效果的影响。

建立卫星姿态控制系统的外在干扰力与力矩的数学模型,针对系统的执行部件与测量部件,设置了不同的故障模式并建立相应的数学模型。

最后,根据卫星姿态控制系统部件的几种典型的故障模式,建立相应的simulink仿真模型,设置卫星姿态控制系统仿真参数与设定该系统的状态观测点,对这几种仿真模型进行数值仿真模拟并采集系统的健康数据与故障数据。

利用聚类分析的方法对该姿态控制系统进行故障诊断。

【英文摘要】The Fault Diagnosis Technique (FDT) applied to satellite systems is a special technique adopted to ensure the proper functions of the systems. It can not only realize the intelligent repaire, but the diagnosis experiences can also optimize the observation points of the satellite systems. With the developing tendency of more complex and intelligent of satellite systems, FDT plays a more important role and is pushed to reach a higher of the complexity of the modern satellite systems, the model-based FDT, which has a complicated modeling process and poor flexibility, can not always obtain a good result of fault diagnosis; while the cluster-analysis-based on history process data FDT (CBHD-FDT), which makes use of historical data and does not need the physical structure of systems, can overcome those short comings. Therefore, the research in this paper is using the CBHD-FDT, and exploring and applying the historical data gathered by the satellite monitering system. The main content of this research is as follows:Firstly, the characteristics of the extraction method into the data distrubuted features of CBHD-FDT are taken into deep research. Then two kindscluster analysis algorithms are studied, using which the expected fitting data and the actual abnormal data are analyzed, a variance model is built, and the initial CBHD-FDT model is , the numerical simulation is performed on a standard aircraft control system, from which both the healthy and faulty data are obtained. Further, CBHD-FDT is applied to diagnose this system, based on the results of which, the modeling process and diagnosis reasoning steps can be ’s more, combined with the actural structure of the satellite attitude controling system, the mathematical models of the main components are analyzed in details, with the influence of the flywheel friction torque on the attitude angle controling effects. The analytical models of the external disturbance forces and torques loading on the satellite attitude controling system are established. In addition, in view of specific operating and measuring components, the corresponding mathematical models are set under different faulure , based on several typical failure modes of a satellite attitude controling system, after the parameters of the simulation models and the state-observation points of the system are set, the simulation model is to be built in SIMULINK. From the running of the simulation, both healthy and faulty data of the system can be gathered, and CBHD-FDT is applied to perform the fault diagnosis of this controling system.【关键词】故障诊断控制系统仿真历史数据聚类分析【英文关键词】fault diagnosis simulation of controling system historical data cluster analysis【目录】基于聚类分析的卫星姿态控制系统故障诊断方法研究摘要4-5 ABSTRACT 5 第1章绪论 8-20 课题背景和意义8-9 故障诊断技术国内外进展 9-17 基于定性模型的故障诊断方法 10-12 基于定量模型的故障诊断方法 12-13 基于历史数据的故障诊断方法 13-14 基于历史数据的定量模型故障诊断方法 14-17 发展趋势 17-18 本文工作总结 18-20 第2章基于聚类分析的故障诊断方法 20-33 引言 20 聚类算法的选取标准 20-22 聚类算法可行性 21 聚类质量评估21-22 聚类分析的数据结构和数据类型 22-24 聚类分析的数据结构 22-23 聚类分析的数据类型 23-24 聚类算法模型建立 24-26 基于划分的方法 24-25 基于密度的方法 25两种方法的比较 25-26 建立基于聚类分析的故障诊断模型 26-32 健康状态的聚类学习 28-30 系统异常状态监测 30-32 本章小结 32-33 第3章基于聚类分析的控制系统故障诊断 33-46引言 33 数据矢量的属性对异常检测结果的影响 33-40 输入为常值时 34-36 输入为正弦波时 36-38 输入为方波时38-40 健康数据的选择对故障检测结果的影响 40-43 输入为常值 40-41 输入为方波 41-42 输入为正弦波 42-43基于聚类算法的复合故障检测 43-44 本章小结 44-46 第4章卫星姿态控制系统仿真模型的建立 46-57 引言 46 数学模型 46-55 卫星姿态控制系统的数学模型 46-53 空间环境力与力矩 53-55 控制系统的故障建模 55-56 本章小结56-57 第5章卫星控制系统的故障诊断 57-65 引言 57 卫星姿态控制系统的数值模拟 57-64 系统仿真模型 57-58 系统仿真参数 58 系统健康状态学习 58-59 故障分类与建模 59-60 基于聚类算法的故障检测 60-62 测量部件的故障检测 62-64 本章小结 64-65 结论 65-66 参考文献66-70 致谢 70。

第２１卷第３期燃气涡轮试验与研究Ｖｏｌ．２１，Ｎｏ．３２００８年８月ＧａｓＴｕｒｂｉｎｅＥｘｐｅｒｉｍｅｎｔａｎｄＲｅｓｅａｒｃｈＡｕｇ．，２００８基于模型的航天器推进系统故障诊断邵继业，徐敏强，王日新（哈尔滨工业大学航天学院，黑龙江哈尔滨１５０００１）摘要：本文设计了基于模型的推进系统的故障诊断系统。

根据推进系统的结构和行为模型，分析出各组件所处的工作状态及组件之间的连接关系，利用ＪＭＰＬ建模语言建立了推进系统各组件的定性模型。

把推进系统模型和系统场景文件输入给诊断推理引擎，可实现推进系统的实时诊断。

诊断结果表明，建立的推进系统模型是准确可靠的，开发的诊断系统能有效地找出故障组件并确定故障组件的状态。

关键词：基于模型；推进系统；故障诊断中图分类号：Ｖ４３０文献标识码：Ａ文章编号：１６７２－２６２０（２００８）０３－００４７－０３Ｍｏｄｅｌ－ｂａｓｅｄＦａｕｌｔＤｉａｇｎｏｓｉｓＳｙｓｔｅｍｆｏｒＳｐａｃｅｃｒａｆｔＰｒｏｐｕｌｓｉｏｎＳｙｓｔｅｍＳＨＡＯＪｉ－ｙｅ，ＸＵＭｉｎ－ｑｉａｎｇ，ＷＡＮＧＲｉ－ｘｉｎ（ＳｃｈｏｏｌｏｆＡｓｔｒｏｎａｕｔｉｃｓ，ＨａｒｂｉｎＩｎｓｔｉｔｕｔｅｏｆＴｅｃｈｎｏｌｏｇｙ，Ｈａｒｂｉｎ１５０００１，Ｃｈｉｎａ）Ａｂｓｔｒａｃｔ：Ｔｈｉｓｐａｐｅｒｄｅｓｉｇｎｅｄａｍｏｄｅｌ－ｂａｓｅｄｆａｕｌｔｄｉａｇｎｏｓｉｓｓｙｓｔｅｍｆｏｒｓｐａｃｅｃｒａｆｔｐｒｏｐｕｌｓｉｏｎｓｙｓｔｅｍ．Ｂａｓｅｄｏｎｔｈｅａｒｃｈｉｔｅｃｔｕｒｅａｎｄｂｅｈａｖｉｏｒｍｏｄｅｌｏｆｔｈｅｐｒｏｐｕｌｓｉｏｎｓｙｓｔｅｍ，ｔｈｉｓｐａｐｅｒａｎａｌｙｚｅｄｔｈｅｗｏｒｋｉｎｇｓｔａｔｅｓｏｆｃｏｍｐｏｎｅｎｔｓａｎｄｃｏｎｎｅｃｔｉｏｎｓｂｅｔｗｅｅｎｔｈｅｃｏｍｐｏｎｅｎｔｓ．Ｔｈｅｗｏｒｋｉｎｇｍｏｄｅｌｏｆｔｈｅｐｒｏｐｕｌｓｉｏｎｓｙｓ－ｔｅｍｗａｓｂｕｉｌｔｗｉｔｈＪＭＰＬｍｏｄｅｌｌａｎｇｕａｇｅ．Ｔｈｅｒｅａｓｏｎｉｎｇｅｎｇｉｎｅｔｏｏｋｔｈｅｓｙｓｔｅｍｍｏｄｅｌａｎｄｔｈｅｒｅａｌｔｉｍｅｓｃｅｎａｒｉｏｓａｓｉｎｐｕｔｓ，ｔｈｅｎｔｈｅｍｏｎｉｔｏｒｉｎｇａｎｄｄｉａｇｎｏｓｉｓｏｆｔｈｅｐｒｏｐｕｌｓｉｏｎｓｙｓｔｅｍｃｏｕｌｄｂｅａｃｈｉｅｖｅｄ．Ｔｈｅｒｅｓｕｌｔｓｐｒｏｖｅｄｔｈａｔｔｈｅｓｙｓｔｅｍｍｏｄｅｌｗａｓａｃｃｕｒａｔｅａｎｄｔｈｅｄｉａｇｎｏｓｉｓｓｙｓｔｅｍｄｅｖｅｌｏｐｅｄｃｏｕｌｄｉｄｅｎｔｉｆｙｔｈｅｆａｕｌｔｙｃｏｍｐｏｎｅｎｔｓａｎｄｔｈｅｍｏｄｅｏｆｔｈｅｆａｕｌｔｙｃｏｍｐｏｎｅｎｔｓｅｆｆｅｃｔｉｖｅｌｙ．Ｋｅｙｗｏｒｄｓ：ｍｏｄｅｌ－ｂａｓｅｄ；ｐｒｏｐｕｌｓｉｏｎｓｙｓｔｅｍ；ｆａｕｌｔｄｉａｇｎｏｓｉｓ１引言由于航天器推进系统非常复杂，在系统的设计、制造和安装过程中很难保证万无一失，这样，故障诊断便成为提高其可靠性的一个重要手段，国内外在这方面做了很多工作［１，２］。

《基于数据驱动的卫星姿控系统故障诊断研究》一、引言随着科技的飞速发展，卫星技术在军事、民用等领域的应用越来越广泛。

卫星姿控系统作为卫星的关键组成部分，其性能和稳定性对卫星的正常运行至关重要。

因此，卫星姿控系统的故障诊断显得尤为重要。

传统的故障诊断方法往往依赖于经验知识和专家系统，而基于数据驱动的故障诊断方法则通过分析大量数据，实现故障的自动诊断和预测。

本文将基于数据驱动的卫星姿控系统故障诊断进行研究，以期提高卫星姿控系统的可靠性和稳定性。

二、研究背景及意义随着卫星系统的日益复杂化，传统的故障诊断方法已难以满足现代卫星系统的需求。

基于数据驱动的故障诊断方法通过收集和分析卫星姿控系统的运行数据，提取有用的信息，实现对故障的快速诊断和预测。

该方法具有以下优点：1. 自动化程度高：无需人工干预，可实现故障的自动诊断和预测。

2. 准确性高：通过分析大量数据，提高故障诊断的准确性。

3. 实时性强：可实时监测卫星姿控系统的运行状态，及时发现故障并进行处理。

因此，基于数据驱动的卫星姿控系统故障诊断研究具有重要的理论意义和实际应用价值。

三、研究内容与方法1. 数据收集与预处理首先，收集卫星姿控系统的运行数据，包括姿态、速度、加速度等参数。

然后，对数据进行清洗、去噪、归一化等预处理操作，为后续的分析提供高质量的数据。

2. 特征提取与选择通过分析预处理后的数据，提取与故障相关的特征，如峰值、谷值、均值等。

利用特征选择算法，选择对故障诊断有重要影响的特征，降低数据的冗余性。

3. 故障诊断模型构建采用机器学习、深度学习等算法，构建基于数据驱动的故障诊断模型。

通过对模型的训练和优化，提高模型的诊断准确性和泛化能力。

4. 实验与结果分析利用实际卫星姿控系统的运行数据对模型进行测试，分析模型的诊断准确率、误诊率等指标。

通过对比传统方法和基于数据驱动的方法，评估基于数据驱动的故障诊断方法的效果。

四、实验结果与分析1. 实验数据与设置本实验采用某型卫星姿控系统的实际运行数据。

基于切换模糊化的滑模变结构auv姿态控制本文旨在研究基于切换模糊化的滑模变结构自主水下船（AUV）姿态控制方案。

首先，提出了基于多传感器数据和变结构控制的AUV 姿态控制建模框架；其次，分别对各个变结构系统模糊化进行了研究；最后，将多传感器数据驱动的全局反馈控制与模糊化的变结构控制结合，实现了基于切换模糊化的全局反馈滑模变结构AUV姿态控制，该技术可以实现智能的姿态控制，提升AUV的运动性能和稳定性。

近年来，滑模变结构技术已广泛应用于AUV，具有自主运动、耐环境干扰以及良好的平稳性等优势。

但是，目前AUV滑模变结构控制存在易受外界干扰、智能受限等问题，为了解决这些问题，我们提出了基于切换模糊化的滑模变结构AUV控制器。

首先，提出了基于多传感器数据和变结构控制的AUV姿态控制建模框架，给出了AUV三维姿态仿真和实际运动仿真系统，使用MATLAB / Simulink模拟和实现控制器，并实现了基于多传感器数据融合的AUV三维姿态实时估计等内容。

其次，对每个变结构系统模糊化进行了深入研究，提出了基于加权采样和切换模糊化2种方法，来处理变结构参数的模糊化过程。

最后，结合多传感器数据驱动的全局反馈控制和模糊化的变结构控制，实现了基于切换模糊化的全局反馈滑模变结构AUV姿态控制，在MATLAB中验证了其控制性能，该技术可以实现智能的姿态控制，提升AUV的运动性能和稳定性。

本文提出了基于切换模糊化的滑模变结构AUV姿态控制技术，以满足AUV系统需求。

研究表明，相比传统的控制方法，切换模糊化技术可以改善AUV系统的性能，实现智能化的姿态控制。

然而，本文仍有一些局限性，包括未考虑AUV变结构系统模糊化过程中的复杂物理约束，以及模糊化参数的精确选择问题，未来的工作将是解决这些问题。

本文探讨了基于切换模糊化的滑模变结构AUV姿态控制方案，提出了基于多传感器数据驱动和模糊化变结构控制的AUV控制器，并通过模拟验证，从而实现智能的姿态控制，提高AUV的系统稳定性和可控性。

The Method for Fault Diagnosis of Electric Actuator for a Type ofSelf-propelled Rocket Launcher Based on the Fuzzy ClusteringXu Bing, Wang Hong-li, Wang Han- Bing, Liu Fu-LiDept. of 5, Wuhan Mechanical Technology, Wuhan 430075, Chinaemail:*****************.cnKeywords: rocket launcher; fuzzy clustering; faulty diagnosis; fuzzy relationAbstract. Fuzzy theory is a forceful analytic tool to the faulty diagnosis of complicated system. Afuzzy clustering method is introduced on the basis of original fault diagnosis. The initial sort isacquired by fuzzy equivalent matrix and F-statistic, and the initial iterative matrix for the fuzzyc-means is also acquired. So it gets the most optimize class matrix and the fuzzy clustering center.The practical example shows the result calculated by fuzzy clustering is good agreement with theactual inspections, and the method can be applied to the faulty diagnosis of electric operationsystem of self-propelled rocket launcher. It can distinguish complicated faults more efficiently andaccurately, and it has a good foreground.ForewordElectric actuator applied to rocket launcher is an extremely complicated system, to the extent thatbattle effectiveness of rocket launcher will be impaired directly in case of failure with electricoperation system, which further leads to degraded technical supporting capacity of the army. Inpractice, a faulty rocket is usually disused by storing in depot, this not only prevents the army fromroutine drilling, but also substantially weakens its battle effectiveness.In this article, the theory of fuzzy clustering analysis is applied in a comprehensive manner,standard fault samples taken and data collected from practical operation for testing are used asclassified samples, fault diagnosis of electric actuator for rocket launcher is conducted withtransitive closure method based on fuzzy equivalent matrices and fuzzy C-means-basedclassification, which is proved to bring out crucial significance and remarkable military benefit inimproving battle performance as well as technical condition and level of rocket launcher.Fundamental & Methodology for Fault Diagnosis with Fuzzy ClusteringBasic Thinking.The method applied herein is that, by assuming normal operation of tested system,a set of data can be tested at first. Depending upon system functions, internal correlation existsamong these sampling points, thus the method of fuzzy clustering may be applied in analyzingsampling points to derive standard functional model for normal system. Then, fuzzy clustering oftested values is conducted during actual operation of the system; in case of system fault, clusteringcenter surely offsets standard model under normal condition, and the fault with those systemfunctions as corresponding membership can be diagnosed on the basis of analysis and calculation offuzzy distance. Obviously, this is a method of model identification in comparison with standardmodel.Clustering Method.1) Data StandardizationAssume domain of discourse {}12,,,n U x x x = as the objects to be classified, where11x a =,22x a =, ,11n n x a =,111n x b +=,122n x b +=, ,2n n x b =,1a ,2a , ,1n a =standard fault sample,1b ,2b , ,2n b =fault sample to be tested; each object consists of m indexes of characteristic forexpression of its properties,()12,,,i i i im x x x x = ()1,2,,i n = ; then, original data matrix is derived;International Power, Electronics and Materials Engineering Conference (IPEMEC 2015)and then, standardization processing is performed with original data matrix derived. Usually, standard data is compressed to interval[0,1], with range transformation: {}{}{}'111min max min ik ik i n ik ik ik i ni n x x x x x ≤≤≤≤≤≤−=− （1） 2) Rating (Construct Fuzzy Similar Matrix)Construct fuzzy similar matrix ()ij R γ=,where ij γ=similar coefficient, to describe similaritybetween samples i and j . Angle cosine is applied:22111()/(()())m mm ij ik jk ik jk k k k x x x x γ====∑∑∑ （2）3) ClusteringClustering requires fuzzy relation R to be fuzzy equivalent relation, fuzzy equivalent relationmatrix R can be derived from square root calculation,i.e. Derivation in sequence 2R ,4R , , 2k R ,until 122k k R R −=, then 2k R R =. Determination of Best Threshold λ. Best value λ is determined by F - statistics; assume c as number of classification corresponding to value λ, number of samples class j is nj , and samplesclass j are expressed as ()()()12,,,j j j njx x x , clustering center of samples class j is vector ()()()()()12,,,j j j j mx x x x = , where ()j k x =mean value of number k characteristics, namely, ()()11nj j j k ik i x x nj ==∑(1,2,,)k m = . F -statistics: ()21()()211||||()||||()nj j j njr j j i i j j nj x x r c F nj x x n r ===−−=−−∑∑∑ （3） Where ()||||j x x −=()()21()m j j k k xx =−∑=distance between ()j xand x , ()()||||j j i i x x −=distance between ()j i x of samples class j and center ()j x .F -Statistics is subjected to F-distribution with DOF of c -1,n -; numerator means distancebetween classes, and denominator means distance of samples in the same class; therefore, bigger F value suggests bigger distance and bigger difference between classes for better classification.Fuzzy C-mean Clustering Algorithm.⑴ Class C of samples can be derived from value λ corresponding to F maximum, thenclassification is transformed to corresponding initial membership matrix (0)c A before iterative algorithm.⑵ Calculate clustering center ()q iV from 11()()n m i ik k ik k i V r x r αα===∑∑ ()q i V (q =iterative times). ⑶ Calculate *ij r based on ()q i V 1*2(1)1||||||||m ij j i j k k r x V x V α−−= =−−∑ to derive new classification matrix *(1)q c A +, where α is taken to 1<α<+∞.⑷ Randomly given a small positive number ε, and assume (1)||||q q A A ε+−<, iteration closes,otherwise, assume 1q q =+ and go back to ⑵ for further iteration, and finally best classification matrix A and clustering center V are derived.Analysis on the Conclusion of Fault Diagnosis.One best classification is derived after fuzzyclustering analysis. If a given sample to be tested belongs to the same class as standard sample of a fault, indicating occurrence of such fault; however, when samples to be tested belong to the sameclass as several classes of standard fault samples, indicating similarity of samples to be tested to several classes of standard fault samples, namely, possibility of concurrence of these faults. If samples to be tested do not belong to the same class as any class of standard fault, it indicates a new class of fault. By case analysis and diagnosis, standard fault samples can be improved further to increase confidence of fault diagnosis.Fault Diagnosis of Electric Actuator for Self-propelled Rocket LauncherFor various systems of rocket launcher, and in occurrence of one phenomenon, it’s likely to be a fault by many causes; likewise, multiple fault causes are likely to result in as much fault phenomena; for example, poor contact of moving contact for manual resistance controller is one cause of the fault that could lead to a number of faults with rocket, including unstable operation, abrupt shutdown, enlargement of motor output voltage surge, and etc; therefore, relation between fault and phenomenon is intricate. To determine main cause of certain phenomenon, the extent of relation between the cause of fault and phenomenon of fault should have to be identified at first, i.e. a weighing coefficient to be assessed based upon empirical and statistic results; working process of whole fuzzy diagnosis is shown in the chart below:Chart 1 Fuzzy Diagnosis ProcessTable 1 contains original data used for fault diagnosis based on fuzzy clustering.Table 1. Original Data Standard fault sample (SFS)Sample Characteristic Distribution Type of FaultStandard fault sample 1 Standard fault sample 2 Standard fault sample 3 Standard fault sample 4 Standard fault sample 5 Standard fault sample 6 Standard fault sample 7 Sample a to be tested Sample b to be tested Sample c to be tested 4 2 2 0 0 3 3 1 1 0 0 2 4 0 0 2 2 0 0 1 3 0 4 3 3 1 10 0 2 0 0 0 0 0 2 2 0 0 00 0 0 4 0 0 0 4 0 31 0 0 32 1 13 1 00 2 1 1 1 2 00 2 02 0 03 3 1 13 0 00 2 4 0 0 3 30 0 10 3 6 0 0 3 30 0 2 Controller faultControl panel faultPolarized relay failureIntermediary relayfailure Variable resistor short-circuit Angle retainer malfunction Enlarged motor fault Type of fault TBD Type of fault TBD Type of fault TBDCollection of data on removed faults by the army Conclusion & analysis of practice experience Determinat ion of weighing coefficient Determination of one fuzzy matrix Computer-based diagnosis Interactive man-mach ine dialogue Output of diagnosis resultA =R = R λ=0.9559=R =By standardization and calculation of similarity coefficient from range transformation and anglecosine of original data contained in Table 1, fuzzy matrix and fuzzy equivalent matrix are derived, based on which F-statistic value corresponding to best λ value is derived.Fuzzy matrix:1.0000 0.6719 0.5599 0.6396 0.0942 0.4139 0.5449 0.4461 0.7242 0.66300.6719 1.0000 0.5836 0.5252 0.0870 0.1457 0.5754 0.1293 0.9813 0.99820.5599 0.5836 1.0000 0.2020 0.4016 0.5603 0.4426 0.6466 0.5490 0.59340.6396 0.5252 0.2020 1.0000 0.0000 0.2774 0.3651 0.2462 0.6794 0.51830.0942 0.0870 0.4016 0.0000 1.0000 0.7351 0.1613 0.6525 0.0750 0.11450.4139 0.1457 0.5603 0.2774 0.7351 1.0000 0.4557 0.9559 0.1884 0.14380.5449 0.5754 0.4426 0.3651 0.1613 0.4557 1.0000 0.3596 0.5788 0.56780.4461 0.1293 0.6466 0.2462 0.6525 0.9559 0.3596 1.0000 0.1672 0.12760.7242 0.9813 0.5490 0.6794 0.0750 0.1884 0.5788 0.1672 1.0000 0.97810.6630 0.9982 0.5934 0.5183 0.1145 0.1438 0.5678 0.1276 0.9781 1.0000 Fuzzy equivalent matrix:1.0000 0.7242 0.5934 0.6794 0.5934 0.5934 0.5788 0.5934 0.7242 0.72420.7242 1.0000 0.5934 0.6794 0.5934 0.5934 0.5788 0.5934 0.9813 0.99820.5934 0.5934 1.0000 0.5934 0.6466 0.6466 0.5788 0.6466 0.5934 0.59340.6794 0.6794 0.5934 1.0000 0.5599 0.5603 0.5788 0.5934 0.6794 0.67940.5934 0.5934 0.6466 0.5599 1.0000 0.7351 0.5754 0.7351 0.5934 0.59340.5934 0.5934 0.6466 0.5603 0.7351 1.0000 0.5754 0.9559 0.5934 0.59340.5788 0.5788 0.5788 0.5788 0.5754 0.5754 1.0000 0.5788 0.5788 0.57880.5934 0.5934 0.6466 0.5934 0.7351 0.9559 0.5788 1.0000 0.5934 0.59340.7242 0.9813 0.5934 0.6794 0.5934 0.5934 0.5788 0.5934 1.0000 0.98130.7242 0.9982 0.5934 0.6794 0.5934 0.5934 0.5788 0.5934 0.9813 1.0000F statistic value and corresponding λ value:λ 0.7242 0.5934 0.6794 0.9813 0.9982 0.6466 0.7351 0.9559F 11.1157 1.4751 11.6086 32.2437 32.4937 10.9798 8.7882 34.5249Through analysis of fuzzy equivalent matrix and F statistic value, when λ=0.9559, bestclassification matrix is obtained, original data are classified into 7 types, namely, {1},{2,b,c},{3},{4},{5},{6,a},{7}; in this case, classification matrix is expressed as follows:1 0 0 0 0 0 0 0 0 00 1 0 0 0 0 0 0 1 10 0 1 0 0 0 0 0 0 00 0 0 1 0 0 0 0 0 00 0 0 0 1 0 0 0 0 00 0 0 0 0 1 0 1 0 00 0 0 0 0 0 1 0 0 00 0 0 0 0 1 0 1 0 00 1 0 0 0 0 0 0 1 10 1 0 0 0 0 0 0 1 1 By further calculation of above classification results, final classification matrix and finalclustering center are derived, as follows: Final classification matrix: 0.0841 0.1089 0.0060 0.0072 0.0090 0.1068 0.0891 0.0638 0.1048 0.11690.3832 0.2871 0.5928 0.2950 0.3694 0.3786 0.3861 0.2908 0.3905 0.0130 0.1028 0.1089 0.0659 0.2086 0.0991 0.1068 0.1089 0.2057 0.1048 0.2468 0.0374 0.0396 0.0240 0.0288 0.0360 0.1553 0.1584 0.0284 0.0381 0.0519 0.2243 0.0396 0.0240 0.1151 0.1441 0.1359 0.1386 0.0426 0.0571 0.2078 0.0467 0.2475 0.1497 0.2518 0.2252 0.0485 0.0495 0.2482 0.1429 0.32470.1215 0.1683 0.1377 0.0935 0.1171 0.0680 0.0693 0.1206 0.1619 0.0390In final clustering:V =0.0001 0.0000 0.9993 0.0342 0.0000 0.0000 0.12830.0000 0.0000 0.0001 0.0096 0.0001 1.0000 0.03000.0001 0.0000 0.0002 0.0233 0.9998 0.0000 0.07800.0001 0.0000 0.0002 0.8928 0.0000 0.0000 0.6539 0.9996 0.0000 0.0002 0.0371 0.0001 0.0000 0.1028 0.0000 1.0000 0.0000 0.0030 0.0000 0.0000 0.00700.0001 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 Diagnosis outcome based on above analysis is that, data a to be tested indicates angle retainermalfunction, while data b and c to be tested indicate control panel fault, base on which engineeringtechnician can work out method and steps for removal of such faults, so as to eliminate any faultsdiscovered in a timely manner. It’s obvious that this fuzzy clustering is easy for realization ofreal-time and intelligent fault diagnosis. Case examination suggests consistent theoreticalcalculation and on-site inspection. This is the evidence that fuzzy clustering is potential for practicalapplication by providing a desirable approach in fault diagnosis of complex system.ConclusionIn this discourse, a new method for fault diagnosis of electric actuator for self-propelled rocketlauncher is expounded, which, based on fuzzy clustering, helps realizing application of uncertain orinsufficient knowledge of human in subjective and practical experiences, so as to succeed in themost accurate classification determination for types of fault thereon. During rocket launcher drillingof the army, type of fault can be determined accurately and quickly with this method in case of fault,and the operator can remove such fault quickly by adopting appropriate method in terms of type offault, which greatly improves technical supporting capacity of rocket launcher and generatesremarkable economic and military benefits.References[1] Brandt M,Kharas Y .An error convergence simulation study of hardvs.fuzzy C-means clustering. In:IEEEInter-NationalConference on Fuzzy System.Orlando:The IEEE Neural NetworksCouncil and the IEEE Orlando Section,1994.1835-1839[2] Sam J,Chan Laiwan.Alternative membership function for sequential fuzzy clustering.In:IEEEInternational Con-Ferenc on Fuzzy System.Orlando:The IEEE Neural Networks Council and theIEEE Orlando Section,1994.1846-1851[3] Wang Shuhai et al, Mechanism of Rocket Launcher [M]; Beijing: Chinese People's LiberationArmy Publishing House, 2000[4] Xu Zhangsui, Fang Liqing, Wang Xiwu et al, Fundamental & Application of Fault InformationDiagnosis, [M]; Beijing: National defense Industry Press, 2000[5] Li Hongxing, Wang Qun, Duan Qinzhi et al, Methodology & Application of Engineering FuzzyMathematics [M]; Tianjin: Tianjin Science and Technology Press, 1993[6] Wu Jinpei, Fuzzy Diagnosis Theory and Application [M]; Beijing: Science Press, 1995[7] Gao Xinbo, Fuzzy Clustering Analysis and Application [M]; Xi’an: Xi'an University ofElectronic Science and Technology Press, 2004。