基于时序溯源的嵌入式数控系统软件可靠性评估方法研究

The software reliability is an important issue for embedded Numerical Control (NC) systems.Embedded NC system software includes hardware drivers, interrupts and other modules.Therefore,the logic of the program running is complex and there are many types of G code used in Computer Numerical Control(CNC) machining. The operating path is uncertainty and the execution order of the modules is random. It’s difficult in software system assessment.This project intends to conduct research on reliability evaluation method of embedded NC software using Markov Chain Model.However, Building the operating profile and quantifying the transition probability is the difficult point of the Markov Chain Model. The time-series traceability method was put forward to rebuild program timing diagram, to build an operating profile and quantify the transition probability. The Time compensation method was given to eliminate the system running real-time effects by the monitoring and data output of the program key point, and restore the CPU's logical sequential relationship on the time axis.Aiming at the monitoring dataset in a interpolation cycle, the data clustering analysis method was studied to automatic identify the fault, and scale adaptive wavelet transform method was used to detect the singularity of motor driving signal.Through the research of the project, the key problem of the operating profile and transfer matrix in the Markov chain model was solved, as well as big data automatic identification.The basic theoretical and key technical support were provided for embedded NC software reliability assessment, testing, software development and debugging in this research.

嵌入式数控软件的可靠性是一个重要问题，由于其包含电机驱动、插补中断等诸多模块，程序运行逻辑复杂，且数控加工G代码种类繁多，程序运行路径不确定性大，软件各模块的执行顺序具有随机性，软件系统可靠性评估困难。拟采用马氏链模型开展嵌入式数控软件的可靠性评估方法研究，而马氏链模型的关键在于运行剖面的确定及转移概率的量化。提出时序溯源的方法，重溯程序运行时序图，构建运行剖面，量化转移概率；采用时间补偿的程序运行逻辑恢复方法，消除关键点数据的监测、输出对系统运行实时性影响，还原CPU在时间轴上的运行逻辑顺序；针对时间轴上以插补为周期的监测数据集，研究数据聚类分析方法自动识辨故障，及尺度自适应小波变换方法检测电机驱动信号奇异点。通过该项目研究，以期解决马氏链模型中运行剖面确定、大数据的故障自动识别等关键问题。本项目的研究将为嵌入式数控系统软件可靠性评估、测试、软件开发、调试等提供基础理论与关键技术支撑。

嵌入式系统因体积小，便携性好，成本低，可扩展性高，被广泛运用在航空航天控制、汽车电子、工业控制等领域；但嵌入式软件的设计自主性比较大，稳定性得不到保障。本项目以多轴联动嵌入式数控系统为对象来开展嵌入式软件可靠性评估方法研究，有着理论和现实意义。.首先，针对嵌入式数控系统软件模块化逻辑分析的复杂性和困难程度，提出了基于改进littlewood马尔科夫链的嵌入式数控系统可靠性预计方法。改进的模型与传统的littlewood马尔科夫链模型相比，在工程实际中应用更方便，系数矩阵更容易确定；该模型更适合基于嵌入式数控系统软件模块化可靠性的分析与研究。.然后，采用尺度自适应的小波变换技术，采用距离最大准则确定了尺度自适应，利用遗传算法计算出最佳小波尺度，使电机驱动波在最佳小波尺度下分解，准确定位故障位置；对电机驱动波信号进行检测，并进行故障诊断和预测分析。.之后，为了进一步兼顾电机运行时的高效平稳，提出了动态前瞻后顾规划算法及一套相对完整的加减速规划流程。并根据数控系统软件模块化的特点，提出、设立了关键数据监测点，通过“插桩方式”得到了相应的参数值，采集各关键点信息，并通过串口输出获得了系统运行时间点，得到数控系统软件运行的故障数据信息，统计了数控系统各功能模块的故障率，以及相邻模块的转移矩阵。.接着，提出了一种CNN-RF集成学习的故障诊断方法，该方法采用Stacking集成策略将卷积神经网络（CNN）和随机森林（RF）融合，不仅能准确提取数据集中的数据特征，而且能针对数据集中故障数据数量不足的问题使误差量均衡化，以提高诊断的准确性。.最后，得到调整后的时序图后，根据重溯的时序图谱，建立插补程序运行剖面，统计模块相互之间的转移概率；根据监测的大数据及故障识别结果，计算模块可靠度及模块转接失效率，构建运行剖面和转移矩阵，建立基于运行剖面的Littewood markov链模型。最后通过建立的Littewood markov链模型计算得到嵌入式数控系统的整体失效率。