基于改进慢特征分析的CSTR故障诊断方法与实验平台

Improved slow feature analysis-based CSTR fault diagnosis method and experimental platform

针对传统慢特征分析(SFA)无法充分解析连续搅拌釜式反应器(CSTR)非线性特性问题,提出一种改进的慢特征分析故障诊断方法——随机傅里叶SFA(RFSFA),并开发了相应的仿真实验平台。该方法引入随机傅里叶映射技术实现过程变量的非线性变换,进而利用慢特征分析建立非线性统计监控模型。为了避免模型随机参数的影响,应用贝叶斯推理理论构建了集成学习模型。为验证该方法的有效性,设计了一个CSTR故障模拟与算法测试实验平台,包括正常工况模拟、故障工况模拟、故障检测等多个子系统。测试结果表明,RFSFA方法具有比传统SFA方法更好的故障检测性能,所开发的实验平台易于操作,开放性好,能够很好地验证算法的有效性。

Aiming at the problem that traditional slow feature analysis(SFA) can not fully analyze the nonlinear characteristics of continuous stirred tank reactor(CSTR), an improved fault diagnosis method of slow feature analysis random Fourier SFA(RFSFA) is proposed, and the corresponding simulation experimental platform is developed. In this method, random Fourier mapping technology is introduced to realize the nonlinear transformation of process variables, and then the nonlinear statistical monitoring model is established by using slow feature analysis. In order to overcome the influence of model random parameters, an ensemble learning model is constructed by using Bayesian inference theory. In order to verify the effectiveness of this method, an experimental platform for CSTR fault simulation and algorithm testing is designed, including the subsystems of normal condition simulation, fault condition simulation, fault detection, etc. The testing results show that the proposed RFSFA method has better fault detection performance than the traditional SFA method. The developed experimental platform is easy to operate along with good openness, and can well verify the effectiveness of the algorithm.