为解决磁致伸缩传感器测试平台动态响应过程中系统状态数据很难包括在故障诊断策略训练数据集中，导致数据驱动故障诊断策略诊断结果可靠性低的问题，提出一种基于稳态识别的数据驱动故障诊断策略。首先，设计一种稳态识别策略,得到测试平台稳态运行过程中的系统状态数据。采用能够处理单类分类问题的支持向量数据描述（support
vector data description，SVDD）对测试平台进行稳态识别，并在SVDD训练过程中采用粒子群优化（particle swarm optimization，PSO）对SVDD的超参数进行优化。其次，采用基于PSO优化的支持向量机（support vector machine，SVM）算法，通过测试平台稳态运行过程中的状态数据进行故障诊断，提高故障诊断的正确率和诊断结果的可靠性。最后，利用Simulink中Simscape模块，建立测试平台仿真模型，对基于稳态识别的故障诊断策略进行验证。测试结果表明：
与未包含稳态识别的基于SVM的数据驱动故障诊断策略相比，不同模式下基于稳态识别的故障诊断策略正确率平均提高5%，该策略在测试平台故障诊断的有效性得到验证。
Incorporating system state data from the dynamic
response phase of a magnetostrictive sensor testing platform into training
datasets for fault diagnosis strategies is challenging, often resulting in low
reliability of data-driven diagnostic results. To
address this issue, a data-driven fault diagnostic
strategy based on steady-state identification is
proposed. First, a steady-state identification strategy
is designed to obtain system-state data generated
during the steady-state operation of the test platform.
Support vector data description (SVDD), which can handle single-class classification problems, is used for steady-state identification of the test platform. During the training
process, particle swarm optimization (PSO) is utilized to optimize the
hyperparameters of the SVDD. Second, a support vector machine (SVM) optimized
by PSO is adopted to perform fault diagnosis based on the state data collected
during the steady-state operation of the test platform
to enhance fault diagnosis accuracy and the reliability of the diagnostic
results. Finally, a simulation model of the test platform is established using
the Simscape module in Simulink to validate the proposed steady-state identification-based fault diagnostic
strategy. The simulation results demonstrate that compared with the SVM-based data-driven fault diagnosis strategy
without steady-state identification, the proposed
strategy achieves an average accuracy improvement of 5% across different modes,
confirming the effectiveness of the proposed method for fault diagnosis in the
test platform.
