Abstract:
With the increasing of detection and storage data dimension in industrial process, the traditional detection methods are encountering difficulties such as slow processing speed and unconspicuous feature extraction. Therefore, it is quite necessary and important to research the fault detection technology on high-dimensional data. To this end, this paper proposes a fault detection method based on adaptive sparse representation and locality preserving projects (ASRLPP). Firstly, the sparse dictionary learning algorithm is used to construct the residual space for feature extraction, which can make the global feature of the data more obvious. Then, the locality preserving projections (LPP) algorithm is used to reduce the dimension of the data in the residual space. LPP can effectively preserve the local features of data. Finally,
T2 statistics are used to calculate the control limit for monitoring. In the process of monitoring, the adaptive updating rules are introduced to update the initial training data, which can improve the efficiency and accuracy of fault detection by dynamically updating the control limits. Additionally, it is shown via the numerical example test and tennessee-eastman(TE) process simulation that the proposed ASRLPP algorithm is superior to the LPP and sparse residual distance(SRD) algorithms and has better fault detection ability in industrial process.