Process Monitoring for New Mode with Limited Data in Multimode Processes Based on Transfer Component Analysis

In order to satisfy different production needs, working modes are often adjusted in real industrial processes, which may lead to the emergence of new working modes with a small amount of modeling data. However, most of the traditional process monitoring algorithm requires that there are sufficient data to establish the reliable models. To address the above issue, a process monitoring algorithm is proposed in this work to transfer the common information from the known mode with sufficient modeling data to the new mode with limited data in multimode processes. Firstly, a reference mode is selected by evaluating the similarity between new mode and the known modes. Then, the common information is extracted and projected into a manifold subspace by using transfer component analysis. Based on the transfer features, two statistics, i.e., T² and SPE, are defined to analyze the process status and identify the fault. The effectiveness of the proposed method is verified by Tennessee Eastman (TE) process.