Improved FP-Growth Based Fault Diagnosis Methods With Engineering Applications

The FP-growth algorithm is very effective for mining frequent itemsets. However, huge frequent pattern trees are generated due to repeat F-list searching, consuming a large amount of memory and leading to a low efficiency. In response to the demands of industrial process fault diagnosis, this thesis presents an improved FP-growth algorithm, termed as UFP algorithm (upgraded FP-growth). The main research work is as follows.1. The traditional FP-growth algorithm is improved. First, we use support function to map the support rate with each item in order to improve the operating efficiency. Second, taking advantage of the keyword filtering technology, the F-list is divided into two parts, key-item list and non-critical list, ensuring that all the association rules reveal valid information. Last, the whole database are separated into subsets according to the first frequent itemsets, constructing condition pattern trees directly which save lots of memory space.2. Regarding the process fault diagnosis problem, an UFP algorithm based approach is created. The rules of setting parameters'thresholds, the implementing steps and flow chart in the context of fault diagnosis are explicitly introduced. The proposed approach is applied to TE process for fault diagnosis. The association rules are generated before they are visualized with MFC. Consequently, the comparison experiments on process fault diagnosis using PCA, KPCA and the proposed algorithm are carried out, demonstrating the superiority of UFP.3. The UFP algorithm is applied to a nuclear power verification platform, successfully performing the diagnosis of 24 common failures and demonstrating the applicability of UFP algorithm in the process industrial integrated automation system. Moreover, the configuration interface of the nuclear power project is developed using MFDraw software rendering system. The integration of a variety of functions, such as data acquisition, transmission, pretreatment, as well as trend displaying, association rules mining and data visualization are realized, offering the convenience for the process monitoring and management.