Data Mining And Condition Recognition On Characteristic Data Of Quayside Container Crane

With the more and more applications of online condition monitoring & assessing system on quayside container cranes (QCCs) in the container terminals, huge numbers of condition data of QCCs are produced and kept in the databases. To be different from the data in fault diagnosis, these monitoring data are sampled when QCCs are working, which results in that the QCCs'dynamic health conditions may be hidden behind them. Whereas, the sizes of these data are commonly very huge, and they look unsystematic and immethodical, so the traditional methods can not be utilized to deal with them. Recently, the technologies of data mining are developed rapidly because of its special advantages in knowledge discovery, so it seems that it maybe a good method to find out new knowledge about QCC's dynamic health conditions by data mining and use these knowledge to do condition reorganizations, which will not only break the bottleneck of traditional methods, but also will utilize the condition data in database adequately. Therefore, it is very valuable and interesting topic for theoretical researches and engineering applications.As the main production of a scientific research project of Shanghai Education Committee (Project No,: 2004095): the researches of data mining and health forecast on monitoring data of QCCs, the paper focuses on the mechanical condition reorganizations by the monitoring data in NetCMAS (Condition Monitoring & Assessing System on Network) based on the methods and technologies of data mining. Two main aspects were developed: quantitative association rule mining on the online monitoring data was carried out to learn how to recognize and evaluate the crane's conditions and its changes; on the other hand, according to the high dimension of the characteristics vectors, the attribute reduction and neural network were utilized to condition reorganization and visualization. In the end of the paper, a simple research on forecast pattern on crane's conditions was made.The main takes of this paper are how to monitor QCC's condition features and their changes based on the huge size of monitoring data, which include the association rules mining, the dimension reduction based on the rough set, the clustering and visualizing based on the neural network. The researches make some valuable explorations on knowledge discovery in QCCs'monitoring data and help to research more deeply. The main contents of this paper are outlined as the followings:1. The Crane Monitoring and Assessing System on Network (NetCMAS) was designed to realize the remote online monitoring and assessing to QCCs in some container terminals. The applications of NetCMAS provide a data source to the researches of knowledge discovery and condition reorganization in the monitoring data by data mining technologies; as well ensure all of the research results in the paper are believable.2. The quantitative association rules mining was carried out on the online monitoring data, and the use these rules as the characteristics to make reorganizations on mechanical characteristics of cranes.According to the characteristics of monitoring data, MCA-MQAR (Mining Quantitative Associations Rules based on Modified Competitive Algorithm) is proposed; According to the disadvantages of the MCA-MQAR, a new algorithm--GA-MQAR (Mining Quantitative Association Rules based on Genetic Algorithm) is proposed. Mining rules are carried out by two above algorithms individually, and the experiment results show that the interesting rules can be found from the data by both methods. The paper prefer to the last algorithm to make the future mining after compares between the results of these two algorithms.The paper analyzes the association rules found from the real data in NetCMAS, and partitions are made on the discovered rules by the similarity as the measurement; In the first time, the association rules are proposed as the features to recognize and evaluate QCCs'conditions and the changes.3. Aiming at the problems of high dimensions of vector during the condition reorganization, the attribute reduction based on the rough set is firstly carried out, then the neural network can be used to condition reorganization and visualization;During the research of dimension reduction on high-dimensional attributions of QCCs, according to the characteristics that the information table has not decision information in NetCMAS, an new algorithm is proposed for the attribution reduction in these kind of information table; at the same time, the Wallace measurement is proposed to evaluate the influences on the clustering precision after reduction., which makes up the limits that there is not a suitable method to evaluate the reduction results. The reduction results on the data of hoist driving system and trolley driving system verified the validities of this method. The modified self-organization network-Growing Cell Structure (GCS) is proposed to cluster and visualize the monitoring information during the research of condition reorganization and visualization of QCC's conditions. The paper breaks through the bottleneck that the traditional visualization only realized in 2D, and realizes the clustering and visualizing in 3D. The researches on the data of hoist system and trolley system individually by GCS found that the results in 3D can get higher precision and better visualization result than that in 2D.4. At last, a forecast model on vibration severities of QCC's driving system is developed primarily based on the research of support vector machine. The single-step and mutil-step forecast are made on vibration severities of some QCC's driving systems by this model. The comparing results discovered that the model based on SVM can get more satisfied results both in single-step and mutil-step forecasts.