The Gear Fault Diagnosis Based On Data Warehouse

Enterprises in metallurgy and petrified industry have installed the on-line and off-line monitoring systems for their significant equipments, and large databases appear incidentally. So data warehouse technique is introduced in fault diagnosis. Takes gear case as the instance, the paper uses the data mining to generate rules distinguishing the faults of gear case, and then uses these rules to diagnose unknown fault. In particular, the main contents and achievements in this thesis can be summed up into the following:1. The establishment of gear fault data warehouseThe paper studies several gear cases in drive lab of Wuhan University of Science and Technology. The signal of normal gear pair, the one has been worn seriously, the one which has circular pitch error and the one whose pinion has a ruptured tooth is acquired independently. After FFT transformation, the amplitude values of 0-0.4 multiple engage frequency, 0.4-0.5 multiple engage frequency, 0.5-1 multiple engage frequency, 1 multiple engage frequency, 2 multiple engage frequency, 3 multiple engage frequency, 4 multiple engage frequency and greater then 4 multiple engage frequency of every signal are obtained respectively. All signal is stored in SQL database, then is purified, cleaned and transferred. At last, gear fault data warehouse is established by star pattern in Microsoft SQL Server 2000. The data in data warehouse is observed in multi-angles and multi-level by slicing, dicing, rolling up, rolling down and pivoting of OLAP. So the feature of every signal can be comprehended completely.2. The establishment of diagnosis model about gear faultThe history fault data of gear case is analysed by clustering supplied by Analysis Services of Microsoft SQL Server 2000 , and the data which has the same property and fault type is classified as a group. Traversing 8 property and outcome of this cluster obtained can get a rule. Rules whose degree of support or belief doesn't meet the demand are deleted and the diagnosis model is established from the rest rules. Moreover, the paper compiles a decision tree-making program by CAMM algorithm in MATLAB in order to compare with the clustering.Regard the history fault data of gear case as specimen, at first the information gain of every decision property is calculated at current decision point and the decision property which has the greatest information gain is selected as the decision point. Then the assembly is subdivided according to the value of its decision property. Delete the groups whose degree of support for sort property classification is less then the designated value. Repeat until the decision tree generate. Delete the branch whose degree of belief is less then the designated value and lastly get the ideal decision...