Bayesian Network In Fault Diagnosis

At present, for the purpose of safety and high efficiency, the reality of the installation of real-time monitoring system to significant equipment has come into being in many sizeable enterprises, all of information about the machine are obtained by the sensors and stored effectively, then large-scale databases and data warehouses come into being. What we concern lies in what valuable we can find from these data warehouses and how to present the knowledge. Considering our concerns, the bayes network, which is one method of data mining, is introduced to the fault diagnosis fields. The problem how to find knowledge from the databases is solved.On the base of statistics, the bayes network is a method of data mining. In essence the bayes network is a directed acyclic graph presenting directly the reliance relations among many variables. It depicts the cause and effect relations by a directed acyclic graph and the chummy relations by a conditional probability distribution table among all nodes. Moreover, we can incorporate the prior knowledge into current data effectively and get a more reasonable result. Especially when the current data are scarce or hard to obtain, the advantage of the bayes network is evident.According to the properties of fault diagnosis, we construct a bayes network model which is made up of two-layer nodes. The upper layer nodes represent the fault nodes and the lower ones represent the syndrome nodes. Furthermore, each of the variables corresponding to the nodes is binary vector. On the suppose that the network structure is known, according to the fact the data are complete or not, varied leaning algorithms are adopted to adjust the conditional probability distribution table and make it more accordant to a specified machine. At the same time, a simplified inference algorithm is used to calculate the posterior probabilities of each fault, so the problem how to recognize and classify the fault data is solved very well. Finally, we test the networks model based on the experiment data. The results show that the leaning of the networks model is effective and the calculated results accord to expert knowledge very well.The reason we introduce the bayes network to fault diagnosis field lies in that the bayes network has its distinct predominance as follows. First, the bayes network has its solid theoretic foundation; second, the bayes network has its mature probability reasoning algorithms; third, the bayes network is liable to present the problems about fault diagnosis; finally, the bayes network has a powerful leaning capability.On the platform of Matlab6.1 and Window2000, a fault diagnosis software is developed by using data mining methods.