Study On Fault Diagnosis Of Mine Fan Based On The Rough Sets And ANN

The main ventilator of coal mine, as one of the key aspects in the ventilation system, is known as the "mine heart". How to timely detect failure symptoms,accurately determine the fan’s operation condition and accurately analysis the failures have become urgent problems for fan fault diagnosis. The issue takes mine rotating axial fan as the study object and proposes the establishment of coal mine main fan fault model on the basis of rough set and LVQ neural network.Then the issue runs simulation comparison on diagnosis model with combination of Rough Sets and BP network model, which shows that the model is suitable for real-time fault diagnosis with high recognition accuracy and high efficiency.The issue also enables real-time LVQ network fault diagnosis system on the basis of communication between Kingview and Matlab of DDE.After in-depth study of the main fan failure mechanism, this issue runs analysis on the essential characteristics of the fault selection.The issue carries out discussion on the characteristics of the failure mechanism with combination of rough set faults method topic. With the reasonable choice of the fault properties, the issue selects the parallel diagnostic method where vibration signal is selected as the main characteristics signal, temperature, noise, outlet cabinet secondary voltage and current are selected as auxiliary characteristic signal. The issue also runs detailed analysis on the methods for original data collection on site.To make a small storage space of attribute reduction, reduce the amount of reduction calculation and obtain the optimal reduction set, this paper presents an improved binary reduction algorithm with discernibility on matrix attribute, and the algorithm has been verified by instance. Because rough sets can only deal with discrete objects, and the raw data collected is continuous data, this paper analyzes the application keys of several common rough set discreteing method and proposes heuristic SOM discrete model, application of SOM to discrete and at the same time calculating the importance of property conditions.The Mtalab 2010 a is adopted for simulation comparison between rough set proposed in this paper and fault system model combined with LVQ. Firstly the rough set theory is used to pre-process the raw data collected, delete redundant features, dig out the minimal set of attributes of all the fault features and reduce the input dimension of LVQ network. As for the problem that there are not prioridetermining formulae for neurons in the hidden layer of LVQ neural network, this paper proposes a method by using cross-validation algorithm to determine the number of hidden layer neurons. The minimal set of attributes and all sample of attribute sets are entered LVQ neural network for training, the comparison shows that LVQ neural network diagnostic of the minimum set of attributes is real-time,with high prediction accuracy. While the minimum set of attributes are input LVQ neural network and BP neural network for training contrast. The simulation results show that within reasonable target error, LVQ neural network is better than BP network in real-time and accurate diagnosis. Finally, through DDE technology, data sharing among processes between PC and Matlab Kingview is achieved and real-time fault diagnosis system based on neural network LVQ is established.