Application Of Multi-Sensor Information Fusion In Tool Diagnosis

Cutting tool wear states not only directly affected machining accuracy and surface quality of the work products surface quality, but also affected the processing cost and production efficiency and so on. Thus, states monitoring of tool wear is a critical technology areas in the cutting processes. The information which is obtained by single-sensor is very limited. Multi-sensor information fusion can be more comprehensive and accurate information on surroundings, which makes up a single sensor for the one-sidedness of the information. Multi-sensor information fusion combines complementary or redundant information of a plurality of sensors, in order to get the consistency description of the object. According to the character of tool teal diagnosis, a new fault diagnosis method based on rough set and neural network is presented.Acoustic emission signals and current signals are very suitable and effective monitoring signals for the tool states monitoring system, because they not only have high sensitivity,but also are easy for installation. What’s more, they are associated with tool cutting states in high degree of correlation. The experiment analysis results show that, it’s effective to process acoustic emission signals with time-frequency analysis and Experience Mode Decomposition (EMD) and process current signal with wavelet packet transform decomposition. The main features include the proceeding processed results. As the additional feature of cutting speed, cutting depth, and feed rate are inclined, too.To overcome the problem of structure complexity and long training time in neural network method for fault diagnosis of tool diagnosis with multi-sensor, a new fault diagnosis method based on rough set and neural network is presented. At first, the rough set is used to choose the influencing factors input into the neutral network. Then, the genetic algorithm is used to overcome the shortcoming of the BP algorithm, such as slowness and convergence to local minimum. The model is applied into tool wear diagnose, the self-organizing map method is used to get the discrete attributes fist, then an adaptive genetic algorithm is designed for attribute reduction. Finally, the results of the attribute reduction is regard as the inputs of the neural network.The results show that, The designed model is better than the neural network in the training error and test error. What’s more, learning times that requests the same error of Neural Network is significantly lower than neutral network. The model achieves a good result and quick learning speed, proves the effectiveness and feasibility of this model.