سیستم هوشمند برای نظارت بر وضعیت چرخ آسیاب

The neural network and fuzzy logic are used to classify the condition of the grinding wheel cutting abilities for the external cylindrical grinding process. For each measuring signal a few statistical and spectral features are calculated and used as an input for data selection and classification procedures. First, a feed forward back propagation neural network was implemented to perform feature selection task from the multiple sensor system. Next, a neural network based fuzzy logic decision system for sensor integration in grinding wheel condition monitoring is discussed.

The paper points at the design and implementation of a neural network and fuzzy logic based system combining the outputs of several sensors for grinding wheel condition monitoring. It can be assumed that in the case of grinding processes, the state of the process during a single grinding wheel life period is only a function of the changes in the wheel cutting ability. This is why the wheel condition monitoring plays a crucial role in any automated supervising system for a grinding process. A successful grinding wheel condition monitoring depends to a great extent on reliable and robust sensors used for this purpose [1], [2], [4] and [5]. In the absence of human operators, the sensors must have the ability to recognize process abnormalities and initiate corrective action. There are various signals which correlate to the condition of the process and they are the subject of different sensing and processing techniques. Each of these signals is able to provide a feature related to the phenomenon of interest although at varying reliability. So to collect the maximum amount of information about the state of a process from a number of different sensors is the best solution. To introduce such an idea to practice an intelligent sensing system embodying strategies for sensor fusion should be implemented [1], [3], [4] and [5]. In this study, a monitoring system with multiple sensors is proposed and the performance of it is experimentally evaluated. This system includes the measurements of vibration, acoustic emission and grinding forces. They generate the useful signals for the grinding wheel wear monitoring but the best configuration of the signals and signal processing methods has to be selected. It is done by a feed forward back propagation neural network. After a tuning procedure of the network it was established that the number of informative features is much smaller than the initially used set of features. The same neural network can also be applied in the decision making procedure because, at the same time, it is able to model grinding wheel wear. Besides, a neural network based fuzzy logic decision system for sensor integration in grinding wheel condition monitoring is discussed. In order to evaluate the proposed procedures, the data collected while grinding with a range of cutting parameters were used. The fresh, worn and partly worn grinding wheel was observed during the experiments. For each measuring signal a few statistical and spectral features are calculated and used as an input for data selection and classification procedures.

An efficient way of application of a neural network and fuzzy logic based systems combining the outputs of several sensors for grinding wheel condition monitoring during external cylindrical grinding process has been presented in the paper. It was confirmed that weigh pruning can be an useful tool for the FFBP neural network structure optimization in feature selection and modeling tasks. A neuro-fuzzy algorithm seems to be the only way of effective application for fuzzy logic based systems with many input variables as in the case of grinding wheel condition monitoring. However, it should be mentioned that performance of such systems can be lower than neural network based systems and their potential for knowledge extraction is limited. Future works should be concentrated on development of other artificial intelligence methods for application in flexible grinding systems. The workpiece quality parameters as the out-of-roundness, roughness and microhardness should be incorporated into such systems as measures to classify grinding wheel wear.