Turning Tool Wear Monitoring Based On GA-BP Neural Network

Along with the continuous improvement of manufacturing industry, the importance of the mechanical fault diagnosis technology is more and more obvious. Tool is one of the most important factors in the machining process. Tool wear can affect the quality, cost and the production rate of the product. The operate-person observed tool and replaced it in the early period of machining. However, in automatic manufacturing system, the tool-failure causes invalid of machine function and the whole system's failure. Therefore, the prediction of tool wear and failure is very urgent. Due to the diversity of processing conditions, the variability of cutting and other factors, it makes the tool condition monitoring become a major link in the entire production process monitoring.The cutting force signals and vibration signals are good means to research tool wear. Cutting force signals come from the machining directly, which are highly relevant to the tool wear, and the vibration signals are easy to get. According to the characteristics of these two signals, we set up an experiment system of the turning tool wear monitoring and collect a variety of fault data using force sensor and vibratory sensor. Vibration signals were mainly analysed by their power spectrum, particularly for the high-frequency part of the signal change; force signals were mainly analysed by the Z direction, because their direction perpendicular to the workpiece rotation, and the signals were most evident. We collected these singals and took their feature extractions, and then got these feature extractions normalized, prepared for the intelligent diagnosis.In the process of the fault diagnosis, most collections can be distinguished through the diagnosis of BP Neural Network, but still some can't be distinguished, and the constringency rate of the BP network is relatively slow. Utilizing the advantage of BP Neural Network and Genetic Algorithm.Genetic Algorithm is applied in the weight training of BP Neural Network. This efficient method can overcome the lack of BP Neural Network, make greater accuracy and reduce the calculating time.Due to the practical constraints, this GA-BP Neural Network for tool wear condition monitoring system has not yet been achieved in practical applications for online real-time monitoring. In addition, the neural network fault diagnosis model-based diagnosis ability is totally dependent on the knowledge which was already existed, due to the collected examples and empirical knowledge are limited, when a new heterogeneous symptoms occur, there may not be the right symptoms to match, which will made us to get a wrong result, so this GA-BP Neural Network method still needs further study.