The Investigation Of Tool Wear Monitoring Technique In High-Speed Cutting Process Based On Artificial Intelligent

Tool condition monitoring is one of the most important techniques to be developed in the automatic cutting processes as it can help to prevent damages of machine tools and workpieces. There are two main obstacles which have not been effectively solved in this research field. First, in varying cutting conditions most of the monitoring systems are not so reliable due to the used monitoring models or strategies with poor adaptability. The second is the difficulty and expense of obtaining train data. This study has tried to eliminate these obstacles by introducing artificial intelligent technologies into the tool-state monitoring research. A tool state detection system is built based on the combinative application of fuzzy inference and self-organizing neural network. Then a real-time intelligent monitoring system is proposed. Neural network based rule extraction methods are used to obtain effective fuzzy rules. This study is aimed to find out the complicated corresponding relationship between sensor signals and tool wear state, furthermore to provide a new clue for impletation of on-line tool wear monitoring.1. For metal turning operation, a novel tool state detection system called Fuzzy-SOM-TWC is established. A group of self-organizing map (SOM) neural networks is established based on the typical cutting condition combinations. For a specific cutting condition combination, the fuzzy logic method is used to select an optimum trained SOM network. The proposed system is used to classify tool-states based on the in-time measurement of force, acoustic emission (AE), and motor current signals. A highly correct classification of the tool wear status is obtained by testing the system with a series data samples under freely selected cutting conditions. The comparison tests with supervised BP network model were carried out. The results showed that the Fuzzy-SOM-TWC kept a higher correct identification rate with shortage of train samples and had a better computational efficiency.2. Based on the analysis of the tool wear rule, a novel method is proposed for monitoring of tool wear via three-direction components of the cutting forces signals generated in high speed milling process. The monitoring system with self-learning capability can automatically identify different tool states and estimate the tool wear value. To a much great extend, the pre-designed"teaching"or"training"phase which is necessary for most other monitoring method can be avoided. Features in different time and/or frequency domain were extracted by using combinations of signal processing techniques, such as time-domain averaging, discrete wavelet transform and Hilbert spectrum analysis. Correlation analysis method was used to select correlative better features. The real-time intelligent monitoring system was built on the cycle process of linear fitting and MD calculating. A series of experiment application results on a CNC vertical milling machine tool showed that the proposed method is efficient for condition monitoring of cutting tools.3. An approach of automatic sensory feature selection methods is proposed, aiming at aiding the systematic design of condition monitoring systems for machining operations and reducing the system's development time and cost. Force, acceleration, sound and acoustic emission sensors are used to acquire signals in high speed milling process. The time domain, frequency domain and wavelet analysis technique are applied to processing the signals. A new calculation method of sensitive coefficient considering both sensitivity and stability could enhance the quality of sensory features and improve the monitoring efficacy. The results show that the suggested algorithm can be applied for an automated, self-learning monitoring system for the selection of the most sensitive sensors and signal processing methods for tool faults.4. To solve the knowledge acquisition problem of the so called EXHARMIL expert system, two different rule extraction methods respectively based on Fuzzy neural network (FNN) and Knowledge based neural network (KBNN) are proposed. The estimation process showed that the optimized combination of fuzzy rules provided a smaller estimation error in comparison to the rule sets obtained by simulated annealing (SA) method.