| Summary: | 碩士 === 國立成功大學 === 機械工程學系 === 107 === Machine tools play key roles in modern manufacturing industry. The quality of machined products are largely depended on the status of machines in various aspects. As a result, appropriate condition monitoring would be essential for both quality control and life assessment. With recent advancement of computer science, artificial intelligence (AI) becomes an alternative choice for establishing diagnostic model. AI provides a decision-making system by using multiple sensor features to predict the states of machines, especially for the machines without physical model. While many research works focus on the adjustment of model parameters or trying different algorithm to improve accuracy, the domain knowledge of machine failures is rarely studied. As a result, an artificial neural network based status monitoring system which is combined with comprehensive investigation of sensor features should be developed. Moreover, for modern machine tools, high ratio of down time is attributed to tool failure. In addition, the complexity of machining operation makes development of a model which can universally applied to different operations by curve fitting difficult. Hence, tool condition assessment is taken as a scenario in this work. For achieving above addressed goal, the experimental system must be setup first. For establishing a low cost wireless communication system, a four channels data transmission module based on Arduino board and Bluetooth is developed. Meanwhile, to access cutting tool condition from physical signal, the multi-sensor environment and data acquisition system are configured. In addition, the signal processing and feature extraction schemes are also addressed. Meanwhile, to obtain the corresponding domain knowledge of tool failure, a number of machining experiments are designed and executed. Through the effort of investigating the relation between tool conditions and sensor indexes by sensor index evaluation, six indexes which are more sensitive to tool condition can be listed to initially establish a diagnostic process. Finally, a multilayer perceptron (MLP) model is adopted to carry out condition assessment, and three models trained by different input features are compared to examine the feasibility of integrating domain knowledge and AI. In the near future, with more data collected, it is expected that more sophisticated models would be developed for better predicting the tool condition. Meanwhile, this concept can be further applied to other sub-systems which are also lack of physical models for establishing status diagnostic model to enhance the manufacturing reliability.
|
|---|
