<p>A new approach for automated tool condition monitoring in machining by using fuzzy neural networks is proposed. The Multiple Principal Component (MPC) fuzzy neural networks are built based on three major components of soft computation, namely fuzzy logic, neural networks, and probability reasoning.</p> <p>The system architecture is a partially connected neural network with fuzzy classification at neurons and fuzzy membership grades for interconnections. Principal component analyses in multiple directions are implemented tor the feature extraction and the "maximum partition". The partitions of the learning samples are based on the distributions of the monitoring indices in the principal component directions. A fuzzy membership function is used to measure uncertainties in the sampled data and to form "soft boundaries" between the classes. A processing clement in the network is connected to others through the fuzzy membership grades and other information available. The partial connections make short training times and short routines in classifications.</p> <p>Three major issues in developing the MPC fuzzy neural networks are supervised classification, unsupervised classification and knowledge updating. The system obtains the knowledge about classifications by learning. The learning samples are obtained from cutting tests performed through a reasonable range of cutting conditions.</p> <p>Several sensors are used for monitoring feature extraction. The signals from different types of sensors at different locations insure that the most significant information about the changes in tool conditions is collected. Metal cutting mechanics are first considered for the sensor selection and the sensor allocation. The measured signals are further analyzed and the monitoring features are extracted. These indices are the inputs for the fuzzy neural networks. The tool conditions considered include sharp tool, tool breakage, and a few selected stages of tool wear. The experimental results in turning and drilling have shown good performance of the proposed monitoring system in these tests.</p> / Doctor of Philosophy (PhD)
Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/6937 |
Date | 11 1900 |
Creators | Li, Shengmu |
Contributors | Elbestawi, M.A., Mechanical Engineering |
Source Sets | McMaster University |
Detected Language | English |
Type | thesis |
Page generated in 0.0025 seconds