Information Technology Reference
In-Depth Information
wafer optical model. The converter is then used to correct the pyrometer reading
value T p ( k +1) to T c ( k +1).
The neural-fuzzy network used for this purpose was the SONFIN, which has a
fuzzy rule-based network possessing neural learning ability. Compared with other
existing neural-fuzzy networks, a major characteristic of this network is that no
preassignment and design of fuzzy rules are required. The rules are constructed
automatically during the training process. Besides, SONFIN can overcome both the
difficulty of finding a number of proper rules for the fuzzy logic controllers and the
overtuned and slow convergence phenomena of backpropagation neural networks.
SONFIN can also optimally determine the consequent of fuzzy IF-THEN rules
during its structure learning phase, and it also outperforms the pure neural
networks greatly, both in learning speed and accuracy.
6.8.3 Application for Tool Wear Monitoring
In automated manufacturing systems, such as flexible manufacturing systems, one
of the most important issues is the detection of tool wear during the cutting process
to avoid poor quality in the product or even damage to the workpiece or the
machine. It will be shown that a neuro-fuzzy model, based on a prediction
technique, can be applied for monitoring tool wear in the drilling process.
The alternating direction of the cutting force leads to vibrations of the machine
structure. These vibrations will change owing to the tool wear conditions. Despite
the relatively harsh environment in the proximity of the cutting zone, the vibrations
can be measured conveniently by accelerometers at a comparably affordable price.
Neural networks have, for a long time, been used for classification of various
signals. However, because of many limitations, including the slow training
performance of neural networks, alternatively a neural network with fuzzy
inference has been used because of its much faster learning ability. The latter is
nothing but a neuro-fuzzy type of hybrid learning network. Using such a network a
new drill condition monitoring method is described, as proposed by Li et al.
(2000). The method is based on spectral analysis of the vibration signal. The
results are used to generate a set of indices for monitoring , utilizing the fact that
the frequency distribution of vibration changes as the tool wears. The nonlinear
relationship between the tool wear condition and these monitoring indices is
modelled using a hybrid neuro-fuzzy network. The hybrid network selected in this
case has five inputs and five outputs. The inputs to the network are the monitoring
indices based on the vibration signal of the drilling process. It is to be noted that
the mean value of each frequency band can be used to characterize the different
tool conditions. The monitoring indices selected as network inputs are summarized
in Table 6.4(a). The content of the Table 6.4(a) is read follows:
x 1 = the r.m.s value of the signal in the frequency band [0, 300] Hz.
Unlike the inputs of the network, the tool wear condition of the network was
divided into five states represented by five fuzzy membership functions (MF),
namely initial wear, normal wear, acceptable wear, severe wear and failure. Based
on the flank wear of the tool, these conditions are summarized in the Table 6.4(b).
Search WWH ::




Custom Search