26.6 Conclusion

Back

Techniques for achieving TCM with vibration-based properties were presented in this chapter. The

sensing methods that have proved to be effective for TCM are force, acceleration, and AE. The sensors

employed must comply with certain requirements such as robustness and cost-effectiveness. Sensors

must be installed as close as possible to the point of metal removal in order to avoid signal-to-noise ratio

problems. Various techniques exist to condition and process the signals in analogue and digital formats.

The aim of signal processing is to generate wear sensitive features from the vibration signals. This could

be done by time, frequency, joint time – frequency, and statistical analysis. Feature selection can be

automated with a variety of procedures, but care must be taken when using these to avoid selection of

linearly dependent data.

The selected features can be used to establish a model of tool wear. Numerous research papers have

shown that NNs should be used due to the many advantages of NN modeling. The training and testing

procedures of NNs are of utmost importance if the system is considered for industrial implementation.

Care must be taken not to overtrain the networks because they will lose their ability to generalize.

Furthermore, NNs cannot be expected to perform well if they are tested with previously unseen

machining parameters. They should also be trained with the minimum and maximum tool wear that is

26-20 Vibration and Shock Handbook

© 2005 by Taylor & Francis Group, LLC

expected. Future work should be directed towards incorporating numerical machining models into the

wear monitoring system to normalize the data with respect to machining parameters. If this can be

achieved, the amount of training data required for an effective TCMS will be reduced, which in turn will

provide a better solution to TCM for the manufacturing industry.

Навигация