**6. Conclusion**

The overall goal of the research was to be able to detect an unacceptable surface produced during a machining operation and then attempt to provide additional information to the machine operator regarding the type of defect, the degree of the defect, and the possible source of the defect. In manufacturing, a defect that extends above the surface such as a ridge along the surface is usually much easier to deal with (repair) than a defect that extends below the surface such as a gouge or fiber tear-out. It is also desirable to determine if a surface defect is periodic, random-like, or localized in nature. In addition, it is also desirable to determine if the defect is stationary or non-stationary **as referenced to the surface of one specimen** (it has been shown that a wood machining operation in which tool wear occurs is technically a non-stationary process when considering multiple specimens).

As discussed previously an example of a periodic surface are the knife marks from a planer or moulder. An example of a random-like surface would be fuzzy grain. An example of a localized surface defect would be a dent or a ridge caused by a chip in the cutting tool. The difference between a stationary or a non-stationary defect is that a stationary defect would extend along the entire length of the workpiece whereas a non-stationary defect would occur only along a portion of the workpiece.

This research compared various JTFA techniques including the STFT as well as numerous discrete wavelet transforms (DWT) on their ability to detect where in time a periodicity exists on the surface of a wood or wood-based composite product. This research concluded that the Harmonic DWT or HWT worked best from an efficiency in computational time as well as its ability to detect both low frequency periodicity as well as localized defects. From the time-frequency plots it can be seen the HWT can differentiate between extreme conditions and can provide the user with comprehensive information about the type of surface that has been scanned. The difference between the periodic and non-periodic situations can be determined by setting a threshold and counting the number of data excursions above the threshold to indicate whether the signal has a periodic component or not. A single threshold crossing could indicate a scratch or other localized defect. Since only periodic components below 50 marks per inch are typically of interest, only lower frequency bins need to be monitored for periodic components. The frequency bins representing periodicities (knife marks) greater than 50 marks per inch can be grouped together and used to monitor overall roughness. A two tier fuzzy logic scheme was devised to determine if the surface profile had a periodicity or was localized and / or if the surface defect was stationary or non-stationary in nature.

Current and future work includes collecting data on the ability of the system to perform in a variety of manufacturing environments and at a variety of manufacturing speeds.
