Signal processing and pattern recognition for surface roughness assessment in multiple sensor monitoring of robot-assisted polishing

Polishing processes have steadily evolved from largely manual operations to automated processes based on robotized systems. Sensor monitoring can be a viable solution for process control in order to achieve more accurate, robust, and reliable automated polishing operations. In this paper, an acoustic emission-, strain-, and current-based sensor-monitoring system was employed during robot-assisted polishing of steel bars for online assessment of workpiece surface roughness. Two feature extraction procedures, a conventional one based on statistics and an advanced one based on wavelet packet transform, were applied to the sensor signals detected during polishing. The extracted relevant features were utilized to construct different types of pattern feature vectors (basic and sensor fusion pattern vectors) to be fed to a neural network pattern recognition paradigm in order to make a decision on polished part surface roughness-level acceptability.