Experimental Setup

The plunge grinding process is one of the popular abrasive finishing processes employed in the industry. We have deployed the sensors on a Next Generation Precision Grinder (NGPG) which is capable of replicating the surface finishes as observed in the industry. It is fitted with two accelerometers which captures the vibration in the normal and tangential direction during machining and a power sensor which measures the spindle power during machining. The accelerometers are sampled at a frequency of 10 kHz and the power sensor is sampled at 1000 Hz. The sensor signals are streamed to the PI System for storage and further analysis. The Performance Dashboard retrieves the data from the PI System and display the different metrics for online monitoring of the machine.

Cylindrical plunge grinding machine for manufacturing precision components

Accelerometers location for measuring vibrations during machining.

Calculation of Surface Roughness

The highly dynamic and random nature of the grinding process is known to adversely affect our ability to predict the surface quality during the process. However, by extracting various features from the accelerometers and power sensor signals and using advance data analytics we have been able to predict the surface roughness values for the product. 20 features like mean, standard deviation, kurtosis, skewness, total energy, etc. were extracted from the time portraits and FFT of the sensors as depicted in tables below. Along with these the feed rate, wheel speed and work speed were also taken into consideration to build a random forest model.

Time domain features from the sensor

Frequency domain features from the sensor

The random forest model is trained on numerous experiments to predict the surface roughness value. It is observed that the trained model can predict the surface roughness with an accuracy of 90%. For more information refer to this paper.

CYBER-SECURITY ASSURANCE