Image Credits: asharkyu/shutterstock.com
There is a growing need for very reliable, high speed and non-contact measurement systems which are capable of accurately measuring the roughness of surfaces. When developing machinery and other industrial equipment materials, the presence of small spaces, micro peaks and/or valleys, are indicating features of surface roughness or unevenness. These can significantly impact the performance and lifespan of these products
The Machine Vision-Based Method
Among the various non-contact measurement methods that have been developed, the machine vision-based method has seen a particular increase in popularity over the past few years. When this technique is employed, an image of the sample’s surface roughness is initially captured. A specialized software then utilizes a feature index to correlate the roughness parameters that have been extracted from the image. Once the relationship between the index and surface roughness has been established, researchers are able to determine the surface roughness characteristics of an unknown sample.
Several analytical techniques have been developed according to the machine vision-based methodology. For example, a 2013 study discussed the Gray Level co-occurrence Matrix Support Vector Machine (GLCM-SVM). This sought to improve the sensitivity and accuracy of the machine vision-based method by utilizing a microscopic imaging system and an artificial neural network (ANN) in order to detect surface roughness.
Non-Gaussian Surface Digital Simulation Technology
Machine vision-based methods provide important information on the roughness of samples. However, unfortunately they often require users to construct a functional relationship between the image feature index and surface roughness measurements of a wide range of known samples at uniform intervals. As a result of this requirement, this type of non-contact surface measurement technique can cost researchers a significant amount of time and money, as they must collect these labeled training samples and perform their measurement analysis.
To address these limitations, a group of researchers from Hunan University in China developed a dual approach that uses both non-Gaussian surface digital simulation technology and computer optical simulation technology. A virtual sample with a specified roughness was designed and constructed through a combined approach of both non-Gaussian surface digital simulation and three-dimensional (3D) entity modeling technology. Following this, the researchers captured an image of the surface of their virtual sample through image simulation and actual imaging techniques. At this point, an image feature index distribution discrepancy value was established using a similar mechanism to that which is seen in traditional machine vision-based methods.
To create this value, the researchers compared the images of both the virtual and unknown real samples of interest using a technique known as transfer kernel learning. Transfer kernel learning is a unique technique that applies the distinct patterns that have been learned in one domain (which, in this case, would be the created virtual image), to a problem present in a different, but relevant domain (such as the actual samples of similar compositions). The experimental results of this study demonstrated that this combination of novel techniques provided an accuracy of over 90% for the group’s surface roughness predictions. This accuracy level is almost identical to that which is achieved by traditional measurement methods.
Various different technology companies around the world offer non-contact surface roughness measurement devices. These are epitomized by Novacam’s metrology systems, which utilize fiber-based probes in their design. Novacam’s fiber probes have numerous advantages, including the ability to function autonomously; a precision rate which is better than 1 micrometer (µm); the capability of acquiring extensive profiles and providing accurate measurements within small-diameter tubes and bores; the capacity to measure surfaces which exhibit a wide variety of characteristics such as steps, grooves, channels, and holes; and the ability to maintain their accuracy capabilities when exposed to extreme temperatures and/or radiation.
Sources and Further Reading
- Zhang, H., Liu, J., Chen, S., & Wang, W. (2018). Novel roughness measurement for grinding surfaces using simulated data by transfer kernel learning. Applied Soft Computing 73; 508-519. DOI: 10.1016/j.asoc.2018.08.042.
- Liu, W., Tu, X., Jia, Z., Wang, W., Ma, X., & Bi, X. (2013). An improved surface roughness measurement method for micro-heterogenous texture in deep hole based on gray-level co-occurrence matrix and support vector machine. The International Journal of Advanced Manufacturing Technology 69(1-4); 583-593. DOI: 10.1007/s00170-013-5048-0.
- “Roughness” – Novacam