Scientists Use Machine Learning to Eliminate Error in Steel Classification

Scientist in Saarbrücken has used new machine learning techniques to develop revolutionary quality control procedures that are more accurate and objective than current standards. The researchers have published their findings in Scientific Reports, the open-access journal associated with the highly regarded scientific journal Nature.

The research was led by scientists from two different disciplines. In order for collaboration to properly take place, both computer scientists and materials researchers had to teach each other to speak the same language.

“It took a fair amount of time before the computer scientists had understood why the internal structures of a material and their representation in image form play such an important role for materials scientists,” (http://go.nature.com/2opzaUf) says Dominik Britz, a Ph.D. student in the Department of Functional Materials at Saarland University.

These internal structures were important to the research because they are very closely linked with the properties of the material. Modern steels are being supplied in wider varieties and are becoming more complex as time goes on, which is leading to error tolerances becoming even tighter. This makes it difficult for engineers developing new steels that need to meet the strict quality requirements.

Seyed Majid Azimi at the Max Planck Institute for Informatics started by explaining to Britz how the machine learning methods that he uses are able to produce more accurate results than any other image analysis process utilized by materials scientists. To achieve the results, Azimi feeds his high-performance computer with image data that had previously been manually classified by experts. This data is used to train the computer models which are then tested by comparing them against other sets of manually classified image data.

“‘Manufacturing special steels is an extremely complex process that depends on many individual factors including the chemical composition of the material, the rolling process used and the types of heat treatment that the material is subjected to. Every stage of the production process influences the internal structure of the steel,” explains Britz. (http://go.nature.com/2opzaUf)

When discussing the internal structure of a material, materials scientists refer to it as the “Microstructure”. The microstructure is composed of tiny crystallites with particular structures known as grains. These grains are highly complex and vary in a multitude of ways. The structures are then classified using by the material development team using microscopic images with specifically prepared samples.

Classifying a material involves comparing the newly taken microscope images and comparing them with reference images that show a typical geometrical microstructure. Experienced engineers are able to discern which particular steel microstructure they are looking at but this skill is developed after significant trial and error.

“But even these practiced experts will sometimes make an incorrect call, as the differences between the images are sometimes barely discernible with the naked eye. Although humans are pretty good at distinguishing small relative differences, we are not very good at recognizing absolute geometric standards,” explains Professor Frank Muecklich, who supervised the study. (http://go.nature.com/2opzaUf)

To solve this issue, the materials scientists were interested in developing a process that was less prone to human error and could be used at any level of expertise. They found that machine learning allows computers to recognize the complex patterns very quickly as well as assign the geometry of the microstructures in microscope images. By utilizing their new approach, the research team in Saarbrücken was able to determine the microstructures of low-carbon steel at an unprecedented level of accuracy.

“When using our system for microstructural classification, we achieved a level of accuracy of around 93 percent. With conventional methods, only about 50 percent of the material samples are correctly classified,” says Muecklich. (http://go.nature.com/2opzaUf)

This fascinating new system is surely going to have a monumental impact on the manufacturing industry and improve the trust in newly developed metals. While the human eye is a full-proof tool, machine learning eliminates any doubt an engineer may have about their material. Manufacturing Talk Radio is going to keep a close eye on this research so be sure to check back for more information.

Sources:

www.manufacturing.net/news/2018/02/computer-scientists-and-materials-researchers-collaborate-optimize-steel-classification

www.nature.com/articles/s41598-018-20037-5