Not a member yet? Register for full benefits!

Simple Yet Highly Effective Robotic Boost to Mass Spectrometry

Like the name suggests, mass spectrometry is the science of analysing a sample, in order to ascertain the masses of its component molecules and atoms. It is a highly useful aspect of material science, as well as a highly labour intensive one.

Since samples come in all sizes, shapes and materials, it's not a field that lends itself well to easy automation. Whilst the scanning process itself is done on a tiny sample collecting that sample from the surface of the sample is a very different matter. A given sample will often need to be tested from multiple different areas, and samples collected from angles that depend greatly on the shape of the sample, so it is a task that the human hand and eye has been traditionally superior to the robot in.

That of course has changed, or this article would not exist. A proof-of-concept early prototype artificial intelligence, developed by researchers within two very different departments at Georgia Tech in the United States.

Professor Facundo Fernandez at the the School of Chemistry and Biochemistry, which typically deals with mass spectrometry as a standard analytic tool, partnered with a robotics group in the College of Computing to try and come up with a way to automate the procedure. Whilst still a long way from an automatic sample analysis tool, they have managed to prove it can be done, by essentially partnering a learning neural network up with a 15 degrees of freedom robotic arm, and to a large part letting the AI itself work out the size and shape of the sample, and the optimum way to approach it.

On left, Professor Facundo Fernandez. On right, Rachel Bennett a member of his research team using the touchscreen interface to their new creation.

On bottom: The creation itself, a sampling proof of concept self-teaching AI/robot arm combination that is capable of analying a sample's size and shape, then determinging the best way to approach an area for sample collection before directing the arm to do just that.
Credit: Rob Felt

"You see the object on a monitor and then you can point and click and take a sample from a particular spot and the robot will go there," said professor Fernandez. "We're using an acupuncture needle that will touch very carefully on the surface of the object and then the robot will turn around and put the material inside of a high resolution mass spectrometer."

"Other people have used an acupuncture needle to poke a sample and then put that in mass spec, but nobody has tried to do a systematic, three-dimensional surface experiment," Fernandez said. "We are trying to push the limits."

To show that the system was capable of probing a three-dimensional object, the researchers imprinted ink patterns on the surfaces of polystyrene spheres. The team then used the robotic arm to model the surfaces, probe specific regions, and see if samples collected were sufficient for mass spectrometry analysis. The researchers were able to detect inks of different colours and create a 3-D image of the object with sufficient sensitivity for their proof-of-principle setup, Fernandez said.

The lead from the College of Computing, Henrik Christensen, the executive director of the Institute for Robotics and Intelligent Machines (IRIM) at Georgia Tech had perhaps predictably, a far better idea of the sheer scope of the potential for the proof of concept system that had been created, a n idea the biochemists naturally would not have considered in their relatively simple quest to automate the sample collecting process:

"The initial findings of this study mark a significant step toward using robots for three-dimensional surface experiments on geological material. We are using the repeatability and accuracy of robots to achieve new capabilities that have numerous applications in biomedical areas such as dermatology."

He's understating the situation more than a little of course. An artificial intelligence paired with a robotic system that can analyse the shape , size, even potentially density and texture of a sample and then decide based on both its own prior experience and relevant database knowledge how to proceed with that sample would have a plethora of applications across the sciences, and perhaps most notably in medicine.

Replace 'sample' with 'patient' and give the AI a comprehensive database of human anatomy. It can then be told what is wrong with the patient's body, and become a surgical assistant, assessing where best to cut for the particular shape of that particular patient, and when the cut is made, analyse the topology inside the patient, and pick out any potential problem sites before they become problem sites.

What is being done, basically, is an attempt to have the intelligence read the 'lay of the land' and apply reasoning skill to determine the best course of action, just like a human would.

We are a long, long, long way from that eventuality of course, but this is definitely a step along that road.

Back with the biochemists, the long-term goals are much simpler. Their term goal is for a highly automated system that works in partnership with a human researcher. The researcher would place the sample within range of the robotic arm, and tell the system where they would like samples taken from. The system would then clean, prepare, and slice the sample area before depositing it into the mass spectrometer – regardless of the size or shape of the sample it would analyse it determine the best point of entry, and do the grunt work.

Ultimately a system that actually selected the areas to analyse without being told, and displayed them as recommended sample areas for the researcher to okay would be ideal, as it creates a feedback circuit in which the highly trained sensors of the AI might well pick up points of interest that a human researcher would miss, but at the same time the researcher never loses control over the process.

Creating such a relatively 'lesser' system is of course far beyond us for the moment, but every step towards it will of course be another great step on the road towards the greater goal, that of an AI that can analyse and reason visually as well as a trained human does.

In close-up we see the robotic arm approaching a sample with an accupuncture needle, to take a piece with the utmost precision from a specific area. It has been told where to take the sample; it has not been told the shape of the object or how best to approach the sample area. That, it is working out for itself.
Credit: Rob Felt



Robotic arm probes chemistry of 3-D objects by mass spectrometry

Robotic Plasma Probe Ionization Mass Spectrometry (RoPPI-MS) of Non-Planar Surfaces (Paper, Paywalled)

Staff Comments


Untitled Document .