How AI and ML is helping Mechanical Engineers Work Smarter!

How AI and ML is helping Mechanical Engineers in the field:

Artificial intelligence and machine learning are really ubiquitous and exciting technologies, and I really view them as another really important tool for mechanical engineers. Because, what’s the role of the mechanical engineer? We build the physical devices you interact with, whether or not it’s a car, it’s your Nest thermostat, it’s a medical device that your surgeon’s using. And what artificial intelligence is going to let mechanical engineers do is to take it to the next level to develop a better device, to better understand that physical phenomenon. And I think mechanical engineers, since we’re the physical connection: we design the products that we all interact with, we play just a central role in sort of translating that technology to make the world a better place. I think that every student should take a course in AI and machine learning. And this is certainly true across engineering.AI and machine learning is a new tool that is not going away, and it is going to help inform engineers how to do their job better.And it’s going to be something that they need to understand because many of the tools they use will be embedded with these methods and these techniques.And so everyone started to look at how we can leverage that strength for fuel cells.Professor Burak Kara and Professor Amir Farimani in mechanical engineering are looking at how to use machine learning to both improve materials design, and improve the operational control of the fuel cells and the vehicles, and improve both their performance and their durability.Their lab at CMU is focusing on using machine learning for molecular discovery, which is a pretty difficult problem because creating a functional mapping between the material geometry and topology and chemistry and basically the properties are a very difficult task.So to that end, what we do is that we train on multiple data that are generated either via simulations or via experiments. And then they made a predictive model that, if we give a molecule or material, it willits properties? So they are using deep neural networks or graph convolutional neural networks in order to be able to model this functional mapping. At Carnegie Mellon University  they are able to develop explainable, reliable, and verifiable AI products by bridging mathematics and innovation. Their current projects focus on autonomous vehicles and smart cities. In the future, the hope to translate the safe AI technologies we are developing to other fields.

 It’s imperative that we teach our next generation of students to really learn how to leverage the data that’s acquired from these sensing systems that are really at the heart of artificial intelligence. And this is exactly why mechanical engineering and artificial intelligence go hand-in-hand: because the systems that we will be creating in the 21st century are a blend of these. We want our students to be able to be leaders and at the cutting edge of technology. And so they need to understand these methods as they move forward, so that they can understand how to apply them in practice. In mechanical engineering we’re infusing concepts of artificial intelligence and machine learning across our curriculum, both our undergraduate and graduate curriculum. And we’re doing this because 21st century mechanical engineers need these additional tools in their toolkit that we need to make sure all graduates are experts in. 

Case studies of how AI and ML has helped Mechanical Engineers in their work:

It is a well-known fact that compressed air is the  most expensive utility on the shop floor. However,   non availability of compressed air just  at the right time is even more expensive.  Learn how a car company used Machine Learning  and Artificial Intelligence solutions   to predict compressor failures and saved millions.  Today's story is based in mid two thousands back  in the US. During such times at a car company,   it's compressor was failing more than once  a year without any notice. The company was  losing millions because of non availability of  compressed air, because a car was getting rolled  off it's plant every 25 seconds. The plant in  charge of this car company was a worried man since non availability of compressed air meant the  painting or finishing operations were affected. The pneumatic tools remained ineffective. Even  the breathing air filters remained offline,  endangering the health of workers on the shop  floor. Now let's understand what exactly was  the problem. This car manufacturing company had  a multi-million dollar compressed air system  consisting of 12 centrifugal air compressors,  almost of the size of a small truck.  The blades of these compressors rotated at almost  20,000 RPM, almost a fraction of an inch from it's  housing. Now, if you can imagine what will happen  if the blade touches the housing. However, the  blades wobble was not the cause of the crash. The crash or surge was the result of  a reverse flow, which happened because the normal  direction of the airflow from the compressor to  a pipe was reversed and the air inside the pipe  pushed back. 

And this is where our hero Mechanical  Engineer steps in. This engineer back in the mid two  thousands thinks about a futuristic world where  all the machines are connected with each other to  a central unit and provide data for intelligent  maintenance. He thinks that all the machines  are like patients and the data coming out of its  sensors is like blood. Hidden inside the data is  meaning about the deviant behaviour of a machine or  process unlocking this meaning leads to predicting  quality, determining the stability of a machine or  a process and predicting future failures. And this  exactly is what industry 4.0 is today. Do you know where and when the  term industry 4.0 was coined?The term "Industrie 4.0", shortened to I4. 0 or simply I4, originated in 2011 from a project in the high-tech strategy of the German government and specifically relates to that project policy, rather than a wider notion of a Fourth Industrial Revolution of 4IR.

Our mechanical engineering hero, along with  a team of his data analysts joined the plant  in-charge and some skilled maintenance workers  and got to solve this problem at  the car manufacturing company.  And they started with asking simple questions  to solve this complex issue. Questions like  Can we predict a reverse flow before it causes a  crash? Can we see something developing in the data  or time to predict this failure? The team decided  to install sensors on various locations on the  compressor and they collected data and analysed  the data for months together from a compressor  and tried to look for anomalies. Sometimes  the team would create surge like conditions  or approach surge conditions on purpose and  see how it reflected in the data set. Finally, the  team's hard work paid off and they found the root  cause that was causing this failure. From the  data, the team understood that the best predictor  of an incipient surge like conditions happened at  stage two, that is in a four stage compressor  the deviant behaviour started at stage two,  which led to eventual failure of the complete  system. Now the very fact that they found this  deviant behaviour starting earlier, it provided  ample time and warning to take corrective action.  With this data and the proof from the design of the experiment led to the company investing in good  quality sensors for all the compressors. And they  started getting data from all the compressors.  Now this data was fed into a central intelligent  unit and an algorithm was developed to predict the  failure earlier. Further tests confirm that this  phenomenon was observed across all the compressors, that is the wobble or the surge like conditions  happened at stage two of every compressor, which  led to failure of that particular compressor. So that means corrective action could be taken  on time and probably the compressor of the  deviant compressor could be bypassed. Or it  could be backed off. What was the result of all  this data gathering and analysis and development of  machine learning algorithms? The result was  that after implementation of this solution,  there was absolutely no unplanned crashing  or shut down of the compressed air system.  Now this meant better predictability and  control of the system, which played a huge part  in improving the bottom line for the company.  

Conclusion:

Now, why did I choose to tell this story?  I am not a machine learning or  artificial intelligence expert. I'm a mechanical engineer. But this story,  which is based in two thousands, it just shows  that as mechanical engineers, we should not create  boundaries in thinking. These are new technologies  that are coming up and we as mechanical engineers  need to find where we can implement or  use these tools to solve our problems. It is an unfound mistake in thinking that these  new tools like machine learning are going to  disrupt the job market. No, this is a challenge  or a new opportunity for budding new mechanical  engineers to come up with innovative solutions  to current problems.We must learn all these skills and come up as a best engineer.