Hello everyone,

My name is Akshay, and I'd like to share my experiences and insights from my journey in mechanical engineering and how I combined it with computer science, specifically machine learning and deep learning. I hope that my story and the lessons I've learned can help guide students who are interested in pursuing careers in these fields.

Educational Background

I began my M.Tech in Mechanical Engineering in 2018, specializing in Computer Integrated Manufacturing. In this specialization, I learned about integrating computer science with mechanical engineering, particularly for industrial applications. Some of the key areas I studied were:

• Supply Chain Management: Understanding how goods and services flow within industries.

• Factory Automation: Learning how to automate manufacturing processes.

• Soft Computing Techniques: Exploring computational methods like neural networks and fuzzy logic.

• Basics of Mechatronics: Combining mechanical engineering with electronics and computer control systems.

Master's Project at BHEL

In 2019, during the second year of my M.Tech, I started my project at the R&D division of Bharat Heavy Electricals Limited (BHEL) in Hyderabad. During this period, I learned about machine learning and how it can be helpful for industries such as thermal power plants.

In a thermal power plant, the basic components are the turbine and generator units. The power output of the plant depends on the capacity and performance of these units. My work focused on automatically identifying the critical parameters that influence the output of the turbine and generator units, and ultimately, the power plant.

Using machine learning classifiers like Random Forest, Decision Trees, and Linear Neural Networks, I identified the most critical parameters in the thermal power plant. I also quantified how much each parameter influences the power output. This was my major research work during my master's.

I completed my M.Tech in 2020 and then decided to pursue a Ph.D.

Integrating Computer Science with Mechanical Engineering

In today's world, everything is moving towards automation. We are controlling many things through computers and electronic parts. In mechanical engineering, this means using advanced techniques from computer science to improve processes.

For example, in the manufacturing industry, there is a high need for automation because there are several processes that cannot be manually controlled efficiently. In the car industry, there are several conveyor belts. Finding the exact location where parts need to be fitted is often handled manually. However, if we replace manual labor with robots, which are ultimately controlled by humans but can operate automatically, we can reduce hazards and improve the accuracy of placing parts.

It's all about reducing manpower and using manpower efficiently. Automation also makes manual operations less harmful. For instance, in a thermal power plant, you need to regulate temperature and pressure. There is a high chance of leaks in these regions. By placing temperature sensors that can automatically control switches when there's a temperature rise, we can alert human operators promptly.

Key Subjects and Skills for Students

If you want to pursue work that combines mechanical engineering with computer science, there are certain subjects and skills you should focus on:

1. Fundamentals of Machine Learning: Understanding basic algorithms and how they can be applied to data.

2. Fundamentals of Deep Learning: Learning about neural networks and how they can model complex patterns.

3. Fundamentals of Mechatronics: Combining mechanical systems with electronics and computer control.

These foundational skills are necessary to deploy any algorithm in industries, especially in R&D.

Projects and Tools Used

In my Ph.D. journey, I have worked on many projects. Initially, I started with machine learning using RStudio, which is an open-source platform that users can easily download. Then I shifted to Python, using the Anaconda interface with tools like Spyder and Jupyter Notebook.

Project Ideas for Students:

1. Classification Tasks: This involves identifying the class of a particular object. For example, I worked on medical images to classify them as normal or abnormal. Based on image features, we can use machine learning to identify the class of the image. This is a basic task known as low-level vision.

2. Segmentation: In this task, we classify each pixel of an image into a particular class. We assign a class to each of those pixels. This is a higher-level vision task and is computationally heavier.

3. Detection: Here, we need to find out the regions of abnormality inside an image. For instance, in a chest X-ray scan, if there is a lesion, we have to identify the correct coordinates where the lesion is present.

4. Multimodal Tasks: Combining different types of data, such as language and vision models.

Students can start with classification because it is simple to understand. Then they can move to multi-label classification when an image has multiple labels. After that, they can explore segmentation and detection, which require higher computational capacity.

Interdisciplinary Research: Mechanical Engineering and Computer Science

My area of research is interdisciplinary between mechanical engineering and computer science. You might wonder what the mechanical part is in my work. I focus on patients located in remote areas, such as rural India, who have limited capacity to go to clinics and spend money on expensive scans like MRI (which costs about 8,000 INR) or X-ray (which costs about 1,000 INR). I aim to replicate these scans using biomechanical parameters.

Currently, I am working on knee osteoarthritis, which is the degeneration of the knees with age. This degeneration might occur early for some individuals and later for others. My approach is to reduce the severity of knee osteoarthritis by detecting it at an early stage.

In clinical practice, diagnosis is usually done based on clinical scans like X-rays and MRIs. My area is to replicate those biomarkers identified in knee osteoarthritis using biomechanical parameters, such as walking patterns.

When there is knee degradation, the individual may walk slightly differently than normal. My goal is to identify the forces, movements, and anthropometric or demographic parameters that influence the walking of the individual. Without any clinical scan, we can get the grade of the disease. This involves biomechanics.

Gait and Motion Analysis Lab at IIT

At IIT, we have an established Gait and Motion Analysis Lab, funded by NECBH. This setup is what I am currently using. We conduct non-invasive scans by placing sensors on different anatomical positions of the body. Mostly, they are infrared (IR) based reflective markers placed at different positions. The person needs only to walk on a mat, and cameras record the location of these markers.

Based on the spatiotemporal relationships, we can identify abnormalities associated with the walk. Additionally, we take foot scans, which are also non-invasive techniques, to support our decisions.

How It Works:

• Sensors: Placed on the body to collect data on movement.

• Force Sensors: Measure the forces coming from the ground onto the foot.

• Kinematic Analysis: Using the collected data to calculate overall body forces.

Additional Subjects for Students

To gain knowledge in this area, students need to focus on:

1. Mechanics: The basics of mechanical engineering, involving the study of forces, movements, and analysis of free-body diagrams.

2. Biomechanics: This involves applying biology to mechanics. For example, understanding the forces acting on tendons, which are living tissues, and how their properties affect these forces.

3. Robotics:

   • Kinematics: Study of motion without considering forces.

   • Dynamics: Study of motion considering forces.

   • Inverse Kinematics and Dynamics: Calculating the required joint movements to achieve a desired position.

Robotics helps in calculating the forces acting on different parts of the human body by modeling it as a series of connected links.

Pursuing a Ph.D.

I appeared for the GATE exam in 2019. Initially, I didn't get admission because it was my first year of M.Tech. In the second year, I enrolled in the Ph.D. program using my GATE score and the percentage I obtained in M.Tech.

Choosing My Field and Admission Process

While working on machine learning in thermal power plants during my M.Tech, I realized that machine learning would grow significantly in the coming years due to its potential. I felt that the time had come for automation.

I observed many conferences and journals focusing on automation in fields like medical imaging and mechanical engineering. I decided to pursue more detailed education in deep learning.

Since I was interested in biomedical engineering during my undergraduate studies, I chose to research in the clinical area, where there is a need for healthcare solutions in India.

Finding a Ph.D. Guide

I started looking for colleges offering programs in biomedical engineering or mechanical engineering with a machine learning aspect. It was challenging because there are fewer faculty members specializing in machine learning within mechanical engineering.

I researched colleges, departments, and faculty expertise areas. By thoroughly studying their profiles, I began applying to colleges and emailing professors.

Ph.D. Journey and Research Work

I enrolled in the Ph.D. program in 2020, during the COVID-19 pandemic. Initially, I had to complete coursework of 24 credits. I took subjects like Biomechanics, Robotics, and Deep Learning.

After completing the coursework in about three and a half months, I went for the comprehensive exam, which is essential to validate your Ph.D. candidacy. There was a state-of-the-art seminar where I presented what I planned to do in my Ph.D., my end goals, and the objectives to achieve those goals over four years.

In each year, specific sub-objectives need to be fulfilled. In this process, I guided many B.Tech and M.Tech students. As a Ph.D. student, you often require help from B.Tech and M.Tech students.

Projects and Guidance:

• Medical Image Analysis: Guided students working on multi-label classification, detection, and segmentation tasks.

• Mechanical Engineering Projects: Assisted in creating knee braces to support knee osteoarthritis patients.

• Remote Data Collection: Developed setups for collecting plantar pressure images (foot images) from remote locations.

Publications:

• Published a paper in ICBJIP related to multi-label classification.

• Have three papers under review on segmentation, classification, and key slice extraction.

Advice for Students

From my experience, I would like to share some insights:

1. Build Foundational Skills:

   • Machine Learning and Deep Learning: Have a strong understanding of these areas.

   • Application Area Knowledge: Whether it's manufacturing, materials, or biomedical engineering, know the basics of your field.

2. Understand the Application Area:

   • The application area helps make your deep learning model deployable.

   • Understanding the essentials allows you to create new architectures tailored to specific problems.

3. Integration of Knowledge:

   • Combining foundational skills with application knowledge is crucial.

   • This integration enables you to develop solutions that can be effectively deployed in the industry.

4. Model Deployment:

   • For industry deployment, models should have fewer parameters to be less computationally intensive and provide quicker results.

Both the foundational subjects (deep learning and machine learning) and the application subjects need to be thoroughly studied.

The Future of AI and Machine Learning

I see a lot of research happening in this area in recent years. Some key developments include:

• 2012: Introduction of AlexNet, a basic architecture of deep neural networks.

• 2015: Proposal of Generative Adversarial Networks (GANs), significantly impacting synthetic image generation.

• 2016: Introduction of attention mechanisms in deep neural networks, improving model performance.

• 2022: Emergence of language vision models like ChatGPT, widely used for various tasks.

Many companies are investing in AI, and governments are boosting initiatives like autonomous vehicles. AI applications are expanding in fields like space, telecommunications, and healthcare. Research papers in every field now often include machine learning as a basic component.

Final Thoughts

Thank you very much for your time. I hope that sharing my journey and insights has been helpful. If you're interested in integrating mechanical engineering with computer science, especially in emerging fields like healthcare, I encourage you to build a strong foundation in both areas.

Remember that application knowledge is just as important as machine learning and deep learning skills. Combining these will allow you to create solutions that can be deployed in the industry and make a real difference.

Wishing you all the best in your endeavors!

Warm regards,

Akshay