This is
Kinan ALI

While robotics is about abstract math, this project was about reliability. I designed a custom PCB based on the ATmega16 microcontroller to replace aging mechanical timers in industrial machines.

Key Engineering Challenges:

• Power Control: Implemented TRIAC-triggered control for high-voltage motors.
• PID on 8-bit: wrote a PID speed controller in C from scratch (no fancy libraries!).
• Simulation: Fully validated the logic in Proteus before manufacturing.

This project taught me that in the industrial world, code doesn't just need to work—it needs to survive voltage spikes and EMI.

1. The "Perfect Model" Trap

We are taught in school that if you know the Mass, Length, and Inertia, you can control anything. I took that to heart. I derived the Jacobian matrices and calculated the Singularities (the robot's "Bermuda Triangles" where it loses control).

Fig 1: The "Danger Zones" (Singularities) where the robot locks up. I had to map these to avoid them.

But when I ran the classical controller, reality hit. The cables pulled slightly, the friction wasn't perfectly linear, and my "perfect" model started to drift. It turns out, the real world doesn't care about your differential equations.

2. Enter the Neural Network

So I tried something different. I threw away the physics book. I programmed the robot to move randomly—literally stumbling around its workspace like a drunk toddler—to collect a dataset of "Cause and Effect" (Input/Output pairs).

Fig 2: The "Drunk Toddler" phase. Collecting random walk data to train the brain.

I fed this chaotic mess into a Neural Network. The idea was simple: instead of telling the robot how to move with math, let it learn how to move from experience.

3. The Victory

The results were humbling. The Neural Network didn't care about friction coefficients or exact link masses. It just learned. As you can see in the spiral test below, the NN (Red/Green) tracked the path way better than the classical Open Loop controller (Blue), which drifted off into space.

Lesson Learned: The NN promised to cut model uncertainty, and it delivered. It shifted my engineering challenge from "Modeling Complex Physics" to "Hyperparameter Tuning & Data Collection." Sometimes, it's better to be street smart (Data-Driven) than book smart (Model-Based).

Simulations are like Instagram: they show you the perfect life. Reality is the oscilloscope showing you 100V spikes because you chose a "lazy" diode. My goal was to build a Flyback converter to boost 10V to 24V. Why Flyback? It's the Swiss Army Knife of power electronics: versatile, isolated, and efficient.

1. The "Magic Smoke" Risk

When you switch a transistor off fast, the inductance fights back. My initial math said a standard RCD Snubber circuit would handle the voltage spikes. I built the prototype, turned it on, and...

Fig 1: The "Spaghetti" prototype. If it looks messy, it's working.

The oscilloscope showed 100V spikes threatening to fry my MOSFET. The culprit? I used a standard MUR460 diode. It was too slow! It couldn't catch the spike in time. I swapped it for a 1N4148 (a tiny, fast signal diode), and the spikes vanished instantly. Crisis averted.

2. The "Reflex" System

I didn't just want it to work; I wanted it to survive. What happens if you short the output wires? Usually, fire. I designed a protection circuit using a logic AND gate and a comparator. It acts like a nervous system reflex—the moment current spikes, it kills the transistor before the fuse even knows what happened.

Lesson Learned: In power electronics, your components don't care about your theory. Always over-spec your capacitors, and never trust a slow diode with a fast job.