Before I built the leak detector, I thought sensors just worked. You connect them, they give you data, you use that data. Simple.
That is not how it goes.
The problem
The SW-420 vibration sensor is a small, cheap component. When something vibrates near it, the output goes HIGH. When it stops, it goes LOW. On paper this is exactly what you need to detect a leaking pipe — the leak creates a constant vibration, the sensor picks it up, you trigger an alarm.
In reality, everything vibrates. A truck driving past the road outside. Someone walking nearby. Wind hitting the surface. Even just tapping the table the circuit is sitting on.
During early testing, the alarm was going off constantly. Not because of leaks. Because of everything else.
First attempt at fixing it — thresholds
My first idea was to add a threshold. Only trigger if the vibration is strong enough. Set a minimum magnitude using the ADXL345 accelerometer and ignore anything below it.
This helped a little. But trucks and heavy footsteps still crossed the threshold. The alarm still fired randomly.
The actual fix — time
The key insight was simple once I thought about it: a real leak is continuous. A truck passes in under a second. A footstep lasts 0.3 seconds. But a pipe leak hisses 24 hours a day without stopping.
So instead of asking "is the vibration strong enough right now?" I changed the question to "has the vibration been strong enough for the last 2 seconds?"
I added a timer. The system only confirms a leak after the signal stays above the threshold for 2 full continuous seconds. If the vibration drops even once during those 2 seconds, the timer resets.
False alarms dropped to zero.
The rolling average
There was still one small problem. The raw accelerometer reading fluctuates even during real vibration. It would dip briefly below the threshold for one reading, then come back up. This was causing the timer to reset unnecessarily.
The fix was a 1-second rolling average. Instead of looking at the current reading, the system looks at the average of the last 10 readings (one per 100ms). A single dip only affects the average by 10% — not enough to reset the timer.
After adding this, the countdown ran smoothly through continuous vibration with no resets.
What I learned
Real-world data is messy. The clean signal you imagine in your head when you design a system rarely exists in practice. You have to expect noise, plan for it, and build filtering into the system from the start.
The 2-second timer and the rolling average are both simple ideas. But they are the difference between a circuit that triggers randomly and one that works reliably.
Simple solutions, when applied to the right problem, work better than complex ones.

This was part of my Vibration Leak Detector mini project, completed in my first year at DIT University under Dr. Nafees Ahmed.