Week 10: Proving the System Works , Data and Testing
Moving from building the project to measuring exactly how well it works through formal testing, including speed, accuracy, and stress tests.
This week was all about moving from building the project to measuring exactly how well it works. In a BEng project, it isn't enough to say "it works", you have to prove it with real numbers. I spent the week running formal tests to see how A.E.G.I.S. handles different real-world situations.
1. The Testing Plan
To make sure my final report is solid, I focused on three main areas:
- Accuracy: Does the system catch a fall every time?
- Speed: How long does it take from the moment someone hits the floor to the "Red Alert" appearing?
- False Alarms: Can the system tell the difference between a person falling and someone just dropping a bag or sitting down quickly?
I performed 100 separate tests to get a clear picture of the system's performance.
2. The Results: Wi-Fi vs. The Full System
I compared using just the Wi-Fi sensor against using my "Bimodal" (Wi-Fi + Radar) system. The results showed exactly why the dual-sensor approach is better.
| Measurement | Wi-Fi Sensor Only | A.E.G.I.S. (Dual Sensors) |
|---|---|---|
| Fall Detection Rate | 92% | 96% |
| False Alarms | 18% | 2% |
| Alert Speed | 380ms | 425ms |
The Takeaway: Adding the radar makes the system slightly slower (by only 45 milliseconds), but it makes it much more accurate. It almost completely stopped the false alarms that happened when I only used Wi-Fi.
3. The 8-Hour Stress Test
A safety system needs to stay on 24/7. To test this, I ran the Raspberry Pi 5 for 8 hours straight while it processed data.
- Heat: The CPU temperature stayed around 56°C, which is very safe.
- Stability: The software didn't crash or slow down, even after processing thousands of signals. This proves the system is reliable enough for a real home.
4. Smoothing Out the Data
During testing, I noticed the Wi-Fi numbers would sometimes "flicker" because signals were bouncing off the walls. To fix this, I added a Moving Average to the code.
Instead of the system reacting to one tiny spike in data, it now looks at the last few frames to make sure the movement is real. This stopped the dashboard from jumping around and made the "Confidence" score much more stable.
— Week 10 Data SmoothingWeek 10 Checklist
- Finished Testing: Completed 100 test runs for the final report.
- Proved the Logic: Showed that the Radar successfully stops false alarms.
- Stress Tested: Confirmed the Pi 5 can run all day without overheating.
- Cleaned the Code: Added filters to make the data more accurate.
What’s Next: Now that I have all my data, I am moving into Week 11: The Final Write-up. I’ll be spending my time finishing the last chapters of my dissertation and preparing the final hardware for the demonstration.