We both contributed equally to the completion of the project. We had multiple zoom meetings over which we brainstormed and worked on the various tasks. We worked on separate repos and wrote our code for every task to make the most of this learning experience.
The greeting string issued by the server to the client upon first connecting : ECE Senior Capstone IoT simulator
1) Add Python code to Websockets client that saves the JSON data to a text file as it comes in (message by message)
python3 -m sp_iotsim.server
python3 -m sp_iotsim.client -l log.txtWe added code to src/sp_iotsim/client.py to collect data in a text file, in this case log.txt.
python3 analyze.py log.txtWe added code to analyze.py to complete this task. Most of the tasks were done using pandas built-in functions.
| Room | Median | Variance |
|---|---|---|
| Lab1 | 21.01 | 7.41 |
| Room | Median | Variance |
|---|---|---|
| Lab1 | 5.00 | 5.05 |
5) What is the mean and variance of the time interval of the sensor readings? Please plot its probability distribution function. Does it mimic a well-known distribution for connection intervals in large systems?
| Mean | Variance |
|---|---|
| 0.968014 | 1.013983 |
Yes, it mimics the Erlang distribution which is used to predict waiting times in queuing systems. Erlang distrubution is also specific case of a Gamma distrubtion. Since the time intervals for all three rooms are the same, we are using one probability distribution function graph to represent them.
1) Implement an algorithm that detects anomalies in temperature sensor data. Print the percent of "bad" data points and determine the temperature median and variance with these bad data points discarded--the same room you did in Task 2 Question 1.
python3 algorithm.py log.txtFraction of "Bad Readings" in temperature sensor data for Lab1:
| Room | Percentage | Fraction |
|---|---|---|
| Lab1 | 1.11% | 6/540 |
The temperature median and variance with these bad data points discarded:
| Room | Median | Variance |
|---|---|---|
| Lab1 | 21.01 | 0.31 |
Not necessarily. A minute change caused by opening doors or changing the AC does not indicate a failed sensor. However, a drastic change would indicate a failed sensor (big change → sensor fails). The rate of change needs to be consistent with reality for example it cannot go from 27 celsius to 41 or from 27 celsius to -26.
Upper and Lower bounds were calculated using 68–95–99.7 rule, which would result in 95.45% of the values to lie within two standard deviations of the mean. This was used to filter the data in task 3. This was completed in algorithm.py.
Upper bounds : mean + (2 * standard deviation)
Lower bounds : mean - (2 * standard deviation)
| room | Upper Bound | Lower Bound |
|---|---|---|
| lab1 | 15.51 | 26.40 |
| class1 | 10.41 | 44.52 |
| office | 15.24 | 30.73 |
As stated before, the simulation follows the Erlang distribution used in queueing systems and alike. The simulation also takes into account that sensors, at times, fail to correctly collect information from their surroundings.
The simulation is deficient because it does not account for the time of the day and the change in temperature from opening or closing of the doors in the room. In other words, the simulation can have more realistic features to it.
3) How is the difficulty of initially using this Python websockets library as compared to a compiled language e.g. C++ websockets
Python websockets library was convenient to use since the functions were well documented and easy to find compared to C++ websockets.
4) Would it be better to have the server poll the sensors, or the sensors reach out to the server when they have data?
It is better for the server to poll the sensor to keep complicated situations to occur especially as the number of sensors increase.