Input: Difference between revisions

From Control Systems Technology Group
Jump to navigation Jump to search
No edit summary
No edit summary
 
(16 intermediate revisions by 3 users not shown)
Line 1: Line 1:
Back to main page: [[PRE2015_3_Groep4]]
Back to the main page: [[PRE2015_3_Groep4]]


To the code: [[Code]]
To the algorithm: [[Algorithm]]


To the output: [[Output]]
To the output: [[Output]]


== Brightness ==
The brightness is measured by a sensor inside the bedroom. In the simulation this will be represented by an integer light strength. The value is determined by superposing the sunlight that falls through the blinds and the light generated by our wake-up light. It is assumed there are no other light sources in the bedroom. It is also assumed that the window is always vertical and perpendicular to the suns rays.


The intensity of the sunlight is calculated based on the date and time. It is assumed the sky is clear, and there are no obstacles above the horizon.


The light coming from the wake-up light is regulated by the controller, giving an output strength, as a fraction of the full output strength( 250 lx <ref>''Philips.com, Wake-up Light HF3470/60 | Philips, consulted February 2016,'' http://www.usa.philips.com/c-p/HF3470_60/wake-up-light/ </ref>). These two values are then added to give the input brightness.
== Gathering Real-Time Data ==
Gathering Real-Time data is rather easy using the Arduino. When connected with a computer through USB, an Arduino can print values directly to a Serial Monitor. Thankfully, there is an add-on<ref>''Arduino Support for Matlab,'' http://nl.mathworks.com/hardware-support/arduino-matlab.html</ref> for Matlab, which enables Matlab to read the Serial Monitor, and also return commands to the Arduino.


== Temperature ==
The Arduino needs some code to measure the input of the sound sensor and print the values to the Serial Monitor. A figure of the code can be found at the page [[code]].
The room is to be modeled as a 3x4x2.5m box filled with air, surrounded by outside air on one side and the house all other sides. This proved to be more difficult than thought, therefor we will start with a simpler, linear model for temperature change:


<math>dT = k*(T - T_\infty)</math>
Our intent was to have as much of the program running on the Arduino itself. That means that you, in theory, would be able to have the Arduino as a standalone product. In our case, that would mean you start the program from your computer, and then would be able to unplug your Arduino since it is able to run the whole code by itself. This was however not possible because of multiple reasons. One of the reasons discussed here is saving the gathered data to read it.


With k a heat transition constant (estimated at k = -0.03), T the room temperature and T<sub>&infin;</sub> the temperature the room assumes when not regulated, 12&deg;C. When the heating is on an additional term is introduced:
For a more accurate calculation of the wake-up moment, as many measurements as able should be saved. The problem is the size of the information, and saving that information on the Arduino. The used Arduino (Arduino Mega<ref>''The Arduino Mega,'' https://www.arduino.cc/en/Main/ArduinoBoardMega2560</ref>) has up to 256 KB of storage. Not all of this can be used since a lot would already be taken up by the program itself, and 8 KB for the bootloader. If we then want to review the last 30 minutes to choose the best moment to wake up (the larger the reviewed period, the more accurate the decision), the sample time has to be reduced dramatically. Reducing the sample time results in an inaccurate measurement, leading to the program not able to choose the correct moment to wake the person. For this reason, we chose to keep the laptop in the loop for this prototype, since a laptop has more than enough storage to store samples of the whole night. In our tests, the files containing all the measurements had a size of ~130 MB.


<math>dT_{heating} = h*(T_r - T)</math>
So, the data will be stored on the computer. This will be done with the aid of the program Gobetwino <ref>''Gobetwino,'' http://mikmo.dk/gobetwino.html </ref>. This program logged the data in a .txt file. And this file can (while Gobetwino is still logging) be further processed in Matlab.


With h a convection constant (estimated at h = 0.02) and T<sub>r</sub> the radiator temperature at 50&deg;C
In the figure below, the Arduino-code is shown for gathering data. See [[Measurement plan and experiments]] for the made trade-offs in for example sample frequency.  
== Sleep model ==
The Sleep Cycle App provides graphs to the user that show his measured sleep behavior during that night. Here's an example:


[[File:sleepcycle_irregular_sleep.png]]
[[File:workingvarfreq.JPG]]
 
To model this pattern, we assumed that if the graph is split up in pieces with the peaks and valleys as cutting points, each piece will be sinusoidal with a random period and amplitude. Furthermore, we made the assumption that the period is normally distributed with mean 90 minutes or 5400 seconds (mean duration of one sleep cycle) and standard deviation 30 minutes or 1800 seconds. At first, we assumed that the amplitude is uniformly distributed, but people sleep deeper just after going to sleep and lighter just before waking up, so we had to come up with our own probability distribution. We decided to use the following probability density function:
 
<math>f(x) = rc \left( x - \frac{1}{2} \right) + 1 \qquad 0 \le x \le 1 \quad \and \quad -2 \le rc \le 2</math>
 
<math>P(0 \le X \le x) = \int\limits_{0}^{x} f(x)dx = \frac{1}{2} rc \left( x^2 + \left( \frac{2}{rc} - 1 \right) x \right)</math>
 
So X is a fraction and the probability of X = x depends on the coefficient rc. How closer rc is to 2, how closer E[X] is to 1 and how closer rc is to -2, how closer E[X] is to 0. To get a random value for x, the second function needs to be rewritten, which gives:
 
<math>x = \frac{1}{2} - \frac{1}{rc} \pm \sqrt{ \frac{2}{rc} P(0 \le X \le x) + \frac{1}{rc^2} - \frac{1}{rc} + \frac{1}{4} }</math>
 
0 ≤ x ≤ 1, so for positive rc the ± becomes a + and for negative rc a -. If rc = 0, then this is a uniform distribution. If a value for P(0 ≤ X ≤ x) is taken from a uniform distribution with range [0,1], then a random value for x can be calculated. To model de transition from generally deep sleep to light sleep, the value of rc must decrease from 2 to -2 over time. For this we used the following function:
 
<math>rc = \frac{4}{1 + e^{\frac{t - 2,5T_{gem}}{20000}}} - 2</math>
 
t is the time elapsed since going to bed and is represented by the number of ticks and Tgem is the mean duration of one sleep cycle and is in this case 90 minutes. A person needs an average of five complete sleep cycles for a good night's rest, so the point of inflection is 2.5*Tgem.
 
=== Script ===
 
globals [y
          Tgem
          start
          T
          p
          rc
          x
          R
          time
          stage
          test]
to Setup
  clear-all
  set y 100
  set Tgem (90 * 60)
  reset-ticks
end
to Go
  foreach [1 2] [
    set start y
    set T -1
    while [T < 0] [set T (random-normal Tgem 1800)]
    set p (random-float 1.0)
    set rc (4 / (1 + exp((ticks - (2.5 * Tgem)) / 20000)) - 2)
    ifelse rc > 0
      [set x (0.5 - (1 / rc) + sqrt((2.0 / rc) * p + (rc ^ -2) - (rc ^ -1) + 0.25))]
      [set x (0.5 - (1 / rc) - sqrt((2.0 / rc) * p + (rc ^ -2) - (rc ^ -1) + 0.25))]
    ifelse ? = 1
      [set R (x * start / 2)]
      [set R ((x - 1) * (100 - start) / 2)]
    set time (n-values (T / 2) [?])
    foreach time [
      set y (R * (cos (360 * ? / T)) + start - R)
      set stage ((y - (y mod 25)) / 25 - 3)
      set test (1 / (1 + exp(ticks)))
      tick
    ]
  ]
end
 
=== Result ===
Running the script gives these two graphs:
 
[[File:SlaapGrafieken.png]]
 
The first graph shows how deep you sleep on a percentage scale and the second graph translates that to the sleep stage you're in. Stage 0 means being awake and corresponds with 100%-75% on the first graph. Afterwards come stage -1, -2 and -3 which are actually stage 1, 2 and 3, the three stages of the sleep cycle. Stage 1 can also correspond with REM sleep, because they both behave the same based on the sound you make during those stages.
 
== User Settings ==


== References ==
== References ==
<references/>
<references/>

Latest revision as of 19:12, 20 April 2016

Back to the main page: PRE2015_3_Groep4

To the algorithm: Algorithm

To the output: Output


Gathering Real-Time Data

Gathering Real-Time data is rather easy using the Arduino. When connected with a computer through USB, an Arduino can print values directly to a Serial Monitor. Thankfully, there is an add-on[1] for Matlab, which enables Matlab to read the Serial Monitor, and also return commands to the Arduino.

The Arduino needs some code to measure the input of the sound sensor and print the values to the Serial Monitor. A figure of the code can be found at the page code.

Our intent was to have as much of the program running on the Arduino itself. That means that you, in theory, would be able to have the Arduino as a standalone product. In our case, that would mean you start the program from your computer, and then would be able to unplug your Arduino since it is able to run the whole code by itself. This was however not possible because of multiple reasons. One of the reasons discussed here is saving the gathered data to read it.

For a more accurate calculation of the wake-up moment, as many measurements as able should be saved. The problem is the size of the information, and saving that information on the Arduino. The used Arduino (Arduino Mega[2]) has up to 256 KB of storage. Not all of this can be used since a lot would already be taken up by the program itself, and 8 KB for the bootloader. If we then want to review the last 30 minutes to choose the best moment to wake up (the larger the reviewed period, the more accurate the decision), the sample time has to be reduced dramatically. Reducing the sample time results in an inaccurate measurement, leading to the program not able to choose the correct moment to wake the person. For this reason, we chose to keep the laptop in the loop for this prototype, since a laptop has more than enough storage to store samples of the whole night. In our tests, the files containing all the measurements had a size of ~130 MB.

So, the data will be stored on the computer. This will be done with the aid of the program Gobetwino [3]. This program logged the data in a .txt file. And this file can (while Gobetwino is still logging) be further processed in Matlab.

In the figure below, the Arduino-code is shown for gathering data. See Measurement plan and experiments for the made trade-offs in for example sample frequency.

Workingvarfreq.JPG

References

  1. ↑ Arduino Support for Matlab, http://nl.mathworks.com/hardware-support/arduino-matlab.html
  2. ↑ The Arduino Mega, https://www.arduino.cc/en/Main/ArduinoBoardMega2560
  3. ↑ Gobetwino, http://mikmo.dk/gobetwino.html
Retrieved from "https://dsdwiki.wtb.tue.nl/index.php?title=Input&oldid=26234"