I am currently doing a school project where i collect data using the Bosch XDK 110 sensor and send the data to ThingSpeak, which i then need to analyse the data and figure out the predictive algorithm. I am currently using sample data to work on to try it out first. However i am unsure of how to do data analysis. May i request help from anybody who knows to help out.
The question is too general, that is why it is not possible to give you precise answer, but I will try to give you some tips.
First, answer for yourself for these two questions:
How often you will read temperature data? Once per month/week/day/hour/minute?
How many data points you want to predict?
Generally, analysing temperature data you can expect periodical behaviour. So, you may assume model: some cyclical trend (maybe even some deterministic function like sin or cos) and some noise.
Check (google) Holt-Winters seasonal model.
I believe thingspeak allows data insertion once every 15 seconds for the trial account, however, not sure why, the bosch sends out data about every 1 min. The temperature will always be monitored, as i am doing a Industry 4.0 project, doing machine monitoring. So, it's like a live monitoring thing.
I am, however, unsure of how many data points to predict as i've never done this before. But for my current project, i can say, perhaps just a small amount should be enough.
My plan is: To get the average working temperature of the machine, in my case, a 3d printer. So when there is a anomaly in the temperature, it sends out notifications via thingHTTP and thingTweet to notify the user.
Your plan doesn't sound like a prediction model. It is more like an alert when present situation is different than normal/average behaviour.
You can calculate simple mean of your temperature AVG measurements and its standard deviations SD.
Then you can rise an alarm if present temperature does not belong into interval (AVG - alpha1*SD, AVG + alpha2*SD), where alpha1 and alpha2 are some constants.
Those constants depends how much dispersion from the mean you are able accept as normal. alpha1 and alpha2 could be equal, but it is up to you if you want symmetrical interval or not. For example, you can accept AVG - 3*SD temperature, but you will want to rise an alarm if temperature exceeds value AVG+2*SD.
Add some complexity:
Will you be measuring the temperature 24h per day or just when 3d printer is on?
If you think about measuring the temperature 24h per day you'll have to face cyclical behaviour - you will have different values temperature at night and during the day and also it does matter if the 3d printer is on or off.
So, the simple model should be divided (your measurments) on some time intervals conditionally if your 3d printer is on or off.
I think i made a mistake in try to explain what im doing but i see you got the right idea. But i need a predictive graph for my project as well. As for my project, we are aiming to integrate Industry 4.0 into it. Attached is the link to what Industry 4.0 is aiming for.
As for that, it requires 24h monitoring as it will mostly be used on heavy machineries used in factories. However, i forgot about the fact that im using a 3d printer, therefore it will only be monitored when it is on.
Most Users Ever Online: 114
Currently Browsing this Page:
Guest Posters: 1
Administrators: Hans, lee