0
$\begingroup$

Consider a feature set containing values of operating condition of an automation unit such as pressure of the valve, temperature, fuel consumption. I want to discover unknown relationship between these variables, to come up with some functional relationship as well.

Another case can be to detect fire from sensor nodes which collects information by sensors, e.g., temperature, light, and smoke. But the signature profile is not known. In this example, it is intuitive to know when a fire occurs which is when temperature, light and smoke all three are present then there is a fire hazard. Therefore, there is a need to discoversome correlation between the attributes. Which data mining approach is suitable to discover such relation/pattern?

I want a machine learning method for feature learning so as to 1) identify any patterns that aren't obvious and 2) allow an algorithm to identify signatures of patterns in the data. For example, to see a correlation between the pressure and temperature to suggest equipment failure, say if temperature is high and pressure is low or both high or both low. These kinds of patterns and correlation need to be discovered from the data.

Question1: Can somebody please suggest some data mining approaches to solve this kind of forecast/ alert problem.

Question2: Is this a pattern recognition problem -- if so, is it supervised or unsupervised?

Question3: Can somebody please point out some examples/research articles which are related to this kind of a problem such that I can get a clear picture on how to solve it. Than you very much.

$\endgroup$
  • $\begingroup$ Cross-posted: stackoverflow.com/q/49212685/781723, cs.stackexchange.com/q/89210/755, stats.stackexchange.com/q/332854/2921. Please do not post the same question on multiple sites. Each community should have an honest shot at answering without anybody's time being wasted. $\endgroup$ – D.W. Mar 11 '18 at 19:08
  • $\begingroup$ What do you want the output to look like? Are you looking for an equation or something like that? Or are you really wanting a model that will let you detect future deviations from normal (i.e., future instances that don't follow the patterns that seem to hold in the past)? $\endgroup$ – D.W. Mar 11 '18 at 19:08
  • $\begingroup$ @D.W.: I want to discover if there is any correlation between the variables which denote the operating conditions of the machinery unit. I want to discover some interesting pattern which may indicate a fault or not a fault. However, I don't know what combination of variables will cause a fault. Can you suggest approaches which are applicable to this problem? thank you $\endgroup$ – SKM Mar 11 '18 at 21:08
1
$\begingroup$

Couple a months ago there was a presentation on Data Science Warsaw meetup on similar topic (data mining in energy industry). Unfortunately the presentation is in Polish. What follows will be based on the presentation.

Q1, Q2:

Anomaly detection seems like a good match, if you have systems which are fine most of the time, but can exhibit abnormal faulty behavior. The presentation mentions using autoencoders for such application.

If you have labeled data (whether it should be alarming or not) you might be able to use classification. Of course then you'd need to be careful about your data, because if you have faulty/not faulty situation, then most likely you have mostly not faulty data which will result in severe class imbalance (see Class imbalance in Supervised Machine Learning ).

Q3:

I googled 'outlier detection industrial applications' and got this article. I think that this query, and also one with 'anomaly detection' might be of interest to you.

$\endgroup$
  • $\begingroup$ Than you for your reply. I don't have information about anomalous data records. I want to find if valve pressure changes effects the temperature, or if pressure changes effects the fuel consumption etc....these kind of unknown relationship. So does this fall under pattern recognition and data mining? $\endgroup$ – SKM Mar 11 '18 at 17:55
  • $\begingroup$ What you mean by 'alerts' then? $\endgroup$ – Jakub Bartczuk Mar 11 '18 at 17:56
  • $\begingroup$ I want a machine learning method for feature learning so as to 1) identify any patterns that aren't obvious and 2) allow an algorithm to identify signatures of patterns in the data. For example, to see a correlation between the pressure and temperature to suggest equipment failure, say if temperature is high and pressure is low or both high or both low. These kinds of patterns and correlation need to be discovered from the data. $\endgroup$ – SKM Mar 11 '18 at 18:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.