Test to prove change in data structure over time? Hello its my first time here!
I have a data of engine oil with its 18 parameters. We measure them every 1000km. It is being measured 11 times.
At each measurment each of parameters change. 
How I can prove that the actual structure is changing? (I think i has to compare 18 variables over 11 times?)
I was thinking about using ANOVA (or similiar) and making 18 groups with 11 measurments, then comparing those groups. What should I do? It is needed to show some kind of statistic "evidence" to prove that the structure is changing.
Thanks in advance for help!
UPDATE: I actually used now similiarity analysis with Bray-Curtis indices. I wonder if anybody agrees with it? How I could use it for any statistics now?
 A: Have a look at ARIMA models for modelling each of your parameters independently. 
Since you have multiple predictors Vector Autoregression might provide a more descriptive model if you suspect there are relationships between your 18 parameters that you can use for predicting future values of each of the 18 parameters based on the historical values of all of your 18 parameters.
As an example, assume you have a VAR model that uses your 18 parameters at times (t-2) and (t-1) to predict your 18 parameters at time t:
$$
\{y_1, \dots, y_{18}\}_{t-2},\{y_n, \dots, y_{18}\}_{t-1} \rightarrow \{y_n, \dots, y_{18}\}_t
$$
You could then iterate t from t=3...11 to build a succession of VAR models of the predicted parameters at t. If the predictions for any of the parameters differed from their observed values at time t, you could conclude that there was a structural change in the differing parameters. The number of lags might have to be larger than 2 (and therefore your starting t > 3) in order to have a robust enough model to detect a structural change. 
The error measure between the predicted and observed values would then be measure of the size of the structural change (i.e. a change not successfully predicted) across all 18 parameters.
Finally, if you use R, the "forecast" package has some very powerful tools for building ARIMA models. The R package "vars" makes it very easy to fit and analyse vector auto-regression models as it produces a list of linear models.
