Understanding the difference of using an AFT model and ALT in Predicting Time to Failure

When I started my research in survival analysis, I began to learn about using non-parametric, semi parametric and parametric methods. I was experimenting my data using these approaches. However the examples were mainly coming from the medical domain which has a different kind of data structure. Eventually I understood that my data is of an accelerated life testing data where I have several a combination of several parameters that are replicated. A close example would be the motor data provided in R but only for 1 parameter. My data would be close to this example: http://www.scielo.org.co/scielo.php?script=sci_arttext&pid=S0012-73532015000300019

I have so far been using the survival regression analysis in R with a weibull distribution to predict the time to failure for the machines. So it looks like below:

weibull_model <- survreg(surv_func~a1+a2+a3+a4, data=train, dist="weibull")

I split the data into train and test so that I can use the model trained on the train data to predict on the test data.

But I do feel that what I am trying to achieve may not be accurate because it seems as if I am parameterising the model across all the test conditions and not according to the different scales in a1, a2 etc.. which is what I noticed when dealing with ALT data.

I would appreciate advice on wheter it makes sense to apply an AFT model on ALT data and whether what I am doing above is acceptable. I get time predictions but if the way i am building the model is wrong, then it would be misleading. Thank you.