1st of all, your impulse responses appear quite unstable. Have you checked out VAR / View / Lag Structure / Lag exclusion test and Lag length criteria? Have you verified the stability of the model under VAR / View / Lag Structure / AR Roots Table?
Interpretation of IRFs: I think they are not correctly estimated, but let me suppose they are correct as they are (I mean sign and evolution...) Then, both your variables yield a positive response (increase) to own positive shock (unexpected increase). The same appears in the case of mutual shocks.
(I do not quite understand why it is interesting to analyze fuel and manufacturing exports in a bivariate VAR.)
Good luck with your work!