Many years ago I was really stumped for an understanding of the idea of transforms, convolution, correlation, modulation etc etc.
The thing that tipped the balance for me (to allow me to visualise the process) was when I looked again at the identity:
∫(-∞ to +∞) A Cos(ax) Cos(bx) dx is zero except for a = b, when it equals A
(If you avoid using the exponential notation, it is easier to explain.)
If you accept that any function f(t) can actually be represented by an infinite sum of simple harmonic functions then putting f(t) in the above gives you
∫(-∞ to +∞) f(t)Cos(bt) dt, which is the amplitude of the Cos(bx) component of f(x). That is what the Fourier transform does - it shows the correlation between your function (as a function of time) and the harmonic function with a particular angular frequency b over the whole range of b values - which shows it as a function of frequency.
I really must get to grips with writing equations better! but I think the above says what I mean. You need to tart it up a bit, with exponential notation but the basic idea is the correlation between the time function and what is, effectively, a swept frequency - to give the frequency function.