OK, in general you have a linear equation whose coefficient are function of a variable "a", that is:
Y=m(a)*x+n(a)
so you want to determine the funtions m(a) and n(a)
from your data we have (ordered by increasing value of "a"):
m(a): m(0.5)=0.092, m(1)=0.106, m(2)=0.028
n(a): n(0.5)=-0.03, n(1)=0.103, n(2)=0.078
we can see that "m" and "n" arent' linear with respect to the variable "a", so you can't have an expression like that you showed (I mean: Y=pX+qa+r) unless you can accept very large errors.
You can instead use a quadratic model; that is m(a)=hm*a^2+km*a+qm and n(a)=hn*a^2+kn*a+qn. This because it's always possible to fit a parabola given 3 points.
numerically you have to solve (with respect to h,k and q) the two linear systems of equations:
0.092 =0.25*hm+0.5*km+qm
0.106=hm+km+qm
0.028=4*hm+2*km+qm
-0.03 =0.25*hn+0.5*kn+qn
0.103=hn+kn+qn
0.078=4*hn+2*kn+qn
Then Y=m(a)*x+n(a) will be:
Y=(hm*a^2+km*a+qm)*x+hn*a^2+kn*a+qn
NOTE: this last will be exact only in the three point (a=0.5, 1 and 2 mm)