I am trying to understand the smooth classification of $n$-disk bundles over $S^n$. As vector bundles, these are classified by $\pi_{n-1}(SO(n))$ via the clutching construction but I am interested in their smooth type. For example, the map $\pi_{n-1}(SO(n)) \rightarrow \pi_{n-1}(Diff(D^n))$ might not be injective in which case two distinct vector bundles would be the same as smooth manifolds. What is known about the smooth type of these bundles? The homology of the total space is just $\mathbb{Z}$ in degree $0, n$ and the intersection form is determined by a single integer, which equals $HJ: \pi_{n-1}(SO(n)) \rightarrow \mathbb{Z}$, where $J: \pi_{n-1}(SO(n)) \rightarrow \pi_{2n-1}(S^n)$ and $H: \pi_{2n-1}(S^n) \rightarrow \mathbb{Z}$ is the Hopf invariant.
These $n$-disk bundles over $S^n$ also correspond to handlebodies with a single critical point of index 0 and n and are $(n-1)$-connected $2n$ manifolds as studied by Wall. Wall came up with an invariant of handlebody presentations; in the case of these disk bundles, Wall's invariant is just an element of $\pi_{n-1}(SO(n))$. However, as explained earlier, these don't seem to be really diffeomorphism invariants (although Wall calls them that..). The problem is that Wall's invariant is invariant under handleslides (change of basis) but he seems to ignore the possibility of birth-death moves that create cancelling $n-1, n$ handles and then more handleslides. Am I correct in thinking this?
Edit See my other question for the general case: What does Wall's classification classify?.
I can't find an example which is not behind a paywall, but in
One finds an argument (attributed to Haefliger and Levine) that there is an embedding of $S^{11}$ into $\mathbb{R}^{17}$ for which the normal bundle is not isomorphic to a product, but whose total space is diffeomorphic to the product. Later on in the same paper (pg 927, first paragraph), there are other related examples.