The necessary condition is indeed the one given by Charles.
This is true if
$$\dim X=\dim\overline{\pi_L(X)}+\dim\overline{\pi_L^{-1}(\pi_L(x))}\tag{$\star$}$$
1
Assume that $(\star)$ holds. The target is smooth, so in particular $\pi_L(x)$ is a complete intersection. This implies that then (by the condition $(\star$)), $\overline{\pi_L^{-1}(\pi_L(x))}$ is also a complete intersection (at least near $x$), but if it is smooth, then it follows that then $X$ has to be smooth near $x$. (This is an iteration of the idea, that if a Cartier divisor is smooth, then so is the ambient space and essentially the same as what Charles suggested.)
2
Here is an example that without assuming $(\star)$ the statement is not true.
Let $X\subseteq \mathbb P^3_{u:v:w:t}$ defined by $uv=w^2$ and let $x=[0:0:0:1]$. Let $L=\{[1:0:0:0]\}$ and $M=(t=0)\subset \mathbb P^3$. Now $(\pi_L)|_X$ is dominant onto $M$, so $\overline{\pi_L(X)}=M\simeq \mathbb P^2_{v:w:t}$, in particular smooth. Furthermore, $\pi_L(x)=[0:0:1]$ and hence $\overline{\pi_L^{-1}(\pi_L(x))}=\{[\alpha:0:0:\beta]|[\alpha,\beta]\in \mathbb P^1\}\simeq \mathbb P^1$, also smooth. However, $X$ is obviously not smooth at $x$.
Remark: One can also see where the above argument breaks down for this example. Since $(\star)$ is not assumed, it does not follows that $\overline{\pi_L^{-1}(\pi_L(x))}$ is a complete intersection (and it isn't).