Laser beam alignment: best practices I have to build a complex optical setup consisting of many mirrors, lenses, beam-splitters, multiple wavelengths lasers to be coupled, etc.
I am looking for the "best practices" to efficiently (and with the best precision) align a laser beam and the procedure to follow (tips, tricks, conventions). 
Edit: The used lasers are in the visible range (blue & red). My question is more like a "How to" in order to optimize the overall alignment, the lens placing for beam collimation, and to avoid aberrations. Of course, safety is extremely important (be extra careful while dealing with non-visible lasers), but this is not what I am interested in.
In which order are the components added to the setup? How to best check the alignment of each individual component?
 A: Safe Alignment
Since you're using visible wavelengths, you can align everything looking at it through webcams. They're so cheap and they see very well for many applications out to 11oonm. Put them all over your apparatus and only look at the computer screen as you align. There's little to no temptation to peek as you can almost always arrange a webcam to see far better than you could if you put your face right up near what you're aligning. The only "peek" risk is the urge to look at your hands when they're out of sight. This can be overcome by (1) wearing the appropriate laser safety googles and (2) practicing alignment with a low power class 1 laser so you get into the habit of focussing solely on the computer screen image. 
If you can't get a computer screen near your apparatus, simply set up a web server on your computer giving a feed of the webcam's output. Then you can browse to the website on your phone or tablet that you stand next to you. Make sure that the latter is on a stand so that it can't fall over into the beam. Another solution is to get hold of some 3D goggles and either download an app to project the webcam output into them or write one yourself. I use a "Google Cardboard" holder to turn my phone into goggles, and, since its a Windows phone, it's very easy to access the relevant APIs through MS Visual Studio written apps. I'm sure Android is just as easy. Then you can stand next to your apparatus and "look direct at it" with zero chance of a laser eye injury. 
For class 4 you'll need to be a little careful with your skin, especially when your hands are out of sight whilst you're looking in goggles or at a computer screen. You may wish to consider all remote, motorized actuation of your adjustment. I would say this is compulsory for anything above 1 watt power. For low end class 4 under a watt and longer wavelengths (down to about 550nm), I personally deem the risk acceptable to use my hands. You must decide for yourself. The beam is about as harmful as a small flame - and also as painful - which is good because one pulls ones hand away if it gets in the beam and any burn is thus minor. Shorter wavelengths 550nm and under bring the risk of photochemical damage and the attendant - small but nonzero - risk of skin cancer from a laser burn. As I said, I am comfortable with the risk for lower power, longer wavelength class 4 lasers but you must decide for yourself - I don't earn any money as a hand product model for Estée Lauder or Elizabeth Aarden and I'm in good health so small injuries are of little bother to me. Also, ONLY decide for yourself - never for a colleague and NEVER for a subordinate. If you have a student or subordinate doing this work for you, you MUST use motorized remote stages for any class 4 laser. Don't even ask a subordinate, it's grossly unfair: they simply cannot, by dint of your relationship with them, answer impartially.
If you do use your hands, make sure to take all rings and watches off so that the beam doesn't end up in someone else's eyes. There should, however, be no idle bystanders throughout alignment - only people who are truly needed for the procedure and who understand laser safety. That said, an experimenter whose job it is to switch the laser on and off as you need it and who can watch what you're doing is a good idea.
It is also best practice to do laser alignment with the laboratory lights switched on. Use bright lighting. The aim here is to shrink the eye's pupil to about 1mm, the small aperture it uses on a bright, sunlit day. If you're aligning in the dark, the pupil is about 7mm in diameter: that means that the probability of a laser's getting in the eye is fifty fold higher.
Below are some shots of webcam arrangements I have used in the past. You need either to get a friendly machinist to make brackets for you to attach them to the various Newport/ Thorlabs/ Edmund /.... stages and supports and allow them to swivvel so that you can adjust their view easily. It's easy to screw directly into the webcam's plastic housing. You should get some ideas from the below.



Telescopes
User Martin Beckett makes the point that I've also used:

If you have space it's often easier to replace the laser with a telescope and a camera (look for survey level telescopes) then you can adjust a component while looking at the image of a target at the output. 

Martin means here a builder's Dumpy Level. They are compact and good quality ones can now be bought for a few hundred dollars. They are also available with standard camera tripod mounts: check this before you get one because the builder's tripods are generally too bulky to use in a laboratory. I usually dispense with the tripod altogether and mount the level on the optical bench on a tip /tilt + XYZ stage. A riflescope of comparable magnification (say 32x and over) can also be used here - this may be a better shape (long and thin) for some applications. 
If you use a dumpy level, you may also wish to get hold of some fiber bundle lighting kits. These let you shine light into your apparatus so that you can see through the level at alignment targets. Surprisingly little light gets through a high magnification telescope (well duh! it's gathering light from a tiny solid angle, but still it always catches me by surprise) so you will probably need lighting to see properly.
The dumpy level is used exactly as you intuitively think - to "fare the photon's path". Wherever the crosshairs land will be where the laser beam will go, when the laser's cavity is aligned with the dumpy level's optical axis.
This last point shows a second aspect to this technique: you must design a method to bring the laser's cavity to alignment with the dumpy level's optical axis after you've aligned everything with the level. Naïvely, this means mounting the laser on a tip/tilt XY stage - the simplest and the best method - but this is impracticable for many bulky laboratory lasers. If so, you will need to pass the beam from the laser through a crossed periscope for your X and Y translational degrees of freedom as well as a tip/tilt mirror. All of these degrees of freedom are coupled, so this is not a trivial exercise. You will need an alignment plan (see below) for this step - it can end up being harder than skipping the level altogether. But sometimes, this last step is easy - once aligned with each other, the components are sometimes quite tolerant of the laser's alignment.
The general rule here is: the use of a level only saves work to set up spatial relationships between components other than the laser. It does not help align the laser. 
For example, a Michelson interferometer requires an exquisitely precise angular relationships between the planes of the reflectors and the plane of the beamsplitter. A level will help you set this up beautifully. The Michelson, once aligned, is pretty tolerant of the laser's direction. You can easily have a sizeable fraction of a degree perturbation to the laser's alignment and an aligned Michelson will simply self correct for it. 
So some applications will benefit from a level, others not.
Alignment Lasers
Similar to the use of a level is the use of alignment lasers. That is, one uses a IEC60825 Class 1 laser (NOT Class 1M) of the same wavelength as the final to do one's alignment with. One does have the same problem at the end as with the level: you must design an alignment procedure to bring the final laser's optical axis into alignment with that of the alignment laser when the latter is replaced with the former. I generally have a package made where I package an alignment laser with a collimator and other optics to achieve the same beamwidth as the final laser will output.
Alignment lasers have obvious safety benefits and you can work so much faster when there is only a Class 1 laser threat. Another benefit is that the beam is often of much higher quality - and coherence - than the final. This lets you use extremely useful tools such as the Shearing Interferometer (which I talk about below) for collimation. Large, high power lasers often do not have the coherence length to allow the use of this vital tool.
Before Procedure Begins
With any complicated system, you need to design an alignment procedure at the outset and calculate theoretically exactly what you will see as you bring each part of the kit into alignment. This, naturally, also includes an alignment criterion - how you decide when alignment is done. Fringe shapes and how they change as you twiddle knobs, intensity profiles, the lot.
When you get to a stage in your alignment that does not agree with theory, stop, make sure you understand what's wrong and reconcile the two (theory and practice) before you go forward. You'll learn a great deal about your setup this way.
Collimation
By far the most effective - and safest - way to collimate a beam is through the use of a Shearing Interferometer. You don't have to aim a class 4 beam across a laboratory to check its divergence and with a shearplate the accuracy of the collimation you can achieve is equivalent to that gotten by checking the beam divergence over hundreds of meters or even kilometers of axial beam length. So it's waaay more effective and safer than many of the collimation setups aiming beams across laboratories I see.
Again, the fringe pattern must be viewed with a camera. You simply use the CCD chip in the camera and remove the webcam's  lens: the fringe pattern is a structured collimated beam. You can use the shutter speed to help you adjust brightness but you will almost certainly need to add an attenuator to prevent CCD chip damage.
The only catch is that you need to check that the lasers are coherent enough to use the shearplate: this means that they need to have an axial coherence length of a centimeter or so, since most shearplates are several millimeters thick. The $1/e$ coherence length for a Lorentzian lineshape is:
$$L_c=\frac{\lambda_0^2}{\pi\,\Delta\lambda} = \frac{c}{\pi\,\Delta\nu}$$
where $\lambda_0$ is the center wavelength, $\Delta\lambda$ the full width half maximum wavelength spread and $\Delta\nu$ the full width half maximum frequency spread. Therefore, for a 1cm coherence length, you need a 633nm laser to have a wavelength spread of about 0.013nm, equivalent to 9.5GHz linewidth. This may be a tall ask for some of the more powerful lasers. If so, you will need to use an alignment laser discussed above as the imperative for avoiding the across-the-laboratory collimation method becomes stronger with increasing laser power.
You can make your own shearing interferometer from a microscope slide but the easiest ones to use have a wedged shearplate so you're probably going to have to buy one for the reasons I discuss below. A diagram from the wikipedia page I linked is below

so that the device is creating interference between the two Fresnel reflected beams from either side of a parallel plate. The device is basically plotting a fringe pattern for the phase map $\Delta\,\nabla_\vec{n} \varphi$ where $\varphi$ is the phase field in the incoming beam, $\Delta$ the magnitude and $\vec{n}$ the direction of the displacement between the two beams. Converging/ diverging beams have, to lowest order:
$$\varphi(\vec{r}) = \varphi + \vec{k}_\perp \cdot \vec{r} + \frac{1}{2}\,\vec{r}^T \kappa\,\vec{r}\tag{1}$$
where $\vec{k}_\perp$ is the component of the wavevector normal to the nominal direction (and thus measures the beam's tilt) and $\kappa$ is the curvature matrix (related to the curvature tensor). Therefore, the shearplate's fringes are set by the field 
$$\Delta\,\nabla_\vec{n} \varphi \propto \vec{k}_\perp \cdot \vec{n} + \langle \kappa\vec{n},\,\vec{r}\rangle\tag{2}$$
which gives a linear fringe pattern.  Collimation is the elimination of $\kappa$. So, from the above equation, you adjust your optics until the linear fringes disappear. That's for a flat shearplate. It is hard to find the point where the last fringe disappears accurately, so commercial shearplates have a wedge between the two surfaces. That way, there is always a linear term in (2): the absence of the $\langle \kappa\vec{n},\,\vec{r}\rangle$ means that the fringe direction is set by the wedge and when $\kappa$ is nonzero they rotate to some other direction so you simply adjust until you have linear fringes pointing in the direction indicated on the instrument.
One thing to be aware of, especially with lasers, is that they cannot always be collimated with an axisymmetric lens like a microscope objective. This happens if the principal curvatures in $\kappa$ above are unequal, and therefore the introduction of a constant curvature into the beam by the use of a collimating axisymmetric lens cannot annul both principal curvatures at once. This is the phenomenon of astigmatism and if you suspect this situation, you must use the shearing interferometer in two positions, rotated through $90^\circ$ about the input optical axis, from one another to check that both principal wavefront curvatures have been annulled. If there is astigmatism, it can only be corrected using cylindrical lenses.
Here is a cross section of a setup I have used with a camera, whowing the relative positions. The tested beam comes in through the enclosed tube from the left and there is a beam stop on the right.

A companion instrument is the autocollimator. I don't use this much as it's pretty much redundant if one uses a dumpy level and shearing interferometer. But it can be compact and convenient, so it's worth looking into. 
The Astronomer's Star Test
Beam quality - lack of aberration - is ideally assessed by a point diffraction interferometer. Here's my own writeup based on a commercial device here. This is a device that uses a pinhole to create a reference beam from the input beam, and thus can measure wavefront aberration of a laser beam.
An interferometer is a bulky and awkward instrument to get into a system, so you may be tempted to use a Hartmann wavefront sensor; see also the Wiki page here. In my experience, although they require little experience to use, one can get a much more accurate confirmation of low aberration with the Astronomer's Star Test and you need only a CCD screen, a focussing lens and your brain to do it.
Again, I emphasize, you must do the star test either with a camera or IEC60825 class 1 (NOT class 1M) alignment laser! Read up on this from the many amateur astronomer's websites. With experience, you will get confirmation of aberration down to as low as 0.02 waves RMS easily. A Hartmann sensor is good down to about 0.07 waves RMS. A refinement is to compare the intensity of a focal spot to one generated from a beam of the same wavelength and beamwidth, thus measuring the Strehl ratio directly. This can help if you are inexperienced, but an experienced operator will measure the equivalent of about 0.97 Strehl from the shape of the point spread function alone.
For alignment, in most cases, you will get as good a result as with the use of an interferometer.
