If your sample in the laser beam is not randomly oriented, then you can get such an effect. So this would be mostly a problem you can get with larger crystals. This should not be a problem with liquids and unoriented amorphous solids.
Thanks for the answer! So in this case, this specific peak intensity changing cannot indicate the concentration of this component change within the mixture, right? It's better to control the particle size in a certain lower size range, isn't it?
Yes, it's better if your particles are grounded down as finely as you can. In some cases, when samples are hard to grind (like fibers), one can also place the sample on a rotating stage, to physically randomize the orientation.
On the question of "mixture" -- You mentioned you have "one component" so unless something is degrading your material, and there is random orientation within the laser "scattering volume" (where the laser hits the material), you should have the same spectrum each time (of course, there will be the counting statistics error that always occur with scattering experiments).
sorry for the confusion. It's a two inorganic solid system, and during the measurement only one peak intensity of one of the components was fluctuating.
I see. That is a different problem. In that case, you would of course still want the particles to be ground down as finely as possible, and with random orientation. But what's even more important is how well mixed your two components are. If your composition is different from point to point, then the measured spectra would naturally differ as well.
The type of Raman setup may influence relative Raman intensities: use a polarisation scambler to depolarize the incident laser radiation. This should overcome crystal orientation issues.
You may also wish to Raster the radiation with respect to the sample: Snowey Range Instruments have this ability on their Raman instruments. http://www.wysri.com/apps/sers/
@John Clarkson: the polarization scrambler is a good idea. If the crystal sample is a single crystal that is larger than the laser spot size, then the crystal axis parallel to the laser beam direction will not be scrambled, and will also not be represented in the spectrum. Admittedly, that is probably a rare situation, but one that can be remedied by finely grounding down and mixing the crystal.