Choose the time interval based on the patterns that you want to detect. For example, to look for month-to-month patterns in a process, collect data at the same time each month. If you collect data each week, then the monthly pattern may be lost in the noise of the weekly data. If you collect data each quarter, the monthly pattern may be lost when it is averaged out in each quarter.
If you are looking only for general trends or shifts in the data over time, and not for patterns associated with a specific time interval, the length of the interval is less important.
A stationary time series has a mean, variance, and autocorrelation function that are essentially constant through time. The data is non-stationary when there is a large spike at lag 1 that slowly decreases over several lags. If you see this pattern, you should difference the data before you attempt to identify a model. To difference the data, use differences. Once you difference the data, obtain another partial autocorrelation plot.
The same pattern may occur at the seasonal lags. That is, a large correlation occurs at the first season lag and decreases over several seasonal lags. If you see this pattern, you should difference the data using a lag equal to the seasonal length before you attempt to identify a model.