Experimental Methods in Earthquake Engineering - Dynamic Behavior of Concrete and Seismic Engineering - page 182

As we have already seen, in a PSD test, it is essential for the displacement
control to be carried out in an accurate way. As a matter of fact, during the step by
step process, errors can accumulate until they distort the response of the structure.
Moreover, if the test procedure is applied to a structure, the response of which
strongly depends on the stress rate, a sharp measure of the speed will also be
important in order to correctly assess the reaction forces linked to the phenomenon.
We should also note that the parameters characteristic of the tested structures are
generally not well known. As a consequence, the control algorithms have to work
within an extremely variable environment including sudden and generally
completely unforeseen disruptions. In such situations, whenever we try to impose a
position tracking control, the PID fixed gain algorithm cannot operate efficiently
enough on the whole possible behavior range of the tested structure. It then becomes
necessary to use alternative control algorithms.
Thus, to ensure sufficient accuracy of the displacement and speed of the jacks, a
recent adaptive algorithm developed at the University of Bristol [STO 90a, STO
90b, STO 92] has been inserted into the control systems. The principles of the MCS
( Minimal Control System ) algorithm which ensures a minimum synthesis of the
control are the following:
- no assessment of the parameters of the system to control is required (not even
on-line). The control algorithm requires neither any adaptive observer nor any
control by recovery;
- the parameters (unknown) of the system to control can vary with time, even
with a relatively wide bandwidth;
- the unknown quantities (such as the speed), which are highly contaminated by
noise, can anyway retroact on the system to control;
- the implementation of the algorithm is very simple.
5.3.3. Sub-structuration for the continuous method
With the conventional PSD method, running the test was mainly asynchronous.
A test step in the laboratory could last between 1 and 4 seconds depending on the
size of the imposed displacement. This context was extremely advantageous for sub-
structuration, since the loading process of the numerical sub-structure could also
operate in a completely asynchronous way. At last, as both processes wait for each
other, the calculation time of the digital part could even become longer than the
implementing time of a loading step in the lab. It was then quite easy to carry out
iterative non-linear calculations at the level of the numerical process.