Blog
ARTIFICIAL INTELLIGENCE: AN EXTRA HELP WITH DAM SECURITY
We have applied techniques such as neural networks or Bayesian networks at the La Aceña dam to analyse the behaviour of this infrastructure
Dams are feats of civil engineering whose design, construction and operation can only be understood with a factor of supreme importance: security. To guarantee this, all these concrete masses, which can withstand the pressure exerted on their walls by bodies of water, have protocols for detecting and, where appropriate, correcting any anomaly that may arise.
This infrastructure is monitored, controlled and sounded non-stop. For example, at the El Atazar dam, the largest in the Region of Madrid, 130,000 data are collected annually about its behaviour at 2,300 measurement points. This information is instantly cross-referenced using various mathematical and statistical models to verify the fact that the operation of the dam is indeed within “normal” parameters (the regulations require that maximum admissible values be specified). It’s like that with them all.
For more than two decades, at Canal de Isabel II we have used predictive models of various types to estimate the ideal behaviour of each dam, mainly based on the temperature and the quantity of water dammed at any given time, though other magnitudes are also borne in mind.
We therefore have many years’ experience in sounding and very reliable models; however, we wanted to go a step further, by introducing artificial intelligence. The fact is that neural networks, Bayesian networks or decision trees are modern tools that make it possible to establish the normal operating range of a dam with enormous precision. Not only that, they are also used to identify the complex dependency relationships between their various elements (a distinguishing quality).
Therefore, we have recently applied these techniques at La Aceña, a dam from which we take 15,000 measurements annually (reservoir level, temperature of the concrete, leakage, displacement, deformation, etc.). In this way, we have updated the historical sounding data, analysed the plans and location of the instruments and measuring devices and, finally, thanks to artificial intelligence, built a rigorous model of what the dam’s normal behaviour should be.
Paradoxically, the results generated by these advanced software are almost a copy of the conventional statistical models we have always worked with, which, without a doubt, reinforces the validity of the surveillance techniques used in buildings that are monitored in great detail. It could hardly be otherwise.