4.2 A Systemic View on Tailings Dams Failure Processes
39
4.2 A Systemic View on Tailings Dams Failure Processes
In this section we present a systemic approach of the “failure chain process” during the service life of dams (Oboni and Oboni 2016b). Investigations, design and
construction (IDC) and then management, monitoring and water balance control
(extended IDC, or e-IDC) of the dam are analysed with a probabilistic causality
analysis based on publicly available incident and accidents data from the last hundred years. The predictive model, geared toward filling the gap between common
practice and the “path to zero failures” goal (Mount Polley Report 2015), accommodates data-mining analytics. The model “constructs” the probability of failure of
a dam which is consistent with factual historical world-data. The causality of various factors entering in the e-IDC process and other elements in the dam’s service
life can then be individually discussed with a sensitivity analysis. We then show
where and how e-IDC process mitigative actions can benefit the most, with practical
example. Special attention is spent on common cause failure (CCF) in operations,
risk assessment and peer reviewing and inspections of tailings dams. CCF means
that a “hidden” shared cause may lead parallel components to fail, annihilating the
theoretical redundancy they have (or were designed for) (Mahesh 2014).
As we will see below, historic records show seepage and internal erosion are
less frequent failure triggers than stability and seismic loading, likely due to historic
state-of-the-art knowledge and understanding. Thus we have decided to focus the
attention on those in this book. Interestingly, and for the same state-of-the-art reasons,
liquefaction does not appear as a cause of failure in “statistical studies” published
before the mid-1990s (we discuss liquefaction in Hazard Identification, see Sect. 9.3).
4.2.1 Design Process and Common Practice Risk Assessment
After each failure the mining community sees codes evolve and imposes tougher
criteria and dam’s specifications (TSM Task Force 2015). Factors of Safety (FoS) are
eventually increased by empirical consensus among experts, whereas risk assessment
methods have generally remained unchanged over almost half a century.
The relation between FoS and the probability of failure is often misunderstood,
together with the multidimensional nature of potential damages to the environment,
infrastructure and human beings. Codes that allow designers to use, for example,
FoS = 1 under some pseudo-static seismic condition (CDA 2014, their Tables 3–5)
actually accept that a tailings dam undergoing that seismic event would have the
same chance of surviving/failing than a coin toss (p f = 0.5). If the considered
quake had a probability of 1/100 then the estimate of the risk under seismic loading would be pfs = 0.5/100 = 5 × 10
−3 times the consequences of the failure
(annually), respectively 10
−3 times the consequences, if the quake has a probability of 1/500 annually. These are certainly not safe conditions with respect to
public expectation or published tolerance thresholds (Oboni and Oboni 2013). Fail-
39
4.2 A Systemic View on Tailings Dams Failure Processes
In this section we present a systemic approach of the “failure chain process” during the service life of dams (Oboni and Oboni 2016b). Investigations, design and
construction (IDC) and then management, monitoring and water balance control
(extended IDC, or e-IDC) of the dam are analysed with a probabilistic causality
analysis based on publicly available incident and accidents data from the last hundred years. The predictive model, geared toward filling the gap between common
practice and the “path to zero failures” goal (Mount Polley Report 2015), accommodates data-mining analytics. The model “constructs” the probability of failure of
a dam which is consistent with factual historical world-data. The causality of various factors entering in the e-IDC process and other elements in the dam’s service
life can then be individually discussed with a sensitivity analysis. We then show
where and how e-IDC process mitigative actions can benefit the most, with practical
example. Special attention is spent on common cause failure (CCF) in operations,
risk assessment and peer reviewing and inspections of tailings dams. CCF means
that a “hidden” shared cause may lead parallel components to fail, annihilating the
theoretical redundancy they have (or were designed for) (Mahesh 2014).
As we will see below, historic records show seepage and internal erosion are
less frequent failure triggers than stability and seismic loading, likely due to historic
state-of-the-art knowledge and understanding. Thus we have decided to focus the
attention on those in this book. Interestingly, and for the same state-of-the-art reasons,
liquefaction does not appear as a cause of failure in “statistical studies” published
before the mid-1990s (we discuss liquefaction in Hazard Identification, see Sect. 9.3).
4.2.1 Design Process and Common Practice Risk Assessment
After each failure the mining community sees codes evolve and imposes tougher
criteria and dam’s specifications (TSM Task Force 2015). Factors of Safety (FoS) are
eventually increased by empirical consensus among experts, whereas risk assessment
methods have generally remained unchanged over almost half a century.
The relation between FoS and the probability of failure is often misunderstood,
together with the multidimensional nature of potential damages to the environment,
infrastructure and human beings. Codes that allow designers to use, for example,
FoS = 1 under some pseudo-static seismic condition (CDA 2014, their Tables 3–5)
actually accept that a tailings dam undergoing that seismic event would have the
same chance of surviving/failing than a coin toss (p f = 0.5). If the considered
quake had a probability of 1/100 then the estimate of the risk under seismic loading would be pfs = 0.5/100 = 5 × 10
−3 times the consequences of the failure
(annually), respectively 10
−3 times the consequences, if the quake has a probability of 1/500 annually. These are certainly not safe conditions with respect to
public expectation or published tolerance thresholds (Oboni and Oboni 2013). Fail-