ArrowHead Solutions – Safety and Operational Risk

ArrowHead Solution Helps Clients Accurately Quantify and Mitigate Operational Risks

Need for Better Understanding Safety and Operational Risks

Safety and operating reliability issues have always been a high-priority concern for energy (and other) companies. However, in recent years there have been a number of highly visible incidents that have highlighted and exacerbated the risks and consequences of failure, catastrophic or otherwise. There have also been a number of significant pipeline incidents, such as the explosion of the gas pipeline in San Bruno, California, resulting in eight deaths, actions and fines against the owner, and a public relations nightmare. The fire at a refinery in Northern California in 2012 resulted in lost lives, tort, loss-of-production and revenue, and property damage. Nearly every year there are refinery accidents that result in fatalities. With the sniper incident in 2013 disabling the Metcalf power station in California, terrorism is another risk to be understood and managed. The 2011 Fukushima nuclear incident (a “perfect storm” event between a massive subduction fault earthquake not far offshore having direct frontal shoreline exposure) portends that natural disasters must be added to the list of prospective precursor events. These events and others emphasize the need for industry to fully understand safety and reliability risks and to assess the adequacy of methods for managing the widening array of risks. It is no longer acceptable to “get by” with judgmental or summary methods. Plant reliability has to be assessed and quantified, and the appropriate safety and failure dimensions have to be understood element by element on a probabilistic basis and mitigated in the most effective and cost effective fashion.

Realistic mitigation of any of the risks, including preventive maintenance, preventive inspection, contingent maintenance or rehabilitation, training for improved human cognition and performance, repairing or retrofitting aging equipment, replacing existing equipment, forensic analysis, improving labor and monitoring processes, total or partial rehabilitation, or addressing potential newly-identified risks, requires investment. There is cost associated with each of these activities.

  • How does one best strategically decide such investments?
  • How does one assess the value of preventive maintenance versus scheduled replacement versus inspection contingent action?
  • Are such investments even effective?
  • How does one understand the relative risk of each danger compared to the others to help prioritize such investment?
  • How does one quantify absolute risk and exposure? (Absolute risk matters a lot. In the final analysis, absolute risk is what one really faces.)
  • How does one assess the best choice among various potential approaches for managing each specific risk?

ArrowHead’s approach circumvents those difficulties.

How Arrowhead Can Help

We have chosen a methodology that provides the required capabilities needed to assess and mitigate operational risk – Bayesian probability and Bayesian networks as shown in the section below. We have developed advanced analytics using this methodology (ArrowHead Bayesian Analytics) which we have used to accurately quantify risk of various types. Examples of the diverse risks for which our analytics are suited include:

  • Spontaneous failure of machines and infrastructure such as pipelines, generation or transmission facilities, mills, and machinery of all types
  • Intentional damage (sabotage, terrorism) to machines and infrastructure such as pipelines, generation or transmission facilities, mills, refineries, offshore platforms, and machinery of all types
  • The specific value of management decisions (e.g., rehabilitation, preventive maintenance) and of short-term decisions (e.g., proper operation or training) as it affects component and total plant failure
  • The consequences (damages) of each given failure sequence – more properly, the probability distribution over consequences
  • The deleterious effects (costs) of such damages, and the probability distribution over damage costs under each mitigation strategy one might pursue
  • Force majeure damage sequences such as hurricanes, storms, earthquakes

This ArrowHead Solution helps clients to specifically and accurately quantify:

  • The true and correct probability distribution (which we call the “authentic” probability distribution over plant failure)
  • The authentic probability distribution over damages caused by plant failure
  • The authentic probability distribution over damage costs caused by plant failure
  • The specific impact of management decisions related to the foregoing probability distributions so you can know the specific safety consequences and improvements from your actions such as:
    • Inspections and contingent rehabilitation or retrofit
    • Routine preventive maintenance
    • Periodic, regularly scheduled part replacement (like changing an oil filter every 4,000 miles)
    • Expensive blanket rehabilitation strategies (change every oil filter in the fleet no matter what)
    • Personnel training
    • Specific steps to thwart human intervention, terrorism, and sabotage (e.g., fences, security, armed guards, night and day visual monitoring, and automated security systems)

This ArrowHead Solution enables clients to use an imminently updatable and expandable approach. You can see from the influence diagrams below that you could ultimately expand to hundreds or more precursor failure variables and probabilities and truly refine your understanding of the true risk you face and the true impact of your mitigation strategy.

Better Alternatives for Understanding Safety and Operational Risks

Bayesian networks are directed graphs of components and their prospective failures. Nodes represent component or precursor failure/non failure events and these are connected via arcs which signify direct dependencies between the linked events. The strengths of these dependencies are quantified by conditional probabilities inserted within the event nodes. Such conditional probabilities cannot be garnered from statistical analysis. Such graphical structures, known also as influence diagrams or belief networks, are used for representing expert knowledge or engineering failure analysis. The graph is used to represent and estimate the posterior probability of unknown variables given other variables (evidence), through a process known as Bayesian probabilistic reasoning, which “rolls up” the explicit, intrinsic, embedded fundamental precursor events (the nodes) and their probabilities.

This approach works reliably when few or no statistics exist (which is always the case with plant failures). The key to the success is formulation and computation of the diagram so that most or all the failure precursors are included and correctly interrelated. The dynamics of accident sequences is critical, and assembling it for and with plant owners is crucial.

For more detail on this offering and its methods, please contact us.

Our Staff

Our staff includes experts with years of experience helping clients to understand and assess risk issues to inform their decisions. They have deep knowledge of statistics, probability, economics, and decision analysis, and they have years of experience using ArrowHead Bayesian Analytics to help clients.

A Unique Solution

This solution is highly differentiated from other methods and approaches. The ArrowHead Bayesian Analytics we use are probabilistic models that uniquely meet many of the requirements for quantifying both risk and the effectiveness of risk management techniques in a variety of ways, which include capability to:

  • Capture relationships between variables, as well as true and correct information about their relationships.
  • Combine diverse types of evidence, including both subjective beliefs and objective data – a Bayesian model is agnostic about the type of data in any variable and about the way the variables are defined.
  • Reach decisions based on visible, auditable, intuitive data. There are no hidden variables. The inference mechanism is based on a long established theorem and long published and understood methods.
  • Reason from effect to cause and vice versa. The ArrowHead Bayesian Analytics will update the probability distributions for every unknown variable whenever an observation is entered into a node.
  • Revise probabilities in real-time as new data and understanding surface.

Above all they provide an accurate and transparent way of utilizing available data and accurately quantifying risk and the effectiveness of risk remediation methods.