Failures

Potential failures leading to a loss of containment associated with any plant or section of a plant can be broadly divided into two main categories, namely:

• The possibility of failure associated with each mechanical component of the plant (for example - vessels, pipes, pumps, compressors). These are generic failures and can be caused by such mechanisms as corrosion, vibration or external impact (mechanical or overpressure). A small event (such as a leak) may escalate to a bigger event, by itself causing a larger failure. These failures are known as generic failures and are associated with the inherent properties of a material, rather than a failure associated with any maleoperation etc. These Generic failures are often not under the control of the Industry. An analysis of this class of failures requires consideration of each component under its normal operating conditions.
•  The Second type of failure are those caused by specific operating circumstances. The prime example of this is human error, however it can also include other accidents due, for example, to reaction runaway or equipment overpressure due to loss of control.These are due to operation beyond design.
  
  Both classes may also require consideration of some components under abnormal conditions. Such failures are often associated with operation of equipment beyond the design capability/ capacity, or close to the design for extended periods. Identification of failures in this second class (specific failures) are based on a formal engineering review of possible failure modes, utilising engineering experience, awareness of specific failure modes and knowledge of the systems under review. The results of HAZOP study could be useful in this part of the study.

Estimation of Failure frequency

 

Failure frequency assessment is a very important aspect of risk estimation- it identifies and differentiates the " high " frequency events from the " low " frequency ones- it is the high frequency and " continuous" events that actually contribute to the chronic hazard towards people and environment. There are several techniques for frequency assessment, and these include:

• Review of past historical data compiled over many years through established databases and correcting the same based on technological change, plant environment, safety procedure and safety and environmental systems etc.
• Based on Actual Site Conditions: Where the required data is not available/ irrelevant, use of other methods is necessary- these techniques typically include:based on actual site conditions.
• Event Tree Analysis- this is used to calculate the probability of different possible INCIDENT OUTCOME CASES, assuming the incident has already occurred. A logical Event Tree is constructed for different incident outcome cased. For example, and Release of material from a vent (incident) could be instantaneous or slow (Incident outcome) and could either ignite immediately, later or not at all (Incident outcome case). Each branch can be further broken down if required (e.g. wind in S-SE, N-NE etc. each forming a different Incident outcome case). The frequency of the different cases is evaluated through event probability . Suitable assumptions are made
• Fuzzy Fault Tree Analysis- based on the inaccuracies of data and the availability of data to well define the problem in question to the desired precision. Fuzzy Tree construction offers a framework that models the imprecision in failure probabilities used in FTA. The concept of Fuzzy Top Event Probability (FTEP) is through use of a Fuzzy number, which lies between 0 and 1. for each primary event. The interval of confidence reduces the uncertainty of using upper and lower bounds.
• Fault Tree Analysis- a technique developed at Bell Laboratories in 1961 for Missile launch systems, further developed by Haasl in 1965 at Boeing and later Rasmussen in 1975 for Nuclear Reactors. Nowadays, great application of Fault Tree methods in practiced in the chemical industry. The system assumes that all failures are BINARY in nature (success or failure) or based on ATTRIBUTES, not a continuum. The underlying logic is simple and based on logic gates (AND , OR) to synthesize a failure model of the plant. Minimal CUT SET ANALYSIS is used to provide a qualitative insight into the potential failure modes of a complex system. In addition, a system is assumed to work if all its subsystems are working. FTA is used to estimate the TOP event probability and determines the discrete contribution of events for the event. Fault Trees can be constructed in THREE ways, namely Manually, Algorithmically or Automatically.
  
  The Manual method takes a top down approach and requires great skill and is continues till the analyst is convinced that the model describes the problem in question. Study boundaries need to be clearly defined, hence further development of the Tree is discontinued. Each gate is numbered and the top event probability estimated through simple rules of addition (for OR gate events) or multiplied (for AND gate events) . The method is subjective and a lot lies in the hand of the Analyst.
  The Algorithmic method is a more systematic and less subjective method through decomposition of a system into components and applying a set of generic failure patterns applicable with minor modification. Typical methods for Process systems evaluation include the DIGRAPH method and others. The skill then shifts to modifying or TAILORING a generic pattern rather than developing it.
  The Automatic system enters data from Process PIDs into computer codes to obtain all TOP event probabilities. Some of the common codes include the CAT code (Salem et al), the RIKKE code (Taylor et al.) and the FAULT Propagation code (Martin-Solis et al.) . The use is difficult and success has been limited.

Analysis of the Fault Tree is done through Minimal Cut Set Analysis- the effectiveness of the protective systems, the combination of events leading to failure as well as COMMON MODE failures are easily worked out through Minimum Cut Set Analysis. Where Trees are more complex, inspection of the Tree is too difficult and methods such as BOOLEAN Analysis are necessary. Fault Trees are converted into equivalent BOOLEAN expressions defining TOP event in terms of a combination of lower events. This expression is expanded using the laws of BOOLEAN algebra until it expresses the TOP event as the sum of all minimal Cut Sets. While the algebra is very tedious and requires good practice, automatic procedures exist for evaluation of Boolean terms- some established systems include the MOCUS system of Fussel et al.