Rumsfeld theory states, “we know what we know (known knowns), we know what we do not know (known unknowns), and we do not know what we do not know (unknown unknowns).” The last two statements are the most dangerous statements in the asset integrity industry, as these contribute to uncertain situations and in some cases catastrophic outcomes unless focused action is taken.
By Kelly P. Caillier, Founder and CEO, KDC Reliability
Uncertainty in the asset integrity industry comes in many forms. Quite often, limitations produce uncertainties. A few limitations from an exhaustive list are a lack of needed budget, a lack of qualified resources, a lack of objectivity in the inspection process, a lack of knowledge, or a lack of systematic communication. Processes in general, but specifically speaking to asset integrity, are normally vulnerable due to these select few and many other organizational limitations. It has been studied by many credible investigation fi rms that the most catastrophic failures can be categorized into three sections.
First, lack of understanding of damage mechanisms and their manifestation. That is fundamentally a limitation of knowledge. Second is the lack of a ‘concurrent engineering’ process or communication silos. This is primarily a limitation of systematic communication.
And lastly, the lack of repeatable and reproducible action sets. This is primarily due to a limitation of focus, standards, training, and subjectivity. Once vulnerabilities are known, either an acceptance or improvement effort opportunity exist. Further, once these vulnerabilities are accepted or improved, one must consider who is now responsible for the results. Being confidently responsible for the results can save lives and jobs.
Industry Subjectivity Case Study
Abyss Solutions, an artificial intelligence and machine learning firm, performed a study with a major owner-operator in the Gulf of Mexico and around the world. The study proposed created several degrees of coating degradation and substrate surface rusting and invited several industry subject matter experts (SME), with more experience than normally witnessed in the fi eld, to participate in the study. The SME objective was to access and grade the variation of degradation in the samples and grade the condition with only a three-level grading system: heavy corrosion, moderate corrosion, and low corrosion. The result of the test was resonating. Among the experts with advanced years of training and development, and with only three categories, there was a 40% agreement rate.
There are international standards prepared for this type of assessment such as ISO 4628, the evaluation of degradation of coatings, and assessment of the degree of rusting. The standards utilize a comparison library of photographs displaying various levels of degradation. Currently, the process involves the trained human interacting with photograph sets and conducting a condition evaluation. These results are then applied to validate past predictions and calculate for the future condition; therefore, repeatability and reliability are of the utmost importance.
Market Research Evidence
A majority of the assets in the oil & gas and chemical industries are labeled under the piping system category due to the large number and complexity. According to Elsevier and the study of “Process Safety and Environmental Protection”, the chemical process industries (CPI) process equipment accidents. According to the report, the most frequent equipment in CPI accidents is piping, by 25%.
Additionally, 41 out of 234 incidents, or 18% of CPI piping system failures are due to human and organizational failures. Organizational structure failure comprises of 63% of the cases and human failure contains the other 37%. According to the study, organizational structure failure consists of activities such as poor management of change systems, lack of inspection knowledge, and poor communication.
Human failures consist of activities such as misjudgment, not following procedures, knowledge-based ignorance, and poor training. A poor organizational structure and human inconsistencies when they are not supported by a strong repeatable structure are not reliable and do not contain a foundation for consistency and repeatability.
According to the Joint Research Centre studying corrosion-related accidents in petroleum refineries in 2013, 99 refinery accidents were studied in the EU and OECD countries. In 53 cases, process conditions were identified as contributing to the corrosive conditions preceding the accident and should have been According to the study, 71% of accidents originated in the piping system.
The study further identified that 40 accidents occurred since the year 2000, indicating that major accidents at refineries involving corrosion failure continue to be a particular cause of concern in the 21st century. In this study, a majority of the accidents were due to the damage mechanisms not being properly identified, measured, or ignored. Most importantly, the study revealed that experts sometimes overlooked the manifestation contributors of the process that created accelerated corrosion.
Artificial Intelligence/Machine Learning Solution
In risk-based inspection systems, which have become common in the asset integrity industry, the utilization of repeatable collected data sets is essential in predicting future activities and calculating the frequency of an inspection or data collection activity. The calculations of probability of failure (PoF), especially in medium to high-risk equipment, depend on repeatable certainty and confidence of the data collected.
Currently, as provided in evidence the industry is subpar at this task. With a poor organizational structure and/ or human inconsistencies occurring in various industries, the level of uncertainty is in dire need to be reduced. According to the Joint Research Centre studying corrosion-related accidents in petroleum refineries, failure in risk management was a contributing cause to the vast majority of the 99 accidents studied. Today, with the rise in artificial intelligence and machine learning applications, certainty, and confidence in the change in asset condition can be realized and measured accurately.
For example, some companies utilize a laser scanner and point cloud data to fully contextualize every component in the facility, whether an offshore platform, FPSO, refinery or chemical plant. Laser scans of facilities are not new in nature and have been utilized and supplied benefits in the industry for many years. However, repeatability and reducing uncertainty were still lacking from a measured condition change standpoint.
Currently, machine learning algorithms can automatically detect, classify, measure, and categorize external damage mechanism morphology and most importantly able to measure the manifested change over time. Further, machine learning is used to identify, isolate, and contextualize the data so that all findings can be referenced to individual equipment, allowing the damage query task to be quickly performed.
A smart digital representation is built using actual fi eld data, enabling access to the most up-to-date representation of the facility. High-accuracy models deliver corrosion identification performance with 97% accuracy which was independently tested and validated. Whether the mechanism is external corrosion contributing to uncertainties in client’s fabric maintenance and mechanical integrity inspection as seen in offshore facilities around the world, or corrosion under pipe supports (CUPS) and corrosion under insulation (CUI) as witnessed throughout most facilities internationally, accuracy in the initial inspection and repeatability models in subsequent inspections supply the engineers with the certainty they need to make decisions.
As presented in Figure 1, a case study of an offshore platform was conducted. Every component was contextualized, allowing for the machine learning algorithm to automatically detect defects and accurately classify them. Importantly, 672 discrepancies from in the piping and instrumentation diagrams (P&id) were reconciled, creating an up-to-date P&id.
From an external damage mechanism standpoint, the risk assessment now possesses the ability to make decisions with certainty as these decisions are built on repeatable logic. As presented in Figure 2 during the rescan, the client was able to utilize their budget effectively, select the systems that needed immediate repair and apply an aggressive inspection frequency on the remaining other critical systems realizing maximum uptime.
New systems have now moved in and can be categorized and acquire a properly applied inspection frequency. Most commonly, as presented in industry studies, these systems are normally mid-inspection cycles when they fail because the degradation was not predicted correctly.
Conclusion
Uncertainty overwhelmingly still exists in the industry today. Uncertainty, however categorized, has also contributed to costly and even catastrophic events in the past. Specifically, in risk based systems, uncertainty has contributed to poor prediction models and again has caused avoidable yet disastrous events. As the infrastructure continues to age this uncertainty will compound and result in serious effects.
A solution has arrived in the artificial intelligence and machine learning models. Allowing the system to be a tool in the toolbox of inspectors, engineers, and even facility management will create enhanced decision-making capabilities. Strengthening the foundation of ground truth in the facilities will allow teams to apply focus correctly and make timely decisions saving the company resources and money. Now vulnerabilities are known and can be detected. Either an acceptance or improvement effort opportunity exists. The company is now responsible for the results. The actions taken from now on will produce results that can save lives and jobs.