November 15, 2020
By: Lyat Avidor Peleg
Increase Refinery Availability by Predicting Failures Before They Happen
Refineries and petrochemical plants have always faced competition, but this is increasing, due to the falling oil prices and other industry changes. With an abundance of data about their operations, refineries and petrochemical plants can use this information to increase their availability, providing them with an additional competitive advantage. We’d like to examine how improved technology like predictive maintenance in a refinery can help refineries and petrochemical plants become cutting edge “plants of tomorrow,” which will positively affect plant availability.
Maintaining high availability provides plants with a strong competitive lead, but there are many other benefits. One of those advantages is improved profitability. According to recent research, unplanned downtime costs the industry at least $2 million over the last 3 years. “A manufacturer’s bottom line can include up to 800 hours of downtime which translates into millions of dollars in revenue loss. Minimizing downtime in manufacturing is just as pivotal as maximizing quality and output to maintain contribution margins,” according to one source. Another reason high availability is so important is that it improves the overall safety of the plant, thus preventing injuries or even death; and when crises are averted, the refinery or plant avoids the ensuing negative publicity.
Plant fires and explosions can cause major damage, but there are many other failures that can occur during production processes leading to production loss and quality issues. For example, bolted joint and seal failure due to excessive vibration or relief valve chattering can lead to mechanical integrity failure.
Even Small Failures Affect Availability
Catastrophic incidents can be extremely costly to refineries and petrochemical plants. Arxiv cites research by the Aberdeen group showing that unplanned downtime costs between $10k and $250k per hour, or $50 billion annually, and equipment failure causes 42% of unplanned downtime. However, equipment failure does not have to be catastrophic to be problematic. Insidious losses from minor failures reduce productivity and are practically untraceable using traditional methods. These problems, if not treated, can eventually lead to massive breakdowns.
At any given time, in any given plant, there are usually a number of minor parts/processes that aren’t operating as they should. Most of the time, a minor malfunction doesn’t cause a serious incident, making it easy for maintenance teams to take a more laid-back approach to low-profile issues. However, when multiple minors errors or faults line up together, like “layers of Swiss cheese with holes in random places,” it can lead to catastrophe.
For example, in 2009 a large explosion at the Caribbean Petroleum Refinery in Puerto Rico caused extensive damage. It occurred when a tanker ship was offloading fuel into a tank farm, and the tank overflowed, causing a vapor cloud which then ignited and caused an explosion. Upon investigation, it was found that the tank side gauge transmitter had failed, and furthermore that it was often out of service. Maintenance teams may well have become accustomed to the fact that it generally wasn’t working because up until now, it hadn’t appeared to be an urgent issue.
A similarly “unimportant” event contributed greatly to the fatal explosion at the Deepwater Horizon Rig in 2010, where one of the emergency disconnect systems wasn’t functional because a miswired solenoid had drained a crucial battery. Every item that’s used in a project or receives maintenance is meant to be verified as in full working order, but a single solenoid is not high on the list of priorities, so it went overlooked.
As you can see from even these few examples (and there are hundreds more), there are plenty of minor things that can go wrong in a refinery or petrochemical plant, and a new method is needed to predict these failures. Traditionally, plants have operated with the logic “if something is broken, fix it.” Now, thanks to predictive analytics the paradigm can change to: “something is a little ‘off,’ let’s investigate and prevent it from deteriorating further.”
Leveraging Plant Data for Increased Availability
Refineries and petrochemical plants collect mountains of data, so why not leverage it to improve availability? With sensors tracking almost every element of the plant, and vast quantities of data collected in the historian, it is important to make sense of it all. However, current technological advances in predictive analytics can be applied to this data, alerting engineers to many potential failures and thus enabling them to prevent them. By identifying anomalies in the plant’s data and applying algorithms to determine the relevance of these anomalies, plant operations managers can focus on predictive monitoring and maintenance, rather than simply doing “preventive” maintenance according to a predetermined schedule or fixing things that have already broken.
Significant progress has been made in other industries by using artificial intelligence and machine learning for predictive monitoring, but solutions designed for discrete industries are not effective enough for refineries or petrochemical plants, unless they are adapted to the unique needs of process industries.
What to Look for in Predictive Monitoring
Artificial intelligence and machine learning are popular buzzwords, but an effective predictive monitoring solution for refineries and petrochemical plants will go far beyond mere anomaly detection using AI and ML. Some of the key aspects to watch out for include:
Cover everything in the plant, not just selected equipment. It can be tempting to cut corners by focusing on just the “important” areas of the plant. However, with every aspect of the plant connected to every other, you don’t know where the next failure will show up, and you can’t afford to miss it. It’s crucial to look for the unforeseen and unexpected incidents that could surprise you, not just what is obvious and “under the lantern.”
Add context by bridging the gap between anomalies in the data to true problems in the plant. It’s the connection between the different parameters, never just one anomaly, that predicts a failure. An effective solution will cluster the anomalies it finds, adding an additional cognitive layer that associates the identified issue with a specific area of the plant, to shorten time to resolution.
Avoid alert fatigue. If the system sends out too many alerts that don’t serve a useful purpose, operators will simply start to ignore them. When the system connects anomalies with the broader plant context, engineers can spot the most relevant alerts from those that simply show noise in the data. It’s particularly important in a process plant where anomalies are constantly cropping up and there is no real “normal.”
Rapid implementation. Once the plant management has decided that it is important to implement a system to enable predictive maintenance in the refinery, time is of the essence. Predictive monitoring systems that can be implemented quickly, without touching the core business of the plant, requiring just minimal consultation with the plant engineers, mean that you can enjoy the benefits as soon as possible without disrupting production.
No need for a data scientist. The operators or process managers know the plant inside and out, and they are the ideal people to investigate the alerts and resolve them.
On-premise or cloud offering. Some companies will prefer cloud implementations to offload the IT burden, others will prefer on-premise, so it is important that the option the company prefers is available to them.
Improving availability and overall equipment effectiveness by using predictive monitoring has many benefits. Not only can it make your plant more competitive, but it can keep your employees and the surrounding areas safe from unwanted explosions, fires, and other hazardous issues. This can ultimately help you to avoid negative publicity and propel your plant into the future of Industry 4.0. The answer is to predict failures before they happen.