J. Armstead-English, Seeq, Chicago, Illinois
Continuous improvement is a well-established practice and methodology that examines how to make processes increasingly more efficient. It encompasses the evaluation of current procedures and the identification of ways to improve operational effectiveness to avoid periods of nonproductivity.
In the oil and gas and petrochemicals industries, companies are always looking for ways to streamline efficiencies and perfect processes. In their operations, there are certain nonnegotiable protocols and procedures; however, viewing all steps through the lens of continuous improvement can empower teams to maximize effectiveness at each point.
Modern industrial operations rely on automated equipment for productivity, and companies have traditionally employed routine maintenance procedures to keep assets running reliably. However, maintenance typically requires brief periods of downtime that interrupt production. Regular, short periods of planned downtime are certainly preferrable to unexpected failures that result in outages of unknown duration, but there are ways to limit these interruptions to only what is required to minimize unanticipated downtime risk while maximizing profitability.
Conventional maintenance shortcomings. Often, plants contain assets or critical pieces of equipment that require routine and pre-planned maintenance to ensure reliability. For example, distillation towers—which are used to separate components within liquid mixtures based on their different boiling points—are subject to frequent fouling, rendering them a common bottleneck in operations (FIG. 1). Fouling occurs as organic deposits, side reactions and particulate matter build up inside the tower, which causes increased pressure drop downstream, greater energy consumption due to reduced heat transfer efficiency and other issues that impede the intended process, resulting in reduced capacity and potential lost production.
As plant personnel record periods of downtime, engineering and maintenance staff typically appoint planned maintenance windows at pre-scheduled intervals to reduce the chance of future upsets. Unfortunately, fouling occurrences are still unpredictable due to the variety of external factors responsible for their cause, severely impacting reliability, quality and production.
Predicting impending failures by monitoring conditions and performing calculations is a better approach because it enables scheduling maintenance only when it is required to prevent downtime as a result of process failures. However, employing effective predictive practices requires the right tools. Engineers often rely on spreadsheets to log data and perform calculations, but these are insufficient to identify root causes, discover complex problems and reliably predict failures in modern plants.
Effective analysis requires reviewing historical composition, temperature, pressure profile, flowrate and a plethora of other data—establishing sufficient predictive models in spreadsheets is complex at best, and nearly impossible at worst.
Predictive solutions. Addressing these and other challenges, machine-learning (ML)-equipped advanced analytics platforms significantly ease data integration, contextualization and predictive modeling procedures. This helps produce accurate process health predictions that empower teams to strategically schedule maintenance when required to prevent downtime. These software platforms leverage multivariate analytics to improve operational reliability.
Leading predictive models can forecast failure events and other process issues based on complex patterns within historical and real-time data. These tools are also effective for identifying anomalies in data that can precede numerous types of process upsets.
Upon implementation, these platforms collect data from field sensors, process historians, data lakes, enterprise resource planning and asset management systems (among other sources), then cleanse and organize the information. The results are combined to create a comprehensive dataset, helping fill in operational and maintenance gaps left within individual sources.
At this stage, the platform also automatically flags outliers and inconsistencies. Users can interact with the contextualized data using graphs, charts and heat maps to visualize trends, correlation and patterns, and to help piece together a well-understood overall plant story (FIG. 2).
Beyond these basic tools for amalgamating information, understanding process interactions and generating insights, modern advanced analytics platforms leverage ML models that self-improve over time to produce failure probabilities and communicate optimal maintenance intervals. These steps help minimize upkeep costs, facilitating just-in-time maintenance to stave off failure while preventing unnecessary downtime.
The required complex multivariate calculations execute in the shadow of a no-code front end to help plant personnel gain clear insights regardless of their analytic or programming skill levels, enhancing operational decision-making. Additionally, these platforms provide user interfaces designed to streamline workflows around process performance.
Predicting fouling in a distillation tower. A global specialty chemicals company leveraged the author’s company’s no-code, ML advanced analytics platforma to predict fouling in its distillation towers and improve operational efficiency. The company’s subject matter experts (SMEs) used the platform’s multivariate modeling capability to analyze numerous operating parameters and conditions to identify and predict fouling instances, including composition, temperature, pressure, flow data and more.
Armed with this information, the ML model produced nearly immediate insights covering multiple contributing factors, in addition to discernment regarding these factors’ interactions. The platform also provided plant personnel with high-level summary metrics, alongside trend views and recommended maintenance actions.
Using the software tools, the team began by identifying baseline operating conditions, along with target periods for analysis. SMEs used the platform’s signal selection tool to identify and remove low-variance signals, along with occurrences of high intercorrelations among multiple independent variables. They then ranked signal importance. These collective efforts emphasized the variables contributing most to fouling, and they highlighted causal relationships to compare dynamics in the stages before, during and after fouling as a method of determining root causes of these unfavorable outcomes (FIG. 3).
The manufacturer has perfected the ML model over time to the point where it now identifies conditions that lead to fouling two months before serious anomalies appear, providing plenty of time to act and perform required upkeep steps. These predictions communicate proper downtime planning as part of the company’s overall maintenance schedule.
Additionally, this manufacturer is now aware of the key contributors to fouling, which is primarily caused by condenser temperature variances. These findings have prompted operational changes that extend runtime between cleanings, increasing profitability through both enhanced efficiency and reduced downtime that was previously prompted by more frequent routine maintenance.
Digital tools to innovate. Data analyses, calculations and modeling were historically conducted using spreadsheets and other cumbersome tools. However, with the ever-increasing amount of information and data repositories in modern plant environments, efficient means of automation are necessary to parse out what the information truly means and why it matters.
ML-equipped advanced analytics platforms empower SMEs to solve some of the toughest process problems efficiently, implement effective process improvements, transition from reactive to predictive maintenance strategies, increase productivity and uptime, and operate more profitably. HP
NOTE
Seeq’s Process Health Solution
Janelle Armstead-English is the Chemicals Industry Principal at Seeq Corp. She has an engineering, market research, sales and product management background and earned dual BS degrees in chemical engineering and mathematics from the University of Pittsburgh. Armstead-English has nearly two decades of experience working with various chemical and petrochemical processors like Honeywell UOP and Praxair (now Linde). In her current role, she enjoys analyzing the ever-changing chemical and petrochemical markets, and understanding the challenges around digital transformation in these industries.
